xref: /linux/net/mptcp/pm_netlink.c (revision 4adfc94d4aeca1177e1188ba83c20ed581523fe1)
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 
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 *
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 
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 
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 
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 
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 
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 *
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 *
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 
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 
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 
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 
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 
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 *
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 
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 
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 *
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 
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 
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  */
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 
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 
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 *
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 *
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 
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 
616 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
617 {
618 	mptcp_pm_create_subflow_or_signal_addr(msk);
619 }
620 
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  */
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 
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 	int i, nr;
689 
690 	add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
691 	subflows_max = mptcp_pm_get_subflows_max(msk);
692 
693 	pr_debug("accepted %d:%d remote family %d",
694 		 msk->pm.add_addr_accepted, add_addr_accept_max,
695 		 msk->pm.remote.family);
696 
697 	remote = msk->pm.remote;
698 	mptcp_pm_announce_addr(msk, &remote, true);
699 	mptcp_pm_nl_addr_send_ack(msk);
700 
701 	if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
702 		return;
703 
704 	/* pick id 0 port, if none is provided the remote address */
705 	if (!remote.port)
706 		remote.port = sk->sk_dport;
707 
708 	/* connect to the specified remote address, using whatever
709 	 * local address the routing configuration will pick.
710 	 */
711 	nr = fill_local_addresses_vec(msk, &remote, addrs);
712 	if (nr == 0)
713 		return;
714 
715 	msk->pm.add_addr_accepted++;
716 	if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
717 	    msk->pm.subflows >= subflows_max)
718 		WRITE_ONCE(msk->pm.accept_addr, false);
719 
720 	spin_unlock_bh(&msk->pm.lock);
721 	for (i = 0; i < nr; i++)
722 		__mptcp_subflow_connect(sk, &addrs[i], &remote);
723 	spin_lock_bh(&msk->pm.lock);
724 }
725 
726 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
727 {
728 	struct mptcp_subflow_context *subflow;
729 
730 	msk_owned_by_me(msk);
731 	lockdep_assert_held(&msk->pm.lock);
732 
733 	if (!mptcp_pm_should_add_signal(msk) &&
734 	    !mptcp_pm_should_rm_signal(msk))
735 		return;
736 
737 	subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
738 	if (subflow)
739 		mptcp_pm_send_ack(msk, subflow, false, false);
740 }
741 
742 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
743 				 struct mptcp_addr_info *addr,
744 				 struct mptcp_addr_info *rem,
745 				 u8 bkup)
746 {
747 	struct mptcp_subflow_context *subflow;
748 
749 	pr_debug("bkup=%d", bkup);
750 
751 	mptcp_for_each_subflow(msk, subflow) {
752 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
753 		struct mptcp_addr_info local, remote;
754 
755 		mptcp_local_address((struct sock_common *)ssk, &local);
756 		if (!mptcp_addresses_equal(&local, addr, addr->port))
757 			continue;
758 
759 		if (rem && rem->family != AF_UNSPEC) {
760 			remote_address((struct sock_common *)ssk, &remote);
761 			if (!mptcp_addresses_equal(&remote, rem, rem->port))
762 				continue;
763 		}
764 
765 		__mptcp_pm_send_ack(msk, subflow, true, bkup);
766 		return 0;
767 	}
768 
769 	return -EINVAL;
770 }
771 
772 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id)
773 {
774 	return local_id == id || (!local_id && msk->mpc_endpoint_id == id);
775 }
776 
777 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
778 					   const struct mptcp_rm_list *rm_list,
779 					   enum linux_mptcp_mib_field rm_type)
780 {
781 	struct mptcp_subflow_context *subflow, *tmp;
782 	struct sock *sk = (struct sock *)msk;
783 	u8 i;
784 
785 	pr_debug("%s rm_list_nr %d",
786 		 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
787 
788 	msk_owned_by_me(msk);
789 
790 	if (sk->sk_state == TCP_LISTEN)
791 		return;
792 
793 	if (!rm_list->nr)
794 		return;
795 
796 	if (list_empty(&msk->conn_list))
797 		return;
798 
799 	for (i = 0; i < rm_list->nr; i++) {
800 		u8 rm_id = rm_list->ids[i];
801 		bool removed = false;
802 
803 		mptcp_for_each_subflow_safe(msk, subflow, tmp) {
804 			struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
805 			u8 remote_id = READ_ONCE(subflow->remote_id);
806 			int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
807 			u8 id = subflow_get_local_id(subflow);
808 
809 			if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id)
810 				continue;
811 			if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id))
812 				continue;
813 
814 			pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u",
815 				 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
816 				 i, rm_id, id, remote_id, msk->mpc_endpoint_id);
817 			spin_unlock_bh(&msk->pm.lock);
818 			mptcp_subflow_shutdown(sk, ssk, how);
819 
820 			/* the following takes care of updating the subflows counter */
821 			mptcp_close_ssk(sk, ssk, subflow);
822 			spin_lock_bh(&msk->pm.lock);
823 
824 			removed = true;
825 			__MPTCP_INC_STATS(sock_net(sk), rm_type);
826 		}
827 		if (rm_type == MPTCP_MIB_RMSUBFLOW)
828 			__set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap);
829 		if (!removed)
830 			continue;
831 
832 		if (!mptcp_pm_is_kernel(msk))
833 			continue;
834 
835 		if (rm_type == MPTCP_MIB_RMADDR) {
836 			msk->pm.add_addr_accepted--;
837 			WRITE_ONCE(msk->pm.accept_addr, true);
838 		} else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
839 			msk->pm.local_addr_used--;
840 		}
841 	}
842 }
843 
844 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
845 {
846 	mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
847 }
848 
849 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
850 				     const struct mptcp_rm_list *rm_list)
851 {
852 	mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
853 }
854 
855 void mptcp_pm_nl_work(struct mptcp_sock *msk)
856 {
857 	struct mptcp_pm_data *pm = &msk->pm;
858 
859 	msk_owned_by_me(msk);
860 
861 	if (!(pm->status & MPTCP_PM_WORK_MASK))
862 		return;
863 
864 	spin_lock_bh(&msk->pm.lock);
865 
866 	pr_debug("msk=%p status=%x", msk, pm->status);
867 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
868 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
869 		mptcp_pm_nl_add_addr_received(msk);
870 	}
871 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
872 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
873 		mptcp_pm_nl_addr_send_ack(msk);
874 	}
875 	if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
876 		pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
877 		mptcp_pm_nl_rm_addr_received(msk);
878 	}
879 	if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
880 		pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
881 		mptcp_pm_nl_fully_established(msk);
882 	}
883 	if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
884 		pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
885 		mptcp_pm_nl_subflow_established(msk);
886 	}
887 
888 	spin_unlock_bh(&msk->pm.lock);
889 }
890 
891 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
892 {
893 	return (entry->flags &
894 		(MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
895 		MPTCP_PM_ADDR_FLAG_SIGNAL;
896 }
897 
898 /* caller must ensure the RCU grace period is already elapsed */
899 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
900 {
901 	if (entry->lsk)
902 		sock_release(entry->lsk);
903 	kfree(entry);
904 }
905 
906 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
907 					     struct mptcp_pm_addr_entry *entry,
908 					     bool needs_id)
909 {
910 	struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
911 	unsigned int addr_max;
912 	int ret = -EINVAL;
913 
914 	spin_lock_bh(&pernet->lock);
915 	/* to keep the code simple, don't do IDR-like allocation for address ID,
916 	 * just bail when we exceed limits
917 	 */
918 	if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
919 		pernet->next_id = 1;
920 	if (pernet->addrs >= MPTCP_PM_ADDR_MAX) {
921 		ret = -ERANGE;
922 		goto out;
923 	}
924 	if (test_bit(entry->addr.id, pernet->id_bitmap)) {
925 		ret = -EBUSY;
926 		goto out;
927 	}
928 
929 	/* do not insert duplicate address, differentiate on port only
930 	 * singled addresses
931 	 */
932 	if (!address_use_port(entry))
933 		entry->addr.port = 0;
934 	list_for_each_entry(cur, &pernet->local_addr_list, list) {
935 		if (mptcp_addresses_equal(&cur->addr, &entry->addr,
936 					  cur->addr.port || entry->addr.port)) {
937 			/* allow replacing the exiting endpoint only if such
938 			 * endpoint is an implicit one and the user-space
939 			 * did not provide an endpoint id
940 			 */
941 			if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) {
942 				ret = -EEXIST;
943 				goto out;
944 			}
945 			if (entry->addr.id)
946 				goto out;
947 
948 			pernet->addrs--;
949 			entry->addr.id = cur->addr.id;
950 			list_del_rcu(&cur->list);
951 			del_entry = cur;
952 			break;
953 		}
954 	}
955 
956 	if (!entry->addr.id && needs_id) {
957 find_next:
958 		entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
959 						    MPTCP_PM_MAX_ADDR_ID + 1,
960 						    pernet->next_id);
961 		if (!entry->addr.id && pernet->next_id != 1) {
962 			pernet->next_id = 1;
963 			goto find_next;
964 		}
965 	}
966 
967 	if (!entry->addr.id && needs_id)
968 		goto out;
969 
970 	__set_bit(entry->addr.id, pernet->id_bitmap);
971 	if (entry->addr.id > pernet->next_id)
972 		pernet->next_id = entry->addr.id;
973 
974 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
975 		addr_max = pernet->add_addr_signal_max;
976 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
977 	}
978 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
979 		addr_max = pernet->local_addr_max;
980 		WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
981 	}
982 
983 	pernet->addrs++;
984 	if (!entry->addr.port)
985 		list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
986 	else
987 		list_add_rcu(&entry->list, &pernet->local_addr_list);
988 	ret = entry->addr.id;
989 
990 out:
991 	spin_unlock_bh(&pernet->lock);
992 
993 	/* just replaced an existing entry, free it */
994 	if (del_entry) {
995 		synchronize_rcu();
996 		__mptcp_pm_release_addr_entry(del_entry);
997 	}
998 	return ret;
999 }
1000 
1001 static struct lock_class_key mptcp_slock_keys[2];
1002 static struct lock_class_key mptcp_keys[2];
1003 
1004 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
1005 					    struct mptcp_pm_addr_entry *entry)
1006 {
1007 	bool is_ipv6 = sk->sk_family == AF_INET6;
1008 	int addrlen = sizeof(struct sockaddr_in);
1009 	struct sockaddr_storage addr;
1010 	struct sock *newsk, *ssk;
1011 	int backlog = 1024;
1012 	int err;
1013 
1014 	err = sock_create_kern(sock_net(sk), entry->addr.family,
1015 			       SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1016 	if (err)
1017 		return err;
1018 
1019 	newsk = entry->lsk->sk;
1020 	if (!newsk)
1021 		return -EINVAL;
1022 
1023 	/* The subflow socket lock is acquired in a nested to the msk one
1024 	 * in several places, even by the TCP stack, and this msk is a kernel
1025 	 * socket: lockdep complains. Instead of propagating the _nested
1026 	 * modifiers in several places, re-init the lock class for the msk
1027 	 * socket to an mptcp specific one.
1028 	 */
1029 	sock_lock_init_class_and_name(newsk,
1030 				      is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1031 				      &mptcp_slock_keys[is_ipv6],
1032 				      is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1033 				      &mptcp_keys[is_ipv6]);
1034 
1035 	lock_sock(newsk);
1036 	ssk = __mptcp_nmpc_sk(mptcp_sk(newsk));
1037 	release_sock(newsk);
1038 	if (IS_ERR(ssk))
1039 		return PTR_ERR(ssk);
1040 
1041 	mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1042 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1043 	if (entry->addr.family == AF_INET6)
1044 		addrlen = sizeof(struct sockaddr_in6);
1045 #endif
1046 	if (ssk->sk_family == AF_INET)
1047 		err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1048 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1049 	else if (ssk->sk_family == AF_INET6)
1050 		err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1051 #endif
1052 	if (err)
1053 		return err;
1054 
1055 	/* We don't use mptcp_set_state() here because it needs to be called
1056 	 * under the msk socket lock. For the moment, that will not bring
1057 	 * anything more than only calling inet_sk_state_store(), because the
1058 	 * old status is known (TCP_CLOSE).
1059 	 */
1060 	inet_sk_state_store(newsk, TCP_LISTEN);
1061 	lock_sock(ssk);
1062 	err = __inet_listen_sk(ssk, backlog);
1063 	if (!err)
1064 		mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
1065 	release_sock(ssk);
1066 	return err;
1067 }
1068 
1069 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc)
1070 {
1071 	struct mptcp_pm_addr_entry *entry;
1072 	struct pm_nl_pernet *pernet;
1073 	int ret = -1;
1074 
1075 	pernet = pm_nl_get_pernet_from_msk(msk);
1076 
1077 	rcu_read_lock();
1078 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1079 		if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) {
1080 			ret = entry->addr.id;
1081 			break;
1082 		}
1083 	}
1084 	rcu_read_unlock();
1085 	if (ret >= 0)
1086 		return ret;
1087 
1088 	/* address not found, add to local list */
1089 	entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1090 	if (!entry)
1091 		return -ENOMEM;
1092 
1093 	entry->addr = *skc;
1094 	entry->addr.id = 0;
1095 	entry->addr.port = 0;
1096 	entry->ifindex = 0;
1097 	entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1098 	entry->lsk = NULL;
1099 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true);
1100 	if (ret < 0)
1101 		kfree(entry);
1102 
1103 	return ret;
1104 }
1105 
1106 #define MPTCP_PM_CMD_GRP_OFFSET       0
1107 #define MPTCP_PM_EV_GRP_OFFSET        1
1108 
1109 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1110 	[MPTCP_PM_CMD_GRP_OFFSET]	= { .name = MPTCP_PM_CMD_GRP_NAME, },
1111 	[MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1112 					    .flags = GENL_MCAST_CAP_NET_ADMIN,
1113 					  },
1114 };
1115 
1116 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1117 {
1118 	struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1119 	struct sock *sk = (struct sock *)msk;
1120 	unsigned int active_max_loss_cnt;
1121 	struct net *net = sock_net(sk);
1122 	unsigned int stale_loss_cnt;
1123 	bool slow;
1124 
1125 	stale_loss_cnt = mptcp_stale_loss_cnt(net);
1126 	if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1127 		return;
1128 
1129 	/* look for another available subflow not in loss state */
1130 	active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1131 	mptcp_for_each_subflow(msk, iter) {
1132 		if (iter != subflow && mptcp_subflow_active(iter) &&
1133 		    iter->stale_count < active_max_loss_cnt) {
1134 			/* we have some alternatives, try to mark this subflow as idle ...*/
1135 			slow = lock_sock_fast(ssk);
1136 			if (!tcp_rtx_and_write_queues_empty(ssk)) {
1137 				subflow->stale = 1;
1138 				__mptcp_retransmit_pending_data(sk);
1139 				MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE);
1140 			}
1141 			unlock_sock_fast(ssk, slow);
1142 
1143 			/* always try to push the pending data regardless of re-injections:
1144 			 * we can possibly use backup subflows now, and subflow selection
1145 			 * is cheap under the msk socket lock
1146 			 */
1147 			__mptcp_push_pending(sk, 0);
1148 			return;
1149 		}
1150 	}
1151 }
1152 
1153 static int mptcp_pm_family_to_addr(int family)
1154 {
1155 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1156 	if (family == AF_INET6)
1157 		return MPTCP_PM_ADDR_ATTR_ADDR6;
1158 #endif
1159 	return MPTCP_PM_ADDR_ATTR_ADDR4;
1160 }
1161 
1162 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1163 				       const struct nlattr *attr,
1164 				       struct genl_info *info,
1165 				       struct mptcp_addr_info *addr,
1166 				       bool require_family)
1167 {
1168 	int err, addr_addr;
1169 
1170 	if (!attr) {
1171 		GENL_SET_ERR_MSG(info, "missing address info");
1172 		return -EINVAL;
1173 	}
1174 
1175 	/* no validation needed - was already done via nested policy */
1176 	err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1177 					  mptcp_pm_address_nl_policy, info->extack);
1178 	if (err)
1179 		return err;
1180 
1181 	if (tb[MPTCP_PM_ADDR_ATTR_ID])
1182 		addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1183 
1184 	if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1185 		if (!require_family)
1186 			return 0;
1187 
1188 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1189 				    "missing family");
1190 		return -EINVAL;
1191 	}
1192 
1193 	addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1194 	if (addr->family != AF_INET
1195 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1196 	    && addr->family != AF_INET6
1197 #endif
1198 	    ) {
1199 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1200 				    "unknown address family");
1201 		return -EINVAL;
1202 	}
1203 	addr_addr = mptcp_pm_family_to_addr(addr->family);
1204 	if (!tb[addr_addr]) {
1205 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1206 				    "missing address data");
1207 		return -EINVAL;
1208 	}
1209 
1210 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1211 	if (addr->family == AF_INET6)
1212 		addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1213 	else
1214 #endif
1215 		addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1216 
1217 	if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1218 		addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1219 
1220 	return 0;
1221 }
1222 
1223 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1224 			struct mptcp_addr_info *addr)
1225 {
1226 	struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1227 
1228 	memset(addr, 0, sizeof(*addr));
1229 
1230 	return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1231 }
1232 
1233 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1234 			 bool require_family,
1235 			 struct mptcp_pm_addr_entry *entry)
1236 {
1237 	struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1238 	int err;
1239 
1240 	memset(entry, 0, sizeof(*entry));
1241 
1242 	err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1243 	if (err)
1244 		return err;
1245 
1246 	if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1247 		u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1248 
1249 		entry->ifindex = val;
1250 	}
1251 
1252 	if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1253 		entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1254 
1255 	if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1256 		entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1257 
1258 	return 0;
1259 }
1260 
1261 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1262 {
1263 	return pm_nl_get_pernet(genl_info_net(info));
1264 }
1265 
1266 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1267 {
1268 	struct mptcp_sock *msk;
1269 	long s_slot = 0, s_num = 0;
1270 
1271 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1272 		struct sock *sk = (struct sock *)msk;
1273 
1274 		if (!READ_ONCE(msk->fully_established) ||
1275 		    mptcp_pm_is_userspace(msk))
1276 			goto next;
1277 
1278 		lock_sock(sk);
1279 		spin_lock_bh(&msk->pm.lock);
1280 		mptcp_pm_create_subflow_or_signal_addr(msk);
1281 		spin_unlock_bh(&msk->pm.lock);
1282 		release_sock(sk);
1283 
1284 next:
1285 		sock_put(sk);
1286 		cond_resched();
1287 	}
1288 
1289 	return 0;
1290 }
1291 
1292 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr,
1293 				      struct genl_info *info)
1294 {
1295 	struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1296 
1297 	if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1298 					 mptcp_pm_address_nl_policy, info->extack) &&
1299 	    tb[MPTCP_PM_ADDR_ATTR_ID])
1300 		return true;
1301 	return false;
1302 }
1303 
1304 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info)
1305 {
1306 	struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1307 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1308 	struct mptcp_pm_addr_entry addr, *entry;
1309 	int ret;
1310 
1311 	ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1312 	if (ret < 0)
1313 		return ret;
1314 
1315 	if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1316 		GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1317 		return -EINVAL;
1318 	}
1319 
1320 	if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1321 	    addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1322 		GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1323 		return -EINVAL;
1324 	}
1325 
1326 	if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1327 		GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1328 		return -EINVAL;
1329 	}
1330 
1331 	entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1332 	if (!entry) {
1333 		GENL_SET_ERR_MSG(info, "can't allocate addr");
1334 		return -ENOMEM;
1335 	}
1336 
1337 	*entry = addr;
1338 	if (entry->addr.port) {
1339 		ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1340 		if (ret) {
1341 			GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret);
1342 			goto out_free;
1343 		}
1344 	}
1345 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry,
1346 						!mptcp_pm_has_addr_attr_id(attr, info));
1347 	if (ret < 0) {
1348 		GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret);
1349 		goto out_free;
1350 	}
1351 
1352 	mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1353 	return 0;
1354 
1355 out_free:
1356 	__mptcp_pm_release_addr_entry(entry);
1357 	return ret;
1358 }
1359 
1360 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1361 					    u8 *flags, int *ifindex)
1362 {
1363 	struct mptcp_pm_addr_entry *entry;
1364 	struct sock *sk = (struct sock *)msk;
1365 	struct net *net = sock_net(sk);
1366 
1367 	rcu_read_lock();
1368 	entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1369 	if (entry) {
1370 		*flags = entry->flags;
1371 		*ifindex = entry->ifindex;
1372 	}
1373 	rcu_read_unlock();
1374 
1375 	return 0;
1376 }
1377 
1378 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1379 				      const struct mptcp_addr_info *addr)
1380 {
1381 	struct mptcp_pm_add_entry *entry;
1382 
1383 	entry = mptcp_pm_del_add_timer(msk, addr, false);
1384 	if (entry) {
1385 		list_del(&entry->list);
1386 		kfree(entry);
1387 		return true;
1388 	}
1389 
1390 	return false;
1391 }
1392 
1393 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1394 				      const struct mptcp_addr_info *addr,
1395 				      bool force)
1396 {
1397 	struct mptcp_rm_list list = { .nr = 0 };
1398 	bool ret;
1399 
1400 	list.ids[list.nr++] = addr->id;
1401 
1402 	ret = remove_anno_list_by_saddr(msk, addr);
1403 	if (ret || force) {
1404 		spin_lock_bh(&msk->pm.lock);
1405 		mptcp_pm_remove_addr(msk, &list);
1406 		spin_unlock_bh(&msk->pm.lock);
1407 	}
1408 	return ret;
1409 }
1410 
1411 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1412 						   const struct mptcp_pm_addr_entry *entry)
1413 {
1414 	const struct mptcp_addr_info *addr = &entry->addr;
1415 	struct mptcp_rm_list list = { .nr = 0 };
1416 	long s_slot = 0, s_num = 0;
1417 	struct mptcp_sock *msk;
1418 
1419 	pr_debug("remove_id=%d", addr->id);
1420 
1421 	list.ids[list.nr++] = addr->id;
1422 
1423 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1424 		struct sock *sk = (struct sock *)msk;
1425 		bool remove_subflow;
1426 
1427 		if (mptcp_pm_is_userspace(msk))
1428 			goto next;
1429 
1430 		if (list_empty(&msk->conn_list)) {
1431 			mptcp_pm_remove_anno_addr(msk, addr, false);
1432 			goto next;
1433 		}
1434 
1435 		lock_sock(sk);
1436 		remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1437 		mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1438 					  !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1439 		if (remove_subflow)
1440 			mptcp_pm_remove_subflow(msk, &list);
1441 		release_sock(sk);
1442 
1443 next:
1444 		sock_put(sk);
1445 		cond_resched();
1446 	}
1447 
1448 	return 0;
1449 }
1450 
1451 static int mptcp_nl_remove_id_zero_address(struct net *net,
1452 					   struct mptcp_addr_info *addr)
1453 {
1454 	struct mptcp_rm_list list = { .nr = 0 };
1455 	long s_slot = 0, s_num = 0;
1456 	struct mptcp_sock *msk;
1457 
1458 	list.ids[list.nr++] = 0;
1459 
1460 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1461 		struct sock *sk = (struct sock *)msk;
1462 		struct mptcp_addr_info msk_local;
1463 
1464 		if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1465 			goto next;
1466 
1467 		mptcp_local_address((struct sock_common *)msk, &msk_local);
1468 		if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1469 			goto next;
1470 
1471 		lock_sock(sk);
1472 		spin_lock_bh(&msk->pm.lock);
1473 		mptcp_pm_remove_addr(msk, &list);
1474 		mptcp_pm_nl_rm_subflow_received(msk, &list);
1475 		spin_unlock_bh(&msk->pm.lock);
1476 		release_sock(sk);
1477 
1478 next:
1479 		sock_put(sk);
1480 		cond_resched();
1481 	}
1482 
1483 	return 0;
1484 }
1485 
1486 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info)
1487 {
1488 	struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1489 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1490 	struct mptcp_pm_addr_entry addr, *entry;
1491 	unsigned int addr_max;
1492 	int ret;
1493 
1494 	ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1495 	if (ret < 0)
1496 		return ret;
1497 
1498 	/* the zero id address is special: the first address used by the msk
1499 	 * always gets such an id, so different subflows can have different zero
1500 	 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1501 	 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1502 	 */
1503 	if (addr.addr.id == 0)
1504 		return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1505 
1506 	spin_lock_bh(&pernet->lock);
1507 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1508 	if (!entry) {
1509 		GENL_SET_ERR_MSG(info, "address not found");
1510 		spin_unlock_bh(&pernet->lock);
1511 		return -EINVAL;
1512 	}
1513 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1514 		addr_max = pernet->add_addr_signal_max;
1515 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1516 	}
1517 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1518 		addr_max = pernet->local_addr_max;
1519 		WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1520 	}
1521 
1522 	pernet->addrs--;
1523 	list_del_rcu(&entry->list);
1524 	__clear_bit(entry->addr.id, pernet->id_bitmap);
1525 	spin_unlock_bh(&pernet->lock);
1526 
1527 	mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1528 	synchronize_rcu();
1529 	__mptcp_pm_release_addr_entry(entry);
1530 
1531 	return ret;
1532 }
1533 
1534 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1535 {
1536 	struct mptcp_rm_list alist = { .nr = 0 };
1537 	struct mptcp_pm_addr_entry *entry;
1538 
1539 	list_for_each_entry(entry, rm_list, list) {
1540 		if ((remove_anno_list_by_saddr(msk, &entry->addr) ||
1541 		     lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) &&
1542 		    alist.nr < MPTCP_RM_IDS_MAX)
1543 			alist.ids[alist.nr++] = entry->addr.id;
1544 	}
1545 
1546 	if (alist.nr) {
1547 		spin_lock_bh(&msk->pm.lock);
1548 		mptcp_pm_remove_addr(msk, &alist);
1549 		spin_unlock_bh(&msk->pm.lock);
1550 	}
1551 }
1552 
1553 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1554 					struct list_head *rm_list)
1555 {
1556 	struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1557 	struct mptcp_pm_addr_entry *entry;
1558 
1559 	list_for_each_entry(entry, rm_list, list) {
1560 		if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1561 		    slist.nr < MPTCP_RM_IDS_MAX)
1562 			slist.ids[slist.nr++] = entry->addr.id;
1563 
1564 		if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1565 		    alist.nr < MPTCP_RM_IDS_MAX)
1566 			alist.ids[alist.nr++] = entry->addr.id;
1567 	}
1568 
1569 	if (alist.nr) {
1570 		spin_lock_bh(&msk->pm.lock);
1571 		mptcp_pm_remove_addr(msk, &alist);
1572 		spin_unlock_bh(&msk->pm.lock);
1573 	}
1574 	if (slist.nr)
1575 		mptcp_pm_remove_subflow(msk, &slist);
1576 }
1577 
1578 static void mptcp_nl_remove_addrs_list(struct net *net,
1579 				       struct list_head *rm_list)
1580 {
1581 	long s_slot = 0, s_num = 0;
1582 	struct mptcp_sock *msk;
1583 
1584 	if (list_empty(rm_list))
1585 		return;
1586 
1587 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1588 		struct sock *sk = (struct sock *)msk;
1589 
1590 		if (!mptcp_pm_is_userspace(msk)) {
1591 			lock_sock(sk);
1592 			mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1593 			release_sock(sk);
1594 		}
1595 
1596 		sock_put(sk);
1597 		cond_resched();
1598 	}
1599 }
1600 
1601 /* caller must ensure the RCU grace period is already elapsed */
1602 static void __flush_addrs(struct list_head *list)
1603 {
1604 	while (!list_empty(list)) {
1605 		struct mptcp_pm_addr_entry *cur;
1606 
1607 		cur = list_entry(list->next,
1608 				 struct mptcp_pm_addr_entry, list);
1609 		list_del_rcu(&cur->list);
1610 		__mptcp_pm_release_addr_entry(cur);
1611 	}
1612 }
1613 
1614 static void __reset_counters(struct pm_nl_pernet *pernet)
1615 {
1616 	WRITE_ONCE(pernet->add_addr_signal_max, 0);
1617 	WRITE_ONCE(pernet->add_addr_accept_max, 0);
1618 	WRITE_ONCE(pernet->local_addr_max, 0);
1619 	pernet->addrs = 0;
1620 }
1621 
1622 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info)
1623 {
1624 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1625 	LIST_HEAD(free_list);
1626 
1627 	spin_lock_bh(&pernet->lock);
1628 	list_splice_init(&pernet->local_addr_list, &free_list);
1629 	__reset_counters(pernet);
1630 	pernet->next_id = 1;
1631 	bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1632 	spin_unlock_bh(&pernet->lock);
1633 	mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1634 	synchronize_rcu();
1635 	__flush_addrs(&free_list);
1636 	return 0;
1637 }
1638 
1639 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1640 			      struct mptcp_pm_addr_entry *entry)
1641 {
1642 	struct mptcp_addr_info *addr = &entry->addr;
1643 	struct nlattr *attr;
1644 
1645 	attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1646 	if (!attr)
1647 		return -EMSGSIZE;
1648 
1649 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1650 		goto nla_put_failure;
1651 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1652 		goto nla_put_failure;
1653 	if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1654 		goto nla_put_failure;
1655 	if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1656 		goto nla_put_failure;
1657 	if (entry->ifindex &&
1658 	    nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1659 		goto nla_put_failure;
1660 
1661 	if (addr->family == AF_INET &&
1662 	    nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1663 			    addr->addr.s_addr))
1664 		goto nla_put_failure;
1665 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1666 	else if (addr->family == AF_INET6 &&
1667 		 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1668 		goto nla_put_failure;
1669 #endif
1670 	nla_nest_end(skb, attr);
1671 	return 0;
1672 
1673 nla_put_failure:
1674 	nla_nest_cancel(skb, attr);
1675 	return -EMSGSIZE;
1676 }
1677 
1678 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info)
1679 {
1680 	struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1681 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1682 	struct mptcp_pm_addr_entry addr, *entry;
1683 	struct sk_buff *msg;
1684 	void *reply;
1685 	int ret;
1686 
1687 	ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1688 	if (ret < 0)
1689 		return ret;
1690 
1691 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1692 	if (!msg)
1693 		return -ENOMEM;
1694 
1695 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1696 				  info->genlhdr->cmd);
1697 	if (!reply) {
1698 		GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1699 		ret = -EMSGSIZE;
1700 		goto fail;
1701 	}
1702 
1703 	spin_lock_bh(&pernet->lock);
1704 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1705 	if (!entry) {
1706 		GENL_SET_ERR_MSG(info, "address not found");
1707 		ret = -EINVAL;
1708 		goto unlock_fail;
1709 	}
1710 
1711 	ret = mptcp_nl_fill_addr(msg, entry);
1712 	if (ret)
1713 		goto unlock_fail;
1714 
1715 	genlmsg_end(msg, reply);
1716 	ret = genlmsg_reply(msg, info);
1717 	spin_unlock_bh(&pernet->lock);
1718 	return ret;
1719 
1720 unlock_fail:
1721 	spin_unlock_bh(&pernet->lock);
1722 
1723 fail:
1724 	nlmsg_free(msg);
1725 	return ret;
1726 }
1727 
1728 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg,
1729 				struct netlink_callback *cb)
1730 {
1731 	struct net *net = sock_net(msg->sk);
1732 	struct mptcp_pm_addr_entry *entry;
1733 	struct pm_nl_pernet *pernet;
1734 	int id = cb->args[0];
1735 	void *hdr;
1736 	int i;
1737 
1738 	pernet = pm_nl_get_pernet(net);
1739 
1740 	spin_lock_bh(&pernet->lock);
1741 	for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1742 		if (test_bit(i, pernet->id_bitmap)) {
1743 			entry = __lookup_addr_by_id(pernet, i);
1744 			if (!entry)
1745 				break;
1746 
1747 			if (entry->addr.id <= id)
1748 				continue;
1749 
1750 			hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1751 					  cb->nlh->nlmsg_seq, &mptcp_genl_family,
1752 					  NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1753 			if (!hdr)
1754 				break;
1755 
1756 			if (mptcp_nl_fill_addr(msg, entry) < 0) {
1757 				genlmsg_cancel(msg, hdr);
1758 				break;
1759 			}
1760 
1761 			id = entry->addr.id;
1762 			genlmsg_end(msg, hdr);
1763 		}
1764 	}
1765 	spin_unlock_bh(&pernet->lock);
1766 
1767 	cb->args[0] = id;
1768 	return msg->len;
1769 }
1770 
1771 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1772 {
1773 	struct nlattr *attr = info->attrs[id];
1774 
1775 	if (!attr)
1776 		return 0;
1777 
1778 	*limit = nla_get_u32(attr);
1779 	if (*limit > MPTCP_PM_ADDR_MAX) {
1780 		GENL_SET_ERR_MSG(info, "limit greater than maximum");
1781 		return -EINVAL;
1782 	}
1783 	return 0;
1784 }
1785 
1786 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info)
1787 {
1788 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1789 	unsigned int rcv_addrs, subflows;
1790 	int ret;
1791 
1792 	spin_lock_bh(&pernet->lock);
1793 	rcv_addrs = pernet->add_addr_accept_max;
1794 	ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1795 	if (ret)
1796 		goto unlock;
1797 
1798 	subflows = pernet->subflows_max;
1799 	ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1800 	if (ret)
1801 		goto unlock;
1802 
1803 	WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1804 	WRITE_ONCE(pernet->subflows_max, subflows);
1805 
1806 unlock:
1807 	spin_unlock_bh(&pernet->lock);
1808 	return ret;
1809 }
1810 
1811 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info)
1812 {
1813 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1814 	struct sk_buff *msg;
1815 	void *reply;
1816 
1817 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1818 	if (!msg)
1819 		return -ENOMEM;
1820 
1821 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1822 				  MPTCP_PM_CMD_GET_LIMITS);
1823 	if (!reply)
1824 		goto fail;
1825 
1826 	if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1827 			READ_ONCE(pernet->add_addr_accept_max)))
1828 		goto fail;
1829 
1830 	if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1831 			READ_ONCE(pernet->subflows_max)))
1832 		goto fail;
1833 
1834 	genlmsg_end(msg, reply);
1835 	return genlmsg_reply(msg, info);
1836 
1837 fail:
1838 	GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1839 	nlmsg_free(msg);
1840 	return -EMSGSIZE;
1841 }
1842 
1843 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1844 				 struct mptcp_addr_info *addr)
1845 {
1846 	struct mptcp_rm_list list = { .nr = 0 };
1847 
1848 	list.ids[list.nr++] = addr->id;
1849 
1850 	spin_lock_bh(&msk->pm.lock);
1851 	mptcp_pm_nl_rm_subflow_received(msk, &list);
1852 	mptcp_pm_create_subflow_or_signal_addr(msk);
1853 	spin_unlock_bh(&msk->pm.lock);
1854 }
1855 
1856 static int mptcp_nl_set_flags(struct net *net,
1857 			      struct mptcp_addr_info *addr,
1858 			      u8 bkup, u8 changed)
1859 {
1860 	long s_slot = 0, s_num = 0;
1861 	struct mptcp_sock *msk;
1862 	int ret = -EINVAL;
1863 
1864 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1865 		struct sock *sk = (struct sock *)msk;
1866 
1867 		if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1868 			goto next;
1869 
1870 		lock_sock(sk);
1871 		if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1872 			ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
1873 		if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1874 			mptcp_pm_nl_fullmesh(msk, addr);
1875 		release_sock(sk);
1876 
1877 next:
1878 		sock_put(sk);
1879 		cond_resched();
1880 	}
1881 
1882 	return ret;
1883 }
1884 
1885 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup)
1886 {
1887 	struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
1888 	u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1889 			   MPTCP_PM_ADDR_FLAG_FULLMESH;
1890 	struct mptcp_pm_addr_entry *entry;
1891 	u8 lookup_by_id = 0;
1892 
1893 	if (addr->addr.family == AF_UNSPEC) {
1894 		lookup_by_id = 1;
1895 		if (!addr->addr.id)
1896 			return -EOPNOTSUPP;
1897 	}
1898 
1899 	spin_lock_bh(&pernet->lock);
1900 	entry = __lookup_addr(pernet, &addr->addr, lookup_by_id);
1901 	if (!entry) {
1902 		spin_unlock_bh(&pernet->lock);
1903 		return -EINVAL;
1904 	}
1905 	if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1906 	    (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1907 		spin_unlock_bh(&pernet->lock);
1908 		return -EINVAL;
1909 	}
1910 
1911 	changed = (addr->flags ^ entry->flags) & mask;
1912 	entry->flags = (entry->flags & ~mask) | (addr->flags & mask);
1913 	*addr = *entry;
1914 	spin_unlock_bh(&pernet->lock);
1915 
1916 	mptcp_nl_set_flags(net, &addr->addr, bkup, changed);
1917 	return 0;
1918 }
1919 
1920 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info)
1921 {
1922 	struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
1923 	struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, };
1924 	struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
1925 	struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
1926 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1927 	struct net *net = sock_net(skb->sk);
1928 	u8 bkup = 0;
1929 	int ret;
1930 
1931 	ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1932 	if (ret < 0)
1933 		return ret;
1934 
1935 	if (attr_rem) {
1936 		ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
1937 		if (ret < 0)
1938 			return ret;
1939 	}
1940 
1941 	if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1942 		bkup = 1;
1943 
1944 	return mptcp_pm_set_flags(net, token, &addr, &remote, bkup);
1945 }
1946 
1947 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1948 {
1949 	genlmsg_multicast_netns(&mptcp_genl_family, net,
1950 				nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1951 }
1952 
1953 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1954 {
1955 	return genl_has_listeners(&mptcp_genl_family,
1956 				  sock_net((const struct sock *)msk),
1957 				  MPTCP_PM_EV_GRP_OFFSET);
1958 }
1959 
1960 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1961 {
1962 	const struct inet_sock *issk = inet_sk(ssk);
1963 	const struct mptcp_subflow_context *sf;
1964 
1965 	if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1966 		return -EMSGSIZE;
1967 
1968 	switch (ssk->sk_family) {
1969 	case AF_INET:
1970 		if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1971 			return -EMSGSIZE;
1972 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1973 			return -EMSGSIZE;
1974 		break;
1975 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1976 	case AF_INET6: {
1977 		const struct ipv6_pinfo *np = inet6_sk(ssk);
1978 
1979 		if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1980 			return -EMSGSIZE;
1981 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1982 			return -EMSGSIZE;
1983 		break;
1984 	}
1985 #endif
1986 	default:
1987 		WARN_ON_ONCE(1);
1988 		return -EMSGSIZE;
1989 	}
1990 
1991 	if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1992 		return -EMSGSIZE;
1993 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1994 		return -EMSGSIZE;
1995 
1996 	sf = mptcp_subflow_ctx(ssk);
1997 	if (WARN_ON_ONCE(!sf))
1998 		return -EINVAL;
1999 
2000 	if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf)))
2001 		return -EMSGSIZE;
2002 
2003 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
2004 		return -EMSGSIZE;
2005 
2006 	return 0;
2007 }
2008 
2009 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
2010 					 const struct mptcp_sock *msk,
2011 					 const struct sock *ssk)
2012 {
2013 	const struct sock *sk = (const struct sock *)msk;
2014 	const struct mptcp_subflow_context *sf;
2015 	u8 sk_err;
2016 
2017 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2018 		return -EMSGSIZE;
2019 
2020 	if (mptcp_event_add_subflow(skb, ssk))
2021 		return -EMSGSIZE;
2022 
2023 	sf = mptcp_subflow_ctx(ssk);
2024 	if (WARN_ON_ONCE(!sf))
2025 		return -EINVAL;
2026 
2027 	if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2028 		return -EMSGSIZE;
2029 
2030 	if (ssk->sk_bound_dev_if &&
2031 	    nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2032 		return -EMSGSIZE;
2033 
2034 	sk_err = READ_ONCE(ssk->sk_err);
2035 	if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2036 	    nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2037 		return -EMSGSIZE;
2038 
2039 	return 0;
2040 }
2041 
2042 static int mptcp_event_sub_established(struct sk_buff *skb,
2043 				       const struct mptcp_sock *msk,
2044 				       const struct sock *ssk)
2045 {
2046 	return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2047 }
2048 
2049 static int mptcp_event_sub_closed(struct sk_buff *skb,
2050 				  const struct mptcp_sock *msk,
2051 				  const struct sock *ssk)
2052 {
2053 	const struct mptcp_subflow_context *sf;
2054 
2055 	if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2056 		return -EMSGSIZE;
2057 
2058 	sf = mptcp_subflow_ctx(ssk);
2059 	if (!sf->reset_seen)
2060 		return 0;
2061 
2062 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2063 		return -EMSGSIZE;
2064 
2065 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2066 		return -EMSGSIZE;
2067 
2068 	return 0;
2069 }
2070 
2071 static int mptcp_event_created(struct sk_buff *skb,
2072 			       const struct mptcp_sock *msk,
2073 			       const struct sock *ssk)
2074 {
2075 	int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2076 
2077 	if (err)
2078 		return err;
2079 
2080 	if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2081 		return -EMSGSIZE;
2082 
2083 	return mptcp_event_add_subflow(skb, ssk);
2084 }
2085 
2086 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2087 {
2088 	struct net *net = sock_net((const struct sock *)msk);
2089 	struct nlmsghdr *nlh;
2090 	struct sk_buff *skb;
2091 
2092 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2093 		return;
2094 
2095 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2096 	if (!skb)
2097 		return;
2098 
2099 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2100 	if (!nlh)
2101 		goto nla_put_failure;
2102 
2103 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2104 		goto nla_put_failure;
2105 
2106 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2107 		goto nla_put_failure;
2108 
2109 	genlmsg_end(skb, nlh);
2110 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2111 	return;
2112 
2113 nla_put_failure:
2114 	nlmsg_free(skb);
2115 }
2116 
2117 void mptcp_event_addr_announced(const struct sock *ssk,
2118 				const struct mptcp_addr_info *info)
2119 {
2120 	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2121 	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2122 	struct net *net = sock_net(ssk);
2123 	struct nlmsghdr *nlh;
2124 	struct sk_buff *skb;
2125 
2126 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2127 		return;
2128 
2129 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2130 	if (!skb)
2131 		return;
2132 
2133 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2134 			  MPTCP_EVENT_ANNOUNCED);
2135 	if (!nlh)
2136 		goto nla_put_failure;
2137 
2138 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2139 		goto nla_put_failure;
2140 
2141 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2142 		goto nla_put_failure;
2143 
2144 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2145 			 info->port == 0 ?
2146 			 inet_sk(ssk)->inet_dport :
2147 			 info->port))
2148 		goto nla_put_failure;
2149 
2150 	switch (info->family) {
2151 	case AF_INET:
2152 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2153 			goto nla_put_failure;
2154 		break;
2155 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2156 	case AF_INET6:
2157 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2158 			goto nla_put_failure;
2159 		break;
2160 #endif
2161 	default:
2162 		WARN_ON_ONCE(1);
2163 		goto nla_put_failure;
2164 	}
2165 
2166 	genlmsg_end(skb, nlh);
2167 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2168 	return;
2169 
2170 nla_put_failure:
2171 	nlmsg_free(skb);
2172 }
2173 
2174 void mptcp_event_pm_listener(const struct sock *ssk,
2175 			     enum mptcp_event_type event)
2176 {
2177 	const struct inet_sock *issk = inet_sk(ssk);
2178 	struct net *net = sock_net(ssk);
2179 	struct nlmsghdr *nlh;
2180 	struct sk_buff *skb;
2181 
2182 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2183 		return;
2184 
2185 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2186 	if (!skb)
2187 		return;
2188 
2189 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event);
2190 	if (!nlh)
2191 		goto nla_put_failure;
2192 
2193 	if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2194 		goto nla_put_failure;
2195 
2196 	if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2197 		goto nla_put_failure;
2198 
2199 	switch (ssk->sk_family) {
2200 	case AF_INET:
2201 		if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2202 			goto nla_put_failure;
2203 		break;
2204 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2205 	case AF_INET6: {
2206 		const struct ipv6_pinfo *np = inet6_sk(ssk);
2207 
2208 		if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2209 			goto nla_put_failure;
2210 		break;
2211 	}
2212 #endif
2213 	default:
2214 		WARN_ON_ONCE(1);
2215 		goto nla_put_failure;
2216 	}
2217 
2218 	genlmsg_end(skb, nlh);
2219 	mptcp_nl_mcast_send(net, skb, GFP_KERNEL);
2220 	return;
2221 
2222 nla_put_failure:
2223 	nlmsg_free(skb);
2224 }
2225 
2226 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2227 		 const struct sock *ssk, gfp_t gfp)
2228 {
2229 	struct net *net = sock_net((const struct sock *)msk);
2230 	struct nlmsghdr *nlh;
2231 	struct sk_buff *skb;
2232 
2233 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2234 		return;
2235 
2236 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2237 	if (!skb)
2238 		return;
2239 
2240 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2241 	if (!nlh)
2242 		goto nla_put_failure;
2243 
2244 	switch (type) {
2245 	case MPTCP_EVENT_UNSPEC:
2246 		WARN_ON_ONCE(1);
2247 		break;
2248 	case MPTCP_EVENT_CREATED:
2249 	case MPTCP_EVENT_ESTABLISHED:
2250 		if (mptcp_event_created(skb, msk, ssk) < 0)
2251 			goto nla_put_failure;
2252 		break;
2253 	case MPTCP_EVENT_CLOSED:
2254 		if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2255 			goto nla_put_failure;
2256 		break;
2257 	case MPTCP_EVENT_ANNOUNCED:
2258 	case MPTCP_EVENT_REMOVED:
2259 		/* call mptcp_event_addr_announced()/removed instead */
2260 		WARN_ON_ONCE(1);
2261 		break;
2262 	case MPTCP_EVENT_SUB_ESTABLISHED:
2263 	case MPTCP_EVENT_SUB_PRIORITY:
2264 		if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2265 			goto nla_put_failure;
2266 		break;
2267 	case MPTCP_EVENT_SUB_CLOSED:
2268 		if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2269 			goto nla_put_failure;
2270 		break;
2271 	case MPTCP_EVENT_LISTENER_CREATED:
2272 	case MPTCP_EVENT_LISTENER_CLOSED:
2273 		break;
2274 	}
2275 
2276 	genlmsg_end(skb, nlh);
2277 	mptcp_nl_mcast_send(net, skb, gfp);
2278 	return;
2279 
2280 nla_put_failure:
2281 	nlmsg_free(skb);
2282 }
2283 
2284 static struct genl_family mptcp_genl_family __ro_after_init = {
2285 	.name		= MPTCP_PM_NAME,
2286 	.version	= MPTCP_PM_VER,
2287 	.netnsok	= true,
2288 	.module		= THIS_MODULE,
2289 	.ops		= mptcp_pm_nl_ops,
2290 	.n_ops		= ARRAY_SIZE(mptcp_pm_nl_ops),
2291 	.resv_start_op	= MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2292 	.mcgrps		= mptcp_pm_mcgrps,
2293 	.n_mcgrps	= ARRAY_SIZE(mptcp_pm_mcgrps),
2294 };
2295 
2296 static int __net_init pm_nl_init_net(struct net *net)
2297 {
2298 	struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2299 
2300 	INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2301 
2302 	/* Cit. 2 subflows ought to be enough for anybody. */
2303 	pernet->subflows_max = 2;
2304 	pernet->next_id = 1;
2305 	pernet->stale_loss_cnt = 4;
2306 	spin_lock_init(&pernet->lock);
2307 
2308 	/* No need to initialize other pernet fields, the struct is zeroed at
2309 	 * allocation time.
2310 	 */
2311 
2312 	return 0;
2313 }
2314 
2315 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2316 {
2317 	struct net *net;
2318 
2319 	list_for_each_entry(net, net_list, exit_list) {
2320 		struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2321 
2322 		/* net is removed from namespace list, can't race with
2323 		 * other modifiers, also netns core already waited for a
2324 		 * RCU grace period.
2325 		 */
2326 		__flush_addrs(&pernet->local_addr_list);
2327 	}
2328 }
2329 
2330 static struct pernet_operations mptcp_pm_pernet_ops = {
2331 	.init = pm_nl_init_net,
2332 	.exit_batch = pm_nl_exit_net,
2333 	.id = &pm_nl_pernet_id,
2334 	.size = sizeof(struct pm_nl_pernet),
2335 };
2336 
2337 void __init mptcp_pm_nl_init(void)
2338 {
2339 	if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2340 		panic("Failed to register MPTCP PM pernet subsystem.\n");
2341 
2342 	if (genl_register_family(&mptcp_genl_family))
2343 		panic("Failed to register MPTCP PM netlink family\n");
2344 }
2345