xref: /openbmc/linux/net/mptcp/pm_netlink.c (revision 8e74a48d)
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/netns/generic.h>
13 #include <net/mptcp.h>
14 #include <net/genetlink.h>
15 #include <uapi/linux/mptcp.h>
16 
17 #include "protocol.h"
18 #include "mib.h"
19 
20 /* forward declaration */
21 static struct genl_family mptcp_genl_family;
22 
23 static int pm_nl_pernet_id;
24 
25 struct mptcp_pm_addr_entry {
26 	struct list_head	list;
27 	struct mptcp_addr_info	addr;
28 	u8			flags;
29 	int			ifindex;
30 	struct socket		*lsk;
31 };
32 
33 struct mptcp_pm_add_entry {
34 	struct list_head	list;
35 	struct mptcp_addr_info	addr;
36 	struct timer_list	add_timer;
37 	struct mptcp_sock	*sock;
38 	u8			retrans_times;
39 };
40 
41 struct pm_nl_pernet {
42 	/* protects pernet updates */
43 	spinlock_t		lock;
44 	struct list_head	local_addr_list;
45 	unsigned int		addrs;
46 	unsigned int		stale_loss_cnt;
47 	unsigned int		add_addr_signal_max;
48 	unsigned int		add_addr_accept_max;
49 	unsigned int		local_addr_max;
50 	unsigned int		subflows_max;
51 	unsigned int		next_id;
52 	DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
53 };
54 
55 #define MPTCP_PM_ADDR_MAX	8
56 #define ADD_ADDR_RETRANS_MAX	3
57 
58 static bool addresses_equal(const struct mptcp_addr_info *a,
59 			    const struct mptcp_addr_info *b, bool use_port)
60 {
61 	bool addr_equals = false;
62 
63 	if (a->family == b->family) {
64 		if (a->family == AF_INET)
65 			addr_equals = a->addr.s_addr == b->addr.s_addr;
66 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
67 		else
68 			addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
69 	} else if (a->family == AF_INET) {
70 		if (ipv6_addr_v4mapped(&b->addr6))
71 			addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
72 	} else if (b->family == AF_INET) {
73 		if (ipv6_addr_v4mapped(&a->addr6))
74 			addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
75 #endif
76 	}
77 
78 	if (!addr_equals)
79 		return false;
80 	if (!use_port)
81 		return true;
82 
83 	return a->port == b->port;
84 }
85 
86 static bool address_zero(const struct mptcp_addr_info *addr)
87 {
88 	struct mptcp_addr_info zero;
89 
90 	memset(&zero, 0, sizeof(zero));
91 	zero.family = addr->family;
92 
93 	return addresses_equal(addr, &zero, true);
94 }
95 
96 static void local_address(const struct sock_common *skc,
97 			  struct mptcp_addr_info *addr)
98 {
99 	addr->family = skc->skc_family;
100 	addr->port = htons(skc->skc_num);
101 	if (addr->family == AF_INET)
102 		addr->addr.s_addr = skc->skc_rcv_saddr;
103 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
104 	else if (addr->family == AF_INET6)
105 		addr->addr6 = skc->skc_v6_rcv_saddr;
106 #endif
107 }
108 
109 static void remote_address(const struct sock_common *skc,
110 			   struct mptcp_addr_info *addr)
111 {
112 	addr->family = skc->skc_family;
113 	addr->port = skc->skc_dport;
114 	if (addr->family == AF_INET)
115 		addr->addr.s_addr = skc->skc_daddr;
116 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
117 	else if (addr->family == AF_INET6)
118 		addr->addr6 = skc->skc_v6_daddr;
119 #endif
120 }
121 
122 static bool lookup_subflow_by_saddr(const struct list_head *list,
123 				    struct mptcp_addr_info *saddr)
124 {
125 	struct mptcp_subflow_context *subflow;
126 	struct mptcp_addr_info cur;
127 	struct sock_common *skc;
128 
129 	list_for_each_entry(subflow, list, node) {
130 		skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
131 
132 		local_address(skc, &cur);
133 		if (addresses_equal(&cur, saddr, saddr->port))
134 			return true;
135 	}
136 
137 	return false;
138 }
139 
140 static bool lookup_subflow_by_daddr(const struct list_head *list,
141 				    struct mptcp_addr_info *daddr)
142 {
143 	struct mptcp_subflow_context *subflow;
144 	struct mptcp_addr_info cur;
145 	struct sock_common *skc;
146 
147 	list_for_each_entry(subflow, list, node) {
148 		skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
149 
150 		remote_address(skc, &cur);
151 		if (addresses_equal(&cur, daddr, daddr->port))
152 			return true;
153 	}
154 
155 	return false;
156 }
157 
158 static struct mptcp_pm_addr_entry *
159 select_local_address(const struct pm_nl_pernet *pernet,
160 		     struct mptcp_sock *msk)
161 {
162 	struct mptcp_pm_addr_entry *entry, *ret = NULL;
163 	struct sock *sk = (struct sock *)msk;
164 
165 	msk_owned_by_me(msk);
166 
167 	rcu_read_lock();
168 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
169 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
170 			continue;
171 
172 		if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
173 			continue;
174 
175 		if (entry->addr.family != sk->sk_family) {
176 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
177 			if ((entry->addr.family == AF_INET &&
178 			     !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
179 			    (sk->sk_family == AF_INET &&
180 			     !ipv6_addr_v4mapped(&entry->addr.addr6)))
181 #endif
182 				continue;
183 		}
184 
185 		ret = entry;
186 		break;
187 	}
188 	rcu_read_unlock();
189 	return ret;
190 }
191 
192 static struct mptcp_pm_addr_entry *
193 select_signal_address(struct pm_nl_pernet *pernet, struct mptcp_sock *msk)
194 {
195 	struct mptcp_pm_addr_entry *entry, *ret = NULL;
196 
197 	rcu_read_lock();
198 	/* do not keep any additional per socket state, just signal
199 	 * the address list in order.
200 	 * Note: removal from the local address list during the msk life-cycle
201 	 * can lead to additional addresses not being announced.
202 	 */
203 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
204 		if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
205 			continue;
206 
207 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
208 			continue;
209 
210 		ret = entry;
211 		break;
212 	}
213 	rcu_read_unlock();
214 	return ret;
215 }
216 
217 unsigned int mptcp_pm_get_add_addr_signal_max(struct mptcp_sock *msk)
218 {
219 	struct pm_nl_pernet *pernet;
220 
221 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
222 	return READ_ONCE(pernet->add_addr_signal_max);
223 }
224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
225 
226 unsigned int mptcp_pm_get_add_addr_accept_max(struct mptcp_sock *msk)
227 {
228 	struct pm_nl_pernet *pernet;
229 
230 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
231 	return READ_ONCE(pernet->add_addr_accept_max);
232 }
233 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
234 
235 unsigned int mptcp_pm_get_subflows_max(struct mptcp_sock *msk)
236 {
237 	struct pm_nl_pernet *pernet;
238 
239 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
240 	return READ_ONCE(pernet->subflows_max);
241 }
242 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
243 
244 unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk)
245 {
246 	struct pm_nl_pernet *pernet;
247 
248 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
249 	return READ_ONCE(pernet->local_addr_max);
250 }
251 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
252 
253 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
254 {
255 	struct pm_nl_pernet *pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
256 
257 	if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
258 	    (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
259 			       MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
260 		WRITE_ONCE(msk->pm.work_pending, false);
261 		return false;
262 	}
263 	return true;
264 }
265 
266 struct mptcp_pm_add_entry *
267 mptcp_lookup_anno_list_by_saddr(struct mptcp_sock *msk,
268 				struct mptcp_addr_info *addr)
269 {
270 	struct mptcp_pm_add_entry *entry;
271 
272 	lockdep_assert_held(&msk->pm.lock);
273 
274 	list_for_each_entry(entry, &msk->pm.anno_list, list) {
275 		if (addresses_equal(&entry->addr, addr, true))
276 			return entry;
277 	}
278 
279 	return NULL;
280 }
281 
282 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
283 {
284 	struct mptcp_pm_add_entry *entry;
285 	struct mptcp_addr_info saddr;
286 	bool ret = false;
287 
288 	local_address((struct sock_common *)sk, &saddr);
289 
290 	spin_lock_bh(&msk->pm.lock);
291 	list_for_each_entry(entry, &msk->pm.anno_list, list) {
292 		if (addresses_equal(&entry->addr, &saddr, true)) {
293 			ret = true;
294 			goto out;
295 		}
296 	}
297 
298 out:
299 	spin_unlock_bh(&msk->pm.lock);
300 	return ret;
301 }
302 
303 static void mptcp_pm_add_timer(struct timer_list *timer)
304 {
305 	struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
306 	struct mptcp_sock *msk = entry->sock;
307 	struct sock *sk = (struct sock *)msk;
308 
309 	pr_debug("msk=%p", msk);
310 
311 	if (!msk)
312 		return;
313 
314 	if (inet_sk_state_load(sk) == TCP_CLOSE)
315 		return;
316 
317 	if (!entry->addr.id)
318 		return;
319 
320 	if (mptcp_pm_should_add_signal_addr(msk)) {
321 		sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
322 		goto out;
323 	}
324 
325 	spin_lock_bh(&msk->pm.lock);
326 
327 	if (!mptcp_pm_should_add_signal_addr(msk)) {
328 		pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
329 		mptcp_pm_announce_addr(msk, &entry->addr, false);
330 		mptcp_pm_add_addr_send_ack(msk);
331 		entry->retrans_times++;
332 	}
333 
334 	if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
335 		sk_reset_timer(sk, timer,
336 			       jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
337 
338 	spin_unlock_bh(&msk->pm.lock);
339 
340 	if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
341 		mptcp_pm_subflow_established(msk);
342 
343 out:
344 	__sock_put(sk);
345 }
346 
347 struct mptcp_pm_add_entry *
348 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
349 		       struct mptcp_addr_info *addr, bool check_id)
350 {
351 	struct mptcp_pm_add_entry *entry;
352 	struct sock *sk = (struct sock *)msk;
353 
354 	spin_lock_bh(&msk->pm.lock);
355 	entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
356 	if (entry && (!check_id || entry->addr.id == addr->id))
357 		entry->retrans_times = ADD_ADDR_RETRANS_MAX;
358 	spin_unlock_bh(&msk->pm.lock);
359 
360 	if (entry && (!check_id || entry->addr.id == addr->id))
361 		sk_stop_timer_sync(sk, &entry->add_timer);
362 
363 	return entry;
364 }
365 
366 static bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
367 				     struct mptcp_pm_addr_entry *entry)
368 {
369 	struct mptcp_pm_add_entry *add_entry = NULL;
370 	struct sock *sk = (struct sock *)msk;
371 	struct net *net = sock_net(sk);
372 
373 	lockdep_assert_held(&msk->pm.lock);
374 
375 	if (mptcp_lookup_anno_list_by_saddr(msk, &entry->addr))
376 		return false;
377 
378 	add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
379 	if (!add_entry)
380 		return false;
381 
382 	list_add(&add_entry->list, &msk->pm.anno_list);
383 
384 	add_entry->addr = entry->addr;
385 	add_entry->sock = msk;
386 	add_entry->retrans_times = 0;
387 
388 	timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
389 	sk_reset_timer(sk, &add_entry->add_timer,
390 		       jiffies + mptcp_get_add_addr_timeout(net));
391 
392 	return true;
393 }
394 
395 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
396 {
397 	struct mptcp_pm_add_entry *entry, *tmp;
398 	struct sock *sk = (struct sock *)msk;
399 	LIST_HEAD(free_list);
400 
401 	pr_debug("msk=%p", msk);
402 
403 	spin_lock_bh(&msk->pm.lock);
404 	list_splice_init(&msk->pm.anno_list, &free_list);
405 	spin_unlock_bh(&msk->pm.lock);
406 
407 	list_for_each_entry_safe(entry, tmp, &free_list, list) {
408 		sk_stop_timer_sync(sk, &entry->add_timer);
409 		kfree(entry);
410 	}
411 }
412 
413 static bool lookup_address_in_vec(struct mptcp_addr_info *addrs, unsigned int nr,
414 				  struct mptcp_addr_info *addr)
415 {
416 	int i;
417 
418 	for (i = 0; i < nr; i++) {
419 		if (addresses_equal(&addrs[i], addr, addr->port))
420 			return true;
421 	}
422 
423 	return false;
424 }
425 
426 /* Fill all the remote addresses into the array addrs[],
427  * and return the array size.
428  */
429 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, bool fullmesh,
430 					      struct mptcp_addr_info *addrs)
431 {
432 	bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
433 	struct sock *sk = (struct sock *)msk, *ssk;
434 	struct mptcp_subflow_context *subflow;
435 	struct mptcp_addr_info remote = { 0 };
436 	unsigned int subflows_max;
437 	int i = 0;
438 
439 	subflows_max = mptcp_pm_get_subflows_max(msk);
440 	remote_address((struct sock_common *)sk, &remote);
441 
442 	/* Non-fullmesh endpoint, fill in the single entry
443 	 * corresponding to the primary MPC subflow remote address
444 	 */
445 	if (!fullmesh) {
446 		if (deny_id0)
447 			return 0;
448 
449 		msk->pm.subflows++;
450 		addrs[i++] = remote;
451 	} else {
452 		mptcp_for_each_subflow(msk, subflow) {
453 			ssk = mptcp_subflow_tcp_sock(subflow);
454 			remote_address((struct sock_common *)ssk, &addrs[i]);
455 			if (deny_id0 && addresses_equal(&addrs[i], &remote, false))
456 				continue;
457 
458 			if (!lookup_address_in_vec(addrs, i, &addrs[i]) &&
459 			    msk->pm.subflows < subflows_max) {
460 				msk->pm.subflows++;
461 				i++;
462 			}
463 		}
464 	}
465 
466 	return i;
467 }
468 
469 static struct mptcp_pm_addr_entry *
470 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
471 {
472 	struct mptcp_pm_addr_entry *entry;
473 
474 	list_for_each_entry(entry, &pernet->local_addr_list, list) {
475 		if (entry->addr.id == id)
476 			return entry;
477 	}
478 	return NULL;
479 }
480 
481 static int
482 lookup_id_by_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *addr)
483 {
484 	struct mptcp_pm_addr_entry *entry;
485 	int ret = -1;
486 
487 	rcu_read_lock();
488 	list_for_each_entry(entry, &pernet->local_addr_list, list) {
489 		if (addresses_equal(&entry->addr, addr, entry->addr.port)) {
490 			ret = entry->addr.id;
491 			break;
492 		}
493 	}
494 	rcu_read_unlock();
495 	return ret;
496 }
497 
498 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
499 {
500 	struct sock *sk = (struct sock *)msk;
501 	struct mptcp_pm_addr_entry *local;
502 	unsigned int add_addr_signal_max;
503 	unsigned int local_addr_max;
504 	struct pm_nl_pernet *pernet;
505 	unsigned int subflows_max;
506 
507 	pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
508 
509 	add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
510 	local_addr_max = mptcp_pm_get_local_addr_max(msk);
511 	subflows_max = mptcp_pm_get_subflows_max(msk);
512 
513 	/* do lazy endpoint usage accounting for the MPC subflows */
514 	if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
515 		struct mptcp_addr_info mpc_addr;
516 		int mpc_id;
517 
518 		local_address((struct sock_common *)msk->first, &mpc_addr);
519 		mpc_id = lookup_id_by_addr(pernet, &mpc_addr);
520 		if (mpc_id >= 0)
521 			__clear_bit(mpc_id, msk->pm.id_avail_bitmap);
522 
523 		msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
524 	}
525 
526 	pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
527 		 msk->pm.local_addr_used, local_addr_max,
528 		 msk->pm.add_addr_signaled, add_addr_signal_max,
529 		 msk->pm.subflows, subflows_max);
530 
531 	/* check first for announce */
532 	if (msk->pm.add_addr_signaled < add_addr_signal_max) {
533 		local = select_signal_address(pernet, msk);
534 
535 		if (local) {
536 			if (mptcp_pm_alloc_anno_list(msk, local)) {
537 				__clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
538 				msk->pm.add_addr_signaled++;
539 				mptcp_pm_announce_addr(msk, &local->addr, false);
540 				mptcp_pm_nl_addr_send_ack(msk);
541 			}
542 		}
543 	}
544 
545 	/* check if should create a new subflow */
546 	while (msk->pm.local_addr_used < local_addr_max &&
547 	       msk->pm.subflows < subflows_max) {
548 		struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
549 		bool fullmesh;
550 		int i, nr;
551 
552 		local = select_local_address(pernet, msk);
553 		if (!local)
554 			break;
555 
556 		fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
557 
558 		msk->pm.local_addr_used++;
559 		nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
560 		if (nr)
561 			__clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
562 		spin_unlock_bh(&msk->pm.lock);
563 		for (i = 0; i < nr; i++)
564 			__mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
565 		spin_lock_bh(&msk->pm.lock);
566 	}
567 	mptcp_pm_nl_check_work_pending(msk);
568 }
569 
570 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
571 {
572 	mptcp_pm_create_subflow_or_signal_addr(msk);
573 }
574 
575 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
576 {
577 	mptcp_pm_create_subflow_or_signal_addr(msk);
578 }
579 
580 /* Fill all the local addresses into the array addrs[],
581  * and return the array size.
582  */
583 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
584 					     struct mptcp_addr_info *addrs)
585 {
586 	struct sock *sk = (struct sock *)msk;
587 	struct mptcp_pm_addr_entry *entry;
588 	struct mptcp_addr_info local;
589 	struct pm_nl_pernet *pernet;
590 	unsigned int subflows_max;
591 	int i = 0;
592 
593 	pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
594 	subflows_max = mptcp_pm_get_subflows_max(msk);
595 
596 	rcu_read_lock();
597 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
598 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
599 			continue;
600 
601 		if (entry->addr.family != sk->sk_family) {
602 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
603 			if ((entry->addr.family == AF_INET &&
604 			     !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
605 			    (sk->sk_family == AF_INET &&
606 			     !ipv6_addr_v4mapped(&entry->addr.addr6)))
607 #endif
608 				continue;
609 		}
610 
611 		if (msk->pm.subflows < subflows_max) {
612 			msk->pm.subflows++;
613 			addrs[i++] = entry->addr;
614 		}
615 	}
616 	rcu_read_unlock();
617 
618 	/* If the array is empty, fill in the single
619 	 * 'IPADDRANY' local address
620 	 */
621 	if (!i) {
622 		memset(&local, 0, sizeof(local));
623 		local.family = msk->pm.remote.family;
624 
625 		msk->pm.subflows++;
626 		addrs[i++] = local;
627 	}
628 
629 	return i;
630 }
631 
632 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
633 {
634 	struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
635 	struct sock *sk = (struct sock *)msk;
636 	unsigned int add_addr_accept_max;
637 	struct mptcp_addr_info remote;
638 	unsigned int subflows_max;
639 	int i, nr;
640 
641 	add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
642 	subflows_max = mptcp_pm_get_subflows_max(msk);
643 
644 	pr_debug("accepted %d:%d remote family %d",
645 		 msk->pm.add_addr_accepted, add_addr_accept_max,
646 		 msk->pm.remote.family);
647 
648 	if (lookup_subflow_by_daddr(&msk->conn_list, &msk->pm.remote))
649 		goto add_addr_echo;
650 
651 	/* connect to the specified remote address, using whatever
652 	 * local address the routing configuration will pick.
653 	 */
654 	remote = msk->pm.remote;
655 	if (!remote.port)
656 		remote.port = sk->sk_dport;
657 	nr = fill_local_addresses_vec(msk, addrs);
658 
659 	msk->pm.add_addr_accepted++;
660 	if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
661 	    msk->pm.subflows >= subflows_max)
662 		WRITE_ONCE(msk->pm.accept_addr, false);
663 
664 	spin_unlock_bh(&msk->pm.lock);
665 	for (i = 0; i < nr; i++)
666 		__mptcp_subflow_connect(sk, &addrs[i], &remote);
667 	spin_lock_bh(&msk->pm.lock);
668 
669 add_addr_echo:
670 	mptcp_pm_announce_addr(msk, &msk->pm.remote, true);
671 	mptcp_pm_nl_addr_send_ack(msk);
672 }
673 
674 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
675 {
676 	struct mptcp_subflow_context *subflow;
677 
678 	msk_owned_by_me(msk);
679 	lockdep_assert_held(&msk->pm.lock);
680 
681 	if (!mptcp_pm_should_add_signal(msk) &&
682 	    !mptcp_pm_should_rm_signal(msk))
683 		return;
684 
685 	subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
686 	if (subflow) {
687 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
688 
689 		spin_unlock_bh(&msk->pm.lock);
690 		pr_debug("send ack for %s",
691 			 mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
692 
693 		mptcp_subflow_send_ack(ssk);
694 		spin_lock_bh(&msk->pm.lock);
695 	}
696 }
697 
698 static int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
699 					struct mptcp_addr_info *addr,
700 					u8 bkup)
701 {
702 	struct mptcp_subflow_context *subflow;
703 
704 	pr_debug("bkup=%d", bkup);
705 
706 	mptcp_for_each_subflow(msk, subflow) {
707 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
708 		struct sock *sk = (struct sock *)msk;
709 		struct mptcp_addr_info local;
710 
711 		local_address((struct sock_common *)ssk, &local);
712 		if (!addresses_equal(&local, addr, addr->port))
713 			continue;
714 
715 		subflow->backup = bkup;
716 		subflow->send_mp_prio = 1;
717 		subflow->request_bkup = bkup;
718 		__MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
719 
720 		spin_unlock_bh(&msk->pm.lock);
721 		pr_debug("send ack for mp_prio");
722 		mptcp_subflow_send_ack(ssk);
723 		spin_lock_bh(&msk->pm.lock);
724 
725 		return 0;
726 	}
727 
728 	return -EINVAL;
729 }
730 
731 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
732 					   const struct mptcp_rm_list *rm_list,
733 					   enum linux_mptcp_mib_field rm_type)
734 {
735 	struct mptcp_subflow_context *subflow, *tmp;
736 	struct sock *sk = (struct sock *)msk;
737 	u8 i;
738 
739 	pr_debug("%s rm_list_nr %d",
740 		 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
741 
742 	msk_owned_by_me(msk);
743 
744 	if (sk->sk_state == TCP_LISTEN)
745 		return;
746 
747 	if (!rm_list->nr)
748 		return;
749 
750 	if (list_empty(&msk->conn_list))
751 		return;
752 
753 	for (i = 0; i < rm_list->nr; i++) {
754 		bool removed = false;
755 
756 		list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
757 			struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
758 			int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
759 			u8 id = subflow->local_id;
760 
761 			if (rm_type == MPTCP_MIB_RMADDR)
762 				id = subflow->remote_id;
763 
764 			if (rm_list->ids[i] != id)
765 				continue;
766 
767 			pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
768 				 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
769 				 i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
770 			spin_unlock_bh(&msk->pm.lock);
771 			mptcp_subflow_shutdown(sk, ssk, how);
772 
773 			/* the following takes care of updating the subflows counter */
774 			mptcp_close_ssk(sk, ssk, subflow);
775 			spin_lock_bh(&msk->pm.lock);
776 
777 			removed = true;
778 			__MPTCP_INC_STATS(sock_net(sk), rm_type);
779 		}
780 		__set_bit(rm_list->ids[1], msk->pm.id_avail_bitmap);
781 		if (!removed)
782 			continue;
783 
784 		if (rm_type == MPTCP_MIB_RMADDR) {
785 			msk->pm.add_addr_accepted--;
786 			WRITE_ONCE(msk->pm.accept_addr, true);
787 		} else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
788 			msk->pm.local_addr_used--;
789 		}
790 	}
791 }
792 
793 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
794 {
795 	mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
796 }
797 
798 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
799 				     const struct mptcp_rm_list *rm_list)
800 {
801 	mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
802 }
803 
804 void mptcp_pm_nl_work(struct mptcp_sock *msk)
805 {
806 	struct mptcp_pm_data *pm = &msk->pm;
807 
808 	msk_owned_by_me(msk);
809 
810 	if (!(pm->status & MPTCP_PM_WORK_MASK))
811 		return;
812 
813 	spin_lock_bh(&msk->pm.lock);
814 
815 	pr_debug("msk=%p status=%x", msk, pm->status);
816 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
817 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
818 		mptcp_pm_nl_add_addr_received(msk);
819 	}
820 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
821 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
822 		mptcp_pm_nl_addr_send_ack(msk);
823 	}
824 	if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
825 		pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
826 		mptcp_pm_nl_rm_addr_received(msk);
827 	}
828 	if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
829 		pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
830 		mptcp_pm_nl_fully_established(msk);
831 	}
832 	if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
833 		pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
834 		mptcp_pm_nl_subflow_established(msk);
835 	}
836 
837 	spin_unlock_bh(&msk->pm.lock);
838 }
839 
840 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
841 {
842 	return (entry->flags &
843 		(MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
844 		MPTCP_PM_ADDR_FLAG_SIGNAL;
845 }
846 
847 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
848 					     struct mptcp_pm_addr_entry *entry)
849 {
850 	struct mptcp_pm_addr_entry *cur;
851 	unsigned int addr_max;
852 	int ret = -EINVAL;
853 
854 	spin_lock_bh(&pernet->lock);
855 	/* to keep the code simple, don't do IDR-like allocation for address ID,
856 	 * just bail when we exceed limits
857 	 */
858 	if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
859 		pernet->next_id = 1;
860 	if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
861 		goto out;
862 	if (test_bit(entry->addr.id, pernet->id_bitmap))
863 		goto out;
864 
865 	/* do not insert duplicate address, differentiate on port only
866 	 * singled addresses
867 	 */
868 	list_for_each_entry(cur, &pernet->local_addr_list, list) {
869 		if (addresses_equal(&cur->addr, &entry->addr,
870 				    address_use_port(entry) &&
871 				    address_use_port(cur)))
872 			goto out;
873 	}
874 
875 	if (!entry->addr.id) {
876 find_next:
877 		entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
878 						    MPTCP_PM_MAX_ADDR_ID + 1,
879 						    pernet->next_id);
880 		if (!entry->addr.id && pernet->next_id != 1) {
881 			pernet->next_id = 1;
882 			goto find_next;
883 		}
884 	}
885 
886 	if (!entry->addr.id)
887 		goto out;
888 
889 	__set_bit(entry->addr.id, pernet->id_bitmap);
890 	if (entry->addr.id > pernet->next_id)
891 		pernet->next_id = entry->addr.id;
892 
893 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
894 		addr_max = pernet->add_addr_signal_max;
895 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
896 	}
897 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
898 		addr_max = pernet->local_addr_max;
899 		WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
900 	}
901 
902 	pernet->addrs++;
903 	list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
904 	ret = entry->addr.id;
905 
906 out:
907 	spin_unlock_bh(&pernet->lock);
908 	return ret;
909 }
910 
911 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
912 					    struct mptcp_pm_addr_entry *entry)
913 {
914 	struct sockaddr_storage addr;
915 	struct mptcp_sock *msk;
916 	struct socket *ssock;
917 	int backlog = 1024;
918 	int err;
919 
920 	err = sock_create_kern(sock_net(sk), entry->addr.family,
921 			       SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
922 	if (err)
923 		return err;
924 
925 	msk = mptcp_sk(entry->lsk->sk);
926 	if (!msk) {
927 		err = -EINVAL;
928 		goto out;
929 	}
930 
931 	ssock = __mptcp_nmpc_socket(msk);
932 	if (!ssock) {
933 		err = -EINVAL;
934 		goto out;
935 	}
936 
937 	mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
938 	err = kernel_bind(ssock, (struct sockaddr *)&addr,
939 			  sizeof(struct sockaddr_in));
940 	if (err) {
941 		pr_warn("kernel_bind error, err=%d", err);
942 		goto out;
943 	}
944 
945 	err = kernel_listen(ssock, backlog);
946 	if (err) {
947 		pr_warn("kernel_listen error, err=%d", err);
948 		goto out;
949 	}
950 
951 	return 0;
952 
953 out:
954 	sock_release(entry->lsk);
955 	return err;
956 }
957 
958 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
959 {
960 	struct mptcp_pm_addr_entry *entry;
961 	struct mptcp_addr_info skc_local;
962 	struct mptcp_addr_info msk_local;
963 	struct pm_nl_pernet *pernet;
964 	int ret = -1;
965 
966 	if (WARN_ON_ONCE(!msk))
967 		return -1;
968 
969 	/* The 0 ID mapping is defined by the first subflow, copied into the msk
970 	 * addr
971 	 */
972 	local_address((struct sock_common *)msk, &msk_local);
973 	local_address((struct sock_common *)skc, &skc_local);
974 	if (addresses_equal(&msk_local, &skc_local, false))
975 		return 0;
976 
977 	if (address_zero(&skc_local))
978 		return 0;
979 
980 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
981 
982 	rcu_read_lock();
983 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
984 		if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
985 			ret = entry->addr.id;
986 			break;
987 		}
988 	}
989 	rcu_read_unlock();
990 	if (ret >= 0)
991 		return ret;
992 
993 	/* address not found, add to local list */
994 	entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
995 	if (!entry)
996 		return -ENOMEM;
997 
998 	entry->addr = skc_local;
999 	entry->addr.id = 0;
1000 	entry->addr.port = 0;
1001 	entry->ifindex = 0;
1002 	entry->flags = 0;
1003 	entry->lsk = NULL;
1004 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1005 	if (ret < 0)
1006 		kfree(entry);
1007 
1008 	return ret;
1009 }
1010 
1011 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
1012 {
1013 	struct mptcp_pm_data *pm = &msk->pm;
1014 	bool subflows;
1015 
1016 	subflows = !!mptcp_pm_get_subflows_max(msk);
1017 	WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
1018 		   !!mptcp_pm_get_add_addr_signal_max(msk));
1019 	WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
1020 	WRITE_ONCE(pm->accept_subflow, subflows);
1021 }
1022 
1023 #define MPTCP_PM_CMD_GRP_OFFSET       0
1024 #define MPTCP_PM_EV_GRP_OFFSET        1
1025 
1026 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1027 	[MPTCP_PM_CMD_GRP_OFFSET]	= { .name = MPTCP_PM_CMD_GRP_NAME, },
1028 	[MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1029 					    .flags = GENL_UNS_ADMIN_PERM,
1030 					  },
1031 };
1032 
1033 static const struct nla_policy
1034 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1035 	[MPTCP_PM_ADDR_ATTR_FAMILY]	= { .type	= NLA_U16,	},
1036 	[MPTCP_PM_ADDR_ATTR_ID]		= { .type	= NLA_U8,	},
1037 	[MPTCP_PM_ADDR_ATTR_ADDR4]	= { .type	= NLA_U32,	},
1038 	[MPTCP_PM_ADDR_ATTR_ADDR6]	=
1039 		NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1040 	[MPTCP_PM_ADDR_ATTR_PORT]	= { .type	= NLA_U16	},
1041 	[MPTCP_PM_ADDR_ATTR_FLAGS]	= { .type	= NLA_U32	},
1042 	[MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type	= NLA_S32	},
1043 };
1044 
1045 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1046 	[MPTCP_PM_ATTR_ADDR]		=
1047 					NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1048 	[MPTCP_PM_ATTR_RCV_ADD_ADDRS]	= { .type	= NLA_U32,	},
1049 	[MPTCP_PM_ATTR_SUBFLOWS]	= { .type	= NLA_U32,	},
1050 };
1051 
1052 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1053 {
1054 	struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1055 	struct sock *sk = (struct sock *)msk;
1056 	unsigned int active_max_loss_cnt;
1057 	struct net *net = sock_net(sk);
1058 	unsigned int stale_loss_cnt;
1059 	bool slow;
1060 
1061 	stale_loss_cnt = mptcp_stale_loss_cnt(net);
1062 	if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1063 		return;
1064 
1065 	/* look for another available subflow not in loss state */
1066 	active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1067 	mptcp_for_each_subflow(msk, iter) {
1068 		if (iter != subflow && mptcp_subflow_active(iter) &&
1069 		    iter->stale_count < active_max_loss_cnt) {
1070 			/* we have some alternatives, try to mark this subflow as idle ...*/
1071 			slow = lock_sock_fast(ssk);
1072 			if (!tcp_rtx_and_write_queues_empty(ssk)) {
1073 				subflow->stale = 1;
1074 				__mptcp_retransmit_pending_data(sk);
1075 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1076 			}
1077 			unlock_sock_fast(ssk, slow);
1078 
1079 			/* always try to push the pending data regarless of re-injections:
1080 			 * we can possibly use backup subflows now, and subflow selection
1081 			 * is cheap under the msk socket lock
1082 			 */
1083 			__mptcp_push_pending(sk, 0);
1084 			return;
1085 		}
1086 	}
1087 }
1088 
1089 static int mptcp_pm_family_to_addr(int family)
1090 {
1091 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1092 	if (family == AF_INET6)
1093 		return MPTCP_PM_ADDR_ATTR_ADDR6;
1094 #endif
1095 	return MPTCP_PM_ADDR_ATTR_ADDR4;
1096 }
1097 
1098 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1099 			       bool require_family,
1100 			       struct mptcp_pm_addr_entry *entry)
1101 {
1102 	struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1103 	int err, addr_addr;
1104 
1105 	if (!attr) {
1106 		GENL_SET_ERR_MSG(info, "missing address info");
1107 		return -EINVAL;
1108 	}
1109 
1110 	/* no validation needed - was already done via nested policy */
1111 	err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1112 					  mptcp_pm_addr_policy, info->extack);
1113 	if (err)
1114 		return err;
1115 
1116 	memset(entry, 0, sizeof(*entry));
1117 	if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1118 		if (!require_family)
1119 			goto skip_family;
1120 
1121 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1122 				    "missing family");
1123 		return -EINVAL;
1124 	}
1125 
1126 	entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1127 	if (entry->addr.family != AF_INET
1128 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1129 	    && entry->addr.family != AF_INET6
1130 #endif
1131 	    ) {
1132 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1133 				    "unknown address family");
1134 		return -EINVAL;
1135 	}
1136 	addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1137 	if (!tb[addr_addr]) {
1138 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1139 				    "missing address data");
1140 		return -EINVAL;
1141 	}
1142 
1143 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1144 	if (entry->addr.family == AF_INET6)
1145 		entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1146 	else
1147 #endif
1148 		entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1149 
1150 skip_family:
1151 	if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1152 		u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1153 
1154 		entry->ifindex = val;
1155 	}
1156 
1157 	if (tb[MPTCP_PM_ADDR_ATTR_ID])
1158 		entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1159 
1160 	if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1161 		entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1162 
1163 	if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1164 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1165 			NL_SET_ERR_MSG_ATTR(info->extack, attr,
1166 					    "flags must have signal when using port");
1167 			return -EINVAL;
1168 		}
1169 		entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1170 	}
1171 
1172 	return 0;
1173 }
1174 
1175 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1176 {
1177 	return net_generic(genl_info_net(info), pm_nl_pernet_id);
1178 }
1179 
1180 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1181 {
1182 	struct mptcp_sock *msk;
1183 	long s_slot = 0, s_num = 0;
1184 
1185 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1186 		struct sock *sk = (struct sock *)msk;
1187 
1188 		if (!READ_ONCE(msk->fully_established))
1189 			goto next;
1190 
1191 		lock_sock(sk);
1192 		spin_lock_bh(&msk->pm.lock);
1193 		mptcp_pm_create_subflow_or_signal_addr(msk);
1194 		spin_unlock_bh(&msk->pm.lock);
1195 		release_sock(sk);
1196 
1197 next:
1198 		sock_put(sk);
1199 		cond_resched();
1200 	}
1201 
1202 	return 0;
1203 }
1204 
1205 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1206 {
1207 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1208 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1209 	struct mptcp_pm_addr_entry addr, *entry;
1210 	int ret;
1211 
1212 	ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1213 	if (ret < 0)
1214 		return ret;
1215 
1216 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1217 	if (!entry) {
1218 		GENL_SET_ERR_MSG(info, "can't allocate addr");
1219 		return -ENOMEM;
1220 	}
1221 
1222 	*entry = addr;
1223 	if (entry->addr.port) {
1224 		ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1225 		if (ret) {
1226 			GENL_SET_ERR_MSG(info, "create listen socket error");
1227 			kfree(entry);
1228 			return ret;
1229 		}
1230 	}
1231 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1232 	if (ret < 0) {
1233 		GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1234 		if (entry->lsk)
1235 			sock_release(entry->lsk);
1236 		kfree(entry);
1237 		return ret;
1238 	}
1239 
1240 	mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1241 
1242 	return 0;
1243 }
1244 
1245 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1246 					 u8 *flags, int *ifindex)
1247 {
1248 	struct mptcp_pm_addr_entry *entry;
1249 
1250 	*flags = 0;
1251 	*ifindex = 0;
1252 
1253 	if (id) {
1254 		rcu_read_lock();
1255 		entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1256 		if (entry) {
1257 			*flags = entry->flags;
1258 			*ifindex = entry->ifindex;
1259 		}
1260 		rcu_read_unlock();
1261 	}
1262 
1263 	return 0;
1264 }
1265 
1266 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1267 				      struct mptcp_addr_info *addr)
1268 {
1269 	struct mptcp_pm_add_entry *entry;
1270 
1271 	entry = mptcp_pm_del_add_timer(msk, addr, false);
1272 	if (entry) {
1273 		list_del(&entry->list);
1274 		kfree(entry);
1275 		return true;
1276 	}
1277 
1278 	return false;
1279 }
1280 
1281 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1282 				      struct mptcp_addr_info *addr,
1283 				      bool force)
1284 {
1285 	struct mptcp_rm_list list = { .nr = 0 };
1286 	bool ret;
1287 
1288 	list.ids[list.nr++] = addr->id;
1289 
1290 	ret = remove_anno_list_by_saddr(msk, addr);
1291 	if (ret || force) {
1292 		spin_lock_bh(&msk->pm.lock);
1293 		mptcp_pm_remove_addr(msk, &list);
1294 		spin_unlock_bh(&msk->pm.lock);
1295 	}
1296 	return ret;
1297 }
1298 
1299 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1300 						   struct mptcp_addr_info *addr)
1301 {
1302 	struct mptcp_sock *msk;
1303 	long s_slot = 0, s_num = 0;
1304 	struct mptcp_rm_list list = { .nr = 0 };
1305 
1306 	pr_debug("remove_id=%d", addr->id);
1307 
1308 	list.ids[list.nr++] = addr->id;
1309 
1310 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1311 		struct sock *sk = (struct sock *)msk;
1312 		bool remove_subflow;
1313 
1314 		if (list_empty(&msk->conn_list)) {
1315 			mptcp_pm_remove_anno_addr(msk, addr, false);
1316 			goto next;
1317 		}
1318 
1319 		lock_sock(sk);
1320 		remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1321 		mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1322 		if (remove_subflow)
1323 			mptcp_pm_remove_subflow(msk, &list);
1324 		release_sock(sk);
1325 
1326 next:
1327 		sock_put(sk);
1328 		cond_resched();
1329 	}
1330 
1331 	return 0;
1332 }
1333 
1334 /* caller must ensure the RCU grace period is already elapsed */
1335 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1336 {
1337 	if (entry->lsk)
1338 		sock_release(entry->lsk);
1339 	kfree(entry);
1340 }
1341 
1342 static int mptcp_nl_remove_id_zero_address(struct net *net,
1343 					   struct mptcp_addr_info *addr)
1344 {
1345 	struct mptcp_rm_list list = { .nr = 0 };
1346 	long s_slot = 0, s_num = 0;
1347 	struct mptcp_sock *msk;
1348 
1349 	list.ids[list.nr++] = 0;
1350 
1351 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1352 		struct sock *sk = (struct sock *)msk;
1353 		struct mptcp_addr_info msk_local;
1354 
1355 		if (list_empty(&msk->conn_list))
1356 			goto next;
1357 
1358 		local_address((struct sock_common *)msk, &msk_local);
1359 		if (!addresses_equal(&msk_local, addr, addr->port))
1360 			goto next;
1361 
1362 		lock_sock(sk);
1363 		spin_lock_bh(&msk->pm.lock);
1364 		mptcp_pm_remove_addr(msk, &list);
1365 		mptcp_pm_nl_rm_subflow_received(msk, &list);
1366 		spin_unlock_bh(&msk->pm.lock);
1367 		release_sock(sk);
1368 
1369 next:
1370 		sock_put(sk);
1371 		cond_resched();
1372 	}
1373 
1374 	return 0;
1375 }
1376 
1377 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1378 {
1379 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1380 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1381 	struct mptcp_pm_addr_entry addr, *entry;
1382 	unsigned int addr_max;
1383 	int ret;
1384 
1385 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1386 	if (ret < 0)
1387 		return ret;
1388 
1389 	/* the zero id address is special: the first address used by the msk
1390 	 * always gets such an id, so different subflows can have different zero
1391 	 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1392 	 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1393 	 */
1394 	if (addr.addr.id == 0)
1395 		return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1396 
1397 	spin_lock_bh(&pernet->lock);
1398 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1399 	if (!entry) {
1400 		GENL_SET_ERR_MSG(info, "address not found");
1401 		spin_unlock_bh(&pernet->lock);
1402 		return -EINVAL;
1403 	}
1404 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1405 		addr_max = pernet->add_addr_signal_max;
1406 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1407 	}
1408 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1409 		addr_max = pernet->local_addr_max;
1410 		WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1411 	}
1412 
1413 	pernet->addrs--;
1414 	list_del_rcu(&entry->list);
1415 	__clear_bit(entry->addr.id, pernet->id_bitmap);
1416 	spin_unlock_bh(&pernet->lock);
1417 
1418 	mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1419 	synchronize_rcu();
1420 	__mptcp_pm_release_addr_entry(entry);
1421 
1422 	return ret;
1423 }
1424 
1425 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1426 					       struct list_head *rm_list)
1427 {
1428 	struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1429 	struct mptcp_pm_addr_entry *entry;
1430 
1431 	list_for_each_entry(entry, rm_list, list) {
1432 		if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1433 		    alist.nr < MPTCP_RM_IDS_MAX &&
1434 		    slist.nr < MPTCP_RM_IDS_MAX) {
1435 			alist.ids[alist.nr++] = entry->addr.id;
1436 			slist.ids[slist.nr++] = entry->addr.id;
1437 		} else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1438 			 alist.nr < MPTCP_RM_IDS_MAX) {
1439 			alist.ids[alist.nr++] = entry->addr.id;
1440 		}
1441 	}
1442 
1443 	if (alist.nr) {
1444 		spin_lock_bh(&msk->pm.lock);
1445 		mptcp_pm_remove_addr(msk, &alist);
1446 		spin_unlock_bh(&msk->pm.lock);
1447 	}
1448 	if (slist.nr)
1449 		mptcp_pm_remove_subflow(msk, &slist);
1450 }
1451 
1452 static void mptcp_nl_remove_addrs_list(struct net *net,
1453 				       struct list_head *rm_list)
1454 {
1455 	long s_slot = 0, s_num = 0;
1456 	struct mptcp_sock *msk;
1457 
1458 	if (list_empty(rm_list))
1459 		return;
1460 
1461 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1462 		struct sock *sk = (struct sock *)msk;
1463 
1464 		lock_sock(sk);
1465 		mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1466 		release_sock(sk);
1467 
1468 		sock_put(sk);
1469 		cond_resched();
1470 	}
1471 }
1472 
1473 /* caller must ensure the RCU grace period is already elapsed */
1474 static void __flush_addrs(struct list_head *list)
1475 {
1476 	while (!list_empty(list)) {
1477 		struct mptcp_pm_addr_entry *cur;
1478 
1479 		cur = list_entry(list->next,
1480 				 struct mptcp_pm_addr_entry, list);
1481 		list_del_rcu(&cur->list);
1482 		__mptcp_pm_release_addr_entry(cur);
1483 	}
1484 }
1485 
1486 static void __reset_counters(struct pm_nl_pernet *pernet)
1487 {
1488 	WRITE_ONCE(pernet->add_addr_signal_max, 0);
1489 	WRITE_ONCE(pernet->add_addr_accept_max, 0);
1490 	WRITE_ONCE(pernet->local_addr_max, 0);
1491 	pernet->addrs = 0;
1492 }
1493 
1494 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1495 {
1496 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1497 	LIST_HEAD(free_list);
1498 
1499 	spin_lock_bh(&pernet->lock);
1500 	list_splice_init(&pernet->local_addr_list, &free_list);
1501 	__reset_counters(pernet);
1502 	pernet->next_id = 1;
1503 	bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1504 	spin_unlock_bh(&pernet->lock);
1505 	mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1506 	synchronize_rcu();
1507 	__flush_addrs(&free_list);
1508 	return 0;
1509 }
1510 
1511 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1512 			      struct mptcp_pm_addr_entry *entry)
1513 {
1514 	struct mptcp_addr_info *addr = &entry->addr;
1515 	struct nlattr *attr;
1516 
1517 	attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1518 	if (!attr)
1519 		return -EMSGSIZE;
1520 
1521 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1522 		goto nla_put_failure;
1523 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1524 		goto nla_put_failure;
1525 	if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1526 		goto nla_put_failure;
1527 	if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1528 		goto nla_put_failure;
1529 	if (entry->ifindex &&
1530 	    nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1531 		goto nla_put_failure;
1532 
1533 	if (addr->family == AF_INET &&
1534 	    nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1535 			    addr->addr.s_addr))
1536 		goto nla_put_failure;
1537 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1538 	else if (addr->family == AF_INET6 &&
1539 		 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1540 		goto nla_put_failure;
1541 #endif
1542 	nla_nest_end(skb, attr);
1543 	return 0;
1544 
1545 nla_put_failure:
1546 	nla_nest_cancel(skb, attr);
1547 	return -EMSGSIZE;
1548 }
1549 
1550 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1551 {
1552 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1553 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1554 	struct mptcp_pm_addr_entry addr, *entry;
1555 	struct sk_buff *msg;
1556 	void *reply;
1557 	int ret;
1558 
1559 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1560 	if (ret < 0)
1561 		return ret;
1562 
1563 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1564 	if (!msg)
1565 		return -ENOMEM;
1566 
1567 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1568 				  info->genlhdr->cmd);
1569 	if (!reply) {
1570 		GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1571 		ret = -EMSGSIZE;
1572 		goto fail;
1573 	}
1574 
1575 	spin_lock_bh(&pernet->lock);
1576 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1577 	if (!entry) {
1578 		GENL_SET_ERR_MSG(info, "address not found");
1579 		ret = -EINVAL;
1580 		goto unlock_fail;
1581 	}
1582 
1583 	ret = mptcp_nl_fill_addr(msg, entry);
1584 	if (ret)
1585 		goto unlock_fail;
1586 
1587 	genlmsg_end(msg, reply);
1588 	ret = genlmsg_reply(msg, info);
1589 	spin_unlock_bh(&pernet->lock);
1590 	return ret;
1591 
1592 unlock_fail:
1593 	spin_unlock_bh(&pernet->lock);
1594 
1595 fail:
1596 	nlmsg_free(msg);
1597 	return ret;
1598 }
1599 
1600 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1601 				   struct netlink_callback *cb)
1602 {
1603 	struct net *net = sock_net(msg->sk);
1604 	struct mptcp_pm_addr_entry *entry;
1605 	struct pm_nl_pernet *pernet;
1606 	int id = cb->args[0];
1607 	void *hdr;
1608 	int i;
1609 
1610 	pernet = net_generic(net, pm_nl_pernet_id);
1611 
1612 	spin_lock_bh(&pernet->lock);
1613 	for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1614 		if (test_bit(i, pernet->id_bitmap)) {
1615 			entry = __lookup_addr_by_id(pernet, i);
1616 			if (!entry)
1617 				break;
1618 
1619 			if (entry->addr.id <= id)
1620 				continue;
1621 
1622 			hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1623 					  cb->nlh->nlmsg_seq, &mptcp_genl_family,
1624 					  NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1625 			if (!hdr)
1626 				break;
1627 
1628 			if (mptcp_nl_fill_addr(msg, entry) < 0) {
1629 				genlmsg_cancel(msg, hdr);
1630 				break;
1631 			}
1632 
1633 			id = entry->addr.id;
1634 			genlmsg_end(msg, hdr);
1635 		}
1636 	}
1637 	spin_unlock_bh(&pernet->lock);
1638 
1639 	cb->args[0] = id;
1640 	return msg->len;
1641 }
1642 
1643 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1644 {
1645 	struct nlattr *attr = info->attrs[id];
1646 
1647 	if (!attr)
1648 		return 0;
1649 
1650 	*limit = nla_get_u32(attr);
1651 	if (*limit > MPTCP_PM_ADDR_MAX) {
1652 		GENL_SET_ERR_MSG(info, "limit greater than maximum");
1653 		return -EINVAL;
1654 	}
1655 	return 0;
1656 }
1657 
1658 static int
1659 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1660 {
1661 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1662 	unsigned int rcv_addrs, subflows;
1663 	int ret;
1664 
1665 	spin_lock_bh(&pernet->lock);
1666 	rcv_addrs = pernet->add_addr_accept_max;
1667 	ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1668 	if (ret)
1669 		goto unlock;
1670 
1671 	subflows = pernet->subflows_max;
1672 	ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1673 	if (ret)
1674 		goto unlock;
1675 
1676 	WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1677 	WRITE_ONCE(pernet->subflows_max, subflows);
1678 
1679 unlock:
1680 	spin_unlock_bh(&pernet->lock);
1681 	return ret;
1682 }
1683 
1684 static int
1685 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1686 {
1687 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1688 	struct sk_buff *msg;
1689 	void *reply;
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 				  MPTCP_PM_CMD_GET_LIMITS);
1697 	if (!reply)
1698 		goto fail;
1699 
1700 	if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1701 			READ_ONCE(pernet->add_addr_accept_max)))
1702 		goto fail;
1703 
1704 	if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1705 			READ_ONCE(pernet->subflows_max)))
1706 		goto fail;
1707 
1708 	genlmsg_end(msg, reply);
1709 	return genlmsg_reply(msg, info);
1710 
1711 fail:
1712 	GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1713 	nlmsg_free(msg);
1714 	return -EMSGSIZE;
1715 }
1716 
1717 static int mptcp_nl_addr_backup(struct net *net,
1718 				struct mptcp_addr_info *addr,
1719 				u8 bkup)
1720 {
1721 	long s_slot = 0, s_num = 0;
1722 	struct mptcp_sock *msk;
1723 	int ret = -EINVAL;
1724 
1725 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1726 		struct sock *sk = (struct sock *)msk;
1727 
1728 		if (list_empty(&msk->conn_list))
1729 			goto next;
1730 
1731 		lock_sock(sk);
1732 		spin_lock_bh(&msk->pm.lock);
1733 		ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1734 		spin_unlock_bh(&msk->pm.lock);
1735 		release_sock(sk);
1736 
1737 next:
1738 		sock_put(sk);
1739 		cond_resched();
1740 	}
1741 
1742 	return ret;
1743 }
1744 
1745 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1746 {
1747 	struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1748 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1749 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1750 	struct net *net = sock_net(skb->sk);
1751 	u8 bkup = 0, lookup_by_id = 0;
1752 	int ret;
1753 
1754 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1755 	if (ret < 0)
1756 		return ret;
1757 
1758 	if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1759 		bkup = 1;
1760 	if (addr.addr.family == AF_UNSPEC) {
1761 		lookup_by_id = 1;
1762 		if (!addr.addr.id)
1763 			return -EOPNOTSUPP;
1764 	}
1765 
1766 	list_for_each_entry(entry, &pernet->local_addr_list, list) {
1767 		if ((!lookup_by_id && addresses_equal(&entry->addr, &addr.addr, true)) ||
1768 		    (lookup_by_id && entry->addr.id == addr.addr.id)) {
1769 			mptcp_nl_addr_backup(net, &entry->addr, bkup);
1770 
1771 			if (bkup)
1772 				entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1773 			else
1774 				entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1775 		}
1776 	}
1777 
1778 	return 0;
1779 }
1780 
1781 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1782 {
1783 	genlmsg_multicast_netns(&mptcp_genl_family, net,
1784 				nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1785 }
1786 
1787 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1788 {
1789 	const struct inet_sock *issk = inet_sk(ssk);
1790 	const struct mptcp_subflow_context *sf;
1791 
1792 	if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1793 		return -EMSGSIZE;
1794 
1795 	switch (ssk->sk_family) {
1796 	case AF_INET:
1797 		if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1798 			return -EMSGSIZE;
1799 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1800 			return -EMSGSIZE;
1801 		break;
1802 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1803 	case AF_INET6: {
1804 		const struct ipv6_pinfo *np = inet6_sk(ssk);
1805 
1806 		if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1807 			return -EMSGSIZE;
1808 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1809 			return -EMSGSIZE;
1810 		break;
1811 	}
1812 #endif
1813 	default:
1814 		WARN_ON_ONCE(1);
1815 		return -EMSGSIZE;
1816 	}
1817 
1818 	if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1819 		return -EMSGSIZE;
1820 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1821 		return -EMSGSIZE;
1822 
1823 	sf = mptcp_subflow_ctx(ssk);
1824 	if (WARN_ON_ONCE(!sf))
1825 		return -EINVAL;
1826 
1827 	if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1828 		return -EMSGSIZE;
1829 
1830 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1831 		return -EMSGSIZE;
1832 
1833 	return 0;
1834 }
1835 
1836 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1837 					 const struct mptcp_sock *msk,
1838 					 const struct sock *ssk)
1839 {
1840 	const struct sock *sk = (const struct sock *)msk;
1841 	const struct mptcp_subflow_context *sf;
1842 	u8 sk_err;
1843 
1844 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1845 		return -EMSGSIZE;
1846 
1847 	if (mptcp_event_add_subflow(skb, ssk))
1848 		return -EMSGSIZE;
1849 
1850 	sf = mptcp_subflow_ctx(ssk);
1851 	if (WARN_ON_ONCE(!sf))
1852 		return -EINVAL;
1853 
1854 	if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1855 		return -EMSGSIZE;
1856 
1857 	if (ssk->sk_bound_dev_if &&
1858 	    nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1859 		return -EMSGSIZE;
1860 
1861 	sk_err = ssk->sk_err;
1862 	if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1863 	    nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1864 		return -EMSGSIZE;
1865 
1866 	return 0;
1867 }
1868 
1869 static int mptcp_event_sub_established(struct sk_buff *skb,
1870 				       const struct mptcp_sock *msk,
1871 				       const struct sock *ssk)
1872 {
1873 	return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1874 }
1875 
1876 static int mptcp_event_sub_closed(struct sk_buff *skb,
1877 				  const struct mptcp_sock *msk,
1878 				  const struct sock *ssk)
1879 {
1880 	const struct mptcp_subflow_context *sf;
1881 
1882 	if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1883 		return -EMSGSIZE;
1884 
1885 	sf = mptcp_subflow_ctx(ssk);
1886 	if (!sf->reset_seen)
1887 		return 0;
1888 
1889 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1890 		return -EMSGSIZE;
1891 
1892 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1893 		return -EMSGSIZE;
1894 
1895 	return 0;
1896 }
1897 
1898 static int mptcp_event_created(struct sk_buff *skb,
1899 			       const struct mptcp_sock *msk,
1900 			       const struct sock *ssk)
1901 {
1902 	int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1903 
1904 	if (err)
1905 		return err;
1906 
1907 	return mptcp_event_add_subflow(skb, ssk);
1908 }
1909 
1910 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1911 {
1912 	struct net *net = sock_net((const struct sock *)msk);
1913 	struct nlmsghdr *nlh;
1914 	struct sk_buff *skb;
1915 
1916 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1917 		return;
1918 
1919 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1920 	if (!skb)
1921 		return;
1922 
1923 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1924 	if (!nlh)
1925 		goto nla_put_failure;
1926 
1927 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1928 		goto nla_put_failure;
1929 
1930 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1931 		goto nla_put_failure;
1932 
1933 	genlmsg_end(skb, nlh);
1934 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1935 	return;
1936 
1937 nla_put_failure:
1938 	kfree_skb(skb);
1939 }
1940 
1941 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1942 				const struct mptcp_addr_info *info)
1943 {
1944 	struct net *net = sock_net((const struct sock *)msk);
1945 	struct nlmsghdr *nlh;
1946 	struct sk_buff *skb;
1947 
1948 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1949 		return;
1950 
1951 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1952 	if (!skb)
1953 		return;
1954 
1955 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1956 			  MPTCP_EVENT_ANNOUNCED);
1957 	if (!nlh)
1958 		goto nla_put_failure;
1959 
1960 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1961 		goto nla_put_failure;
1962 
1963 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
1964 		goto nla_put_failure;
1965 
1966 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
1967 		goto nla_put_failure;
1968 
1969 	switch (info->family) {
1970 	case AF_INET:
1971 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
1972 			goto nla_put_failure;
1973 		break;
1974 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1975 	case AF_INET6:
1976 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
1977 			goto nla_put_failure;
1978 		break;
1979 #endif
1980 	default:
1981 		WARN_ON_ONCE(1);
1982 		goto nla_put_failure;
1983 	}
1984 
1985 	genlmsg_end(skb, nlh);
1986 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1987 	return;
1988 
1989 nla_put_failure:
1990 	kfree_skb(skb);
1991 }
1992 
1993 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
1994 		 const struct sock *ssk, gfp_t gfp)
1995 {
1996 	struct net *net = sock_net((const struct sock *)msk);
1997 	struct nlmsghdr *nlh;
1998 	struct sk_buff *skb;
1999 
2000 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2001 		return;
2002 
2003 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2004 	if (!skb)
2005 		return;
2006 
2007 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2008 	if (!nlh)
2009 		goto nla_put_failure;
2010 
2011 	switch (type) {
2012 	case MPTCP_EVENT_UNSPEC:
2013 		WARN_ON_ONCE(1);
2014 		break;
2015 	case MPTCP_EVENT_CREATED:
2016 	case MPTCP_EVENT_ESTABLISHED:
2017 		if (mptcp_event_created(skb, msk, ssk) < 0)
2018 			goto nla_put_failure;
2019 		break;
2020 	case MPTCP_EVENT_CLOSED:
2021 		if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2022 			goto nla_put_failure;
2023 		break;
2024 	case MPTCP_EVENT_ANNOUNCED:
2025 	case MPTCP_EVENT_REMOVED:
2026 		/* call mptcp_event_addr_announced()/removed instead */
2027 		WARN_ON_ONCE(1);
2028 		break;
2029 	case MPTCP_EVENT_SUB_ESTABLISHED:
2030 	case MPTCP_EVENT_SUB_PRIORITY:
2031 		if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2032 			goto nla_put_failure;
2033 		break;
2034 	case MPTCP_EVENT_SUB_CLOSED:
2035 		if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2036 			goto nla_put_failure;
2037 		break;
2038 	}
2039 
2040 	genlmsg_end(skb, nlh);
2041 	mptcp_nl_mcast_send(net, skb, gfp);
2042 	return;
2043 
2044 nla_put_failure:
2045 	kfree_skb(skb);
2046 }
2047 
2048 static const struct genl_small_ops mptcp_pm_ops[] = {
2049 	{
2050 		.cmd    = MPTCP_PM_CMD_ADD_ADDR,
2051 		.doit   = mptcp_nl_cmd_add_addr,
2052 		.flags  = GENL_ADMIN_PERM,
2053 	},
2054 	{
2055 		.cmd    = MPTCP_PM_CMD_DEL_ADDR,
2056 		.doit   = mptcp_nl_cmd_del_addr,
2057 		.flags  = GENL_ADMIN_PERM,
2058 	},
2059 	{
2060 		.cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2061 		.doit   = mptcp_nl_cmd_flush_addrs,
2062 		.flags  = GENL_ADMIN_PERM,
2063 	},
2064 	{
2065 		.cmd    = MPTCP_PM_CMD_GET_ADDR,
2066 		.doit   = mptcp_nl_cmd_get_addr,
2067 		.dumpit   = mptcp_nl_cmd_dump_addrs,
2068 	},
2069 	{
2070 		.cmd    = MPTCP_PM_CMD_SET_LIMITS,
2071 		.doit   = mptcp_nl_cmd_set_limits,
2072 		.flags  = GENL_ADMIN_PERM,
2073 	},
2074 	{
2075 		.cmd    = MPTCP_PM_CMD_GET_LIMITS,
2076 		.doit   = mptcp_nl_cmd_get_limits,
2077 	},
2078 	{
2079 		.cmd    = MPTCP_PM_CMD_SET_FLAGS,
2080 		.doit   = mptcp_nl_cmd_set_flags,
2081 		.flags  = GENL_ADMIN_PERM,
2082 	},
2083 };
2084 
2085 static struct genl_family mptcp_genl_family __ro_after_init = {
2086 	.name		= MPTCP_PM_NAME,
2087 	.version	= MPTCP_PM_VER,
2088 	.maxattr	= MPTCP_PM_ATTR_MAX,
2089 	.policy		= mptcp_pm_policy,
2090 	.netnsok	= true,
2091 	.module		= THIS_MODULE,
2092 	.small_ops	= mptcp_pm_ops,
2093 	.n_small_ops	= ARRAY_SIZE(mptcp_pm_ops),
2094 	.mcgrps		= mptcp_pm_mcgrps,
2095 	.n_mcgrps	= ARRAY_SIZE(mptcp_pm_mcgrps),
2096 };
2097 
2098 static int __net_init pm_nl_init_net(struct net *net)
2099 {
2100 	struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2101 
2102 	INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2103 
2104 	/* Cit. 2 subflows ought to be enough for anybody. */
2105 	pernet->subflows_max = 2;
2106 	pernet->next_id = 1;
2107 	pernet->stale_loss_cnt = 4;
2108 	spin_lock_init(&pernet->lock);
2109 
2110 	/* No need to initialize other pernet fields, the struct is zeroed at
2111 	 * allocation time.
2112 	 */
2113 
2114 	return 0;
2115 }
2116 
2117 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2118 {
2119 	struct net *net;
2120 
2121 	list_for_each_entry(net, net_list, exit_list) {
2122 		struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2123 
2124 		/* net is removed from namespace list, can't race with
2125 		 * other modifiers, also netns core already waited for a
2126 		 * RCU grace period.
2127 		 */
2128 		__flush_addrs(&pernet->local_addr_list);
2129 	}
2130 }
2131 
2132 static struct pernet_operations mptcp_pm_pernet_ops = {
2133 	.init = pm_nl_init_net,
2134 	.exit_batch = pm_nl_exit_net,
2135 	.id = &pm_nl_pernet_id,
2136 	.size = sizeof(struct pm_nl_pernet),
2137 };
2138 
2139 void __init mptcp_pm_nl_init(void)
2140 {
2141 	if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2142 		panic("Failed to register MPTCP PM pernet subsystem.\n");
2143 
2144 	if (genl_register_family(&mptcp_genl_family))
2145 		panic("Failed to register MPTCP PM netlink family\n");
2146 }
2147