xref: /openbmc/linux/net/mptcp/pm_netlink.c (revision dfc66bef)
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 #define MAX_ADDR_ID		255
42 #define BITMAP_SZ DIV_ROUND_UP(MAX_ADDR_ID + 1, BITS_PER_LONG)
43 
44 struct pm_nl_pernet {
45 	/* protects pernet updates */
46 	spinlock_t		lock;
47 	struct list_head	local_addr_list;
48 	unsigned int		addrs;
49 	unsigned int		stale_loss_cnt;
50 	unsigned int		add_addr_signal_max;
51 	unsigned int		add_addr_accept_max;
52 	unsigned int		local_addr_max;
53 	unsigned int		subflows_max;
54 	unsigned int		next_id;
55 	unsigned long		id_bitmap[BITMAP_SZ];
56 };
57 
58 #define MPTCP_PM_ADDR_MAX	8
59 #define ADD_ADDR_RETRANS_MAX	3
60 
61 static bool addresses_equal(const struct mptcp_addr_info *a,
62 			    struct mptcp_addr_info *b, bool use_port)
63 {
64 	bool addr_equals = false;
65 
66 	if (a->family == b->family) {
67 		if (a->family == AF_INET)
68 			addr_equals = a->addr.s_addr == b->addr.s_addr;
69 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
70 		else
71 			addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
72 	} else if (a->family == AF_INET) {
73 		if (ipv6_addr_v4mapped(&b->addr6))
74 			addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
75 	} else if (b->family == AF_INET) {
76 		if (ipv6_addr_v4mapped(&a->addr6))
77 			addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
78 #endif
79 	}
80 
81 	if (!addr_equals)
82 		return false;
83 	if (!use_port)
84 		return true;
85 
86 	return a->port == b->port;
87 }
88 
89 static bool address_zero(const struct mptcp_addr_info *addr)
90 {
91 	struct mptcp_addr_info zero;
92 
93 	memset(&zero, 0, sizeof(zero));
94 	zero.family = addr->family;
95 
96 	return addresses_equal(addr, &zero, true);
97 }
98 
99 static void local_address(const struct sock_common *skc,
100 			  struct mptcp_addr_info *addr)
101 {
102 	addr->family = skc->skc_family;
103 	addr->port = htons(skc->skc_num);
104 	if (addr->family == AF_INET)
105 		addr->addr.s_addr = skc->skc_rcv_saddr;
106 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
107 	else if (addr->family == AF_INET6)
108 		addr->addr6 = skc->skc_v6_rcv_saddr;
109 #endif
110 }
111 
112 static void remote_address(const struct sock_common *skc,
113 			   struct mptcp_addr_info *addr)
114 {
115 	addr->family = skc->skc_family;
116 	addr->port = skc->skc_dport;
117 	if (addr->family == AF_INET)
118 		addr->addr.s_addr = skc->skc_daddr;
119 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
120 	else if (addr->family == AF_INET6)
121 		addr->addr6 = skc->skc_v6_daddr;
122 #endif
123 }
124 
125 static bool lookup_subflow_by_saddr(const struct list_head *list,
126 				    struct mptcp_addr_info *saddr)
127 {
128 	struct mptcp_subflow_context *subflow;
129 	struct mptcp_addr_info cur;
130 	struct sock_common *skc;
131 
132 	list_for_each_entry(subflow, list, node) {
133 		skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
134 
135 		local_address(skc, &cur);
136 		if (addresses_equal(&cur, saddr, saddr->port))
137 			return true;
138 	}
139 
140 	return false;
141 }
142 
143 static bool lookup_subflow_by_daddr(const struct list_head *list,
144 				    struct mptcp_addr_info *daddr)
145 {
146 	struct mptcp_subflow_context *subflow;
147 	struct mptcp_addr_info cur;
148 	struct sock_common *skc;
149 
150 	list_for_each_entry(subflow, list, node) {
151 		skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
152 
153 		remote_address(skc, &cur);
154 		if (addresses_equal(&cur, daddr, daddr->port))
155 			return true;
156 	}
157 
158 	return false;
159 }
160 
161 static struct mptcp_pm_addr_entry *
162 select_local_address(const struct pm_nl_pernet *pernet,
163 		     struct mptcp_sock *msk)
164 {
165 	struct mptcp_pm_addr_entry *entry, *ret = NULL;
166 	struct sock *sk = (struct sock *)msk;
167 
168 	msk_owned_by_me(msk);
169 
170 	rcu_read_lock();
171 	__mptcp_flush_join_list(msk);
172 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
173 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
174 			continue;
175 
176 		if (entry->addr.family != sk->sk_family) {
177 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
178 			if ((entry->addr.family == AF_INET &&
179 			     !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
180 			    (sk->sk_family == AF_INET &&
181 			     !ipv6_addr_v4mapped(&entry->addr.addr6)))
182 #endif
183 				continue;
184 		}
185 
186 		/* avoid any address already in use by subflows and
187 		 * pending join
188 		 */
189 		if (!lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) {
190 			ret = entry;
191 			break;
192 		}
193 	}
194 	rcu_read_unlock();
195 	return ret;
196 }
197 
198 static struct mptcp_pm_addr_entry *
199 select_signal_address(struct pm_nl_pernet *pernet, unsigned int pos)
200 {
201 	struct mptcp_pm_addr_entry *entry, *ret = NULL;
202 	int i = 0;
203 
204 	rcu_read_lock();
205 	/* do not keep any additional per socket state, just signal
206 	 * the address list in order.
207 	 * Note: removal from the local address list during the msk life-cycle
208 	 * can lead to additional addresses not being announced.
209 	 */
210 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
211 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
212 			continue;
213 		if (i++ == pos) {
214 			ret = entry;
215 			break;
216 		}
217 	}
218 	rcu_read_unlock();
219 	return ret;
220 }
221 
222 unsigned int mptcp_pm_get_add_addr_signal_max(struct mptcp_sock *msk)
223 {
224 	struct pm_nl_pernet *pernet;
225 
226 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
227 	return READ_ONCE(pernet->add_addr_signal_max);
228 }
229 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
230 
231 unsigned int mptcp_pm_get_add_addr_accept_max(struct mptcp_sock *msk)
232 {
233 	struct pm_nl_pernet *pernet;
234 
235 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
236 	return READ_ONCE(pernet->add_addr_accept_max);
237 }
238 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
239 
240 unsigned int mptcp_pm_get_subflows_max(struct mptcp_sock *msk)
241 {
242 	struct pm_nl_pernet *pernet;
243 
244 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
245 	return READ_ONCE(pernet->subflows_max);
246 }
247 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
248 
249 unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk)
250 {
251 	struct pm_nl_pernet *pernet;
252 
253 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
254 	return READ_ONCE(pernet->local_addr_max);
255 }
256 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
257 
258 static void check_work_pending(struct mptcp_sock *msk)
259 {
260 	if (msk->pm.add_addr_signaled == mptcp_pm_get_add_addr_signal_max(msk) &&
261 	    (msk->pm.local_addr_used == mptcp_pm_get_local_addr_max(msk) ||
262 	     msk->pm.subflows == mptcp_pm_get_subflows_max(msk)))
263 		WRITE_ONCE(msk->pm.work_pending, false);
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 	struct sock *sk = (struct sock *)msk, *ssk;
433 	struct mptcp_subflow_context *subflow;
434 	struct mptcp_addr_info remote = { 0 };
435 	unsigned int subflows_max;
436 	int i = 0;
437 
438 	subflows_max = mptcp_pm_get_subflows_max(msk);
439 
440 	/* Non-fullmesh endpoint, fill in the single entry
441 	 * corresponding to the primary MPC subflow remote address
442 	 */
443 	if (!fullmesh) {
444 		remote_address((struct sock_common *)sk, &remote);
445 		msk->pm.subflows++;
446 		addrs[i++] = remote;
447 	} else {
448 		mptcp_for_each_subflow(msk, subflow) {
449 			ssk = mptcp_subflow_tcp_sock(subflow);
450 			remote_address((struct sock_common *)ssk, &remote);
451 			if (!lookup_address_in_vec(addrs, i, &remote) &&
452 			    msk->pm.subflows < subflows_max) {
453 				msk->pm.subflows++;
454 				addrs[i++] = remote;
455 			}
456 		}
457 	}
458 
459 	return i;
460 }
461 
462 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
463 {
464 	struct sock *sk = (struct sock *)msk;
465 	struct mptcp_pm_addr_entry *local;
466 	unsigned int add_addr_signal_max;
467 	unsigned int local_addr_max;
468 	struct pm_nl_pernet *pernet;
469 	unsigned int subflows_max;
470 
471 	pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
472 
473 	add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
474 	local_addr_max = mptcp_pm_get_local_addr_max(msk);
475 	subflows_max = mptcp_pm_get_subflows_max(msk);
476 
477 	pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
478 		 msk->pm.local_addr_used, local_addr_max,
479 		 msk->pm.add_addr_signaled, add_addr_signal_max,
480 		 msk->pm.subflows, subflows_max);
481 
482 	/* check first for announce */
483 	if (msk->pm.add_addr_signaled < add_addr_signal_max) {
484 		local = select_signal_address(pernet,
485 					      msk->pm.add_addr_signaled);
486 
487 		if (local) {
488 			if (mptcp_pm_alloc_anno_list(msk, local)) {
489 				msk->pm.add_addr_signaled++;
490 				mptcp_pm_announce_addr(msk, &local->addr, false);
491 				mptcp_pm_nl_addr_send_ack(msk);
492 			}
493 		} else {
494 			/* pick failed, avoid fourther attempts later */
495 			msk->pm.local_addr_used = add_addr_signal_max;
496 		}
497 
498 		check_work_pending(msk);
499 	}
500 
501 	/* check if should create a new subflow */
502 	if (msk->pm.local_addr_used < local_addr_max &&
503 	    msk->pm.subflows < subflows_max &&
504 	    !READ_ONCE(msk->pm.remote_deny_join_id0)) {
505 		local = select_local_address(pernet, msk);
506 		if (local) {
507 			bool fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
508 			struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
509 			int i, nr;
510 
511 			msk->pm.local_addr_used++;
512 			check_work_pending(msk);
513 			nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
514 			spin_unlock_bh(&msk->pm.lock);
515 			for (i = 0; i < nr; i++)
516 				__mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
517 			spin_lock_bh(&msk->pm.lock);
518 			return;
519 		}
520 
521 		/* lookup failed, avoid fourther attempts later */
522 		msk->pm.local_addr_used = local_addr_max;
523 		check_work_pending(msk);
524 	}
525 }
526 
527 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
528 {
529 	mptcp_pm_create_subflow_or_signal_addr(msk);
530 }
531 
532 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
533 {
534 	mptcp_pm_create_subflow_or_signal_addr(msk);
535 }
536 
537 /* Fill all the local addresses into the array addrs[],
538  * and return the array size.
539  */
540 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
541 					     struct mptcp_addr_info *addrs)
542 {
543 	struct sock *sk = (struct sock *)msk;
544 	struct mptcp_pm_addr_entry *entry;
545 	struct mptcp_addr_info local;
546 	struct pm_nl_pernet *pernet;
547 	unsigned int subflows_max;
548 	int i = 0;
549 
550 	pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
551 	subflows_max = mptcp_pm_get_subflows_max(msk);
552 
553 	rcu_read_lock();
554 	__mptcp_flush_join_list(msk);
555 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
556 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
557 			continue;
558 
559 		if (entry->addr.family != sk->sk_family) {
560 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
561 			if ((entry->addr.family == AF_INET &&
562 			     !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
563 			    (sk->sk_family == AF_INET &&
564 			     !ipv6_addr_v4mapped(&entry->addr.addr6)))
565 #endif
566 				continue;
567 		}
568 
569 		if (msk->pm.subflows < subflows_max) {
570 			msk->pm.subflows++;
571 			addrs[i++] = entry->addr;
572 		}
573 	}
574 	rcu_read_unlock();
575 
576 	/* If the array is empty, fill in the single
577 	 * 'IPADDRANY' local address
578 	 */
579 	if (!i) {
580 		memset(&local, 0, sizeof(local));
581 		local.family = msk->pm.remote.family;
582 
583 		msk->pm.subflows++;
584 		addrs[i++] = local;
585 	}
586 
587 	return i;
588 }
589 
590 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
591 {
592 	struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
593 	struct sock *sk = (struct sock *)msk;
594 	unsigned int add_addr_accept_max;
595 	struct mptcp_addr_info remote;
596 	unsigned int subflows_max;
597 	int i, nr;
598 
599 	add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
600 	subflows_max = mptcp_pm_get_subflows_max(msk);
601 
602 	pr_debug("accepted %d:%d remote family %d",
603 		 msk->pm.add_addr_accepted, add_addr_accept_max,
604 		 msk->pm.remote.family);
605 
606 	if (lookup_subflow_by_daddr(&msk->conn_list, &msk->pm.remote))
607 		goto add_addr_echo;
608 
609 	/* connect to the specified remote address, using whatever
610 	 * local address the routing configuration will pick.
611 	 */
612 	remote = msk->pm.remote;
613 	if (!remote.port)
614 		remote.port = sk->sk_dport;
615 	nr = fill_local_addresses_vec(msk, addrs);
616 
617 	msk->pm.add_addr_accepted++;
618 	if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
619 	    msk->pm.subflows >= subflows_max)
620 		WRITE_ONCE(msk->pm.accept_addr, false);
621 
622 	spin_unlock_bh(&msk->pm.lock);
623 	for (i = 0; i < nr; i++)
624 		__mptcp_subflow_connect(sk, &addrs[i], &remote);
625 	spin_lock_bh(&msk->pm.lock);
626 
627 add_addr_echo:
628 	mptcp_pm_announce_addr(msk, &msk->pm.remote, true);
629 	mptcp_pm_nl_addr_send_ack(msk);
630 }
631 
632 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
633 {
634 	struct mptcp_subflow_context *subflow;
635 
636 	msk_owned_by_me(msk);
637 	lockdep_assert_held(&msk->pm.lock);
638 
639 	if (!mptcp_pm_should_add_signal(msk) &&
640 	    !mptcp_pm_should_rm_signal(msk))
641 		return;
642 
643 	__mptcp_flush_join_list(msk);
644 	subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
645 	if (subflow) {
646 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
647 
648 		spin_unlock_bh(&msk->pm.lock);
649 		pr_debug("send ack for %s",
650 			 mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
651 
652 		mptcp_subflow_send_ack(ssk);
653 		spin_lock_bh(&msk->pm.lock);
654 	}
655 }
656 
657 static int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
658 					struct mptcp_addr_info *addr,
659 					u8 bkup)
660 {
661 	struct mptcp_subflow_context *subflow;
662 
663 	pr_debug("bkup=%d", bkup);
664 
665 	mptcp_for_each_subflow(msk, subflow) {
666 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
667 		struct sock *sk = (struct sock *)msk;
668 		struct mptcp_addr_info local;
669 
670 		local_address((struct sock_common *)ssk, &local);
671 		if (!addresses_equal(&local, addr, addr->port))
672 			continue;
673 
674 		subflow->backup = bkup;
675 		subflow->send_mp_prio = 1;
676 		subflow->request_bkup = bkup;
677 		__MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
678 
679 		spin_unlock_bh(&msk->pm.lock);
680 		pr_debug("send ack for mp_prio");
681 		mptcp_subflow_send_ack(ssk);
682 		spin_lock_bh(&msk->pm.lock);
683 
684 		return 0;
685 	}
686 
687 	return -EINVAL;
688 }
689 
690 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
691 					   const struct mptcp_rm_list *rm_list,
692 					   enum linux_mptcp_mib_field rm_type)
693 {
694 	struct mptcp_subflow_context *subflow, *tmp;
695 	struct sock *sk = (struct sock *)msk;
696 	u8 i;
697 
698 	pr_debug("%s rm_list_nr %d",
699 		 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
700 
701 	msk_owned_by_me(msk);
702 
703 	if (sk->sk_state == TCP_LISTEN)
704 		return;
705 
706 	if (!rm_list->nr)
707 		return;
708 
709 	if (list_empty(&msk->conn_list))
710 		return;
711 
712 	for (i = 0; i < rm_list->nr; i++) {
713 		list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
714 			struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
715 			int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
716 			u8 id = subflow->local_id;
717 
718 			if (rm_type == MPTCP_MIB_RMADDR)
719 				id = subflow->remote_id;
720 
721 			if (rm_list->ids[i] != id)
722 				continue;
723 
724 			pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
725 				 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
726 				 i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
727 			spin_unlock_bh(&msk->pm.lock);
728 			mptcp_subflow_shutdown(sk, ssk, how);
729 			mptcp_close_ssk(sk, ssk, subflow);
730 			spin_lock_bh(&msk->pm.lock);
731 
732 			if (rm_type == MPTCP_MIB_RMADDR) {
733 				msk->pm.add_addr_accepted--;
734 				WRITE_ONCE(msk->pm.accept_addr, true);
735 			} else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
736 				msk->pm.local_addr_used--;
737 			}
738 			msk->pm.subflows--;
739 			__MPTCP_INC_STATS(sock_net(sk), rm_type);
740 		}
741 	}
742 }
743 
744 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
745 {
746 	mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
747 }
748 
749 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
750 				     const struct mptcp_rm_list *rm_list)
751 {
752 	mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
753 }
754 
755 void mptcp_pm_nl_work(struct mptcp_sock *msk)
756 {
757 	struct mptcp_pm_data *pm = &msk->pm;
758 
759 	msk_owned_by_me(msk);
760 
761 	spin_lock_bh(&msk->pm.lock);
762 
763 	pr_debug("msk=%p status=%x", msk, pm->status);
764 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
765 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
766 		mptcp_pm_nl_add_addr_received(msk);
767 	}
768 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
769 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
770 		mptcp_pm_nl_addr_send_ack(msk);
771 	}
772 	if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
773 		pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
774 		mptcp_pm_nl_rm_addr_received(msk);
775 	}
776 	if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
777 		pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
778 		mptcp_pm_nl_fully_established(msk);
779 	}
780 	if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
781 		pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
782 		mptcp_pm_nl_subflow_established(msk);
783 	}
784 
785 	spin_unlock_bh(&msk->pm.lock);
786 }
787 
788 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
789 {
790 	return (entry->flags &
791 		(MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
792 		MPTCP_PM_ADDR_FLAG_SIGNAL;
793 }
794 
795 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
796 					     struct mptcp_pm_addr_entry *entry)
797 {
798 	struct mptcp_pm_addr_entry *cur;
799 	unsigned int addr_max;
800 	int ret = -EINVAL;
801 
802 	spin_lock_bh(&pernet->lock);
803 	/* to keep the code simple, don't do IDR-like allocation for address ID,
804 	 * just bail when we exceed limits
805 	 */
806 	if (pernet->next_id == MAX_ADDR_ID)
807 		pernet->next_id = 1;
808 	if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
809 		goto out;
810 	if (test_bit(entry->addr.id, pernet->id_bitmap))
811 		goto out;
812 
813 	/* do not insert duplicate address, differentiate on port only
814 	 * singled addresses
815 	 */
816 	list_for_each_entry(cur, &pernet->local_addr_list, list) {
817 		if (addresses_equal(&cur->addr, &entry->addr,
818 				    address_use_port(entry) &&
819 				    address_use_port(cur)))
820 			goto out;
821 	}
822 
823 	if (!entry->addr.id) {
824 find_next:
825 		entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
826 						    MAX_ADDR_ID + 1,
827 						    pernet->next_id);
828 		if ((!entry->addr.id || entry->addr.id > MAX_ADDR_ID) &&
829 		    pernet->next_id != 1) {
830 			pernet->next_id = 1;
831 			goto find_next;
832 		}
833 	}
834 
835 	if (!entry->addr.id || entry->addr.id > MAX_ADDR_ID)
836 		goto out;
837 
838 	__set_bit(entry->addr.id, pernet->id_bitmap);
839 	if (entry->addr.id > pernet->next_id)
840 		pernet->next_id = entry->addr.id;
841 
842 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
843 		addr_max = pernet->add_addr_signal_max;
844 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
845 	}
846 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
847 		addr_max = pernet->local_addr_max;
848 		WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
849 	}
850 
851 	pernet->addrs++;
852 	list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
853 	ret = entry->addr.id;
854 
855 out:
856 	spin_unlock_bh(&pernet->lock);
857 	return ret;
858 }
859 
860 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
861 					    struct mptcp_pm_addr_entry *entry)
862 {
863 	struct sockaddr_storage addr;
864 	struct mptcp_sock *msk;
865 	struct socket *ssock;
866 	int backlog = 1024;
867 	int err;
868 
869 	err = sock_create_kern(sock_net(sk), entry->addr.family,
870 			       SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
871 	if (err)
872 		return err;
873 
874 	msk = mptcp_sk(entry->lsk->sk);
875 	if (!msk) {
876 		err = -EINVAL;
877 		goto out;
878 	}
879 
880 	ssock = __mptcp_nmpc_socket(msk);
881 	if (!ssock) {
882 		err = -EINVAL;
883 		goto out;
884 	}
885 
886 	mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
887 	err = kernel_bind(ssock, (struct sockaddr *)&addr,
888 			  sizeof(struct sockaddr_in));
889 	if (err) {
890 		pr_warn("kernel_bind error, err=%d", err);
891 		goto out;
892 	}
893 
894 	err = kernel_listen(ssock, backlog);
895 	if (err) {
896 		pr_warn("kernel_listen error, err=%d", err);
897 		goto out;
898 	}
899 
900 	return 0;
901 
902 out:
903 	sock_release(entry->lsk);
904 	return err;
905 }
906 
907 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
908 {
909 	struct mptcp_pm_addr_entry *entry;
910 	struct mptcp_addr_info skc_local;
911 	struct mptcp_addr_info msk_local;
912 	struct pm_nl_pernet *pernet;
913 	int ret = -1;
914 
915 	if (WARN_ON_ONCE(!msk))
916 		return -1;
917 
918 	/* The 0 ID mapping is defined by the first subflow, copied into the msk
919 	 * addr
920 	 */
921 	local_address((struct sock_common *)msk, &msk_local);
922 	local_address((struct sock_common *)skc, &skc_local);
923 	if (addresses_equal(&msk_local, &skc_local, false))
924 		return 0;
925 
926 	if (address_zero(&skc_local))
927 		return 0;
928 
929 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
930 
931 	rcu_read_lock();
932 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
933 		if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
934 			ret = entry->addr.id;
935 			break;
936 		}
937 	}
938 	rcu_read_unlock();
939 	if (ret >= 0)
940 		return ret;
941 
942 	/* address not found, add to local list */
943 	entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
944 	if (!entry)
945 		return -ENOMEM;
946 
947 	entry->addr = skc_local;
948 	entry->addr.id = 0;
949 	entry->addr.port = 0;
950 	entry->ifindex = 0;
951 	entry->flags = 0;
952 	entry->lsk = NULL;
953 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
954 	if (ret < 0)
955 		kfree(entry);
956 
957 	return ret;
958 }
959 
960 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
961 {
962 	struct mptcp_pm_data *pm = &msk->pm;
963 	bool subflows;
964 
965 	subflows = !!mptcp_pm_get_subflows_max(msk);
966 	WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
967 		   !!mptcp_pm_get_add_addr_signal_max(msk));
968 	WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
969 	WRITE_ONCE(pm->accept_subflow, subflows);
970 }
971 
972 #define MPTCP_PM_CMD_GRP_OFFSET       0
973 #define MPTCP_PM_EV_GRP_OFFSET        1
974 
975 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
976 	[MPTCP_PM_CMD_GRP_OFFSET]	= { .name = MPTCP_PM_CMD_GRP_NAME, },
977 	[MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
978 					    .flags = GENL_UNS_ADMIN_PERM,
979 					  },
980 };
981 
982 static const struct nla_policy
983 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
984 	[MPTCP_PM_ADDR_ATTR_FAMILY]	= { .type	= NLA_U16,	},
985 	[MPTCP_PM_ADDR_ATTR_ID]		= { .type	= NLA_U8,	},
986 	[MPTCP_PM_ADDR_ATTR_ADDR4]	= { .type	= NLA_U32,	},
987 	[MPTCP_PM_ADDR_ATTR_ADDR6]	=
988 		NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
989 	[MPTCP_PM_ADDR_ATTR_PORT]	= { .type	= NLA_U16	},
990 	[MPTCP_PM_ADDR_ATTR_FLAGS]	= { .type	= NLA_U32	},
991 	[MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type	= NLA_S32	},
992 };
993 
994 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
995 	[MPTCP_PM_ATTR_ADDR]		=
996 					NLA_POLICY_NESTED(mptcp_pm_addr_policy),
997 	[MPTCP_PM_ATTR_RCV_ADD_ADDRS]	= { .type	= NLA_U32,	},
998 	[MPTCP_PM_ATTR_SUBFLOWS]	= { .type	= NLA_U32,	},
999 };
1000 
1001 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1002 {
1003 	struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1004 	struct sock *sk = (struct sock *)msk;
1005 	unsigned int active_max_loss_cnt;
1006 	struct net *net = sock_net(sk);
1007 	unsigned int stale_loss_cnt;
1008 	bool slow;
1009 
1010 	stale_loss_cnt = mptcp_stale_loss_cnt(net);
1011 	if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1012 		return;
1013 
1014 	/* look for another available subflow not in loss state */
1015 	active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1016 	mptcp_for_each_subflow(msk, iter) {
1017 		if (iter != subflow && mptcp_subflow_active(iter) &&
1018 		    iter->stale_count < active_max_loss_cnt) {
1019 			/* we have some alternatives, try to mark this subflow as idle ...*/
1020 			slow = lock_sock_fast(ssk);
1021 			if (!tcp_rtx_and_write_queues_empty(ssk)) {
1022 				subflow->stale = 1;
1023 				__mptcp_retransmit_pending_data(sk);
1024 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1025 			}
1026 			unlock_sock_fast(ssk, slow);
1027 
1028 			/* always try to push the pending data regarless of re-injections:
1029 			 * we can possibly use backup subflows now, and subflow selection
1030 			 * is cheap under the msk socket lock
1031 			 */
1032 			__mptcp_push_pending(sk, 0);
1033 			return;
1034 		}
1035 	}
1036 }
1037 
1038 static int mptcp_pm_family_to_addr(int family)
1039 {
1040 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1041 	if (family == AF_INET6)
1042 		return MPTCP_PM_ADDR_ATTR_ADDR6;
1043 #endif
1044 	return MPTCP_PM_ADDR_ATTR_ADDR4;
1045 }
1046 
1047 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1048 			       bool require_family,
1049 			       struct mptcp_pm_addr_entry *entry)
1050 {
1051 	struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1052 	int err, addr_addr;
1053 
1054 	if (!attr) {
1055 		GENL_SET_ERR_MSG(info, "missing address info");
1056 		return -EINVAL;
1057 	}
1058 
1059 	/* no validation needed - was already done via nested policy */
1060 	err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1061 					  mptcp_pm_addr_policy, info->extack);
1062 	if (err)
1063 		return err;
1064 
1065 	memset(entry, 0, sizeof(*entry));
1066 	if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1067 		if (!require_family)
1068 			goto skip_family;
1069 
1070 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1071 				    "missing family");
1072 		return -EINVAL;
1073 	}
1074 
1075 	entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1076 	if (entry->addr.family != AF_INET
1077 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1078 	    && entry->addr.family != AF_INET6
1079 #endif
1080 	    ) {
1081 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1082 				    "unknown address family");
1083 		return -EINVAL;
1084 	}
1085 	addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1086 	if (!tb[addr_addr]) {
1087 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1088 				    "missing address data");
1089 		return -EINVAL;
1090 	}
1091 
1092 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1093 	if (entry->addr.family == AF_INET6)
1094 		entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1095 	else
1096 #endif
1097 		entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1098 
1099 skip_family:
1100 	if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1101 		u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1102 
1103 		entry->ifindex = val;
1104 	}
1105 
1106 	if (tb[MPTCP_PM_ADDR_ATTR_ID])
1107 		entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1108 
1109 	if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1110 		entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1111 
1112 	if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1113 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1114 			NL_SET_ERR_MSG_ATTR(info->extack, attr,
1115 					    "flags must have signal when using port");
1116 			return -EINVAL;
1117 		}
1118 		entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1119 	}
1120 
1121 	return 0;
1122 }
1123 
1124 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1125 {
1126 	return net_generic(genl_info_net(info), pm_nl_pernet_id);
1127 }
1128 
1129 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1130 {
1131 	struct mptcp_sock *msk;
1132 	long s_slot = 0, s_num = 0;
1133 
1134 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1135 		struct sock *sk = (struct sock *)msk;
1136 
1137 		if (!READ_ONCE(msk->fully_established))
1138 			goto next;
1139 
1140 		lock_sock(sk);
1141 		spin_lock_bh(&msk->pm.lock);
1142 		mptcp_pm_create_subflow_or_signal_addr(msk);
1143 		spin_unlock_bh(&msk->pm.lock);
1144 		release_sock(sk);
1145 
1146 next:
1147 		sock_put(sk);
1148 		cond_resched();
1149 	}
1150 
1151 	return 0;
1152 }
1153 
1154 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1155 {
1156 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1157 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1158 	struct mptcp_pm_addr_entry addr, *entry;
1159 	int ret;
1160 
1161 	ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1162 	if (ret < 0)
1163 		return ret;
1164 
1165 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1166 	if (!entry) {
1167 		GENL_SET_ERR_MSG(info, "can't allocate addr");
1168 		return -ENOMEM;
1169 	}
1170 
1171 	*entry = addr;
1172 	if (entry->addr.port) {
1173 		ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1174 		if (ret) {
1175 			GENL_SET_ERR_MSG(info, "create listen socket error");
1176 			kfree(entry);
1177 			return ret;
1178 		}
1179 	}
1180 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1181 	if (ret < 0) {
1182 		GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1183 		if (entry->lsk)
1184 			sock_release(entry->lsk);
1185 		kfree(entry);
1186 		return ret;
1187 	}
1188 
1189 	mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1190 
1191 	return 0;
1192 }
1193 
1194 static struct mptcp_pm_addr_entry *
1195 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
1196 {
1197 	struct mptcp_pm_addr_entry *entry;
1198 
1199 	list_for_each_entry(entry, &pernet->local_addr_list, list) {
1200 		if (entry->addr.id == id)
1201 			return entry;
1202 	}
1203 	return NULL;
1204 }
1205 
1206 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1207 					 u8 *flags, int *ifindex)
1208 {
1209 	struct mptcp_pm_addr_entry *entry;
1210 
1211 	*flags = 0;
1212 	*ifindex = 0;
1213 
1214 	if (id) {
1215 		rcu_read_lock();
1216 		entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1217 		if (entry) {
1218 			*flags = entry->flags;
1219 			*ifindex = entry->ifindex;
1220 		}
1221 		rcu_read_unlock();
1222 	}
1223 
1224 	return 0;
1225 }
1226 
1227 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1228 				      struct mptcp_addr_info *addr)
1229 {
1230 	struct mptcp_pm_add_entry *entry;
1231 
1232 	entry = mptcp_pm_del_add_timer(msk, addr, false);
1233 	if (entry) {
1234 		list_del(&entry->list);
1235 		kfree(entry);
1236 		return true;
1237 	}
1238 
1239 	return false;
1240 }
1241 
1242 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1243 				      struct mptcp_addr_info *addr,
1244 				      bool force)
1245 {
1246 	struct mptcp_rm_list list = { .nr = 0 };
1247 	bool ret;
1248 
1249 	list.ids[list.nr++] = addr->id;
1250 
1251 	ret = remove_anno_list_by_saddr(msk, addr);
1252 	if (ret || force) {
1253 		spin_lock_bh(&msk->pm.lock);
1254 		mptcp_pm_remove_addr(msk, &list);
1255 		spin_unlock_bh(&msk->pm.lock);
1256 	}
1257 	return ret;
1258 }
1259 
1260 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1261 						   struct mptcp_addr_info *addr)
1262 {
1263 	struct mptcp_sock *msk;
1264 	long s_slot = 0, s_num = 0;
1265 	struct mptcp_rm_list list = { .nr = 0 };
1266 
1267 	pr_debug("remove_id=%d", addr->id);
1268 
1269 	list.ids[list.nr++] = addr->id;
1270 
1271 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1272 		struct sock *sk = (struct sock *)msk;
1273 		bool remove_subflow;
1274 
1275 		if (list_empty(&msk->conn_list)) {
1276 			mptcp_pm_remove_anno_addr(msk, addr, false);
1277 			goto next;
1278 		}
1279 
1280 		lock_sock(sk);
1281 		remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1282 		mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1283 		if (remove_subflow)
1284 			mptcp_pm_remove_subflow(msk, &list);
1285 		release_sock(sk);
1286 
1287 next:
1288 		sock_put(sk);
1289 		cond_resched();
1290 	}
1291 
1292 	return 0;
1293 }
1294 
1295 /* caller must ensure the RCU grace period is already elapsed */
1296 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1297 {
1298 	if (entry->lsk)
1299 		sock_release(entry->lsk);
1300 	kfree(entry);
1301 }
1302 
1303 static int mptcp_nl_remove_id_zero_address(struct net *net,
1304 					   struct mptcp_addr_info *addr)
1305 {
1306 	struct mptcp_rm_list list = { .nr = 0 };
1307 	long s_slot = 0, s_num = 0;
1308 	struct mptcp_sock *msk;
1309 
1310 	list.ids[list.nr++] = 0;
1311 
1312 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1313 		struct sock *sk = (struct sock *)msk;
1314 		struct mptcp_addr_info msk_local;
1315 
1316 		if (list_empty(&msk->conn_list))
1317 			goto next;
1318 
1319 		local_address((struct sock_common *)msk, &msk_local);
1320 		if (!addresses_equal(&msk_local, addr, addr->port))
1321 			goto next;
1322 
1323 		lock_sock(sk);
1324 		spin_lock_bh(&msk->pm.lock);
1325 		mptcp_pm_remove_addr(msk, &list);
1326 		mptcp_pm_nl_rm_subflow_received(msk, &list);
1327 		spin_unlock_bh(&msk->pm.lock);
1328 		release_sock(sk);
1329 
1330 next:
1331 		sock_put(sk);
1332 		cond_resched();
1333 	}
1334 
1335 	return 0;
1336 }
1337 
1338 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1339 {
1340 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1341 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1342 	struct mptcp_pm_addr_entry addr, *entry;
1343 	unsigned int addr_max;
1344 	int ret;
1345 
1346 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1347 	if (ret < 0)
1348 		return ret;
1349 
1350 	/* the zero id address is special: the first address used by the msk
1351 	 * always gets such an id, so different subflows can have different zero
1352 	 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1353 	 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1354 	 */
1355 	if (addr.addr.id == 0)
1356 		return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1357 
1358 	spin_lock_bh(&pernet->lock);
1359 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1360 	if (!entry) {
1361 		GENL_SET_ERR_MSG(info, "address not found");
1362 		spin_unlock_bh(&pernet->lock);
1363 		return -EINVAL;
1364 	}
1365 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1366 		addr_max = pernet->add_addr_signal_max;
1367 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1368 	}
1369 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1370 		addr_max = pernet->local_addr_max;
1371 		WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1372 	}
1373 
1374 	pernet->addrs--;
1375 	list_del_rcu(&entry->list);
1376 	__clear_bit(entry->addr.id, pernet->id_bitmap);
1377 	spin_unlock_bh(&pernet->lock);
1378 
1379 	mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1380 	synchronize_rcu();
1381 	__mptcp_pm_release_addr_entry(entry);
1382 
1383 	return ret;
1384 }
1385 
1386 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1387 					       struct list_head *rm_list)
1388 {
1389 	struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1390 	struct mptcp_pm_addr_entry *entry;
1391 
1392 	list_for_each_entry(entry, rm_list, list) {
1393 		if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1394 		    alist.nr < MPTCP_RM_IDS_MAX &&
1395 		    slist.nr < MPTCP_RM_IDS_MAX) {
1396 			alist.ids[alist.nr++] = entry->addr.id;
1397 			slist.ids[slist.nr++] = entry->addr.id;
1398 		} else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1399 			 alist.nr < MPTCP_RM_IDS_MAX) {
1400 			alist.ids[alist.nr++] = entry->addr.id;
1401 		}
1402 	}
1403 
1404 	if (alist.nr) {
1405 		spin_lock_bh(&msk->pm.lock);
1406 		mptcp_pm_remove_addr(msk, &alist);
1407 		spin_unlock_bh(&msk->pm.lock);
1408 	}
1409 	if (slist.nr)
1410 		mptcp_pm_remove_subflow(msk, &slist);
1411 }
1412 
1413 static void mptcp_nl_remove_addrs_list(struct net *net,
1414 				       struct list_head *rm_list)
1415 {
1416 	long s_slot = 0, s_num = 0;
1417 	struct mptcp_sock *msk;
1418 
1419 	if (list_empty(rm_list))
1420 		return;
1421 
1422 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1423 		struct sock *sk = (struct sock *)msk;
1424 
1425 		lock_sock(sk);
1426 		mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1427 		release_sock(sk);
1428 
1429 		sock_put(sk);
1430 		cond_resched();
1431 	}
1432 }
1433 
1434 /* caller must ensure the RCU grace period is already elapsed */
1435 static void __flush_addrs(struct list_head *list)
1436 {
1437 	while (!list_empty(list)) {
1438 		struct mptcp_pm_addr_entry *cur;
1439 
1440 		cur = list_entry(list->next,
1441 				 struct mptcp_pm_addr_entry, list);
1442 		list_del_rcu(&cur->list);
1443 		__mptcp_pm_release_addr_entry(cur);
1444 	}
1445 }
1446 
1447 static void __reset_counters(struct pm_nl_pernet *pernet)
1448 {
1449 	WRITE_ONCE(pernet->add_addr_signal_max, 0);
1450 	WRITE_ONCE(pernet->add_addr_accept_max, 0);
1451 	WRITE_ONCE(pernet->local_addr_max, 0);
1452 	pernet->addrs = 0;
1453 }
1454 
1455 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1456 {
1457 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1458 	LIST_HEAD(free_list);
1459 
1460 	spin_lock_bh(&pernet->lock);
1461 	list_splice_init(&pernet->local_addr_list, &free_list);
1462 	__reset_counters(pernet);
1463 	pernet->next_id = 1;
1464 	bitmap_zero(pernet->id_bitmap, MAX_ADDR_ID + 1);
1465 	spin_unlock_bh(&pernet->lock);
1466 	mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1467 	synchronize_rcu();
1468 	__flush_addrs(&free_list);
1469 	return 0;
1470 }
1471 
1472 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1473 			      struct mptcp_pm_addr_entry *entry)
1474 {
1475 	struct mptcp_addr_info *addr = &entry->addr;
1476 	struct nlattr *attr;
1477 
1478 	attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1479 	if (!attr)
1480 		return -EMSGSIZE;
1481 
1482 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1483 		goto nla_put_failure;
1484 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1485 		goto nla_put_failure;
1486 	if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1487 		goto nla_put_failure;
1488 	if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1489 		goto nla_put_failure;
1490 	if (entry->ifindex &&
1491 	    nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1492 		goto nla_put_failure;
1493 
1494 	if (addr->family == AF_INET &&
1495 	    nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1496 			    addr->addr.s_addr))
1497 		goto nla_put_failure;
1498 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1499 	else if (addr->family == AF_INET6 &&
1500 		 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1501 		goto nla_put_failure;
1502 #endif
1503 	nla_nest_end(skb, attr);
1504 	return 0;
1505 
1506 nla_put_failure:
1507 	nla_nest_cancel(skb, attr);
1508 	return -EMSGSIZE;
1509 }
1510 
1511 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1512 {
1513 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1514 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1515 	struct mptcp_pm_addr_entry addr, *entry;
1516 	struct sk_buff *msg;
1517 	void *reply;
1518 	int ret;
1519 
1520 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1521 	if (ret < 0)
1522 		return ret;
1523 
1524 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1525 	if (!msg)
1526 		return -ENOMEM;
1527 
1528 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1529 				  info->genlhdr->cmd);
1530 	if (!reply) {
1531 		GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1532 		ret = -EMSGSIZE;
1533 		goto fail;
1534 	}
1535 
1536 	spin_lock_bh(&pernet->lock);
1537 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1538 	if (!entry) {
1539 		GENL_SET_ERR_MSG(info, "address not found");
1540 		ret = -EINVAL;
1541 		goto unlock_fail;
1542 	}
1543 
1544 	ret = mptcp_nl_fill_addr(msg, entry);
1545 	if (ret)
1546 		goto unlock_fail;
1547 
1548 	genlmsg_end(msg, reply);
1549 	ret = genlmsg_reply(msg, info);
1550 	spin_unlock_bh(&pernet->lock);
1551 	return ret;
1552 
1553 unlock_fail:
1554 	spin_unlock_bh(&pernet->lock);
1555 
1556 fail:
1557 	nlmsg_free(msg);
1558 	return ret;
1559 }
1560 
1561 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1562 				   struct netlink_callback *cb)
1563 {
1564 	struct net *net = sock_net(msg->sk);
1565 	struct mptcp_pm_addr_entry *entry;
1566 	struct pm_nl_pernet *pernet;
1567 	int id = cb->args[0];
1568 	void *hdr;
1569 	int i;
1570 
1571 	pernet = net_generic(net, pm_nl_pernet_id);
1572 
1573 	spin_lock_bh(&pernet->lock);
1574 	for (i = id; i < MAX_ADDR_ID + 1; i++) {
1575 		if (test_bit(i, pernet->id_bitmap)) {
1576 			entry = __lookup_addr_by_id(pernet, i);
1577 			if (!entry)
1578 				break;
1579 
1580 			if (entry->addr.id <= id)
1581 				continue;
1582 
1583 			hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1584 					  cb->nlh->nlmsg_seq, &mptcp_genl_family,
1585 					  NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1586 			if (!hdr)
1587 				break;
1588 
1589 			if (mptcp_nl_fill_addr(msg, entry) < 0) {
1590 				genlmsg_cancel(msg, hdr);
1591 				break;
1592 			}
1593 
1594 			id = entry->addr.id;
1595 			genlmsg_end(msg, hdr);
1596 		}
1597 	}
1598 	spin_unlock_bh(&pernet->lock);
1599 
1600 	cb->args[0] = id;
1601 	return msg->len;
1602 }
1603 
1604 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1605 {
1606 	struct nlattr *attr = info->attrs[id];
1607 
1608 	if (!attr)
1609 		return 0;
1610 
1611 	*limit = nla_get_u32(attr);
1612 	if (*limit > MPTCP_PM_ADDR_MAX) {
1613 		GENL_SET_ERR_MSG(info, "limit greater than maximum");
1614 		return -EINVAL;
1615 	}
1616 	return 0;
1617 }
1618 
1619 static int
1620 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1621 {
1622 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1623 	unsigned int rcv_addrs, subflows;
1624 	int ret;
1625 
1626 	spin_lock_bh(&pernet->lock);
1627 	rcv_addrs = pernet->add_addr_accept_max;
1628 	ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1629 	if (ret)
1630 		goto unlock;
1631 
1632 	subflows = pernet->subflows_max;
1633 	ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1634 	if (ret)
1635 		goto unlock;
1636 
1637 	WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1638 	WRITE_ONCE(pernet->subflows_max, subflows);
1639 
1640 unlock:
1641 	spin_unlock_bh(&pernet->lock);
1642 	return ret;
1643 }
1644 
1645 static int
1646 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1647 {
1648 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1649 	struct sk_buff *msg;
1650 	void *reply;
1651 
1652 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1653 	if (!msg)
1654 		return -ENOMEM;
1655 
1656 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1657 				  MPTCP_PM_CMD_GET_LIMITS);
1658 	if (!reply)
1659 		goto fail;
1660 
1661 	if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1662 			READ_ONCE(pernet->add_addr_accept_max)))
1663 		goto fail;
1664 
1665 	if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1666 			READ_ONCE(pernet->subflows_max)))
1667 		goto fail;
1668 
1669 	genlmsg_end(msg, reply);
1670 	return genlmsg_reply(msg, info);
1671 
1672 fail:
1673 	GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1674 	nlmsg_free(msg);
1675 	return -EMSGSIZE;
1676 }
1677 
1678 static int mptcp_nl_addr_backup(struct net *net,
1679 				struct mptcp_addr_info *addr,
1680 				u8 bkup)
1681 {
1682 	long s_slot = 0, s_num = 0;
1683 	struct mptcp_sock *msk;
1684 	int ret = -EINVAL;
1685 
1686 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1687 		struct sock *sk = (struct sock *)msk;
1688 
1689 		if (list_empty(&msk->conn_list))
1690 			goto next;
1691 
1692 		lock_sock(sk);
1693 		spin_lock_bh(&msk->pm.lock);
1694 		ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1695 		spin_unlock_bh(&msk->pm.lock);
1696 		release_sock(sk);
1697 
1698 next:
1699 		sock_put(sk);
1700 		cond_resched();
1701 	}
1702 
1703 	return ret;
1704 }
1705 
1706 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1707 {
1708 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1709 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1710 	struct mptcp_pm_addr_entry addr, *entry;
1711 	struct net *net = sock_net(skb->sk);
1712 	u8 bkup = 0;
1713 	int ret;
1714 
1715 	ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1716 	if (ret < 0)
1717 		return ret;
1718 
1719 	if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1720 		bkup = 1;
1721 
1722 	list_for_each_entry(entry, &pernet->local_addr_list, list) {
1723 		if (addresses_equal(&entry->addr, &addr.addr, true)) {
1724 			mptcp_nl_addr_backup(net, &entry->addr, bkup);
1725 
1726 			if (bkup)
1727 				entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1728 			else
1729 				entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1730 		}
1731 	}
1732 
1733 	return 0;
1734 }
1735 
1736 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1737 {
1738 	genlmsg_multicast_netns(&mptcp_genl_family, net,
1739 				nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1740 }
1741 
1742 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1743 {
1744 	const struct inet_sock *issk = inet_sk(ssk);
1745 	const struct mptcp_subflow_context *sf;
1746 
1747 	if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1748 		return -EMSGSIZE;
1749 
1750 	switch (ssk->sk_family) {
1751 	case AF_INET:
1752 		if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1753 			return -EMSGSIZE;
1754 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1755 			return -EMSGSIZE;
1756 		break;
1757 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1758 	case AF_INET6: {
1759 		const struct ipv6_pinfo *np = inet6_sk(ssk);
1760 
1761 		if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1762 			return -EMSGSIZE;
1763 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1764 			return -EMSGSIZE;
1765 		break;
1766 	}
1767 #endif
1768 	default:
1769 		WARN_ON_ONCE(1);
1770 		return -EMSGSIZE;
1771 	}
1772 
1773 	if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1774 		return -EMSGSIZE;
1775 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1776 		return -EMSGSIZE;
1777 
1778 	sf = mptcp_subflow_ctx(ssk);
1779 	if (WARN_ON_ONCE(!sf))
1780 		return -EINVAL;
1781 
1782 	if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1783 		return -EMSGSIZE;
1784 
1785 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1786 		return -EMSGSIZE;
1787 
1788 	return 0;
1789 }
1790 
1791 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1792 					 const struct mptcp_sock *msk,
1793 					 const struct sock *ssk)
1794 {
1795 	const struct sock *sk = (const struct sock *)msk;
1796 	const struct mptcp_subflow_context *sf;
1797 	u8 sk_err;
1798 
1799 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1800 		return -EMSGSIZE;
1801 
1802 	if (mptcp_event_add_subflow(skb, ssk))
1803 		return -EMSGSIZE;
1804 
1805 	sf = mptcp_subflow_ctx(ssk);
1806 	if (WARN_ON_ONCE(!sf))
1807 		return -EINVAL;
1808 
1809 	if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1810 		return -EMSGSIZE;
1811 
1812 	if (ssk->sk_bound_dev_if &&
1813 	    nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1814 		return -EMSGSIZE;
1815 
1816 	sk_err = ssk->sk_err;
1817 	if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1818 	    nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1819 		return -EMSGSIZE;
1820 
1821 	return 0;
1822 }
1823 
1824 static int mptcp_event_sub_established(struct sk_buff *skb,
1825 				       const struct mptcp_sock *msk,
1826 				       const struct sock *ssk)
1827 {
1828 	return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1829 }
1830 
1831 static int mptcp_event_sub_closed(struct sk_buff *skb,
1832 				  const struct mptcp_sock *msk,
1833 				  const struct sock *ssk)
1834 {
1835 	const struct mptcp_subflow_context *sf;
1836 
1837 	if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1838 		return -EMSGSIZE;
1839 
1840 	sf = mptcp_subflow_ctx(ssk);
1841 	if (!sf->reset_seen)
1842 		return 0;
1843 
1844 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1845 		return -EMSGSIZE;
1846 
1847 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1848 		return -EMSGSIZE;
1849 
1850 	return 0;
1851 }
1852 
1853 static int mptcp_event_created(struct sk_buff *skb,
1854 			       const struct mptcp_sock *msk,
1855 			       const struct sock *ssk)
1856 {
1857 	int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1858 
1859 	if (err)
1860 		return err;
1861 
1862 	return mptcp_event_add_subflow(skb, ssk);
1863 }
1864 
1865 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1866 {
1867 	struct net *net = sock_net((const struct sock *)msk);
1868 	struct nlmsghdr *nlh;
1869 	struct sk_buff *skb;
1870 
1871 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1872 		return;
1873 
1874 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1875 	if (!skb)
1876 		return;
1877 
1878 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1879 	if (!nlh)
1880 		goto nla_put_failure;
1881 
1882 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1883 		goto nla_put_failure;
1884 
1885 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1886 		goto nla_put_failure;
1887 
1888 	genlmsg_end(skb, nlh);
1889 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1890 	return;
1891 
1892 nla_put_failure:
1893 	kfree_skb(skb);
1894 }
1895 
1896 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1897 				const struct mptcp_addr_info *info)
1898 {
1899 	struct net *net = sock_net((const struct sock *)msk);
1900 	struct nlmsghdr *nlh;
1901 	struct sk_buff *skb;
1902 
1903 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1904 		return;
1905 
1906 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1907 	if (!skb)
1908 		return;
1909 
1910 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1911 			  MPTCP_EVENT_ANNOUNCED);
1912 	if (!nlh)
1913 		goto nla_put_failure;
1914 
1915 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1916 		goto nla_put_failure;
1917 
1918 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
1919 		goto nla_put_failure;
1920 
1921 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
1922 		goto nla_put_failure;
1923 
1924 	switch (info->family) {
1925 	case AF_INET:
1926 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
1927 			goto nla_put_failure;
1928 		break;
1929 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1930 	case AF_INET6:
1931 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
1932 			goto nla_put_failure;
1933 		break;
1934 #endif
1935 	default:
1936 		WARN_ON_ONCE(1);
1937 		goto nla_put_failure;
1938 	}
1939 
1940 	genlmsg_end(skb, nlh);
1941 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1942 	return;
1943 
1944 nla_put_failure:
1945 	kfree_skb(skb);
1946 }
1947 
1948 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
1949 		 const struct sock *ssk, gfp_t gfp)
1950 {
1951 	struct net *net = sock_net((const struct sock *)msk);
1952 	struct nlmsghdr *nlh;
1953 	struct sk_buff *skb;
1954 
1955 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1956 		return;
1957 
1958 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
1959 	if (!skb)
1960 		return;
1961 
1962 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
1963 	if (!nlh)
1964 		goto nla_put_failure;
1965 
1966 	switch (type) {
1967 	case MPTCP_EVENT_UNSPEC:
1968 		WARN_ON_ONCE(1);
1969 		break;
1970 	case MPTCP_EVENT_CREATED:
1971 	case MPTCP_EVENT_ESTABLISHED:
1972 		if (mptcp_event_created(skb, msk, ssk) < 0)
1973 			goto nla_put_failure;
1974 		break;
1975 	case MPTCP_EVENT_CLOSED:
1976 		if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
1977 			goto nla_put_failure;
1978 		break;
1979 	case MPTCP_EVENT_ANNOUNCED:
1980 	case MPTCP_EVENT_REMOVED:
1981 		/* call mptcp_event_addr_announced()/removed instead */
1982 		WARN_ON_ONCE(1);
1983 		break;
1984 	case MPTCP_EVENT_SUB_ESTABLISHED:
1985 	case MPTCP_EVENT_SUB_PRIORITY:
1986 		if (mptcp_event_sub_established(skb, msk, ssk) < 0)
1987 			goto nla_put_failure;
1988 		break;
1989 	case MPTCP_EVENT_SUB_CLOSED:
1990 		if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
1991 			goto nla_put_failure;
1992 		break;
1993 	}
1994 
1995 	genlmsg_end(skb, nlh);
1996 	mptcp_nl_mcast_send(net, skb, gfp);
1997 	return;
1998 
1999 nla_put_failure:
2000 	kfree_skb(skb);
2001 }
2002 
2003 static const struct genl_small_ops mptcp_pm_ops[] = {
2004 	{
2005 		.cmd    = MPTCP_PM_CMD_ADD_ADDR,
2006 		.doit   = mptcp_nl_cmd_add_addr,
2007 		.flags  = GENL_ADMIN_PERM,
2008 	},
2009 	{
2010 		.cmd    = MPTCP_PM_CMD_DEL_ADDR,
2011 		.doit   = mptcp_nl_cmd_del_addr,
2012 		.flags  = GENL_ADMIN_PERM,
2013 	},
2014 	{
2015 		.cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2016 		.doit   = mptcp_nl_cmd_flush_addrs,
2017 		.flags  = GENL_ADMIN_PERM,
2018 	},
2019 	{
2020 		.cmd    = MPTCP_PM_CMD_GET_ADDR,
2021 		.doit   = mptcp_nl_cmd_get_addr,
2022 		.dumpit   = mptcp_nl_cmd_dump_addrs,
2023 	},
2024 	{
2025 		.cmd    = MPTCP_PM_CMD_SET_LIMITS,
2026 		.doit   = mptcp_nl_cmd_set_limits,
2027 		.flags  = GENL_ADMIN_PERM,
2028 	},
2029 	{
2030 		.cmd    = MPTCP_PM_CMD_GET_LIMITS,
2031 		.doit   = mptcp_nl_cmd_get_limits,
2032 	},
2033 	{
2034 		.cmd    = MPTCP_PM_CMD_SET_FLAGS,
2035 		.doit   = mptcp_nl_cmd_set_flags,
2036 		.flags  = GENL_ADMIN_PERM,
2037 	},
2038 };
2039 
2040 static struct genl_family mptcp_genl_family __ro_after_init = {
2041 	.name		= MPTCP_PM_NAME,
2042 	.version	= MPTCP_PM_VER,
2043 	.maxattr	= MPTCP_PM_ATTR_MAX,
2044 	.policy		= mptcp_pm_policy,
2045 	.netnsok	= true,
2046 	.module		= THIS_MODULE,
2047 	.small_ops	= mptcp_pm_ops,
2048 	.n_small_ops	= ARRAY_SIZE(mptcp_pm_ops),
2049 	.mcgrps		= mptcp_pm_mcgrps,
2050 	.n_mcgrps	= ARRAY_SIZE(mptcp_pm_mcgrps),
2051 };
2052 
2053 static int __net_init pm_nl_init_net(struct net *net)
2054 {
2055 	struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2056 
2057 	INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2058 
2059 	/* Cit. 2 subflows ought to be enough for anybody. */
2060 	pernet->subflows_max = 2;
2061 	pernet->next_id = 1;
2062 	pernet->stale_loss_cnt = 4;
2063 	spin_lock_init(&pernet->lock);
2064 
2065 	/* No need to initialize other pernet fields, the struct is zeroed at
2066 	 * allocation time.
2067 	 */
2068 
2069 	return 0;
2070 }
2071 
2072 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2073 {
2074 	struct net *net;
2075 
2076 	list_for_each_entry(net, net_list, exit_list) {
2077 		struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2078 
2079 		/* net is removed from namespace list, can't race with
2080 		 * other modifiers, also netns core already waited for a
2081 		 * RCU grace period.
2082 		 */
2083 		__flush_addrs(&pernet->local_addr_list);
2084 	}
2085 }
2086 
2087 static struct pernet_operations mptcp_pm_pernet_ops = {
2088 	.init = pm_nl_init_net,
2089 	.exit_batch = pm_nl_exit_net,
2090 	.id = &pm_nl_pernet_id,
2091 	.size = sizeof(struct pm_nl_pernet),
2092 };
2093 
2094 void __init mptcp_pm_nl_init(void)
2095 {
2096 	if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2097 		panic("Failed to register MPTCP PM pernet subsystem.\n");
2098 
2099 	if (genl_register_family(&mptcp_genl_family))
2100 		panic("Failed to register MPTCP PM netlink family\n");
2101 }
2102