xref: /openbmc/linux/net/mptcp/pm_netlink.c (revision 8931ddd8)
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 struct mptcp_pm_addr_entry *
482 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
483 	      bool lookup_by_id)
484 {
485 	struct mptcp_pm_addr_entry *entry;
486 
487 	list_for_each_entry(entry, &pernet->local_addr_list, list) {
488 		if ((!lookup_by_id && addresses_equal(&entry->addr, info, true)) ||
489 		    (lookup_by_id && entry->addr.id == info->id))
490 			return entry;
491 	}
492 	return NULL;
493 }
494 
495 static int
496 lookup_id_by_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *addr)
497 {
498 	struct mptcp_pm_addr_entry *entry;
499 	int ret = -1;
500 
501 	rcu_read_lock();
502 	list_for_each_entry(entry, &pernet->local_addr_list, list) {
503 		if (addresses_equal(&entry->addr, addr, entry->addr.port)) {
504 			ret = entry->addr.id;
505 			break;
506 		}
507 	}
508 	rcu_read_unlock();
509 	return ret;
510 }
511 
512 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
513 {
514 	struct sock *sk = (struct sock *)msk;
515 	struct mptcp_pm_addr_entry *local;
516 	unsigned int add_addr_signal_max;
517 	unsigned int local_addr_max;
518 	struct pm_nl_pernet *pernet;
519 	unsigned int subflows_max;
520 
521 	pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
522 
523 	add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
524 	local_addr_max = mptcp_pm_get_local_addr_max(msk);
525 	subflows_max = mptcp_pm_get_subflows_max(msk);
526 
527 	/* do lazy endpoint usage accounting for the MPC subflows */
528 	if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
529 		struct mptcp_addr_info mpc_addr;
530 		int mpc_id;
531 
532 		local_address((struct sock_common *)msk->first, &mpc_addr);
533 		mpc_id = lookup_id_by_addr(pernet, &mpc_addr);
534 		if (mpc_id >= 0)
535 			__clear_bit(mpc_id, msk->pm.id_avail_bitmap);
536 
537 		msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
538 	}
539 
540 	pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
541 		 msk->pm.local_addr_used, local_addr_max,
542 		 msk->pm.add_addr_signaled, add_addr_signal_max,
543 		 msk->pm.subflows, subflows_max);
544 
545 	/* check first for announce */
546 	if (msk->pm.add_addr_signaled < add_addr_signal_max) {
547 		local = select_signal_address(pernet, msk);
548 
549 		if (local) {
550 			if (mptcp_pm_alloc_anno_list(msk, local)) {
551 				__clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
552 				msk->pm.add_addr_signaled++;
553 				mptcp_pm_announce_addr(msk, &local->addr, false);
554 				mptcp_pm_nl_addr_send_ack(msk);
555 			}
556 		}
557 	}
558 
559 	/* check if should create a new subflow */
560 	while (msk->pm.local_addr_used < local_addr_max &&
561 	       msk->pm.subflows < subflows_max) {
562 		struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
563 		bool fullmesh;
564 		int i, nr;
565 
566 		local = select_local_address(pernet, msk);
567 		if (!local)
568 			break;
569 
570 		fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
571 
572 		msk->pm.local_addr_used++;
573 		nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
574 		if (nr)
575 			__clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
576 		spin_unlock_bh(&msk->pm.lock);
577 		for (i = 0; i < nr; i++)
578 			__mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
579 		spin_lock_bh(&msk->pm.lock);
580 	}
581 	mptcp_pm_nl_check_work_pending(msk);
582 }
583 
584 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
585 {
586 	mptcp_pm_create_subflow_or_signal_addr(msk);
587 }
588 
589 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
590 {
591 	mptcp_pm_create_subflow_or_signal_addr(msk);
592 }
593 
594 /* Fill all the local addresses into the array addrs[],
595  * and return the array size.
596  */
597 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
598 					     struct mptcp_addr_info *addrs)
599 {
600 	struct sock *sk = (struct sock *)msk;
601 	struct mptcp_pm_addr_entry *entry;
602 	struct mptcp_addr_info local;
603 	struct pm_nl_pernet *pernet;
604 	unsigned int subflows_max;
605 	int i = 0;
606 
607 	pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
608 	subflows_max = mptcp_pm_get_subflows_max(msk);
609 
610 	rcu_read_lock();
611 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
612 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
613 			continue;
614 
615 		if (entry->addr.family != sk->sk_family) {
616 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
617 			if ((entry->addr.family == AF_INET &&
618 			     !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
619 			    (sk->sk_family == AF_INET &&
620 			     !ipv6_addr_v4mapped(&entry->addr.addr6)))
621 #endif
622 				continue;
623 		}
624 
625 		if (msk->pm.subflows < subflows_max) {
626 			msk->pm.subflows++;
627 			addrs[i++] = entry->addr;
628 		}
629 	}
630 	rcu_read_unlock();
631 
632 	/* If the array is empty, fill in the single
633 	 * 'IPADDRANY' local address
634 	 */
635 	if (!i) {
636 		memset(&local, 0, sizeof(local));
637 		local.family = msk->pm.remote.family;
638 
639 		msk->pm.subflows++;
640 		addrs[i++] = local;
641 	}
642 
643 	return i;
644 }
645 
646 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
647 {
648 	struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
649 	struct sock *sk = (struct sock *)msk;
650 	unsigned int add_addr_accept_max;
651 	struct mptcp_addr_info remote;
652 	unsigned int subflows_max;
653 	int i, nr;
654 
655 	add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
656 	subflows_max = mptcp_pm_get_subflows_max(msk);
657 
658 	pr_debug("accepted %d:%d remote family %d",
659 		 msk->pm.add_addr_accepted, add_addr_accept_max,
660 		 msk->pm.remote.family);
661 
662 	if (lookup_subflow_by_daddr(&msk->conn_list, &msk->pm.remote))
663 		goto add_addr_echo;
664 
665 	/* connect to the specified remote address, using whatever
666 	 * local address the routing configuration will pick.
667 	 */
668 	remote = msk->pm.remote;
669 	if (!remote.port)
670 		remote.port = sk->sk_dport;
671 	nr = fill_local_addresses_vec(msk, addrs);
672 
673 	msk->pm.add_addr_accepted++;
674 	if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
675 	    msk->pm.subflows >= subflows_max)
676 		WRITE_ONCE(msk->pm.accept_addr, false);
677 
678 	spin_unlock_bh(&msk->pm.lock);
679 	for (i = 0; i < nr; i++)
680 		__mptcp_subflow_connect(sk, &addrs[i], &remote);
681 	spin_lock_bh(&msk->pm.lock);
682 
683 add_addr_echo:
684 	mptcp_pm_announce_addr(msk, &msk->pm.remote, true);
685 	mptcp_pm_nl_addr_send_ack(msk);
686 }
687 
688 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
689 {
690 	struct mptcp_subflow_context *subflow;
691 
692 	msk_owned_by_me(msk);
693 	lockdep_assert_held(&msk->pm.lock);
694 
695 	if (!mptcp_pm_should_add_signal(msk) &&
696 	    !mptcp_pm_should_rm_signal(msk))
697 		return;
698 
699 	subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
700 	if (subflow) {
701 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
702 
703 		spin_unlock_bh(&msk->pm.lock);
704 		pr_debug("send ack for %s",
705 			 mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
706 
707 		mptcp_subflow_send_ack(ssk);
708 		spin_lock_bh(&msk->pm.lock);
709 	}
710 }
711 
712 static int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
713 					struct mptcp_addr_info *addr,
714 					u8 bkup)
715 {
716 	struct mptcp_subflow_context *subflow;
717 
718 	pr_debug("bkup=%d", bkup);
719 
720 	mptcp_for_each_subflow(msk, subflow) {
721 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
722 		struct sock *sk = (struct sock *)msk;
723 		struct mptcp_addr_info local;
724 
725 		local_address((struct sock_common *)ssk, &local);
726 		if (!addresses_equal(&local, addr, addr->port))
727 			continue;
728 
729 		subflow->backup = bkup;
730 		subflow->send_mp_prio = 1;
731 		subflow->request_bkup = bkup;
732 		__MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
733 
734 		spin_unlock_bh(&msk->pm.lock);
735 		pr_debug("send ack for mp_prio");
736 		mptcp_subflow_send_ack(ssk);
737 		spin_lock_bh(&msk->pm.lock);
738 
739 		return 0;
740 	}
741 
742 	return -EINVAL;
743 }
744 
745 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
746 					   const struct mptcp_rm_list *rm_list,
747 					   enum linux_mptcp_mib_field rm_type)
748 {
749 	struct mptcp_subflow_context *subflow, *tmp;
750 	struct sock *sk = (struct sock *)msk;
751 	u8 i;
752 
753 	pr_debug("%s rm_list_nr %d",
754 		 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
755 
756 	msk_owned_by_me(msk);
757 
758 	if (sk->sk_state == TCP_LISTEN)
759 		return;
760 
761 	if (!rm_list->nr)
762 		return;
763 
764 	if (list_empty(&msk->conn_list))
765 		return;
766 
767 	for (i = 0; i < rm_list->nr; i++) {
768 		bool removed = false;
769 
770 		list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
771 			struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
772 			int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
773 			u8 id = subflow->local_id;
774 
775 			if (rm_type == MPTCP_MIB_RMADDR)
776 				id = subflow->remote_id;
777 
778 			if (rm_list->ids[i] != id)
779 				continue;
780 
781 			pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
782 				 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
783 				 i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
784 			spin_unlock_bh(&msk->pm.lock);
785 			mptcp_subflow_shutdown(sk, ssk, how);
786 
787 			/* the following takes care of updating the subflows counter */
788 			mptcp_close_ssk(sk, ssk, subflow);
789 			spin_lock_bh(&msk->pm.lock);
790 
791 			removed = true;
792 			__MPTCP_INC_STATS(sock_net(sk), rm_type);
793 		}
794 		__set_bit(rm_list->ids[i], msk->pm.id_avail_bitmap);
795 		if (!removed)
796 			continue;
797 
798 		if (rm_type == MPTCP_MIB_RMADDR) {
799 			msk->pm.add_addr_accepted--;
800 			WRITE_ONCE(msk->pm.accept_addr, true);
801 		} else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
802 			msk->pm.local_addr_used--;
803 		}
804 	}
805 }
806 
807 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
808 {
809 	mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
810 }
811 
812 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
813 				     const struct mptcp_rm_list *rm_list)
814 {
815 	mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
816 }
817 
818 void mptcp_pm_nl_work(struct mptcp_sock *msk)
819 {
820 	struct mptcp_pm_data *pm = &msk->pm;
821 
822 	msk_owned_by_me(msk);
823 
824 	if (!(pm->status & MPTCP_PM_WORK_MASK))
825 		return;
826 
827 	spin_lock_bh(&msk->pm.lock);
828 
829 	pr_debug("msk=%p status=%x", msk, pm->status);
830 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
831 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
832 		mptcp_pm_nl_add_addr_received(msk);
833 	}
834 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
835 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
836 		mptcp_pm_nl_addr_send_ack(msk);
837 	}
838 	if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
839 		pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
840 		mptcp_pm_nl_rm_addr_received(msk);
841 	}
842 	if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
843 		pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
844 		mptcp_pm_nl_fully_established(msk);
845 	}
846 	if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
847 		pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
848 		mptcp_pm_nl_subflow_established(msk);
849 	}
850 
851 	spin_unlock_bh(&msk->pm.lock);
852 }
853 
854 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
855 {
856 	return (entry->flags &
857 		(MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
858 		MPTCP_PM_ADDR_FLAG_SIGNAL;
859 }
860 
861 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
862 					     struct mptcp_pm_addr_entry *entry)
863 {
864 	struct mptcp_pm_addr_entry *cur;
865 	unsigned int addr_max;
866 	int ret = -EINVAL;
867 
868 	spin_lock_bh(&pernet->lock);
869 	/* to keep the code simple, don't do IDR-like allocation for address ID,
870 	 * just bail when we exceed limits
871 	 */
872 	if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
873 		pernet->next_id = 1;
874 	if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
875 		goto out;
876 	if (test_bit(entry->addr.id, pernet->id_bitmap))
877 		goto out;
878 
879 	/* do not insert duplicate address, differentiate on port only
880 	 * singled addresses
881 	 */
882 	list_for_each_entry(cur, &pernet->local_addr_list, list) {
883 		if (addresses_equal(&cur->addr, &entry->addr,
884 				    address_use_port(entry) &&
885 				    address_use_port(cur)))
886 			goto out;
887 	}
888 
889 	if (!entry->addr.id) {
890 find_next:
891 		entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
892 						    MPTCP_PM_MAX_ADDR_ID + 1,
893 						    pernet->next_id);
894 		if (!entry->addr.id && pernet->next_id != 1) {
895 			pernet->next_id = 1;
896 			goto find_next;
897 		}
898 	}
899 
900 	if (!entry->addr.id)
901 		goto out;
902 
903 	__set_bit(entry->addr.id, pernet->id_bitmap);
904 	if (entry->addr.id > pernet->next_id)
905 		pernet->next_id = entry->addr.id;
906 
907 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
908 		addr_max = pernet->add_addr_signal_max;
909 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
910 	}
911 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
912 		addr_max = pernet->local_addr_max;
913 		WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
914 	}
915 
916 	pernet->addrs++;
917 	list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
918 	ret = entry->addr.id;
919 
920 out:
921 	spin_unlock_bh(&pernet->lock);
922 	return ret;
923 }
924 
925 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
926 					    struct mptcp_pm_addr_entry *entry)
927 {
928 	struct sockaddr_storage addr;
929 	struct mptcp_sock *msk;
930 	struct socket *ssock;
931 	int backlog = 1024;
932 	int err;
933 
934 	err = sock_create_kern(sock_net(sk), entry->addr.family,
935 			       SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
936 	if (err)
937 		return err;
938 
939 	msk = mptcp_sk(entry->lsk->sk);
940 	if (!msk) {
941 		err = -EINVAL;
942 		goto out;
943 	}
944 
945 	ssock = __mptcp_nmpc_socket(msk);
946 	if (!ssock) {
947 		err = -EINVAL;
948 		goto out;
949 	}
950 
951 	mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
952 	err = kernel_bind(ssock, (struct sockaddr *)&addr,
953 			  sizeof(struct sockaddr_in));
954 	if (err) {
955 		pr_warn("kernel_bind error, err=%d", err);
956 		goto out;
957 	}
958 
959 	err = kernel_listen(ssock, backlog);
960 	if (err) {
961 		pr_warn("kernel_listen error, err=%d", err);
962 		goto out;
963 	}
964 
965 	return 0;
966 
967 out:
968 	sock_release(entry->lsk);
969 	return err;
970 }
971 
972 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
973 {
974 	struct mptcp_pm_addr_entry *entry;
975 	struct mptcp_addr_info skc_local;
976 	struct mptcp_addr_info msk_local;
977 	struct pm_nl_pernet *pernet;
978 	int ret = -1;
979 
980 	if (WARN_ON_ONCE(!msk))
981 		return -1;
982 
983 	/* The 0 ID mapping is defined by the first subflow, copied into the msk
984 	 * addr
985 	 */
986 	local_address((struct sock_common *)msk, &msk_local);
987 	local_address((struct sock_common *)skc, &skc_local);
988 	if (addresses_equal(&msk_local, &skc_local, false))
989 		return 0;
990 
991 	if (address_zero(&skc_local))
992 		return 0;
993 
994 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
995 
996 	rcu_read_lock();
997 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
998 		if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
999 			ret = entry->addr.id;
1000 			break;
1001 		}
1002 	}
1003 	rcu_read_unlock();
1004 	if (ret >= 0)
1005 		return ret;
1006 
1007 	/* address not found, add to local list */
1008 	entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1009 	if (!entry)
1010 		return -ENOMEM;
1011 
1012 	entry->addr = skc_local;
1013 	entry->addr.id = 0;
1014 	entry->addr.port = 0;
1015 	entry->ifindex = 0;
1016 	entry->flags = 0;
1017 	entry->lsk = NULL;
1018 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1019 	if (ret < 0)
1020 		kfree(entry);
1021 
1022 	return ret;
1023 }
1024 
1025 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
1026 {
1027 	struct mptcp_pm_data *pm = &msk->pm;
1028 	bool subflows;
1029 
1030 	subflows = !!mptcp_pm_get_subflows_max(msk);
1031 	WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
1032 		   !!mptcp_pm_get_add_addr_signal_max(msk));
1033 	WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
1034 	WRITE_ONCE(pm->accept_subflow, subflows);
1035 }
1036 
1037 #define MPTCP_PM_CMD_GRP_OFFSET       0
1038 #define MPTCP_PM_EV_GRP_OFFSET        1
1039 
1040 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1041 	[MPTCP_PM_CMD_GRP_OFFSET]	= { .name = MPTCP_PM_CMD_GRP_NAME, },
1042 	[MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1043 					    .flags = GENL_UNS_ADMIN_PERM,
1044 					  },
1045 };
1046 
1047 static const struct nla_policy
1048 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1049 	[MPTCP_PM_ADDR_ATTR_FAMILY]	= { .type	= NLA_U16,	},
1050 	[MPTCP_PM_ADDR_ATTR_ID]		= { .type	= NLA_U8,	},
1051 	[MPTCP_PM_ADDR_ATTR_ADDR4]	= { .type	= NLA_U32,	},
1052 	[MPTCP_PM_ADDR_ATTR_ADDR6]	=
1053 		NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1054 	[MPTCP_PM_ADDR_ATTR_PORT]	= { .type	= NLA_U16	},
1055 	[MPTCP_PM_ADDR_ATTR_FLAGS]	= { .type	= NLA_U32	},
1056 	[MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type	= NLA_S32	},
1057 };
1058 
1059 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1060 	[MPTCP_PM_ATTR_ADDR]		=
1061 					NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1062 	[MPTCP_PM_ATTR_RCV_ADD_ADDRS]	= { .type	= NLA_U32,	},
1063 	[MPTCP_PM_ATTR_SUBFLOWS]	= { .type	= NLA_U32,	},
1064 };
1065 
1066 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1067 {
1068 	struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1069 	struct sock *sk = (struct sock *)msk;
1070 	unsigned int active_max_loss_cnt;
1071 	struct net *net = sock_net(sk);
1072 	unsigned int stale_loss_cnt;
1073 	bool slow;
1074 
1075 	stale_loss_cnt = mptcp_stale_loss_cnt(net);
1076 	if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1077 		return;
1078 
1079 	/* look for another available subflow not in loss state */
1080 	active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1081 	mptcp_for_each_subflow(msk, iter) {
1082 		if (iter != subflow && mptcp_subflow_active(iter) &&
1083 		    iter->stale_count < active_max_loss_cnt) {
1084 			/* we have some alternatives, try to mark this subflow as idle ...*/
1085 			slow = lock_sock_fast(ssk);
1086 			if (!tcp_rtx_and_write_queues_empty(ssk)) {
1087 				subflow->stale = 1;
1088 				__mptcp_retransmit_pending_data(sk);
1089 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1090 			}
1091 			unlock_sock_fast(ssk, slow);
1092 
1093 			/* always try to push the pending data regarless of re-injections:
1094 			 * we can possibly use backup subflows now, and subflow selection
1095 			 * is cheap under the msk socket lock
1096 			 */
1097 			__mptcp_push_pending(sk, 0);
1098 			return;
1099 		}
1100 	}
1101 }
1102 
1103 static int mptcp_pm_family_to_addr(int family)
1104 {
1105 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1106 	if (family == AF_INET6)
1107 		return MPTCP_PM_ADDR_ATTR_ADDR6;
1108 #endif
1109 	return MPTCP_PM_ADDR_ATTR_ADDR4;
1110 }
1111 
1112 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1113 			       bool require_family,
1114 			       struct mptcp_pm_addr_entry *entry)
1115 {
1116 	struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1117 	int err, addr_addr;
1118 
1119 	if (!attr) {
1120 		GENL_SET_ERR_MSG(info, "missing address info");
1121 		return -EINVAL;
1122 	}
1123 
1124 	/* no validation needed - was already done via nested policy */
1125 	err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1126 					  mptcp_pm_addr_policy, info->extack);
1127 	if (err)
1128 		return err;
1129 
1130 	memset(entry, 0, sizeof(*entry));
1131 	if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1132 		if (!require_family)
1133 			goto skip_family;
1134 
1135 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1136 				    "missing family");
1137 		return -EINVAL;
1138 	}
1139 
1140 	entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1141 	if (entry->addr.family != AF_INET
1142 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1143 	    && entry->addr.family != AF_INET6
1144 #endif
1145 	    ) {
1146 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1147 				    "unknown address family");
1148 		return -EINVAL;
1149 	}
1150 	addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1151 	if (!tb[addr_addr]) {
1152 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1153 				    "missing address data");
1154 		return -EINVAL;
1155 	}
1156 
1157 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1158 	if (entry->addr.family == AF_INET6)
1159 		entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1160 	else
1161 #endif
1162 		entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1163 
1164 skip_family:
1165 	if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1166 		u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1167 
1168 		entry->ifindex = val;
1169 	}
1170 
1171 	if (tb[MPTCP_PM_ADDR_ATTR_ID])
1172 		entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1173 
1174 	if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1175 		entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1176 
1177 	if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1178 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1179 			NL_SET_ERR_MSG_ATTR(info->extack, attr,
1180 					    "flags must have signal when using port");
1181 			return -EINVAL;
1182 		}
1183 		entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1184 	}
1185 
1186 	return 0;
1187 }
1188 
1189 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1190 {
1191 	return net_generic(genl_info_net(info), pm_nl_pernet_id);
1192 }
1193 
1194 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1195 {
1196 	struct mptcp_sock *msk;
1197 	long s_slot = 0, s_num = 0;
1198 
1199 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1200 		struct sock *sk = (struct sock *)msk;
1201 
1202 		if (!READ_ONCE(msk->fully_established))
1203 			goto next;
1204 
1205 		lock_sock(sk);
1206 		spin_lock_bh(&msk->pm.lock);
1207 		mptcp_pm_create_subflow_or_signal_addr(msk);
1208 		spin_unlock_bh(&msk->pm.lock);
1209 		release_sock(sk);
1210 
1211 next:
1212 		sock_put(sk);
1213 		cond_resched();
1214 	}
1215 
1216 	return 0;
1217 }
1218 
1219 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1220 {
1221 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1222 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1223 	struct mptcp_pm_addr_entry addr, *entry;
1224 	int ret;
1225 
1226 	ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1227 	if (ret < 0)
1228 		return ret;
1229 
1230 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1231 	if (!entry) {
1232 		GENL_SET_ERR_MSG(info, "can't allocate addr");
1233 		return -ENOMEM;
1234 	}
1235 
1236 	*entry = addr;
1237 	if (entry->addr.port) {
1238 		ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1239 		if (ret) {
1240 			GENL_SET_ERR_MSG(info, "create listen socket error");
1241 			kfree(entry);
1242 			return ret;
1243 		}
1244 	}
1245 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1246 	if (ret < 0) {
1247 		GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1248 		if (entry->lsk)
1249 			sock_release(entry->lsk);
1250 		kfree(entry);
1251 		return ret;
1252 	}
1253 
1254 	mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1255 
1256 	return 0;
1257 }
1258 
1259 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1260 					 u8 *flags, int *ifindex)
1261 {
1262 	struct mptcp_pm_addr_entry *entry;
1263 
1264 	*flags = 0;
1265 	*ifindex = 0;
1266 
1267 	if (id) {
1268 		rcu_read_lock();
1269 		entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1270 		if (entry) {
1271 			*flags = entry->flags;
1272 			*ifindex = entry->ifindex;
1273 		}
1274 		rcu_read_unlock();
1275 	}
1276 
1277 	return 0;
1278 }
1279 
1280 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1281 				      struct mptcp_addr_info *addr)
1282 {
1283 	struct mptcp_pm_add_entry *entry;
1284 
1285 	entry = mptcp_pm_del_add_timer(msk, addr, false);
1286 	if (entry) {
1287 		list_del(&entry->list);
1288 		kfree(entry);
1289 		return true;
1290 	}
1291 
1292 	return false;
1293 }
1294 
1295 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1296 				      struct mptcp_addr_info *addr,
1297 				      bool force)
1298 {
1299 	struct mptcp_rm_list list = { .nr = 0 };
1300 	bool ret;
1301 
1302 	list.ids[list.nr++] = addr->id;
1303 
1304 	ret = remove_anno_list_by_saddr(msk, addr);
1305 	if (ret || force) {
1306 		spin_lock_bh(&msk->pm.lock);
1307 		mptcp_pm_remove_addr(msk, &list);
1308 		spin_unlock_bh(&msk->pm.lock);
1309 	}
1310 	return ret;
1311 }
1312 
1313 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1314 						   struct mptcp_addr_info *addr)
1315 {
1316 	struct mptcp_sock *msk;
1317 	long s_slot = 0, s_num = 0;
1318 	struct mptcp_rm_list list = { .nr = 0 };
1319 
1320 	pr_debug("remove_id=%d", addr->id);
1321 
1322 	list.ids[list.nr++] = addr->id;
1323 
1324 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1325 		struct sock *sk = (struct sock *)msk;
1326 		bool remove_subflow;
1327 
1328 		if (list_empty(&msk->conn_list)) {
1329 			mptcp_pm_remove_anno_addr(msk, addr, false);
1330 			goto next;
1331 		}
1332 
1333 		lock_sock(sk);
1334 		remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1335 		mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1336 		if (remove_subflow)
1337 			mptcp_pm_remove_subflow(msk, &list);
1338 		release_sock(sk);
1339 
1340 next:
1341 		sock_put(sk);
1342 		cond_resched();
1343 	}
1344 
1345 	return 0;
1346 }
1347 
1348 /* caller must ensure the RCU grace period is already elapsed */
1349 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1350 {
1351 	if (entry->lsk)
1352 		sock_release(entry->lsk);
1353 	kfree(entry);
1354 }
1355 
1356 static int mptcp_nl_remove_id_zero_address(struct net *net,
1357 					   struct mptcp_addr_info *addr)
1358 {
1359 	struct mptcp_rm_list list = { .nr = 0 };
1360 	long s_slot = 0, s_num = 0;
1361 	struct mptcp_sock *msk;
1362 
1363 	list.ids[list.nr++] = 0;
1364 
1365 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1366 		struct sock *sk = (struct sock *)msk;
1367 		struct mptcp_addr_info msk_local;
1368 
1369 		if (list_empty(&msk->conn_list))
1370 			goto next;
1371 
1372 		local_address((struct sock_common *)msk, &msk_local);
1373 		if (!addresses_equal(&msk_local, addr, addr->port))
1374 			goto next;
1375 
1376 		lock_sock(sk);
1377 		spin_lock_bh(&msk->pm.lock);
1378 		mptcp_pm_remove_addr(msk, &list);
1379 		mptcp_pm_nl_rm_subflow_received(msk, &list);
1380 		spin_unlock_bh(&msk->pm.lock);
1381 		release_sock(sk);
1382 
1383 next:
1384 		sock_put(sk);
1385 		cond_resched();
1386 	}
1387 
1388 	return 0;
1389 }
1390 
1391 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1392 {
1393 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1394 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1395 	struct mptcp_pm_addr_entry addr, *entry;
1396 	unsigned int addr_max;
1397 	int ret;
1398 
1399 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1400 	if (ret < 0)
1401 		return ret;
1402 
1403 	/* the zero id address is special: the first address used by the msk
1404 	 * always gets such an id, so different subflows can have different zero
1405 	 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1406 	 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1407 	 */
1408 	if (addr.addr.id == 0)
1409 		return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1410 
1411 	spin_lock_bh(&pernet->lock);
1412 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1413 	if (!entry) {
1414 		GENL_SET_ERR_MSG(info, "address not found");
1415 		spin_unlock_bh(&pernet->lock);
1416 		return -EINVAL;
1417 	}
1418 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1419 		addr_max = pernet->add_addr_signal_max;
1420 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1421 	}
1422 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1423 		addr_max = pernet->local_addr_max;
1424 		WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1425 	}
1426 
1427 	pernet->addrs--;
1428 	list_del_rcu(&entry->list);
1429 	__clear_bit(entry->addr.id, pernet->id_bitmap);
1430 	spin_unlock_bh(&pernet->lock);
1431 
1432 	mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1433 	synchronize_rcu();
1434 	__mptcp_pm_release_addr_entry(entry);
1435 
1436 	return ret;
1437 }
1438 
1439 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1440 					       struct list_head *rm_list)
1441 {
1442 	struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1443 	struct mptcp_pm_addr_entry *entry;
1444 
1445 	list_for_each_entry(entry, rm_list, list) {
1446 		if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1447 		    alist.nr < MPTCP_RM_IDS_MAX &&
1448 		    slist.nr < MPTCP_RM_IDS_MAX) {
1449 			alist.ids[alist.nr++] = entry->addr.id;
1450 			slist.ids[slist.nr++] = entry->addr.id;
1451 		} else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1452 			 alist.nr < MPTCP_RM_IDS_MAX) {
1453 			alist.ids[alist.nr++] = entry->addr.id;
1454 		}
1455 	}
1456 
1457 	if (alist.nr) {
1458 		spin_lock_bh(&msk->pm.lock);
1459 		mptcp_pm_remove_addr(msk, &alist);
1460 		spin_unlock_bh(&msk->pm.lock);
1461 	}
1462 	if (slist.nr)
1463 		mptcp_pm_remove_subflow(msk, &slist);
1464 }
1465 
1466 static void mptcp_nl_remove_addrs_list(struct net *net,
1467 				       struct list_head *rm_list)
1468 {
1469 	long s_slot = 0, s_num = 0;
1470 	struct mptcp_sock *msk;
1471 
1472 	if (list_empty(rm_list))
1473 		return;
1474 
1475 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1476 		struct sock *sk = (struct sock *)msk;
1477 
1478 		lock_sock(sk);
1479 		mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1480 		release_sock(sk);
1481 
1482 		sock_put(sk);
1483 		cond_resched();
1484 	}
1485 }
1486 
1487 /* caller must ensure the RCU grace period is already elapsed */
1488 static void __flush_addrs(struct list_head *list)
1489 {
1490 	while (!list_empty(list)) {
1491 		struct mptcp_pm_addr_entry *cur;
1492 
1493 		cur = list_entry(list->next,
1494 				 struct mptcp_pm_addr_entry, list);
1495 		list_del_rcu(&cur->list);
1496 		__mptcp_pm_release_addr_entry(cur);
1497 	}
1498 }
1499 
1500 static void __reset_counters(struct pm_nl_pernet *pernet)
1501 {
1502 	WRITE_ONCE(pernet->add_addr_signal_max, 0);
1503 	WRITE_ONCE(pernet->add_addr_accept_max, 0);
1504 	WRITE_ONCE(pernet->local_addr_max, 0);
1505 	pernet->addrs = 0;
1506 }
1507 
1508 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1509 {
1510 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1511 	LIST_HEAD(free_list);
1512 
1513 	spin_lock_bh(&pernet->lock);
1514 	list_splice_init(&pernet->local_addr_list, &free_list);
1515 	__reset_counters(pernet);
1516 	pernet->next_id = 1;
1517 	bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1518 	spin_unlock_bh(&pernet->lock);
1519 	mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1520 	synchronize_rcu();
1521 	__flush_addrs(&free_list);
1522 	return 0;
1523 }
1524 
1525 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1526 			      struct mptcp_pm_addr_entry *entry)
1527 {
1528 	struct mptcp_addr_info *addr = &entry->addr;
1529 	struct nlattr *attr;
1530 
1531 	attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1532 	if (!attr)
1533 		return -EMSGSIZE;
1534 
1535 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1536 		goto nla_put_failure;
1537 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1538 		goto nla_put_failure;
1539 	if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1540 		goto nla_put_failure;
1541 	if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1542 		goto nla_put_failure;
1543 	if (entry->ifindex &&
1544 	    nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1545 		goto nla_put_failure;
1546 
1547 	if (addr->family == AF_INET &&
1548 	    nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1549 			    addr->addr.s_addr))
1550 		goto nla_put_failure;
1551 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1552 	else if (addr->family == AF_INET6 &&
1553 		 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1554 		goto nla_put_failure;
1555 #endif
1556 	nla_nest_end(skb, attr);
1557 	return 0;
1558 
1559 nla_put_failure:
1560 	nla_nest_cancel(skb, attr);
1561 	return -EMSGSIZE;
1562 }
1563 
1564 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1565 {
1566 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1567 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1568 	struct mptcp_pm_addr_entry addr, *entry;
1569 	struct sk_buff *msg;
1570 	void *reply;
1571 	int ret;
1572 
1573 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1574 	if (ret < 0)
1575 		return ret;
1576 
1577 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1578 	if (!msg)
1579 		return -ENOMEM;
1580 
1581 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1582 				  info->genlhdr->cmd);
1583 	if (!reply) {
1584 		GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1585 		ret = -EMSGSIZE;
1586 		goto fail;
1587 	}
1588 
1589 	spin_lock_bh(&pernet->lock);
1590 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1591 	if (!entry) {
1592 		GENL_SET_ERR_MSG(info, "address not found");
1593 		ret = -EINVAL;
1594 		goto unlock_fail;
1595 	}
1596 
1597 	ret = mptcp_nl_fill_addr(msg, entry);
1598 	if (ret)
1599 		goto unlock_fail;
1600 
1601 	genlmsg_end(msg, reply);
1602 	ret = genlmsg_reply(msg, info);
1603 	spin_unlock_bh(&pernet->lock);
1604 	return ret;
1605 
1606 unlock_fail:
1607 	spin_unlock_bh(&pernet->lock);
1608 
1609 fail:
1610 	nlmsg_free(msg);
1611 	return ret;
1612 }
1613 
1614 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1615 				   struct netlink_callback *cb)
1616 {
1617 	struct net *net = sock_net(msg->sk);
1618 	struct mptcp_pm_addr_entry *entry;
1619 	struct pm_nl_pernet *pernet;
1620 	int id = cb->args[0];
1621 	void *hdr;
1622 	int i;
1623 
1624 	pernet = net_generic(net, pm_nl_pernet_id);
1625 
1626 	spin_lock_bh(&pernet->lock);
1627 	for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1628 		if (test_bit(i, pernet->id_bitmap)) {
1629 			entry = __lookup_addr_by_id(pernet, i);
1630 			if (!entry)
1631 				break;
1632 
1633 			if (entry->addr.id <= id)
1634 				continue;
1635 
1636 			hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1637 					  cb->nlh->nlmsg_seq, &mptcp_genl_family,
1638 					  NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1639 			if (!hdr)
1640 				break;
1641 
1642 			if (mptcp_nl_fill_addr(msg, entry) < 0) {
1643 				genlmsg_cancel(msg, hdr);
1644 				break;
1645 			}
1646 
1647 			id = entry->addr.id;
1648 			genlmsg_end(msg, hdr);
1649 		}
1650 	}
1651 	spin_unlock_bh(&pernet->lock);
1652 
1653 	cb->args[0] = id;
1654 	return msg->len;
1655 }
1656 
1657 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1658 {
1659 	struct nlattr *attr = info->attrs[id];
1660 
1661 	if (!attr)
1662 		return 0;
1663 
1664 	*limit = nla_get_u32(attr);
1665 	if (*limit > MPTCP_PM_ADDR_MAX) {
1666 		GENL_SET_ERR_MSG(info, "limit greater than maximum");
1667 		return -EINVAL;
1668 	}
1669 	return 0;
1670 }
1671 
1672 static int
1673 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1674 {
1675 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1676 	unsigned int rcv_addrs, subflows;
1677 	int ret;
1678 
1679 	spin_lock_bh(&pernet->lock);
1680 	rcv_addrs = pernet->add_addr_accept_max;
1681 	ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1682 	if (ret)
1683 		goto unlock;
1684 
1685 	subflows = pernet->subflows_max;
1686 	ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1687 	if (ret)
1688 		goto unlock;
1689 
1690 	WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1691 	WRITE_ONCE(pernet->subflows_max, subflows);
1692 
1693 unlock:
1694 	spin_unlock_bh(&pernet->lock);
1695 	return ret;
1696 }
1697 
1698 static int
1699 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1700 {
1701 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1702 	struct sk_buff *msg;
1703 	void *reply;
1704 
1705 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1706 	if (!msg)
1707 		return -ENOMEM;
1708 
1709 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1710 				  MPTCP_PM_CMD_GET_LIMITS);
1711 	if (!reply)
1712 		goto fail;
1713 
1714 	if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1715 			READ_ONCE(pernet->add_addr_accept_max)))
1716 		goto fail;
1717 
1718 	if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1719 			READ_ONCE(pernet->subflows_max)))
1720 		goto fail;
1721 
1722 	genlmsg_end(msg, reply);
1723 	return genlmsg_reply(msg, info);
1724 
1725 fail:
1726 	GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1727 	nlmsg_free(msg);
1728 	return -EMSGSIZE;
1729 }
1730 
1731 static int mptcp_nl_addr_backup(struct net *net,
1732 				struct mptcp_addr_info *addr,
1733 				u8 bkup)
1734 {
1735 	long s_slot = 0, s_num = 0;
1736 	struct mptcp_sock *msk;
1737 	int ret = -EINVAL;
1738 
1739 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1740 		struct sock *sk = (struct sock *)msk;
1741 
1742 		if (list_empty(&msk->conn_list))
1743 			goto next;
1744 
1745 		lock_sock(sk);
1746 		spin_lock_bh(&msk->pm.lock);
1747 		ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1748 		spin_unlock_bh(&msk->pm.lock);
1749 		release_sock(sk);
1750 
1751 next:
1752 		sock_put(sk);
1753 		cond_resched();
1754 	}
1755 
1756 	return ret;
1757 }
1758 
1759 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1760 {
1761 	struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1762 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1763 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1764 	struct net *net = sock_net(skb->sk);
1765 	u8 bkup = 0, lookup_by_id = 0;
1766 	int ret;
1767 
1768 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1769 	if (ret < 0)
1770 		return ret;
1771 
1772 	if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1773 		bkup = 1;
1774 	if (addr.addr.family == AF_UNSPEC) {
1775 		lookup_by_id = 1;
1776 		if (!addr.addr.id)
1777 			return -EOPNOTSUPP;
1778 	}
1779 
1780 	spin_lock_bh(&pernet->lock);
1781 	entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1782 	if (!entry) {
1783 		spin_unlock_bh(&pernet->lock);
1784 		return -EINVAL;
1785 	}
1786 
1787 	if (bkup)
1788 		entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1789 	else
1790 		entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1791 	addr = *entry;
1792 	spin_unlock_bh(&pernet->lock);
1793 
1794 	mptcp_nl_addr_backup(net, &addr.addr, bkup);
1795 	return 0;
1796 }
1797 
1798 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1799 {
1800 	genlmsg_multicast_netns(&mptcp_genl_family, net,
1801 				nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1802 }
1803 
1804 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1805 {
1806 	const struct inet_sock *issk = inet_sk(ssk);
1807 	const struct mptcp_subflow_context *sf;
1808 
1809 	if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1810 		return -EMSGSIZE;
1811 
1812 	switch (ssk->sk_family) {
1813 	case AF_INET:
1814 		if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1815 			return -EMSGSIZE;
1816 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1817 			return -EMSGSIZE;
1818 		break;
1819 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1820 	case AF_INET6: {
1821 		const struct ipv6_pinfo *np = inet6_sk(ssk);
1822 
1823 		if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1824 			return -EMSGSIZE;
1825 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1826 			return -EMSGSIZE;
1827 		break;
1828 	}
1829 #endif
1830 	default:
1831 		WARN_ON_ONCE(1);
1832 		return -EMSGSIZE;
1833 	}
1834 
1835 	if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1836 		return -EMSGSIZE;
1837 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1838 		return -EMSGSIZE;
1839 
1840 	sf = mptcp_subflow_ctx(ssk);
1841 	if (WARN_ON_ONCE(!sf))
1842 		return -EINVAL;
1843 
1844 	if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1845 		return -EMSGSIZE;
1846 
1847 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1848 		return -EMSGSIZE;
1849 
1850 	return 0;
1851 }
1852 
1853 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1854 					 const struct mptcp_sock *msk,
1855 					 const struct sock *ssk)
1856 {
1857 	const struct sock *sk = (const struct sock *)msk;
1858 	const struct mptcp_subflow_context *sf;
1859 	u8 sk_err;
1860 
1861 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1862 		return -EMSGSIZE;
1863 
1864 	if (mptcp_event_add_subflow(skb, ssk))
1865 		return -EMSGSIZE;
1866 
1867 	sf = mptcp_subflow_ctx(ssk);
1868 	if (WARN_ON_ONCE(!sf))
1869 		return -EINVAL;
1870 
1871 	if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1872 		return -EMSGSIZE;
1873 
1874 	if (ssk->sk_bound_dev_if &&
1875 	    nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1876 		return -EMSGSIZE;
1877 
1878 	sk_err = ssk->sk_err;
1879 	if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1880 	    nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1881 		return -EMSGSIZE;
1882 
1883 	return 0;
1884 }
1885 
1886 static int mptcp_event_sub_established(struct sk_buff *skb,
1887 				       const struct mptcp_sock *msk,
1888 				       const struct sock *ssk)
1889 {
1890 	return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1891 }
1892 
1893 static int mptcp_event_sub_closed(struct sk_buff *skb,
1894 				  const struct mptcp_sock *msk,
1895 				  const struct sock *ssk)
1896 {
1897 	const struct mptcp_subflow_context *sf;
1898 
1899 	if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1900 		return -EMSGSIZE;
1901 
1902 	sf = mptcp_subflow_ctx(ssk);
1903 	if (!sf->reset_seen)
1904 		return 0;
1905 
1906 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1907 		return -EMSGSIZE;
1908 
1909 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1910 		return -EMSGSIZE;
1911 
1912 	return 0;
1913 }
1914 
1915 static int mptcp_event_created(struct sk_buff *skb,
1916 			       const struct mptcp_sock *msk,
1917 			       const struct sock *ssk)
1918 {
1919 	int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1920 
1921 	if (err)
1922 		return err;
1923 
1924 	return mptcp_event_add_subflow(skb, ssk);
1925 }
1926 
1927 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1928 {
1929 	struct net *net = sock_net((const struct sock *)msk);
1930 	struct nlmsghdr *nlh;
1931 	struct sk_buff *skb;
1932 
1933 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1934 		return;
1935 
1936 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1937 	if (!skb)
1938 		return;
1939 
1940 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1941 	if (!nlh)
1942 		goto nla_put_failure;
1943 
1944 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1945 		goto nla_put_failure;
1946 
1947 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1948 		goto nla_put_failure;
1949 
1950 	genlmsg_end(skb, nlh);
1951 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1952 	return;
1953 
1954 nla_put_failure:
1955 	kfree_skb(skb);
1956 }
1957 
1958 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1959 				const struct mptcp_addr_info *info)
1960 {
1961 	struct net *net = sock_net((const struct sock *)msk);
1962 	struct nlmsghdr *nlh;
1963 	struct sk_buff *skb;
1964 
1965 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1966 		return;
1967 
1968 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1969 	if (!skb)
1970 		return;
1971 
1972 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1973 			  MPTCP_EVENT_ANNOUNCED);
1974 	if (!nlh)
1975 		goto nla_put_failure;
1976 
1977 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1978 		goto nla_put_failure;
1979 
1980 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
1981 		goto nla_put_failure;
1982 
1983 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
1984 		goto nla_put_failure;
1985 
1986 	switch (info->family) {
1987 	case AF_INET:
1988 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
1989 			goto nla_put_failure;
1990 		break;
1991 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1992 	case AF_INET6:
1993 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
1994 			goto nla_put_failure;
1995 		break;
1996 #endif
1997 	default:
1998 		WARN_ON_ONCE(1);
1999 		goto nla_put_failure;
2000 	}
2001 
2002 	genlmsg_end(skb, nlh);
2003 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2004 	return;
2005 
2006 nla_put_failure:
2007 	kfree_skb(skb);
2008 }
2009 
2010 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2011 		 const struct sock *ssk, gfp_t gfp)
2012 {
2013 	struct net *net = sock_net((const struct sock *)msk);
2014 	struct nlmsghdr *nlh;
2015 	struct sk_buff *skb;
2016 
2017 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2018 		return;
2019 
2020 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2021 	if (!skb)
2022 		return;
2023 
2024 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2025 	if (!nlh)
2026 		goto nla_put_failure;
2027 
2028 	switch (type) {
2029 	case MPTCP_EVENT_UNSPEC:
2030 		WARN_ON_ONCE(1);
2031 		break;
2032 	case MPTCP_EVENT_CREATED:
2033 	case MPTCP_EVENT_ESTABLISHED:
2034 		if (mptcp_event_created(skb, msk, ssk) < 0)
2035 			goto nla_put_failure;
2036 		break;
2037 	case MPTCP_EVENT_CLOSED:
2038 		if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2039 			goto nla_put_failure;
2040 		break;
2041 	case MPTCP_EVENT_ANNOUNCED:
2042 	case MPTCP_EVENT_REMOVED:
2043 		/* call mptcp_event_addr_announced()/removed instead */
2044 		WARN_ON_ONCE(1);
2045 		break;
2046 	case MPTCP_EVENT_SUB_ESTABLISHED:
2047 	case MPTCP_EVENT_SUB_PRIORITY:
2048 		if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2049 			goto nla_put_failure;
2050 		break;
2051 	case MPTCP_EVENT_SUB_CLOSED:
2052 		if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2053 			goto nla_put_failure;
2054 		break;
2055 	}
2056 
2057 	genlmsg_end(skb, nlh);
2058 	mptcp_nl_mcast_send(net, skb, gfp);
2059 	return;
2060 
2061 nla_put_failure:
2062 	kfree_skb(skb);
2063 }
2064 
2065 static const struct genl_small_ops mptcp_pm_ops[] = {
2066 	{
2067 		.cmd    = MPTCP_PM_CMD_ADD_ADDR,
2068 		.doit   = mptcp_nl_cmd_add_addr,
2069 		.flags  = GENL_ADMIN_PERM,
2070 	},
2071 	{
2072 		.cmd    = MPTCP_PM_CMD_DEL_ADDR,
2073 		.doit   = mptcp_nl_cmd_del_addr,
2074 		.flags  = GENL_ADMIN_PERM,
2075 	},
2076 	{
2077 		.cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2078 		.doit   = mptcp_nl_cmd_flush_addrs,
2079 		.flags  = GENL_ADMIN_PERM,
2080 	},
2081 	{
2082 		.cmd    = MPTCP_PM_CMD_GET_ADDR,
2083 		.doit   = mptcp_nl_cmd_get_addr,
2084 		.dumpit   = mptcp_nl_cmd_dump_addrs,
2085 	},
2086 	{
2087 		.cmd    = MPTCP_PM_CMD_SET_LIMITS,
2088 		.doit   = mptcp_nl_cmd_set_limits,
2089 		.flags  = GENL_ADMIN_PERM,
2090 	},
2091 	{
2092 		.cmd    = MPTCP_PM_CMD_GET_LIMITS,
2093 		.doit   = mptcp_nl_cmd_get_limits,
2094 	},
2095 	{
2096 		.cmd    = MPTCP_PM_CMD_SET_FLAGS,
2097 		.doit   = mptcp_nl_cmd_set_flags,
2098 		.flags  = GENL_ADMIN_PERM,
2099 	},
2100 };
2101 
2102 static struct genl_family mptcp_genl_family __ro_after_init = {
2103 	.name		= MPTCP_PM_NAME,
2104 	.version	= MPTCP_PM_VER,
2105 	.maxattr	= MPTCP_PM_ATTR_MAX,
2106 	.policy		= mptcp_pm_policy,
2107 	.netnsok	= true,
2108 	.module		= THIS_MODULE,
2109 	.small_ops	= mptcp_pm_ops,
2110 	.n_small_ops	= ARRAY_SIZE(mptcp_pm_ops),
2111 	.mcgrps		= mptcp_pm_mcgrps,
2112 	.n_mcgrps	= ARRAY_SIZE(mptcp_pm_mcgrps),
2113 };
2114 
2115 static int __net_init pm_nl_init_net(struct net *net)
2116 {
2117 	struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2118 
2119 	INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2120 
2121 	/* Cit. 2 subflows ought to be enough for anybody. */
2122 	pernet->subflows_max = 2;
2123 	pernet->next_id = 1;
2124 	pernet->stale_loss_cnt = 4;
2125 	spin_lock_init(&pernet->lock);
2126 
2127 	/* No need to initialize other pernet fields, the struct is zeroed at
2128 	 * allocation time.
2129 	 */
2130 
2131 	return 0;
2132 }
2133 
2134 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2135 {
2136 	struct net *net;
2137 
2138 	list_for_each_entry(net, net_list, exit_list) {
2139 		struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2140 
2141 		/* net is removed from namespace list, can't race with
2142 		 * other modifiers, also netns core already waited for a
2143 		 * RCU grace period.
2144 		 */
2145 		__flush_addrs(&pernet->local_addr_list);
2146 	}
2147 }
2148 
2149 static struct pernet_operations mptcp_pm_pernet_ops = {
2150 	.init = pm_nl_init_net,
2151 	.exit_batch = pm_nl_exit_net,
2152 	.id = &pm_nl_pernet_id,
2153 	.size = sizeof(struct pm_nl_pernet),
2154 };
2155 
2156 void __init mptcp_pm_nl_init(void)
2157 {
2158 	if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2159 		panic("Failed to register MPTCP PM pernet subsystem.\n");
2160 
2161 	if (genl_register_family(&mptcp_genl_family))
2162 		panic("Failed to register MPTCP PM netlink family\n");
2163 }
2164