xref: /openbmc/linux/net/mptcp/pm_netlink.c (revision 3a9a6f3d)
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 		/* due to racing events on both ends we can reach here while
550 		 * previous add address is still running: if we invoke now
551 		 * mptcp_pm_announce_addr(), that will fail and the
552 		 * corresponding id will be marked as used.
553 		 * Instead let the PM machinery reschedule us when the
554 		 * current address announce will be completed.
555 		 */
556 		if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
557 			return;
558 
559 		if (local) {
560 			if (mptcp_pm_alloc_anno_list(msk, local)) {
561 				__clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
562 				msk->pm.add_addr_signaled++;
563 				mptcp_pm_announce_addr(msk, &local->addr, false);
564 				mptcp_pm_nl_addr_send_ack(msk);
565 			}
566 		}
567 	}
568 
569 	/* check if should create a new subflow */
570 	while (msk->pm.local_addr_used < local_addr_max &&
571 	       msk->pm.subflows < subflows_max) {
572 		struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
573 		bool fullmesh;
574 		int i, nr;
575 
576 		local = select_local_address(pernet, msk);
577 		if (!local)
578 			break;
579 
580 		fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
581 
582 		msk->pm.local_addr_used++;
583 		nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
584 		if (nr)
585 			__clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
586 		spin_unlock_bh(&msk->pm.lock);
587 		for (i = 0; i < nr; i++)
588 			__mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
589 		spin_lock_bh(&msk->pm.lock);
590 	}
591 	mptcp_pm_nl_check_work_pending(msk);
592 }
593 
594 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
595 {
596 	mptcp_pm_create_subflow_or_signal_addr(msk);
597 }
598 
599 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
600 {
601 	mptcp_pm_create_subflow_or_signal_addr(msk);
602 }
603 
604 /* Fill all the local addresses into the array addrs[],
605  * and return the array size.
606  */
607 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
608 					     struct mptcp_addr_info *addrs)
609 {
610 	struct sock *sk = (struct sock *)msk;
611 	struct mptcp_pm_addr_entry *entry;
612 	struct mptcp_addr_info local;
613 	struct pm_nl_pernet *pernet;
614 	unsigned int subflows_max;
615 	int i = 0;
616 
617 	pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
618 	subflows_max = mptcp_pm_get_subflows_max(msk);
619 
620 	rcu_read_lock();
621 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
622 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
623 			continue;
624 
625 		if (entry->addr.family != sk->sk_family) {
626 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
627 			if ((entry->addr.family == AF_INET &&
628 			     !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
629 			    (sk->sk_family == AF_INET &&
630 			     !ipv6_addr_v4mapped(&entry->addr.addr6)))
631 #endif
632 				continue;
633 		}
634 
635 		if (msk->pm.subflows < subflows_max) {
636 			msk->pm.subflows++;
637 			addrs[i++] = entry->addr;
638 		}
639 	}
640 	rcu_read_unlock();
641 
642 	/* If the array is empty, fill in the single
643 	 * 'IPADDRANY' local address
644 	 */
645 	if (!i) {
646 		memset(&local, 0, sizeof(local));
647 		local.family = msk->pm.remote.family;
648 
649 		msk->pm.subflows++;
650 		addrs[i++] = local;
651 	}
652 
653 	return i;
654 }
655 
656 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
657 {
658 	struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
659 	struct sock *sk = (struct sock *)msk;
660 	unsigned int add_addr_accept_max;
661 	struct mptcp_addr_info remote;
662 	unsigned int subflows_max;
663 	bool reset_port = false;
664 	int i, nr;
665 
666 	add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
667 	subflows_max = mptcp_pm_get_subflows_max(msk);
668 
669 	pr_debug("accepted %d:%d remote family %d",
670 		 msk->pm.add_addr_accepted, add_addr_accept_max,
671 		 msk->pm.remote.family);
672 
673 	remote = msk->pm.remote;
674 	if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
675 		goto add_addr_echo;
676 
677 	/* pick id 0 port, if none is provided the remote address */
678 	if (!remote.port) {
679 		reset_port = true;
680 		remote.port = sk->sk_dport;
681 	}
682 
683 	/* connect to the specified remote address, using whatever
684 	 * local address the routing configuration will pick.
685 	 */
686 	nr = fill_local_addresses_vec(msk, addrs);
687 
688 	msk->pm.add_addr_accepted++;
689 	if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
690 	    msk->pm.subflows >= subflows_max)
691 		WRITE_ONCE(msk->pm.accept_addr, false);
692 
693 	spin_unlock_bh(&msk->pm.lock);
694 	for (i = 0; i < nr; i++)
695 		__mptcp_subflow_connect(sk, &addrs[i], &remote);
696 	spin_lock_bh(&msk->pm.lock);
697 
698 	/* be sure to echo exactly the received address */
699 	if (reset_port)
700 		remote.port = 0;
701 
702 add_addr_echo:
703 	mptcp_pm_announce_addr(msk, &remote, true);
704 	mptcp_pm_nl_addr_send_ack(msk);
705 }
706 
707 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
708 {
709 	struct mptcp_subflow_context *subflow;
710 
711 	msk_owned_by_me(msk);
712 	lockdep_assert_held(&msk->pm.lock);
713 
714 	if (!mptcp_pm_should_add_signal(msk) &&
715 	    !mptcp_pm_should_rm_signal(msk))
716 		return;
717 
718 	subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
719 	if (subflow) {
720 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
721 
722 		spin_unlock_bh(&msk->pm.lock);
723 		pr_debug("send ack for %s",
724 			 mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
725 
726 		mptcp_subflow_send_ack(ssk);
727 		spin_lock_bh(&msk->pm.lock);
728 	}
729 }
730 
731 static int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
732 					struct mptcp_addr_info *addr,
733 					u8 bkup)
734 {
735 	struct mptcp_subflow_context *subflow;
736 
737 	pr_debug("bkup=%d", bkup);
738 
739 	mptcp_for_each_subflow(msk, subflow) {
740 		struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
741 		struct sock *sk = (struct sock *)msk;
742 		struct mptcp_addr_info local;
743 
744 		local_address((struct sock_common *)ssk, &local);
745 		if (!addresses_equal(&local, addr, addr->port))
746 			continue;
747 
748 		subflow->backup = bkup;
749 		subflow->send_mp_prio = 1;
750 		subflow->request_bkup = bkup;
751 		__MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
752 
753 		spin_unlock_bh(&msk->pm.lock);
754 		pr_debug("send ack for mp_prio");
755 		mptcp_subflow_send_ack(ssk);
756 		spin_lock_bh(&msk->pm.lock);
757 
758 		return 0;
759 	}
760 
761 	return -EINVAL;
762 }
763 
764 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
765 					   const struct mptcp_rm_list *rm_list,
766 					   enum linux_mptcp_mib_field rm_type)
767 {
768 	struct mptcp_subflow_context *subflow, *tmp;
769 	struct sock *sk = (struct sock *)msk;
770 	u8 i;
771 
772 	pr_debug("%s rm_list_nr %d",
773 		 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
774 
775 	msk_owned_by_me(msk);
776 
777 	if (sk->sk_state == TCP_LISTEN)
778 		return;
779 
780 	if (!rm_list->nr)
781 		return;
782 
783 	if (list_empty(&msk->conn_list))
784 		return;
785 
786 	for (i = 0; i < rm_list->nr; i++) {
787 		bool removed = false;
788 
789 		list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
790 			struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
791 			int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
792 			u8 id = subflow->local_id;
793 
794 			if (rm_type == MPTCP_MIB_RMADDR)
795 				id = subflow->remote_id;
796 
797 			if (rm_list->ids[i] != id)
798 				continue;
799 
800 			pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
801 				 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
802 				 i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
803 			spin_unlock_bh(&msk->pm.lock);
804 			mptcp_subflow_shutdown(sk, ssk, how);
805 
806 			/* the following takes care of updating the subflows counter */
807 			mptcp_close_ssk(sk, ssk, subflow);
808 			spin_lock_bh(&msk->pm.lock);
809 
810 			removed = true;
811 			__MPTCP_INC_STATS(sock_net(sk), rm_type);
812 		}
813 		__set_bit(rm_list->ids[i], msk->pm.id_avail_bitmap);
814 		if (!removed)
815 			continue;
816 
817 		if (rm_type == MPTCP_MIB_RMADDR) {
818 			msk->pm.add_addr_accepted--;
819 			WRITE_ONCE(msk->pm.accept_addr, true);
820 		} else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
821 			msk->pm.local_addr_used--;
822 		}
823 	}
824 }
825 
826 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
827 {
828 	mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
829 }
830 
831 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
832 				     const struct mptcp_rm_list *rm_list)
833 {
834 	mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
835 }
836 
837 void mptcp_pm_nl_work(struct mptcp_sock *msk)
838 {
839 	struct mptcp_pm_data *pm = &msk->pm;
840 
841 	msk_owned_by_me(msk);
842 
843 	if (!(pm->status & MPTCP_PM_WORK_MASK))
844 		return;
845 
846 	spin_lock_bh(&msk->pm.lock);
847 
848 	pr_debug("msk=%p status=%x", msk, pm->status);
849 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
850 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
851 		mptcp_pm_nl_add_addr_received(msk);
852 	}
853 	if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
854 		pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
855 		mptcp_pm_nl_addr_send_ack(msk);
856 	}
857 	if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
858 		pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
859 		mptcp_pm_nl_rm_addr_received(msk);
860 	}
861 	if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
862 		pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
863 		mptcp_pm_nl_fully_established(msk);
864 	}
865 	if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
866 		pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
867 		mptcp_pm_nl_subflow_established(msk);
868 	}
869 
870 	spin_unlock_bh(&msk->pm.lock);
871 }
872 
873 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
874 {
875 	return (entry->flags &
876 		(MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
877 		MPTCP_PM_ADDR_FLAG_SIGNAL;
878 }
879 
880 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
881 					     struct mptcp_pm_addr_entry *entry)
882 {
883 	struct mptcp_pm_addr_entry *cur;
884 	unsigned int addr_max;
885 	int ret = -EINVAL;
886 
887 	spin_lock_bh(&pernet->lock);
888 	/* to keep the code simple, don't do IDR-like allocation for address ID,
889 	 * just bail when we exceed limits
890 	 */
891 	if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
892 		pernet->next_id = 1;
893 	if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
894 		goto out;
895 	if (test_bit(entry->addr.id, pernet->id_bitmap))
896 		goto out;
897 
898 	/* do not insert duplicate address, differentiate on port only
899 	 * singled addresses
900 	 */
901 	list_for_each_entry(cur, &pernet->local_addr_list, list) {
902 		if (addresses_equal(&cur->addr, &entry->addr,
903 				    address_use_port(entry) &&
904 				    address_use_port(cur)))
905 			goto out;
906 	}
907 
908 	if (!entry->addr.id) {
909 find_next:
910 		entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
911 						    MPTCP_PM_MAX_ADDR_ID + 1,
912 						    pernet->next_id);
913 		if (!entry->addr.id && pernet->next_id != 1) {
914 			pernet->next_id = 1;
915 			goto find_next;
916 		}
917 	}
918 
919 	if (!entry->addr.id)
920 		goto out;
921 
922 	__set_bit(entry->addr.id, pernet->id_bitmap);
923 	if (entry->addr.id > pernet->next_id)
924 		pernet->next_id = entry->addr.id;
925 
926 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
927 		addr_max = pernet->add_addr_signal_max;
928 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
929 	}
930 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
931 		addr_max = pernet->local_addr_max;
932 		WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
933 	}
934 
935 	pernet->addrs++;
936 	list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
937 	ret = entry->addr.id;
938 
939 out:
940 	spin_unlock_bh(&pernet->lock);
941 	return ret;
942 }
943 
944 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
945 					    struct mptcp_pm_addr_entry *entry)
946 {
947 	int addrlen = sizeof(struct sockaddr_in);
948 	struct sockaddr_storage addr;
949 	struct mptcp_sock *msk;
950 	struct socket *ssock;
951 	int backlog = 1024;
952 	int err;
953 
954 	err = sock_create_kern(sock_net(sk), entry->addr.family,
955 			       SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
956 	if (err)
957 		return err;
958 
959 	msk = mptcp_sk(entry->lsk->sk);
960 	if (!msk) {
961 		err = -EINVAL;
962 		goto out;
963 	}
964 
965 	ssock = __mptcp_nmpc_socket(msk);
966 	if (!ssock) {
967 		err = -EINVAL;
968 		goto out;
969 	}
970 
971 	mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
972 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
973 	if (entry->addr.family == AF_INET6)
974 		addrlen = sizeof(struct sockaddr_in6);
975 #endif
976 	err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
977 	if (err) {
978 		pr_warn("kernel_bind error, err=%d", err);
979 		goto out;
980 	}
981 
982 	err = kernel_listen(ssock, backlog);
983 	if (err) {
984 		pr_warn("kernel_listen error, err=%d", err);
985 		goto out;
986 	}
987 
988 	return 0;
989 
990 out:
991 	sock_release(entry->lsk);
992 	return err;
993 }
994 
995 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
996 {
997 	struct mptcp_pm_addr_entry *entry;
998 	struct mptcp_addr_info skc_local;
999 	struct mptcp_addr_info msk_local;
1000 	struct pm_nl_pernet *pernet;
1001 	int ret = -1;
1002 
1003 	if (WARN_ON_ONCE(!msk))
1004 		return -1;
1005 
1006 	/* The 0 ID mapping is defined by the first subflow, copied into the msk
1007 	 * addr
1008 	 */
1009 	local_address((struct sock_common *)msk, &msk_local);
1010 	local_address((struct sock_common *)skc, &skc_local);
1011 	if (addresses_equal(&msk_local, &skc_local, false))
1012 		return 0;
1013 
1014 	if (address_zero(&skc_local))
1015 		return 0;
1016 
1017 	pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
1018 
1019 	rcu_read_lock();
1020 	list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1021 		if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1022 			ret = entry->addr.id;
1023 			break;
1024 		}
1025 	}
1026 	rcu_read_unlock();
1027 	if (ret >= 0)
1028 		return ret;
1029 
1030 	/* address not found, add to local list */
1031 	entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1032 	if (!entry)
1033 		return -ENOMEM;
1034 
1035 	entry->addr = skc_local;
1036 	entry->addr.id = 0;
1037 	entry->addr.port = 0;
1038 	entry->ifindex = 0;
1039 	entry->flags = 0;
1040 	entry->lsk = NULL;
1041 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1042 	if (ret < 0)
1043 		kfree(entry);
1044 
1045 	return ret;
1046 }
1047 
1048 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
1049 {
1050 	struct mptcp_pm_data *pm = &msk->pm;
1051 	bool subflows;
1052 
1053 	subflows = !!mptcp_pm_get_subflows_max(msk);
1054 	WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
1055 		   !!mptcp_pm_get_add_addr_signal_max(msk));
1056 	WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
1057 	WRITE_ONCE(pm->accept_subflow, subflows);
1058 }
1059 
1060 #define MPTCP_PM_CMD_GRP_OFFSET       0
1061 #define MPTCP_PM_EV_GRP_OFFSET        1
1062 
1063 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1064 	[MPTCP_PM_CMD_GRP_OFFSET]	= { .name = MPTCP_PM_CMD_GRP_NAME, },
1065 	[MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1066 					    .flags = GENL_UNS_ADMIN_PERM,
1067 					  },
1068 };
1069 
1070 static const struct nla_policy
1071 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1072 	[MPTCP_PM_ADDR_ATTR_FAMILY]	= { .type	= NLA_U16,	},
1073 	[MPTCP_PM_ADDR_ATTR_ID]		= { .type	= NLA_U8,	},
1074 	[MPTCP_PM_ADDR_ATTR_ADDR4]	= { .type	= NLA_U32,	},
1075 	[MPTCP_PM_ADDR_ATTR_ADDR6]	=
1076 		NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1077 	[MPTCP_PM_ADDR_ATTR_PORT]	= { .type	= NLA_U16	},
1078 	[MPTCP_PM_ADDR_ATTR_FLAGS]	= { .type	= NLA_U32	},
1079 	[MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type	= NLA_S32	},
1080 };
1081 
1082 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1083 	[MPTCP_PM_ATTR_ADDR]		=
1084 					NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1085 	[MPTCP_PM_ATTR_RCV_ADD_ADDRS]	= { .type	= NLA_U32,	},
1086 	[MPTCP_PM_ATTR_SUBFLOWS]	= { .type	= NLA_U32,	},
1087 };
1088 
1089 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1090 {
1091 	struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1092 	struct sock *sk = (struct sock *)msk;
1093 	unsigned int active_max_loss_cnt;
1094 	struct net *net = sock_net(sk);
1095 	unsigned int stale_loss_cnt;
1096 	bool slow;
1097 
1098 	stale_loss_cnt = mptcp_stale_loss_cnt(net);
1099 	if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1100 		return;
1101 
1102 	/* look for another available subflow not in loss state */
1103 	active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1104 	mptcp_for_each_subflow(msk, iter) {
1105 		if (iter != subflow && mptcp_subflow_active(iter) &&
1106 		    iter->stale_count < active_max_loss_cnt) {
1107 			/* we have some alternatives, try to mark this subflow as idle ...*/
1108 			slow = lock_sock_fast(ssk);
1109 			if (!tcp_rtx_and_write_queues_empty(ssk)) {
1110 				subflow->stale = 1;
1111 				__mptcp_retransmit_pending_data(sk);
1112 				MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1113 			}
1114 			unlock_sock_fast(ssk, slow);
1115 
1116 			/* always try to push the pending data regarless of re-injections:
1117 			 * we can possibly use backup subflows now, and subflow selection
1118 			 * is cheap under the msk socket lock
1119 			 */
1120 			__mptcp_push_pending(sk, 0);
1121 			return;
1122 		}
1123 	}
1124 }
1125 
1126 static int mptcp_pm_family_to_addr(int family)
1127 {
1128 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1129 	if (family == AF_INET6)
1130 		return MPTCP_PM_ADDR_ATTR_ADDR6;
1131 #endif
1132 	return MPTCP_PM_ADDR_ATTR_ADDR4;
1133 }
1134 
1135 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1136 			       bool require_family,
1137 			       struct mptcp_pm_addr_entry *entry)
1138 {
1139 	struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1140 	int err, addr_addr;
1141 
1142 	if (!attr) {
1143 		GENL_SET_ERR_MSG(info, "missing address info");
1144 		return -EINVAL;
1145 	}
1146 
1147 	/* no validation needed - was already done via nested policy */
1148 	err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1149 					  mptcp_pm_addr_policy, info->extack);
1150 	if (err)
1151 		return err;
1152 
1153 	memset(entry, 0, sizeof(*entry));
1154 	if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1155 		if (!require_family)
1156 			goto skip_family;
1157 
1158 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1159 				    "missing family");
1160 		return -EINVAL;
1161 	}
1162 
1163 	entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1164 	if (entry->addr.family != AF_INET
1165 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1166 	    && entry->addr.family != AF_INET6
1167 #endif
1168 	    ) {
1169 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1170 				    "unknown address family");
1171 		return -EINVAL;
1172 	}
1173 	addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1174 	if (!tb[addr_addr]) {
1175 		NL_SET_ERR_MSG_ATTR(info->extack, attr,
1176 				    "missing address data");
1177 		return -EINVAL;
1178 	}
1179 
1180 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1181 	if (entry->addr.family == AF_INET6)
1182 		entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1183 	else
1184 #endif
1185 		entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1186 
1187 skip_family:
1188 	if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1189 		u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1190 
1191 		entry->ifindex = val;
1192 	}
1193 
1194 	if (tb[MPTCP_PM_ADDR_ATTR_ID])
1195 		entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1196 
1197 	if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1198 		entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1199 
1200 	if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1201 		if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1202 			NL_SET_ERR_MSG_ATTR(info->extack, attr,
1203 					    "flags must have signal when using port");
1204 			return -EINVAL;
1205 		}
1206 		entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1207 	}
1208 
1209 	return 0;
1210 }
1211 
1212 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1213 {
1214 	return net_generic(genl_info_net(info), pm_nl_pernet_id);
1215 }
1216 
1217 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1218 {
1219 	struct mptcp_sock *msk;
1220 	long s_slot = 0, s_num = 0;
1221 
1222 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1223 		struct sock *sk = (struct sock *)msk;
1224 
1225 		if (!READ_ONCE(msk->fully_established))
1226 			goto next;
1227 
1228 		lock_sock(sk);
1229 		spin_lock_bh(&msk->pm.lock);
1230 		mptcp_pm_create_subflow_or_signal_addr(msk);
1231 		spin_unlock_bh(&msk->pm.lock);
1232 		release_sock(sk);
1233 
1234 next:
1235 		sock_put(sk);
1236 		cond_resched();
1237 	}
1238 
1239 	return 0;
1240 }
1241 
1242 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1243 {
1244 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1245 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1246 	struct mptcp_pm_addr_entry addr, *entry;
1247 	int ret;
1248 
1249 	ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1250 	if (ret < 0)
1251 		return ret;
1252 
1253 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1254 	if (!entry) {
1255 		GENL_SET_ERR_MSG(info, "can't allocate addr");
1256 		return -ENOMEM;
1257 	}
1258 
1259 	*entry = addr;
1260 	if (entry->addr.port) {
1261 		ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1262 		if (ret) {
1263 			GENL_SET_ERR_MSG(info, "create listen socket error");
1264 			kfree(entry);
1265 			return ret;
1266 		}
1267 	}
1268 	ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1269 	if (ret < 0) {
1270 		GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1271 		if (entry->lsk)
1272 			sock_release(entry->lsk);
1273 		kfree(entry);
1274 		return ret;
1275 	}
1276 
1277 	mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1278 
1279 	return 0;
1280 }
1281 
1282 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1283 					 u8 *flags, int *ifindex)
1284 {
1285 	struct mptcp_pm_addr_entry *entry;
1286 
1287 	*flags = 0;
1288 	*ifindex = 0;
1289 
1290 	if (id) {
1291 		rcu_read_lock();
1292 		entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1293 		if (entry) {
1294 			*flags = entry->flags;
1295 			*ifindex = entry->ifindex;
1296 		}
1297 		rcu_read_unlock();
1298 	}
1299 
1300 	return 0;
1301 }
1302 
1303 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1304 				      struct mptcp_addr_info *addr)
1305 {
1306 	struct mptcp_pm_add_entry *entry;
1307 
1308 	entry = mptcp_pm_del_add_timer(msk, addr, false);
1309 	if (entry) {
1310 		list_del(&entry->list);
1311 		kfree(entry);
1312 		return true;
1313 	}
1314 
1315 	return false;
1316 }
1317 
1318 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1319 				      struct mptcp_addr_info *addr,
1320 				      bool force)
1321 {
1322 	struct mptcp_rm_list list = { .nr = 0 };
1323 	bool ret;
1324 
1325 	list.ids[list.nr++] = addr->id;
1326 
1327 	ret = remove_anno_list_by_saddr(msk, addr);
1328 	if (ret || force) {
1329 		spin_lock_bh(&msk->pm.lock);
1330 		mptcp_pm_remove_addr(msk, &list);
1331 		spin_unlock_bh(&msk->pm.lock);
1332 	}
1333 	return ret;
1334 }
1335 
1336 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1337 						   struct mptcp_addr_info *addr)
1338 {
1339 	struct mptcp_sock *msk;
1340 	long s_slot = 0, s_num = 0;
1341 	struct mptcp_rm_list list = { .nr = 0 };
1342 
1343 	pr_debug("remove_id=%d", addr->id);
1344 
1345 	list.ids[list.nr++] = addr->id;
1346 
1347 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1348 		struct sock *sk = (struct sock *)msk;
1349 		bool remove_subflow;
1350 
1351 		if (list_empty(&msk->conn_list)) {
1352 			mptcp_pm_remove_anno_addr(msk, addr, false);
1353 			goto next;
1354 		}
1355 
1356 		lock_sock(sk);
1357 		remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1358 		mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1359 		if (remove_subflow)
1360 			mptcp_pm_remove_subflow(msk, &list);
1361 		release_sock(sk);
1362 
1363 next:
1364 		sock_put(sk);
1365 		cond_resched();
1366 	}
1367 
1368 	return 0;
1369 }
1370 
1371 /* caller must ensure the RCU grace period is already elapsed */
1372 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1373 {
1374 	if (entry->lsk)
1375 		sock_release(entry->lsk);
1376 	kfree(entry);
1377 }
1378 
1379 static int mptcp_nl_remove_id_zero_address(struct net *net,
1380 					   struct mptcp_addr_info *addr)
1381 {
1382 	struct mptcp_rm_list list = { .nr = 0 };
1383 	long s_slot = 0, s_num = 0;
1384 	struct mptcp_sock *msk;
1385 
1386 	list.ids[list.nr++] = 0;
1387 
1388 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1389 		struct sock *sk = (struct sock *)msk;
1390 		struct mptcp_addr_info msk_local;
1391 
1392 		if (list_empty(&msk->conn_list))
1393 			goto next;
1394 
1395 		local_address((struct sock_common *)msk, &msk_local);
1396 		if (!addresses_equal(&msk_local, addr, addr->port))
1397 			goto next;
1398 
1399 		lock_sock(sk);
1400 		spin_lock_bh(&msk->pm.lock);
1401 		mptcp_pm_remove_addr(msk, &list);
1402 		mptcp_pm_nl_rm_subflow_received(msk, &list);
1403 		spin_unlock_bh(&msk->pm.lock);
1404 		release_sock(sk);
1405 
1406 next:
1407 		sock_put(sk);
1408 		cond_resched();
1409 	}
1410 
1411 	return 0;
1412 }
1413 
1414 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1415 {
1416 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1417 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1418 	struct mptcp_pm_addr_entry addr, *entry;
1419 	unsigned int addr_max;
1420 	int ret;
1421 
1422 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1423 	if (ret < 0)
1424 		return ret;
1425 
1426 	/* the zero id address is special: the first address used by the msk
1427 	 * always gets such an id, so different subflows can have different zero
1428 	 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1429 	 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1430 	 */
1431 	if (addr.addr.id == 0)
1432 		return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1433 
1434 	spin_lock_bh(&pernet->lock);
1435 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1436 	if (!entry) {
1437 		GENL_SET_ERR_MSG(info, "address not found");
1438 		spin_unlock_bh(&pernet->lock);
1439 		return -EINVAL;
1440 	}
1441 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1442 		addr_max = pernet->add_addr_signal_max;
1443 		WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1444 	}
1445 	if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1446 		addr_max = pernet->local_addr_max;
1447 		WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1448 	}
1449 
1450 	pernet->addrs--;
1451 	list_del_rcu(&entry->list);
1452 	__clear_bit(entry->addr.id, pernet->id_bitmap);
1453 	spin_unlock_bh(&pernet->lock);
1454 
1455 	mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1456 	synchronize_rcu();
1457 	__mptcp_pm_release_addr_entry(entry);
1458 
1459 	return ret;
1460 }
1461 
1462 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1463 					       struct list_head *rm_list)
1464 {
1465 	struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1466 	struct mptcp_pm_addr_entry *entry;
1467 
1468 	list_for_each_entry(entry, rm_list, list) {
1469 		if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1470 		    alist.nr < MPTCP_RM_IDS_MAX &&
1471 		    slist.nr < MPTCP_RM_IDS_MAX) {
1472 			alist.ids[alist.nr++] = entry->addr.id;
1473 			slist.ids[slist.nr++] = entry->addr.id;
1474 		} else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1475 			 alist.nr < MPTCP_RM_IDS_MAX) {
1476 			alist.ids[alist.nr++] = entry->addr.id;
1477 		}
1478 	}
1479 
1480 	if (alist.nr) {
1481 		spin_lock_bh(&msk->pm.lock);
1482 		mptcp_pm_remove_addr(msk, &alist);
1483 		spin_unlock_bh(&msk->pm.lock);
1484 	}
1485 	if (slist.nr)
1486 		mptcp_pm_remove_subflow(msk, &slist);
1487 }
1488 
1489 static void mptcp_nl_remove_addrs_list(struct net *net,
1490 				       struct list_head *rm_list)
1491 {
1492 	long s_slot = 0, s_num = 0;
1493 	struct mptcp_sock *msk;
1494 
1495 	if (list_empty(rm_list))
1496 		return;
1497 
1498 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1499 		struct sock *sk = (struct sock *)msk;
1500 
1501 		lock_sock(sk);
1502 		mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1503 		release_sock(sk);
1504 
1505 		sock_put(sk);
1506 		cond_resched();
1507 	}
1508 }
1509 
1510 /* caller must ensure the RCU grace period is already elapsed */
1511 static void __flush_addrs(struct list_head *list)
1512 {
1513 	while (!list_empty(list)) {
1514 		struct mptcp_pm_addr_entry *cur;
1515 
1516 		cur = list_entry(list->next,
1517 				 struct mptcp_pm_addr_entry, list);
1518 		list_del_rcu(&cur->list);
1519 		__mptcp_pm_release_addr_entry(cur);
1520 	}
1521 }
1522 
1523 static void __reset_counters(struct pm_nl_pernet *pernet)
1524 {
1525 	WRITE_ONCE(pernet->add_addr_signal_max, 0);
1526 	WRITE_ONCE(pernet->add_addr_accept_max, 0);
1527 	WRITE_ONCE(pernet->local_addr_max, 0);
1528 	pernet->addrs = 0;
1529 }
1530 
1531 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1532 {
1533 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1534 	LIST_HEAD(free_list);
1535 
1536 	spin_lock_bh(&pernet->lock);
1537 	list_splice_init(&pernet->local_addr_list, &free_list);
1538 	__reset_counters(pernet);
1539 	pernet->next_id = 1;
1540 	bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1541 	spin_unlock_bh(&pernet->lock);
1542 	mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1543 	synchronize_rcu();
1544 	__flush_addrs(&free_list);
1545 	return 0;
1546 }
1547 
1548 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1549 			      struct mptcp_pm_addr_entry *entry)
1550 {
1551 	struct mptcp_addr_info *addr = &entry->addr;
1552 	struct nlattr *attr;
1553 
1554 	attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1555 	if (!attr)
1556 		return -EMSGSIZE;
1557 
1558 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1559 		goto nla_put_failure;
1560 	if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1561 		goto nla_put_failure;
1562 	if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1563 		goto nla_put_failure;
1564 	if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1565 		goto nla_put_failure;
1566 	if (entry->ifindex &&
1567 	    nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1568 		goto nla_put_failure;
1569 
1570 	if (addr->family == AF_INET &&
1571 	    nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1572 			    addr->addr.s_addr))
1573 		goto nla_put_failure;
1574 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1575 	else if (addr->family == AF_INET6 &&
1576 		 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1577 		goto nla_put_failure;
1578 #endif
1579 	nla_nest_end(skb, attr);
1580 	return 0;
1581 
1582 nla_put_failure:
1583 	nla_nest_cancel(skb, attr);
1584 	return -EMSGSIZE;
1585 }
1586 
1587 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1588 {
1589 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1590 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1591 	struct mptcp_pm_addr_entry addr, *entry;
1592 	struct sk_buff *msg;
1593 	void *reply;
1594 	int ret;
1595 
1596 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1597 	if (ret < 0)
1598 		return ret;
1599 
1600 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1601 	if (!msg)
1602 		return -ENOMEM;
1603 
1604 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1605 				  info->genlhdr->cmd);
1606 	if (!reply) {
1607 		GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1608 		ret = -EMSGSIZE;
1609 		goto fail;
1610 	}
1611 
1612 	spin_lock_bh(&pernet->lock);
1613 	entry = __lookup_addr_by_id(pernet, addr.addr.id);
1614 	if (!entry) {
1615 		GENL_SET_ERR_MSG(info, "address not found");
1616 		ret = -EINVAL;
1617 		goto unlock_fail;
1618 	}
1619 
1620 	ret = mptcp_nl_fill_addr(msg, entry);
1621 	if (ret)
1622 		goto unlock_fail;
1623 
1624 	genlmsg_end(msg, reply);
1625 	ret = genlmsg_reply(msg, info);
1626 	spin_unlock_bh(&pernet->lock);
1627 	return ret;
1628 
1629 unlock_fail:
1630 	spin_unlock_bh(&pernet->lock);
1631 
1632 fail:
1633 	nlmsg_free(msg);
1634 	return ret;
1635 }
1636 
1637 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1638 				   struct netlink_callback *cb)
1639 {
1640 	struct net *net = sock_net(msg->sk);
1641 	struct mptcp_pm_addr_entry *entry;
1642 	struct pm_nl_pernet *pernet;
1643 	int id = cb->args[0];
1644 	void *hdr;
1645 	int i;
1646 
1647 	pernet = net_generic(net, pm_nl_pernet_id);
1648 
1649 	spin_lock_bh(&pernet->lock);
1650 	for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1651 		if (test_bit(i, pernet->id_bitmap)) {
1652 			entry = __lookup_addr_by_id(pernet, i);
1653 			if (!entry)
1654 				break;
1655 
1656 			if (entry->addr.id <= id)
1657 				continue;
1658 
1659 			hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1660 					  cb->nlh->nlmsg_seq, &mptcp_genl_family,
1661 					  NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1662 			if (!hdr)
1663 				break;
1664 
1665 			if (mptcp_nl_fill_addr(msg, entry) < 0) {
1666 				genlmsg_cancel(msg, hdr);
1667 				break;
1668 			}
1669 
1670 			id = entry->addr.id;
1671 			genlmsg_end(msg, hdr);
1672 		}
1673 	}
1674 	spin_unlock_bh(&pernet->lock);
1675 
1676 	cb->args[0] = id;
1677 	return msg->len;
1678 }
1679 
1680 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1681 {
1682 	struct nlattr *attr = info->attrs[id];
1683 
1684 	if (!attr)
1685 		return 0;
1686 
1687 	*limit = nla_get_u32(attr);
1688 	if (*limit > MPTCP_PM_ADDR_MAX) {
1689 		GENL_SET_ERR_MSG(info, "limit greater than maximum");
1690 		return -EINVAL;
1691 	}
1692 	return 0;
1693 }
1694 
1695 static int
1696 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1697 {
1698 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1699 	unsigned int rcv_addrs, subflows;
1700 	int ret;
1701 
1702 	spin_lock_bh(&pernet->lock);
1703 	rcv_addrs = pernet->add_addr_accept_max;
1704 	ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1705 	if (ret)
1706 		goto unlock;
1707 
1708 	subflows = pernet->subflows_max;
1709 	ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1710 	if (ret)
1711 		goto unlock;
1712 
1713 	WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1714 	WRITE_ONCE(pernet->subflows_max, subflows);
1715 
1716 unlock:
1717 	spin_unlock_bh(&pernet->lock);
1718 	return ret;
1719 }
1720 
1721 static int
1722 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1723 {
1724 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1725 	struct sk_buff *msg;
1726 	void *reply;
1727 
1728 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1729 	if (!msg)
1730 		return -ENOMEM;
1731 
1732 	reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1733 				  MPTCP_PM_CMD_GET_LIMITS);
1734 	if (!reply)
1735 		goto fail;
1736 
1737 	if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1738 			READ_ONCE(pernet->add_addr_accept_max)))
1739 		goto fail;
1740 
1741 	if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1742 			READ_ONCE(pernet->subflows_max)))
1743 		goto fail;
1744 
1745 	genlmsg_end(msg, reply);
1746 	return genlmsg_reply(msg, info);
1747 
1748 fail:
1749 	GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1750 	nlmsg_free(msg);
1751 	return -EMSGSIZE;
1752 }
1753 
1754 static int mptcp_nl_addr_backup(struct net *net,
1755 				struct mptcp_addr_info *addr,
1756 				u8 bkup)
1757 {
1758 	long s_slot = 0, s_num = 0;
1759 	struct mptcp_sock *msk;
1760 	int ret = -EINVAL;
1761 
1762 	while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1763 		struct sock *sk = (struct sock *)msk;
1764 
1765 		if (list_empty(&msk->conn_list))
1766 			goto next;
1767 
1768 		lock_sock(sk);
1769 		spin_lock_bh(&msk->pm.lock);
1770 		ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1771 		spin_unlock_bh(&msk->pm.lock);
1772 		release_sock(sk);
1773 
1774 next:
1775 		sock_put(sk);
1776 		cond_resched();
1777 	}
1778 
1779 	return ret;
1780 }
1781 
1782 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1783 {
1784 	struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1785 	struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1786 	struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1787 	struct net *net = sock_net(skb->sk);
1788 	u8 bkup = 0, lookup_by_id = 0;
1789 	int ret;
1790 
1791 	ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1792 	if (ret < 0)
1793 		return ret;
1794 
1795 	if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1796 		bkup = 1;
1797 	if (addr.addr.family == AF_UNSPEC) {
1798 		lookup_by_id = 1;
1799 		if (!addr.addr.id)
1800 			return -EOPNOTSUPP;
1801 	}
1802 
1803 	spin_lock_bh(&pernet->lock);
1804 	entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1805 	if (!entry) {
1806 		spin_unlock_bh(&pernet->lock);
1807 		return -EINVAL;
1808 	}
1809 
1810 	if (bkup)
1811 		entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1812 	else
1813 		entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1814 	addr = *entry;
1815 	spin_unlock_bh(&pernet->lock);
1816 
1817 	mptcp_nl_addr_backup(net, &addr.addr, bkup);
1818 	return 0;
1819 }
1820 
1821 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1822 {
1823 	genlmsg_multicast_netns(&mptcp_genl_family, net,
1824 				nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1825 }
1826 
1827 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1828 {
1829 	const struct inet_sock *issk = inet_sk(ssk);
1830 	const struct mptcp_subflow_context *sf;
1831 
1832 	if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1833 		return -EMSGSIZE;
1834 
1835 	switch (ssk->sk_family) {
1836 	case AF_INET:
1837 		if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1838 			return -EMSGSIZE;
1839 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1840 			return -EMSGSIZE;
1841 		break;
1842 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1843 	case AF_INET6: {
1844 		const struct ipv6_pinfo *np = inet6_sk(ssk);
1845 
1846 		if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1847 			return -EMSGSIZE;
1848 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1849 			return -EMSGSIZE;
1850 		break;
1851 	}
1852 #endif
1853 	default:
1854 		WARN_ON_ONCE(1);
1855 		return -EMSGSIZE;
1856 	}
1857 
1858 	if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1859 		return -EMSGSIZE;
1860 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1861 		return -EMSGSIZE;
1862 
1863 	sf = mptcp_subflow_ctx(ssk);
1864 	if (WARN_ON_ONCE(!sf))
1865 		return -EINVAL;
1866 
1867 	if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1868 		return -EMSGSIZE;
1869 
1870 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1871 		return -EMSGSIZE;
1872 
1873 	return 0;
1874 }
1875 
1876 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1877 					 const struct mptcp_sock *msk,
1878 					 const struct sock *ssk)
1879 {
1880 	const struct sock *sk = (const struct sock *)msk;
1881 	const struct mptcp_subflow_context *sf;
1882 	u8 sk_err;
1883 
1884 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1885 		return -EMSGSIZE;
1886 
1887 	if (mptcp_event_add_subflow(skb, ssk))
1888 		return -EMSGSIZE;
1889 
1890 	sf = mptcp_subflow_ctx(ssk);
1891 	if (WARN_ON_ONCE(!sf))
1892 		return -EINVAL;
1893 
1894 	if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1895 		return -EMSGSIZE;
1896 
1897 	if (ssk->sk_bound_dev_if &&
1898 	    nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1899 		return -EMSGSIZE;
1900 
1901 	sk_err = ssk->sk_err;
1902 	if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1903 	    nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1904 		return -EMSGSIZE;
1905 
1906 	return 0;
1907 }
1908 
1909 static int mptcp_event_sub_established(struct sk_buff *skb,
1910 				       const struct mptcp_sock *msk,
1911 				       const struct sock *ssk)
1912 {
1913 	return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1914 }
1915 
1916 static int mptcp_event_sub_closed(struct sk_buff *skb,
1917 				  const struct mptcp_sock *msk,
1918 				  const struct sock *ssk)
1919 {
1920 	const struct mptcp_subflow_context *sf;
1921 
1922 	if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1923 		return -EMSGSIZE;
1924 
1925 	sf = mptcp_subflow_ctx(ssk);
1926 	if (!sf->reset_seen)
1927 		return 0;
1928 
1929 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1930 		return -EMSGSIZE;
1931 
1932 	if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1933 		return -EMSGSIZE;
1934 
1935 	return 0;
1936 }
1937 
1938 static int mptcp_event_created(struct sk_buff *skb,
1939 			       const struct mptcp_sock *msk,
1940 			       const struct sock *ssk)
1941 {
1942 	int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1943 
1944 	if (err)
1945 		return err;
1946 
1947 	return mptcp_event_add_subflow(skb, ssk);
1948 }
1949 
1950 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1951 {
1952 	struct net *net = sock_net((const struct sock *)msk);
1953 	struct nlmsghdr *nlh;
1954 	struct sk_buff *skb;
1955 
1956 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1957 		return;
1958 
1959 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1960 	if (!skb)
1961 		return;
1962 
1963 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1964 	if (!nlh)
1965 		goto nla_put_failure;
1966 
1967 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1968 		goto nla_put_failure;
1969 
1970 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1971 		goto nla_put_failure;
1972 
1973 	genlmsg_end(skb, nlh);
1974 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1975 	return;
1976 
1977 nla_put_failure:
1978 	kfree_skb(skb);
1979 }
1980 
1981 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1982 				const struct mptcp_addr_info *info)
1983 {
1984 	struct net *net = sock_net((const struct sock *)msk);
1985 	struct nlmsghdr *nlh;
1986 	struct sk_buff *skb;
1987 
1988 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1989 		return;
1990 
1991 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1992 	if (!skb)
1993 		return;
1994 
1995 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1996 			  MPTCP_EVENT_ANNOUNCED);
1997 	if (!nlh)
1998 		goto nla_put_failure;
1999 
2000 	if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2001 		goto nla_put_failure;
2002 
2003 	if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2004 		goto nla_put_failure;
2005 
2006 	if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
2007 		goto nla_put_failure;
2008 
2009 	switch (info->family) {
2010 	case AF_INET:
2011 		if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2012 			goto nla_put_failure;
2013 		break;
2014 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2015 	case AF_INET6:
2016 		if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2017 			goto nla_put_failure;
2018 		break;
2019 #endif
2020 	default:
2021 		WARN_ON_ONCE(1);
2022 		goto nla_put_failure;
2023 	}
2024 
2025 	genlmsg_end(skb, nlh);
2026 	mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2027 	return;
2028 
2029 nla_put_failure:
2030 	kfree_skb(skb);
2031 }
2032 
2033 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2034 		 const struct sock *ssk, gfp_t gfp)
2035 {
2036 	struct net *net = sock_net((const struct sock *)msk);
2037 	struct nlmsghdr *nlh;
2038 	struct sk_buff *skb;
2039 
2040 	if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2041 		return;
2042 
2043 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2044 	if (!skb)
2045 		return;
2046 
2047 	nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2048 	if (!nlh)
2049 		goto nla_put_failure;
2050 
2051 	switch (type) {
2052 	case MPTCP_EVENT_UNSPEC:
2053 		WARN_ON_ONCE(1);
2054 		break;
2055 	case MPTCP_EVENT_CREATED:
2056 	case MPTCP_EVENT_ESTABLISHED:
2057 		if (mptcp_event_created(skb, msk, ssk) < 0)
2058 			goto nla_put_failure;
2059 		break;
2060 	case MPTCP_EVENT_CLOSED:
2061 		if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2062 			goto nla_put_failure;
2063 		break;
2064 	case MPTCP_EVENT_ANNOUNCED:
2065 	case MPTCP_EVENT_REMOVED:
2066 		/* call mptcp_event_addr_announced()/removed instead */
2067 		WARN_ON_ONCE(1);
2068 		break;
2069 	case MPTCP_EVENT_SUB_ESTABLISHED:
2070 	case MPTCP_EVENT_SUB_PRIORITY:
2071 		if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2072 			goto nla_put_failure;
2073 		break;
2074 	case MPTCP_EVENT_SUB_CLOSED:
2075 		if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2076 			goto nla_put_failure;
2077 		break;
2078 	}
2079 
2080 	genlmsg_end(skb, nlh);
2081 	mptcp_nl_mcast_send(net, skb, gfp);
2082 	return;
2083 
2084 nla_put_failure:
2085 	kfree_skb(skb);
2086 }
2087 
2088 static const struct genl_small_ops mptcp_pm_ops[] = {
2089 	{
2090 		.cmd    = MPTCP_PM_CMD_ADD_ADDR,
2091 		.doit   = mptcp_nl_cmd_add_addr,
2092 		.flags  = GENL_ADMIN_PERM,
2093 	},
2094 	{
2095 		.cmd    = MPTCP_PM_CMD_DEL_ADDR,
2096 		.doit   = mptcp_nl_cmd_del_addr,
2097 		.flags  = GENL_ADMIN_PERM,
2098 	},
2099 	{
2100 		.cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2101 		.doit   = mptcp_nl_cmd_flush_addrs,
2102 		.flags  = GENL_ADMIN_PERM,
2103 	},
2104 	{
2105 		.cmd    = MPTCP_PM_CMD_GET_ADDR,
2106 		.doit   = mptcp_nl_cmd_get_addr,
2107 		.dumpit   = mptcp_nl_cmd_dump_addrs,
2108 	},
2109 	{
2110 		.cmd    = MPTCP_PM_CMD_SET_LIMITS,
2111 		.doit   = mptcp_nl_cmd_set_limits,
2112 		.flags  = GENL_ADMIN_PERM,
2113 	},
2114 	{
2115 		.cmd    = MPTCP_PM_CMD_GET_LIMITS,
2116 		.doit   = mptcp_nl_cmd_get_limits,
2117 	},
2118 	{
2119 		.cmd    = MPTCP_PM_CMD_SET_FLAGS,
2120 		.doit   = mptcp_nl_cmd_set_flags,
2121 		.flags  = GENL_ADMIN_PERM,
2122 	},
2123 };
2124 
2125 static struct genl_family mptcp_genl_family __ro_after_init = {
2126 	.name		= MPTCP_PM_NAME,
2127 	.version	= MPTCP_PM_VER,
2128 	.maxattr	= MPTCP_PM_ATTR_MAX,
2129 	.policy		= mptcp_pm_policy,
2130 	.netnsok	= true,
2131 	.module		= THIS_MODULE,
2132 	.small_ops	= mptcp_pm_ops,
2133 	.n_small_ops	= ARRAY_SIZE(mptcp_pm_ops),
2134 	.mcgrps		= mptcp_pm_mcgrps,
2135 	.n_mcgrps	= ARRAY_SIZE(mptcp_pm_mcgrps),
2136 };
2137 
2138 static int __net_init pm_nl_init_net(struct net *net)
2139 {
2140 	struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2141 
2142 	INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2143 
2144 	/* Cit. 2 subflows ought to be enough for anybody. */
2145 	pernet->subflows_max = 2;
2146 	pernet->next_id = 1;
2147 	pernet->stale_loss_cnt = 4;
2148 	spin_lock_init(&pernet->lock);
2149 
2150 	/* No need to initialize other pernet fields, the struct is zeroed at
2151 	 * allocation time.
2152 	 */
2153 
2154 	return 0;
2155 }
2156 
2157 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2158 {
2159 	struct net *net;
2160 
2161 	list_for_each_entry(net, net_list, exit_list) {
2162 		struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2163 
2164 		/* net is removed from namespace list, can't race with
2165 		 * other modifiers, also netns core already waited for a
2166 		 * RCU grace period.
2167 		 */
2168 		__flush_addrs(&pernet->local_addr_list);
2169 	}
2170 }
2171 
2172 static struct pernet_operations mptcp_pm_pernet_ops = {
2173 	.init = pm_nl_init_net,
2174 	.exit_batch = pm_nl_exit_net,
2175 	.id = &pm_nl_pernet_id,
2176 	.size = sizeof(struct pm_nl_pernet),
2177 };
2178 
2179 void __init mptcp_pm_nl_init(void)
2180 {
2181 	if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2182 		panic("Failed to register MPTCP PM pernet subsystem.\n");
2183 
2184 	if (genl_register_family(&mptcp_genl_family))
2185 		panic("Failed to register MPTCP PM netlink family\n");
2186 }
2187