1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (c) 2020, Red Hat, Inc.
7 #define pr_fmt(fmt) "MPTCP: " fmt
9 #include <linux/inet.h>
10 #include <linux/kernel.h>
12 #include <net/netns/generic.h>
13 #include <net/mptcp.h>
14 #include <net/genetlink.h>
15 #include <uapi/linux/mptcp.h>
20 /* forward declaration */
21 static struct genl_family mptcp_genl_family;
23 static int pm_nl_pernet_id;
25 struct mptcp_pm_add_entry {
26 struct list_head list;
27 struct mptcp_addr_info addr;
28 struct timer_list add_timer;
29 struct mptcp_sock *sock;
34 /* protects pernet updates */
36 struct list_head local_addr_list;
38 unsigned int stale_loss_cnt;
39 unsigned int add_addr_signal_max;
40 unsigned int add_addr_accept_max;
41 unsigned int local_addr_max;
42 unsigned int subflows_max;
44 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
47 #define MPTCP_PM_ADDR_MAX 8
48 #define ADD_ADDR_RETRANS_MAX 3
50 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net)
52 return net_generic(net, pm_nl_pernet_id);
55 static struct pm_nl_pernet *
56 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk)
58 return pm_nl_get_pernet(sock_net((struct sock *)msk));
61 bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
62 const struct mptcp_addr_info *b, bool use_port)
64 bool addr_equals = false;
66 if (a->family == b->family) {
67 if (a->family == AF_INET)
68 addr_equals = a->addr.s_addr == b->addr.s_addr;
69 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
71 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
72 } else if (a->family == AF_INET) {
73 if (ipv6_addr_v4mapped(&b->addr6))
74 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
75 } else if (b->family == AF_INET) {
76 if (ipv6_addr_v4mapped(&a->addr6))
77 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
86 return a->port == b->port;
89 static void local_address(const struct sock_common *skc,
90 struct mptcp_addr_info *addr)
92 addr->family = skc->skc_family;
93 addr->port = htons(skc->skc_num);
94 if (addr->family == AF_INET)
95 addr->addr.s_addr = skc->skc_rcv_saddr;
96 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
97 else if (addr->family == AF_INET6)
98 addr->addr6 = skc->skc_v6_rcv_saddr;
102 static void remote_address(const struct sock_common *skc,
103 struct mptcp_addr_info *addr)
105 addr->family = skc->skc_family;
106 addr->port = skc->skc_dport;
107 if (addr->family == AF_INET)
108 addr->addr.s_addr = skc->skc_daddr;
109 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
110 else if (addr->family == AF_INET6)
111 addr->addr6 = skc->skc_v6_daddr;
115 static bool lookup_subflow_by_saddr(const struct list_head *list,
116 const struct mptcp_addr_info *saddr)
118 struct mptcp_subflow_context *subflow;
119 struct mptcp_addr_info cur;
120 struct sock_common *skc;
122 list_for_each_entry(subflow, list, node) {
123 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
125 local_address(skc, &cur);
126 if (mptcp_addresses_equal(&cur, saddr, saddr->port))
133 static bool lookup_subflow_by_daddr(const struct list_head *list,
134 const struct mptcp_addr_info *daddr)
136 struct mptcp_subflow_context *subflow;
137 struct mptcp_addr_info cur;
138 struct sock_common *skc;
140 list_for_each_entry(subflow, list, node) {
141 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
143 remote_address(skc, &cur);
144 if (mptcp_addresses_equal(&cur, daddr, daddr->port))
151 static struct mptcp_pm_addr_entry *
152 select_local_address(const struct pm_nl_pernet *pernet,
153 const struct mptcp_sock *msk)
155 const struct sock *sk = (const struct sock *)msk;
156 struct mptcp_pm_addr_entry *entry, *ret = NULL;
158 msk_owned_by_me(msk);
161 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
162 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
165 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
168 if (entry->addr.family != sk->sk_family) {
169 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
170 if ((entry->addr.family == AF_INET &&
171 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
172 (sk->sk_family == AF_INET &&
173 !ipv6_addr_v4mapped(&entry->addr.addr6)))
185 static struct mptcp_pm_addr_entry *
186 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk)
188 struct mptcp_pm_addr_entry *entry, *ret = NULL;
191 /* do not keep any additional per socket state, just signal
192 * the address list in order.
193 * Note: removal from the local address list during the msk life-cycle
194 * can lead to additional addresses not being announced.
196 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
197 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
200 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
210 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk)
212 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
214 return READ_ONCE(pernet->add_addr_signal_max);
216 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
218 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk)
220 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
222 return READ_ONCE(pernet->add_addr_accept_max);
224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
226 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk)
228 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
230 return READ_ONCE(pernet->subflows_max);
232 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
234 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk)
236 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
238 return READ_ONCE(pernet->local_addr_max);
240 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
242 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
244 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
246 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
247 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
248 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
249 WRITE_ONCE(msk->pm.work_pending, false);
255 struct mptcp_pm_add_entry *
256 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
257 const struct mptcp_addr_info *addr)
259 struct mptcp_pm_add_entry *entry;
261 lockdep_assert_held(&msk->pm.lock);
263 list_for_each_entry(entry, &msk->pm.anno_list, list) {
264 if (mptcp_addresses_equal(&entry->addr, addr, true))
271 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
273 struct mptcp_pm_add_entry *entry;
274 struct mptcp_addr_info saddr;
277 local_address((struct sock_common *)sk, &saddr);
279 spin_lock_bh(&msk->pm.lock);
280 list_for_each_entry(entry, &msk->pm.anno_list, list) {
281 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) {
288 spin_unlock_bh(&msk->pm.lock);
292 static void mptcp_pm_add_timer(struct timer_list *timer)
294 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
295 struct mptcp_sock *msk = entry->sock;
296 struct sock *sk = (struct sock *)msk;
298 pr_debug("msk=%p", msk);
303 if (inet_sk_state_load(sk) == TCP_CLOSE)
309 if (mptcp_pm_should_add_signal_addr(msk)) {
310 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
314 spin_lock_bh(&msk->pm.lock);
316 if (!mptcp_pm_should_add_signal_addr(msk)) {
317 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
318 mptcp_pm_announce_addr(msk, &entry->addr, false);
319 mptcp_pm_add_addr_send_ack(msk);
320 entry->retrans_times++;
323 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
324 sk_reset_timer(sk, timer,
325 jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
327 spin_unlock_bh(&msk->pm.lock);
329 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
330 mptcp_pm_subflow_established(msk);
336 struct mptcp_pm_add_entry *
337 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
338 const struct mptcp_addr_info *addr, bool check_id)
340 struct mptcp_pm_add_entry *entry;
341 struct sock *sk = (struct sock *)msk;
343 spin_lock_bh(&msk->pm.lock);
344 entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
345 if (entry && (!check_id || entry->addr.id == addr->id))
346 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
347 spin_unlock_bh(&msk->pm.lock);
349 if (entry && (!check_id || entry->addr.id == addr->id))
350 sk_stop_timer_sync(sk, &entry->add_timer);
355 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
356 const struct mptcp_pm_addr_entry *entry)
358 struct mptcp_pm_add_entry *add_entry = NULL;
359 struct sock *sk = (struct sock *)msk;
360 struct net *net = sock_net(sk);
362 lockdep_assert_held(&msk->pm.lock);
364 add_entry = mptcp_lookup_anno_list_by_saddr(msk, &entry->addr);
367 if (mptcp_pm_is_kernel(msk))
370 sk_reset_timer(sk, &add_entry->add_timer,
371 jiffies + mptcp_get_add_addr_timeout(net));
375 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
379 list_add(&add_entry->list, &msk->pm.anno_list);
381 add_entry->addr = entry->addr;
382 add_entry->sock = msk;
383 add_entry->retrans_times = 0;
385 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
386 sk_reset_timer(sk, &add_entry->add_timer,
387 jiffies + mptcp_get_add_addr_timeout(net));
392 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
394 struct mptcp_pm_add_entry *entry, *tmp;
395 struct sock *sk = (struct sock *)msk;
396 LIST_HEAD(free_list);
398 pr_debug("msk=%p", msk);
400 spin_lock_bh(&msk->pm.lock);
401 list_splice_init(&msk->pm.anno_list, &free_list);
402 spin_unlock_bh(&msk->pm.lock);
404 list_for_each_entry_safe(entry, tmp, &free_list, list) {
405 sk_stop_timer_sync(sk, &entry->add_timer);
410 /* Fill all the remote addresses into the array addrs[],
411 * and return the array size.
413 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk,
414 struct mptcp_addr_info *local,
416 struct mptcp_addr_info *addrs)
418 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
419 struct sock *sk = (struct sock *)msk, *ssk;
420 struct mptcp_subflow_context *subflow;
421 struct mptcp_addr_info remote = { 0 };
422 unsigned int subflows_max;
425 subflows_max = mptcp_pm_get_subflows_max(msk);
426 remote_address((struct sock_common *)sk, &remote);
428 /* Non-fullmesh endpoint, fill in the single entry
429 * corresponding to the primary MPC subflow remote address
438 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
440 /* Forbid creation of new subflows matching existing
441 * ones, possibly already created by incoming ADD_ADDR
443 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
444 mptcp_for_each_subflow(msk, subflow)
445 if (READ_ONCE(subflow->local_id) == local->id)
446 __set_bit(subflow->remote_id, unavail_id);
448 mptcp_for_each_subflow(msk, subflow) {
449 ssk = mptcp_subflow_tcp_sock(subflow);
450 remote_address((struct sock_common *)ssk, &addrs[i]);
451 addrs[i].id = READ_ONCE(subflow->remote_id);
452 if (deny_id0 && !addrs[i].id)
455 if (msk->pm.subflows < subflows_max) {
456 /* forbid creating multiple address towards
459 __set_bit(addrs[i].id, unavail_id);
469 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
470 bool prio, bool backup)
472 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
475 pr_debug("send ack for %s",
476 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr"));
478 slow = lock_sock_fast(ssk);
480 if (subflow->backup != backup)
481 msk->last_snd = NULL;
483 subflow->send_mp_prio = 1;
484 subflow->backup = backup;
485 subflow->request_bkup = backup;
488 __mptcp_subflow_send_ack(ssk);
489 unlock_sock_fast(ssk, slow);
492 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
493 bool prio, bool backup)
495 spin_unlock_bh(&msk->pm.lock);
496 __mptcp_pm_send_ack(msk, subflow, prio, backup);
497 spin_lock_bh(&msk->pm.lock);
500 static struct mptcp_pm_addr_entry *
501 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
503 struct mptcp_pm_addr_entry *entry;
505 list_for_each_entry(entry, &pernet->local_addr_list, list) {
506 if (entry->addr.id == id)
512 static struct mptcp_pm_addr_entry *
513 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
516 struct mptcp_pm_addr_entry *entry;
518 list_for_each_entry(entry, &pernet->local_addr_list, list) {
519 if ((!lookup_by_id &&
520 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) ||
521 (lookup_by_id && entry->addr.id == info->id))
527 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
529 struct sock *sk = (struct sock *)msk;
530 struct mptcp_pm_addr_entry *local;
531 unsigned int add_addr_signal_max;
532 unsigned int local_addr_max;
533 struct pm_nl_pernet *pernet;
534 unsigned int subflows_max;
536 pernet = pm_nl_get_pernet(sock_net(sk));
538 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
539 local_addr_max = mptcp_pm_get_local_addr_max(msk);
540 subflows_max = mptcp_pm_get_subflows_max(msk);
542 /* do lazy endpoint usage accounting for the MPC subflows */
543 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
544 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
545 struct mptcp_pm_addr_entry *entry;
546 struct mptcp_addr_info mpc_addr;
549 local_address((struct sock_common *)msk->first, &mpc_addr);
551 entry = __lookup_addr(pernet, &mpc_addr, false);
553 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap);
554 msk->mpc_endpoint_id = entry->addr.id;
555 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
560 mptcp_pm_send_ack(msk, subflow, true, backup);
562 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
565 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
566 msk->pm.local_addr_used, local_addr_max,
567 msk->pm.add_addr_signaled, add_addr_signal_max,
568 msk->pm.subflows, subflows_max);
570 /* check first for announce */
571 if (msk->pm.add_addr_signaled < add_addr_signal_max) {
572 local = select_signal_address(pernet, msk);
574 /* due to racing events on both ends we can reach here while
575 * previous add address is still running: if we invoke now
576 * mptcp_pm_announce_addr(), that will fail and the
577 * corresponding id will be marked as used.
578 * Instead let the PM machinery reschedule us when the
579 * current address announce will be completed.
581 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
585 if (mptcp_pm_alloc_anno_list(msk, local)) {
586 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
587 msk->pm.add_addr_signaled++;
588 mptcp_pm_announce_addr(msk, &local->addr, false);
589 mptcp_pm_nl_addr_send_ack(msk);
594 /* check if should create a new subflow */
595 while (msk->pm.local_addr_used < local_addr_max &&
596 msk->pm.subflows < subflows_max) {
597 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
601 local = select_local_address(pernet, msk);
605 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
607 msk->pm.local_addr_used++;
608 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs);
610 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
611 spin_unlock_bh(&msk->pm.lock);
612 for (i = 0; i < nr; i++)
613 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
614 spin_lock_bh(&msk->pm.lock);
616 mptcp_pm_nl_check_work_pending(msk);
619 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
621 mptcp_pm_create_subflow_or_signal_addr(msk);
624 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
626 mptcp_pm_create_subflow_or_signal_addr(msk);
629 /* Fill all the local addresses into the array addrs[],
630 * and return the array size.
632 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
633 struct mptcp_addr_info *addrs)
635 struct sock *sk = (struct sock *)msk;
636 struct mptcp_pm_addr_entry *entry;
637 struct mptcp_addr_info local;
638 struct pm_nl_pernet *pernet;
639 unsigned int subflows_max;
642 pernet = pm_nl_get_pernet_from_msk(msk);
643 subflows_max = mptcp_pm_get_subflows_max(msk);
646 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
647 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
650 if (entry->addr.family != sk->sk_family) {
651 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
652 if ((entry->addr.family == AF_INET &&
653 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
654 (sk->sk_family == AF_INET &&
655 !ipv6_addr_v4mapped(&entry->addr.addr6)))
660 if (msk->pm.subflows < subflows_max) {
662 addrs[i++] = entry->addr;
667 /* If the array is empty, fill in the single
668 * 'IPADDRANY' local address
671 memset(&local, 0, sizeof(local));
672 local.family = msk->pm.remote.family;
681 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
683 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
684 struct sock *sk = (struct sock *)msk;
685 unsigned int add_addr_accept_max;
686 struct mptcp_addr_info remote;
687 unsigned int subflows_max;
690 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
691 subflows_max = mptcp_pm_get_subflows_max(msk);
693 pr_debug("accepted %d:%d remote family %d",
694 msk->pm.add_addr_accepted, add_addr_accept_max,
695 msk->pm.remote.family);
697 remote = msk->pm.remote;
698 mptcp_pm_announce_addr(msk, &remote, true);
699 mptcp_pm_nl_addr_send_ack(msk);
701 if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
704 /* pick id 0 port, if none is provided the remote address */
706 remote.port = sk->sk_dport;
708 /* connect to the specified remote address, using whatever
709 * local address the routing configuration will pick.
711 nr = fill_local_addresses_vec(msk, addrs);
713 msk->pm.add_addr_accepted++;
714 if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
715 msk->pm.subflows >= subflows_max)
716 WRITE_ONCE(msk->pm.accept_addr, false);
718 spin_unlock_bh(&msk->pm.lock);
719 for (i = 0; i < nr; i++)
720 __mptcp_subflow_connect(sk, &addrs[i], &remote);
721 spin_lock_bh(&msk->pm.lock);
724 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
726 struct mptcp_subflow_context *subflow;
728 msk_owned_by_me(msk);
729 lockdep_assert_held(&msk->pm.lock);
731 if (!mptcp_pm_should_add_signal(msk) &&
732 !mptcp_pm_should_rm_signal(msk))
735 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
737 mptcp_pm_send_ack(msk, subflow, false, false);
740 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
741 struct mptcp_addr_info *addr,
742 struct mptcp_addr_info *rem,
745 struct mptcp_subflow_context *subflow;
747 pr_debug("bkup=%d", bkup);
749 mptcp_for_each_subflow(msk, subflow) {
750 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
751 struct mptcp_addr_info local, remote;
753 local_address((struct sock_common *)ssk, &local);
754 if (!mptcp_addresses_equal(&local, addr, addr->port))
757 if (rem && rem->family != AF_UNSPEC) {
758 remote_address((struct sock_common *)ssk, &remote);
759 if (!mptcp_addresses_equal(&remote, rem, rem->port))
763 __mptcp_pm_send_ack(msk, subflow, true, bkup);
770 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id)
772 return local_id == id || (!local_id && msk->mpc_endpoint_id == id);
775 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
776 const struct mptcp_rm_list *rm_list,
777 enum linux_mptcp_mib_field rm_type)
779 struct mptcp_subflow_context *subflow, *tmp;
780 struct sock *sk = (struct sock *)msk;
783 pr_debug("%s rm_list_nr %d",
784 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
786 msk_owned_by_me(msk);
788 if (sk->sk_state == TCP_LISTEN)
794 if (list_empty(&msk->conn_list))
797 for (i = 0; i < rm_list->nr; i++) {
798 u8 rm_id = rm_list->ids[i];
799 bool removed = false;
801 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
802 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
803 u8 remote_id = READ_ONCE(subflow->remote_id);
804 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
805 u8 id = subflow_get_local_id(subflow);
807 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id)
809 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id))
812 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u",
813 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
814 i, rm_id, id, remote_id, msk->mpc_endpoint_id);
815 spin_unlock_bh(&msk->pm.lock);
816 mptcp_subflow_shutdown(sk, ssk, how);
818 /* the following takes care of updating the subflows counter */
819 mptcp_close_ssk(sk, ssk, subflow);
820 spin_lock_bh(&msk->pm.lock);
823 __MPTCP_INC_STATS(sock_net(sk), rm_type);
825 if (rm_type == MPTCP_MIB_RMSUBFLOW)
826 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap);
830 if (!mptcp_pm_is_kernel(msk))
833 if (rm_type == MPTCP_MIB_RMADDR) {
834 msk->pm.add_addr_accepted--;
835 WRITE_ONCE(msk->pm.accept_addr, true);
836 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
837 msk->pm.local_addr_used--;
842 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
844 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
847 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
848 const struct mptcp_rm_list *rm_list)
850 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
853 void mptcp_pm_nl_work(struct mptcp_sock *msk)
855 struct mptcp_pm_data *pm = &msk->pm;
857 msk_owned_by_me(msk);
859 if (!(pm->status & MPTCP_PM_WORK_MASK))
862 spin_lock_bh(&msk->pm.lock);
864 pr_debug("msk=%p status=%x", msk, pm->status);
865 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
866 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
867 mptcp_pm_nl_add_addr_received(msk);
869 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
870 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
871 mptcp_pm_nl_addr_send_ack(msk);
873 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
874 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
875 mptcp_pm_nl_rm_addr_received(msk);
877 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
878 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
879 mptcp_pm_nl_fully_established(msk);
881 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
882 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
883 mptcp_pm_nl_subflow_established(msk);
886 spin_unlock_bh(&msk->pm.lock);
889 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
891 return (entry->flags &
892 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
893 MPTCP_PM_ADDR_FLAG_SIGNAL;
896 /* caller must ensure the RCU grace period is already elapsed */
897 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
900 sock_release(entry->lsk);
904 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
905 struct mptcp_pm_addr_entry *entry,
908 struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
909 unsigned int addr_max;
912 spin_lock_bh(&pernet->lock);
913 /* to keep the code simple, don't do IDR-like allocation for address ID,
914 * just bail when we exceed limits
916 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
918 if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
920 if (test_bit(entry->addr.id, pernet->id_bitmap))
923 /* do not insert duplicate address, differentiate on port only
926 if (!address_use_port(entry))
927 entry->addr.port = 0;
928 list_for_each_entry(cur, &pernet->local_addr_list, list) {
929 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
930 cur->addr.port || entry->addr.port)) {
931 /* allow replacing the exiting endpoint only if such
932 * endpoint is an implicit one and the user-space
933 * did not provide an endpoint id
935 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT))
941 entry->addr.id = cur->addr.id;
942 list_del_rcu(&cur->list);
948 if (!entry->addr.id && needs_id) {
950 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
951 MPTCP_PM_MAX_ADDR_ID + 1,
953 if (!entry->addr.id && pernet->next_id != 1) {
959 if (!entry->addr.id && needs_id)
962 __set_bit(entry->addr.id, pernet->id_bitmap);
963 if (entry->addr.id > pernet->next_id)
964 pernet->next_id = entry->addr.id;
966 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
967 addr_max = pernet->add_addr_signal_max;
968 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
970 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
971 addr_max = pernet->local_addr_max;
972 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
976 if (!entry->addr.port)
977 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
979 list_add_rcu(&entry->list, &pernet->local_addr_list);
980 ret = entry->addr.id;
983 spin_unlock_bh(&pernet->lock);
985 /* just replaced an existing entry, free it */
988 __mptcp_pm_release_addr_entry(del_entry);
993 static struct lock_class_key mptcp_slock_keys[2];
994 static struct lock_class_key mptcp_keys[2];
996 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
997 struct mptcp_pm_addr_entry *entry)
999 bool is_ipv6 = sk->sk_family == AF_INET6;
1000 int addrlen = sizeof(struct sockaddr_in);
1001 struct sockaddr_storage addr;
1002 struct socket *ssock;
1007 err = sock_create_kern(sock_net(sk), entry->addr.family,
1008 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1012 newsk = entry->lsk->sk;
1016 /* The subflow socket lock is acquired in a nested to the msk one
1017 * in several places, even by the TCP stack, and this msk is a kernel
1018 * socket: lockdep complains. Instead of propagating the _nested
1019 * modifiers in several places, re-init the lock class for the msk
1020 * socket to an mptcp specific one.
1022 sock_lock_init_class_and_name(newsk,
1023 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1024 &mptcp_slock_keys[is_ipv6],
1025 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1026 &mptcp_keys[is_ipv6]);
1029 ssock = __mptcp_nmpc_socket(mptcp_sk(newsk));
1030 release_sock(newsk);
1034 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1035 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1036 if (entry->addr.family == AF_INET6)
1037 addrlen = sizeof(struct sockaddr_in6);
1039 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
1041 pr_warn("kernel_bind error, err=%d", err);
1045 inet_sk_state_store(newsk, TCP_LISTEN);
1046 err = kernel_listen(ssock, backlog);
1048 pr_warn("kernel_listen error, err=%d", err);
1055 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
1057 struct mptcp_pm_addr_entry *entry;
1058 struct mptcp_addr_info skc_local;
1059 struct mptcp_addr_info msk_local;
1060 struct pm_nl_pernet *pernet;
1063 if (WARN_ON_ONCE(!msk))
1066 /* The 0 ID mapping is defined by the first subflow, copied into the msk
1069 local_address((struct sock_common *)msk, &msk_local);
1070 local_address((struct sock_common *)skc, &skc_local);
1071 if (mptcp_addresses_equal(&msk_local, &skc_local, false))
1074 if (mptcp_pm_is_userspace(msk))
1075 return mptcp_userspace_pm_get_local_id(msk, &skc_local);
1077 pernet = pm_nl_get_pernet_from_msk(msk);
1080 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1081 if (mptcp_addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1082 ret = entry->addr.id;
1090 /* address not found, add to local list */
1091 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1095 entry->addr = skc_local;
1097 entry->addr.port = 0;
1099 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1101 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true);
1108 #define MPTCP_PM_CMD_GRP_OFFSET 0
1109 #define MPTCP_PM_EV_GRP_OFFSET 1
1111 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1112 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, },
1113 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME,
1114 .flags = GENL_UNS_ADMIN_PERM,
1118 static const struct nla_policy
1119 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1120 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, },
1121 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, },
1122 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, },
1123 [MPTCP_PM_ADDR_ATTR_ADDR6] =
1124 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1125 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 },
1126 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 },
1127 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 },
1130 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1131 [MPTCP_PM_ATTR_ADDR] =
1132 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1133 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, },
1134 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, },
1135 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, },
1136 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, },
1137 [MPTCP_PM_ATTR_ADDR_REMOTE] =
1138 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1141 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1143 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1144 struct sock *sk = (struct sock *)msk;
1145 unsigned int active_max_loss_cnt;
1146 struct net *net = sock_net(sk);
1147 unsigned int stale_loss_cnt;
1150 stale_loss_cnt = mptcp_stale_loss_cnt(net);
1151 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1154 /* look for another available subflow not in loss state */
1155 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1156 mptcp_for_each_subflow(msk, iter) {
1157 if (iter != subflow && mptcp_subflow_active(iter) &&
1158 iter->stale_count < active_max_loss_cnt) {
1159 /* we have some alternatives, try to mark this subflow as idle ...*/
1160 slow = lock_sock_fast(ssk);
1161 if (!tcp_rtx_and_write_queues_empty(ssk)) {
1163 __mptcp_retransmit_pending_data(sk);
1164 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1166 unlock_sock_fast(ssk, slow);
1168 /* always try to push the pending data regardless of re-injections:
1169 * we can possibly use backup subflows now, and subflow selection
1170 * is cheap under the msk socket lock
1172 __mptcp_push_pending(sk, 0);
1178 static int mptcp_pm_family_to_addr(int family)
1180 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1181 if (family == AF_INET6)
1182 return MPTCP_PM_ADDR_ATTR_ADDR6;
1184 return MPTCP_PM_ADDR_ATTR_ADDR4;
1187 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1188 const struct nlattr *attr,
1189 struct genl_info *info,
1190 struct mptcp_addr_info *addr,
1191 bool require_family)
1196 GENL_SET_ERR_MSG(info, "missing address info");
1200 /* no validation needed - was already done via nested policy */
1201 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1202 mptcp_pm_addr_policy, info->extack);
1206 if (tb[MPTCP_PM_ADDR_ATTR_ID])
1207 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1209 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1210 if (!require_family)
1213 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1218 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1219 if (addr->family != AF_INET
1220 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1221 && addr->family != AF_INET6
1224 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1225 "unknown address family");
1228 addr_addr = mptcp_pm_family_to_addr(addr->family);
1229 if (!tb[addr_addr]) {
1230 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1231 "missing address data");
1235 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1236 if (addr->family == AF_INET6)
1237 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1240 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1242 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1243 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1248 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1249 struct mptcp_addr_info *addr)
1251 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1253 memset(addr, 0, sizeof(*addr));
1255 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1258 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1259 bool require_family,
1260 struct mptcp_pm_addr_entry *entry)
1262 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1265 memset(entry, 0, sizeof(*entry));
1267 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1271 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1272 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1274 entry->ifindex = val;
1277 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1278 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1280 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1281 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1286 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1288 return pm_nl_get_pernet(genl_info_net(info));
1291 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1293 struct mptcp_sock *msk;
1294 long s_slot = 0, s_num = 0;
1296 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1297 struct sock *sk = (struct sock *)msk;
1299 if (!READ_ONCE(msk->fully_established) ||
1300 mptcp_pm_is_userspace(msk))
1304 spin_lock_bh(&msk->pm.lock);
1305 mptcp_pm_create_subflow_or_signal_addr(msk);
1306 spin_unlock_bh(&msk->pm.lock);
1317 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr,
1318 struct genl_info *info)
1320 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1322 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1323 mptcp_pm_addr_policy, info->extack) &&
1324 tb[MPTCP_PM_ADDR_ATTR_ID])
1329 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1331 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1332 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1333 struct mptcp_pm_addr_entry addr, *entry;
1336 ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1340 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1341 GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1345 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1346 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1347 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1351 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1352 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1356 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1358 GENL_SET_ERR_MSG(info, "can't allocate addr");
1363 if (entry->addr.port) {
1364 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1366 GENL_SET_ERR_MSG(info, "create listen socket error");
1370 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry,
1371 !mptcp_pm_has_addr_attr_id(attr, info));
1373 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1377 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1381 __mptcp_pm_release_addr_entry(entry);
1385 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1386 u8 *flags, int *ifindex)
1388 struct mptcp_pm_addr_entry *entry;
1389 struct sock *sk = (struct sock *)msk;
1390 struct net *net = sock_net(sk);
1396 if (mptcp_pm_is_userspace(msk))
1397 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk,
1403 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1405 *flags = entry->flags;
1406 *ifindex = entry->ifindex;
1414 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1415 const struct mptcp_addr_info *addr)
1417 struct mptcp_pm_add_entry *entry;
1419 entry = mptcp_pm_del_add_timer(msk, addr, false);
1421 list_del(&entry->list);
1429 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1430 const struct mptcp_addr_info *addr,
1433 struct mptcp_rm_list list = { .nr = 0 };
1436 list.ids[list.nr++] = addr->id;
1438 ret = remove_anno_list_by_saddr(msk, addr);
1440 spin_lock_bh(&msk->pm.lock);
1441 mptcp_pm_remove_addr(msk, &list);
1442 spin_unlock_bh(&msk->pm.lock);
1447 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1448 const struct mptcp_pm_addr_entry *entry)
1450 const struct mptcp_addr_info *addr = &entry->addr;
1451 struct mptcp_rm_list list = { .nr = 0 };
1452 long s_slot = 0, s_num = 0;
1453 struct mptcp_sock *msk;
1455 pr_debug("remove_id=%d", addr->id);
1457 list.ids[list.nr++] = addr->id;
1459 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1460 struct sock *sk = (struct sock *)msk;
1461 bool remove_subflow;
1463 if (mptcp_pm_is_userspace(msk))
1466 if (list_empty(&msk->conn_list)) {
1467 mptcp_pm_remove_anno_addr(msk, addr, false);
1472 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1473 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1474 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1476 mptcp_pm_remove_subflow(msk, &list);
1487 static int mptcp_nl_remove_id_zero_address(struct net *net,
1488 struct mptcp_addr_info *addr)
1490 struct mptcp_rm_list list = { .nr = 0 };
1491 long s_slot = 0, s_num = 0;
1492 struct mptcp_sock *msk;
1494 list.ids[list.nr++] = 0;
1496 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1497 struct sock *sk = (struct sock *)msk;
1498 struct mptcp_addr_info msk_local;
1500 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1503 local_address((struct sock_common *)msk, &msk_local);
1504 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1508 spin_lock_bh(&msk->pm.lock);
1509 mptcp_pm_remove_addr(msk, &list);
1510 mptcp_pm_nl_rm_subflow_received(msk, &list);
1511 spin_unlock_bh(&msk->pm.lock);
1522 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1524 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1525 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1526 struct mptcp_pm_addr_entry addr, *entry;
1527 unsigned int addr_max;
1530 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1534 /* the zero id address is special: the first address used by the msk
1535 * always gets such an id, so different subflows can have different zero
1536 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1537 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1539 if (addr.addr.id == 0)
1540 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1542 spin_lock_bh(&pernet->lock);
1543 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1545 GENL_SET_ERR_MSG(info, "address not found");
1546 spin_unlock_bh(&pernet->lock);
1549 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1550 addr_max = pernet->add_addr_signal_max;
1551 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1553 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1554 addr_max = pernet->local_addr_max;
1555 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1559 list_del_rcu(&entry->list);
1560 __clear_bit(entry->addr.id, pernet->id_bitmap);
1561 spin_unlock_bh(&pernet->lock);
1563 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1565 __mptcp_pm_release_addr_entry(entry);
1570 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1572 struct mptcp_rm_list alist = { .nr = 0 };
1573 struct mptcp_pm_addr_entry *entry;
1575 list_for_each_entry(entry, rm_list, list) {
1576 if ((remove_anno_list_by_saddr(msk, &entry->addr) ||
1577 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) &&
1578 alist.nr < MPTCP_RM_IDS_MAX)
1579 alist.ids[alist.nr++] = entry->addr.id;
1583 spin_lock_bh(&msk->pm.lock);
1584 mptcp_pm_remove_addr(msk, &alist);
1585 spin_unlock_bh(&msk->pm.lock);
1589 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1590 struct list_head *rm_list)
1592 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1593 struct mptcp_pm_addr_entry *entry;
1595 list_for_each_entry(entry, rm_list, list) {
1596 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1597 slist.nr < MPTCP_RM_IDS_MAX)
1598 slist.ids[slist.nr++] = entry->addr.id;
1600 if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1601 alist.nr < MPTCP_RM_IDS_MAX)
1602 alist.ids[alist.nr++] = entry->addr.id;
1606 spin_lock_bh(&msk->pm.lock);
1607 mptcp_pm_remove_addr(msk, &alist);
1608 spin_unlock_bh(&msk->pm.lock);
1611 mptcp_pm_remove_subflow(msk, &slist);
1614 static void mptcp_nl_remove_addrs_list(struct net *net,
1615 struct list_head *rm_list)
1617 long s_slot = 0, s_num = 0;
1618 struct mptcp_sock *msk;
1620 if (list_empty(rm_list))
1623 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1624 struct sock *sk = (struct sock *)msk;
1626 if (!mptcp_pm_is_userspace(msk)) {
1628 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1637 /* caller must ensure the RCU grace period is already elapsed */
1638 static void __flush_addrs(struct list_head *list)
1640 while (!list_empty(list)) {
1641 struct mptcp_pm_addr_entry *cur;
1643 cur = list_entry(list->next,
1644 struct mptcp_pm_addr_entry, list);
1645 list_del_rcu(&cur->list);
1646 __mptcp_pm_release_addr_entry(cur);
1650 static void __reset_counters(struct pm_nl_pernet *pernet)
1652 WRITE_ONCE(pernet->add_addr_signal_max, 0);
1653 WRITE_ONCE(pernet->add_addr_accept_max, 0);
1654 WRITE_ONCE(pernet->local_addr_max, 0);
1658 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1660 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1661 LIST_HEAD(free_list);
1663 spin_lock_bh(&pernet->lock);
1664 list_splice_init(&pernet->local_addr_list, &free_list);
1665 __reset_counters(pernet);
1666 pernet->next_id = 1;
1667 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1668 spin_unlock_bh(&pernet->lock);
1669 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1671 __flush_addrs(&free_list);
1675 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1676 struct mptcp_pm_addr_entry *entry)
1678 struct mptcp_addr_info *addr = &entry->addr;
1679 struct nlattr *attr;
1681 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1685 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1686 goto nla_put_failure;
1687 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1688 goto nla_put_failure;
1689 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1690 goto nla_put_failure;
1691 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1692 goto nla_put_failure;
1693 if (entry->ifindex &&
1694 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1695 goto nla_put_failure;
1697 if (addr->family == AF_INET &&
1698 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1700 goto nla_put_failure;
1701 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1702 else if (addr->family == AF_INET6 &&
1703 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1704 goto nla_put_failure;
1706 nla_nest_end(skb, attr);
1710 nla_nest_cancel(skb, attr);
1714 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1716 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1717 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1718 struct mptcp_pm_addr_entry addr, *entry;
1719 struct sk_buff *msg;
1723 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1727 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1731 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1732 info->genlhdr->cmd);
1734 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1739 spin_lock_bh(&pernet->lock);
1740 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1742 GENL_SET_ERR_MSG(info, "address not found");
1747 ret = mptcp_nl_fill_addr(msg, entry);
1751 genlmsg_end(msg, reply);
1752 ret = genlmsg_reply(msg, info);
1753 spin_unlock_bh(&pernet->lock);
1757 spin_unlock_bh(&pernet->lock);
1764 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1765 struct netlink_callback *cb)
1767 struct net *net = sock_net(msg->sk);
1768 struct mptcp_pm_addr_entry *entry;
1769 struct pm_nl_pernet *pernet;
1770 int id = cb->args[0];
1774 pernet = pm_nl_get_pernet(net);
1776 spin_lock_bh(&pernet->lock);
1777 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1778 if (test_bit(i, pernet->id_bitmap)) {
1779 entry = __lookup_addr_by_id(pernet, i);
1783 if (entry->addr.id <= id)
1786 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1787 cb->nlh->nlmsg_seq, &mptcp_genl_family,
1788 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1792 if (mptcp_nl_fill_addr(msg, entry) < 0) {
1793 genlmsg_cancel(msg, hdr);
1797 id = entry->addr.id;
1798 genlmsg_end(msg, hdr);
1801 spin_unlock_bh(&pernet->lock);
1807 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1809 struct nlattr *attr = info->attrs[id];
1814 *limit = nla_get_u32(attr);
1815 if (*limit > MPTCP_PM_ADDR_MAX) {
1816 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1823 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1825 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1826 unsigned int rcv_addrs, subflows;
1829 spin_lock_bh(&pernet->lock);
1830 rcv_addrs = pernet->add_addr_accept_max;
1831 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1835 subflows = pernet->subflows_max;
1836 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1840 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1841 WRITE_ONCE(pernet->subflows_max, subflows);
1844 spin_unlock_bh(&pernet->lock);
1849 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1851 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1852 struct sk_buff *msg;
1855 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1859 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1860 MPTCP_PM_CMD_GET_LIMITS);
1864 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1865 READ_ONCE(pernet->add_addr_accept_max)))
1868 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1869 READ_ONCE(pernet->subflows_max)))
1872 genlmsg_end(msg, reply);
1873 return genlmsg_reply(msg, info);
1876 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1881 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1882 struct mptcp_addr_info *addr)
1884 struct mptcp_rm_list list = { .nr = 0 };
1886 list.ids[list.nr++] = addr->id;
1888 spin_lock_bh(&msk->pm.lock);
1889 mptcp_pm_nl_rm_subflow_received(msk, &list);
1890 mptcp_pm_create_subflow_or_signal_addr(msk);
1891 spin_unlock_bh(&msk->pm.lock);
1894 static int mptcp_nl_set_flags(struct net *net,
1895 struct mptcp_addr_info *addr,
1896 u8 bkup, u8 changed)
1898 long s_slot = 0, s_num = 0;
1899 struct mptcp_sock *msk;
1902 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1903 struct sock *sk = (struct sock *)msk;
1905 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1909 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1910 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
1911 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1912 mptcp_pm_nl_fullmesh(msk, addr);
1923 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1925 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1926 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
1927 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
1928 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
1929 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1930 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1931 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1932 MPTCP_PM_ADDR_FLAG_FULLMESH;
1933 struct net *net = sock_net(skb->sk);
1934 u8 bkup = 0, lookup_by_id = 0;
1937 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1942 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
1947 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1949 if (addr.addr.family == AF_UNSPEC) {
1956 return mptcp_userspace_pm_set_flags(sock_net(skb->sk),
1957 token, &addr, &remote, bkup);
1959 spin_lock_bh(&pernet->lock);
1960 entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1962 spin_unlock_bh(&pernet->lock);
1965 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1966 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1967 spin_unlock_bh(&pernet->lock);
1971 changed = (addr.flags ^ entry->flags) & mask;
1972 entry->flags = (entry->flags & ~mask) | (addr.flags & mask);
1974 spin_unlock_bh(&pernet->lock);
1976 mptcp_nl_set_flags(net, &addr.addr, bkup, changed);
1980 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1982 genlmsg_multicast_netns(&mptcp_genl_family, net,
1983 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1986 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1988 return genl_has_listeners(&mptcp_genl_family,
1989 sock_net((const struct sock *)msk),
1990 MPTCP_PM_EV_GRP_OFFSET);
1993 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1995 const struct inet_sock *issk = inet_sk(ssk);
1996 const struct mptcp_subflow_context *sf;
1998 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2001 switch (ssk->sk_family) {
2003 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2005 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
2008 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2010 const struct ipv6_pinfo *np = inet6_sk(ssk);
2012 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2014 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
2024 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2026 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
2029 sf = mptcp_subflow_ctx(ssk);
2030 if (WARN_ON_ONCE(!sf))
2033 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf)))
2036 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
2042 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
2043 const struct mptcp_sock *msk,
2044 const struct sock *ssk)
2046 const struct sock *sk = (const struct sock *)msk;
2047 const struct mptcp_subflow_context *sf;
2050 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2053 if (mptcp_event_add_subflow(skb, ssk))
2056 sf = mptcp_subflow_ctx(ssk);
2057 if (WARN_ON_ONCE(!sf))
2060 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2063 if (ssk->sk_bound_dev_if &&
2064 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2067 sk_err = READ_ONCE(ssk->sk_err);
2068 if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2069 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2075 static int mptcp_event_sub_established(struct sk_buff *skb,
2076 const struct mptcp_sock *msk,
2077 const struct sock *ssk)
2079 return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2082 static int mptcp_event_sub_closed(struct sk_buff *skb,
2083 const struct mptcp_sock *msk,
2084 const struct sock *ssk)
2086 const struct mptcp_subflow_context *sf;
2088 if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2091 sf = mptcp_subflow_ctx(ssk);
2092 if (!sf->reset_seen)
2095 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2098 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2104 static int mptcp_event_created(struct sk_buff *skb,
2105 const struct mptcp_sock *msk,
2106 const struct sock *ssk)
2108 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2113 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2116 return mptcp_event_add_subflow(skb, ssk);
2119 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2121 struct net *net = sock_net((const struct sock *)msk);
2122 struct nlmsghdr *nlh;
2123 struct sk_buff *skb;
2125 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2128 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2132 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2134 goto nla_put_failure;
2136 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2137 goto nla_put_failure;
2139 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2140 goto nla_put_failure;
2142 genlmsg_end(skb, nlh);
2143 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2150 void mptcp_event_addr_announced(const struct sock *ssk,
2151 const struct mptcp_addr_info *info)
2153 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2154 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2155 struct net *net = sock_net(ssk);
2156 struct nlmsghdr *nlh;
2157 struct sk_buff *skb;
2159 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2162 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2166 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2167 MPTCP_EVENT_ANNOUNCED);
2169 goto nla_put_failure;
2171 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2172 goto nla_put_failure;
2174 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2175 goto nla_put_failure;
2177 if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2179 inet_sk(ssk)->inet_dport :
2181 goto nla_put_failure;
2183 switch (info->family) {
2185 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2186 goto nla_put_failure;
2188 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2190 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2191 goto nla_put_failure;
2196 goto nla_put_failure;
2199 genlmsg_end(skb, nlh);
2200 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2207 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2208 const struct sock *ssk, gfp_t gfp)
2210 struct net *net = sock_net((const struct sock *)msk);
2211 struct nlmsghdr *nlh;
2212 struct sk_buff *skb;
2214 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2217 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2221 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2223 goto nla_put_failure;
2226 case MPTCP_EVENT_UNSPEC:
2229 case MPTCP_EVENT_CREATED:
2230 case MPTCP_EVENT_ESTABLISHED:
2231 if (mptcp_event_created(skb, msk, ssk) < 0)
2232 goto nla_put_failure;
2234 case MPTCP_EVENT_CLOSED:
2235 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2236 goto nla_put_failure;
2238 case MPTCP_EVENT_ANNOUNCED:
2239 case MPTCP_EVENT_REMOVED:
2240 /* call mptcp_event_addr_announced()/removed instead */
2243 case MPTCP_EVENT_SUB_ESTABLISHED:
2244 case MPTCP_EVENT_SUB_PRIORITY:
2245 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2246 goto nla_put_failure;
2248 case MPTCP_EVENT_SUB_CLOSED:
2249 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2250 goto nla_put_failure;
2254 genlmsg_end(skb, nlh);
2255 mptcp_nl_mcast_send(net, skb, gfp);
2262 static const struct genl_small_ops mptcp_pm_ops[] = {
2264 .cmd = MPTCP_PM_CMD_ADD_ADDR,
2265 .doit = mptcp_nl_cmd_add_addr,
2266 .flags = GENL_UNS_ADMIN_PERM,
2269 .cmd = MPTCP_PM_CMD_DEL_ADDR,
2270 .doit = mptcp_nl_cmd_del_addr,
2271 .flags = GENL_UNS_ADMIN_PERM,
2274 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS,
2275 .doit = mptcp_nl_cmd_flush_addrs,
2276 .flags = GENL_UNS_ADMIN_PERM,
2279 .cmd = MPTCP_PM_CMD_GET_ADDR,
2280 .doit = mptcp_nl_cmd_get_addr,
2281 .dumpit = mptcp_nl_cmd_dump_addrs,
2284 .cmd = MPTCP_PM_CMD_SET_LIMITS,
2285 .doit = mptcp_nl_cmd_set_limits,
2286 .flags = GENL_UNS_ADMIN_PERM,
2289 .cmd = MPTCP_PM_CMD_GET_LIMITS,
2290 .doit = mptcp_nl_cmd_get_limits,
2293 .cmd = MPTCP_PM_CMD_SET_FLAGS,
2294 .doit = mptcp_nl_cmd_set_flags,
2295 .flags = GENL_UNS_ADMIN_PERM,
2298 .cmd = MPTCP_PM_CMD_ANNOUNCE,
2299 .doit = mptcp_nl_cmd_announce,
2300 .flags = GENL_UNS_ADMIN_PERM,
2303 .cmd = MPTCP_PM_CMD_REMOVE,
2304 .doit = mptcp_nl_cmd_remove,
2305 .flags = GENL_UNS_ADMIN_PERM,
2308 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE,
2309 .doit = mptcp_nl_cmd_sf_create,
2310 .flags = GENL_UNS_ADMIN_PERM,
2313 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY,
2314 .doit = mptcp_nl_cmd_sf_destroy,
2315 .flags = GENL_UNS_ADMIN_PERM,
2319 static struct genl_family mptcp_genl_family __ro_after_init = {
2320 .name = MPTCP_PM_NAME,
2321 .version = MPTCP_PM_VER,
2322 .maxattr = MPTCP_PM_ATTR_MAX,
2323 .policy = mptcp_pm_policy,
2325 .module = THIS_MODULE,
2326 .small_ops = mptcp_pm_ops,
2327 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops),
2328 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2329 .mcgrps = mptcp_pm_mcgrps,
2330 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps),
2333 static int __net_init pm_nl_init_net(struct net *net)
2335 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2337 INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2339 /* Cit. 2 subflows ought to be enough for anybody. */
2340 pernet->subflows_max = 2;
2341 pernet->next_id = 1;
2342 pernet->stale_loss_cnt = 4;
2343 spin_lock_init(&pernet->lock);
2345 /* No need to initialize other pernet fields, the struct is zeroed at
2352 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2356 list_for_each_entry(net, net_list, exit_list) {
2357 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2359 /* net is removed from namespace list, can't race with
2360 * other modifiers, also netns core already waited for a
2363 __flush_addrs(&pernet->local_addr_list);
2367 static struct pernet_operations mptcp_pm_pernet_ops = {
2368 .init = pm_nl_init_net,
2369 .exit_batch = pm_nl_exit_net,
2370 .id = &pm_nl_pernet_id,
2371 .size = sizeof(struct pm_nl_pernet),
2374 void __init mptcp_pm_nl_init(void)
2376 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2377 panic("Failed to register MPTCP PM pernet subsystem.\n");
2379 if (genl_register_family(&mptcp_genl_family))
2380 panic("Failed to register MPTCP PM netlink family\n");