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 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr,
411 const struct mptcp_addr_info *addr)
415 for (i = 0; i < nr; i++) {
416 if (mptcp_addresses_equal(&addrs[i], addr, addr->port))
423 /* Fill all the remote addresses into the array addrs[],
424 * and return the array size.
426 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, bool fullmesh,
427 struct mptcp_addr_info *addrs)
429 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
430 struct sock *sk = (struct sock *)msk, *ssk;
431 struct mptcp_subflow_context *subflow;
432 struct mptcp_addr_info remote = { 0 };
433 unsigned int subflows_max;
436 subflows_max = mptcp_pm_get_subflows_max(msk);
437 remote_address((struct sock_common *)sk, &remote);
439 /* Non-fullmesh endpoint, fill in the single entry
440 * corresponding to the primary MPC subflow remote address
449 mptcp_for_each_subflow(msk, subflow) {
450 ssk = mptcp_subflow_tcp_sock(subflow);
451 remote_address((struct sock_common *)ssk, &addrs[i]);
452 if (deny_id0 && mptcp_addresses_equal(&addrs[i], &remote, false))
455 if (!lookup_address_in_vec(addrs, i, &addrs[i]) &&
456 msk->pm.subflows < subflows_max) {
466 static struct mptcp_pm_addr_entry *
467 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
469 struct mptcp_pm_addr_entry *entry;
471 list_for_each_entry(entry, &pernet->local_addr_list, list) {
472 if (entry->addr.id == id)
478 static struct mptcp_pm_addr_entry *
479 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
482 struct mptcp_pm_addr_entry *entry;
484 list_for_each_entry(entry, &pernet->local_addr_list, list) {
485 if ((!lookup_by_id && mptcp_addresses_equal(&entry->addr, info, true)) ||
486 (lookup_by_id && entry->addr.id == info->id))
493 lookup_id_by_addr(const struct pm_nl_pernet *pernet, const struct mptcp_addr_info *addr)
495 const struct mptcp_pm_addr_entry *entry;
499 list_for_each_entry(entry, &pernet->local_addr_list, list) {
500 if (mptcp_addresses_equal(&entry->addr, addr, entry->addr.port)) {
501 ret = entry->addr.id;
509 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
511 struct sock *sk = (struct sock *)msk;
512 struct mptcp_pm_addr_entry *local;
513 unsigned int add_addr_signal_max;
514 unsigned int local_addr_max;
515 struct pm_nl_pernet *pernet;
516 unsigned int subflows_max;
518 pernet = pm_nl_get_pernet(sock_net(sk));
520 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
521 local_addr_max = mptcp_pm_get_local_addr_max(msk);
522 subflows_max = mptcp_pm_get_subflows_max(msk);
524 /* do lazy endpoint usage accounting for the MPC subflows */
525 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
526 struct mptcp_addr_info mpc_addr;
529 local_address((struct sock_common *)msk->first, &mpc_addr);
530 mpc_id = lookup_id_by_addr(pernet, &mpc_addr);
532 __clear_bit(mpc_id, msk->pm.id_avail_bitmap);
534 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
537 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
538 msk->pm.local_addr_used, local_addr_max,
539 msk->pm.add_addr_signaled, add_addr_signal_max,
540 msk->pm.subflows, subflows_max);
542 /* check first for announce */
543 if (msk->pm.add_addr_signaled < add_addr_signal_max) {
544 local = select_signal_address(pernet, msk);
546 /* due to racing events on both ends we can reach here while
547 * previous add address is still running: if we invoke now
548 * mptcp_pm_announce_addr(), that will fail and the
549 * corresponding id will be marked as used.
550 * Instead let the PM machinery reschedule us when the
551 * current address announce will be completed.
553 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
557 if (mptcp_pm_alloc_anno_list(msk, local)) {
558 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
559 msk->pm.add_addr_signaled++;
560 mptcp_pm_announce_addr(msk, &local->addr, false);
561 mptcp_pm_nl_addr_send_ack(msk);
566 /* check if should create a new subflow */
567 while (msk->pm.local_addr_used < local_addr_max &&
568 msk->pm.subflows < subflows_max) {
569 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
573 local = select_local_address(pernet, msk);
577 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
579 msk->pm.local_addr_used++;
580 nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
582 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
583 spin_unlock_bh(&msk->pm.lock);
584 for (i = 0; i < nr; i++)
585 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
586 spin_lock_bh(&msk->pm.lock);
588 mptcp_pm_nl_check_work_pending(msk);
591 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
593 mptcp_pm_create_subflow_or_signal_addr(msk);
596 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
598 mptcp_pm_create_subflow_or_signal_addr(msk);
601 /* Fill all the local addresses into the array addrs[],
602 * and return the array size.
604 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
605 struct mptcp_addr_info *addrs)
607 struct sock *sk = (struct sock *)msk;
608 struct mptcp_pm_addr_entry *entry;
609 struct mptcp_addr_info local;
610 struct pm_nl_pernet *pernet;
611 unsigned int subflows_max;
614 pernet = pm_nl_get_pernet_from_msk(msk);
615 subflows_max = mptcp_pm_get_subflows_max(msk);
618 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
619 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
622 if (entry->addr.family != sk->sk_family) {
623 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
624 if ((entry->addr.family == AF_INET &&
625 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
626 (sk->sk_family == AF_INET &&
627 !ipv6_addr_v4mapped(&entry->addr.addr6)))
632 if (msk->pm.subflows < subflows_max) {
634 addrs[i++] = entry->addr;
639 /* If the array is empty, fill in the single
640 * 'IPADDRANY' local address
643 memset(&local, 0, sizeof(local));
644 local.family = msk->pm.remote.family;
653 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
655 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
656 struct sock *sk = (struct sock *)msk;
657 unsigned int add_addr_accept_max;
658 struct mptcp_addr_info remote;
659 unsigned int subflows_max;
662 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
663 subflows_max = mptcp_pm_get_subflows_max(msk);
665 pr_debug("accepted %d:%d remote family %d",
666 msk->pm.add_addr_accepted, add_addr_accept_max,
667 msk->pm.remote.family);
669 remote = msk->pm.remote;
670 mptcp_pm_announce_addr(msk, &remote, true);
671 mptcp_pm_nl_addr_send_ack(msk);
673 if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
676 /* pick id 0 port, if none is provided the remote address */
678 remote.port = sk->sk_dport;
680 /* connect to the specified remote address, using whatever
681 * local address the routing configuration will pick.
683 nr = fill_local_addresses_vec(msk, addrs);
685 msk->pm.add_addr_accepted++;
686 if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
687 msk->pm.subflows >= subflows_max)
688 WRITE_ONCE(msk->pm.accept_addr, false);
690 spin_unlock_bh(&msk->pm.lock);
691 for (i = 0; i < nr; i++)
692 __mptcp_subflow_connect(sk, &addrs[i], &remote);
693 spin_lock_bh(&msk->pm.lock);
696 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
698 struct mptcp_subflow_context *subflow;
700 msk_owned_by_me(msk);
701 lockdep_assert_held(&msk->pm.lock);
703 if (!mptcp_pm_should_add_signal(msk) &&
704 !mptcp_pm_should_rm_signal(msk))
707 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
709 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
711 spin_unlock_bh(&msk->pm.lock);
712 pr_debug("send ack for %s",
713 mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
715 mptcp_subflow_send_ack(ssk);
716 spin_lock_bh(&msk->pm.lock);
720 static int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
721 struct mptcp_addr_info *addr,
724 struct mptcp_subflow_context *subflow;
726 pr_debug("bkup=%d", bkup);
728 mptcp_for_each_subflow(msk, subflow) {
729 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
730 struct sock *sk = (struct sock *)msk;
731 struct mptcp_addr_info local;
733 local_address((struct sock_common *)ssk, &local);
734 if (!mptcp_addresses_equal(&local, addr, addr->port))
737 if (subflow->backup != bkup)
738 msk->last_snd = NULL;
739 subflow->backup = bkup;
740 subflow->send_mp_prio = 1;
741 subflow->request_bkup = bkup;
742 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
744 spin_unlock_bh(&msk->pm.lock);
745 pr_debug("send ack for mp_prio");
746 mptcp_subflow_send_ack(ssk);
747 spin_lock_bh(&msk->pm.lock);
755 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
756 const struct mptcp_rm_list *rm_list,
757 enum linux_mptcp_mib_field rm_type)
759 struct mptcp_subflow_context *subflow, *tmp;
760 struct sock *sk = (struct sock *)msk;
763 pr_debug("%s rm_list_nr %d",
764 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
766 msk_owned_by_me(msk);
768 if (sk->sk_state == TCP_LISTEN)
774 if (list_empty(&msk->conn_list))
777 for (i = 0; i < rm_list->nr; i++) {
778 bool removed = false;
780 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
781 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
782 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
783 u8 id = subflow->local_id;
785 if (rm_type == MPTCP_MIB_RMADDR)
786 id = subflow->remote_id;
788 if (rm_list->ids[i] != id)
791 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
792 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
793 i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
794 spin_unlock_bh(&msk->pm.lock);
795 mptcp_subflow_shutdown(sk, ssk, how);
797 /* the following takes care of updating the subflows counter */
798 mptcp_close_ssk(sk, ssk, subflow);
799 spin_lock_bh(&msk->pm.lock);
802 __MPTCP_INC_STATS(sock_net(sk), rm_type);
804 __set_bit(rm_list->ids[i], msk->pm.id_avail_bitmap);
808 if (!mptcp_pm_is_kernel(msk))
811 if (rm_type == MPTCP_MIB_RMADDR) {
812 msk->pm.add_addr_accepted--;
813 WRITE_ONCE(msk->pm.accept_addr, true);
814 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
815 msk->pm.local_addr_used--;
820 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
822 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
825 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
826 const struct mptcp_rm_list *rm_list)
828 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
831 void mptcp_pm_nl_work(struct mptcp_sock *msk)
833 struct mptcp_pm_data *pm = &msk->pm;
835 msk_owned_by_me(msk);
837 if (!(pm->status & MPTCP_PM_WORK_MASK))
840 spin_lock_bh(&msk->pm.lock);
842 pr_debug("msk=%p status=%x", msk, pm->status);
843 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
844 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
845 mptcp_pm_nl_add_addr_received(msk);
847 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
848 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
849 mptcp_pm_nl_addr_send_ack(msk);
851 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
852 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
853 mptcp_pm_nl_rm_addr_received(msk);
855 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
856 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
857 mptcp_pm_nl_fully_established(msk);
859 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
860 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
861 mptcp_pm_nl_subflow_established(msk);
864 spin_unlock_bh(&msk->pm.lock);
867 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
869 return (entry->flags &
870 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
871 MPTCP_PM_ADDR_FLAG_SIGNAL;
874 /* caller must ensure the RCU grace period is already elapsed */
875 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
878 sock_release(entry->lsk);
882 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
883 struct mptcp_pm_addr_entry *entry)
885 struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
886 unsigned int addr_max;
889 spin_lock_bh(&pernet->lock);
890 /* to keep the code simple, don't do IDR-like allocation for address ID,
891 * just bail when we exceed limits
893 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
895 if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
897 if (test_bit(entry->addr.id, pernet->id_bitmap))
900 /* do not insert duplicate address, differentiate on port only
903 list_for_each_entry(cur, &pernet->local_addr_list, list) {
904 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
905 address_use_port(entry) &&
906 address_use_port(cur))) {
907 /* allow replacing the exiting endpoint only if such
908 * endpoint is an implicit one and the user-space
909 * did not provide an endpoint id
911 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT))
917 entry->addr.id = cur->addr.id;
918 list_del_rcu(&cur->list);
924 if (!entry->addr.id) {
926 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
927 MPTCP_PM_MAX_ADDR_ID + 1,
929 if (!entry->addr.id && pernet->next_id != 1) {
938 __set_bit(entry->addr.id, pernet->id_bitmap);
939 if (entry->addr.id > pernet->next_id)
940 pernet->next_id = entry->addr.id;
942 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
943 addr_max = pernet->add_addr_signal_max;
944 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
946 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
947 addr_max = pernet->local_addr_max;
948 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
952 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
953 ret = entry->addr.id;
956 spin_unlock_bh(&pernet->lock);
958 /* just replaced an existing entry, free it */
961 __mptcp_pm_release_addr_entry(del_entry);
966 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
967 struct mptcp_pm_addr_entry *entry)
969 int addrlen = sizeof(struct sockaddr_in);
970 struct sockaddr_storage addr;
971 struct mptcp_sock *msk;
972 struct socket *ssock;
976 err = sock_create_kern(sock_net(sk), entry->addr.family,
977 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
981 msk = mptcp_sk(entry->lsk->sk);
987 ssock = __mptcp_nmpc_socket(msk);
993 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
994 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
995 if (entry->addr.family == AF_INET6)
996 addrlen = sizeof(struct sockaddr_in6);
998 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
1000 pr_warn("kernel_bind error, err=%d", err);
1004 err = kernel_listen(ssock, backlog);
1006 pr_warn("kernel_listen error, err=%d", err);
1013 sock_release(entry->lsk);
1017 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
1019 struct mptcp_pm_addr_entry *entry;
1020 struct mptcp_addr_info skc_local;
1021 struct mptcp_addr_info msk_local;
1022 struct pm_nl_pernet *pernet;
1025 if (WARN_ON_ONCE(!msk))
1028 /* The 0 ID mapping is defined by the first subflow, copied into the msk
1031 local_address((struct sock_common *)msk, &msk_local);
1032 local_address((struct sock_common *)skc, &skc_local);
1033 if (mptcp_addresses_equal(&msk_local, &skc_local, false))
1036 if (mptcp_pm_is_userspace(msk))
1037 return mptcp_userspace_pm_get_local_id(msk, &skc_local);
1039 pernet = pm_nl_get_pernet_from_msk(msk);
1042 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1043 if (mptcp_addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1044 ret = entry->addr.id;
1052 /* address not found, add to local list */
1053 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1057 entry->addr = skc_local;
1059 entry->addr.port = 0;
1061 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1063 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1070 #define MPTCP_PM_CMD_GRP_OFFSET 0
1071 #define MPTCP_PM_EV_GRP_OFFSET 1
1073 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1074 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, },
1075 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME,
1076 .flags = GENL_UNS_ADMIN_PERM,
1080 static const struct nla_policy
1081 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1082 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, },
1083 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, },
1084 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, },
1085 [MPTCP_PM_ADDR_ATTR_ADDR6] =
1086 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1087 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 },
1088 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 },
1089 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 },
1092 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1093 [MPTCP_PM_ATTR_ADDR] =
1094 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1095 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, },
1096 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, },
1097 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, },
1098 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, },
1099 [MPTCP_PM_ATTR_ADDR_REMOTE] =
1100 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1103 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1105 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1106 struct sock *sk = (struct sock *)msk;
1107 unsigned int active_max_loss_cnt;
1108 struct net *net = sock_net(sk);
1109 unsigned int stale_loss_cnt;
1112 stale_loss_cnt = mptcp_stale_loss_cnt(net);
1113 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1116 /* look for another available subflow not in loss state */
1117 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1118 mptcp_for_each_subflow(msk, iter) {
1119 if (iter != subflow && mptcp_subflow_active(iter) &&
1120 iter->stale_count < active_max_loss_cnt) {
1121 /* we have some alternatives, try to mark this subflow as idle ...*/
1122 slow = lock_sock_fast(ssk);
1123 if (!tcp_rtx_and_write_queues_empty(ssk)) {
1125 __mptcp_retransmit_pending_data(sk);
1126 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1128 unlock_sock_fast(ssk, slow);
1130 /* always try to push the pending data regarless of re-injections:
1131 * we can possibly use backup subflows now, and subflow selection
1132 * is cheap under the msk socket lock
1134 __mptcp_push_pending(sk, 0);
1140 static int mptcp_pm_family_to_addr(int family)
1142 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1143 if (family == AF_INET6)
1144 return MPTCP_PM_ADDR_ATTR_ADDR6;
1146 return MPTCP_PM_ADDR_ATTR_ADDR4;
1149 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1150 const struct nlattr *attr,
1151 struct genl_info *info,
1152 struct mptcp_addr_info *addr,
1153 bool require_family)
1158 GENL_SET_ERR_MSG(info, "missing address info");
1162 /* no validation needed - was already done via nested policy */
1163 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1164 mptcp_pm_addr_policy, info->extack);
1168 if (tb[MPTCP_PM_ADDR_ATTR_ID])
1169 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1171 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1172 if (!require_family)
1175 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1180 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1181 if (addr->family != AF_INET
1182 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1183 && addr->family != AF_INET6
1186 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1187 "unknown address family");
1190 addr_addr = mptcp_pm_family_to_addr(addr->family);
1191 if (!tb[addr_addr]) {
1192 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1193 "missing address data");
1197 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1198 if (addr->family == AF_INET6)
1199 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1202 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1204 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1205 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1210 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1211 struct mptcp_addr_info *addr)
1213 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1215 memset(addr, 0, sizeof(*addr));
1217 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1220 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1221 bool require_family,
1222 struct mptcp_pm_addr_entry *entry)
1224 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1227 memset(entry, 0, sizeof(*entry));
1229 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1233 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1234 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1236 entry->ifindex = val;
1239 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1240 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1242 if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1243 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1248 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1250 return pm_nl_get_pernet(genl_info_net(info));
1253 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1255 struct mptcp_sock *msk;
1256 long s_slot = 0, s_num = 0;
1258 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1259 struct sock *sk = (struct sock *)msk;
1261 if (!READ_ONCE(msk->fully_established) ||
1262 mptcp_pm_is_userspace(msk))
1266 spin_lock_bh(&msk->pm.lock);
1267 mptcp_pm_create_subflow_or_signal_addr(msk);
1268 spin_unlock_bh(&msk->pm.lock);
1279 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1281 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1282 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1283 struct mptcp_pm_addr_entry addr, *entry;
1286 ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1290 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1291 GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1295 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1296 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1297 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1301 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1302 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1306 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1308 GENL_SET_ERR_MSG(info, "can't allocate addr");
1313 if (entry->addr.port) {
1314 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1316 GENL_SET_ERR_MSG(info, "create listen socket error");
1321 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1323 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1325 sock_release(entry->lsk);
1330 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1335 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1336 u8 *flags, int *ifindex)
1338 struct mptcp_pm_addr_entry *entry;
1339 struct sock *sk = (struct sock *)msk;
1340 struct net *net = sock_net(sk);
1346 if (mptcp_pm_is_userspace(msk))
1347 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk,
1353 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1355 *flags = entry->flags;
1356 *ifindex = entry->ifindex;
1364 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1365 const struct mptcp_addr_info *addr)
1367 struct mptcp_pm_add_entry *entry;
1369 entry = mptcp_pm_del_add_timer(msk, addr, false);
1371 list_del(&entry->list);
1379 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1380 const struct mptcp_addr_info *addr,
1383 struct mptcp_rm_list list = { .nr = 0 };
1386 list.ids[list.nr++] = addr->id;
1388 ret = remove_anno_list_by_saddr(msk, addr);
1390 spin_lock_bh(&msk->pm.lock);
1391 mptcp_pm_remove_addr(msk, &list);
1392 spin_unlock_bh(&msk->pm.lock);
1397 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1398 const struct mptcp_pm_addr_entry *entry)
1400 const struct mptcp_addr_info *addr = &entry->addr;
1401 struct mptcp_rm_list list = { .nr = 0 };
1402 long s_slot = 0, s_num = 0;
1403 struct mptcp_sock *msk;
1405 pr_debug("remove_id=%d", addr->id);
1407 list.ids[list.nr++] = addr->id;
1409 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1410 struct sock *sk = (struct sock *)msk;
1411 bool remove_subflow;
1413 if (mptcp_pm_is_userspace(msk))
1416 if (list_empty(&msk->conn_list)) {
1417 mptcp_pm_remove_anno_addr(msk, addr, false);
1422 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1423 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1424 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1426 mptcp_pm_remove_subflow(msk, &list);
1437 static int mptcp_nl_remove_id_zero_address(struct net *net,
1438 struct mptcp_addr_info *addr)
1440 struct mptcp_rm_list list = { .nr = 0 };
1441 long s_slot = 0, s_num = 0;
1442 struct mptcp_sock *msk;
1444 list.ids[list.nr++] = 0;
1446 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1447 struct sock *sk = (struct sock *)msk;
1448 struct mptcp_addr_info msk_local;
1450 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1453 local_address((struct sock_common *)msk, &msk_local);
1454 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1458 spin_lock_bh(&msk->pm.lock);
1459 mptcp_pm_remove_addr(msk, &list);
1460 mptcp_pm_nl_rm_subflow_received(msk, &list);
1461 spin_unlock_bh(&msk->pm.lock);
1472 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1474 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1475 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1476 struct mptcp_pm_addr_entry addr, *entry;
1477 unsigned int addr_max;
1480 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1484 /* the zero id address is special: the first address used by the msk
1485 * always gets such an id, so different subflows can have different zero
1486 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1487 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1489 if (addr.addr.id == 0)
1490 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1492 spin_lock_bh(&pernet->lock);
1493 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1495 GENL_SET_ERR_MSG(info, "address not found");
1496 spin_unlock_bh(&pernet->lock);
1499 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1500 addr_max = pernet->add_addr_signal_max;
1501 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1503 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1504 addr_max = pernet->local_addr_max;
1505 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1509 list_del_rcu(&entry->list);
1510 __clear_bit(entry->addr.id, pernet->id_bitmap);
1511 spin_unlock_bh(&pernet->lock);
1513 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1515 __mptcp_pm_release_addr_entry(entry);
1520 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1521 struct list_head *rm_list)
1523 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1524 struct mptcp_pm_addr_entry *entry;
1526 list_for_each_entry(entry, rm_list, list) {
1527 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1528 slist.nr < MPTCP_RM_IDS_MAX)
1529 slist.ids[slist.nr++] = entry->addr.id;
1531 if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1532 alist.nr < MPTCP_RM_IDS_MAX)
1533 alist.ids[alist.nr++] = entry->addr.id;
1537 spin_lock_bh(&msk->pm.lock);
1538 mptcp_pm_remove_addr(msk, &alist);
1539 spin_unlock_bh(&msk->pm.lock);
1542 mptcp_pm_remove_subflow(msk, &slist);
1545 static void mptcp_nl_remove_addrs_list(struct net *net,
1546 struct list_head *rm_list)
1548 long s_slot = 0, s_num = 0;
1549 struct mptcp_sock *msk;
1551 if (list_empty(rm_list))
1554 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1555 struct sock *sk = (struct sock *)msk;
1557 if (!mptcp_pm_is_userspace(msk)) {
1559 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1568 /* caller must ensure the RCU grace period is already elapsed */
1569 static void __flush_addrs(struct list_head *list)
1571 while (!list_empty(list)) {
1572 struct mptcp_pm_addr_entry *cur;
1574 cur = list_entry(list->next,
1575 struct mptcp_pm_addr_entry, list);
1576 list_del_rcu(&cur->list);
1577 __mptcp_pm_release_addr_entry(cur);
1581 static void __reset_counters(struct pm_nl_pernet *pernet)
1583 WRITE_ONCE(pernet->add_addr_signal_max, 0);
1584 WRITE_ONCE(pernet->add_addr_accept_max, 0);
1585 WRITE_ONCE(pernet->local_addr_max, 0);
1589 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1591 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1592 LIST_HEAD(free_list);
1594 spin_lock_bh(&pernet->lock);
1595 list_splice_init(&pernet->local_addr_list, &free_list);
1596 __reset_counters(pernet);
1597 pernet->next_id = 1;
1598 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1599 spin_unlock_bh(&pernet->lock);
1600 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1602 __flush_addrs(&free_list);
1606 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1607 struct mptcp_pm_addr_entry *entry)
1609 struct mptcp_addr_info *addr = &entry->addr;
1610 struct nlattr *attr;
1612 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1616 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1617 goto nla_put_failure;
1618 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1619 goto nla_put_failure;
1620 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1621 goto nla_put_failure;
1622 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1623 goto nla_put_failure;
1624 if (entry->ifindex &&
1625 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1626 goto nla_put_failure;
1628 if (addr->family == AF_INET &&
1629 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1631 goto nla_put_failure;
1632 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1633 else if (addr->family == AF_INET6 &&
1634 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1635 goto nla_put_failure;
1637 nla_nest_end(skb, attr);
1641 nla_nest_cancel(skb, attr);
1645 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1647 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1648 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1649 struct mptcp_pm_addr_entry addr, *entry;
1650 struct sk_buff *msg;
1654 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1658 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1662 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1663 info->genlhdr->cmd);
1665 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1670 spin_lock_bh(&pernet->lock);
1671 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1673 GENL_SET_ERR_MSG(info, "address not found");
1678 ret = mptcp_nl_fill_addr(msg, entry);
1682 genlmsg_end(msg, reply);
1683 ret = genlmsg_reply(msg, info);
1684 spin_unlock_bh(&pernet->lock);
1688 spin_unlock_bh(&pernet->lock);
1695 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1696 struct netlink_callback *cb)
1698 struct net *net = sock_net(msg->sk);
1699 struct mptcp_pm_addr_entry *entry;
1700 struct pm_nl_pernet *pernet;
1701 int id = cb->args[0];
1705 pernet = pm_nl_get_pernet(net);
1707 spin_lock_bh(&pernet->lock);
1708 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1709 if (test_bit(i, pernet->id_bitmap)) {
1710 entry = __lookup_addr_by_id(pernet, i);
1714 if (entry->addr.id <= id)
1717 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1718 cb->nlh->nlmsg_seq, &mptcp_genl_family,
1719 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1723 if (mptcp_nl_fill_addr(msg, entry) < 0) {
1724 genlmsg_cancel(msg, hdr);
1728 id = entry->addr.id;
1729 genlmsg_end(msg, hdr);
1732 spin_unlock_bh(&pernet->lock);
1738 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1740 struct nlattr *attr = info->attrs[id];
1745 *limit = nla_get_u32(attr);
1746 if (*limit > MPTCP_PM_ADDR_MAX) {
1747 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1754 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1756 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1757 unsigned int rcv_addrs, subflows;
1760 spin_lock_bh(&pernet->lock);
1761 rcv_addrs = pernet->add_addr_accept_max;
1762 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1766 subflows = pernet->subflows_max;
1767 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1771 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1772 WRITE_ONCE(pernet->subflows_max, subflows);
1775 spin_unlock_bh(&pernet->lock);
1780 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1782 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1783 struct sk_buff *msg;
1786 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1790 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1791 MPTCP_PM_CMD_GET_LIMITS);
1795 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1796 READ_ONCE(pernet->add_addr_accept_max)))
1799 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1800 READ_ONCE(pernet->subflows_max)))
1803 genlmsg_end(msg, reply);
1804 return genlmsg_reply(msg, info);
1807 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1812 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1813 struct mptcp_addr_info *addr)
1815 struct mptcp_rm_list list = { .nr = 0 };
1817 list.ids[list.nr++] = addr->id;
1819 mptcp_pm_nl_rm_subflow_received(msk, &list);
1820 mptcp_pm_create_subflow_or_signal_addr(msk);
1823 static int mptcp_nl_set_flags(struct net *net,
1824 struct mptcp_addr_info *addr,
1825 u8 bkup, u8 changed)
1827 long s_slot = 0, s_num = 0;
1828 struct mptcp_sock *msk;
1831 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1832 struct sock *sk = (struct sock *)msk;
1834 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1838 spin_lock_bh(&msk->pm.lock);
1839 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1840 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1841 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1842 mptcp_pm_nl_fullmesh(msk, addr);
1843 spin_unlock_bh(&msk->pm.lock);
1854 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1856 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1857 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1858 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1859 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1860 MPTCP_PM_ADDR_FLAG_FULLMESH;
1861 struct net *net = sock_net(skb->sk);
1862 u8 bkup = 0, lookup_by_id = 0;
1865 ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1869 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1871 if (addr.addr.family == AF_UNSPEC) {
1877 spin_lock_bh(&pernet->lock);
1878 entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1880 spin_unlock_bh(&pernet->lock);
1883 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1884 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1885 spin_unlock_bh(&pernet->lock);
1889 changed = (addr.flags ^ entry->flags) & mask;
1890 entry->flags = (entry->flags & ~mask) | (addr.flags & mask);
1892 spin_unlock_bh(&pernet->lock);
1894 mptcp_nl_set_flags(net, &addr.addr, bkup, changed);
1898 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1900 genlmsg_multicast_netns(&mptcp_genl_family, net,
1901 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1904 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1906 return genl_has_listeners(&mptcp_genl_family,
1907 sock_net((const struct sock *)msk),
1908 MPTCP_PM_EV_GRP_OFFSET);
1911 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1913 const struct inet_sock *issk = inet_sk(ssk);
1914 const struct mptcp_subflow_context *sf;
1916 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1919 switch (ssk->sk_family) {
1921 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1923 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1926 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1928 const struct ipv6_pinfo *np = inet6_sk(ssk);
1930 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1932 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1942 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1944 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1947 sf = mptcp_subflow_ctx(ssk);
1948 if (WARN_ON_ONCE(!sf))
1951 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1954 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1960 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1961 const struct mptcp_sock *msk,
1962 const struct sock *ssk)
1964 const struct sock *sk = (const struct sock *)msk;
1965 const struct mptcp_subflow_context *sf;
1968 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1971 if (mptcp_event_add_subflow(skb, ssk))
1974 sf = mptcp_subflow_ctx(ssk);
1975 if (WARN_ON_ONCE(!sf))
1978 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1981 if (ssk->sk_bound_dev_if &&
1982 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1985 sk_err = ssk->sk_err;
1986 if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1987 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1993 static int mptcp_event_sub_established(struct sk_buff *skb,
1994 const struct mptcp_sock *msk,
1995 const struct sock *ssk)
1997 return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2000 static int mptcp_event_sub_closed(struct sk_buff *skb,
2001 const struct mptcp_sock *msk,
2002 const struct sock *ssk)
2004 const struct mptcp_subflow_context *sf;
2006 if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2009 sf = mptcp_subflow_ctx(ssk);
2010 if (!sf->reset_seen)
2013 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2016 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2022 static int mptcp_event_created(struct sk_buff *skb,
2023 const struct mptcp_sock *msk,
2024 const struct sock *ssk)
2026 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2031 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2034 return mptcp_event_add_subflow(skb, ssk);
2037 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2039 struct net *net = sock_net((const struct sock *)msk);
2040 struct nlmsghdr *nlh;
2041 struct sk_buff *skb;
2043 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2046 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2050 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2052 goto nla_put_failure;
2054 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2055 goto nla_put_failure;
2057 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2058 goto nla_put_failure;
2060 genlmsg_end(skb, nlh);
2061 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2068 void mptcp_event_addr_announced(const struct sock *ssk,
2069 const struct mptcp_addr_info *info)
2071 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2072 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2073 struct net *net = sock_net(ssk);
2074 struct nlmsghdr *nlh;
2075 struct sk_buff *skb;
2077 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2080 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2084 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2085 MPTCP_EVENT_ANNOUNCED);
2087 goto nla_put_failure;
2089 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2090 goto nla_put_failure;
2092 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2093 goto nla_put_failure;
2095 if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2097 inet_sk(ssk)->inet_dport :
2099 goto nla_put_failure;
2101 switch (info->family) {
2103 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2104 goto nla_put_failure;
2106 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2108 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2109 goto nla_put_failure;
2114 goto nla_put_failure;
2117 genlmsg_end(skb, nlh);
2118 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2125 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2126 const struct sock *ssk, gfp_t gfp)
2128 struct net *net = sock_net((const struct sock *)msk);
2129 struct nlmsghdr *nlh;
2130 struct sk_buff *skb;
2132 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2135 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2139 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2141 goto nla_put_failure;
2144 case MPTCP_EVENT_UNSPEC:
2147 case MPTCP_EVENT_CREATED:
2148 case MPTCP_EVENT_ESTABLISHED:
2149 if (mptcp_event_created(skb, msk, ssk) < 0)
2150 goto nla_put_failure;
2152 case MPTCP_EVENT_CLOSED:
2153 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2154 goto nla_put_failure;
2156 case MPTCP_EVENT_ANNOUNCED:
2157 case MPTCP_EVENT_REMOVED:
2158 /* call mptcp_event_addr_announced()/removed instead */
2161 case MPTCP_EVENT_SUB_ESTABLISHED:
2162 case MPTCP_EVENT_SUB_PRIORITY:
2163 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2164 goto nla_put_failure;
2166 case MPTCP_EVENT_SUB_CLOSED:
2167 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2168 goto nla_put_failure;
2172 genlmsg_end(skb, nlh);
2173 mptcp_nl_mcast_send(net, skb, gfp);
2180 static const struct genl_small_ops mptcp_pm_ops[] = {
2182 .cmd = MPTCP_PM_CMD_ADD_ADDR,
2183 .doit = mptcp_nl_cmd_add_addr,
2184 .flags = GENL_ADMIN_PERM,
2187 .cmd = MPTCP_PM_CMD_DEL_ADDR,
2188 .doit = mptcp_nl_cmd_del_addr,
2189 .flags = GENL_ADMIN_PERM,
2192 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS,
2193 .doit = mptcp_nl_cmd_flush_addrs,
2194 .flags = GENL_ADMIN_PERM,
2197 .cmd = MPTCP_PM_CMD_GET_ADDR,
2198 .doit = mptcp_nl_cmd_get_addr,
2199 .dumpit = mptcp_nl_cmd_dump_addrs,
2202 .cmd = MPTCP_PM_CMD_SET_LIMITS,
2203 .doit = mptcp_nl_cmd_set_limits,
2204 .flags = GENL_ADMIN_PERM,
2207 .cmd = MPTCP_PM_CMD_GET_LIMITS,
2208 .doit = mptcp_nl_cmd_get_limits,
2211 .cmd = MPTCP_PM_CMD_SET_FLAGS,
2212 .doit = mptcp_nl_cmd_set_flags,
2213 .flags = GENL_ADMIN_PERM,
2216 .cmd = MPTCP_PM_CMD_ANNOUNCE,
2217 .doit = mptcp_nl_cmd_announce,
2218 .flags = GENL_ADMIN_PERM,
2221 .cmd = MPTCP_PM_CMD_REMOVE,
2222 .doit = mptcp_nl_cmd_remove,
2223 .flags = GENL_ADMIN_PERM,
2226 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE,
2227 .doit = mptcp_nl_cmd_sf_create,
2228 .flags = GENL_ADMIN_PERM,
2231 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY,
2232 .doit = mptcp_nl_cmd_sf_destroy,
2233 .flags = GENL_ADMIN_PERM,
2237 static struct genl_family mptcp_genl_family __ro_after_init = {
2238 .name = MPTCP_PM_NAME,
2239 .version = MPTCP_PM_VER,
2240 .maxattr = MPTCP_PM_ATTR_MAX,
2241 .policy = mptcp_pm_policy,
2243 .module = THIS_MODULE,
2244 .small_ops = mptcp_pm_ops,
2245 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops),
2246 .mcgrps = mptcp_pm_mcgrps,
2247 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps),
2250 static int __net_init pm_nl_init_net(struct net *net)
2252 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2254 INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2256 /* Cit. 2 subflows ought to be enough for anybody. */
2257 pernet->subflows_max = 2;
2258 pernet->next_id = 1;
2259 pernet->stale_loss_cnt = 4;
2260 spin_lock_init(&pernet->lock);
2262 /* No need to initialize other pernet fields, the struct is zeroed at
2269 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2273 list_for_each_entry(net, net_list, exit_list) {
2274 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2276 /* net is removed from namespace list, can't race with
2277 * other modifiers, also netns core already waited for a
2280 __flush_addrs(&pernet->local_addr_list);
2284 static struct pernet_operations mptcp_pm_pernet_ops = {
2285 .init = pm_nl_init_net,
2286 .exit_batch = pm_nl_exit_net,
2287 .id = &pm_nl_pernet_id,
2288 .size = sizeof(struct pm_nl_pernet),
2291 void __init mptcp_pm_nl_init(void)
2293 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2294 panic("Failed to register MPTCP PM pernet subsystem.\n");
2296 if (genl_register_family(&mptcp_genl_family))
2297 panic("Failed to register MPTCP PM netlink family\n");