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_addr_entry {
26 struct list_head list;
27 struct mptcp_addr_info addr;
33 struct mptcp_pm_add_entry {
34 struct list_head list;
35 struct mptcp_addr_info addr;
36 struct timer_list add_timer;
37 struct mptcp_sock *sock;
42 /* protects pernet updates */
44 struct list_head local_addr_list;
46 unsigned int stale_loss_cnt;
47 unsigned int add_addr_signal_max;
48 unsigned int add_addr_accept_max;
49 unsigned int local_addr_max;
50 unsigned int subflows_max;
52 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
55 #define MPTCP_PM_ADDR_MAX 8
56 #define ADD_ADDR_RETRANS_MAX 3
58 static bool addresses_equal(const struct mptcp_addr_info *a,
59 const struct mptcp_addr_info *b, bool use_port)
61 bool addr_equals = false;
63 if (a->family == b->family) {
64 if (a->family == AF_INET)
65 addr_equals = a->addr.s_addr == b->addr.s_addr;
66 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
68 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
69 } else if (a->family == AF_INET) {
70 if (ipv6_addr_v4mapped(&b->addr6))
71 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
72 } else if (b->family == AF_INET) {
73 if (ipv6_addr_v4mapped(&a->addr6))
74 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
83 return a->port == b->port;
86 static bool address_zero(const struct mptcp_addr_info *addr)
88 struct mptcp_addr_info zero;
90 memset(&zero, 0, sizeof(zero));
91 zero.family = addr->family;
93 return addresses_equal(addr, &zero, true);
96 static void local_address(const struct sock_common *skc,
97 struct mptcp_addr_info *addr)
99 addr->family = skc->skc_family;
100 addr->port = htons(skc->skc_num);
101 if (addr->family == AF_INET)
102 addr->addr.s_addr = skc->skc_rcv_saddr;
103 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
104 else if (addr->family == AF_INET6)
105 addr->addr6 = skc->skc_v6_rcv_saddr;
109 static void remote_address(const struct sock_common *skc,
110 struct mptcp_addr_info *addr)
112 addr->family = skc->skc_family;
113 addr->port = skc->skc_dport;
114 if (addr->family == AF_INET)
115 addr->addr.s_addr = skc->skc_daddr;
116 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
117 else if (addr->family == AF_INET6)
118 addr->addr6 = skc->skc_v6_daddr;
122 static bool lookup_subflow_by_saddr(const struct list_head *list,
123 struct mptcp_addr_info *saddr)
125 struct mptcp_subflow_context *subflow;
126 struct mptcp_addr_info cur;
127 struct sock_common *skc;
129 list_for_each_entry(subflow, list, node) {
130 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
132 local_address(skc, &cur);
133 if (addresses_equal(&cur, saddr, saddr->port))
140 static bool lookup_subflow_by_daddr(const struct list_head *list,
141 struct mptcp_addr_info *daddr)
143 struct mptcp_subflow_context *subflow;
144 struct mptcp_addr_info cur;
145 struct sock_common *skc;
147 list_for_each_entry(subflow, list, node) {
148 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
150 remote_address(skc, &cur);
151 if (addresses_equal(&cur, daddr, daddr->port))
158 static struct mptcp_pm_addr_entry *
159 select_local_address(const struct pm_nl_pernet *pernet,
160 struct mptcp_sock *msk)
162 struct mptcp_pm_addr_entry *entry, *ret = NULL;
163 struct sock *sk = (struct sock *)msk;
165 msk_owned_by_me(msk);
168 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
169 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
172 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
175 if (entry->addr.family != sk->sk_family) {
176 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
177 if ((entry->addr.family == AF_INET &&
178 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
179 (sk->sk_family == AF_INET &&
180 !ipv6_addr_v4mapped(&entry->addr.addr6)))
192 static struct mptcp_pm_addr_entry *
193 select_signal_address(struct pm_nl_pernet *pernet, struct mptcp_sock *msk)
195 struct mptcp_pm_addr_entry *entry, *ret = NULL;
198 /* do not keep any additional per socket state, just signal
199 * the address list in order.
200 * Note: removal from the local address list during the msk life-cycle
201 * can lead to additional addresses not being announced.
203 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
204 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
207 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
217 unsigned int mptcp_pm_get_add_addr_signal_max(struct mptcp_sock *msk)
219 struct pm_nl_pernet *pernet;
221 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
222 return READ_ONCE(pernet->add_addr_signal_max);
224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
226 unsigned int mptcp_pm_get_add_addr_accept_max(struct mptcp_sock *msk)
228 struct pm_nl_pernet *pernet;
230 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
231 return READ_ONCE(pernet->add_addr_accept_max);
233 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
235 unsigned int mptcp_pm_get_subflows_max(struct mptcp_sock *msk)
237 struct pm_nl_pernet *pernet;
239 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
240 return READ_ONCE(pernet->subflows_max);
242 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
244 unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk)
246 struct pm_nl_pernet *pernet;
248 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
249 return READ_ONCE(pernet->local_addr_max);
251 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
253 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
255 struct pm_nl_pernet *pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
257 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
258 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
259 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
260 WRITE_ONCE(msk->pm.work_pending, false);
266 struct mptcp_pm_add_entry *
267 mptcp_lookup_anno_list_by_saddr(struct mptcp_sock *msk,
268 struct mptcp_addr_info *addr)
270 struct mptcp_pm_add_entry *entry;
272 lockdep_assert_held(&msk->pm.lock);
274 list_for_each_entry(entry, &msk->pm.anno_list, list) {
275 if (addresses_equal(&entry->addr, addr, true))
282 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
284 struct mptcp_pm_add_entry *entry;
285 struct mptcp_addr_info saddr;
288 local_address((struct sock_common *)sk, &saddr);
290 spin_lock_bh(&msk->pm.lock);
291 list_for_each_entry(entry, &msk->pm.anno_list, list) {
292 if (addresses_equal(&entry->addr, &saddr, true)) {
299 spin_unlock_bh(&msk->pm.lock);
303 static void mptcp_pm_add_timer(struct timer_list *timer)
305 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
306 struct mptcp_sock *msk = entry->sock;
307 struct sock *sk = (struct sock *)msk;
309 pr_debug("msk=%p", msk);
314 if (inet_sk_state_load(sk) == TCP_CLOSE)
320 if (mptcp_pm_should_add_signal_addr(msk)) {
321 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
325 spin_lock_bh(&msk->pm.lock);
327 if (!mptcp_pm_should_add_signal_addr(msk)) {
328 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
329 mptcp_pm_announce_addr(msk, &entry->addr, false);
330 mptcp_pm_add_addr_send_ack(msk);
331 entry->retrans_times++;
334 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
335 sk_reset_timer(sk, timer,
336 jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
338 spin_unlock_bh(&msk->pm.lock);
340 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
341 mptcp_pm_subflow_established(msk);
347 struct mptcp_pm_add_entry *
348 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
349 struct mptcp_addr_info *addr, bool check_id)
351 struct mptcp_pm_add_entry *entry;
352 struct sock *sk = (struct sock *)msk;
354 spin_lock_bh(&msk->pm.lock);
355 entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
356 if (entry && (!check_id || entry->addr.id == addr->id))
357 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
358 spin_unlock_bh(&msk->pm.lock);
360 if (entry && (!check_id || entry->addr.id == addr->id))
361 sk_stop_timer_sync(sk, &entry->add_timer);
366 static bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
367 struct mptcp_pm_addr_entry *entry)
369 struct mptcp_pm_add_entry *add_entry = NULL;
370 struct sock *sk = (struct sock *)msk;
371 struct net *net = sock_net(sk);
373 lockdep_assert_held(&msk->pm.lock);
375 if (mptcp_lookup_anno_list_by_saddr(msk, &entry->addr))
378 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
382 list_add(&add_entry->list, &msk->pm.anno_list);
384 add_entry->addr = entry->addr;
385 add_entry->sock = msk;
386 add_entry->retrans_times = 0;
388 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
389 sk_reset_timer(sk, &add_entry->add_timer,
390 jiffies + mptcp_get_add_addr_timeout(net));
395 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
397 struct mptcp_pm_add_entry *entry, *tmp;
398 struct sock *sk = (struct sock *)msk;
399 LIST_HEAD(free_list);
401 pr_debug("msk=%p", msk);
403 spin_lock_bh(&msk->pm.lock);
404 list_splice_init(&msk->pm.anno_list, &free_list);
405 spin_unlock_bh(&msk->pm.lock);
407 list_for_each_entry_safe(entry, tmp, &free_list, list) {
408 sk_stop_timer_sync(sk, &entry->add_timer);
413 static bool lookup_address_in_vec(struct mptcp_addr_info *addrs, unsigned int nr,
414 struct mptcp_addr_info *addr)
418 for (i = 0; i < nr; i++) {
419 if (addresses_equal(&addrs[i], addr, addr->port))
426 /* Fill all the remote addresses into the array addrs[],
427 * and return the array size.
429 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, bool fullmesh,
430 struct mptcp_addr_info *addrs)
432 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
433 struct sock *sk = (struct sock *)msk, *ssk;
434 struct mptcp_subflow_context *subflow;
435 struct mptcp_addr_info remote = { 0 };
436 unsigned int subflows_max;
439 subflows_max = mptcp_pm_get_subflows_max(msk);
440 remote_address((struct sock_common *)sk, &remote);
442 /* Non-fullmesh endpoint, fill in the single entry
443 * corresponding to the primary MPC subflow remote address
452 mptcp_for_each_subflow(msk, subflow) {
453 ssk = mptcp_subflow_tcp_sock(subflow);
454 remote_address((struct sock_common *)ssk, &addrs[i]);
455 if (deny_id0 && addresses_equal(&addrs[i], &remote, false))
458 if (!lookup_address_in_vec(addrs, i, &addrs[i]) &&
459 msk->pm.subflows < subflows_max) {
469 static struct mptcp_pm_addr_entry *
470 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
472 struct mptcp_pm_addr_entry *entry;
474 list_for_each_entry(entry, &pernet->local_addr_list, list) {
475 if (entry->addr.id == id)
481 static struct mptcp_pm_addr_entry *
482 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
485 struct mptcp_pm_addr_entry *entry;
487 list_for_each_entry(entry, &pernet->local_addr_list, list) {
488 if ((!lookup_by_id && addresses_equal(&entry->addr, info, true)) ||
489 (lookup_by_id && entry->addr.id == info->id))
496 lookup_id_by_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *addr)
498 struct mptcp_pm_addr_entry *entry;
502 list_for_each_entry(entry, &pernet->local_addr_list, list) {
503 if (addresses_equal(&entry->addr, addr, entry->addr.port)) {
504 ret = entry->addr.id;
512 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
514 struct sock *sk = (struct sock *)msk;
515 struct mptcp_pm_addr_entry *local;
516 unsigned int add_addr_signal_max;
517 unsigned int local_addr_max;
518 struct pm_nl_pernet *pernet;
519 unsigned int subflows_max;
521 pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
523 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
524 local_addr_max = mptcp_pm_get_local_addr_max(msk);
525 subflows_max = mptcp_pm_get_subflows_max(msk);
527 /* do lazy endpoint usage accounting for the MPC subflows */
528 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
529 struct mptcp_addr_info mpc_addr;
532 local_address((struct sock_common *)msk->first, &mpc_addr);
533 mpc_id = lookup_id_by_addr(pernet, &mpc_addr);
535 __clear_bit(mpc_id, msk->pm.id_avail_bitmap);
537 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
540 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
541 msk->pm.local_addr_used, local_addr_max,
542 msk->pm.add_addr_signaled, add_addr_signal_max,
543 msk->pm.subflows, subflows_max);
545 /* check first for announce */
546 if (msk->pm.add_addr_signaled < add_addr_signal_max) {
547 local = select_signal_address(pernet, msk);
549 /* due to racing events on both ends we can reach here while
550 * previous add address is still running: if we invoke now
551 * mptcp_pm_announce_addr(), that will fail and the
552 * corresponding id will be marked as used.
553 * Instead let the PM machinery reschedule us when the
554 * current address announce will be completed.
556 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
560 if (mptcp_pm_alloc_anno_list(msk, local)) {
561 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
562 msk->pm.add_addr_signaled++;
563 mptcp_pm_announce_addr(msk, &local->addr, false);
564 mptcp_pm_nl_addr_send_ack(msk);
569 /* check if should create a new subflow */
570 while (msk->pm.local_addr_used < local_addr_max &&
571 msk->pm.subflows < subflows_max) {
572 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
576 local = select_local_address(pernet, msk);
580 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
582 msk->pm.local_addr_used++;
583 nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
585 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
586 spin_unlock_bh(&msk->pm.lock);
587 for (i = 0; i < nr; i++)
588 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
589 spin_lock_bh(&msk->pm.lock);
591 mptcp_pm_nl_check_work_pending(msk);
594 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
596 mptcp_pm_create_subflow_or_signal_addr(msk);
599 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
601 mptcp_pm_create_subflow_or_signal_addr(msk);
604 /* Fill all the local addresses into the array addrs[],
605 * and return the array size.
607 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
608 struct mptcp_addr_info *addrs)
610 struct sock *sk = (struct sock *)msk;
611 struct mptcp_pm_addr_entry *entry;
612 struct mptcp_addr_info local;
613 struct pm_nl_pernet *pernet;
614 unsigned int subflows_max;
617 pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
618 subflows_max = mptcp_pm_get_subflows_max(msk);
621 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
622 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
625 if (entry->addr.family != sk->sk_family) {
626 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
627 if ((entry->addr.family == AF_INET &&
628 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
629 (sk->sk_family == AF_INET &&
630 !ipv6_addr_v4mapped(&entry->addr.addr6)))
635 if (msk->pm.subflows < subflows_max) {
637 addrs[i++] = entry->addr;
642 /* If the array is empty, fill in the single
643 * 'IPADDRANY' local address
646 memset(&local, 0, sizeof(local));
647 local.family = msk->pm.remote.family;
656 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
658 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
659 struct sock *sk = (struct sock *)msk;
660 unsigned int add_addr_accept_max;
661 struct mptcp_addr_info remote;
662 unsigned int subflows_max;
665 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
666 subflows_max = mptcp_pm_get_subflows_max(msk);
668 pr_debug("accepted %d:%d remote family %d",
669 msk->pm.add_addr_accepted, add_addr_accept_max,
670 msk->pm.remote.family);
672 if (lookup_subflow_by_daddr(&msk->conn_list, &msk->pm.remote))
675 /* connect to the specified remote address, using whatever
676 * local address the routing configuration will pick.
678 remote = msk->pm.remote;
680 remote.port = sk->sk_dport;
681 nr = fill_local_addresses_vec(msk, addrs);
683 msk->pm.add_addr_accepted++;
684 if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
685 msk->pm.subflows >= subflows_max)
686 WRITE_ONCE(msk->pm.accept_addr, false);
688 spin_unlock_bh(&msk->pm.lock);
689 for (i = 0; i < nr; i++)
690 __mptcp_subflow_connect(sk, &addrs[i], &remote);
691 spin_lock_bh(&msk->pm.lock);
694 mptcp_pm_announce_addr(msk, &msk->pm.remote, true);
695 mptcp_pm_nl_addr_send_ack(msk);
698 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
700 struct mptcp_subflow_context *subflow;
702 msk_owned_by_me(msk);
703 lockdep_assert_held(&msk->pm.lock);
705 if (!mptcp_pm_should_add_signal(msk) &&
706 !mptcp_pm_should_rm_signal(msk))
709 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
711 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
713 spin_unlock_bh(&msk->pm.lock);
714 pr_debug("send ack for %s",
715 mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
717 mptcp_subflow_send_ack(ssk);
718 spin_lock_bh(&msk->pm.lock);
722 static int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
723 struct mptcp_addr_info *addr,
726 struct mptcp_subflow_context *subflow;
728 pr_debug("bkup=%d", bkup);
730 mptcp_for_each_subflow(msk, subflow) {
731 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
732 struct sock *sk = (struct sock *)msk;
733 struct mptcp_addr_info local;
735 local_address((struct sock_common *)ssk, &local);
736 if (!addresses_equal(&local, addr, addr->port))
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 (rm_type == MPTCP_MIB_RMADDR) {
809 msk->pm.add_addr_accepted--;
810 WRITE_ONCE(msk->pm.accept_addr, true);
811 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
812 msk->pm.local_addr_used--;
817 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
819 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
822 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
823 const struct mptcp_rm_list *rm_list)
825 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
828 void mptcp_pm_nl_work(struct mptcp_sock *msk)
830 struct mptcp_pm_data *pm = &msk->pm;
832 msk_owned_by_me(msk);
834 if (!(pm->status & MPTCP_PM_WORK_MASK))
837 spin_lock_bh(&msk->pm.lock);
839 pr_debug("msk=%p status=%x", msk, pm->status);
840 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
841 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
842 mptcp_pm_nl_add_addr_received(msk);
844 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
845 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
846 mptcp_pm_nl_addr_send_ack(msk);
848 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
849 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
850 mptcp_pm_nl_rm_addr_received(msk);
852 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
853 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
854 mptcp_pm_nl_fully_established(msk);
856 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
857 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
858 mptcp_pm_nl_subflow_established(msk);
861 spin_unlock_bh(&msk->pm.lock);
864 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
866 return (entry->flags &
867 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
868 MPTCP_PM_ADDR_FLAG_SIGNAL;
871 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
872 struct mptcp_pm_addr_entry *entry)
874 struct mptcp_pm_addr_entry *cur;
875 unsigned int addr_max;
878 spin_lock_bh(&pernet->lock);
879 /* to keep the code simple, don't do IDR-like allocation for address ID,
880 * just bail when we exceed limits
882 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
884 if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
886 if (test_bit(entry->addr.id, pernet->id_bitmap))
889 /* do not insert duplicate address, differentiate on port only
892 list_for_each_entry(cur, &pernet->local_addr_list, list) {
893 if (addresses_equal(&cur->addr, &entry->addr,
894 address_use_port(entry) &&
895 address_use_port(cur)))
899 if (!entry->addr.id) {
901 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
902 MPTCP_PM_MAX_ADDR_ID + 1,
904 if (!entry->addr.id && pernet->next_id != 1) {
913 __set_bit(entry->addr.id, pernet->id_bitmap);
914 if (entry->addr.id > pernet->next_id)
915 pernet->next_id = entry->addr.id;
917 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
918 addr_max = pernet->add_addr_signal_max;
919 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
921 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
922 addr_max = pernet->local_addr_max;
923 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
927 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
928 ret = entry->addr.id;
931 spin_unlock_bh(&pernet->lock);
935 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
936 struct mptcp_pm_addr_entry *entry)
938 int addrlen = sizeof(struct sockaddr_in);
939 struct sockaddr_storage addr;
940 struct mptcp_sock *msk;
941 struct socket *ssock;
945 err = sock_create_kern(sock_net(sk), entry->addr.family,
946 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
950 msk = mptcp_sk(entry->lsk->sk);
956 ssock = __mptcp_nmpc_socket(msk);
962 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
963 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
964 if (entry->addr.family == AF_INET6)
965 addrlen = sizeof(struct sockaddr_in6);
967 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
969 pr_warn("kernel_bind error, err=%d", err);
973 err = kernel_listen(ssock, backlog);
975 pr_warn("kernel_listen error, err=%d", err);
982 sock_release(entry->lsk);
986 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
988 struct mptcp_pm_addr_entry *entry;
989 struct mptcp_addr_info skc_local;
990 struct mptcp_addr_info msk_local;
991 struct pm_nl_pernet *pernet;
994 if (WARN_ON_ONCE(!msk))
997 /* The 0 ID mapping is defined by the first subflow, copied into the msk
1000 local_address((struct sock_common *)msk, &msk_local);
1001 local_address((struct sock_common *)skc, &skc_local);
1002 if (addresses_equal(&msk_local, &skc_local, false))
1005 if (address_zero(&skc_local))
1008 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
1011 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1012 if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1013 ret = entry->addr.id;
1021 /* address not found, add to local list */
1022 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1026 entry->addr = skc_local;
1028 entry->addr.port = 0;
1032 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1039 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
1041 struct mptcp_pm_data *pm = &msk->pm;
1044 subflows = !!mptcp_pm_get_subflows_max(msk);
1045 WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
1046 !!mptcp_pm_get_add_addr_signal_max(msk));
1047 WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
1048 WRITE_ONCE(pm->accept_subflow, subflows);
1051 #define MPTCP_PM_CMD_GRP_OFFSET 0
1052 #define MPTCP_PM_EV_GRP_OFFSET 1
1054 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1055 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, },
1056 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME,
1057 .flags = GENL_UNS_ADMIN_PERM,
1061 static const struct nla_policy
1062 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1063 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, },
1064 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, },
1065 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, },
1066 [MPTCP_PM_ADDR_ATTR_ADDR6] =
1067 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1068 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 },
1069 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 },
1070 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 },
1073 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1074 [MPTCP_PM_ATTR_ADDR] =
1075 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1076 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, },
1077 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, },
1080 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1082 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1083 struct sock *sk = (struct sock *)msk;
1084 unsigned int active_max_loss_cnt;
1085 struct net *net = sock_net(sk);
1086 unsigned int stale_loss_cnt;
1089 stale_loss_cnt = mptcp_stale_loss_cnt(net);
1090 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1093 /* look for another available subflow not in loss state */
1094 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1095 mptcp_for_each_subflow(msk, iter) {
1096 if (iter != subflow && mptcp_subflow_active(iter) &&
1097 iter->stale_count < active_max_loss_cnt) {
1098 /* we have some alternatives, try to mark this subflow as idle ...*/
1099 slow = lock_sock_fast(ssk);
1100 if (!tcp_rtx_and_write_queues_empty(ssk)) {
1102 __mptcp_retransmit_pending_data(sk);
1103 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1105 unlock_sock_fast(ssk, slow);
1107 /* always try to push the pending data regarless of re-injections:
1108 * we can possibly use backup subflows now, and subflow selection
1109 * is cheap under the msk socket lock
1111 __mptcp_push_pending(sk, 0);
1117 static int mptcp_pm_family_to_addr(int family)
1119 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1120 if (family == AF_INET6)
1121 return MPTCP_PM_ADDR_ATTR_ADDR6;
1123 return MPTCP_PM_ADDR_ATTR_ADDR4;
1126 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1127 bool require_family,
1128 struct mptcp_pm_addr_entry *entry)
1130 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1134 GENL_SET_ERR_MSG(info, "missing address info");
1138 /* no validation needed - was already done via nested policy */
1139 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1140 mptcp_pm_addr_policy, info->extack);
1144 memset(entry, 0, sizeof(*entry));
1145 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1146 if (!require_family)
1149 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1154 entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1155 if (entry->addr.family != AF_INET
1156 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1157 && entry->addr.family != AF_INET6
1160 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1161 "unknown address family");
1164 addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1165 if (!tb[addr_addr]) {
1166 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1167 "missing address data");
1171 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1172 if (entry->addr.family == AF_INET6)
1173 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1176 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1179 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1180 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1182 entry->ifindex = val;
1185 if (tb[MPTCP_PM_ADDR_ATTR_ID])
1186 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1188 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1189 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1191 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1192 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1193 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1194 "flags must have signal when using port");
1197 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1203 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1205 return net_generic(genl_info_net(info), pm_nl_pernet_id);
1208 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1210 struct mptcp_sock *msk;
1211 long s_slot = 0, s_num = 0;
1213 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1214 struct sock *sk = (struct sock *)msk;
1216 if (!READ_ONCE(msk->fully_established))
1220 spin_lock_bh(&msk->pm.lock);
1221 mptcp_pm_create_subflow_or_signal_addr(msk);
1222 spin_unlock_bh(&msk->pm.lock);
1233 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1235 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1236 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1237 struct mptcp_pm_addr_entry addr, *entry;
1240 ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1244 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1246 GENL_SET_ERR_MSG(info, "can't allocate addr");
1251 if (entry->addr.port) {
1252 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1254 GENL_SET_ERR_MSG(info, "create listen socket error");
1259 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1261 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1263 sock_release(entry->lsk);
1268 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1273 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1274 u8 *flags, int *ifindex)
1276 struct mptcp_pm_addr_entry *entry;
1283 entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1285 *flags = entry->flags;
1286 *ifindex = entry->ifindex;
1294 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1295 struct mptcp_addr_info *addr)
1297 struct mptcp_pm_add_entry *entry;
1299 entry = mptcp_pm_del_add_timer(msk, addr, false);
1301 list_del(&entry->list);
1309 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1310 struct mptcp_addr_info *addr,
1313 struct mptcp_rm_list list = { .nr = 0 };
1316 list.ids[list.nr++] = addr->id;
1318 ret = remove_anno_list_by_saddr(msk, addr);
1320 spin_lock_bh(&msk->pm.lock);
1321 mptcp_pm_remove_addr(msk, &list);
1322 spin_unlock_bh(&msk->pm.lock);
1327 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1328 struct mptcp_addr_info *addr)
1330 struct mptcp_sock *msk;
1331 long s_slot = 0, s_num = 0;
1332 struct mptcp_rm_list list = { .nr = 0 };
1334 pr_debug("remove_id=%d", addr->id);
1336 list.ids[list.nr++] = addr->id;
1338 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1339 struct sock *sk = (struct sock *)msk;
1340 bool remove_subflow;
1342 if (list_empty(&msk->conn_list)) {
1343 mptcp_pm_remove_anno_addr(msk, addr, false);
1348 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1349 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1351 mptcp_pm_remove_subflow(msk, &list);
1362 /* caller must ensure the RCU grace period is already elapsed */
1363 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1366 sock_release(entry->lsk);
1370 static int mptcp_nl_remove_id_zero_address(struct net *net,
1371 struct mptcp_addr_info *addr)
1373 struct mptcp_rm_list list = { .nr = 0 };
1374 long s_slot = 0, s_num = 0;
1375 struct mptcp_sock *msk;
1377 list.ids[list.nr++] = 0;
1379 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1380 struct sock *sk = (struct sock *)msk;
1381 struct mptcp_addr_info msk_local;
1383 if (list_empty(&msk->conn_list))
1386 local_address((struct sock_common *)msk, &msk_local);
1387 if (!addresses_equal(&msk_local, addr, addr->port))
1391 spin_lock_bh(&msk->pm.lock);
1392 mptcp_pm_remove_addr(msk, &list);
1393 mptcp_pm_nl_rm_subflow_received(msk, &list);
1394 spin_unlock_bh(&msk->pm.lock);
1405 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1407 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1408 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1409 struct mptcp_pm_addr_entry addr, *entry;
1410 unsigned int addr_max;
1413 ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1417 /* the zero id address is special: the first address used by the msk
1418 * always gets such an id, so different subflows can have different zero
1419 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1420 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1422 if (addr.addr.id == 0)
1423 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1425 spin_lock_bh(&pernet->lock);
1426 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1428 GENL_SET_ERR_MSG(info, "address not found");
1429 spin_unlock_bh(&pernet->lock);
1432 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1433 addr_max = pernet->add_addr_signal_max;
1434 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1436 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1437 addr_max = pernet->local_addr_max;
1438 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1442 list_del_rcu(&entry->list);
1443 __clear_bit(entry->addr.id, pernet->id_bitmap);
1444 spin_unlock_bh(&pernet->lock);
1446 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1448 __mptcp_pm_release_addr_entry(entry);
1453 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1454 struct list_head *rm_list)
1456 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1457 struct mptcp_pm_addr_entry *entry;
1459 list_for_each_entry(entry, rm_list, list) {
1460 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1461 alist.nr < MPTCP_RM_IDS_MAX &&
1462 slist.nr < MPTCP_RM_IDS_MAX) {
1463 alist.ids[alist.nr++] = entry->addr.id;
1464 slist.ids[slist.nr++] = entry->addr.id;
1465 } else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1466 alist.nr < MPTCP_RM_IDS_MAX) {
1467 alist.ids[alist.nr++] = entry->addr.id;
1472 spin_lock_bh(&msk->pm.lock);
1473 mptcp_pm_remove_addr(msk, &alist);
1474 spin_unlock_bh(&msk->pm.lock);
1477 mptcp_pm_remove_subflow(msk, &slist);
1480 static void mptcp_nl_remove_addrs_list(struct net *net,
1481 struct list_head *rm_list)
1483 long s_slot = 0, s_num = 0;
1484 struct mptcp_sock *msk;
1486 if (list_empty(rm_list))
1489 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1490 struct sock *sk = (struct sock *)msk;
1493 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1501 /* caller must ensure the RCU grace period is already elapsed */
1502 static void __flush_addrs(struct list_head *list)
1504 while (!list_empty(list)) {
1505 struct mptcp_pm_addr_entry *cur;
1507 cur = list_entry(list->next,
1508 struct mptcp_pm_addr_entry, list);
1509 list_del_rcu(&cur->list);
1510 __mptcp_pm_release_addr_entry(cur);
1514 static void __reset_counters(struct pm_nl_pernet *pernet)
1516 WRITE_ONCE(pernet->add_addr_signal_max, 0);
1517 WRITE_ONCE(pernet->add_addr_accept_max, 0);
1518 WRITE_ONCE(pernet->local_addr_max, 0);
1522 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1524 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1525 LIST_HEAD(free_list);
1527 spin_lock_bh(&pernet->lock);
1528 list_splice_init(&pernet->local_addr_list, &free_list);
1529 __reset_counters(pernet);
1530 pernet->next_id = 1;
1531 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1532 spin_unlock_bh(&pernet->lock);
1533 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1535 __flush_addrs(&free_list);
1539 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1540 struct mptcp_pm_addr_entry *entry)
1542 struct mptcp_addr_info *addr = &entry->addr;
1543 struct nlattr *attr;
1545 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1549 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1550 goto nla_put_failure;
1551 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1552 goto nla_put_failure;
1553 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1554 goto nla_put_failure;
1555 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1556 goto nla_put_failure;
1557 if (entry->ifindex &&
1558 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1559 goto nla_put_failure;
1561 if (addr->family == AF_INET &&
1562 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1564 goto nla_put_failure;
1565 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1566 else if (addr->family == AF_INET6 &&
1567 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1568 goto nla_put_failure;
1570 nla_nest_end(skb, attr);
1574 nla_nest_cancel(skb, attr);
1578 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1580 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1581 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1582 struct mptcp_pm_addr_entry addr, *entry;
1583 struct sk_buff *msg;
1587 ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1591 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1595 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1596 info->genlhdr->cmd);
1598 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1603 spin_lock_bh(&pernet->lock);
1604 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1606 GENL_SET_ERR_MSG(info, "address not found");
1611 ret = mptcp_nl_fill_addr(msg, entry);
1615 genlmsg_end(msg, reply);
1616 ret = genlmsg_reply(msg, info);
1617 spin_unlock_bh(&pernet->lock);
1621 spin_unlock_bh(&pernet->lock);
1628 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1629 struct netlink_callback *cb)
1631 struct net *net = sock_net(msg->sk);
1632 struct mptcp_pm_addr_entry *entry;
1633 struct pm_nl_pernet *pernet;
1634 int id = cb->args[0];
1638 pernet = net_generic(net, pm_nl_pernet_id);
1640 spin_lock_bh(&pernet->lock);
1641 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1642 if (test_bit(i, pernet->id_bitmap)) {
1643 entry = __lookup_addr_by_id(pernet, i);
1647 if (entry->addr.id <= id)
1650 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1651 cb->nlh->nlmsg_seq, &mptcp_genl_family,
1652 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1656 if (mptcp_nl_fill_addr(msg, entry) < 0) {
1657 genlmsg_cancel(msg, hdr);
1661 id = entry->addr.id;
1662 genlmsg_end(msg, hdr);
1665 spin_unlock_bh(&pernet->lock);
1671 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1673 struct nlattr *attr = info->attrs[id];
1678 *limit = nla_get_u32(attr);
1679 if (*limit > MPTCP_PM_ADDR_MAX) {
1680 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1687 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1689 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1690 unsigned int rcv_addrs, subflows;
1693 spin_lock_bh(&pernet->lock);
1694 rcv_addrs = pernet->add_addr_accept_max;
1695 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1699 subflows = pernet->subflows_max;
1700 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1704 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1705 WRITE_ONCE(pernet->subflows_max, subflows);
1708 spin_unlock_bh(&pernet->lock);
1713 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1715 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1716 struct sk_buff *msg;
1719 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1723 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1724 MPTCP_PM_CMD_GET_LIMITS);
1728 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1729 READ_ONCE(pernet->add_addr_accept_max)))
1732 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1733 READ_ONCE(pernet->subflows_max)))
1736 genlmsg_end(msg, reply);
1737 return genlmsg_reply(msg, info);
1740 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1745 static int mptcp_nl_addr_backup(struct net *net,
1746 struct mptcp_addr_info *addr,
1749 long s_slot = 0, s_num = 0;
1750 struct mptcp_sock *msk;
1753 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1754 struct sock *sk = (struct sock *)msk;
1756 if (list_empty(&msk->conn_list))
1760 spin_lock_bh(&msk->pm.lock);
1761 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1762 spin_unlock_bh(&msk->pm.lock);
1773 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1775 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1776 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1777 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1778 struct net *net = sock_net(skb->sk);
1779 u8 bkup = 0, lookup_by_id = 0;
1782 ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1786 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1788 if (addr.addr.family == AF_UNSPEC) {
1794 spin_lock_bh(&pernet->lock);
1795 entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1797 spin_unlock_bh(&pernet->lock);
1802 entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1804 entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1806 spin_unlock_bh(&pernet->lock);
1808 mptcp_nl_addr_backup(net, &addr.addr, bkup);
1812 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1814 genlmsg_multicast_netns(&mptcp_genl_family, net,
1815 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1818 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1820 const struct inet_sock *issk = inet_sk(ssk);
1821 const struct mptcp_subflow_context *sf;
1823 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1826 switch (ssk->sk_family) {
1828 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1830 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1833 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1835 const struct ipv6_pinfo *np = inet6_sk(ssk);
1837 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1839 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1849 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1851 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1854 sf = mptcp_subflow_ctx(ssk);
1855 if (WARN_ON_ONCE(!sf))
1858 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1861 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1867 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1868 const struct mptcp_sock *msk,
1869 const struct sock *ssk)
1871 const struct sock *sk = (const struct sock *)msk;
1872 const struct mptcp_subflow_context *sf;
1875 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1878 if (mptcp_event_add_subflow(skb, ssk))
1881 sf = mptcp_subflow_ctx(ssk);
1882 if (WARN_ON_ONCE(!sf))
1885 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1888 if (ssk->sk_bound_dev_if &&
1889 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1892 sk_err = ssk->sk_err;
1893 if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1894 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1900 static int mptcp_event_sub_established(struct sk_buff *skb,
1901 const struct mptcp_sock *msk,
1902 const struct sock *ssk)
1904 return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1907 static int mptcp_event_sub_closed(struct sk_buff *skb,
1908 const struct mptcp_sock *msk,
1909 const struct sock *ssk)
1911 const struct mptcp_subflow_context *sf;
1913 if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1916 sf = mptcp_subflow_ctx(ssk);
1917 if (!sf->reset_seen)
1920 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1923 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1929 static int mptcp_event_created(struct sk_buff *skb,
1930 const struct mptcp_sock *msk,
1931 const struct sock *ssk)
1933 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1938 return mptcp_event_add_subflow(skb, ssk);
1941 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1943 struct net *net = sock_net((const struct sock *)msk);
1944 struct nlmsghdr *nlh;
1945 struct sk_buff *skb;
1947 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1950 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1954 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1956 goto nla_put_failure;
1958 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1959 goto nla_put_failure;
1961 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1962 goto nla_put_failure;
1964 genlmsg_end(skb, nlh);
1965 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1972 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1973 const struct mptcp_addr_info *info)
1975 struct net *net = sock_net((const struct sock *)msk);
1976 struct nlmsghdr *nlh;
1977 struct sk_buff *skb;
1979 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1982 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1986 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1987 MPTCP_EVENT_ANNOUNCED);
1989 goto nla_put_failure;
1991 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1992 goto nla_put_failure;
1994 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
1995 goto nla_put_failure;
1997 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
1998 goto nla_put_failure;
2000 switch (info->family) {
2002 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2003 goto nla_put_failure;
2005 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2007 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2008 goto nla_put_failure;
2013 goto nla_put_failure;
2016 genlmsg_end(skb, nlh);
2017 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2024 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2025 const struct sock *ssk, gfp_t gfp)
2027 struct net *net = sock_net((const struct sock *)msk);
2028 struct nlmsghdr *nlh;
2029 struct sk_buff *skb;
2031 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2034 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2038 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2040 goto nla_put_failure;
2043 case MPTCP_EVENT_UNSPEC:
2046 case MPTCP_EVENT_CREATED:
2047 case MPTCP_EVENT_ESTABLISHED:
2048 if (mptcp_event_created(skb, msk, ssk) < 0)
2049 goto nla_put_failure;
2051 case MPTCP_EVENT_CLOSED:
2052 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2053 goto nla_put_failure;
2055 case MPTCP_EVENT_ANNOUNCED:
2056 case MPTCP_EVENT_REMOVED:
2057 /* call mptcp_event_addr_announced()/removed instead */
2060 case MPTCP_EVENT_SUB_ESTABLISHED:
2061 case MPTCP_EVENT_SUB_PRIORITY:
2062 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2063 goto nla_put_failure;
2065 case MPTCP_EVENT_SUB_CLOSED:
2066 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2067 goto nla_put_failure;
2071 genlmsg_end(skb, nlh);
2072 mptcp_nl_mcast_send(net, skb, gfp);
2079 static const struct genl_small_ops mptcp_pm_ops[] = {
2081 .cmd = MPTCP_PM_CMD_ADD_ADDR,
2082 .doit = mptcp_nl_cmd_add_addr,
2083 .flags = GENL_ADMIN_PERM,
2086 .cmd = MPTCP_PM_CMD_DEL_ADDR,
2087 .doit = mptcp_nl_cmd_del_addr,
2088 .flags = GENL_ADMIN_PERM,
2091 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS,
2092 .doit = mptcp_nl_cmd_flush_addrs,
2093 .flags = GENL_ADMIN_PERM,
2096 .cmd = MPTCP_PM_CMD_GET_ADDR,
2097 .doit = mptcp_nl_cmd_get_addr,
2098 .dumpit = mptcp_nl_cmd_dump_addrs,
2101 .cmd = MPTCP_PM_CMD_SET_LIMITS,
2102 .doit = mptcp_nl_cmd_set_limits,
2103 .flags = GENL_ADMIN_PERM,
2106 .cmd = MPTCP_PM_CMD_GET_LIMITS,
2107 .doit = mptcp_nl_cmd_get_limits,
2110 .cmd = MPTCP_PM_CMD_SET_FLAGS,
2111 .doit = mptcp_nl_cmd_set_flags,
2112 .flags = GENL_ADMIN_PERM,
2116 static struct genl_family mptcp_genl_family __ro_after_init = {
2117 .name = MPTCP_PM_NAME,
2118 .version = MPTCP_PM_VER,
2119 .maxattr = MPTCP_PM_ATTR_MAX,
2120 .policy = mptcp_pm_policy,
2122 .module = THIS_MODULE,
2123 .small_ops = mptcp_pm_ops,
2124 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops),
2125 .mcgrps = mptcp_pm_mcgrps,
2126 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps),
2129 static int __net_init pm_nl_init_net(struct net *net)
2131 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2133 INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2135 /* Cit. 2 subflows ought to be enough for anybody. */
2136 pernet->subflows_max = 2;
2137 pernet->next_id = 1;
2138 pernet->stale_loss_cnt = 4;
2139 spin_lock_init(&pernet->lock);
2141 /* No need to initialize other pernet fields, the struct is zeroed at
2148 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2152 list_for_each_entry(net, net_list, exit_list) {
2153 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2155 /* net is removed from namespace list, can't race with
2156 * other modifiers, also netns core already waited for a
2159 __flush_addrs(&pernet->local_addr_list);
2163 static struct pernet_operations mptcp_pm_pernet_ops = {
2164 .init = pm_nl_init_net,
2165 .exit_batch = pm_nl_exit_net,
2166 .id = &pm_nl_pernet_id,
2167 .size = sizeof(struct pm_nl_pernet),
2170 void __init mptcp_pm_nl_init(void)
2172 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2173 panic("Failed to register MPTCP PM pernet subsystem.\n");
2175 if (genl_register_family(&mptcp_genl_family))
2176 panic("Failed to register MPTCP PM netlink family\n");