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;
41 #define MAX_ADDR_ID 255
42 #define BITMAP_SZ DIV_ROUND_UP(MAX_ADDR_ID + 1, BITS_PER_LONG)
45 /* protects pernet updates */
47 struct list_head local_addr_list;
49 unsigned int stale_loss_cnt;
50 unsigned int add_addr_signal_max;
51 unsigned int add_addr_accept_max;
52 unsigned int local_addr_max;
53 unsigned int subflows_max;
55 unsigned long id_bitmap[BITMAP_SZ];
58 #define MPTCP_PM_ADDR_MAX 8
59 #define ADD_ADDR_RETRANS_MAX 3
61 static bool addresses_equal(const struct mptcp_addr_info *a,
62 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 bool address_zero(const struct mptcp_addr_info *addr)
91 struct mptcp_addr_info zero;
93 memset(&zero, 0, sizeof(zero));
94 zero.family = addr->family;
96 return addresses_equal(addr, &zero, true);
99 static void local_address(const struct sock_common *skc,
100 struct mptcp_addr_info *addr)
102 addr->family = skc->skc_family;
103 addr->port = htons(skc->skc_num);
104 if (addr->family == AF_INET)
105 addr->addr.s_addr = skc->skc_rcv_saddr;
106 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
107 else if (addr->family == AF_INET6)
108 addr->addr6 = skc->skc_v6_rcv_saddr;
112 static void remote_address(const struct sock_common *skc,
113 struct mptcp_addr_info *addr)
115 addr->family = skc->skc_family;
116 addr->port = skc->skc_dport;
117 if (addr->family == AF_INET)
118 addr->addr.s_addr = skc->skc_daddr;
119 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
120 else if (addr->family == AF_INET6)
121 addr->addr6 = skc->skc_v6_daddr;
125 static bool lookup_subflow_by_saddr(const struct list_head *list,
126 struct mptcp_addr_info *saddr)
128 struct mptcp_subflow_context *subflow;
129 struct mptcp_addr_info cur;
130 struct sock_common *skc;
132 list_for_each_entry(subflow, list, node) {
133 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
135 local_address(skc, &cur);
136 if (addresses_equal(&cur, saddr, saddr->port))
143 static bool lookup_subflow_by_daddr(const struct list_head *list,
144 struct mptcp_addr_info *daddr)
146 struct mptcp_subflow_context *subflow;
147 struct mptcp_addr_info cur;
148 struct sock_common *skc;
150 list_for_each_entry(subflow, list, node) {
151 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
153 remote_address(skc, &cur);
154 if (addresses_equal(&cur, daddr, daddr->port))
161 static struct mptcp_pm_addr_entry *
162 select_local_address(const struct pm_nl_pernet *pernet,
163 struct mptcp_sock *msk)
165 struct mptcp_pm_addr_entry *entry, *ret = NULL;
166 struct sock *sk = (struct sock *)msk;
168 msk_owned_by_me(msk);
171 __mptcp_flush_join_list(msk);
172 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
173 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
176 if (entry->addr.family != sk->sk_family) {
177 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
178 if ((entry->addr.family == AF_INET &&
179 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
180 (sk->sk_family == AF_INET &&
181 !ipv6_addr_v4mapped(&entry->addr.addr6)))
186 /* avoid any address already in use by subflows and
189 if (!lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) {
198 static struct mptcp_pm_addr_entry *
199 select_signal_address(struct pm_nl_pernet *pernet, unsigned int pos)
201 struct mptcp_pm_addr_entry *entry, *ret = NULL;
205 /* do not keep any additional per socket state, just signal
206 * the address list in order.
207 * Note: removal from the local address list during the msk life-cycle
208 * can lead to additional addresses not being announced.
210 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
211 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
222 unsigned int mptcp_pm_get_add_addr_signal_max(struct mptcp_sock *msk)
224 struct pm_nl_pernet *pernet;
226 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
227 return READ_ONCE(pernet->add_addr_signal_max);
229 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
231 unsigned int mptcp_pm_get_add_addr_accept_max(struct mptcp_sock *msk)
233 struct pm_nl_pernet *pernet;
235 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
236 return READ_ONCE(pernet->add_addr_accept_max);
238 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
240 unsigned int mptcp_pm_get_subflows_max(struct mptcp_sock *msk)
242 struct pm_nl_pernet *pernet;
244 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
245 return READ_ONCE(pernet->subflows_max);
247 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
249 unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk)
251 struct pm_nl_pernet *pernet;
253 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
254 return READ_ONCE(pernet->local_addr_max);
256 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
258 static void check_work_pending(struct mptcp_sock *msk)
260 if (msk->pm.add_addr_signaled == mptcp_pm_get_add_addr_signal_max(msk) &&
261 (msk->pm.local_addr_used == mptcp_pm_get_local_addr_max(msk) ||
262 msk->pm.subflows == mptcp_pm_get_subflows_max(msk)))
263 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 struct sock *sk = (struct sock *)msk, *ssk;
433 struct mptcp_subflow_context *subflow;
434 struct mptcp_addr_info remote = { 0 };
435 unsigned int subflows_max;
438 subflows_max = mptcp_pm_get_subflows_max(msk);
440 /* Non-fullmesh endpoint, fill in the single entry
441 * corresponding to the primary MPC subflow remote address
444 remote_address((struct sock_common *)sk, &remote);
448 mptcp_for_each_subflow(msk, subflow) {
449 ssk = mptcp_subflow_tcp_sock(subflow);
450 remote_address((struct sock_common *)ssk, &remote);
451 if (!lookup_address_in_vec(addrs, i, &remote) &&
452 msk->pm.subflows < subflows_max) {
462 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
464 struct sock *sk = (struct sock *)msk;
465 struct mptcp_pm_addr_entry *local;
466 unsigned int add_addr_signal_max;
467 unsigned int local_addr_max;
468 struct pm_nl_pernet *pernet;
469 unsigned int subflows_max;
471 pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
473 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
474 local_addr_max = mptcp_pm_get_local_addr_max(msk);
475 subflows_max = mptcp_pm_get_subflows_max(msk);
477 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
478 msk->pm.local_addr_used, local_addr_max,
479 msk->pm.add_addr_signaled, add_addr_signal_max,
480 msk->pm.subflows, subflows_max);
482 /* check first for announce */
483 if (msk->pm.add_addr_signaled < add_addr_signal_max) {
484 local = select_signal_address(pernet,
485 msk->pm.add_addr_signaled);
488 if (mptcp_pm_alloc_anno_list(msk, local)) {
489 msk->pm.add_addr_signaled++;
490 mptcp_pm_announce_addr(msk, &local->addr, false);
491 mptcp_pm_nl_addr_send_ack(msk);
494 /* pick failed, avoid fourther attempts later */
495 msk->pm.local_addr_used = add_addr_signal_max;
498 check_work_pending(msk);
501 /* check if should create a new subflow */
502 if (msk->pm.local_addr_used < local_addr_max &&
503 msk->pm.subflows < subflows_max &&
504 !READ_ONCE(msk->pm.remote_deny_join_id0)) {
505 local = select_local_address(pernet, msk);
507 bool fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
508 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
511 msk->pm.local_addr_used++;
512 check_work_pending(msk);
513 nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
514 spin_unlock_bh(&msk->pm.lock);
515 for (i = 0; i < nr; i++)
516 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
517 spin_lock_bh(&msk->pm.lock);
521 /* lookup failed, avoid fourther attempts later */
522 msk->pm.local_addr_used = local_addr_max;
523 check_work_pending(msk);
527 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
529 mptcp_pm_create_subflow_or_signal_addr(msk);
532 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
534 mptcp_pm_create_subflow_or_signal_addr(msk);
537 /* Fill all the local addresses into the array addrs[],
538 * and return the array size.
540 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
541 struct mptcp_addr_info *addrs)
543 struct sock *sk = (struct sock *)msk;
544 struct mptcp_pm_addr_entry *entry;
545 struct mptcp_addr_info local;
546 struct pm_nl_pernet *pernet;
547 unsigned int subflows_max;
550 pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
551 subflows_max = mptcp_pm_get_subflows_max(msk);
554 __mptcp_flush_join_list(msk);
555 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
556 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
559 if (entry->addr.family != sk->sk_family) {
560 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
561 if ((entry->addr.family == AF_INET &&
562 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
563 (sk->sk_family == AF_INET &&
564 !ipv6_addr_v4mapped(&entry->addr.addr6)))
569 if (msk->pm.subflows < subflows_max) {
571 addrs[i++] = entry->addr;
576 /* If the array is empty, fill in the single
577 * 'IPADDRANY' local address
580 memset(&local, 0, sizeof(local));
581 local.family = msk->pm.remote.family;
590 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
592 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
593 struct sock *sk = (struct sock *)msk;
594 unsigned int add_addr_accept_max;
595 struct mptcp_addr_info remote;
596 unsigned int subflows_max;
599 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
600 subflows_max = mptcp_pm_get_subflows_max(msk);
602 pr_debug("accepted %d:%d remote family %d",
603 msk->pm.add_addr_accepted, add_addr_accept_max,
604 msk->pm.remote.family);
606 if (lookup_subflow_by_daddr(&msk->conn_list, &msk->pm.remote))
609 /* connect to the specified remote address, using whatever
610 * local address the routing configuration will pick.
612 remote = msk->pm.remote;
614 remote.port = sk->sk_dport;
615 nr = fill_local_addresses_vec(msk, addrs);
617 msk->pm.add_addr_accepted++;
618 if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
619 msk->pm.subflows >= subflows_max)
620 WRITE_ONCE(msk->pm.accept_addr, false);
622 spin_unlock_bh(&msk->pm.lock);
623 for (i = 0; i < nr; i++)
624 __mptcp_subflow_connect(sk, &addrs[i], &remote);
625 spin_lock_bh(&msk->pm.lock);
628 mptcp_pm_announce_addr(msk, &msk->pm.remote, true);
629 mptcp_pm_nl_addr_send_ack(msk);
632 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
634 struct mptcp_subflow_context *subflow;
636 msk_owned_by_me(msk);
637 lockdep_assert_held(&msk->pm.lock);
639 if (!mptcp_pm_should_add_signal(msk) &&
640 !mptcp_pm_should_rm_signal(msk))
643 __mptcp_flush_join_list(msk);
644 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
646 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
648 spin_unlock_bh(&msk->pm.lock);
649 pr_debug("send ack for %s",
650 mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
652 mptcp_subflow_send_ack(ssk);
653 spin_lock_bh(&msk->pm.lock);
657 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
658 struct mptcp_addr_info *addr,
661 struct mptcp_subflow_context *subflow;
663 pr_debug("bkup=%d", bkup);
665 mptcp_for_each_subflow(msk, subflow) {
666 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
667 struct sock *sk = (struct sock *)msk;
668 struct mptcp_addr_info local;
670 local_address((struct sock_common *)ssk, &local);
671 if (!addresses_equal(&local, addr, addr->port))
674 subflow->backup = bkup;
675 subflow->send_mp_prio = 1;
676 subflow->request_bkup = bkup;
677 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
679 spin_unlock_bh(&msk->pm.lock);
680 pr_debug("send ack for mp_prio");
681 mptcp_subflow_send_ack(ssk);
682 spin_lock_bh(&msk->pm.lock);
690 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
691 const struct mptcp_rm_list *rm_list,
692 enum linux_mptcp_mib_field rm_type)
694 struct mptcp_subflow_context *subflow, *tmp;
695 struct sock *sk = (struct sock *)msk;
698 pr_debug("%s rm_list_nr %d",
699 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
701 msk_owned_by_me(msk);
706 if (list_empty(&msk->conn_list))
709 for (i = 0; i < rm_list->nr; i++) {
710 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
711 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
712 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
713 u8 id = subflow->local_id;
715 if (rm_type == MPTCP_MIB_RMADDR)
716 id = subflow->remote_id;
718 if (rm_list->ids[i] != id)
721 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
722 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
723 i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
724 spin_unlock_bh(&msk->pm.lock);
725 mptcp_subflow_shutdown(sk, ssk, how);
726 mptcp_close_ssk(sk, ssk, subflow);
727 spin_lock_bh(&msk->pm.lock);
729 if (rm_type == MPTCP_MIB_RMADDR) {
730 msk->pm.add_addr_accepted--;
731 WRITE_ONCE(msk->pm.accept_addr, true);
732 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
733 msk->pm.local_addr_used--;
736 __MPTCP_INC_STATS(sock_net(sk), rm_type);
741 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
743 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
746 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
747 const struct mptcp_rm_list *rm_list)
749 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
752 void mptcp_pm_nl_work(struct mptcp_sock *msk)
754 struct mptcp_pm_data *pm = &msk->pm;
756 msk_owned_by_me(msk);
758 spin_lock_bh(&msk->pm.lock);
760 pr_debug("msk=%p status=%x", msk, pm->status);
761 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
762 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
763 mptcp_pm_nl_add_addr_received(msk);
765 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
766 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
767 mptcp_pm_nl_addr_send_ack(msk);
769 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
770 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
771 mptcp_pm_nl_rm_addr_received(msk);
773 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
774 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
775 mptcp_pm_nl_fully_established(msk);
777 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
778 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
779 mptcp_pm_nl_subflow_established(msk);
782 spin_unlock_bh(&msk->pm.lock);
785 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
787 return (entry->flags &
788 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
789 MPTCP_PM_ADDR_FLAG_SIGNAL;
792 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
793 struct mptcp_pm_addr_entry *entry)
795 struct mptcp_pm_addr_entry *cur;
796 unsigned int addr_max;
799 spin_lock_bh(&pernet->lock);
800 /* to keep the code simple, don't do IDR-like allocation for address ID,
801 * just bail when we exceed limits
803 if (pernet->next_id == MAX_ADDR_ID)
805 if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
807 if (test_bit(entry->addr.id, pernet->id_bitmap))
810 /* do not insert duplicate address, differentiate on port only
813 list_for_each_entry(cur, &pernet->local_addr_list, list) {
814 if (addresses_equal(&cur->addr, &entry->addr,
815 address_use_port(entry) &&
816 address_use_port(cur)))
820 if (!entry->addr.id) {
822 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
825 if ((!entry->addr.id || entry->addr.id > MAX_ADDR_ID) &&
826 pernet->next_id != 1) {
832 if (!entry->addr.id || entry->addr.id > MAX_ADDR_ID)
835 __set_bit(entry->addr.id, pernet->id_bitmap);
836 if (entry->addr.id > pernet->next_id)
837 pernet->next_id = entry->addr.id;
839 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
840 addr_max = pernet->add_addr_signal_max;
841 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
843 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
844 addr_max = pernet->local_addr_max;
845 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
849 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
850 ret = entry->addr.id;
853 spin_unlock_bh(&pernet->lock);
857 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
858 struct mptcp_pm_addr_entry *entry)
860 struct sockaddr_storage addr;
861 struct mptcp_sock *msk;
862 struct socket *ssock;
866 err = sock_create_kern(sock_net(sk), entry->addr.family,
867 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
871 msk = mptcp_sk(entry->lsk->sk);
877 ssock = __mptcp_nmpc_socket(msk);
883 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
884 err = kernel_bind(ssock, (struct sockaddr *)&addr,
885 sizeof(struct sockaddr_in));
887 pr_warn("kernel_bind error, err=%d", err);
891 err = kernel_listen(ssock, backlog);
893 pr_warn("kernel_listen error, err=%d", err);
900 sock_release(entry->lsk);
904 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
906 struct mptcp_pm_addr_entry *entry;
907 struct mptcp_addr_info skc_local;
908 struct mptcp_addr_info msk_local;
909 struct pm_nl_pernet *pernet;
912 if (WARN_ON_ONCE(!msk))
915 /* The 0 ID mapping is defined by the first subflow, copied into the msk
918 local_address((struct sock_common *)msk, &msk_local);
919 local_address((struct sock_common *)skc, &skc_local);
920 if (addresses_equal(&msk_local, &skc_local, false))
923 if (address_zero(&skc_local))
926 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
929 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
930 if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
931 ret = entry->addr.id;
939 /* address not found, add to local list */
940 entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
944 entry->addr = skc_local;
946 entry->addr.port = 0;
950 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
957 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
959 struct mptcp_pm_data *pm = &msk->pm;
962 subflows = !!mptcp_pm_get_subflows_max(msk);
963 WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
964 !!mptcp_pm_get_add_addr_signal_max(msk));
965 WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
966 WRITE_ONCE(pm->accept_subflow, subflows);
969 #define MPTCP_PM_CMD_GRP_OFFSET 0
970 #define MPTCP_PM_EV_GRP_OFFSET 1
972 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
973 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, },
974 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME,
975 .flags = GENL_UNS_ADMIN_PERM,
979 static const struct nla_policy
980 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
981 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, },
982 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, },
983 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, },
984 [MPTCP_PM_ADDR_ATTR_ADDR6] =
985 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
986 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 },
987 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 },
988 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 },
991 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
992 [MPTCP_PM_ATTR_ADDR] =
993 NLA_POLICY_NESTED(mptcp_pm_addr_policy),
994 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, },
995 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, },
998 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1000 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1001 struct sock *sk = (struct sock *)msk;
1002 unsigned int active_max_loss_cnt;
1003 struct net *net = sock_net(sk);
1004 unsigned int stale_loss_cnt;
1007 stale_loss_cnt = mptcp_stale_loss_cnt(net);
1008 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1011 /* look for another available subflow not in loss state */
1012 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1013 mptcp_for_each_subflow(msk, iter) {
1014 if (iter != subflow && mptcp_subflow_active(iter) &&
1015 iter->stale_count < active_max_loss_cnt) {
1016 /* we have some alternatives, try to mark this subflow as idle ...*/
1017 slow = lock_sock_fast(ssk);
1018 if (!tcp_rtx_and_write_queues_empty(ssk)) {
1020 __mptcp_retransmit_pending_data(sk);
1021 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1023 unlock_sock_fast(ssk, slow);
1025 /* always try to push the pending data regarless of re-injections:
1026 * we can possibly use backup subflows now, and subflow selection
1027 * is cheap under the msk socket lock
1029 __mptcp_push_pending(sk, 0);
1035 static int mptcp_pm_family_to_addr(int family)
1037 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1038 if (family == AF_INET6)
1039 return MPTCP_PM_ADDR_ATTR_ADDR6;
1041 return MPTCP_PM_ADDR_ATTR_ADDR4;
1044 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1045 bool require_family,
1046 struct mptcp_pm_addr_entry *entry)
1048 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1052 GENL_SET_ERR_MSG(info, "missing address info");
1056 /* no validation needed - was already done via nested policy */
1057 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1058 mptcp_pm_addr_policy, info->extack);
1062 memset(entry, 0, sizeof(*entry));
1063 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1064 if (!require_family)
1067 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1072 entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1073 if (entry->addr.family != AF_INET
1074 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1075 && entry->addr.family != AF_INET6
1078 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1079 "unknown address family");
1082 addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1083 if (!tb[addr_addr]) {
1084 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1085 "missing address data");
1089 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1090 if (entry->addr.family == AF_INET6)
1091 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1094 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1097 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1098 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1100 entry->ifindex = val;
1103 if (tb[MPTCP_PM_ADDR_ATTR_ID])
1104 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1106 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1107 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1109 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1110 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1111 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1112 "flags must have signal when using port");
1115 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1121 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1123 return net_generic(genl_info_net(info), pm_nl_pernet_id);
1126 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1128 struct mptcp_sock *msk;
1129 long s_slot = 0, s_num = 0;
1131 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1132 struct sock *sk = (struct sock *)msk;
1134 if (!READ_ONCE(msk->fully_established))
1138 spin_lock_bh(&msk->pm.lock);
1139 mptcp_pm_create_subflow_or_signal_addr(msk);
1140 spin_unlock_bh(&msk->pm.lock);
1151 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1153 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1154 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1155 struct mptcp_pm_addr_entry addr, *entry;
1158 ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1162 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1164 GENL_SET_ERR_MSG(info, "can't allocate addr");
1169 if (entry->addr.port) {
1170 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1172 GENL_SET_ERR_MSG(info, "create listen socket error");
1177 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1179 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1181 sock_release(entry->lsk);
1186 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1191 static struct mptcp_pm_addr_entry *
1192 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
1194 struct mptcp_pm_addr_entry *entry;
1196 list_for_each_entry(entry, &pernet->local_addr_list, list) {
1197 if (entry->addr.id == id)
1203 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1204 u8 *flags, int *ifindex)
1206 struct mptcp_pm_addr_entry *entry;
1213 entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1215 *flags = entry->flags;
1216 *ifindex = entry->ifindex;
1224 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1225 struct mptcp_addr_info *addr)
1227 struct mptcp_pm_add_entry *entry;
1229 entry = mptcp_pm_del_add_timer(msk, addr, false);
1231 list_del(&entry->list);
1239 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1240 struct mptcp_addr_info *addr,
1243 struct mptcp_rm_list list = { .nr = 0 };
1246 list.ids[list.nr++] = addr->id;
1248 ret = remove_anno_list_by_saddr(msk, addr);
1250 spin_lock_bh(&msk->pm.lock);
1251 mptcp_pm_remove_addr(msk, &list);
1252 spin_unlock_bh(&msk->pm.lock);
1257 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1258 struct mptcp_addr_info *addr)
1260 struct mptcp_sock *msk;
1261 long s_slot = 0, s_num = 0;
1262 struct mptcp_rm_list list = { .nr = 0 };
1264 pr_debug("remove_id=%d", addr->id);
1266 list.ids[list.nr++] = addr->id;
1268 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1269 struct sock *sk = (struct sock *)msk;
1270 bool remove_subflow;
1272 if (list_empty(&msk->conn_list)) {
1273 mptcp_pm_remove_anno_addr(msk, addr, false);
1278 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1279 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1281 mptcp_pm_remove_subflow(msk, &list);
1292 /* caller must ensure the RCU grace period is already elapsed */
1293 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1296 sock_release(entry->lsk);
1300 static int mptcp_nl_remove_id_zero_address(struct net *net,
1301 struct mptcp_addr_info *addr)
1303 struct mptcp_rm_list list = { .nr = 0 };
1304 long s_slot = 0, s_num = 0;
1305 struct mptcp_sock *msk;
1307 list.ids[list.nr++] = 0;
1309 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1310 struct sock *sk = (struct sock *)msk;
1311 struct mptcp_addr_info msk_local;
1313 if (list_empty(&msk->conn_list))
1316 local_address((struct sock_common *)msk, &msk_local);
1317 if (!addresses_equal(&msk_local, addr, addr->port))
1321 spin_lock_bh(&msk->pm.lock);
1322 mptcp_pm_remove_addr(msk, &list);
1323 mptcp_pm_nl_rm_subflow_received(msk, &list);
1324 spin_unlock_bh(&msk->pm.lock);
1335 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1337 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1338 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1339 struct mptcp_pm_addr_entry addr, *entry;
1340 unsigned int addr_max;
1343 ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1347 /* the zero id address is special: the first address used by the msk
1348 * always gets such an id, so different subflows can have different zero
1349 * id addresses. Additionally zero id is not accounted for in id_bitmap.
1350 * Let's use an 'mptcp_rm_list' instead of the common remove code.
1352 if (addr.addr.id == 0)
1353 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1355 spin_lock_bh(&pernet->lock);
1356 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1358 GENL_SET_ERR_MSG(info, "address not found");
1359 spin_unlock_bh(&pernet->lock);
1362 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1363 addr_max = pernet->add_addr_signal_max;
1364 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1366 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1367 addr_max = pernet->local_addr_max;
1368 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1372 list_del_rcu(&entry->list);
1373 __clear_bit(entry->addr.id, pernet->id_bitmap);
1374 spin_unlock_bh(&pernet->lock);
1376 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1378 __mptcp_pm_release_addr_entry(entry);
1383 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1384 struct list_head *rm_list)
1386 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1387 struct mptcp_pm_addr_entry *entry;
1389 list_for_each_entry(entry, rm_list, list) {
1390 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1391 alist.nr < MPTCP_RM_IDS_MAX &&
1392 slist.nr < MPTCP_RM_IDS_MAX) {
1393 alist.ids[alist.nr++] = entry->addr.id;
1394 slist.ids[slist.nr++] = entry->addr.id;
1395 } else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1396 alist.nr < MPTCP_RM_IDS_MAX) {
1397 alist.ids[alist.nr++] = entry->addr.id;
1402 spin_lock_bh(&msk->pm.lock);
1403 mptcp_pm_remove_addr(msk, &alist);
1404 spin_unlock_bh(&msk->pm.lock);
1407 mptcp_pm_remove_subflow(msk, &slist);
1410 static void mptcp_nl_remove_addrs_list(struct net *net,
1411 struct list_head *rm_list)
1413 long s_slot = 0, s_num = 0;
1414 struct mptcp_sock *msk;
1416 if (list_empty(rm_list))
1419 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1420 struct sock *sk = (struct sock *)msk;
1423 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1431 /* caller must ensure the RCU grace period is already elapsed */
1432 static void __flush_addrs(struct list_head *list)
1434 while (!list_empty(list)) {
1435 struct mptcp_pm_addr_entry *cur;
1437 cur = list_entry(list->next,
1438 struct mptcp_pm_addr_entry, list);
1439 list_del_rcu(&cur->list);
1440 __mptcp_pm_release_addr_entry(cur);
1444 static void __reset_counters(struct pm_nl_pernet *pernet)
1446 WRITE_ONCE(pernet->add_addr_signal_max, 0);
1447 WRITE_ONCE(pernet->add_addr_accept_max, 0);
1448 WRITE_ONCE(pernet->local_addr_max, 0);
1452 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1454 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1455 LIST_HEAD(free_list);
1457 spin_lock_bh(&pernet->lock);
1458 list_splice_init(&pernet->local_addr_list, &free_list);
1459 __reset_counters(pernet);
1460 pernet->next_id = 1;
1461 bitmap_zero(pernet->id_bitmap, MAX_ADDR_ID + 1);
1462 spin_unlock_bh(&pernet->lock);
1463 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1465 __flush_addrs(&free_list);
1469 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1470 struct mptcp_pm_addr_entry *entry)
1472 struct mptcp_addr_info *addr = &entry->addr;
1473 struct nlattr *attr;
1475 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1479 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1480 goto nla_put_failure;
1481 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1482 goto nla_put_failure;
1483 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1484 goto nla_put_failure;
1485 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1486 goto nla_put_failure;
1487 if (entry->ifindex &&
1488 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1489 goto nla_put_failure;
1491 if (addr->family == AF_INET &&
1492 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1494 goto nla_put_failure;
1495 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1496 else if (addr->family == AF_INET6 &&
1497 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1498 goto nla_put_failure;
1500 nla_nest_end(skb, attr);
1504 nla_nest_cancel(skb, attr);
1508 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1510 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1511 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1512 struct mptcp_pm_addr_entry addr, *entry;
1513 struct sk_buff *msg;
1517 ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1521 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1525 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1526 info->genlhdr->cmd);
1528 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1533 spin_lock_bh(&pernet->lock);
1534 entry = __lookup_addr_by_id(pernet, addr.addr.id);
1536 GENL_SET_ERR_MSG(info, "address not found");
1541 ret = mptcp_nl_fill_addr(msg, entry);
1545 genlmsg_end(msg, reply);
1546 ret = genlmsg_reply(msg, info);
1547 spin_unlock_bh(&pernet->lock);
1551 spin_unlock_bh(&pernet->lock);
1558 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1559 struct netlink_callback *cb)
1561 struct net *net = sock_net(msg->sk);
1562 struct mptcp_pm_addr_entry *entry;
1563 struct pm_nl_pernet *pernet;
1564 int id = cb->args[0];
1568 pernet = net_generic(net, pm_nl_pernet_id);
1570 spin_lock_bh(&pernet->lock);
1571 for (i = id; i < MAX_ADDR_ID + 1; i++) {
1572 if (test_bit(i, pernet->id_bitmap)) {
1573 entry = __lookup_addr_by_id(pernet, i);
1577 if (entry->addr.id <= id)
1580 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1581 cb->nlh->nlmsg_seq, &mptcp_genl_family,
1582 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1586 if (mptcp_nl_fill_addr(msg, entry) < 0) {
1587 genlmsg_cancel(msg, hdr);
1591 id = entry->addr.id;
1592 genlmsg_end(msg, hdr);
1595 spin_unlock_bh(&pernet->lock);
1601 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1603 struct nlattr *attr = info->attrs[id];
1608 *limit = nla_get_u32(attr);
1609 if (*limit > MPTCP_PM_ADDR_MAX) {
1610 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1617 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1619 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1620 unsigned int rcv_addrs, subflows;
1623 spin_lock_bh(&pernet->lock);
1624 rcv_addrs = pernet->add_addr_accept_max;
1625 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1629 subflows = pernet->subflows_max;
1630 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1634 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1635 WRITE_ONCE(pernet->subflows_max, subflows);
1638 spin_unlock_bh(&pernet->lock);
1643 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1645 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1646 struct sk_buff *msg;
1649 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1653 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1654 MPTCP_PM_CMD_GET_LIMITS);
1658 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1659 READ_ONCE(pernet->add_addr_accept_max)))
1662 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1663 READ_ONCE(pernet->subflows_max)))
1666 genlmsg_end(msg, reply);
1667 return genlmsg_reply(msg, info);
1670 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1675 static int mptcp_nl_addr_backup(struct net *net,
1676 struct mptcp_addr_info *addr,
1679 long s_slot = 0, s_num = 0;
1680 struct mptcp_sock *msk;
1683 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1684 struct sock *sk = (struct sock *)msk;
1686 if (list_empty(&msk->conn_list))
1690 spin_lock_bh(&msk->pm.lock);
1691 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1692 spin_unlock_bh(&msk->pm.lock);
1703 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1705 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1706 struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1707 struct mptcp_pm_addr_entry addr, *entry;
1708 struct net *net = sock_net(skb->sk);
1712 ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1716 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1719 list_for_each_entry(entry, &pernet->local_addr_list, list) {
1720 if (addresses_equal(&entry->addr, &addr.addr, true)) {
1721 mptcp_nl_addr_backup(net, &entry->addr, bkup);
1724 entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1726 entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1733 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1735 genlmsg_multicast_netns(&mptcp_genl_family, net,
1736 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1739 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1741 const struct inet_sock *issk = inet_sk(ssk);
1742 const struct mptcp_subflow_context *sf;
1744 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1747 switch (ssk->sk_family) {
1749 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1751 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1754 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1756 const struct ipv6_pinfo *np = inet6_sk(ssk);
1758 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1760 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1770 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1772 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1775 sf = mptcp_subflow_ctx(ssk);
1776 if (WARN_ON_ONCE(!sf))
1779 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1782 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1788 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1789 const struct mptcp_sock *msk,
1790 const struct sock *ssk)
1792 const struct sock *sk = (const struct sock *)msk;
1793 const struct mptcp_subflow_context *sf;
1796 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1799 if (mptcp_event_add_subflow(skb, ssk))
1802 sf = mptcp_subflow_ctx(ssk);
1803 if (WARN_ON_ONCE(!sf))
1806 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1809 if (ssk->sk_bound_dev_if &&
1810 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1813 sk_err = ssk->sk_err;
1814 if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1815 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1821 static int mptcp_event_sub_established(struct sk_buff *skb,
1822 const struct mptcp_sock *msk,
1823 const struct sock *ssk)
1825 return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1828 static int mptcp_event_sub_closed(struct sk_buff *skb,
1829 const struct mptcp_sock *msk,
1830 const struct sock *ssk)
1832 const struct mptcp_subflow_context *sf;
1834 if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1837 sf = mptcp_subflow_ctx(ssk);
1838 if (!sf->reset_seen)
1841 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1844 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1850 static int mptcp_event_created(struct sk_buff *skb,
1851 const struct mptcp_sock *msk,
1852 const struct sock *ssk)
1854 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1859 return mptcp_event_add_subflow(skb, ssk);
1862 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1864 struct net *net = sock_net((const struct sock *)msk);
1865 struct nlmsghdr *nlh;
1866 struct sk_buff *skb;
1868 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1871 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1875 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1877 goto nla_put_failure;
1879 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1880 goto nla_put_failure;
1882 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1883 goto nla_put_failure;
1885 genlmsg_end(skb, nlh);
1886 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1893 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1894 const struct mptcp_addr_info *info)
1896 struct net *net = sock_net((const struct sock *)msk);
1897 struct nlmsghdr *nlh;
1898 struct sk_buff *skb;
1900 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1903 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1907 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1908 MPTCP_EVENT_ANNOUNCED);
1910 goto nla_put_failure;
1912 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1913 goto nla_put_failure;
1915 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
1916 goto nla_put_failure;
1918 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
1919 goto nla_put_failure;
1921 switch (info->family) {
1923 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
1924 goto nla_put_failure;
1926 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1928 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
1929 goto nla_put_failure;
1934 goto nla_put_failure;
1937 genlmsg_end(skb, nlh);
1938 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1945 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
1946 const struct sock *ssk, gfp_t gfp)
1948 struct net *net = sock_net((const struct sock *)msk);
1949 struct nlmsghdr *nlh;
1950 struct sk_buff *skb;
1952 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1955 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
1959 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
1961 goto nla_put_failure;
1964 case MPTCP_EVENT_UNSPEC:
1967 case MPTCP_EVENT_CREATED:
1968 case MPTCP_EVENT_ESTABLISHED:
1969 if (mptcp_event_created(skb, msk, ssk) < 0)
1970 goto nla_put_failure;
1972 case MPTCP_EVENT_CLOSED:
1973 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
1974 goto nla_put_failure;
1976 case MPTCP_EVENT_ANNOUNCED:
1977 case MPTCP_EVENT_REMOVED:
1978 /* call mptcp_event_addr_announced()/removed instead */
1981 case MPTCP_EVENT_SUB_ESTABLISHED:
1982 case MPTCP_EVENT_SUB_PRIORITY:
1983 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
1984 goto nla_put_failure;
1986 case MPTCP_EVENT_SUB_CLOSED:
1987 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
1988 goto nla_put_failure;
1992 genlmsg_end(skb, nlh);
1993 mptcp_nl_mcast_send(net, skb, gfp);
2000 static const struct genl_small_ops mptcp_pm_ops[] = {
2002 .cmd = MPTCP_PM_CMD_ADD_ADDR,
2003 .doit = mptcp_nl_cmd_add_addr,
2004 .flags = GENL_ADMIN_PERM,
2007 .cmd = MPTCP_PM_CMD_DEL_ADDR,
2008 .doit = mptcp_nl_cmd_del_addr,
2009 .flags = GENL_ADMIN_PERM,
2012 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS,
2013 .doit = mptcp_nl_cmd_flush_addrs,
2014 .flags = GENL_ADMIN_PERM,
2017 .cmd = MPTCP_PM_CMD_GET_ADDR,
2018 .doit = mptcp_nl_cmd_get_addr,
2019 .dumpit = mptcp_nl_cmd_dump_addrs,
2022 .cmd = MPTCP_PM_CMD_SET_LIMITS,
2023 .doit = mptcp_nl_cmd_set_limits,
2024 .flags = GENL_ADMIN_PERM,
2027 .cmd = MPTCP_PM_CMD_GET_LIMITS,
2028 .doit = mptcp_nl_cmd_get_limits,
2031 .cmd = MPTCP_PM_CMD_SET_FLAGS,
2032 .doit = mptcp_nl_cmd_set_flags,
2033 .flags = GENL_ADMIN_PERM,
2037 static struct genl_family mptcp_genl_family __ro_after_init = {
2038 .name = MPTCP_PM_NAME,
2039 .version = MPTCP_PM_VER,
2040 .maxattr = MPTCP_PM_ATTR_MAX,
2041 .policy = mptcp_pm_policy,
2043 .module = THIS_MODULE,
2044 .small_ops = mptcp_pm_ops,
2045 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops),
2046 .mcgrps = mptcp_pm_mcgrps,
2047 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps),
2050 static int __net_init pm_nl_init_net(struct net *net)
2052 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2054 INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2055 pernet->next_id = 1;
2056 pernet->stale_loss_cnt = 4;
2057 spin_lock_init(&pernet->lock);
2059 /* No need to initialize other pernet fields, the struct is zeroed at
2066 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2070 list_for_each_entry(net, net_list, exit_list) {
2071 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2073 /* net is removed from namespace list, can't race with
2074 * other modifiers, also netns core already waited for a
2077 __flush_addrs(&pernet->local_addr_list);
2081 static struct pernet_operations mptcp_pm_pernet_ops = {
2082 .init = pm_nl_init_net,
2083 .exit_batch = pm_nl_exit_net,
2084 .id = &pm_nl_pernet_id,
2085 .size = sizeof(struct pm_nl_pernet),
2088 void __init mptcp_pm_nl_init(void)
2090 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2091 panic("Failed to register MPTCP PM pernet subsystem.\n");
2093 if (genl_register_family(&mptcp_genl_family))
2094 panic("Failed to register MPTCP PM netlink family\n");