1 // SPDX-License-Identifier: GPL-2.0
4 * Copyright (c) 2017 - 2019, Intel Corporation.
7 #define pr_fmt(fmt) "MPTCP: " fmt
9 #include <linux/kernel.h>
10 #include <linux/module.h>
11 #include <linux/netdevice.h>
12 #include <linux/sched/signal.h>
13 #include <linux/atomic.h>
15 #include <net/inet_common.h>
16 #include <net/inet_hashtables.h>
17 #include <net/protocol.h>
19 #include <net/tcp_states.h>
20 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
21 #include <net/transp_v6.h>
23 #include <net/mptcp.h>
28 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
30 struct mptcp_sock msk;
41 #define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
43 static struct percpu_counter mptcp_sockets_allocated;
45 static void __mptcp_destroy_sock(struct sock *sk);
46 static void __mptcp_check_send_data_fin(struct sock *sk);
48 /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
49 * completed yet or has failed, return the subflow socket.
50 * Otherwise return NULL.
52 static struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
54 if (!msk->subflow || READ_ONCE(msk->can_ack))
60 /* Returns end sequence number of the receiver's advertised window */
61 static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
63 return atomic64_read(&msk->wnd_end);
66 static bool mptcp_is_tcpsk(struct sock *sk)
68 struct socket *sock = sk->sk_socket;
70 if (unlikely(sk->sk_prot == &tcp_prot)) {
71 /* we are being invoked after mptcp_accept() has
72 * accepted a non-mp-capable flow: sk is a tcp_sk,
75 * Hand the socket over to tcp so all further socket ops
78 sock->ops = &inet_stream_ops;
80 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
81 } else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
82 sock->ops = &inet6_stream_ops;
90 static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk)
92 sock_owned_by_me((const struct sock *)msk);
94 if (likely(!__mptcp_check_fallback(msk)))
100 static int __mptcp_socket_create(struct mptcp_sock *msk)
102 struct mptcp_subflow_context *subflow;
103 struct sock *sk = (struct sock *)msk;
104 struct socket *ssock;
107 err = mptcp_subflow_create_socket(sk, &ssock);
111 msk->first = ssock->sk;
112 msk->subflow = ssock;
113 subflow = mptcp_subflow_ctx(ssock->sk);
114 list_add(&subflow->node, &msk->conn_list);
115 sock_hold(ssock->sk);
116 subflow->request_mptcp = 1;
118 /* accept() will wait on first subflow sk_wq, and we always wakes up
121 RCU_INIT_POINTER(msk->first->sk_wq, &sk->sk_socket->wq);
126 static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
128 sk_drops_add(sk, skb);
132 static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
133 struct sk_buff *from)
138 if (MPTCP_SKB_CB(from)->offset ||
139 !skb_try_coalesce(to, from, &fragstolen, &delta))
142 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
143 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
144 to->len, MPTCP_SKB_CB(from)->end_seq);
145 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
146 kfree_skb_partial(from, fragstolen);
147 atomic_add(delta, &sk->sk_rmem_alloc);
148 sk_mem_charge(sk, delta);
152 static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
153 struct sk_buff *from)
155 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
158 return mptcp_try_coalesce((struct sock *)msk, to, from);
161 /* "inspired" by tcp_data_queue_ofo(), main differences:
163 * - don't cope with sacks
165 static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
167 struct sock *sk = (struct sock *)msk;
168 struct rb_node **p, *parent;
169 u64 seq, end_seq, max_seq;
170 struct sk_buff *skb1;
172 seq = MPTCP_SKB_CB(skb)->map_seq;
173 end_seq = MPTCP_SKB_CB(skb)->end_seq;
174 max_seq = READ_ONCE(msk->rcv_wnd_sent);
176 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
177 RB_EMPTY_ROOT(&msk->out_of_order_queue));
178 if (after64(end_seq, max_seq)) {
181 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
182 (unsigned long long)end_seq - (unsigned long)max_seq,
183 (unsigned long long)msk->rcv_wnd_sent);
184 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
188 p = &msk->out_of_order_queue.rb_node;
189 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
190 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
191 rb_link_node(&skb->rbnode, NULL, p);
192 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
193 msk->ooo_last_skb = skb;
197 /* with 2 subflows, adding at end of ooo queue is quite likely
198 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
200 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
201 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
202 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
206 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
207 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
208 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
209 parent = &msk->ooo_last_skb->rbnode;
210 p = &parent->rb_right;
214 /* Find place to insert this segment. Handle overlaps on the way. */
218 skb1 = rb_to_skb(parent);
219 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
220 p = &parent->rb_left;
223 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
224 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
225 /* All the bits are present. Drop. */
227 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
230 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
234 * continue traversing
237 /* skb's seq == skb1's seq and skb covers skb1.
238 * Replace skb1 with skb.
240 rb_replace_node(&skb1->rbnode, &skb->rbnode,
241 &msk->out_of_order_queue);
242 mptcp_drop(sk, skb1);
243 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
246 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
247 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
250 p = &parent->rb_right;
254 /* Insert segment into RB tree. */
255 rb_link_node(&skb->rbnode, parent, p);
256 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
259 /* Remove other segments covered by skb. */
260 while ((skb1 = skb_rb_next(skb)) != NULL) {
261 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
263 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
264 mptcp_drop(sk, skb1);
265 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
267 /* If there is no skb after us, we are the last_skb ! */
269 msk->ooo_last_skb = skb;
273 skb_set_owner_r(skb, sk);
276 static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
277 struct sk_buff *skb, unsigned int offset,
280 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
281 struct sock *sk = (struct sock *)msk;
282 struct sk_buff *tail;
284 __skb_unlink(skb, &ssk->sk_receive_queue);
289 /* try to fetch required memory from subflow */
290 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
291 if (ssk->sk_forward_alloc < skb->truesize)
293 __sk_mem_reclaim(ssk, skb->truesize);
294 if (!sk_rmem_schedule(sk, skb, skb->truesize))
298 /* the skb map_seq accounts for the skb offset:
299 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
302 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
303 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
304 MPTCP_SKB_CB(skb)->offset = offset;
306 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
308 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
309 tail = skb_peek_tail(&sk->sk_receive_queue);
310 if (tail && mptcp_try_coalesce(sk, tail, skb))
313 skb_set_owner_r(skb, sk);
314 __skb_queue_tail(&sk->sk_receive_queue, skb);
316 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
317 mptcp_data_queue_ofo(msk, skb);
321 /* old data, keep it simple and drop the whole pkt, sender
322 * will retransmit as needed, if needed.
324 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
330 static void mptcp_stop_timer(struct sock *sk)
332 struct inet_connection_sock *icsk = inet_csk(sk);
334 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
335 mptcp_sk(sk)->timer_ival = 0;
338 static void mptcp_close_wake_up(struct sock *sk)
340 if (sock_flag(sk, SOCK_DEAD))
343 sk->sk_state_change(sk);
344 if (sk->sk_shutdown == SHUTDOWN_MASK ||
345 sk->sk_state == TCP_CLOSE)
346 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
348 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
351 static void mptcp_check_data_fin_ack(struct sock *sk)
353 struct mptcp_sock *msk = mptcp_sk(sk);
355 if (__mptcp_check_fallback(msk))
358 /* Look for an acknowledged DATA_FIN */
359 if (((1 << sk->sk_state) &
360 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
361 msk->write_seq == atomic64_read(&msk->snd_una)) {
362 mptcp_stop_timer(sk);
364 WRITE_ONCE(msk->snd_data_fin_enable, 0);
366 switch (sk->sk_state) {
368 inet_sk_state_store(sk, TCP_FIN_WAIT2);
372 inet_sk_state_store(sk, TCP_CLOSE);
376 mptcp_close_wake_up(sk);
380 static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
382 struct mptcp_sock *msk = mptcp_sk(sk);
384 if (READ_ONCE(msk->rcv_data_fin) &&
385 ((1 << sk->sk_state) &
386 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
387 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
389 if (msk->ack_seq == rcv_data_fin_seq) {
391 *seq = rcv_data_fin_seq;
400 static void mptcp_set_timeout(const struct sock *sk, const struct sock *ssk)
402 long tout = ssk && inet_csk(ssk)->icsk_pending ?
403 inet_csk(ssk)->icsk_timeout - jiffies : 0;
406 tout = mptcp_sk(sk)->timer_ival;
407 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
410 static bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
412 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
414 /* can't send if JOIN hasn't completed yet (i.e. is usable for mptcp) */
415 if (subflow->request_join && !subflow->fully_established)
418 /* only send if our side has not closed yet */
419 return ((1 << ssk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT));
422 static bool tcp_can_send_ack(const struct sock *ssk)
424 return !((1 << inet_sk_state_load(ssk)) &
425 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE));
428 static void mptcp_send_ack(struct mptcp_sock *msk)
430 struct mptcp_subflow_context *subflow;
432 mptcp_for_each_subflow(msk, subflow) {
433 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
436 if (tcp_can_send_ack(ssk))
442 static bool mptcp_subflow_cleanup_rbuf(struct sock *ssk)
447 ret = tcp_can_send_ack(ssk);
449 tcp_cleanup_rbuf(ssk, 1);
454 static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
456 struct sock *ack_hint = READ_ONCE(msk->ack_hint);
457 struct mptcp_subflow_context *subflow;
459 /* if the hinted ssk is still active, try to use it */
460 if (likely(ack_hint)) {
461 mptcp_for_each_subflow(msk, subflow) {
462 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
464 if (ack_hint == ssk && mptcp_subflow_cleanup_rbuf(ssk))
469 /* otherwise pick the first active subflow */
470 mptcp_for_each_subflow(msk, subflow)
471 if (mptcp_subflow_cleanup_rbuf(mptcp_subflow_tcp_sock(subflow)))
475 static bool mptcp_check_data_fin(struct sock *sk)
477 struct mptcp_sock *msk = mptcp_sk(sk);
478 u64 rcv_data_fin_seq;
481 if (__mptcp_check_fallback(msk) || !msk->first)
484 /* Need to ack a DATA_FIN received from a peer while this side
485 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
486 * msk->rcv_data_fin was set when parsing the incoming options
487 * at the subflow level and the msk lock was not held, so this
488 * is the first opportunity to act on the DATA_FIN and change
491 * If we are caught up to the sequence number of the incoming
492 * DATA_FIN, send the DATA_ACK now and do state transition. If
493 * not caught up, do nothing and let the recv code send DATA_ACK
497 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
498 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
499 WRITE_ONCE(msk->rcv_data_fin, 0);
501 sk->sk_shutdown |= RCV_SHUTDOWN;
502 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
503 set_bit(MPTCP_DATA_READY, &msk->flags);
505 switch (sk->sk_state) {
506 case TCP_ESTABLISHED:
507 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
510 inet_sk_state_store(sk, TCP_CLOSING);
513 inet_sk_state_store(sk, TCP_CLOSE);
516 /* Other states not expected */
522 mptcp_set_timeout(sk, NULL);
524 mptcp_close_wake_up(sk);
529 static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
533 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
534 struct sock *sk = (struct sock *)msk;
535 unsigned int moved = 0;
536 bool more_data_avail;
541 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
543 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
544 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
546 if (unlikely(ssk_rbuf > sk_rbuf)) {
547 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
552 pr_debug("msk=%p ssk=%p", msk, ssk);
555 u32 map_remaining, offset;
556 u32 seq = tp->copied_seq;
560 /* try to move as much data as available */
561 map_remaining = subflow->map_data_len -
562 mptcp_subflow_get_map_offset(subflow);
564 skb = skb_peek(&ssk->sk_receive_queue);
566 /* if no data is found, a racing workqueue/recvmsg
567 * already processed the new data, stop here or we
568 * can enter an infinite loop
575 if (__mptcp_check_fallback(msk)) {
576 /* if we are running under the workqueue, TCP could have
577 * collapsed skbs between dummy map creation and now
578 * be sure to adjust the size
580 map_remaining = skb->len;
581 subflow->map_data_len = skb->len;
584 offset = seq - TCP_SKB_CB(skb)->seq;
585 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
591 if (offset < skb->len) {
592 size_t len = skb->len - offset;
597 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
601 if (WARN_ON_ONCE(map_remaining < len))
605 sk_eat_skb(ssk, skb);
609 WRITE_ONCE(tp->copied_seq, seq);
610 more_data_avail = mptcp_subflow_data_available(ssk);
612 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
616 } while (more_data_avail);
617 WRITE_ONCE(msk->ack_hint, ssk);
623 static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
625 struct sock *sk = (struct sock *)msk;
626 struct sk_buff *skb, *tail;
631 p = rb_first(&msk->out_of_order_queue);
632 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
635 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
639 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
641 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
644 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
648 end_seq = MPTCP_SKB_CB(skb)->end_seq;
649 tail = skb_peek_tail(&sk->sk_receive_queue);
650 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
651 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
653 /* skip overlapping data, if any */
654 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
655 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
657 MPTCP_SKB_CB(skb)->offset += delta;
658 __skb_queue_tail(&sk->sk_receive_queue, skb);
660 msk->ack_seq = end_seq;
666 /* In most cases we will be able to lock the mptcp socket. If its already
667 * owned, we need to defer to the work queue to avoid ABBA deadlock.
669 static void move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
671 struct sock *sk = (struct sock *)msk;
672 unsigned int moved = 0;
674 if (inet_sk_state_load(sk) == TCP_CLOSE)
679 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
680 __mptcp_ofo_queue(msk);
682 /* If the moves have caught up with the DATA_FIN sequence number
683 * it's time to ack the DATA_FIN and change socket state, but
684 * this is not a good place to change state. Let the workqueue
687 if (mptcp_pending_data_fin(sk, NULL))
688 mptcp_schedule_work(sk);
689 mptcp_data_unlock(sk);
692 void mptcp_data_ready(struct sock *sk, struct sock *ssk)
694 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
695 struct mptcp_sock *msk = mptcp_sk(sk);
696 int sk_rbuf, ssk_rbuf;
699 /* move_skbs_to_msk below can legitly clear the data_avail flag,
700 * but we will need later to properly woke the reader, cache its
703 wake = subflow->data_avail == MPTCP_SUBFLOW_DATA_AVAIL;
705 set_bit(MPTCP_DATA_READY, &msk->flags);
707 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
708 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
709 if (unlikely(ssk_rbuf > sk_rbuf))
712 /* over limit? can't append more skbs to msk */
713 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf)
716 move_skbs_to_msk(msk, ssk);
720 sk->sk_data_ready(sk);
723 void __mptcp_flush_join_list(struct mptcp_sock *msk)
725 if (likely(list_empty(&msk->join_list)))
728 spin_lock_bh(&msk->join_list_lock);
729 list_splice_tail_init(&msk->join_list, &msk->conn_list);
730 spin_unlock_bh(&msk->join_list_lock);
733 static bool mptcp_timer_pending(struct sock *sk)
735 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
738 static void mptcp_reset_timer(struct sock *sk)
740 struct inet_connection_sock *icsk = inet_csk(sk);
743 /* prevent rescheduling on close */
744 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
747 /* should never be called with mptcp level timer cleared */
748 tout = READ_ONCE(mptcp_sk(sk)->timer_ival);
749 if (WARN_ON_ONCE(!tout))
751 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
754 bool mptcp_schedule_work(struct sock *sk)
756 if (inet_sk_state_load(sk) != TCP_CLOSE &&
757 schedule_work(&mptcp_sk(sk)->work)) {
758 /* each subflow already holds a reference to the sk, and the
759 * workqueue is invoked by a subflow, so sk can't go away here.
767 void mptcp_data_acked(struct sock *sk)
769 mptcp_reset_timer(sk);
771 if ((test_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags) ||
772 mptcp_send_head(sk) ||
773 (inet_sk_state_load(sk) != TCP_ESTABLISHED)))
774 mptcp_schedule_work(sk);
777 void mptcp_subflow_eof(struct sock *sk)
779 if (!test_and_set_bit(MPTCP_WORK_EOF, &mptcp_sk(sk)->flags))
780 mptcp_schedule_work(sk);
783 static void mptcp_check_for_eof(struct mptcp_sock *msk)
785 struct mptcp_subflow_context *subflow;
786 struct sock *sk = (struct sock *)msk;
789 mptcp_for_each_subflow(msk, subflow)
790 receivers += !subflow->rx_eof;
794 if (!(sk->sk_shutdown & RCV_SHUTDOWN)) {
795 /* hopefully temporary hack: propagate shutdown status
796 * to msk, when all subflows agree on it
798 sk->sk_shutdown |= RCV_SHUTDOWN;
800 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
801 set_bit(MPTCP_DATA_READY, &msk->flags);
802 sk->sk_data_ready(sk);
805 switch (sk->sk_state) {
806 case TCP_ESTABLISHED:
807 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
810 inet_sk_state_store(sk, TCP_CLOSING);
813 inet_sk_state_store(sk, TCP_CLOSE);
818 mptcp_close_wake_up(sk);
821 static bool mptcp_ext_cache_refill(struct mptcp_sock *msk)
823 const struct sock *sk = (const struct sock *)msk;
825 if (!msk->cached_ext)
826 msk->cached_ext = __skb_ext_alloc(sk->sk_allocation);
828 return !!msk->cached_ext;
831 static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
833 struct mptcp_subflow_context *subflow;
834 struct sock *sk = (struct sock *)msk;
836 sock_owned_by_me(sk);
838 mptcp_for_each_subflow(msk, subflow) {
839 if (subflow->data_avail)
840 return mptcp_subflow_tcp_sock(subflow);
846 static bool mptcp_skb_can_collapse_to(u64 write_seq,
847 const struct sk_buff *skb,
848 const struct mptcp_ext *mpext)
850 if (!tcp_skb_can_collapse_to(skb))
853 /* can collapse only if MPTCP level sequence is in order and this
854 * mapping has not been xmitted yet
856 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
860 static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
861 const struct page_frag *pfrag,
862 const struct mptcp_data_frag *df)
864 return df && pfrag->page == df->page &&
865 pfrag->size - pfrag->offset > 0 &&
866 df->data_seq + df->data_len == msk->write_seq;
869 static int mptcp_wmem_with_overhead(int size)
871 return size + ((sizeof(struct mptcp_data_frag) * size) >> PAGE_SHIFT);
874 static void __mptcp_wmem_reserve(struct sock *sk, int size)
876 int amount = mptcp_wmem_with_overhead(size);
877 struct mptcp_sock *msk = mptcp_sk(sk);
879 WARN_ON_ONCE(msk->wmem_reserved);
880 if (amount <= sk->sk_forward_alloc)
883 /* under memory pressure try to reserve at most a single page
884 * otherwise try to reserve the full estimate and fallback
885 * to a single page before entering the error path
887 if ((tcp_under_memory_pressure(sk) && amount > PAGE_SIZE) ||
888 !sk_wmem_schedule(sk, amount)) {
889 if (amount <= PAGE_SIZE)
893 if (!sk_wmem_schedule(sk, amount))
898 msk->wmem_reserved = amount;
899 sk->sk_forward_alloc -= amount;
903 /* we will wait for memory on next allocation */
904 msk->wmem_reserved = -1;
907 static void __mptcp_update_wmem(struct sock *sk)
909 struct mptcp_sock *msk = mptcp_sk(sk);
911 if (!msk->wmem_reserved)
914 if (msk->wmem_reserved < 0)
915 msk->wmem_reserved = 0;
916 if (msk->wmem_reserved > 0) {
917 sk->sk_forward_alloc += msk->wmem_reserved;
918 msk->wmem_reserved = 0;
922 static bool mptcp_wmem_alloc(struct sock *sk, int size)
924 struct mptcp_sock *msk = mptcp_sk(sk);
926 /* check for pre-existing error condition */
927 if (msk->wmem_reserved < 0)
930 if (msk->wmem_reserved >= size)
934 if (!sk_wmem_schedule(sk, size)) {
935 mptcp_data_unlock(sk);
939 sk->sk_forward_alloc -= size;
940 msk->wmem_reserved += size;
941 mptcp_data_unlock(sk);
944 msk->wmem_reserved -= size;
948 static void mptcp_wmem_uncharge(struct sock *sk, int size)
950 struct mptcp_sock *msk = mptcp_sk(sk);
952 if (msk->wmem_reserved < 0)
953 msk->wmem_reserved = 0;
954 msk->wmem_reserved += size;
957 static void dfrag_uncharge(struct sock *sk, int len)
959 sk_mem_uncharge(sk, len);
960 sk_wmem_queued_add(sk, -len);
963 static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
965 int len = dfrag->data_len + dfrag->overhead;
967 list_del(&dfrag->list);
968 dfrag_uncharge(sk, len);
969 put_page(dfrag->page);
972 static void mptcp_clean_una(struct sock *sk)
974 struct mptcp_sock *msk = mptcp_sk(sk);
975 struct mptcp_data_frag *dtmp, *dfrag;
976 bool cleaned = false;
979 /* on fallback we just need to ignore snd_una, as this is really
982 if (__mptcp_check_fallback(msk))
983 atomic64_set(&msk->snd_una, msk->snd_nxt);
986 snd_una = atomic64_read(&msk->snd_una);
988 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
989 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
992 if (WARN_ON_ONCE(dfrag == msk->first_pending))
994 dfrag_clear(sk, dfrag);
998 dfrag = mptcp_rtx_head(sk);
999 if (dfrag && after64(snd_una, dfrag->data_seq)) {
1000 u64 delta = snd_una - dfrag->data_seq;
1002 if (WARN_ON_ONCE(delta > dfrag->already_sent))
1005 dfrag->data_seq += delta;
1006 dfrag->offset += delta;
1007 dfrag->data_len -= delta;
1008 dfrag->already_sent -= delta;
1010 dfrag_uncharge(sk, delta);
1015 if (cleaned && tcp_under_memory_pressure(sk))
1016 sk_mem_reclaim_partial(sk);
1017 mptcp_data_unlock(sk);
1020 static void mptcp_clean_una_wakeup(struct sock *sk)
1022 struct mptcp_sock *msk = mptcp_sk(sk);
1024 mptcp_clean_una(sk);
1026 /* Only wake up writers if a subflow is ready */
1027 if (sk_stream_is_writeable(sk)) {
1028 clear_bit(MPTCP_NOSPACE, &msk->flags);
1029 sk_stream_write_space(sk);
1033 /* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1036 static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1038 struct mptcp_subflow_context *subflow;
1039 struct mptcp_sock *msk = mptcp_sk(sk);
1042 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1043 pfrag, sk->sk_allocation)))
1046 sk_stream_moderate_sndbuf(sk);
1047 mptcp_for_each_subflow(msk, subflow) {
1048 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1051 tcp_enter_memory_pressure(ssk);
1052 sk_stream_moderate_sndbuf(ssk);
1058 static struct mptcp_data_frag *
1059 mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1062 int offset = ALIGN(orig_offset, sizeof(long));
1063 struct mptcp_data_frag *dfrag;
1065 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1066 dfrag->data_len = 0;
1067 dfrag->data_seq = msk->write_seq;
1068 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1069 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
1070 dfrag->already_sent = 0;
1071 dfrag->page = pfrag->page;
1076 struct mptcp_sendmsg_info {
1084 static int mptcp_check_allowed_size(struct mptcp_sock *msk, u64 data_seq,
1087 u64 window_end = mptcp_wnd_end(msk);
1089 if (__mptcp_check_fallback(msk))
1092 if (!before64(data_seq + avail_size, window_end)) {
1093 u64 allowed_size = window_end - data_seq;
1095 return min_t(unsigned int, allowed_size, avail_size);
1101 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
1102 struct mptcp_data_frag *dfrag,
1103 struct mptcp_sendmsg_info *info)
1105 u64 data_seq = dfrag->data_seq + info->sent;
1106 struct mptcp_sock *msk = mptcp_sk(sk);
1107 bool zero_window_probe = false;
1108 struct mptcp_ext *mpext = NULL;
1109 struct sk_buff *skb, *tail;
1110 bool can_collapse = false;
1114 pr_debug("msk=%p ssk=%p sending dfrag at seq=%lld len=%d already sent=%d",
1115 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1117 /* compute send limit */
1118 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
1119 avail_size = info->size_goal;
1120 skb = tcp_write_queue_tail(ssk);
1122 /* Limit the write to the size available in the
1123 * current skb, if any, so that we create at most a new skb.
1124 * Explicitly tells TCP internals to avoid collapsing on later
1125 * queue management operation, to avoid breaking the ext <->
1126 * SSN association set here
1128 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1129 can_collapse = (info->size_goal - skb->len > 0) &&
1130 mptcp_skb_can_collapse_to(data_seq, skb, mpext);
1132 TCP_SKB_CB(skb)->eor = 1;
1134 avail_size = info->size_goal - skb->len;
1137 /* Zero window and all data acked? Probe. */
1138 avail_size = mptcp_check_allowed_size(msk, data_seq, avail_size);
1139 if (avail_size == 0) {
1140 if (skb || atomic64_read(&msk->snd_una) != msk->snd_nxt)
1142 zero_window_probe = true;
1143 data_seq = atomic64_read(&msk->snd_una) - 1;
1147 if (WARN_ON_ONCE(info->sent > info->limit ||
1148 info->limit > dfrag->data_len))
1151 ret = info->limit - info->sent;
1152 tail = tcp_build_frag(ssk, avail_size, info->flags, dfrag->page,
1153 dfrag->offset + info->sent, &ret);
1155 tcp_remove_empty_skb(sk, tcp_write_queue_tail(ssk));
1159 /* if the tail skb is still the cached one, collapsing really happened.
1162 WARN_ON_ONCE(!can_collapse);
1163 mpext->data_len += ret;
1164 WARN_ON_ONCE(zero_window_probe);
1168 mpext = __skb_ext_set(tail, SKB_EXT_MPTCP, msk->cached_ext);
1169 msk->cached_ext = NULL;
1171 memset(mpext, 0, sizeof(*mpext));
1172 mpext->data_seq = data_seq;
1173 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
1174 mpext->data_len = ret;
1178 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1179 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1182 if (zero_window_probe) {
1183 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
1186 tcp_push_pending_frames(ssk);
1189 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
1193 static void mptcp_nospace(struct mptcp_sock *msk)
1195 struct mptcp_subflow_context *subflow;
1197 set_bit(MPTCP_NOSPACE, &msk->flags);
1198 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
1200 mptcp_for_each_subflow(msk, subflow) {
1201 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1202 bool ssk_writeable = sk_stream_is_writeable(ssk);
1203 struct socket *sock = READ_ONCE(ssk->sk_socket);
1205 if (ssk_writeable || !sock)
1208 /* enables ssk->write_space() callbacks */
1209 set_bit(SOCK_NOSPACE, &sock->flags);
1212 /* mptcp_data_acked() could run just before we set the NOSPACE bit,
1213 * so explicitly check for snd_una value
1215 mptcp_clean_una((struct sock *)msk);
1218 #define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1219 sizeof(struct tcphdr) - \
1220 MAX_TCP_OPTION_SPACE - \
1221 sizeof(struct ipv6hdr) - \
1222 sizeof(struct frag_hdr))
1224 struct subflow_send_info {
1229 static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk,
1232 struct subflow_send_info send_info[2];
1233 struct mptcp_subflow_context *subflow;
1234 int i, nr_active = 0;
1239 sock_owned_by_me((struct sock *)msk);
1242 if (!mptcp_ext_cache_refill(msk))
1245 if (__mptcp_check_fallback(msk)) {
1248 *sndbuf = msk->first->sk_sndbuf;
1249 return sk_stream_memory_free(msk->first) ? msk->first : NULL;
1252 /* re-use last subflow, if the burst allow that */
1253 if (msk->last_snd && msk->snd_burst > 0 &&
1254 sk_stream_memory_free(msk->last_snd) &&
1255 mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1256 mptcp_for_each_subflow(msk, subflow) {
1257 ssk = mptcp_subflow_tcp_sock(subflow);
1258 *sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
1260 return msk->last_snd;
1263 /* pick the subflow with the lower wmem/wspace ratio */
1264 for (i = 0; i < 2; ++i) {
1265 send_info[i].ssk = NULL;
1266 send_info[i].ratio = -1;
1268 mptcp_for_each_subflow(msk, subflow) {
1269 ssk = mptcp_subflow_tcp_sock(subflow);
1270 if (!mptcp_subflow_active(subflow))
1273 nr_active += !subflow->backup;
1274 *sndbuf = max(tcp_sk(ssk)->snd_wnd, *sndbuf);
1275 if (!sk_stream_memory_free(subflow->tcp_sock))
1278 pace = READ_ONCE(ssk->sk_pacing_rate);
1282 ratio = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32,
1284 if (ratio < send_info[subflow->backup].ratio) {
1285 send_info[subflow->backup].ssk = ssk;
1286 send_info[subflow->backup].ratio = ratio;
1290 pr_debug("msk=%p nr_active=%d ssk=%p:%lld backup=%p:%lld",
1291 msk, nr_active, send_info[0].ssk, send_info[0].ratio,
1292 send_info[1].ssk, send_info[1].ratio);
1294 /* pick the best backup if no other subflow is active */
1296 send_info[0].ssk = send_info[1].ssk;
1298 if (send_info[0].ssk) {
1299 msk->last_snd = send_info[0].ssk;
1300 msk->snd_burst = min_t(int, MPTCP_SEND_BURST_SIZE,
1301 sk_stream_wspace(msk->last_snd));
1302 return msk->last_snd;
1307 static void mptcp_push_release(struct sock *sk, struct sock *ssk,
1308 struct mptcp_sendmsg_info *info)
1310 mptcp_set_timeout(sk, ssk);
1311 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1315 static void mptcp_push_pending(struct sock *sk, unsigned int flags)
1317 struct sock *prev_ssk = NULL, *ssk = NULL;
1318 struct mptcp_sock *msk = mptcp_sk(sk);
1319 struct mptcp_sendmsg_info info = {
1322 struct mptcp_data_frag *dfrag;
1323 int len, copied = 0;
1326 while ((dfrag = mptcp_send_head(sk))) {
1327 info.sent = dfrag->already_sent;
1328 info.limit = dfrag->data_len;
1329 len = dfrag->data_len - dfrag->already_sent;
1334 __mptcp_flush_join_list(msk);
1335 ssk = mptcp_subflow_get_send(msk, &sndbuf);
1337 /* do auto tuning */
1338 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK) &&
1339 sndbuf > READ_ONCE(sk->sk_sndbuf))
1340 WRITE_ONCE(sk->sk_sndbuf, sndbuf);
1342 /* try to keep the subflow socket lock across
1343 * consecutive xmit on the same socket
1345 if (ssk != prev_ssk && prev_ssk)
1346 mptcp_push_release(sk, prev_ssk, &info);
1350 if (ssk != prev_ssk || !prev_ssk)
1353 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1355 mptcp_push_release(sk, ssk, &info);
1360 dfrag->already_sent += ret;
1361 msk->snd_nxt += ret;
1362 msk->snd_burst -= ret;
1366 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1369 /* at this point we held the socket lock for the last subflow we used */
1371 mptcp_push_release(sk, ssk, &info);
1375 /* start the timer, if it's not pending */
1376 if (!mptcp_timer_pending(sk))
1377 mptcp_reset_timer(sk);
1378 __mptcp_check_send_data_fin(sk);
1382 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1384 struct mptcp_sock *msk = mptcp_sk(sk);
1385 struct page_frag *pfrag;
1390 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
1393 mptcp_lock_sock(sk, __mptcp_wmem_reserve(sk, len));
1395 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1397 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1398 ret = sk_stream_wait_connect(sk, &timeo);
1403 pfrag = sk_page_frag(sk);
1404 mptcp_clean_una(sk);
1406 while (msg_data_left(msg)) {
1407 struct mptcp_data_frag *dfrag;
1408 int frag_truesize = 0;
1409 bool dfrag_collapsed;
1410 size_t psize, offset;
1412 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
1417 /* reuse tail pfrag, if possible, or carve a new one from the
1420 dfrag = mptcp_pending_tail(sk);
1421 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1422 if (!dfrag_collapsed) {
1423 if (!sk_stream_memory_free(sk)) {
1424 mptcp_push_pending(sk, msg->msg_flags);
1425 if (!sk_stream_memory_free(sk))
1426 goto wait_for_memory;
1428 if (!mptcp_page_frag_refill(sk, pfrag))
1429 goto wait_for_memory;
1431 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1432 frag_truesize = dfrag->overhead;
1435 /* we do not bound vs wspace, to allow a single packet.
1436 * memory accounting will prevent execessive memory usage
1439 offset = dfrag->offset + dfrag->data_len;
1440 psize = pfrag->size - offset;
1441 psize = min_t(size_t, psize, msg_data_left(msg));
1442 if (!mptcp_wmem_alloc(sk, psize + frag_truesize))
1443 goto wait_for_memory;
1445 if (copy_page_from_iter(dfrag->page, offset, psize,
1446 &msg->msg_iter) != psize) {
1447 mptcp_wmem_uncharge(sk, psize + frag_truesize);
1452 /* data successfully copied into the write queue */
1454 dfrag->data_len += psize;
1455 frag_truesize += psize;
1456 pfrag->offset += frag_truesize;
1457 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1459 /* charge data on mptcp pending queue to the msk socket
1460 * Note: we charge such data both to sk and ssk
1462 sk_wmem_queued_add(sk, frag_truesize);
1463 if (!dfrag_collapsed) {
1464 get_page(dfrag->page);
1465 list_add_tail(&dfrag->list, &msk->rtx_queue);
1466 if (!msk->first_pending)
1467 WRITE_ONCE(msk->first_pending, dfrag);
1469 pr_debug("msk=%p dfrag at seq=%lld len=%d sent=%d new=%d", msk,
1470 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1473 if (!mptcp_ext_cache_refill(msk))
1474 goto wait_for_memory;
1479 if (mptcp_timer_pending(sk))
1480 mptcp_reset_timer(sk);
1481 ret = sk_stream_wait_memory(sk, &timeo);
1487 mptcp_push_pending(sk, msg->msg_flags);
1491 return copied ? : ret;
1494 static void mptcp_wait_data(struct sock *sk, long *timeo)
1496 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1497 struct mptcp_sock *msk = mptcp_sk(sk);
1499 add_wait_queue(sk_sleep(sk), &wait);
1500 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1502 sk_wait_event(sk, timeo,
1503 test_and_clear_bit(MPTCP_DATA_READY, &msk->flags), &wait);
1505 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1506 remove_wait_queue(sk_sleep(sk), &wait);
1509 static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1513 struct sk_buff *skb;
1516 while ((skb = skb_peek(&msk->receive_queue)) != NULL) {
1517 u32 offset = MPTCP_SKB_CB(skb)->offset;
1518 u32 data_len = skb->len - offset;
1519 u32 count = min_t(size_t, len - copied, data_len);
1522 err = skb_copy_datagram_msg(skb, offset, msg, count);
1523 if (unlikely(err < 0)) {
1531 if (count < data_len) {
1532 MPTCP_SKB_CB(skb)->offset += count;
1536 /* we will bulk release the skb memory later */
1537 skb->destructor = NULL;
1538 msk->rmem_released += skb->truesize;
1539 __skb_unlink(skb, &msk->receive_queue);
1549 /* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1551 * Only difference: Use highest rtt estimate of the subflows in use.
1553 static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1555 struct mptcp_subflow_context *subflow;
1556 struct sock *sk = (struct sock *)msk;
1557 u32 time, advmss = 1;
1560 sock_owned_by_me(sk);
1565 msk->rcvq_space.copied += copied;
1567 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1568 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1570 rtt_us = msk->rcvq_space.rtt_us;
1571 if (rtt_us && time < (rtt_us >> 3))
1575 mptcp_for_each_subflow(msk, subflow) {
1576 const struct tcp_sock *tp;
1580 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1582 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1583 sf_advmss = READ_ONCE(tp->advmss);
1585 rtt_us = max(sf_rtt_us, rtt_us);
1586 advmss = max(sf_advmss, advmss);
1589 msk->rcvq_space.rtt_us = rtt_us;
1590 if (time < (rtt_us >> 3) || rtt_us == 0)
1593 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1596 if (sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf &&
1597 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1601 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1603 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1605 do_div(grow, msk->rcvq_space.space);
1606 rcvwin += (grow << 1);
1608 rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1609 while (tcp_win_from_space(sk, rcvmem) < advmss)
1612 do_div(rcvwin, advmss);
1613 rcvbuf = min_t(u64, rcvwin * rcvmem,
1614 sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
1616 if (rcvbuf > sk->sk_rcvbuf) {
1619 window_clamp = tcp_win_from_space(sk, rcvbuf);
1620 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1622 /* Make subflows follow along. If we do not do this, we
1623 * get drops at subflow level if skbs can't be moved to
1624 * the mptcp rx queue fast enough (announced rcv_win can
1625 * exceed ssk->sk_rcvbuf).
1627 mptcp_for_each_subflow(msk, subflow) {
1631 ssk = mptcp_subflow_tcp_sock(subflow);
1632 slow = lock_sock_fast(ssk);
1633 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1634 tcp_sk(ssk)->window_clamp = window_clamp;
1635 tcp_cleanup_rbuf(ssk, 1);
1636 unlock_sock_fast(ssk, slow);
1641 msk->rcvq_space.space = msk->rcvq_space.copied;
1643 msk->rcvq_space.copied = 0;
1644 msk->rcvq_space.time = mstamp;
1647 static void __mptcp_update_rmem(struct sock *sk)
1649 struct mptcp_sock *msk = mptcp_sk(sk);
1651 if (!msk->rmem_released)
1654 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
1655 sk_mem_uncharge(sk, msk->rmem_released);
1656 msk->rmem_released = 0;
1659 static void __mptcp_splice_receive_queue(struct sock *sk)
1661 struct mptcp_sock *msk = mptcp_sk(sk);
1663 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1666 static bool __mptcp_move_skbs(struct mptcp_sock *msk, unsigned int rcv)
1668 struct sock *sk = (struct sock *)msk;
1669 unsigned int moved = 0;
1672 __mptcp_flush_join_list(msk);
1674 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
1677 /* we can have data pending in the subflows only if the msk
1678 * receive buffer was full at subflow_data_ready() time,
1679 * that is an unlikely slow path.
1684 slowpath = lock_sock_fast(ssk);
1685 mptcp_data_lock(sk);
1686 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
1687 mptcp_data_unlock(sk);
1689 WRITE_ONCE(msk->rmem_pending, min(rcv, moved));
1690 tcp_cleanup_rbuf(ssk, 1);
1691 WRITE_ONCE(msk->rmem_pending, 0);
1693 unlock_sock_fast(ssk, slowpath);
1696 /* acquire the data lock only if some input data is pending */
1698 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
1699 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
1700 mptcp_data_lock(sk);
1701 __mptcp_update_rmem(sk);
1702 ret |= __mptcp_ofo_queue(msk);
1703 __mptcp_splice_receive_queue(sk);
1704 mptcp_data_unlock(sk);
1707 mptcp_check_data_fin((struct sock *)msk);
1708 return !skb_queue_empty(&msk->receive_queue);
1711 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
1712 int nonblock, int flags, int *addr_len)
1714 struct mptcp_sock *msk = mptcp_sk(sk);
1719 if (msg->msg_flags & ~(MSG_WAITALL | MSG_DONTWAIT))
1722 mptcp_lock_sock(sk, __mptcp_splice_receive_queue(sk));
1723 if (unlikely(sk->sk_state == TCP_LISTEN)) {
1728 timeo = sock_rcvtimeo(sk, nonblock);
1730 len = min_t(size_t, len, INT_MAX);
1731 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1734 int bytes_read, old_space;
1736 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied);
1737 if (unlikely(bytes_read < 0)) {
1739 copied = bytes_read;
1743 copied += bytes_read;
1745 if (skb_queue_empty(&msk->receive_queue) &&
1746 __mptcp_move_skbs(msk, len - copied))
1749 /* be sure to advertise window change */
1750 old_space = READ_ONCE(msk->old_wspace);
1751 if ((tcp_space(sk) - old_space) >= old_space)
1752 mptcp_cleanup_rbuf(msk);
1754 /* only the master socket status is relevant here. The exit
1755 * conditions mirror closely tcp_recvmsg()
1757 if (copied >= target)
1762 sk->sk_state == TCP_CLOSE ||
1763 (sk->sk_shutdown & RCV_SHUTDOWN) ||
1765 signal_pending(current))
1769 copied = sock_error(sk);
1773 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
1774 mptcp_check_for_eof(msk);
1776 if (sk->sk_shutdown & RCV_SHUTDOWN) {
1777 /* race breaker: the shutdown could be after the
1778 * previous receive queue check
1780 if (__mptcp_move_skbs(msk, len - copied))
1785 if (sk->sk_state == TCP_CLOSE) {
1795 if (signal_pending(current)) {
1796 copied = sock_intr_errno(timeo);
1801 pr_debug("block timeout %ld", timeo);
1802 mptcp_wait_data(sk, &timeo);
1805 if (skb_queue_empty_lockless(&sk->sk_receive_queue) &&
1806 skb_queue_empty(&msk->receive_queue)) {
1807 /* entire backlog drained, clear DATA_READY. */
1808 clear_bit(MPTCP_DATA_READY, &msk->flags);
1810 /* .. race-breaker: ssk might have gotten new data
1811 * after last __mptcp_move_skbs() returned false.
1813 if (unlikely(__mptcp_move_skbs(msk, 0)))
1814 set_bit(MPTCP_DATA_READY, &msk->flags);
1815 } else if (unlikely(!test_bit(MPTCP_DATA_READY, &msk->flags))) {
1816 /* data to read but mptcp_wait_data() cleared DATA_READY */
1817 set_bit(MPTCP_DATA_READY, &msk->flags);
1820 pr_debug("msk=%p data_ready=%d rx queue empty=%d copied=%d",
1821 msk, test_bit(MPTCP_DATA_READY, &msk->flags),
1822 skb_queue_empty_lockless(&sk->sk_receive_queue), copied);
1823 mptcp_rcv_space_adjust(msk, copied);
1829 static void mptcp_retransmit_handler(struct sock *sk)
1831 struct mptcp_sock *msk = mptcp_sk(sk);
1833 if (atomic64_read(&msk->snd_una) == READ_ONCE(msk->snd_nxt)) {
1834 mptcp_stop_timer(sk);
1836 set_bit(MPTCP_WORK_RTX, &msk->flags);
1837 mptcp_schedule_work(sk);
1841 static void mptcp_retransmit_timer(struct timer_list *t)
1843 struct inet_connection_sock *icsk = from_timer(icsk, t,
1844 icsk_retransmit_timer);
1845 struct sock *sk = &icsk->icsk_inet.sk;
1848 if (!sock_owned_by_user(sk)) {
1849 mptcp_retransmit_handler(sk);
1851 /* delegate our work to tcp_release_cb() */
1852 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED,
1860 static void mptcp_timeout_timer(struct timer_list *t)
1862 struct sock *sk = from_timer(sk, t, sk_timer);
1864 mptcp_schedule_work(sk);
1868 /* Find an idle subflow. Return NULL if there is unacked data at tcp
1871 * A backup subflow is returned only if that is the only kind available.
1873 static struct sock *mptcp_subflow_get_retrans(const struct mptcp_sock *msk)
1875 struct mptcp_subflow_context *subflow;
1876 struct sock *backup = NULL;
1878 sock_owned_by_me((const struct sock *)msk);
1880 if (__mptcp_check_fallback(msk))
1883 mptcp_for_each_subflow(msk, subflow) {
1884 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1886 if (!mptcp_subflow_active(subflow))
1889 /* still data outstanding at TCP level? Don't retransmit. */
1890 if (!tcp_write_queue_empty(ssk)) {
1891 if (inet_csk(ssk)->icsk_ca_state >= TCP_CA_Loss)
1896 if (subflow->backup) {
1908 /* subflow sockets can be either outgoing (connect) or incoming
1911 * Outgoing subflows use in-kernel sockets.
1912 * Incoming subflows do not have their own 'struct socket' allocated,
1913 * so we need to use tcp_close() after detaching them from the mptcp
1916 void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
1917 struct mptcp_subflow_context *subflow)
1919 bool dispose_socket = false;
1920 struct socket *sock;
1922 list_del(&subflow->node);
1926 /* if we are invoked by the msk cleanup code, the subflow is
1929 sock = ssk->sk_socket;
1931 dispose_socket = sock != sk->sk_socket;
1935 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
1936 * the ssk has been already destroyed, we just need to release the
1937 * reference owned by msk;
1939 if (!inet_csk(ssk)->icsk_ulp_ops) {
1940 kfree_rcu(subflow, rcu);
1942 /* otherwise ask tcp do dispose of ssk and subflow ctx */
1943 subflow->disposable = 1;
1944 __tcp_close(ssk, 0);
1946 /* close acquired an extra ref */
1951 iput(SOCK_INODE(sock));
1956 static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
1961 static void pm_work(struct mptcp_sock *msk)
1963 struct mptcp_pm_data *pm = &msk->pm;
1965 spin_lock_bh(&msk->pm.lock);
1967 pr_debug("msk=%p status=%x", msk, pm->status);
1968 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
1969 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
1970 mptcp_pm_nl_add_addr_received(msk);
1972 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
1973 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
1974 mptcp_pm_nl_add_addr_send_ack(msk);
1976 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
1977 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
1978 mptcp_pm_nl_rm_addr_received(msk);
1980 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
1981 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
1982 mptcp_pm_nl_fully_established(msk);
1984 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
1985 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
1986 mptcp_pm_nl_subflow_established(msk);
1989 spin_unlock_bh(&msk->pm.lock);
1992 static void __mptcp_close_subflow(struct mptcp_sock *msk)
1994 struct mptcp_subflow_context *subflow, *tmp;
1996 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
1997 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1999 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2002 __mptcp_close_ssk((struct sock *)msk, ssk, subflow);
2006 static bool mptcp_check_close_timeout(const struct sock *sk)
2008 s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2009 struct mptcp_subflow_context *subflow;
2011 if (delta >= TCP_TIMEWAIT_LEN)
2014 /* if all subflows are in closed status don't bother with additional
2017 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2018 if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2025 static void mptcp_worker(struct work_struct *work)
2027 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2028 struct sock *ssk, *sk = &msk->sk.icsk_inet.sk;
2029 struct mptcp_sendmsg_info info = {};
2030 struct mptcp_data_frag *dfrag;
2035 state = sk->sk_state;
2036 if (unlikely(state == TCP_CLOSE))
2039 mptcp_clean_una_wakeup(sk);
2040 mptcp_check_data_fin_ack(sk);
2041 __mptcp_flush_join_list(msk);
2042 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2043 __mptcp_close_subflow(msk);
2045 if (mptcp_send_head(sk))
2046 mptcp_push_pending(sk, 0);
2051 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2052 mptcp_check_for_eof(msk);
2054 mptcp_check_data_fin(sk);
2056 /* if the msk data is completely acked, or the socket timedout,
2057 * there is no point in keeping around an orphaned sk
2059 if (sock_flag(sk, SOCK_DEAD) &&
2060 (mptcp_check_close_timeout(sk) ||
2061 (state != sk->sk_state &&
2062 ((1 << inet_sk_state_load(sk)) & (TCPF_CLOSE | TCPF_FIN_WAIT2))))) {
2063 inet_sk_state_store(sk, TCP_CLOSE);
2064 __mptcp_destroy_sock(sk);
2068 if (!test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2071 dfrag = mptcp_rtx_head(sk);
2075 if (!mptcp_ext_cache_refill(msk))
2078 ssk = mptcp_subflow_get_retrans(msk);
2084 /* limit retransmission to the bytes already sent on some subflows */
2086 info.limit = dfrag->already_sent;
2087 while (info.sent < dfrag->already_sent) {
2088 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2092 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2096 if (!mptcp_ext_cache_refill(msk))
2100 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2103 mptcp_set_timeout(sk, ssk);
2107 if (!mptcp_timer_pending(sk))
2108 mptcp_reset_timer(sk);
2115 static int __mptcp_init_sock(struct sock *sk)
2117 struct mptcp_sock *msk = mptcp_sk(sk);
2119 spin_lock_init(&msk->join_list_lock);
2121 INIT_LIST_HEAD(&msk->conn_list);
2122 INIT_LIST_HEAD(&msk->join_list);
2123 INIT_LIST_HEAD(&msk->rtx_queue);
2124 INIT_WORK(&msk->work, mptcp_worker);
2125 __skb_queue_head_init(&msk->receive_queue);
2126 msk->out_of_order_queue = RB_ROOT;
2127 msk->first_pending = NULL;
2128 msk->wmem_reserved = 0;
2129 msk->rmem_released = 0;
2131 msk->ack_hint = NULL;
2133 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
2135 mptcp_pm_data_init(msk);
2137 /* re-use the csk retrans timer for MPTCP-level retrans */
2138 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
2139 timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
2143 static int mptcp_init_sock(struct sock *sk)
2145 struct net *net = sock_net(sk);
2148 ret = __mptcp_init_sock(sk);
2152 if (!mptcp_is_enabled(net))
2153 return -ENOPROTOOPT;
2155 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2158 ret = __mptcp_socket_create(mptcp_sk(sk));
2162 sk_sockets_allocated_inc(sk);
2163 sk->sk_rcvbuf = sock_net(sk)->ipv4.sysctl_tcp_rmem[1];
2164 sk->sk_sndbuf = sock_net(sk)->ipv4.sysctl_tcp_wmem[1];
2169 static void __mptcp_clear_xmit(struct sock *sk)
2171 struct mptcp_sock *msk = mptcp_sk(sk);
2172 struct mptcp_data_frag *dtmp, *dfrag;
2174 sk_stop_timer(sk, &msk->sk.icsk_retransmit_timer);
2176 WRITE_ONCE(msk->first_pending, NULL);
2177 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
2178 dfrag_clear(sk, dfrag);
2181 static void mptcp_cancel_work(struct sock *sk)
2183 struct mptcp_sock *msk = mptcp_sk(sk);
2185 if (cancel_work_sync(&msk->work))
2189 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
2193 switch (ssk->sk_state) {
2195 if (!(how & RCV_SHUTDOWN))
2199 tcp_disconnect(ssk, O_NONBLOCK);
2202 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2203 pr_debug("Fallback");
2204 ssk->sk_shutdown |= how;
2205 tcp_shutdown(ssk, how);
2207 pr_debug("Sending DATA_FIN on subflow %p", ssk);
2208 mptcp_set_timeout(sk, ssk);
2217 static const unsigned char new_state[16] = {
2218 /* current state: new state: action: */
2219 [0 /* (Invalid) */] = TCP_CLOSE,
2220 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2221 [TCP_SYN_SENT] = TCP_CLOSE,
2222 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2223 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2224 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2225 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2226 [TCP_CLOSE] = TCP_CLOSE,
2227 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2228 [TCP_LAST_ACK] = TCP_LAST_ACK,
2229 [TCP_LISTEN] = TCP_CLOSE,
2230 [TCP_CLOSING] = TCP_CLOSING,
2231 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2234 static int mptcp_close_state(struct sock *sk)
2236 int next = (int)new_state[sk->sk_state];
2237 int ns = next & TCP_STATE_MASK;
2239 inet_sk_state_store(sk, ns);
2241 return next & TCP_ACTION_FIN;
2244 static void __mptcp_check_send_data_fin(struct sock *sk)
2246 struct mptcp_subflow_context *subflow;
2247 struct mptcp_sock *msk = mptcp_sk(sk);
2249 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2250 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2251 msk->snd_nxt, msk->write_seq);
2253 /* we still need to enqueue subflows or not really shutting down,
2256 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2257 mptcp_send_head(sk))
2260 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2262 /* fallback socket will not get data_fin/ack, can move to the next
2265 if (__mptcp_check_fallback(msk)) {
2266 if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2267 inet_sk_state_store(sk, TCP_CLOSE);
2268 mptcp_close_wake_up(sk);
2269 } else if (sk->sk_state == TCP_FIN_WAIT1) {
2270 inet_sk_state_store(sk, TCP_FIN_WAIT2);
2274 __mptcp_flush_join_list(msk);
2275 mptcp_for_each_subflow(msk, subflow) {
2276 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2278 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
2282 static void __mptcp_wr_shutdown(struct sock *sk)
2284 struct mptcp_sock *msk = mptcp_sk(sk);
2286 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2287 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2288 !!mptcp_send_head(sk));
2290 /* will be ignored by fallback sockets */
2291 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2292 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2294 __mptcp_check_send_data_fin(sk);
2297 static void __mptcp_destroy_sock(struct sock *sk)
2299 struct mptcp_subflow_context *subflow, *tmp;
2300 struct mptcp_sock *msk = mptcp_sk(sk);
2301 LIST_HEAD(conn_list);
2303 pr_debug("msk=%p", msk);
2305 /* be sure to always acquire the join list lock, to sync vs
2306 * mptcp_finish_join().
2308 spin_lock_bh(&msk->join_list_lock);
2309 list_splice_tail_init(&msk->join_list, &msk->conn_list);
2310 spin_unlock_bh(&msk->join_list_lock);
2311 list_splice_init(&msk->conn_list, &conn_list);
2313 __mptcp_clear_xmit(sk);
2314 sk_stop_timer(sk, &sk->sk_timer);
2317 list_for_each_entry_safe(subflow, tmp, &conn_list, node) {
2318 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2319 __mptcp_close_ssk(sk, ssk, subflow);
2322 sk->sk_prot->destroy(sk);
2324 WARN_ON_ONCE(msk->wmem_reserved);
2325 WARN_ON_ONCE(msk->rmem_released);
2326 sk_stream_kill_queues(sk);
2327 xfrm_sk_free_policy(sk);
2328 sk_refcnt_debug_release(sk);
2332 static void mptcp_close(struct sock *sk, long timeout)
2334 struct mptcp_subflow_context *subflow;
2335 bool do_cancel_work = false;
2338 sk->sk_shutdown = SHUTDOWN_MASK;
2340 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2341 inet_sk_state_store(sk, TCP_CLOSE);
2345 if (mptcp_close_state(sk))
2346 __mptcp_wr_shutdown(sk);
2348 sk_stream_wait_close(sk, timeout);
2351 /* orphan all the subflows */
2352 inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
2353 list_for_each_entry(subflow, &mptcp_sk(sk)->conn_list, node) {
2354 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2355 bool slow, dispose_socket;
2356 struct socket *sock;
2358 slow = lock_sock_fast(ssk);
2359 sock = ssk->sk_socket;
2360 dispose_socket = sock && sock != sk->sk_socket;
2362 unlock_sock_fast(ssk, slow);
2364 /* for the outgoing subflows we additionally need to free
2365 * the associated socket
2368 iput(SOCK_INODE(sock));
2373 pr_debug("msk=%p state=%d", sk, sk->sk_state);
2374 if (sk->sk_state == TCP_CLOSE) {
2375 __mptcp_destroy_sock(sk);
2376 do_cancel_work = true;
2378 sk_reset_timer(sk, &sk->sk_timer, jiffies + TCP_TIMEWAIT_LEN);
2382 mptcp_cancel_work(sk);
2386 static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
2388 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2389 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
2390 struct ipv6_pinfo *msk6 = inet6_sk(msk);
2392 msk->sk_v6_daddr = ssk->sk_v6_daddr;
2393 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
2396 msk6->saddr = ssk6->saddr;
2397 msk6->flow_label = ssk6->flow_label;
2401 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
2402 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
2403 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
2404 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
2405 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
2406 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
2409 static int mptcp_disconnect(struct sock *sk, int flags)
2411 /* Should never be called.
2412 * inet_stream_connect() calls ->disconnect, but that
2413 * refers to the subflow socket, not the mptcp one.
2419 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2420 static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
2422 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
2424 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
2428 struct sock *mptcp_sk_clone(const struct sock *sk,
2429 const struct mptcp_options_received *mp_opt,
2430 struct request_sock *req)
2432 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2433 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
2434 struct mptcp_sock *msk;
2440 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2441 if (nsk->sk_family == AF_INET6)
2442 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
2445 __mptcp_init_sock(nsk);
2447 msk = mptcp_sk(nsk);
2448 msk->local_key = subflow_req->local_key;
2449 msk->token = subflow_req->token;
2450 msk->subflow = NULL;
2451 WRITE_ONCE(msk->fully_established, false);
2453 msk->write_seq = subflow_req->idsn + 1;
2454 msk->snd_nxt = msk->write_seq;
2455 atomic64_set(&msk->snd_una, msk->write_seq);
2456 atomic64_set(&msk->wnd_end, msk->snd_nxt + req->rsk_rcv_wnd);
2458 if (mp_opt->mp_capable) {
2459 msk->can_ack = true;
2460 msk->remote_key = mp_opt->sndr_key;
2461 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
2463 WRITE_ONCE(msk->ack_seq, ack_seq);
2464 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
2467 sock_reset_flag(nsk, SOCK_RCU_FREE);
2468 /* will be fully established after successful MPC subflow creation */
2469 inet_sk_state_store(nsk, TCP_SYN_RECV);
2470 bh_unlock_sock(nsk);
2472 /* keep a single reference */
2477 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
2479 const struct tcp_sock *tp = tcp_sk(ssk);
2481 msk->rcvq_space.copied = 0;
2482 msk->rcvq_space.rtt_us = 0;
2484 msk->rcvq_space.time = tp->tcp_mstamp;
2486 /* initial rcv_space offering made to peer */
2487 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
2488 TCP_INIT_CWND * tp->advmss);
2489 if (msk->rcvq_space.space == 0)
2490 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
2492 atomic64_set(&msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
2495 static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
2498 struct mptcp_sock *msk = mptcp_sk(sk);
2499 struct socket *listener;
2502 listener = __mptcp_nmpc_socket(msk);
2503 if (WARN_ON_ONCE(!listener)) {
2508 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
2509 newsk = inet_csk_accept(listener->sk, flags, err, kern);
2513 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
2514 if (sk_is_mptcp(newsk)) {
2515 struct mptcp_subflow_context *subflow;
2516 struct sock *new_mptcp_sock;
2518 subflow = mptcp_subflow_ctx(newsk);
2519 new_mptcp_sock = subflow->conn;
2521 /* is_mptcp should be false if subflow->conn is missing, see
2522 * subflow_syn_recv_sock()
2524 if (WARN_ON_ONCE(!new_mptcp_sock)) {
2525 tcp_sk(newsk)->is_mptcp = 0;
2529 /* acquire the 2nd reference for the owning socket */
2530 sock_hold(new_mptcp_sock);
2531 newsk = new_mptcp_sock;
2532 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
2534 MPTCP_INC_STATS(sock_net(sk),
2535 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
2541 void mptcp_destroy_common(struct mptcp_sock *msk)
2543 struct sock *sk = (struct sock *)msk;
2545 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
2546 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
2548 skb_rbtree_purge(&msk->out_of_order_queue);
2549 mptcp_token_destroy(msk);
2550 mptcp_pm_free_anno_list(msk);
2553 static void mptcp_destroy(struct sock *sk)
2555 struct mptcp_sock *msk = mptcp_sk(sk);
2557 if (msk->cached_ext)
2558 __skb_ext_put(msk->cached_ext);
2560 mptcp_destroy_common(msk);
2561 sk_sockets_allocated_dec(sk);
2564 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname,
2565 sockptr_t optval, unsigned int optlen)
2567 struct sock *sk = (struct sock *)msk;
2568 struct socket *ssock;
2575 ssock = __mptcp_nmpc_socket(msk);
2581 ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen);
2583 if (optname == SO_REUSEPORT)
2584 sk->sk_reuseport = ssock->sk->sk_reuseport;
2585 else if (optname == SO_REUSEADDR)
2586 sk->sk_reuse = ssock->sk->sk_reuse;
2592 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen);
2595 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname,
2596 sockptr_t optval, unsigned int optlen)
2598 struct sock *sk = (struct sock *)msk;
2599 int ret = -EOPNOTSUPP;
2600 struct socket *ssock;
2605 ssock = __mptcp_nmpc_socket(msk);
2611 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen);
2613 sk->sk_ipv6only = ssock->sk->sk_ipv6only;
2622 static int mptcp_setsockopt(struct sock *sk, int level, int optname,
2623 sockptr_t optval, unsigned int optlen)
2625 struct mptcp_sock *msk = mptcp_sk(sk);
2628 pr_debug("msk=%p", msk);
2630 if (level == SOL_SOCKET)
2631 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen);
2633 /* @@ the meaning of setsockopt() when the socket is connected and
2634 * there are multiple subflows is not yet defined. It is up to the
2635 * MPTCP-level socket to configure the subflows until the subflow
2636 * is in TCP fallback, when TCP socket options are passed through
2637 * to the one remaining subflow.
2640 ssk = __mptcp_tcp_fallback(msk);
2643 return tcp_setsockopt(ssk, level, optname, optval, optlen);
2645 if (level == SOL_IPV6)
2646 return mptcp_setsockopt_v6(msk, optname, optval, optlen);
2651 static int mptcp_getsockopt(struct sock *sk, int level, int optname,
2652 char __user *optval, int __user *option)
2654 struct mptcp_sock *msk = mptcp_sk(sk);
2657 pr_debug("msk=%p", msk);
2659 /* @@ the meaning of setsockopt() when the socket is connected and
2660 * there are multiple subflows is not yet defined. It is up to the
2661 * MPTCP-level socket to configure the subflows until the subflow
2662 * is in TCP fallback, when socket options are passed through
2663 * to the one remaining subflow.
2666 ssk = __mptcp_tcp_fallback(msk);
2669 return tcp_getsockopt(ssk, level, optname, optval, option);
2674 #define MPTCP_DEFERRED_ALL (TCPF_WRITE_TIMER_DEFERRED)
2676 /* processes deferred events and flush wmem */
2677 static void mptcp_release_cb(struct sock *sk)
2679 unsigned long flags, nflags;
2681 /* clear any wmem reservation and errors */
2682 __mptcp_update_wmem(sk);
2683 __mptcp_update_rmem(sk);
2686 flags = sk->sk_tsq_flags;
2687 if (!(flags & MPTCP_DEFERRED_ALL))
2689 nflags = flags & ~MPTCP_DEFERRED_ALL;
2690 } while (cmpxchg(&sk->sk_tsq_flags, flags, nflags) != flags);
2692 sock_release_ownership(sk);
2694 if (flags & TCPF_WRITE_TIMER_DEFERRED) {
2695 mptcp_retransmit_handler(sk);
2700 static int mptcp_hash(struct sock *sk)
2702 /* should never be called,
2703 * we hash the TCP subflows not the master socket
2709 static void mptcp_unhash(struct sock *sk)
2711 /* called from sk_common_release(), but nothing to do here */
2714 static int mptcp_get_port(struct sock *sk, unsigned short snum)
2716 struct mptcp_sock *msk = mptcp_sk(sk);
2717 struct socket *ssock;
2719 ssock = __mptcp_nmpc_socket(msk);
2720 pr_debug("msk=%p, subflow=%p", msk, ssock);
2721 if (WARN_ON_ONCE(!ssock))
2724 return inet_csk_get_port(ssock->sk, snum);
2727 void mptcp_finish_connect(struct sock *ssk)
2729 struct mptcp_subflow_context *subflow;
2730 struct mptcp_sock *msk;
2734 subflow = mptcp_subflow_ctx(ssk);
2738 pr_debug("msk=%p, token=%u", sk, subflow->token);
2740 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
2742 subflow->map_seq = ack_seq;
2743 subflow->map_subflow_seq = 1;
2745 /* the socket is not connected yet, no msk/subflow ops can access/race
2746 * accessing the field below
2748 WRITE_ONCE(msk->remote_key, subflow->remote_key);
2749 WRITE_ONCE(msk->local_key, subflow->local_key);
2750 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
2751 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2752 WRITE_ONCE(msk->ack_seq, ack_seq);
2753 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
2754 WRITE_ONCE(msk->can_ack, 1);
2755 atomic64_set(&msk->snd_una, msk->write_seq);
2757 mptcp_pm_new_connection(msk, 0);
2759 mptcp_rcv_space_init(msk, ssk);
2762 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)
2764 write_lock_bh(&sk->sk_callback_lock);
2765 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2766 sk_set_socket(sk, parent);
2767 sk->sk_uid = SOCK_INODE(parent)->i_uid;
2768 write_unlock_bh(&sk->sk_callback_lock);
2771 bool mptcp_finish_join(struct sock *ssk)
2773 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2774 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2775 struct sock *parent = (void *)msk;
2776 struct socket *parent_sock;
2779 pr_debug("msk=%p, subflow=%p", msk, subflow);
2781 /* mptcp socket already closing? */
2782 if (!mptcp_is_fully_established(parent))
2785 if (!msk->pm.server_side)
2788 if (!mptcp_pm_allow_new_subflow(msk))
2791 /* active connections are already on conn_list, and we can't acquire
2793 * use the join list lock as synchronization point and double-check
2794 * msk status to avoid racing with __mptcp_destroy_sock()
2796 spin_lock_bh(&msk->join_list_lock);
2797 ret = inet_sk_state_load(parent) == TCP_ESTABLISHED;
2798 if (ret && !WARN_ON_ONCE(!list_empty(&subflow->node))) {
2799 list_add_tail(&subflow->node, &msk->join_list);
2802 spin_unlock_bh(&msk->join_list_lock);
2806 /* attach to msk socket only after we are sure he will deal with us
2809 parent_sock = READ_ONCE(parent->sk_socket);
2810 if (parent_sock && !ssk->sk_socket)
2811 mptcp_sock_graft(ssk, parent_sock);
2812 subflow->map_seq = READ_ONCE(msk->ack_seq);
2816 static struct proto mptcp_prot = {
2818 .owner = THIS_MODULE,
2819 .init = mptcp_init_sock,
2820 .disconnect = mptcp_disconnect,
2821 .close = mptcp_close,
2822 .accept = mptcp_accept,
2823 .setsockopt = mptcp_setsockopt,
2824 .getsockopt = mptcp_getsockopt,
2825 .shutdown = tcp_shutdown,
2826 .destroy = mptcp_destroy,
2827 .sendmsg = mptcp_sendmsg,
2828 .recvmsg = mptcp_recvmsg,
2829 .release_cb = mptcp_release_cb,
2831 .unhash = mptcp_unhash,
2832 .get_port = mptcp_get_port,
2833 .sockets_allocated = &mptcp_sockets_allocated,
2834 .memory_allocated = &tcp_memory_allocated,
2835 .memory_pressure = &tcp_memory_pressure,
2836 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2837 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2838 .sysctl_mem = sysctl_tcp_mem,
2839 .obj_size = sizeof(struct mptcp_sock),
2840 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2841 .no_autobind = true,
2844 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2846 struct mptcp_sock *msk = mptcp_sk(sock->sk);
2847 struct socket *ssock;
2850 lock_sock(sock->sk);
2851 ssock = __mptcp_nmpc_socket(msk);
2857 err = ssock->ops->bind(ssock, uaddr, addr_len);
2859 mptcp_copy_inaddrs(sock->sk, ssock->sk);
2862 release_sock(sock->sk);
2866 static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
2867 struct mptcp_subflow_context *subflow)
2869 subflow->request_mptcp = 0;
2870 __mptcp_do_fallback(msk);
2873 static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
2874 int addr_len, int flags)
2876 struct mptcp_sock *msk = mptcp_sk(sock->sk);
2877 struct mptcp_subflow_context *subflow;
2878 struct socket *ssock;
2881 lock_sock(sock->sk);
2882 if (sock->state != SS_UNCONNECTED && msk->subflow) {
2883 /* pending connection or invalid state, let existing subflow
2886 ssock = msk->subflow;
2890 ssock = __mptcp_nmpc_socket(msk);
2896 mptcp_token_destroy(msk);
2897 inet_sk_state_store(sock->sk, TCP_SYN_SENT);
2898 subflow = mptcp_subflow_ctx(ssock->sk);
2899 #ifdef CONFIG_TCP_MD5SIG
2900 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
2903 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
2904 mptcp_subflow_early_fallback(msk, subflow);
2906 if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk))
2907 mptcp_subflow_early_fallback(msk, subflow);
2910 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
2911 sock->state = ssock->state;
2913 /* on successful connect, the msk state will be moved to established by
2914 * subflow_finish_connect()
2916 if (!err || err == -EINPROGRESS)
2917 mptcp_copy_inaddrs(sock->sk, ssock->sk);
2919 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
2922 release_sock(sock->sk);
2926 static int mptcp_listen(struct socket *sock, int backlog)
2928 struct mptcp_sock *msk = mptcp_sk(sock->sk);
2929 struct socket *ssock;
2932 pr_debug("msk=%p", msk);
2934 lock_sock(sock->sk);
2935 ssock = __mptcp_nmpc_socket(msk);
2941 mptcp_token_destroy(msk);
2942 inet_sk_state_store(sock->sk, TCP_LISTEN);
2943 sock_set_flag(sock->sk, SOCK_RCU_FREE);
2945 err = ssock->ops->listen(ssock, backlog);
2946 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
2948 mptcp_copy_inaddrs(sock->sk, ssock->sk);
2951 release_sock(sock->sk);
2955 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
2956 int flags, bool kern)
2958 struct mptcp_sock *msk = mptcp_sk(sock->sk);
2959 struct socket *ssock;
2962 pr_debug("msk=%p", msk);
2964 lock_sock(sock->sk);
2965 if (sock->sk->sk_state != TCP_LISTEN)
2968 ssock = __mptcp_nmpc_socket(msk);
2972 clear_bit(MPTCP_DATA_READY, &msk->flags);
2973 sock_hold(ssock->sk);
2974 release_sock(sock->sk);
2976 err = ssock->ops->accept(sock, newsock, flags, kern);
2977 if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
2978 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
2979 struct mptcp_subflow_context *subflow;
2980 struct sock *newsk = newsock->sk;
2983 slowpath = lock_sock_fast(newsk);
2984 mptcp_copy_inaddrs(newsk, msk->first);
2985 mptcp_rcv_space_init(msk, msk->first);
2987 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
2988 * This is needed so NOSPACE flag can be set from tcp stack.
2990 __mptcp_flush_join_list(msk);
2991 mptcp_for_each_subflow(msk, subflow) {
2992 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2994 if (!ssk->sk_socket)
2995 mptcp_sock_graft(ssk, newsock);
2997 unlock_sock_fast(newsk, slowpath);
3000 if (inet_csk_listen_poll(ssock->sk))
3001 set_bit(MPTCP_DATA_READY, &msk->flags);
3002 sock_put(ssock->sk);
3006 release_sock(sock->sk);
3010 static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
3012 return test_bit(MPTCP_DATA_READY, &msk->flags) ? EPOLLIN | EPOLLRDNORM :
3016 static bool __mptcp_check_writeable(struct mptcp_sock *msk)
3018 struct sock *sk = (struct sock *)msk;
3019 bool mptcp_writable;
3021 mptcp_clean_una(sk);
3022 mptcp_writable = sk_stream_is_writeable(sk);
3023 if (!mptcp_writable)
3026 return mptcp_writable;
3029 static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3031 struct sock *sk = (struct sock *)msk;
3035 if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
3038 if (sk_stream_is_writeable(sk))
3039 return EPOLLOUT | EPOLLWRNORM;
3041 slow = lock_sock_fast(sk);
3042 if (__mptcp_check_writeable(msk))
3043 ret = EPOLLOUT | EPOLLWRNORM;
3045 unlock_sock_fast(sk, slow);
3049 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3050 struct poll_table_struct *wait)
3052 struct sock *sk = sock->sk;
3053 struct mptcp_sock *msk;
3058 sock_poll_wait(file, sock, wait);
3060 state = inet_sk_state_load(sk);
3061 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
3062 if (state == TCP_LISTEN)
3063 return mptcp_check_readable(msk);
3065 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3066 mask |= mptcp_check_readable(msk);
3067 mask |= mptcp_check_writeable(msk);
3069 if (sk->sk_shutdown & RCV_SHUTDOWN)
3070 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3075 static int mptcp_shutdown(struct socket *sock, int how)
3077 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3078 struct sock *sk = sock->sk;
3081 pr_debug("sk=%p, how=%d", msk, how);
3086 if ((how & ~SHUTDOWN_MASK) || !how) {
3091 if (sock->state == SS_CONNECTING) {
3092 if ((1 << sk->sk_state) &
3093 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
3094 sock->state = SS_DISCONNECTING;
3096 sock->state = SS_CONNECTED;
3099 sk->sk_shutdown |= how;
3100 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3101 __mptcp_wr_shutdown(sk);
3103 /* Wake up anyone sleeping in poll. */
3104 sk->sk_state_change(sk);
3112 static const struct proto_ops mptcp_stream_ops = {
3114 .owner = THIS_MODULE,
3115 .release = inet_release,
3117 .connect = mptcp_stream_connect,
3118 .socketpair = sock_no_socketpair,
3119 .accept = mptcp_stream_accept,
3120 .getname = inet_getname,
3122 .ioctl = inet_ioctl,
3123 .gettstamp = sock_gettstamp,
3124 .listen = mptcp_listen,
3125 .shutdown = mptcp_shutdown,
3126 .setsockopt = sock_common_setsockopt,
3127 .getsockopt = sock_common_getsockopt,
3128 .sendmsg = inet_sendmsg,
3129 .recvmsg = inet_recvmsg,
3130 .mmap = sock_no_mmap,
3131 .sendpage = inet_sendpage,
3134 static struct inet_protosw mptcp_protosw = {
3135 .type = SOCK_STREAM,
3136 .protocol = IPPROTO_MPTCP,
3137 .prot = &mptcp_prot,
3138 .ops = &mptcp_stream_ops,
3139 .flags = INET_PROTOSW_ICSK,
3142 void __init mptcp_proto_init(void)
3144 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
3146 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3147 panic("Failed to allocate MPTCP pcpu counter\n");
3149 mptcp_subflow_init();
3153 if (proto_register(&mptcp_prot, 1) != 0)
3154 panic("Failed to register MPTCP proto.\n");
3156 inet_register_protosw(&mptcp_protosw);
3158 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
3161 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3162 static const struct proto_ops mptcp_v6_stream_ops = {
3164 .owner = THIS_MODULE,
3165 .release = inet6_release,
3167 .connect = mptcp_stream_connect,
3168 .socketpair = sock_no_socketpair,
3169 .accept = mptcp_stream_accept,
3170 .getname = inet6_getname,
3172 .ioctl = inet6_ioctl,
3173 .gettstamp = sock_gettstamp,
3174 .listen = mptcp_listen,
3175 .shutdown = mptcp_shutdown,
3176 .setsockopt = sock_common_setsockopt,
3177 .getsockopt = sock_common_getsockopt,
3178 .sendmsg = inet6_sendmsg,
3179 .recvmsg = inet6_recvmsg,
3180 .mmap = sock_no_mmap,
3181 .sendpage = inet_sendpage,
3182 #ifdef CONFIG_COMPAT
3183 .compat_ioctl = inet6_compat_ioctl,
3187 static struct proto mptcp_v6_prot;
3189 static void mptcp_v6_destroy(struct sock *sk)
3192 inet6_destroy_sock(sk);
3195 static struct inet_protosw mptcp_v6_protosw = {
3196 .type = SOCK_STREAM,
3197 .protocol = IPPROTO_MPTCP,
3198 .prot = &mptcp_v6_prot,
3199 .ops = &mptcp_v6_stream_ops,
3200 .flags = INET_PROTOSW_ICSK,
3203 int __init mptcp_proto_v6_init(void)
3207 mptcp_v6_prot = mptcp_prot;
3208 strcpy(mptcp_v6_prot.name, "MPTCPv6");
3209 mptcp_v6_prot.slab = NULL;
3210 mptcp_v6_prot.destroy = mptcp_v6_destroy;
3211 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
3213 err = proto_register(&mptcp_v6_prot, 1);
3217 err = inet6_register_protosw(&mptcp_v6_protosw);
3219 proto_unregister(&mptcp_v6_prot);