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>
25 #include <asm/ioctls.h>
29 #define CREATE_TRACE_POINTS
30 #include <trace/events/mptcp.h>
32 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
34 struct mptcp_sock msk;
46 #define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
49 MPTCP_CMSG_TS = BIT(0),
50 MPTCP_CMSG_INQ = BIT(1),
53 static struct percpu_counter mptcp_sockets_allocated ____cacheline_aligned_in_smp;
55 static void __mptcp_destroy_sock(struct sock *sk);
56 static void __mptcp_check_send_data_fin(struct sock *sk);
58 DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions);
59 static struct net_device mptcp_napi_dev;
61 /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
62 * completed yet or has failed, return the subflow socket.
63 * Otherwise return NULL.
65 struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
67 if (!msk->subflow || READ_ONCE(msk->can_ack))
73 /* Returns end sequence number of the receiver's advertised window */
74 static u64 mptcp_wnd_end(const struct mptcp_sock *msk)
76 return READ_ONCE(msk->wnd_end);
79 static bool mptcp_is_tcpsk(struct sock *sk)
81 struct socket *sock = sk->sk_socket;
83 if (unlikely(sk->sk_prot == &tcp_prot)) {
84 /* we are being invoked after mptcp_accept() has
85 * accepted a non-mp-capable flow: sk is a tcp_sk,
88 * Hand the socket over to tcp so all further socket ops
91 sock->ops = &inet_stream_ops;
93 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
94 } else if (unlikely(sk->sk_prot == &tcpv6_prot)) {
95 sock->ops = &inet6_stream_ops;
103 static int __mptcp_socket_create(struct mptcp_sock *msk)
105 struct mptcp_subflow_context *subflow;
106 struct sock *sk = (struct sock *)msk;
107 struct socket *ssock;
110 err = mptcp_subflow_create_socket(sk, &ssock);
114 msk->first = ssock->sk;
115 msk->subflow = ssock;
116 subflow = mptcp_subflow_ctx(ssock->sk);
117 list_add(&subflow->node, &msk->conn_list);
118 sock_hold(ssock->sk);
119 subflow->request_mptcp = 1;
121 /* This is the first subflow, always with id 0 */
122 subflow->local_id_valid = 1;
123 mptcp_sock_graft(msk->first, sk->sk_socket);
128 static void mptcp_drop(struct sock *sk, struct sk_buff *skb)
130 sk_drops_add(sk, skb);
134 static void mptcp_rmem_charge(struct sock *sk, int size)
136 mptcp_sk(sk)->rmem_fwd_alloc -= size;
139 static bool mptcp_try_coalesce(struct sock *sk, struct sk_buff *to,
140 struct sk_buff *from)
145 if (MPTCP_SKB_CB(from)->offset ||
146 !skb_try_coalesce(to, from, &fragstolen, &delta))
149 pr_debug("colesced seq %llx into %llx new len %d new end seq %llx",
150 MPTCP_SKB_CB(from)->map_seq, MPTCP_SKB_CB(to)->map_seq,
151 to->len, MPTCP_SKB_CB(from)->end_seq);
152 MPTCP_SKB_CB(to)->end_seq = MPTCP_SKB_CB(from)->end_seq;
153 kfree_skb_partial(from, fragstolen);
154 atomic_add(delta, &sk->sk_rmem_alloc);
155 mptcp_rmem_charge(sk, delta);
159 static bool mptcp_ooo_try_coalesce(struct mptcp_sock *msk, struct sk_buff *to,
160 struct sk_buff *from)
162 if (MPTCP_SKB_CB(from)->map_seq != MPTCP_SKB_CB(to)->end_seq)
165 return mptcp_try_coalesce((struct sock *)msk, to, from);
168 static void __mptcp_rmem_reclaim(struct sock *sk, int amount)
170 amount >>= SK_MEM_QUANTUM_SHIFT;
171 mptcp_sk(sk)->rmem_fwd_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
172 __sk_mem_reduce_allocated(sk, amount);
175 static void mptcp_rmem_uncharge(struct sock *sk, int size)
177 struct mptcp_sock *msk = mptcp_sk(sk);
180 msk->rmem_fwd_alloc += size;
181 reclaimable = msk->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
183 /* see sk_mem_uncharge() for the rationale behind the following schema */
184 if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD))
185 __mptcp_rmem_reclaim(sk, SK_RECLAIM_CHUNK);
188 static void mptcp_rfree(struct sk_buff *skb)
190 unsigned int len = skb->truesize;
191 struct sock *sk = skb->sk;
193 atomic_sub(len, &sk->sk_rmem_alloc);
194 mptcp_rmem_uncharge(sk, len);
197 static void mptcp_set_owner_r(struct sk_buff *skb, struct sock *sk)
201 skb->destructor = mptcp_rfree;
202 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
203 mptcp_rmem_charge(sk, skb->truesize);
206 /* "inspired" by tcp_data_queue_ofo(), main differences:
208 * - don't cope with sacks
210 static void mptcp_data_queue_ofo(struct mptcp_sock *msk, struct sk_buff *skb)
212 struct sock *sk = (struct sock *)msk;
213 struct rb_node **p, *parent;
214 u64 seq, end_seq, max_seq;
215 struct sk_buff *skb1;
217 seq = MPTCP_SKB_CB(skb)->map_seq;
218 end_seq = MPTCP_SKB_CB(skb)->end_seq;
219 max_seq = atomic64_read(&msk->rcv_wnd_sent);
221 pr_debug("msk=%p seq=%llx limit=%llx empty=%d", msk, seq, max_seq,
222 RB_EMPTY_ROOT(&msk->out_of_order_queue));
223 if (after64(end_seq, max_seq)) {
226 pr_debug("oow by %lld, rcv_wnd_sent %llu\n",
227 (unsigned long long)end_seq - (unsigned long)max_seq,
228 (unsigned long long)atomic64_read(&msk->rcv_wnd_sent));
229 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_NODSSWINDOW);
233 p = &msk->out_of_order_queue.rb_node;
234 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUE);
235 if (RB_EMPTY_ROOT(&msk->out_of_order_queue)) {
236 rb_link_node(&skb->rbnode, NULL, p);
237 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
238 msk->ooo_last_skb = skb;
242 /* with 2 subflows, adding at end of ooo queue is quite likely
243 * Use of ooo_last_skb avoids the O(Log(N)) rbtree lookup.
245 if (mptcp_ooo_try_coalesce(msk, msk->ooo_last_skb, skb)) {
246 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
247 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
251 /* Can avoid an rbtree lookup if we are adding skb after ooo_last_skb */
252 if (!before64(seq, MPTCP_SKB_CB(msk->ooo_last_skb)->end_seq)) {
253 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOQUEUETAIL);
254 parent = &msk->ooo_last_skb->rbnode;
255 p = &parent->rb_right;
259 /* Find place to insert this segment. Handle overlaps on the way. */
263 skb1 = rb_to_skb(parent);
264 if (before64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
265 p = &parent->rb_left;
268 if (before64(seq, MPTCP_SKB_CB(skb1)->end_seq)) {
269 if (!after64(end_seq, MPTCP_SKB_CB(skb1)->end_seq)) {
270 /* All the bits are present. Drop. */
272 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
275 if (after64(seq, MPTCP_SKB_CB(skb1)->map_seq)) {
279 * continue traversing
282 /* skb's seq == skb1's seq and skb covers skb1.
283 * Replace skb1 with skb.
285 rb_replace_node(&skb1->rbnode, &skb->rbnode,
286 &msk->out_of_order_queue);
287 mptcp_drop(sk, skb1);
288 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
291 } else if (mptcp_ooo_try_coalesce(msk, skb1, skb)) {
292 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_OFOMERGE);
295 p = &parent->rb_right;
299 /* Insert segment into RB tree. */
300 rb_link_node(&skb->rbnode, parent, p);
301 rb_insert_color(&skb->rbnode, &msk->out_of_order_queue);
304 /* Remove other segments covered by skb. */
305 while ((skb1 = skb_rb_next(skb)) != NULL) {
306 if (before64(end_seq, MPTCP_SKB_CB(skb1)->end_seq))
308 rb_erase(&skb1->rbnode, &msk->out_of_order_queue);
309 mptcp_drop(sk, skb1);
310 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
312 /* If there is no skb after us, we are the last_skb ! */
314 msk->ooo_last_skb = skb;
318 mptcp_set_owner_r(skb, sk);
321 static bool mptcp_rmem_schedule(struct sock *sk, struct sock *ssk, int size)
323 struct mptcp_sock *msk = mptcp_sk(sk);
326 if (size <= msk->rmem_fwd_alloc)
329 size -= msk->rmem_fwd_alloc;
330 amt = sk_mem_pages(size);
331 amount = amt << SK_MEM_QUANTUM_SHIFT;
332 msk->rmem_fwd_alloc += amount;
333 if (!__sk_mem_raise_allocated(sk, size, amt, SK_MEM_RECV)) {
334 if (ssk->sk_forward_alloc < amount) {
335 msk->rmem_fwd_alloc -= amount;
339 ssk->sk_forward_alloc -= amount;
344 static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
345 struct sk_buff *skb, unsigned int offset,
348 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
349 struct sock *sk = (struct sock *)msk;
350 struct sk_buff *tail;
353 __skb_unlink(skb, &ssk->sk_receive_queue);
358 /* try to fetch required memory from subflow */
359 if (!mptcp_rmem_schedule(sk, ssk, skb->truesize))
362 has_rxtstamp = TCP_SKB_CB(skb)->has_rxtstamp;
364 /* the skb map_seq accounts for the skb offset:
365 * mptcp_subflow_get_mapped_dsn() is based on the current tp->copied_seq
368 MPTCP_SKB_CB(skb)->map_seq = mptcp_subflow_get_mapped_dsn(subflow);
369 MPTCP_SKB_CB(skb)->end_seq = MPTCP_SKB_CB(skb)->map_seq + copy_len;
370 MPTCP_SKB_CB(skb)->offset = offset;
371 MPTCP_SKB_CB(skb)->has_rxtstamp = has_rxtstamp;
373 if (MPTCP_SKB_CB(skb)->map_seq == msk->ack_seq) {
375 WRITE_ONCE(msk->ack_seq, msk->ack_seq + copy_len);
376 tail = skb_peek_tail(&sk->sk_receive_queue);
377 if (tail && mptcp_try_coalesce(sk, tail, skb))
380 mptcp_set_owner_r(skb, sk);
381 __skb_queue_tail(&sk->sk_receive_queue, skb);
383 } else if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq)) {
384 mptcp_data_queue_ofo(msk, skb);
388 /* old data, keep it simple and drop the whole pkt, sender
389 * will retransmit as needed, if needed.
391 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
397 static void mptcp_stop_timer(struct sock *sk)
399 struct inet_connection_sock *icsk = inet_csk(sk);
401 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
402 mptcp_sk(sk)->timer_ival = 0;
405 static void mptcp_close_wake_up(struct sock *sk)
407 if (sock_flag(sk, SOCK_DEAD))
410 sk->sk_state_change(sk);
411 if (sk->sk_shutdown == SHUTDOWN_MASK ||
412 sk->sk_state == TCP_CLOSE)
413 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
415 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
418 static bool mptcp_pending_data_fin_ack(struct sock *sk)
420 struct mptcp_sock *msk = mptcp_sk(sk);
422 return !__mptcp_check_fallback(msk) &&
423 ((1 << sk->sk_state) &
424 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK)) &&
425 msk->write_seq == READ_ONCE(msk->snd_una);
428 static void mptcp_check_data_fin_ack(struct sock *sk)
430 struct mptcp_sock *msk = mptcp_sk(sk);
432 /* Look for an acknowledged DATA_FIN */
433 if (mptcp_pending_data_fin_ack(sk)) {
434 WRITE_ONCE(msk->snd_data_fin_enable, 0);
436 switch (sk->sk_state) {
438 inet_sk_state_store(sk, TCP_FIN_WAIT2);
442 inet_sk_state_store(sk, TCP_CLOSE);
446 mptcp_close_wake_up(sk);
450 static bool mptcp_pending_data_fin(struct sock *sk, u64 *seq)
452 struct mptcp_sock *msk = mptcp_sk(sk);
454 if (READ_ONCE(msk->rcv_data_fin) &&
455 ((1 << sk->sk_state) &
456 (TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2))) {
457 u64 rcv_data_fin_seq = READ_ONCE(msk->rcv_data_fin_seq);
459 if (msk->ack_seq == rcv_data_fin_seq) {
461 *seq = rcv_data_fin_seq;
470 static void mptcp_set_datafin_timeout(const struct sock *sk)
472 struct inet_connection_sock *icsk = inet_csk(sk);
475 retransmits = min_t(u32, icsk->icsk_retransmits,
476 ilog2(TCP_RTO_MAX / TCP_RTO_MIN));
478 mptcp_sk(sk)->timer_ival = TCP_RTO_MIN << retransmits;
481 static void __mptcp_set_timeout(struct sock *sk, long tout)
483 mptcp_sk(sk)->timer_ival = tout > 0 ? tout : TCP_RTO_MIN;
486 static long mptcp_timeout_from_subflow(const struct mptcp_subflow_context *subflow)
488 const struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
490 return inet_csk(ssk)->icsk_pending && !subflow->stale_count ?
491 inet_csk(ssk)->icsk_timeout - jiffies : 0;
494 static void mptcp_set_timeout(struct sock *sk)
496 struct mptcp_subflow_context *subflow;
499 mptcp_for_each_subflow(mptcp_sk(sk), subflow)
500 tout = max(tout, mptcp_timeout_from_subflow(subflow));
501 __mptcp_set_timeout(sk, tout);
504 static inline bool tcp_can_send_ack(const struct sock *ssk)
506 return !((1 << inet_sk_state_load(ssk)) &
507 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_TIME_WAIT | TCPF_CLOSE | TCPF_LISTEN));
510 void __mptcp_subflow_send_ack(struct sock *ssk)
512 if (tcp_can_send_ack(ssk))
516 void mptcp_subflow_send_ack(struct sock *ssk)
520 slow = lock_sock_fast(ssk);
521 __mptcp_subflow_send_ack(ssk);
522 unlock_sock_fast(ssk, slow);
525 static void mptcp_send_ack(struct mptcp_sock *msk)
527 struct mptcp_subflow_context *subflow;
529 mptcp_for_each_subflow(msk, subflow)
530 mptcp_subflow_send_ack(mptcp_subflow_tcp_sock(subflow));
533 static void mptcp_subflow_cleanup_rbuf(struct sock *ssk)
537 slow = lock_sock_fast(ssk);
538 if (tcp_can_send_ack(ssk))
539 tcp_cleanup_rbuf(ssk, 1);
540 unlock_sock_fast(ssk, slow);
543 static bool mptcp_subflow_could_cleanup(const struct sock *ssk, bool rx_empty)
545 const struct inet_connection_sock *icsk = inet_csk(ssk);
546 u8 ack_pending = READ_ONCE(icsk->icsk_ack.pending);
547 const struct tcp_sock *tp = tcp_sk(ssk);
549 return (ack_pending & ICSK_ACK_SCHED) &&
550 ((READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->rcv_wup) >
551 READ_ONCE(icsk->icsk_ack.rcv_mss)) ||
552 (rx_empty && ack_pending &
553 (ICSK_ACK_PUSHED2 | ICSK_ACK_PUSHED)));
556 static void mptcp_cleanup_rbuf(struct mptcp_sock *msk)
558 int old_space = READ_ONCE(msk->old_wspace);
559 struct mptcp_subflow_context *subflow;
560 struct sock *sk = (struct sock *)msk;
561 int space = __mptcp_space(sk);
562 bool cleanup, rx_empty;
564 cleanup = (space > 0) && (space >= (old_space << 1));
565 rx_empty = !__mptcp_rmem(sk);
567 mptcp_for_each_subflow(msk, subflow) {
568 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
570 if (cleanup || mptcp_subflow_could_cleanup(ssk, rx_empty))
571 mptcp_subflow_cleanup_rbuf(ssk);
575 static bool mptcp_check_data_fin(struct sock *sk)
577 struct mptcp_sock *msk = mptcp_sk(sk);
578 u64 rcv_data_fin_seq;
581 if (__mptcp_check_fallback(msk))
584 /* Need to ack a DATA_FIN received from a peer while this side
585 * of the connection is in ESTABLISHED, FIN_WAIT1, or FIN_WAIT2.
586 * msk->rcv_data_fin was set when parsing the incoming options
587 * at the subflow level and the msk lock was not held, so this
588 * is the first opportunity to act on the DATA_FIN and change
591 * If we are caught up to the sequence number of the incoming
592 * DATA_FIN, send the DATA_ACK now and do state transition. If
593 * not caught up, do nothing and let the recv code send DATA_ACK
597 if (mptcp_pending_data_fin(sk, &rcv_data_fin_seq)) {
598 WRITE_ONCE(msk->ack_seq, msk->ack_seq + 1);
599 WRITE_ONCE(msk->rcv_data_fin, 0);
601 sk->sk_shutdown |= RCV_SHUTDOWN;
602 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
604 switch (sk->sk_state) {
605 case TCP_ESTABLISHED:
606 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
609 inet_sk_state_store(sk, TCP_CLOSING);
612 inet_sk_state_store(sk, TCP_CLOSE);
615 /* Other states not expected */
622 mptcp_close_wake_up(sk);
627 static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock *msk,
631 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
632 struct sock *sk = (struct sock *)msk;
633 unsigned int moved = 0;
634 bool more_data_avail;
639 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
641 if (!(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
642 int ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
644 if (unlikely(ssk_rbuf > sk_rbuf)) {
645 WRITE_ONCE(sk->sk_rcvbuf, ssk_rbuf);
650 pr_debug("msk=%p ssk=%p", msk, ssk);
653 u32 map_remaining, offset;
654 u32 seq = tp->copied_seq;
658 /* try to move as much data as available */
659 map_remaining = subflow->map_data_len -
660 mptcp_subflow_get_map_offset(subflow);
662 skb = skb_peek(&ssk->sk_receive_queue);
664 /* if no data is found, a racing workqueue/recvmsg
665 * already processed the new data, stop here or we
666 * can enter an infinite loop
673 if (__mptcp_check_fallback(msk)) {
674 /* if we are running under the workqueue, TCP could have
675 * collapsed skbs between dummy map creation and now
676 * be sure to adjust the size
678 map_remaining = skb->len;
679 subflow->map_data_len = skb->len;
682 offset = seq - TCP_SKB_CB(skb)->seq;
683 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
689 if (offset < skb->len) {
690 size_t len = skb->len - offset;
695 if (__mptcp_move_skb(msk, ssk, skb, offset, len))
699 if (WARN_ON_ONCE(map_remaining < len))
703 sk_eat_skb(ssk, skb);
707 WRITE_ONCE(tp->copied_seq, seq);
708 more_data_avail = mptcp_subflow_data_available(ssk);
710 if (atomic_read(&sk->sk_rmem_alloc) > sk_rbuf) {
714 } while (more_data_avail);
720 static bool __mptcp_ofo_queue(struct mptcp_sock *msk)
722 struct sock *sk = (struct sock *)msk;
723 struct sk_buff *skb, *tail;
728 p = rb_first(&msk->out_of_order_queue);
729 pr_debug("msk=%p empty=%d", msk, RB_EMPTY_ROOT(&msk->out_of_order_queue));
732 if (after64(MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq))
736 rb_erase(&skb->rbnode, &msk->out_of_order_queue);
738 if (unlikely(!after64(MPTCP_SKB_CB(skb)->end_seq,
741 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_DUPDATA);
745 end_seq = MPTCP_SKB_CB(skb)->end_seq;
746 tail = skb_peek_tail(&sk->sk_receive_queue);
747 if (!tail || !mptcp_ooo_try_coalesce(msk, tail, skb)) {
748 int delta = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
750 /* skip overlapping data, if any */
751 pr_debug("uncoalesced seq=%llx ack seq=%llx delta=%d",
752 MPTCP_SKB_CB(skb)->map_seq, msk->ack_seq,
754 MPTCP_SKB_CB(skb)->offset += delta;
755 MPTCP_SKB_CB(skb)->map_seq += delta;
756 __skb_queue_tail(&sk->sk_receive_queue, skb);
758 msk->ack_seq = end_seq;
764 /* In most cases we will be able to lock the mptcp socket. If its already
765 * owned, we need to defer to the work queue to avoid ABBA deadlock.
767 static bool move_skbs_to_msk(struct mptcp_sock *msk, struct sock *ssk)
769 struct sock *sk = (struct sock *)msk;
770 unsigned int moved = 0;
772 __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
773 __mptcp_ofo_queue(msk);
774 if (unlikely(ssk->sk_err)) {
775 if (!sock_owned_by_user(sk))
776 __mptcp_error_report(sk);
778 __set_bit(MPTCP_ERROR_REPORT, &msk->cb_flags);
781 /* If the moves have caught up with the DATA_FIN sequence number
782 * it's time to ack the DATA_FIN and change socket state, but
783 * this is not a good place to change state. Let the workqueue
786 if (mptcp_pending_data_fin(sk, NULL))
787 mptcp_schedule_work(sk);
791 void mptcp_data_ready(struct sock *sk, struct sock *ssk)
793 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
794 struct mptcp_sock *msk = mptcp_sk(sk);
795 int sk_rbuf, ssk_rbuf;
797 /* The peer can send data while we are shutting down this
798 * subflow at msk destruction time, but we must avoid enqueuing
799 * more data to the msk receive queue
801 if (unlikely(subflow->disposable))
804 ssk_rbuf = READ_ONCE(ssk->sk_rcvbuf);
805 sk_rbuf = READ_ONCE(sk->sk_rcvbuf);
806 if (unlikely(ssk_rbuf > sk_rbuf))
809 /* over limit? can't append more skbs to msk, Also, no need to wake-up*/
810 if (__mptcp_rmem(sk) > sk_rbuf) {
811 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RCVPRUNED);
815 /* Wake-up the reader only for in-sequence data */
817 if (move_skbs_to_msk(msk, ssk))
818 sk->sk_data_ready(sk);
820 mptcp_data_unlock(sk);
823 static bool __mptcp_finish_join(struct mptcp_sock *msk, struct sock *ssk)
825 struct sock *sk = (struct sock *)msk;
827 if (sk->sk_state != TCP_ESTABLISHED)
830 /* attach to msk socket only after we are sure we will deal with it
833 if (sk->sk_socket && !ssk->sk_socket)
834 mptcp_sock_graft(ssk, sk->sk_socket);
836 mptcp_propagate_sndbuf((struct sock *)msk, ssk);
837 mptcp_sockopt_sync_locked(msk, ssk);
841 static void __mptcp_flush_join_list(struct sock *sk)
843 struct mptcp_subflow_context *tmp, *subflow;
844 struct mptcp_sock *msk = mptcp_sk(sk);
846 list_for_each_entry_safe(subflow, tmp, &msk->join_list, node) {
847 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
848 bool slow = lock_sock_fast(ssk);
850 list_move_tail(&subflow->node, &msk->conn_list);
851 if (!__mptcp_finish_join(msk, ssk))
852 mptcp_subflow_reset(ssk);
853 unlock_sock_fast(ssk, slow);
857 static bool mptcp_timer_pending(struct sock *sk)
859 return timer_pending(&inet_csk(sk)->icsk_retransmit_timer);
862 static void mptcp_reset_timer(struct sock *sk)
864 struct inet_connection_sock *icsk = inet_csk(sk);
867 /* prevent rescheduling on close */
868 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE))
871 tout = mptcp_sk(sk)->timer_ival;
872 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + tout);
875 bool mptcp_schedule_work(struct sock *sk)
877 if (inet_sk_state_load(sk) != TCP_CLOSE &&
878 schedule_work(&mptcp_sk(sk)->work)) {
879 /* each subflow already holds a reference to the sk, and the
880 * workqueue is invoked by a subflow, so sk can't go away here.
888 void mptcp_subflow_eof(struct sock *sk)
890 if (!test_and_set_bit(MPTCP_WORK_EOF, &mptcp_sk(sk)->flags))
891 mptcp_schedule_work(sk);
894 static void mptcp_check_for_eof(struct mptcp_sock *msk)
896 struct mptcp_subflow_context *subflow;
897 struct sock *sk = (struct sock *)msk;
900 mptcp_for_each_subflow(msk, subflow)
901 receivers += !subflow->rx_eof;
905 if (!(sk->sk_shutdown & RCV_SHUTDOWN)) {
906 /* hopefully temporary hack: propagate shutdown status
907 * to msk, when all subflows agree on it
909 sk->sk_shutdown |= RCV_SHUTDOWN;
911 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
912 sk->sk_data_ready(sk);
915 switch (sk->sk_state) {
916 case TCP_ESTABLISHED:
917 inet_sk_state_store(sk, TCP_CLOSE_WAIT);
920 inet_sk_state_store(sk, TCP_CLOSING);
923 inet_sk_state_store(sk, TCP_CLOSE);
928 mptcp_close_wake_up(sk);
931 static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
933 struct mptcp_subflow_context *subflow;
934 struct sock *sk = (struct sock *)msk;
936 sock_owned_by_me(sk);
938 mptcp_for_each_subflow(msk, subflow) {
939 if (READ_ONCE(subflow->data_avail))
940 return mptcp_subflow_tcp_sock(subflow);
946 static bool mptcp_skb_can_collapse_to(u64 write_seq,
947 const struct sk_buff *skb,
948 const struct mptcp_ext *mpext)
950 if (!tcp_skb_can_collapse_to(skb))
953 /* can collapse only if MPTCP level sequence is in order and this
954 * mapping has not been xmitted yet
956 return mpext && mpext->data_seq + mpext->data_len == write_seq &&
960 /* we can append data to the given data frag if:
961 * - there is space available in the backing page_frag
962 * - the data frag tail matches the current page_frag free offset
963 * - the data frag end sequence number matches the current write seq
965 static bool mptcp_frag_can_collapse_to(const struct mptcp_sock *msk,
966 const struct page_frag *pfrag,
967 const struct mptcp_data_frag *df)
969 return df && pfrag->page == df->page &&
970 pfrag->size - pfrag->offset > 0 &&
971 pfrag->offset == (df->offset + df->data_len) &&
972 df->data_seq + df->data_len == msk->write_seq;
975 static void __mptcp_mem_reclaim_partial(struct sock *sk)
977 int reclaimable = mptcp_sk(sk)->rmem_fwd_alloc - sk_unused_reserved_mem(sk);
979 lockdep_assert_held_once(&sk->sk_lock.slock);
981 if (reclaimable > SK_MEM_QUANTUM)
982 __mptcp_rmem_reclaim(sk, reclaimable - 1);
984 sk_mem_reclaim_partial(sk);
987 static void mptcp_mem_reclaim_partial(struct sock *sk)
990 __mptcp_mem_reclaim_partial(sk);
991 mptcp_data_unlock(sk);
994 static void dfrag_uncharge(struct sock *sk, int len)
996 sk_mem_uncharge(sk, len);
997 sk_wmem_queued_add(sk, -len);
1000 static void dfrag_clear(struct sock *sk, struct mptcp_data_frag *dfrag)
1002 int len = dfrag->data_len + dfrag->overhead;
1004 list_del(&dfrag->list);
1005 dfrag_uncharge(sk, len);
1006 put_page(dfrag->page);
1009 static void __mptcp_clean_una(struct sock *sk)
1011 struct mptcp_sock *msk = mptcp_sk(sk);
1012 struct mptcp_data_frag *dtmp, *dfrag;
1013 bool cleaned = false;
1016 /* on fallback we just need to ignore snd_una, as this is really
1019 if (__mptcp_check_fallback(msk))
1020 msk->snd_una = READ_ONCE(msk->snd_nxt);
1022 snd_una = msk->snd_una;
1023 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list) {
1024 if (after64(dfrag->data_seq + dfrag->data_len, snd_una))
1027 if (unlikely(dfrag == msk->first_pending)) {
1028 /* in recovery mode can see ack after the current snd head */
1029 if (WARN_ON_ONCE(!msk->recovery))
1032 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1035 dfrag_clear(sk, dfrag);
1039 dfrag = mptcp_rtx_head(sk);
1040 if (dfrag && after64(snd_una, dfrag->data_seq)) {
1041 u64 delta = snd_una - dfrag->data_seq;
1043 /* prevent wrap around in recovery mode */
1044 if (unlikely(delta > dfrag->already_sent)) {
1045 if (WARN_ON_ONCE(!msk->recovery))
1047 if (WARN_ON_ONCE(delta > dfrag->data_len))
1049 dfrag->already_sent += delta - dfrag->already_sent;
1052 dfrag->data_seq += delta;
1053 dfrag->offset += delta;
1054 dfrag->data_len -= delta;
1055 dfrag->already_sent -= delta;
1057 dfrag_uncharge(sk, delta);
1061 /* all retransmitted data acked, recovery completed */
1062 if (unlikely(msk->recovery) && after64(msk->snd_una, msk->recovery_snd_nxt))
1063 msk->recovery = false;
1066 if (cleaned && tcp_under_memory_pressure(sk))
1067 __mptcp_mem_reclaim_partial(sk);
1069 if (snd_una == READ_ONCE(msk->snd_nxt) &&
1070 snd_una == READ_ONCE(msk->write_seq)) {
1071 if (mptcp_timer_pending(sk) && !mptcp_data_fin_enabled(msk))
1072 mptcp_stop_timer(sk);
1074 mptcp_reset_timer(sk);
1078 static void __mptcp_clean_una_wakeup(struct sock *sk)
1080 lockdep_assert_held_once(&sk->sk_lock.slock);
1082 __mptcp_clean_una(sk);
1083 mptcp_write_space(sk);
1086 static void mptcp_clean_una_wakeup(struct sock *sk)
1088 mptcp_data_lock(sk);
1089 __mptcp_clean_una_wakeup(sk);
1090 mptcp_data_unlock(sk);
1093 static void mptcp_enter_memory_pressure(struct sock *sk)
1095 struct mptcp_subflow_context *subflow;
1096 struct mptcp_sock *msk = mptcp_sk(sk);
1099 sk_stream_moderate_sndbuf(sk);
1100 mptcp_for_each_subflow(msk, subflow) {
1101 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1104 tcp_enter_memory_pressure(ssk);
1105 sk_stream_moderate_sndbuf(ssk);
1110 /* ensure we get enough memory for the frag hdr, beyond some minimal amount of
1113 static bool mptcp_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1115 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag),
1116 pfrag, sk->sk_allocation)))
1119 mptcp_enter_memory_pressure(sk);
1123 static struct mptcp_data_frag *
1124 mptcp_carve_data_frag(const struct mptcp_sock *msk, struct page_frag *pfrag,
1127 int offset = ALIGN(orig_offset, sizeof(long));
1128 struct mptcp_data_frag *dfrag;
1130 dfrag = (struct mptcp_data_frag *)(page_to_virt(pfrag->page) + offset);
1131 dfrag->data_len = 0;
1132 dfrag->data_seq = msk->write_seq;
1133 dfrag->overhead = offset - orig_offset + sizeof(struct mptcp_data_frag);
1134 dfrag->offset = offset + sizeof(struct mptcp_data_frag);
1135 dfrag->already_sent = 0;
1136 dfrag->page = pfrag->page;
1141 struct mptcp_sendmsg_info {
1147 bool data_lock_held;
1150 static int mptcp_check_allowed_size(const struct mptcp_sock *msk, struct sock *ssk,
1151 u64 data_seq, int avail_size)
1153 u64 window_end = mptcp_wnd_end(msk);
1156 if (__mptcp_check_fallback(msk))
1159 mptcp_snd_wnd = window_end - data_seq;
1160 avail_size = min_t(unsigned int, mptcp_snd_wnd, avail_size);
1162 if (unlikely(tcp_sk(ssk)->snd_wnd < mptcp_snd_wnd)) {
1163 tcp_sk(ssk)->snd_wnd = min_t(u64, U32_MAX, mptcp_snd_wnd);
1164 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_SNDWNDSHARED);
1170 static bool __mptcp_add_ext(struct sk_buff *skb, gfp_t gfp)
1172 struct skb_ext *mpext = __skb_ext_alloc(gfp);
1176 __skb_ext_set(skb, SKB_EXT_MPTCP, mpext);
1180 static struct sk_buff *__mptcp_do_alloc_tx_skb(struct sock *sk, gfp_t gfp)
1182 struct sk_buff *skb;
1184 skb = alloc_skb_fclone(MAX_TCP_HEADER, gfp);
1186 if (likely(__mptcp_add_ext(skb, gfp))) {
1187 skb_reserve(skb, MAX_TCP_HEADER);
1188 skb->ip_summed = CHECKSUM_PARTIAL;
1189 INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
1194 mptcp_enter_memory_pressure(sk);
1199 static struct sk_buff *__mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, gfp_t gfp)
1201 struct sk_buff *skb;
1203 skb = __mptcp_do_alloc_tx_skb(sk, gfp);
1207 if (likely(sk_wmem_schedule(ssk, skb->truesize))) {
1208 tcp_skb_entail(ssk, skb);
1211 tcp_skb_tsorted_anchor_cleanup(skb);
1216 static struct sk_buff *mptcp_alloc_tx_skb(struct sock *sk, struct sock *ssk, bool data_lock_held)
1218 gfp_t gfp = data_lock_held ? GFP_ATOMIC : sk->sk_allocation;
1220 if (unlikely(tcp_under_memory_pressure(sk))) {
1222 __mptcp_mem_reclaim_partial(sk);
1224 mptcp_mem_reclaim_partial(sk);
1226 return __mptcp_alloc_tx_skb(sk, ssk, gfp);
1229 /* note: this always recompute the csum on the whole skb, even
1230 * if we just appended a single frag. More status info needed
1232 static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
1234 struct mptcp_ext *mpext = mptcp_get_ext(skb);
1235 __wsum csum = ~csum_unfold(mpext->csum);
1236 int offset = skb->len - added;
1238 mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
1241 static void mptcp_update_infinite_map(struct mptcp_sock *msk,
1243 struct mptcp_ext *mpext)
1248 mpext->infinite_map = 1;
1249 mpext->data_len = 0;
1251 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPTX);
1252 mptcp_subflow_ctx(ssk)->send_infinite_map = 0;
1254 mptcp_do_fallback(ssk);
1257 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
1258 struct mptcp_data_frag *dfrag,
1259 struct mptcp_sendmsg_info *info)
1261 u64 data_seq = dfrag->data_seq + info->sent;
1262 int offset = dfrag->offset + info->sent;
1263 struct mptcp_sock *msk = mptcp_sk(sk);
1264 bool zero_window_probe = false;
1265 struct mptcp_ext *mpext = NULL;
1266 bool can_coalesce = false;
1267 bool reuse_skb = true;
1268 struct sk_buff *skb;
1272 pr_debug("msk=%p ssk=%p sending dfrag at seq=%llu len=%u already sent=%u",
1273 msk, ssk, dfrag->data_seq, dfrag->data_len, info->sent);
1275 if (WARN_ON_ONCE(info->sent > info->limit ||
1276 info->limit > dfrag->data_len))
1279 if (unlikely(!__tcp_can_send(ssk)))
1282 /* compute send limit */
1283 info->mss_now = tcp_send_mss(ssk, &info->size_goal, info->flags);
1284 copy = info->size_goal;
1286 skb = tcp_write_queue_tail(ssk);
1287 if (skb && copy > skb->len) {
1288 /* Limit the write to the size available in the
1289 * current skb, if any, so that we create at most a new skb.
1290 * Explicitly tells TCP internals to avoid collapsing on later
1291 * queue management operation, to avoid breaking the ext <->
1292 * SSN association set here
1294 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1295 if (!mptcp_skb_can_collapse_to(data_seq, skb, mpext)) {
1296 TCP_SKB_CB(skb)->eor = 1;
1300 i = skb_shinfo(skb)->nr_frags;
1301 can_coalesce = skb_can_coalesce(skb, i, dfrag->page, offset);
1302 if (!can_coalesce && i >= READ_ONCE(sysctl_max_skb_frags)) {
1303 tcp_mark_push(tcp_sk(ssk), skb);
1310 skb = mptcp_alloc_tx_skb(sk, ssk, info->data_lock_held);
1314 i = skb_shinfo(skb)->nr_frags;
1316 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
1319 /* Zero window and all data acked? Probe. */
1320 copy = mptcp_check_allowed_size(msk, ssk, data_seq, copy);
1322 u64 snd_una = READ_ONCE(msk->snd_una);
1324 if (snd_una != msk->snd_nxt) {
1325 tcp_remove_empty_skb(ssk);
1329 zero_window_probe = true;
1330 data_seq = snd_una - 1;
1333 /* all mptcp-level data is acked, no skbs should be present into the
1336 WARN_ON_ONCE(reuse_skb);
1339 copy = min_t(size_t, copy, info->limit - info->sent);
1340 if (!sk_wmem_schedule(ssk, copy)) {
1341 tcp_remove_empty_skb(ssk);
1346 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1348 get_page(dfrag->page);
1349 skb_fill_page_desc(skb, i, dfrag->page, offset, copy);
1353 skb->data_len += copy;
1354 skb->truesize += copy;
1355 sk_wmem_queued_add(ssk, copy);
1356 sk_mem_charge(ssk, copy);
1357 WRITE_ONCE(tcp_sk(ssk)->write_seq, tcp_sk(ssk)->write_seq + copy);
1358 TCP_SKB_CB(skb)->end_seq += copy;
1359 tcp_skb_pcount_set(skb, 0);
1361 /* on skb reuse we just need to update the DSS len */
1363 TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_PSH;
1364 mpext->data_len += copy;
1365 WARN_ON_ONCE(zero_window_probe);
1369 memset(mpext, 0, sizeof(*mpext));
1370 mpext->data_seq = data_seq;
1371 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
1372 mpext->data_len = copy;
1376 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
1377 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1380 if (zero_window_probe) {
1381 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1383 if (READ_ONCE(msk->csum_enabled))
1384 mptcp_update_data_checksum(skb, copy);
1385 tcp_push_pending_frames(ssk);
1389 if (READ_ONCE(msk->csum_enabled))
1390 mptcp_update_data_checksum(skb, copy);
1391 if (mptcp_subflow_ctx(ssk)->send_infinite_map)
1392 mptcp_update_infinite_map(msk, ssk, mpext);
1393 trace_mptcp_sendmsg_frag(mpext);
1394 mptcp_subflow_ctx(ssk)->rel_write_seq += copy;
1398 #define MPTCP_SEND_BURST_SIZE ((1 << 16) - \
1399 sizeof(struct tcphdr) - \
1400 MAX_TCP_OPTION_SPACE - \
1401 sizeof(struct ipv6hdr) - \
1402 sizeof(struct frag_hdr))
1404 struct subflow_send_info {
1409 void mptcp_subflow_set_active(struct mptcp_subflow_context *subflow)
1411 if (!subflow->stale)
1415 MPTCP_INC_STATS(sock_net(mptcp_subflow_tcp_sock(subflow)), MPTCP_MIB_SUBFLOWRECOVER);
1418 bool mptcp_subflow_active(struct mptcp_subflow_context *subflow)
1420 if (unlikely(subflow->stale)) {
1421 u32 rcv_tstamp = READ_ONCE(tcp_sk(mptcp_subflow_tcp_sock(subflow))->rcv_tstamp);
1423 if (subflow->stale_rcv_tstamp == rcv_tstamp)
1426 mptcp_subflow_set_active(subflow);
1428 return __mptcp_subflow_active(subflow);
1431 #define SSK_MODE_ACTIVE 0
1432 #define SSK_MODE_BACKUP 1
1433 #define SSK_MODE_MAX 2
1435 /* implement the mptcp packet scheduler;
1436 * returns the subflow that will transmit the next DSS
1437 * additionally updates the rtx timeout
1439 static struct sock *mptcp_subflow_get_send(struct mptcp_sock *msk)
1441 struct subflow_send_info send_info[SSK_MODE_MAX];
1442 struct mptcp_subflow_context *subflow;
1443 struct sock *sk = (struct sock *)msk;
1444 u32 pace, burst, wmem;
1445 int i, nr_active = 0;
1450 sock_owned_by_me(sk);
1452 if (__mptcp_check_fallback(msk)) {
1455 return __tcp_can_send(msk->first) &&
1456 sk_stream_memory_free(msk->first) ? msk->first : NULL;
1459 /* re-use last subflow, if the burst allow that */
1460 if (msk->last_snd && msk->snd_burst > 0 &&
1461 sk_stream_memory_free(msk->last_snd) &&
1462 mptcp_subflow_active(mptcp_subflow_ctx(msk->last_snd))) {
1463 mptcp_set_timeout(sk);
1464 return msk->last_snd;
1467 /* pick the subflow with the lower wmem/wspace ratio */
1468 for (i = 0; i < SSK_MODE_MAX; ++i) {
1469 send_info[i].ssk = NULL;
1470 send_info[i].linger_time = -1;
1473 mptcp_for_each_subflow(msk, subflow) {
1474 trace_mptcp_subflow_get_send(subflow);
1475 ssk = mptcp_subflow_tcp_sock(subflow);
1476 if (!mptcp_subflow_active(subflow))
1479 tout = max(tout, mptcp_timeout_from_subflow(subflow));
1480 nr_active += !subflow->backup;
1481 pace = subflow->avg_pacing_rate;
1482 if (unlikely(!pace)) {
1483 /* init pacing rate from socket */
1484 subflow->avg_pacing_rate = READ_ONCE(ssk->sk_pacing_rate);
1485 pace = subflow->avg_pacing_rate;
1490 linger_time = div_u64((u64)READ_ONCE(ssk->sk_wmem_queued) << 32, pace);
1491 if (linger_time < send_info[subflow->backup].linger_time) {
1492 send_info[subflow->backup].ssk = ssk;
1493 send_info[subflow->backup].linger_time = linger_time;
1496 __mptcp_set_timeout(sk, tout);
1498 /* pick the best backup if no other subflow is active */
1500 send_info[SSK_MODE_ACTIVE].ssk = send_info[SSK_MODE_BACKUP].ssk;
1502 /* According to the blest algorithm, to avoid HoL blocking for the
1503 * faster flow, we need to:
1504 * - estimate the faster flow linger time
1505 * - use the above to estimate the amount of byte transferred
1506 * by the faster flow
1507 * - check that the amount of queued data is greter than the above,
1508 * otherwise do not use the picked, slower, subflow
1509 * We select the subflow with the shorter estimated time to flush
1510 * the queued mem, which basically ensure the above. We just need
1511 * to check that subflow has a non empty cwin.
1513 ssk = send_info[SSK_MODE_ACTIVE].ssk;
1514 if (!ssk || !sk_stream_memory_free(ssk))
1517 burst = min_t(int, MPTCP_SEND_BURST_SIZE, mptcp_wnd_end(msk) - msk->snd_nxt);
1518 wmem = READ_ONCE(ssk->sk_wmem_queued);
1520 msk->last_snd = NULL;
1524 subflow = mptcp_subflow_ctx(ssk);
1525 subflow->avg_pacing_rate = div_u64((u64)subflow->avg_pacing_rate * wmem +
1526 READ_ONCE(ssk->sk_pacing_rate) * burst,
1528 msk->last_snd = ssk;
1529 msk->snd_burst = burst;
1533 static void mptcp_push_release(struct sock *ssk, struct mptcp_sendmsg_info *info)
1535 tcp_push(ssk, 0, info->mss_now, tcp_sk(ssk)->nonagle, info->size_goal);
1539 static void mptcp_update_post_push(struct mptcp_sock *msk,
1540 struct mptcp_data_frag *dfrag,
1543 u64 snd_nxt_new = dfrag->data_seq;
1545 dfrag->already_sent += sent;
1547 msk->snd_burst -= sent;
1549 snd_nxt_new += dfrag->already_sent;
1551 /* snd_nxt_new can be smaller than snd_nxt in case mptcp
1552 * is recovering after a failover. In that event, this re-sends
1555 * Thus compute snd_nxt_new candidate based on
1556 * the dfrag->data_seq that was sent and the data
1557 * that has been handed to the subflow for transmission
1558 * and skip update in case it was old dfrag.
1560 if (likely(after64(snd_nxt_new, msk->snd_nxt)))
1561 msk->snd_nxt = snd_nxt_new;
1564 void mptcp_check_and_set_pending(struct sock *sk)
1566 if (mptcp_send_head(sk))
1567 mptcp_sk(sk)->push_pending |= BIT(MPTCP_PUSH_PENDING);
1570 void __mptcp_push_pending(struct sock *sk, unsigned int flags)
1572 struct sock *prev_ssk = NULL, *ssk = NULL;
1573 struct mptcp_sock *msk = mptcp_sk(sk);
1574 struct mptcp_sendmsg_info info = {
1577 struct mptcp_data_frag *dfrag;
1578 int len, copied = 0;
1580 while ((dfrag = mptcp_send_head(sk))) {
1581 info.sent = dfrag->already_sent;
1582 info.limit = dfrag->data_len;
1583 len = dfrag->data_len - dfrag->already_sent;
1588 ssk = mptcp_subflow_get_send(msk);
1590 /* First check. If the ssk has changed since
1591 * the last round, release prev_ssk
1593 if (ssk != prev_ssk && prev_ssk)
1594 mptcp_push_release(prev_ssk, &info);
1598 /* Need to lock the new subflow only if different
1599 * from the previous one, otherwise we are still
1600 * helding the relevant lock
1602 if (ssk != prev_ssk)
1605 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1609 mptcp_push_release(ssk, &info);
1617 mptcp_update_post_push(msk, dfrag, ret);
1619 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1622 /* at this point we held the socket lock for the last subflow we used */
1624 mptcp_push_release(ssk, &info);
1627 /* ensure the rtx timer is running */
1628 if (!mptcp_timer_pending(sk))
1629 mptcp_reset_timer(sk);
1631 __mptcp_check_send_data_fin(sk);
1634 static void __mptcp_subflow_push_pending(struct sock *sk, struct sock *ssk)
1636 struct mptcp_sock *msk = mptcp_sk(sk);
1637 struct mptcp_sendmsg_info info = {
1638 .data_lock_held = true,
1640 struct mptcp_data_frag *dfrag;
1641 struct sock *xmit_ssk;
1642 int len, copied = 0;
1646 while ((dfrag = mptcp_send_head(sk))) {
1647 info.sent = dfrag->already_sent;
1648 info.limit = dfrag->data_len;
1649 len = dfrag->data_len - dfrag->already_sent;
1653 /* the caller already invoked the packet scheduler,
1654 * check for a different subflow usage only after
1655 * spooling the first chunk of data
1657 xmit_ssk = first ? ssk : mptcp_subflow_get_send(mptcp_sk(sk));
1660 if (xmit_ssk != ssk) {
1661 mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk),
1662 MPTCP_DELEGATE_SEND);
1666 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
1675 mptcp_update_post_push(msk, dfrag, ret);
1677 WRITE_ONCE(msk->first_pending, mptcp_send_next(sk));
1681 /* __mptcp_alloc_tx_skb could have released some wmem and we are
1682 * not going to flush it via release_sock()
1685 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
1687 if (!mptcp_timer_pending(sk))
1688 mptcp_reset_timer(sk);
1690 if (msk->snd_data_fin_enable &&
1691 msk->snd_nxt + 1 == msk->write_seq)
1692 mptcp_schedule_work(sk);
1696 static void mptcp_set_nospace(struct sock *sk)
1698 /* enable autotune */
1699 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1701 /* will be cleared on avail space */
1702 set_bit(MPTCP_NOSPACE, &mptcp_sk(sk)->flags);
1705 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1707 struct mptcp_sock *msk = mptcp_sk(sk);
1708 struct page_frag *pfrag;
1713 /* we don't support FASTOPEN yet */
1714 if (msg->msg_flags & MSG_FASTOPEN)
1717 /* silently ignore everything else */
1718 msg->msg_flags &= MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL;
1722 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1724 if ((1 << sk->sk_state) & ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
1725 ret = sk_stream_wait_connect(sk, &timeo);
1730 pfrag = sk_page_frag(sk);
1732 while (msg_data_left(msg)) {
1733 int total_ts, frag_truesize = 0;
1734 struct mptcp_data_frag *dfrag;
1735 bool dfrag_collapsed;
1736 size_t psize, offset;
1738 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN)) {
1743 /* reuse tail pfrag, if possible, or carve a new one from the
1746 dfrag = mptcp_pending_tail(sk);
1747 dfrag_collapsed = mptcp_frag_can_collapse_to(msk, pfrag, dfrag);
1748 if (!dfrag_collapsed) {
1749 if (!sk_stream_memory_free(sk))
1750 goto wait_for_memory;
1752 if (!mptcp_page_frag_refill(sk, pfrag))
1753 goto wait_for_memory;
1755 dfrag = mptcp_carve_data_frag(msk, pfrag, pfrag->offset);
1756 frag_truesize = dfrag->overhead;
1759 /* we do not bound vs wspace, to allow a single packet.
1760 * memory accounting will prevent execessive memory usage
1763 offset = dfrag->offset + dfrag->data_len;
1764 psize = pfrag->size - offset;
1765 psize = min_t(size_t, psize, msg_data_left(msg));
1766 total_ts = psize + frag_truesize;
1768 if (!sk_wmem_schedule(sk, total_ts))
1769 goto wait_for_memory;
1771 if (copy_page_from_iter(dfrag->page, offset, psize,
1772 &msg->msg_iter) != psize) {
1777 /* data successfully copied into the write queue */
1778 sk->sk_forward_alloc -= total_ts;
1780 dfrag->data_len += psize;
1781 frag_truesize += psize;
1782 pfrag->offset += frag_truesize;
1783 WRITE_ONCE(msk->write_seq, msk->write_seq + psize);
1785 /* charge data on mptcp pending queue to the msk socket
1786 * Note: we charge such data both to sk and ssk
1788 sk_wmem_queued_add(sk, frag_truesize);
1789 if (!dfrag_collapsed) {
1790 get_page(dfrag->page);
1791 list_add_tail(&dfrag->list, &msk->rtx_queue);
1792 if (!msk->first_pending)
1793 WRITE_ONCE(msk->first_pending, dfrag);
1795 pr_debug("msk=%p dfrag at seq=%llu len=%u sent=%u new=%d", msk,
1796 dfrag->data_seq, dfrag->data_len, dfrag->already_sent,
1802 mptcp_set_nospace(sk);
1803 __mptcp_push_pending(sk, msg->msg_flags);
1804 ret = sk_stream_wait_memory(sk, &timeo);
1810 __mptcp_push_pending(sk, msg->msg_flags);
1814 return copied ? : ret;
1817 static int __mptcp_recvmsg_mskq(struct mptcp_sock *msk,
1819 size_t len, int flags,
1820 struct scm_timestamping_internal *tss,
1823 struct sk_buff *skb, *tmp;
1826 skb_queue_walk_safe(&msk->receive_queue, skb, tmp) {
1827 u32 offset = MPTCP_SKB_CB(skb)->offset;
1828 u32 data_len = skb->len - offset;
1829 u32 count = min_t(size_t, len - copied, data_len);
1832 if (!(flags & MSG_TRUNC)) {
1833 err = skb_copy_datagram_msg(skb, offset, msg, count);
1834 if (unlikely(err < 0)) {
1841 if (MPTCP_SKB_CB(skb)->has_rxtstamp) {
1842 tcp_update_recv_tstamps(skb, tss);
1843 *cmsg_flags |= MPTCP_CMSG_TS;
1848 if (count < data_len) {
1849 if (!(flags & MSG_PEEK)) {
1850 MPTCP_SKB_CB(skb)->offset += count;
1851 MPTCP_SKB_CB(skb)->map_seq += count;
1856 if (!(flags & MSG_PEEK)) {
1857 /* we will bulk release the skb memory later */
1858 skb->destructor = NULL;
1859 WRITE_ONCE(msk->rmem_released, msk->rmem_released + skb->truesize);
1860 __skb_unlink(skb, &msk->receive_queue);
1871 /* receive buffer autotuning. See tcp_rcv_space_adjust for more information.
1873 * Only difference: Use highest rtt estimate of the subflows in use.
1875 static void mptcp_rcv_space_adjust(struct mptcp_sock *msk, int copied)
1877 struct mptcp_subflow_context *subflow;
1878 struct sock *sk = (struct sock *)msk;
1879 u32 time, advmss = 1;
1882 sock_owned_by_me(sk);
1887 msk->rcvq_space.copied += copied;
1889 mstamp = div_u64(tcp_clock_ns(), NSEC_PER_USEC);
1890 time = tcp_stamp_us_delta(mstamp, msk->rcvq_space.time);
1892 rtt_us = msk->rcvq_space.rtt_us;
1893 if (rtt_us && time < (rtt_us >> 3))
1897 mptcp_for_each_subflow(msk, subflow) {
1898 const struct tcp_sock *tp;
1902 tp = tcp_sk(mptcp_subflow_tcp_sock(subflow));
1904 sf_rtt_us = READ_ONCE(tp->rcv_rtt_est.rtt_us);
1905 sf_advmss = READ_ONCE(tp->advmss);
1907 rtt_us = max(sf_rtt_us, rtt_us);
1908 advmss = max(sf_advmss, advmss);
1911 msk->rcvq_space.rtt_us = rtt_us;
1912 if (time < (rtt_us >> 3) || rtt_us == 0)
1915 if (msk->rcvq_space.copied <= msk->rcvq_space.space)
1918 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&
1919 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {
1923 rcvwin = ((u64)msk->rcvq_space.copied << 1) + 16 * advmss;
1925 grow = rcvwin * (msk->rcvq_space.copied - msk->rcvq_space.space);
1927 do_div(grow, msk->rcvq_space.space);
1928 rcvwin += (grow << 1);
1930 rcvmem = SKB_TRUESIZE(advmss + MAX_TCP_HEADER);
1931 while (tcp_win_from_space(sk, rcvmem) < advmss)
1934 do_div(rcvwin, advmss);
1935 rcvbuf = min_t(u64, rcvwin * rcvmem,
1936 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));
1938 if (rcvbuf > sk->sk_rcvbuf) {
1941 window_clamp = tcp_win_from_space(sk, rcvbuf);
1942 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);
1944 /* Make subflows follow along. If we do not do this, we
1945 * get drops at subflow level if skbs can't be moved to
1946 * the mptcp rx queue fast enough (announced rcv_win can
1947 * exceed ssk->sk_rcvbuf).
1949 mptcp_for_each_subflow(msk, subflow) {
1953 ssk = mptcp_subflow_tcp_sock(subflow);
1954 slow = lock_sock_fast(ssk);
1955 WRITE_ONCE(ssk->sk_rcvbuf, rcvbuf);
1956 tcp_sk(ssk)->window_clamp = window_clamp;
1957 tcp_cleanup_rbuf(ssk, 1);
1958 unlock_sock_fast(ssk, slow);
1963 msk->rcvq_space.space = msk->rcvq_space.copied;
1965 msk->rcvq_space.copied = 0;
1966 msk->rcvq_space.time = mstamp;
1969 static void __mptcp_update_rmem(struct sock *sk)
1971 struct mptcp_sock *msk = mptcp_sk(sk);
1973 if (!msk->rmem_released)
1976 atomic_sub(msk->rmem_released, &sk->sk_rmem_alloc);
1977 mptcp_rmem_uncharge(sk, msk->rmem_released);
1978 WRITE_ONCE(msk->rmem_released, 0);
1981 static void __mptcp_splice_receive_queue(struct sock *sk)
1983 struct mptcp_sock *msk = mptcp_sk(sk);
1985 skb_queue_splice_tail_init(&sk->sk_receive_queue, &msk->receive_queue);
1988 static bool __mptcp_move_skbs(struct mptcp_sock *msk)
1990 struct sock *sk = (struct sock *)msk;
1991 unsigned int moved = 0;
1995 struct sock *ssk = mptcp_subflow_recv_lookup(msk);
1998 /* we can have data pending in the subflows only if the msk
1999 * receive buffer was full at subflow_data_ready() time,
2000 * that is an unlikely slow path.
2005 slowpath = lock_sock_fast(ssk);
2006 mptcp_data_lock(sk);
2007 __mptcp_update_rmem(sk);
2008 done = __mptcp_move_skbs_from_subflow(msk, ssk, &moved);
2009 mptcp_data_unlock(sk);
2011 if (unlikely(ssk->sk_err))
2012 __mptcp_error_report(sk);
2013 unlock_sock_fast(ssk, slowpath);
2016 /* acquire the data lock only if some input data is pending */
2018 if (!RB_EMPTY_ROOT(&msk->out_of_order_queue) ||
2019 !skb_queue_empty_lockless(&sk->sk_receive_queue)) {
2020 mptcp_data_lock(sk);
2021 __mptcp_update_rmem(sk);
2022 ret |= __mptcp_ofo_queue(msk);
2023 __mptcp_splice_receive_queue(sk);
2024 mptcp_data_unlock(sk);
2027 mptcp_check_data_fin((struct sock *)msk);
2028 return !skb_queue_empty(&msk->receive_queue);
2031 static unsigned int mptcp_inq_hint(const struct sock *sk)
2033 const struct mptcp_sock *msk = mptcp_sk(sk);
2034 const struct sk_buff *skb;
2036 skb = skb_peek(&msk->receive_queue);
2038 u64 hint_val = msk->ack_seq - MPTCP_SKB_CB(skb)->map_seq;
2040 if (hint_val >= INT_MAX)
2043 return (unsigned int)hint_val;
2046 if (sk->sk_state == TCP_CLOSE || (sk->sk_shutdown & RCV_SHUTDOWN))
2052 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2053 int flags, int *addr_len)
2055 struct mptcp_sock *msk = mptcp_sk(sk);
2056 struct scm_timestamping_internal tss;
2057 int copied = 0, cmsg_flags = 0;
2061 /* MSG_ERRQUEUE is really a no-op till we support IP_RECVERR */
2062 if (unlikely(flags & MSG_ERRQUEUE))
2063 return inet_recv_error(sk, msg, len, addr_len);
2066 if (unlikely(sk->sk_state == TCP_LISTEN)) {
2071 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2073 len = min_t(size_t, len, INT_MAX);
2074 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
2076 if (unlikely(msk->recvmsg_inq))
2077 cmsg_flags = MPTCP_CMSG_INQ;
2079 while (copied < len) {
2082 bytes_read = __mptcp_recvmsg_mskq(msk, msg, len - copied, flags, &tss, &cmsg_flags);
2083 if (unlikely(bytes_read < 0)) {
2085 copied = bytes_read;
2089 copied += bytes_read;
2091 /* be sure to advertise window change */
2092 mptcp_cleanup_rbuf(msk);
2094 if (skb_queue_empty(&msk->receive_queue) && __mptcp_move_skbs(msk))
2097 /* only the master socket status is relevant here. The exit
2098 * conditions mirror closely tcp_recvmsg()
2100 if (copied >= target)
2105 sk->sk_state == TCP_CLOSE ||
2106 (sk->sk_shutdown & RCV_SHUTDOWN) ||
2108 signal_pending(current))
2112 copied = sock_error(sk);
2116 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2117 mptcp_check_for_eof(msk);
2119 if (sk->sk_shutdown & RCV_SHUTDOWN) {
2120 /* race breaker: the shutdown could be after the
2121 * previous receive queue check
2123 if (__mptcp_move_skbs(msk))
2128 if (sk->sk_state == TCP_CLOSE) {
2138 if (signal_pending(current)) {
2139 copied = sock_intr_errno(timeo);
2144 pr_debug("block timeout %ld", timeo);
2145 sk_wait_data(sk, &timeo, NULL);
2149 if (cmsg_flags && copied >= 0) {
2150 if (cmsg_flags & MPTCP_CMSG_TS)
2151 tcp_recv_timestamp(msg, sk, &tss);
2153 if (cmsg_flags & MPTCP_CMSG_INQ) {
2154 unsigned int inq = mptcp_inq_hint(sk);
2156 put_cmsg(msg, SOL_TCP, TCP_CM_INQ, sizeof(inq), &inq);
2160 pr_debug("msk=%p rx queue empty=%d:%d copied=%d",
2161 msk, skb_queue_empty_lockless(&sk->sk_receive_queue),
2162 skb_queue_empty(&msk->receive_queue), copied);
2163 if (!(flags & MSG_PEEK))
2164 mptcp_rcv_space_adjust(msk, copied);
2170 static void mptcp_retransmit_timer(struct timer_list *t)
2172 struct inet_connection_sock *icsk = from_timer(icsk, t,
2173 icsk_retransmit_timer);
2174 struct sock *sk = &icsk->icsk_inet.sk;
2175 struct mptcp_sock *msk = mptcp_sk(sk);
2178 if (!sock_owned_by_user(sk)) {
2179 /* we need a process context to retransmit */
2180 if (!test_and_set_bit(MPTCP_WORK_RTX, &msk->flags))
2181 mptcp_schedule_work(sk);
2183 /* delegate our work to tcp_release_cb() */
2184 __set_bit(MPTCP_RETRANSMIT, &msk->cb_flags);
2190 static void mptcp_timeout_timer(struct timer_list *t)
2192 struct sock *sk = from_timer(sk, t, sk_timer);
2194 mptcp_schedule_work(sk);
2198 /* Find an idle subflow. Return NULL if there is unacked data at tcp
2201 * A backup subflow is returned only if that is the only kind available.
2203 static struct sock *mptcp_subflow_get_retrans(struct mptcp_sock *msk)
2205 struct sock *backup = NULL, *pick = NULL;
2206 struct mptcp_subflow_context *subflow;
2207 int min_stale_count = INT_MAX;
2209 sock_owned_by_me((const struct sock *)msk);
2211 if (__mptcp_check_fallback(msk))
2214 mptcp_for_each_subflow(msk, subflow) {
2215 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2217 if (!__mptcp_subflow_active(subflow))
2220 /* still data outstanding at TCP level? skip this */
2221 if (!tcp_rtx_and_write_queues_empty(ssk)) {
2222 mptcp_pm_subflow_chk_stale(msk, ssk);
2223 min_stale_count = min_t(int, min_stale_count, subflow->stale_count);
2227 if (subflow->backup) {
2240 /* use backup only if there are no progresses anywhere */
2241 return min_stale_count > 1 ? backup : NULL;
2244 static void mptcp_dispose_initial_subflow(struct mptcp_sock *msk)
2247 iput(SOCK_INODE(msk->subflow));
2248 msk->subflow = NULL;
2252 bool __mptcp_retransmit_pending_data(struct sock *sk)
2254 struct mptcp_data_frag *cur, *rtx_head;
2255 struct mptcp_sock *msk = mptcp_sk(sk);
2257 if (__mptcp_check_fallback(mptcp_sk(sk)))
2260 if (tcp_rtx_and_write_queues_empty(sk))
2263 /* the closing socket has some data untransmitted and/or unacked:
2264 * some data in the mptcp rtx queue has not really xmitted yet.
2265 * keep it simple and re-inject the whole mptcp level rtx queue
2267 mptcp_data_lock(sk);
2268 __mptcp_clean_una_wakeup(sk);
2269 rtx_head = mptcp_rtx_head(sk);
2271 mptcp_data_unlock(sk);
2275 msk->recovery_snd_nxt = msk->snd_nxt;
2276 msk->recovery = true;
2277 mptcp_data_unlock(sk);
2279 msk->first_pending = rtx_head;
2282 /* be sure to clear the "sent status" on all re-injected fragments */
2283 list_for_each_entry(cur, &msk->rtx_queue, list) {
2284 if (!cur->already_sent)
2286 cur->already_sent = 0;
2292 /* flags for __mptcp_close_ssk() */
2293 #define MPTCP_CF_PUSH BIT(1)
2294 #define MPTCP_CF_FASTCLOSE BIT(2)
2296 /* subflow sockets can be either outgoing (connect) or incoming
2299 * Outgoing subflows use in-kernel sockets.
2300 * Incoming subflows do not have their own 'struct socket' allocated,
2301 * so we need to use tcp_close() after detaching them from the mptcp
2304 static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2305 struct mptcp_subflow_context *subflow,
2308 struct mptcp_sock *msk = mptcp_sk(sk);
2309 bool need_push, dispose_it;
2311 dispose_it = !msk->subflow || ssk != msk->subflow->sk;
2313 list_del(&subflow->node);
2315 lock_sock_nested(ssk, SINGLE_DEPTH_NESTING);
2317 if (flags & MPTCP_CF_FASTCLOSE)
2318 subflow->send_fastclose = 1;
2320 need_push = (flags & MPTCP_CF_PUSH) && __mptcp_retransmit_pending_data(sk);
2322 tcp_disconnect(ssk, 0);
2323 msk->subflow->state = SS_UNCONNECTED;
2324 mptcp_subflow_ctx_reset(subflow);
2330 /* if we are invoked by the msk cleanup code, the subflow is
2336 subflow->disposable = 1;
2338 /* if ssk hit tcp_done(), tcp_cleanup_ulp() cleared the related ops
2339 * the ssk has been already destroyed, we just need to release the
2340 * reference owned by msk;
2342 if (!inet_csk(ssk)->icsk_ulp_ops) {
2343 kfree_rcu(subflow, rcu);
2345 /* otherwise tcp will dispose of the ssk and subflow ctx */
2346 if (ssk->sk_state == TCP_LISTEN) {
2347 tcp_set_state(ssk, TCP_CLOSE);
2348 mptcp_subflow_queue_clean(ssk);
2349 inet_csk_listen_stop(ssk);
2351 __tcp_close(ssk, 0);
2353 /* close acquired an extra ref */
2360 if (ssk == msk->first)
2364 if (ssk == msk->last_snd)
2365 msk->last_snd = NULL;
2368 __mptcp_push_pending(sk, 0);
2371 void mptcp_close_ssk(struct sock *sk, struct sock *ssk,
2372 struct mptcp_subflow_context *subflow)
2374 if (sk->sk_state == TCP_ESTABLISHED)
2375 mptcp_event(MPTCP_EVENT_SUB_CLOSED, mptcp_sk(sk), ssk, GFP_KERNEL);
2377 /* subflow aborted before reaching the fully_established status
2378 * attempt the creation of the next subflow
2380 mptcp_pm_subflow_check_next(mptcp_sk(sk), ssk, subflow);
2382 __mptcp_close_ssk(sk, ssk, subflow, MPTCP_CF_PUSH);
2385 static unsigned int mptcp_sync_mss(struct sock *sk, u32 pmtu)
2390 static void __mptcp_close_subflow(struct mptcp_sock *msk)
2392 struct mptcp_subflow_context *subflow, *tmp;
2396 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2397 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2399 if (inet_sk_state_load(ssk) != TCP_CLOSE)
2402 /* 'subflow_data_ready' will re-sched once rx queue is empty */
2403 if (!skb_queue_empty_lockless(&ssk->sk_receive_queue))
2406 mptcp_close_ssk((struct sock *)msk, ssk, subflow);
2410 static bool mptcp_check_close_timeout(const struct sock *sk)
2412 s32 delta = tcp_jiffies32 - inet_csk(sk)->icsk_mtup.probe_timestamp;
2413 struct mptcp_subflow_context *subflow;
2415 if (delta >= TCP_TIMEWAIT_LEN)
2418 /* if all subflows are in closed status don't bother with additional
2421 mptcp_for_each_subflow(mptcp_sk(sk), subflow) {
2422 if (inet_sk_state_load(mptcp_subflow_tcp_sock(subflow)) !=
2429 static void mptcp_check_fastclose(struct mptcp_sock *msk)
2431 struct mptcp_subflow_context *subflow, *tmp;
2432 struct sock *sk = &msk->sk.icsk_inet.sk;
2434 if (likely(!READ_ONCE(msk->rcv_fastclose)))
2437 mptcp_token_destroy(msk);
2439 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
2440 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2443 slow = lock_sock_fast(tcp_sk);
2444 if (tcp_sk->sk_state != TCP_CLOSE) {
2445 tcp_send_active_reset(tcp_sk, GFP_ATOMIC);
2446 tcp_set_state(tcp_sk, TCP_CLOSE);
2448 unlock_sock_fast(tcp_sk, slow);
2451 inet_sk_state_store(sk, TCP_CLOSE);
2452 sk->sk_shutdown = SHUTDOWN_MASK;
2453 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
2454 set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags);
2456 mptcp_close_wake_up(sk);
2459 static void __mptcp_retrans(struct sock *sk)
2461 struct mptcp_sock *msk = mptcp_sk(sk);
2462 struct mptcp_sendmsg_info info = {};
2463 struct mptcp_data_frag *dfrag;
2468 mptcp_clean_una_wakeup(sk);
2470 /* first check ssk: need to kick "stale" logic */
2471 ssk = mptcp_subflow_get_retrans(msk);
2472 dfrag = mptcp_rtx_head(sk);
2474 if (mptcp_data_fin_enabled(msk)) {
2475 struct inet_connection_sock *icsk = inet_csk(sk);
2477 icsk->icsk_retransmits++;
2478 mptcp_set_datafin_timeout(sk);
2479 mptcp_send_ack(msk);
2484 if (!mptcp_send_head(sk))
2495 /* limit retransmission to the bytes already sent on some subflows */
2497 info.limit = READ_ONCE(msk->csum_enabled) ? dfrag->data_len : dfrag->already_sent;
2498 while (info.sent < info.limit) {
2499 ret = mptcp_sendmsg_frag(sk, ssk, dfrag, &info);
2503 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RETRANSSEGS);
2508 dfrag->already_sent = max(dfrag->already_sent, info.sent);
2509 tcp_push(ssk, 0, info.mss_now, tcp_sk(ssk)->nonagle,
2511 WRITE_ONCE(msk->allow_infinite_fallback, false);
2517 mptcp_check_and_set_pending(sk);
2519 if (!mptcp_timer_pending(sk))
2520 mptcp_reset_timer(sk);
2523 /* schedule the timeout timer for the relevant event: either close timeout
2524 * or mp_fail timeout. The close timeout takes precedence on the mp_fail one
2526 void mptcp_reset_timeout(struct mptcp_sock *msk, unsigned long fail_tout)
2528 struct sock *sk = (struct sock *)msk;
2529 unsigned long timeout, close_timeout;
2531 if (!fail_tout && !sock_flag(sk, SOCK_DEAD))
2534 close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies + TCP_TIMEWAIT_LEN;
2536 /* the close timeout takes precedence on the fail one, and here at least one of
2539 timeout = sock_flag(sk, SOCK_DEAD) ? close_timeout : fail_tout;
2541 sk_reset_timer(sk, &sk->sk_timer, timeout);
2544 static void mptcp_mp_fail_no_response(struct mptcp_sock *msk)
2546 struct sock *ssk = msk->first;
2552 pr_debug("MP_FAIL doesn't respond, reset the subflow");
2554 slow = lock_sock_fast(ssk);
2555 mptcp_subflow_reset(ssk);
2556 WRITE_ONCE(mptcp_subflow_ctx(ssk)->fail_tout, 0);
2557 unlock_sock_fast(ssk, slow);
2559 mptcp_reset_timeout(msk, 0);
2562 static void mptcp_worker(struct work_struct *work)
2564 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
2565 struct sock *sk = &msk->sk.icsk_inet.sk;
2566 unsigned long fail_tout;
2570 state = sk->sk_state;
2571 if (unlikely(state == TCP_CLOSE))
2574 mptcp_check_data_fin_ack(sk);
2576 mptcp_check_fastclose(msk);
2578 mptcp_pm_nl_work(msk);
2580 if (test_and_clear_bit(MPTCP_WORK_EOF, &msk->flags))
2581 mptcp_check_for_eof(msk);
2583 __mptcp_check_send_data_fin(sk);
2584 mptcp_check_data_fin(sk);
2586 /* There is no point in keeping around an orphaned sk timedout or
2587 * closed, but we need the msk around to reply to incoming DATA_FIN,
2588 * even if it is orphaned and in FIN_WAIT2 state
2590 if (sock_flag(sk, SOCK_DEAD) &&
2591 (mptcp_check_close_timeout(sk) || sk->sk_state == TCP_CLOSE)) {
2592 inet_sk_state_store(sk, TCP_CLOSE);
2593 __mptcp_destroy_sock(sk);
2597 if (test_and_clear_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
2598 __mptcp_close_subflow(msk);
2600 if (test_and_clear_bit(MPTCP_WORK_RTX, &msk->flags))
2601 __mptcp_retrans(sk);
2603 fail_tout = msk->first ? READ_ONCE(mptcp_subflow_ctx(msk->first)->fail_tout) : 0;
2604 if (fail_tout && time_after(jiffies, fail_tout))
2605 mptcp_mp_fail_no_response(msk);
2612 static int __mptcp_init_sock(struct sock *sk)
2614 struct mptcp_sock *msk = mptcp_sk(sk);
2616 INIT_LIST_HEAD(&msk->conn_list);
2617 INIT_LIST_HEAD(&msk->join_list);
2618 INIT_LIST_HEAD(&msk->rtx_queue);
2619 INIT_WORK(&msk->work, mptcp_worker);
2620 __skb_queue_head_init(&msk->receive_queue);
2621 msk->out_of_order_queue = RB_ROOT;
2622 msk->first_pending = NULL;
2623 msk->rmem_fwd_alloc = 0;
2624 WRITE_ONCE(msk->rmem_released, 0);
2625 msk->timer_ival = TCP_RTO_MIN;
2628 inet_csk(sk)->icsk_sync_mss = mptcp_sync_mss;
2629 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
2630 WRITE_ONCE(msk->allow_infinite_fallback, true);
2631 msk->recovery = false;
2633 mptcp_pm_data_init(msk);
2635 /* re-use the csk retrans timer for MPTCP-level retrans */
2636 timer_setup(&msk->sk.icsk_retransmit_timer, mptcp_retransmit_timer, 0);
2637 timer_setup(&sk->sk_timer, mptcp_timeout_timer, 0);
2642 static void mptcp_ca_reset(struct sock *sk)
2644 struct inet_connection_sock *icsk = inet_csk(sk);
2646 tcp_assign_congestion_control(sk);
2647 strcpy(mptcp_sk(sk)->ca_name, icsk->icsk_ca_ops->name);
2649 /* no need to keep a reference to the ops, the name will suffice */
2650 tcp_cleanup_congestion_control(sk);
2651 icsk->icsk_ca_ops = NULL;
2654 static int mptcp_init_sock(struct sock *sk)
2656 struct net *net = sock_net(sk);
2659 ret = __mptcp_init_sock(sk);
2663 if (!mptcp_is_enabled(net))
2664 return -ENOPROTOOPT;
2666 if (unlikely(!net->mib.mptcp_statistics) && !mptcp_mib_alloc(net))
2669 ret = __mptcp_socket_create(mptcp_sk(sk));
2673 /* fetch the ca name; do it outside __mptcp_init_sock(), so that clone will
2674 * propagate the correct value
2678 sk_sockets_allocated_inc(sk);
2679 sk->sk_rcvbuf = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[1]);
2680 sk->sk_sndbuf = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[1]);
2685 static void __mptcp_clear_xmit(struct sock *sk)
2687 struct mptcp_sock *msk = mptcp_sk(sk);
2688 struct mptcp_data_frag *dtmp, *dfrag;
2690 WRITE_ONCE(msk->first_pending, NULL);
2691 list_for_each_entry_safe(dfrag, dtmp, &msk->rtx_queue, list)
2692 dfrag_clear(sk, dfrag);
2695 static void mptcp_cancel_work(struct sock *sk)
2697 struct mptcp_sock *msk = mptcp_sk(sk);
2699 if (cancel_work_sync(&msk->work))
2703 void mptcp_subflow_shutdown(struct sock *sk, struct sock *ssk, int how)
2707 switch (ssk->sk_state) {
2709 if (!(how & RCV_SHUTDOWN))
2713 tcp_disconnect(ssk, O_NONBLOCK);
2716 if (__mptcp_check_fallback(mptcp_sk(sk))) {
2717 pr_debug("Fallback");
2718 ssk->sk_shutdown |= how;
2719 tcp_shutdown(ssk, how);
2721 pr_debug("Sending DATA_FIN on subflow %p", ssk);
2723 if (!mptcp_timer_pending(sk))
2724 mptcp_reset_timer(sk);
2732 static const unsigned char new_state[16] = {
2733 /* current state: new state: action: */
2734 [0 /* (Invalid) */] = TCP_CLOSE,
2735 [TCP_ESTABLISHED] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2736 [TCP_SYN_SENT] = TCP_CLOSE,
2737 [TCP_SYN_RECV] = TCP_FIN_WAIT1 | TCP_ACTION_FIN,
2738 [TCP_FIN_WAIT1] = TCP_FIN_WAIT1,
2739 [TCP_FIN_WAIT2] = TCP_FIN_WAIT2,
2740 [TCP_TIME_WAIT] = TCP_CLOSE, /* should not happen ! */
2741 [TCP_CLOSE] = TCP_CLOSE,
2742 [TCP_CLOSE_WAIT] = TCP_LAST_ACK | TCP_ACTION_FIN,
2743 [TCP_LAST_ACK] = TCP_LAST_ACK,
2744 [TCP_LISTEN] = TCP_CLOSE,
2745 [TCP_CLOSING] = TCP_CLOSING,
2746 [TCP_NEW_SYN_RECV] = TCP_CLOSE, /* should not happen ! */
2749 static int mptcp_close_state(struct sock *sk)
2751 int next = (int)new_state[sk->sk_state];
2752 int ns = next & TCP_STATE_MASK;
2754 inet_sk_state_store(sk, ns);
2756 return next & TCP_ACTION_FIN;
2759 static void __mptcp_check_send_data_fin(struct sock *sk)
2761 struct mptcp_subflow_context *subflow;
2762 struct mptcp_sock *msk = mptcp_sk(sk);
2764 pr_debug("msk=%p snd_data_fin_enable=%d pending=%d snd_nxt=%llu write_seq=%llu",
2765 msk, msk->snd_data_fin_enable, !!mptcp_send_head(sk),
2766 msk->snd_nxt, msk->write_seq);
2768 /* we still need to enqueue subflows or not really shutting down,
2771 if (!msk->snd_data_fin_enable || msk->snd_nxt + 1 != msk->write_seq ||
2772 mptcp_send_head(sk))
2775 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
2777 /* fallback socket will not get data_fin/ack, can move to the next
2780 if (__mptcp_check_fallback(msk)) {
2781 WRITE_ONCE(msk->snd_una, msk->write_seq);
2782 if ((1 << sk->sk_state) & (TCPF_CLOSING | TCPF_LAST_ACK)) {
2783 inet_sk_state_store(sk, TCP_CLOSE);
2784 mptcp_close_wake_up(sk);
2785 } else if (sk->sk_state == TCP_FIN_WAIT1) {
2786 inet_sk_state_store(sk, TCP_FIN_WAIT2);
2790 mptcp_for_each_subflow(msk, subflow) {
2791 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
2793 mptcp_subflow_shutdown(sk, tcp_sk, SEND_SHUTDOWN);
2797 static void __mptcp_wr_shutdown(struct sock *sk)
2799 struct mptcp_sock *msk = mptcp_sk(sk);
2801 pr_debug("msk=%p snd_data_fin_enable=%d shutdown=%x state=%d pending=%d",
2802 msk, msk->snd_data_fin_enable, sk->sk_shutdown, sk->sk_state,
2803 !!mptcp_send_head(sk));
2805 /* will be ignored by fallback sockets */
2806 WRITE_ONCE(msk->write_seq, msk->write_seq + 1);
2807 WRITE_ONCE(msk->snd_data_fin_enable, 1);
2809 __mptcp_check_send_data_fin(sk);
2812 static void __mptcp_destroy_sock(struct sock *sk)
2814 struct mptcp_sock *msk = mptcp_sk(sk);
2816 pr_debug("msk=%p", msk);
2820 mptcp_stop_timer(sk);
2821 sk_stop_timer(sk, &sk->sk_timer);
2824 sk->sk_prot->destroy(sk);
2826 WARN_ON_ONCE(msk->rmem_fwd_alloc);
2827 WARN_ON_ONCE(msk->rmem_released);
2828 sk_stream_kill_queues(sk);
2829 xfrm_sk_free_policy(sk);
2831 sk_refcnt_debug_release(sk);
2835 static void mptcp_close(struct sock *sk, long timeout)
2837 struct mptcp_subflow_context *subflow;
2838 struct mptcp_sock *msk = mptcp_sk(sk);
2839 bool do_cancel_work = false;
2842 sk->sk_shutdown = SHUTDOWN_MASK;
2844 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) {
2845 inet_sk_state_store(sk, TCP_CLOSE);
2849 if (mptcp_close_state(sk))
2850 __mptcp_wr_shutdown(sk);
2852 sk_stream_wait_close(sk, timeout);
2855 /* orphan all the subflows */
2856 inet_csk(sk)->icsk_mtup.probe_timestamp = tcp_jiffies32;
2857 mptcp_for_each_subflow(msk, subflow) {
2858 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
2859 bool slow = lock_sock_fast_nested(ssk);
2861 /* since the close timeout takes precedence on the fail one,
2864 if (ssk == msk->first)
2865 subflow->fail_tout = 0;
2868 unlock_sock_fast(ssk, slow);
2873 pr_debug("msk=%p state=%d", sk, sk->sk_state);
2874 if (mptcp_sk(sk)->token)
2875 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL);
2877 if (sk->sk_state == TCP_CLOSE) {
2878 __mptcp_destroy_sock(sk);
2879 do_cancel_work = true;
2881 mptcp_reset_timeout(msk, 0);
2885 mptcp_cancel_work(sk);
2890 static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
2892 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2893 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
2894 struct ipv6_pinfo *msk6 = inet6_sk(msk);
2896 msk->sk_v6_daddr = ssk->sk_v6_daddr;
2897 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
2900 msk6->saddr = ssk6->saddr;
2901 msk6->flow_label = ssk6->flow_label;
2905 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
2906 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
2907 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
2908 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
2909 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
2910 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
2913 static int mptcp_disconnect(struct sock *sk, int flags)
2915 struct mptcp_sock *msk = mptcp_sk(sk);
2917 inet_sk_state_store(sk, TCP_CLOSE);
2919 mptcp_stop_timer(sk);
2920 sk_stop_timer(sk, &sk->sk_timer);
2922 if (mptcp_sk(sk)->token)
2923 mptcp_event(MPTCP_EVENT_CLOSED, mptcp_sk(sk), NULL, GFP_KERNEL);
2925 /* msk->subflow is still intact, the following will not free the first
2928 mptcp_destroy_common(msk, MPTCP_CF_FASTCLOSE);
2929 msk->last_snd = NULL;
2930 WRITE_ONCE(msk->flags, 0);
2932 msk->push_pending = 0;
2933 msk->recovery = false;
2934 msk->can_ack = false;
2935 msk->fully_established = false;
2936 msk->rcv_data_fin = false;
2937 msk->snd_data_fin_enable = false;
2938 msk->rcv_fastclose = false;
2939 msk->use_64bit_ack = false;
2940 WRITE_ONCE(msk->csum_enabled, mptcp_is_checksum_enabled(sock_net(sk)));
2941 mptcp_pm_data_reset(msk);
2944 sk->sk_shutdown = 0;
2945 sk_error_report(sk);
2949 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2950 static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
2952 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
2954 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
2958 struct sock *mptcp_sk_clone(const struct sock *sk,
2959 const struct mptcp_options_received *mp_opt,
2960 struct request_sock *req)
2962 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
2963 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
2964 struct mptcp_sock *msk;
2970 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2971 if (nsk->sk_family == AF_INET6)
2972 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
2975 __mptcp_init_sock(nsk);
2977 msk = mptcp_sk(nsk);
2978 msk->local_key = subflow_req->local_key;
2979 msk->token = subflow_req->token;
2980 msk->subflow = NULL;
2981 WRITE_ONCE(msk->fully_established, false);
2982 if (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD)
2983 WRITE_ONCE(msk->csum_enabled, true);
2985 msk->write_seq = subflow_req->idsn + 1;
2986 msk->snd_nxt = msk->write_seq;
2987 msk->snd_una = msk->write_seq;
2988 msk->wnd_end = msk->snd_nxt + req->rsk_rcv_wnd;
2989 msk->setsockopt_seq = mptcp_sk(sk)->setsockopt_seq;
2991 if (mp_opt->suboptions & OPTIONS_MPTCP_MPC) {
2992 msk->can_ack = true;
2993 msk->remote_key = mp_opt->sndr_key;
2994 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
2996 WRITE_ONCE(msk->ack_seq, ack_seq);
2997 atomic64_set(&msk->rcv_wnd_sent, ack_seq);
3000 sock_reset_flag(nsk, SOCK_RCU_FREE);
3001 /* will be fully established after successful MPC subflow creation */
3002 inet_sk_state_store(nsk, TCP_SYN_RECV);
3004 security_inet_csk_clone(nsk, req);
3005 bh_unlock_sock(nsk);
3007 /* keep a single reference */
3012 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk)
3014 const struct tcp_sock *tp = tcp_sk(ssk);
3016 msk->rcvq_space.copied = 0;
3017 msk->rcvq_space.rtt_us = 0;
3019 msk->rcvq_space.time = tp->tcp_mstamp;
3021 /* initial rcv_space offering made to peer */
3022 msk->rcvq_space.space = min_t(u32, tp->rcv_wnd,
3023 TCP_INIT_CWND * tp->advmss);
3024 if (msk->rcvq_space.space == 0)
3025 msk->rcvq_space.space = TCP_INIT_CWND * TCP_MSS_DEFAULT;
3027 WRITE_ONCE(msk->wnd_end, msk->snd_nxt + tcp_sk(ssk)->snd_wnd);
3030 static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
3033 struct mptcp_sock *msk = mptcp_sk(sk);
3034 struct socket *listener;
3037 listener = __mptcp_nmpc_socket(msk);
3038 if (WARN_ON_ONCE(!listener)) {
3043 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
3044 newsk = inet_csk_accept(listener->sk, flags, err, kern);
3048 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
3049 if (sk_is_mptcp(newsk)) {
3050 struct mptcp_subflow_context *subflow;
3051 struct sock *new_mptcp_sock;
3053 subflow = mptcp_subflow_ctx(newsk);
3054 new_mptcp_sock = subflow->conn;
3056 /* is_mptcp should be false if subflow->conn is missing, see
3057 * subflow_syn_recv_sock()
3059 if (WARN_ON_ONCE(!new_mptcp_sock)) {
3060 tcp_sk(newsk)->is_mptcp = 0;
3064 /* acquire the 2nd reference for the owning socket */
3065 sock_hold(new_mptcp_sock);
3066 newsk = new_mptcp_sock;
3067 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEPASSIVEACK);
3069 MPTCP_INC_STATS(sock_net(sk),
3070 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
3074 newsk->sk_kern_sock = kern;
3078 void mptcp_destroy_common(struct mptcp_sock *msk, unsigned int flags)
3080 struct mptcp_subflow_context *subflow, *tmp;
3081 struct sock *sk = (struct sock *)msk;
3083 __mptcp_clear_xmit(sk);
3085 /* join list will be eventually flushed (with rst) at sock lock release time */
3086 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node)
3087 __mptcp_close_ssk(sk, mptcp_subflow_tcp_sock(subflow), subflow, flags);
3089 /* move to sk_receive_queue, sk_stream_kill_queues will purge it */
3090 mptcp_data_lock(sk);
3091 skb_queue_splice_tail_init(&msk->receive_queue, &sk->sk_receive_queue);
3092 __skb_queue_purge(&sk->sk_receive_queue);
3093 skb_rbtree_purge(&msk->out_of_order_queue);
3094 mptcp_data_unlock(sk);
3096 /* move all the rx fwd alloc into the sk_mem_reclaim_final in
3097 * inet_sock_destruct() will dispose it
3099 sk->sk_forward_alloc += msk->rmem_fwd_alloc;
3100 msk->rmem_fwd_alloc = 0;
3101 mptcp_token_destroy(msk);
3102 mptcp_pm_free_anno_list(msk);
3103 mptcp_free_local_addr_list(msk);
3106 static void mptcp_destroy(struct sock *sk)
3108 struct mptcp_sock *msk = mptcp_sk(sk);
3110 /* clears msk->subflow, allowing the following to close
3111 * even the initial subflow
3113 mptcp_dispose_initial_subflow(msk);
3114 mptcp_destroy_common(msk, 0);
3115 sk_sockets_allocated_dec(sk);
3118 void __mptcp_data_acked(struct sock *sk)
3120 if (!sock_owned_by_user(sk))
3121 __mptcp_clean_una(sk);
3123 __set_bit(MPTCP_CLEAN_UNA, &mptcp_sk(sk)->cb_flags);
3125 if (mptcp_pending_data_fin_ack(sk))
3126 mptcp_schedule_work(sk);
3129 void __mptcp_check_push(struct sock *sk, struct sock *ssk)
3131 if (!mptcp_send_head(sk))
3134 if (!sock_owned_by_user(sk)) {
3135 struct sock *xmit_ssk = mptcp_subflow_get_send(mptcp_sk(sk));
3137 if (xmit_ssk == ssk)
3138 __mptcp_subflow_push_pending(sk, ssk);
3140 mptcp_subflow_delegate(mptcp_subflow_ctx(xmit_ssk), MPTCP_DELEGATE_SEND);
3142 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3146 #define MPTCP_FLAGS_PROCESS_CTX_NEED (BIT(MPTCP_PUSH_PENDING) | \
3147 BIT(MPTCP_RETRANSMIT) | \
3148 BIT(MPTCP_FLUSH_JOIN_LIST))
3150 /* processes deferred events and flush wmem */
3151 static void mptcp_release_cb(struct sock *sk)
3152 __must_hold(&sk->sk_lock.slock)
3154 struct mptcp_sock *msk = mptcp_sk(sk);
3157 unsigned long flags = (msk->cb_flags & MPTCP_FLAGS_PROCESS_CTX_NEED) |
3162 /* the following actions acquire the subflow socket lock
3164 * 1) can't be invoked in atomic scope
3165 * 2) must avoid ABBA deadlock with msk socket spinlock: the RX
3166 * datapath acquires the msk socket spinlock while helding
3167 * the subflow socket lock
3169 msk->push_pending = 0;
3170 msk->cb_flags &= ~flags;
3171 spin_unlock_bh(&sk->sk_lock.slock);
3172 if (flags & BIT(MPTCP_FLUSH_JOIN_LIST))
3173 __mptcp_flush_join_list(sk);
3174 if (flags & BIT(MPTCP_PUSH_PENDING))
3175 __mptcp_push_pending(sk, 0);
3176 if (flags & BIT(MPTCP_RETRANSMIT))
3177 __mptcp_retrans(sk);
3180 spin_lock_bh(&sk->sk_lock.slock);
3183 if (__test_and_clear_bit(MPTCP_CLEAN_UNA, &msk->cb_flags))
3184 __mptcp_clean_una_wakeup(sk);
3185 if (unlikely(&msk->cb_flags)) {
3186 /* be sure to set the current sk state before tacking actions
3187 * depending on sk_state, that is processing MPTCP_ERROR_REPORT
3189 if (__test_and_clear_bit(MPTCP_CONNECTED, &msk->cb_flags))
3190 __mptcp_set_connected(sk);
3191 if (__test_and_clear_bit(MPTCP_ERROR_REPORT, &msk->cb_flags))
3192 __mptcp_error_report(sk);
3193 if (__test_and_clear_bit(MPTCP_RESET_SCHEDULER, &msk->cb_flags))
3194 msk->last_snd = NULL;
3197 __mptcp_update_rmem(sk);
3200 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
3201 * TCP can't schedule delack timer before the subflow is fully established.
3202 * MPTCP uses the delack timer to do 3rd ack retransmissions
3204 static void schedule_3rdack_retransmission(struct sock *ssk)
3206 struct inet_connection_sock *icsk = inet_csk(ssk);
3207 struct tcp_sock *tp = tcp_sk(ssk);
3208 unsigned long timeout;
3210 if (mptcp_subflow_ctx(ssk)->fully_established)
3213 /* reschedule with a timeout above RTT, as we must look only for drop */
3215 timeout = usecs_to_jiffies(tp->srtt_us >> (3 - 1));
3217 timeout = TCP_TIMEOUT_INIT;
3220 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
3221 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
3222 icsk->icsk_ack.timeout = timeout;
3223 sk_reset_timer(ssk, &icsk->icsk_delack_timer, timeout);
3226 void mptcp_subflow_process_delegated(struct sock *ssk)
3228 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3229 struct sock *sk = subflow->conn;
3231 if (test_bit(MPTCP_DELEGATE_SEND, &subflow->delegated_status)) {
3232 mptcp_data_lock(sk);
3233 if (!sock_owned_by_user(sk))
3234 __mptcp_subflow_push_pending(sk, ssk);
3236 __set_bit(MPTCP_PUSH_PENDING, &mptcp_sk(sk)->cb_flags);
3237 mptcp_data_unlock(sk);
3238 mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_SEND);
3240 if (test_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status)) {
3241 schedule_3rdack_retransmission(ssk);
3242 mptcp_subflow_delegated_done(subflow, MPTCP_DELEGATE_ACK);
3246 static int mptcp_hash(struct sock *sk)
3248 /* should never be called,
3249 * we hash the TCP subflows not the master socket
3255 static void mptcp_unhash(struct sock *sk)
3257 /* called from sk_common_release(), but nothing to do here */
3260 static int mptcp_get_port(struct sock *sk, unsigned short snum)
3262 struct mptcp_sock *msk = mptcp_sk(sk);
3263 struct socket *ssock;
3265 ssock = __mptcp_nmpc_socket(msk);
3266 pr_debug("msk=%p, subflow=%p", msk, ssock);
3267 if (WARN_ON_ONCE(!ssock))
3270 return inet_csk_get_port(ssock->sk, snum);
3273 void mptcp_finish_connect(struct sock *ssk)
3275 struct mptcp_subflow_context *subflow;
3276 struct mptcp_sock *msk;
3280 subflow = mptcp_subflow_ctx(ssk);
3284 pr_debug("msk=%p, token=%u", sk, subflow->token);
3286 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
3288 subflow->map_seq = ack_seq;
3289 subflow->map_subflow_seq = 1;
3291 /* the socket is not connected yet, no msk/subflow ops can access/race
3292 * accessing the field below
3294 WRITE_ONCE(msk->remote_key, subflow->remote_key);
3295 WRITE_ONCE(msk->local_key, subflow->local_key);
3296 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
3297 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
3298 WRITE_ONCE(msk->ack_seq, ack_seq);
3299 WRITE_ONCE(msk->can_ack, 1);
3300 WRITE_ONCE(msk->snd_una, msk->write_seq);
3301 atomic64_set(&msk->rcv_wnd_sent, ack_seq);
3303 mptcp_pm_new_connection(msk, ssk, 0);
3305 mptcp_rcv_space_init(msk, ssk);
3308 void mptcp_sock_graft(struct sock *sk, struct socket *parent)
3310 write_lock_bh(&sk->sk_callback_lock);
3311 rcu_assign_pointer(sk->sk_wq, &parent->wq);
3312 sk_set_socket(sk, parent);
3313 sk->sk_uid = SOCK_INODE(parent)->i_uid;
3314 write_unlock_bh(&sk->sk_callback_lock);
3317 bool mptcp_finish_join(struct sock *ssk)
3319 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
3320 struct mptcp_sock *msk = mptcp_sk(subflow->conn);
3321 struct sock *parent = (void *)msk;
3324 pr_debug("msk=%p, subflow=%p", msk, subflow);
3326 /* mptcp socket already closing? */
3327 if (!mptcp_is_fully_established(parent)) {
3328 subflow->reset_reason = MPTCP_RST_EMPTCP;
3332 if (!list_empty(&subflow->node))
3335 if (!mptcp_pm_allow_new_subflow(msk))
3336 goto err_prohibited;
3338 /* active connections are already on conn_list.
3339 * If we can't acquire msk socket lock here, let the release callback
3342 mptcp_data_lock(parent);
3343 if (!sock_owned_by_user(parent)) {
3344 ret = __mptcp_finish_join(msk, ssk);
3347 list_add_tail(&subflow->node, &msk->conn_list);
3351 list_add_tail(&subflow->node, &msk->join_list);
3352 __set_bit(MPTCP_FLUSH_JOIN_LIST, &msk->cb_flags);
3354 mptcp_data_unlock(parent);
3358 subflow->reset_reason = MPTCP_RST_EPROHIBIT;
3362 subflow->map_seq = READ_ONCE(msk->ack_seq);
3363 WRITE_ONCE(msk->allow_infinite_fallback, false);
3366 mptcp_event(MPTCP_EVENT_SUB_ESTABLISHED, msk, ssk, GFP_ATOMIC);
3370 static void mptcp_shutdown(struct sock *sk, int how)
3372 pr_debug("sk=%p, how=%d", sk, how);
3374 if ((how & SEND_SHUTDOWN) && mptcp_close_state(sk))
3375 __mptcp_wr_shutdown(sk);
3378 static int mptcp_forward_alloc_get(const struct sock *sk)
3380 return sk->sk_forward_alloc + mptcp_sk(sk)->rmem_fwd_alloc;
3383 static int mptcp_ioctl_outq(const struct mptcp_sock *msk, u64 v)
3385 const struct sock *sk = (void *)msk;
3388 if (sk->sk_state == TCP_LISTEN)
3391 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
3394 delta = msk->write_seq - v;
3395 if (__mptcp_check_fallback(msk) && msk->first) {
3396 struct tcp_sock *tp = tcp_sk(msk->first);
3398 /* the first subflow is disconnected after close - see
3399 * __mptcp_close_ssk(). tcp_disconnect() moves the write_seq
3400 * so ignore that status, too.
3402 if (!((1 << msk->first->sk_state) &
3403 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)))
3404 delta += READ_ONCE(tp->write_seq) - tp->snd_una;
3406 if (delta > INT_MAX)
3412 static int mptcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
3414 struct mptcp_sock *msk = mptcp_sk(sk);
3420 if (sk->sk_state == TCP_LISTEN)
3424 __mptcp_move_skbs(msk);
3425 answ = mptcp_inq_hint(sk);
3429 slow = lock_sock_fast(sk);
3430 answ = mptcp_ioctl_outq(msk, READ_ONCE(msk->snd_una));
3431 unlock_sock_fast(sk, slow);
3434 slow = lock_sock_fast(sk);
3435 answ = mptcp_ioctl_outq(msk, msk->snd_nxt);
3436 unlock_sock_fast(sk, slow);
3439 return -ENOIOCTLCMD;
3442 return put_user(answ, (int __user *)arg);
3445 static struct proto mptcp_prot = {
3447 .owner = THIS_MODULE,
3448 .init = mptcp_init_sock,
3449 .disconnect = mptcp_disconnect,
3450 .close = mptcp_close,
3451 .accept = mptcp_accept,
3452 .setsockopt = mptcp_setsockopt,
3453 .getsockopt = mptcp_getsockopt,
3454 .shutdown = mptcp_shutdown,
3455 .destroy = mptcp_destroy,
3456 .sendmsg = mptcp_sendmsg,
3457 .ioctl = mptcp_ioctl,
3458 .recvmsg = mptcp_recvmsg,
3459 .release_cb = mptcp_release_cb,
3461 .unhash = mptcp_unhash,
3462 .get_port = mptcp_get_port,
3463 .forward_alloc_get = mptcp_forward_alloc_get,
3464 .sockets_allocated = &mptcp_sockets_allocated,
3465 .memory_allocated = &tcp_memory_allocated,
3466 .memory_pressure = &tcp_memory_pressure,
3467 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
3468 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
3469 .sysctl_mem = sysctl_tcp_mem,
3470 .obj_size = sizeof(struct mptcp_sock),
3471 .slab_flags = SLAB_TYPESAFE_BY_RCU,
3472 .no_autobind = true,
3475 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3477 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3478 struct socket *ssock;
3481 lock_sock(sock->sk);
3482 ssock = __mptcp_nmpc_socket(msk);
3488 err = ssock->ops->bind(ssock, uaddr, addr_len);
3490 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3493 release_sock(sock->sk);
3497 static void mptcp_subflow_early_fallback(struct mptcp_sock *msk,
3498 struct mptcp_subflow_context *subflow)
3500 subflow->request_mptcp = 0;
3501 __mptcp_do_fallback(msk);
3504 static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
3505 int addr_len, int flags)
3507 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3508 struct mptcp_subflow_context *subflow;
3509 struct socket *ssock;
3512 lock_sock(sock->sk);
3514 if (addr_len < sizeof(uaddr->sa_family))
3517 if (uaddr->sa_family == AF_UNSPEC) {
3518 err = mptcp_disconnect(sock->sk, flags);
3519 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
3524 if (sock->state != SS_UNCONNECTED && msk->subflow) {
3525 /* pending connection or invalid state, let existing subflow
3528 ssock = msk->subflow;
3532 ssock = __mptcp_nmpc_socket(msk);
3536 mptcp_token_destroy(msk);
3537 inet_sk_state_store(sock->sk, TCP_SYN_SENT);
3538 subflow = mptcp_subflow_ctx(ssock->sk);
3539 #ifdef CONFIG_TCP_MD5SIG
3540 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
3543 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
3544 mptcp_subflow_early_fallback(msk, subflow);
3546 if (subflow->request_mptcp && mptcp_token_new_connect(ssock->sk)) {
3547 MPTCP_INC_STATS(sock_net(ssock->sk), MPTCP_MIB_TOKENFALLBACKINIT);
3548 mptcp_subflow_early_fallback(msk, subflow);
3550 if (likely(!__mptcp_check_fallback(msk)))
3551 MPTCP_INC_STATS(sock_net(sock->sk), MPTCP_MIB_MPCAPABLEACTIVE);
3554 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
3555 sock->state = ssock->state;
3557 /* on successful connect, the msk state will be moved to established by
3558 * subflow_finish_connect()
3560 if (!err || err == -EINPROGRESS)
3561 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3563 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3566 release_sock(sock->sk);
3570 static int mptcp_listen(struct socket *sock, int backlog)
3572 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3573 struct socket *ssock;
3576 pr_debug("msk=%p", msk);
3578 lock_sock(sock->sk);
3579 ssock = __mptcp_nmpc_socket(msk);
3585 mptcp_token_destroy(msk);
3586 inet_sk_state_store(sock->sk, TCP_LISTEN);
3587 sock_set_flag(sock->sk, SOCK_RCU_FREE);
3589 err = ssock->ops->listen(ssock, backlog);
3590 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
3592 mptcp_copy_inaddrs(sock->sk, ssock->sk);
3595 release_sock(sock->sk);
3599 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
3600 int flags, bool kern)
3602 struct mptcp_sock *msk = mptcp_sk(sock->sk);
3603 struct socket *ssock;
3606 pr_debug("msk=%p", msk);
3608 ssock = __mptcp_nmpc_socket(msk);
3612 err = ssock->ops->accept(sock, newsock, flags, kern);
3613 if (err == 0 && !mptcp_is_tcpsk(newsock->sk)) {
3614 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
3615 struct mptcp_subflow_context *subflow;
3616 struct sock *newsk = newsock->sk;
3620 /* PM/worker can now acquire the first subflow socket
3621 * lock without racing with listener queue cleanup,
3622 * we can notify it, if needed.
3624 * Even if remote has reset the initial subflow by now
3625 * the refcnt is still at least one.
3627 subflow = mptcp_subflow_ctx(msk->first);
3628 list_add(&subflow->node, &msk->conn_list);
3629 sock_hold(msk->first);
3630 if (mptcp_is_fully_established(newsk))
3631 mptcp_pm_fully_established(msk, msk->first, GFP_KERNEL);
3633 mptcp_copy_inaddrs(newsk, msk->first);
3634 mptcp_rcv_space_init(msk, msk->first);
3635 mptcp_propagate_sndbuf(newsk, msk->first);
3637 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
3638 * This is needed so NOSPACE flag can be set from tcp stack.
3640 mptcp_for_each_subflow(msk, subflow) {
3641 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3643 if (!ssk->sk_socket)
3644 mptcp_sock_graft(ssk, newsock);
3646 release_sock(newsk);
3652 static __poll_t mptcp_check_readable(struct mptcp_sock *msk)
3654 /* Concurrent splices from sk_receive_queue into receive_queue will
3655 * always show at least one non-empty queue when checked in this order.
3657 if (skb_queue_empty_lockless(&((struct sock *)msk)->sk_receive_queue) &&
3658 skb_queue_empty_lockless(&msk->receive_queue))
3661 return EPOLLIN | EPOLLRDNORM;
3664 static __poll_t mptcp_check_writeable(struct mptcp_sock *msk)
3666 struct sock *sk = (struct sock *)msk;
3668 if (unlikely(sk->sk_shutdown & SEND_SHUTDOWN))
3669 return EPOLLOUT | EPOLLWRNORM;
3671 if (sk_stream_is_writeable(sk))
3672 return EPOLLOUT | EPOLLWRNORM;
3674 mptcp_set_nospace(sk);
3675 smp_mb__after_atomic(); /* msk->flags is changed by write_space cb */
3676 if (sk_stream_is_writeable(sk))
3677 return EPOLLOUT | EPOLLWRNORM;
3682 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
3683 struct poll_table_struct *wait)
3685 struct sock *sk = sock->sk;
3686 struct mptcp_sock *msk;
3691 sock_poll_wait(file, sock, wait);
3693 state = inet_sk_state_load(sk);
3694 pr_debug("msk=%p state=%d flags=%lx", msk, state, msk->flags);
3695 if (state == TCP_LISTEN) {
3696 if (WARN_ON_ONCE(!msk->subflow || !msk->subflow->sk))
3699 return inet_csk_listen_poll(msk->subflow->sk);
3702 if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
3703 mask |= mptcp_check_readable(msk);
3704 mask |= mptcp_check_writeable(msk);
3706 if (sk->sk_shutdown == SHUTDOWN_MASK || state == TCP_CLOSE)
3708 if (sk->sk_shutdown & RCV_SHUTDOWN)
3709 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
3711 /* This barrier is coupled with smp_wmb() in tcp_reset() */
3719 static const struct proto_ops mptcp_stream_ops = {
3721 .owner = THIS_MODULE,
3722 .release = inet_release,
3724 .connect = mptcp_stream_connect,
3725 .socketpair = sock_no_socketpair,
3726 .accept = mptcp_stream_accept,
3727 .getname = inet_getname,
3729 .ioctl = inet_ioctl,
3730 .gettstamp = sock_gettstamp,
3731 .listen = mptcp_listen,
3732 .shutdown = inet_shutdown,
3733 .setsockopt = sock_common_setsockopt,
3734 .getsockopt = sock_common_getsockopt,
3735 .sendmsg = inet_sendmsg,
3736 .recvmsg = inet_recvmsg,
3737 .mmap = sock_no_mmap,
3738 .sendpage = inet_sendpage,
3741 static struct inet_protosw mptcp_protosw = {
3742 .type = SOCK_STREAM,
3743 .protocol = IPPROTO_MPTCP,
3744 .prot = &mptcp_prot,
3745 .ops = &mptcp_stream_ops,
3746 .flags = INET_PROTOSW_ICSK,
3749 static int mptcp_napi_poll(struct napi_struct *napi, int budget)
3751 struct mptcp_delegated_action *delegated;
3752 struct mptcp_subflow_context *subflow;
3755 delegated = container_of(napi, struct mptcp_delegated_action, napi);
3756 while ((subflow = mptcp_subflow_delegated_next(delegated)) != NULL) {
3757 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
3759 bh_lock_sock_nested(ssk);
3760 if (!sock_owned_by_user(ssk) &&
3761 mptcp_subflow_has_delegated_action(subflow))
3762 mptcp_subflow_process_delegated(ssk);
3763 /* ... elsewhere tcp_release_cb_override already processed
3764 * the action or will do at next release_sock().
3765 * In both case must dequeue the subflow here - on the same
3766 * CPU that scheduled it.
3768 bh_unlock_sock(ssk);
3771 if (++work_done == budget)
3775 /* always provide a 0 'work_done' argument, so that napi_complete_done
3776 * will not try accessing the NULL napi->dev ptr
3778 napi_complete_done(napi, 0);
3782 void __init mptcp_proto_init(void)
3784 struct mptcp_delegated_action *delegated;
3787 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
3789 if (percpu_counter_init(&mptcp_sockets_allocated, 0, GFP_KERNEL))
3790 panic("Failed to allocate MPTCP pcpu counter\n");
3792 init_dummy_netdev(&mptcp_napi_dev);
3793 for_each_possible_cpu(cpu) {
3794 delegated = per_cpu_ptr(&mptcp_delegated_actions, cpu);
3795 INIT_LIST_HEAD(&delegated->head);
3796 netif_napi_add_tx(&mptcp_napi_dev, &delegated->napi,
3798 napi_enable(&delegated->napi);
3801 mptcp_subflow_init();
3805 if (proto_register(&mptcp_prot, 1) != 0)
3806 panic("Failed to register MPTCP proto.\n");
3808 inet_register_protosw(&mptcp_protosw);
3810 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb) > sizeof_field(struct sk_buff, cb));
3813 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
3814 static const struct proto_ops mptcp_v6_stream_ops = {
3816 .owner = THIS_MODULE,
3817 .release = inet6_release,
3819 .connect = mptcp_stream_connect,
3820 .socketpair = sock_no_socketpair,
3821 .accept = mptcp_stream_accept,
3822 .getname = inet6_getname,
3824 .ioctl = inet6_ioctl,
3825 .gettstamp = sock_gettstamp,
3826 .listen = mptcp_listen,
3827 .shutdown = inet_shutdown,
3828 .setsockopt = sock_common_setsockopt,
3829 .getsockopt = sock_common_getsockopt,
3830 .sendmsg = inet6_sendmsg,
3831 .recvmsg = inet6_recvmsg,
3832 .mmap = sock_no_mmap,
3833 .sendpage = inet_sendpage,
3834 #ifdef CONFIG_COMPAT
3835 .compat_ioctl = inet6_compat_ioctl,
3839 static struct proto mptcp_v6_prot;
3841 static void mptcp_v6_destroy(struct sock *sk)
3844 inet6_destroy_sock(sk);
3847 static struct inet_protosw mptcp_v6_protosw = {
3848 .type = SOCK_STREAM,
3849 .protocol = IPPROTO_MPTCP,
3850 .prot = &mptcp_v6_prot,
3851 .ops = &mptcp_v6_stream_ops,
3852 .flags = INET_PROTOSW_ICSK,
3855 int __init mptcp_proto_v6_init(void)
3859 mptcp_v6_prot = mptcp_prot;
3860 strcpy(mptcp_v6_prot.name, "MPTCPv6");
3861 mptcp_v6_prot.slab = NULL;
3862 mptcp_v6_prot.destroy = mptcp_v6_destroy;
3863 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
3865 err = proto_register(&mptcp_v6_prot, 1);
3869 err = inet6_register_protosw(&mptcp_v6_protosw);
3871 proto_unregister(&mptcp_v6_prot);