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 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
20 #include <net/transp_v6.h>
22 #include <net/mptcp.h>
26 #define MPTCP_SAME_STATE TCP_MAX_STATES
28 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
30 struct mptcp_sock msk
;
39 #define MPTCP_SKB_CB(__skb) ((struct mptcp_skb_cb *)&((__skb)->cb[0]))
41 static struct percpu_counter mptcp_sockets_allocated
;
43 /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
44 * completed yet or has failed, return the subflow socket.
45 * Otherwise return NULL.
47 static struct socket
*__mptcp_nmpc_socket(const struct mptcp_sock
*msk
)
49 if (!msk
->subflow
|| READ_ONCE(msk
->can_ack
))
55 static bool __mptcp_needs_tcp_fallback(const struct mptcp_sock
*msk
)
57 return msk
->first
&& !sk_is_mptcp(msk
->first
);
60 static struct socket
*mptcp_is_tcpsk(struct sock
*sk
)
62 struct socket
*sock
= sk
->sk_socket
;
67 if (unlikely(sk
->sk_prot
== &tcp_prot
)) {
68 /* we are being invoked after mptcp_accept() has
69 * accepted a non-mp-capable flow: sk is a tcp_sk,
72 * Hand the socket over to tcp so all further socket ops
75 sock
->ops
= &inet_stream_ops
;
77 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
78 } else if (unlikely(sk
->sk_prot
== &tcpv6_prot
)) {
79 sock
->ops
= &inet6_stream_ops
;
87 static struct socket
*__mptcp_tcp_fallback(struct mptcp_sock
*msk
)
91 sock_owned_by_me((const struct sock
*)msk
);
93 sock
= mptcp_is_tcpsk((struct sock
*)msk
);
97 if (likely(!__mptcp_needs_tcp_fallback(msk
)))
103 static bool __mptcp_can_create_subflow(const struct mptcp_sock
*msk
)
108 static struct socket
*__mptcp_socket_create(struct mptcp_sock
*msk
, int state
)
110 struct mptcp_subflow_context
*subflow
;
111 struct sock
*sk
= (struct sock
*)msk
;
112 struct socket
*ssock
;
115 ssock
= __mptcp_tcp_fallback(msk
);
119 ssock
= __mptcp_nmpc_socket(msk
);
123 if (!__mptcp_can_create_subflow(msk
))
124 return ERR_PTR(-EINVAL
);
126 err
= mptcp_subflow_create_socket(sk
, &ssock
);
130 msk
->first
= ssock
->sk
;
131 msk
->subflow
= ssock
;
132 subflow
= mptcp_subflow_ctx(ssock
->sk
);
133 list_add(&subflow
->node
, &msk
->conn_list
);
134 subflow
->request_mptcp
= 1;
137 if (state
!= MPTCP_SAME_STATE
)
138 inet_sk_state_store(sk
, state
);
142 static void __mptcp_move_skb(struct mptcp_sock
*msk
, struct sock
*ssk
,
144 unsigned int offset
, size_t copy_len
)
146 struct sock
*sk
= (struct sock
*)msk
;
148 __skb_unlink(skb
, &ssk
->sk_receive_queue
);
149 skb_set_owner_r(skb
, sk
);
150 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
152 msk
->ack_seq
+= copy_len
;
153 MPTCP_SKB_CB(skb
)->offset
= offset
;
156 /* both sockets must be locked */
157 static bool mptcp_subflow_dsn_valid(const struct mptcp_sock
*msk
,
160 struct mptcp_subflow_context
*subflow
= mptcp_subflow_ctx(ssk
);
161 u64 dsn
= mptcp_subflow_get_mapped_dsn(subflow
);
163 /* revalidate data sequence number.
165 * mptcp_subflow_data_available() is usually called
166 * without msk lock. Its unlikely (but possible)
167 * that msk->ack_seq has been advanced since the last
168 * call found in-sequence data.
170 if (likely(dsn
== msk
->ack_seq
))
173 subflow
->data_avail
= 0;
174 return mptcp_subflow_data_available(ssk
);
177 static bool __mptcp_move_skbs_from_subflow(struct mptcp_sock
*msk
,
181 struct mptcp_subflow_context
*subflow
= mptcp_subflow_ctx(ssk
);
182 struct sock
*sk
= (struct sock
*)msk
;
183 unsigned int moved
= 0;
184 bool more_data_avail
;
188 if (!mptcp_subflow_dsn_valid(msk
, ssk
)) {
193 if (!(sk
->sk_userlocks
& SOCK_RCVBUF_LOCK
)) {
194 int rcvbuf
= max(ssk
->sk_rcvbuf
, sk
->sk_rcvbuf
);
196 if (rcvbuf
> sk
->sk_rcvbuf
)
197 sk
->sk_rcvbuf
= rcvbuf
;
202 u32 map_remaining
, offset
;
203 u32 seq
= tp
->copied_seq
;
207 /* try to move as much data as available */
208 map_remaining
= subflow
->map_data_len
-
209 mptcp_subflow_get_map_offset(subflow
);
211 skb
= skb_peek(&ssk
->sk_receive_queue
);
215 offset
= seq
- TCP_SKB_CB(skb
)->seq
;
216 fin
= TCP_SKB_CB(skb
)->tcp_flags
& TCPHDR_FIN
;
222 if (offset
< skb
->len
) {
223 size_t len
= skb
->len
- offset
;
228 __mptcp_move_skb(msk
, ssk
, skb
, offset
, len
);
232 if (WARN_ON_ONCE(map_remaining
< len
))
236 sk_eat_skb(ssk
, skb
);
240 WRITE_ONCE(tp
->copied_seq
, seq
);
241 more_data_avail
= mptcp_subflow_data_available(ssk
);
243 if (atomic_read(&sk
->sk_rmem_alloc
) > READ_ONCE(sk
->sk_rcvbuf
)) {
247 } while (more_data_avail
);
254 /* In most cases we will be able to lock the mptcp socket. If its already
255 * owned, we need to defer to the work queue to avoid ABBA deadlock.
257 static bool move_skbs_to_msk(struct mptcp_sock
*msk
, struct sock
*ssk
)
259 struct sock
*sk
= (struct sock
*)msk
;
260 unsigned int moved
= 0;
262 if (READ_ONCE(sk
->sk_lock
.owned
))
265 if (unlikely(!spin_trylock_bh(&sk
->sk_lock
.slock
)))
268 /* must re-check after taking the lock */
269 if (!READ_ONCE(sk
->sk_lock
.owned
))
270 __mptcp_move_skbs_from_subflow(msk
, ssk
, &moved
);
272 spin_unlock_bh(&sk
->sk_lock
.slock
);
277 void mptcp_data_ready(struct sock
*sk
, struct sock
*ssk
)
279 struct mptcp_sock
*msk
= mptcp_sk(sk
);
281 set_bit(MPTCP_DATA_READY
, &msk
->flags
);
283 if (atomic_read(&sk
->sk_rmem_alloc
) < READ_ONCE(sk
->sk_rcvbuf
) &&
284 move_skbs_to_msk(msk
, ssk
))
287 /* don't schedule if mptcp sk is (still) over limit */
288 if (atomic_read(&sk
->sk_rmem_alloc
) > READ_ONCE(sk
->sk_rcvbuf
))
291 /* mptcp socket is owned, release_cb should retry */
292 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED
,
293 &sk
->sk_tsq_flags
)) {
296 /* need to try again, its possible release_cb() has already
297 * been called after the test_and_set_bit() above.
299 move_skbs_to_msk(msk
, ssk
);
302 sk
->sk_data_ready(sk
);
305 static void __mptcp_flush_join_list(struct mptcp_sock
*msk
)
307 if (likely(list_empty(&msk
->join_list
)))
310 spin_lock_bh(&msk
->join_list_lock
);
311 list_splice_tail_init(&msk
->join_list
, &msk
->conn_list
);
312 spin_unlock_bh(&msk
->join_list_lock
);
315 static void mptcp_set_timeout(const struct sock
*sk
, const struct sock
*ssk
)
317 long tout
= ssk
&& inet_csk(ssk
)->icsk_pending
?
318 inet_csk(ssk
)->icsk_timeout
- jiffies
: 0;
321 tout
= mptcp_sk(sk
)->timer_ival
;
322 mptcp_sk(sk
)->timer_ival
= tout
> 0 ? tout
: TCP_RTO_MIN
;
325 static bool mptcp_timer_pending(struct sock
*sk
)
327 return timer_pending(&inet_csk(sk
)->icsk_retransmit_timer
);
330 static void mptcp_reset_timer(struct sock
*sk
)
332 struct inet_connection_sock
*icsk
= inet_csk(sk
);
335 /* should never be called with mptcp level timer cleared */
336 tout
= READ_ONCE(mptcp_sk(sk
)->timer_ival
);
337 if (WARN_ON_ONCE(!tout
))
339 sk_reset_timer(sk
, &icsk
->icsk_retransmit_timer
, jiffies
+ tout
);
342 void mptcp_data_acked(struct sock
*sk
)
344 mptcp_reset_timer(sk
);
346 if (!sk_stream_is_writeable(sk
) &&
347 schedule_work(&mptcp_sk(sk
)->work
))
351 void mptcp_subflow_eof(struct sock
*sk
)
353 struct mptcp_sock
*msk
= mptcp_sk(sk
);
355 if (!test_and_set_bit(MPTCP_WORK_EOF
, &msk
->flags
) &&
356 schedule_work(&msk
->work
))
360 static void mptcp_stop_timer(struct sock
*sk
)
362 struct inet_connection_sock
*icsk
= inet_csk(sk
);
364 sk_stop_timer(sk
, &icsk
->icsk_retransmit_timer
);
365 mptcp_sk(sk
)->timer_ival
= 0;
368 static bool mptcp_ext_cache_refill(struct mptcp_sock
*msk
)
370 if (!msk
->cached_ext
)
371 msk
->cached_ext
= __skb_ext_alloc();
373 return !!msk
->cached_ext
;
376 static struct sock
*mptcp_subflow_recv_lookup(const struct mptcp_sock
*msk
)
378 struct mptcp_subflow_context
*subflow
;
379 struct sock
*sk
= (struct sock
*)msk
;
381 sock_owned_by_me(sk
);
383 mptcp_for_each_subflow(msk
, subflow
) {
384 if (subflow
->data_avail
)
385 return mptcp_subflow_tcp_sock(subflow
);
391 static bool mptcp_skb_can_collapse_to(u64 write_seq
,
392 const struct sk_buff
*skb
,
393 const struct mptcp_ext
*mpext
)
395 if (!tcp_skb_can_collapse_to(skb
))
398 /* can collapse only if MPTCP level sequence is in order */
399 return mpext
&& mpext
->data_seq
+ mpext
->data_len
== write_seq
;
402 static bool mptcp_frag_can_collapse_to(const struct mptcp_sock
*msk
,
403 const struct page_frag
*pfrag
,
404 const struct mptcp_data_frag
*df
)
406 return df
&& pfrag
->page
== df
->page
&&
407 df
->data_seq
+ df
->data_len
== msk
->write_seq
;
410 static void dfrag_uncharge(struct sock
*sk
, int len
)
412 sk_mem_uncharge(sk
, len
);
413 sk_wmem_queued_add(sk
, -len
);
416 static void dfrag_clear(struct sock
*sk
, struct mptcp_data_frag
*dfrag
)
418 int len
= dfrag
->data_len
+ dfrag
->overhead
;
420 list_del(&dfrag
->list
);
421 dfrag_uncharge(sk
, len
);
422 put_page(dfrag
->page
);
425 static void mptcp_clean_una(struct sock
*sk
)
427 struct mptcp_sock
*msk
= mptcp_sk(sk
);
428 struct mptcp_data_frag
*dtmp
, *dfrag
;
429 u64 snd_una
= atomic64_read(&msk
->snd_una
);
430 bool cleaned
= false;
432 list_for_each_entry_safe(dfrag
, dtmp
, &msk
->rtx_queue
, list
) {
433 if (after64(dfrag
->data_seq
+ dfrag
->data_len
, snd_una
))
436 dfrag_clear(sk
, dfrag
);
440 dfrag
= mptcp_rtx_head(sk
);
441 if (dfrag
&& after64(snd_una
, dfrag
->data_seq
)) {
442 u64 delta
= dfrag
->data_seq
+ dfrag
->data_len
- snd_una
;
444 dfrag
->data_seq
+= delta
;
445 dfrag
->data_len
-= delta
;
447 dfrag_uncharge(sk
, delta
);
452 sk_mem_reclaim_partial(sk
);
454 /* Only wake up writers if a subflow is ready */
455 if (test_bit(MPTCP_SEND_SPACE
, &msk
->flags
))
456 sk_stream_write_space(sk
);
460 /* ensure we get enough memory for the frag hdr, beyond some minimal amount of
463 static bool mptcp_page_frag_refill(struct sock
*sk
, struct page_frag
*pfrag
)
465 if (likely(skb_page_frag_refill(32U + sizeof(struct mptcp_data_frag
),
466 pfrag
, sk
->sk_allocation
)))
469 sk
->sk_prot
->enter_memory_pressure(sk
);
470 sk_stream_moderate_sndbuf(sk
);
474 static struct mptcp_data_frag
*
475 mptcp_carve_data_frag(const struct mptcp_sock
*msk
, struct page_frag
*pfrag
,
478 int offset
= ALIGN(orig_offset
, sizeof(long));
479 struct mptcp_data_frag
*dfrag
;
481 dfrag
= (struct mptcp_data_frag
*)(page_to_virt(pfrag
->page
) + offset
);
483 dfrag
->data_seq
= msk
->write_seq
;
484 dfrag
->overhead
= offset
- orig_offset
+ sizeof(struct mptcp_data_frag
);
485 dfrag
->offset
= offset
+ sizeof(struct mptcp_data_frag
);
486 dfrag
->page
= pfrag
->page
;
491 static int mptcp_sendmsg_frag(struct sock
*sk
, struct sock
*ssk
,
492 struct msghdr
*msg
, struct mptcp_data_frag
*dfrag
,
493 long *timeo
, int *pmss_now
,
496 int mss_now
, avail_size
, size_goal
, offset
, ret
, frag_truesize
= 0;
497 bool dfrag_collapsed
, can_collapse
= false;
498 struct mptcp_sock
*msk
= mptcp_sk(sk
);
499 struct mptcp_ext
*mpext
= NULL
;
500 bool retransmission
= !!dfrag
;
501 struct sk_buff
*skb
, *tail
;
502 struct page_frag
*pfrag
;
507 /* use the mptcp page cache so that we can easily move the data
508 * from one substream to another, but do per subflow memory accounting
509 * Note: pfrag is used only !retransmission, but the compiler if
510 * fooled into a warning if we don't init here
512 pfrag
= sk_page_frag(sk
);
513 while ((!retransmission
&& !mptcp_page_frag_refill(ssk
, pfrag
)) ||
514 !mptcp_ext_cache_refill(msk
)) {
515 ret
= sk_stream_wait_memory(ssk
, timeo
);
519 /* if sk_stream_wait_memory() sleeps snd_una can change
520 * significantly, refresh the rtx queue
524 if (unlikely(__mptcp_needs_tcp_fallback(msk
)))
527 if (!retransmission
) {
528 write_seq
= &msk
->write_seq
;
531 write_seq
= &dfrag
->data_seq
;
535 /* compute copy limit */
536 mss_now
= tcp_send_mss(ssk
, &size_goal
, msg
->msg_flags
);
538 *ps_goal
= size_goal
;
539 avail_size
= size_goal
;
540 skb
= tcp_write_queue_tail(ssk
);
542 mpext
= skb_ext_find(skb
, SKB_EXT_MPTCP
);
544 /* Limit the write to the size available in the
545 * current skb, if any, so that we create at most a new skb.
546 * Explicitly tells TCP internals to avoid collapsing on later
547 * queue management operation, to avoid breaking the ext <->
548 * SSN association set here
550 can_collapse
= (size_goal
- skb
->len
> 0) &&
551 mptcp_skb_can_collapse_to(*write_seq
, skb
, mpext
);
553 TCP_SKB_CB(skb
)->eor
= 1;
555 avail_size
= size_goal
- skb
->len
;
558 if (!retransmission
) {
559 /* reuse tail pfrag, if possible, or carve a new one from the
562 dfrag
= mptcp_rtx_tail(sk
);
563 offset
= pfrag
->offset
;
564 dfrag_collapsed
= mptcp_frag_can_collapse_to(msk
, pfrag
, dfrag
);
565 if (!dfrag_collapsed
) {
566 dfrag
= mptcp_carve_data_frag(msk
, pfrag
, offset
);
567 offset
= dfrag
->offset
;
568 frag_truesize
= dfrag
->overhead
;
570 psize
= min_t(size_t, pfrag
->size
- offset
, avail_size
);
573 pr_debug("left=%zu", msg_data_left(msg
));
574 psize
= copy_page_from_iter(pfrag
->page
, offset
,
575 min_t(size_t, msg_data_left(msg
),
578 pr_debug("left=%zu", msg_data_left(msg
));
582 if (!sk_wmem_schedule(sk
, psize
+ dfrag
->overhead
))
585 offset
= dfrag
->offset
;
586 psize
= min_t(size_t, dfrag
->data_len
, avail_size
);
589 /* tell the TCP stack to delay the push so that we can safely
590 * access the skb after the sendpages call
592 ret
= do_tcp_sendpages(ssk
, page
, offset
, psize
,
593 msg
->msg_flags
| MSG_SENDPAGE_NOTLAST
);
597 frag_truesize
+= ret
;
598 if (!retransmission
) {
599 if (unlikely(ret
< psize
))
600 iov_iter_revert(&msg
->msg_iter
, psize
- ret
);
602 /* send successful, keep track of sent data for mptcp-level
605 dfrag
->data_len
+= ret
;
606 if (!dfrag_collapsed
) {
607 get_page(dfrag
->page
);
608 list_add_tail(&dfrag
->list
, &msk
->rtx_queue
);
609 sk_wmem_queued_add(sk
, frag_truesize
);
611 sk_wmem_queued_add(sk
, ret
);
614 /* charge data on mptcp rtx queue to the master socket
615 * Note: we charge such data both to sk and ssk
617 sk
->sk_forward_alloc
-= frag_truesize
;
620 /* if the tail skb extension is still the cached one, collapsing
621 * really happened. Note: we can't check for 'same skb' as the sk_buff
622 * hdr on tail can be transmitted, freed and re-allocated by the
623 * do_tcp_sendpages() call
625 tail
= tcp_write_queue_tail(ssk
);
626 if (mpext
&& tail
&& mpext
== skb_ext_find(tail
, SKB_EXT_MPTCP
)) {
627 WARN_ON_ONCE(!can_collapse
);
628 mpext
->data_len
+= ret
;
632 skb
= tcp_write_queue_tail(ssk
);
633 mpext
= __skb_ext_set(skb
, SKB_EXT_MPTCP
, msk
->cached_ext
);
634 msk
->cached_ext
= NULL
;
636 memset(mpext
, 0, sizeof(*mpext
));
637 mpext
->data_seq
= *write_seq
;
638 mpext
->subflow_seq
= mptcp_subflow_ctx(ssk
)->rel_write_seq
;
639 mpext
->data_len
= ret
;
643 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
644 mpext
->data_seq
, mpext
->subflow_seq
, mpext
->data_len
,
649 pfrag
->offset
+= frag_truesize
;
651 mptcp_subflow_ctx(ssk
)->rel_write_seq
+= ret
;
656 static struct sock
*mptcp_subflow_get_send(struct mptcp_sock
*msk
)
658 struct mptcp_subflow_context
*subflow
;
659 struct sock
*backup
= NULL
;
661 sock_owned_by_me((const struct sock
*)msk
);
663 mptcp_for_each_subflow(msk
, subflow
) {
664 struct sock
*ssk
= mptcp_subflow_tcp_sock(subflow
);
666 if (!sk_stream_memory_free(ssk
)) {
667 struct socket
*sock
= ssk
->sk_socket
;
670 clear_bit(MPTCP_SEND_SPACE
, &msk
->flags
);
671 smp_mb__after_atomic();
673 /* enables sk->write_space() callbacks */
674 set_bit(SOCK_NOSPACE
, &sock
->flags
);
680 if (subflow
->backup
) {
693 static void ssk_check_wmem(struct mptcp_sock
*msk
, struct sock
*ssk
)
697 if (likely(sk_stream_is_writeable(ssk
)))
700 sock
= READ_ONCE(ssk
->sk_socket
);
703 clear_bit(MPTCP_SEND_SPACE
, &msk
->flags
);
704 smp_mb__after_atomic();
705 /* set NOSPACE only after clearing SEND_SPACE flag */
706 set_bit(SOCK_NOSPACE
, &sock
->flags
);
710 static int mptcp_sendmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
)
712 int mss_now
= 0, size_goal
= 0, ret
= 0;
713 struct mptcp_sock
*msk
= mptcp_sk(sk
);
714 struct socket
*ssock
;
719 if (msg
->msg_flags
& ~(MSG_MORE
| MSG_DONTWAIT
| MSG_NOSIGNAL
))
724 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
726 if ((1 << sk
->sk_state
) & ~(TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
)) {
727 ret
= sk_stream_wait_connect(sk
, &timeo
);
733 ssock
= __mptcp_tcp_fallback(msk
);
734 if (unlikely(ssock
)) {
736 pr_debug("fallback passthrough");
737 ret
= sock_sendmsg(ssock
, msg
);
738 return ret
>= 0 ? ret
+ copied
: (copied
? copied
: ret
);
743 __mptcp_flush_join_list(msk
);
744 ssk
= mptcp_subflow_get_send(msk
);
745 while (!sk_stream_memory_free(sk
) || !ssk
) {
746 ret
= sk_stream_wait_memory(sk
, &timeo
);
752 ssk
= mptcp_subflow_get_send(msk
);
753 if (list_empty(&msk
->conn_list
)) {
759 pr_debug("conn_list->subflow=%p", ssk
);
762 while (msg_data_left(msg
)) {
763 ret
= mptcp_sendmsg_frag(sk
, ssk
, msg
, NULL
, &timeo
, &mss_now
,
767 if (ret
== 0 && unlikely(__mptcp_needs_tcp_fallback(msk
))) {
768 /* Can happen for passive sockets:
769 * 3WHS negotiated MPTCP, but first packet after is
770 * plain TCP (e.g. due to middlebox filtering unknown
782 mptcp_set_timeout(sk
, ssk
);
785 tcp_push(ssk
, msg
->msg_flags
, mss_now
, tcp_sk(ssk
)->nonagle
,
788 /* start the timer, if it's not pending */
789 if (!mptcp_timer_pending(sk
))
790 mptcp_reset_timer(sk
);
793 ssk_check_wmem(msk
, ssk
);
800 static void mptcp_wait_data(struct sock
*sk
, long *timeo
)
802 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
803 struct mptcp_sock
*msk
= mptcp_sk(sk
);
805 add_wait_queue(sk_sleep(sk
), &wait
);
806 sk_set_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
808 sk_wait_event(sk
, timeo
,
809 test_and_clear_bit(MPTCP_DATA_READY
, &msk
->flags
), &wait
);
811 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA
, sk
);
812 remove_wait_queue(sk_sleep(sk
), &wait
);
815 static int __mptcp_recvmsg_mskq(struct mptcp_sock
*msk
,
819 struct sock
*sk
= (struct sock
*)msk
;
823 while ((skb
= skb_peek(&sk
->sk_receive_queue
)) != NULL
) {
824 u32 offset
= MPTCP_SKB_CB(skb
)->offset
;
825 u32 data_len
= skb
->len
- offset
;
826 u32 count
= min_t(size_t, len
- copied
, data_len
);
829 err
= skb_copy_datagram_msg(skb
, offset
, msg
, count
);
830 if (unlikely(err
< 0)) {
838 if (count
< data_len
) {
839 MPTCP_SKB_CB(skb
)->offset
+= count
;
843 __skb_unlink(skb
, &sk
->sk_receive_queue
);
853 static bool __mptcp_move_skbs(struct mptcp_sock
*msk
)
855 unsigned int moved
= 0;
859 struct sock
*ssk
= mptcp_subflow_recv_lookup(msk
);
865 done
= __mptcp_move_skbs_from_subflow(msk
, ssk
, &moved
);
872 static int mptcp_recvmsg(struct sock
*sk
, struct msghdr
*msg
, size_t len
,
873 int nonblock
, int flags
, int *addr_len
)
875 struct mptcp_sock
*msk
= mptcp_sk(sk
);
876 struct socket
*ssock
;
881 if (msg
->msg_flags
& ~(MSG_WAITALL
| MSG_DONTWAIT
))
885 ssock
= __mptcp_tcp_fallback(msk
);
886 if (unlikely(ssock
)) {
889 pr_debug("fallback-read subflow=%p",
890 mptcp_subflow_ctx(ssock
->sk
));
891 copied
= sock_recvmsg(ssock
, msg
, flags
);
895 timeo
= sock_rcvtimeo(sk
, nonblock
);
897 len
= min_t(size_t, len
, INT_MAX
);
898 target
= sock_rcvlowat(sk
, flags
& MSG_WAITALL
, len
);
899 __mptcp_flush_join_list(msk
);
901 while (len
> (size_t)copied
) {
904 bytes_read
= __mptcp_recvmsg_mskq(msk
, msg
, len
- copied
);
905 if (unlikely(bytes_read
< 0)) {
911 copied
+= bytes_read
;
913 if (skb_queue_empty(&sk
->sk_receive_queue
) &&
914 __mptcp_move_skbs(msk
))
917 /* only the master socket status is relevant here. The exit
918 * conditions mirror closely tcp_recvmsg()
920 if (copied
>= target
)
925 sk
->sk_state
== TCP_CLOSE
||
926 (sk
->sk_shutdown
& RCV_SHUTDOWN
) ||
928 signal_pending(current
))
932 copied
= sock_error(sk
);
936 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
939 if (sk
->sk_state
== TCP_CLOSE
) {
949 if (signal_pending(current
)) {
950 copied
= sock_intr_errno(timeo
);
955 pr_debug("block timeout %ld", timeo
);
956 mptcp_wait_data(sk
, &timeo
);
957 if (unlikely(__mptcp_tcp_fallback(msk
)))
961 if (skb_queue_empty(&sk
->sk_receive_queue
)) {
962 /* entire backlog drained, clear DATA_READY. */
963 clear_bit(MPTCP_DATA_READY
, &msk
->flags
);
965 /* .. race-breaker: ssk might have gotten new data
966 * after last __mptcp_move_skbs() returned false.
968 if (unlikely(__mptcp_move_skbs(msk
)))
969 set_bit(MPTCP_DATA_READY
, &msk
->flags
);
970 } else if (unlikely(!test_bit(MPTCP_DATA_READY
, &msk
->flags
))) {
971 /* data to read but mptcp_wait_data() cleared DATA_READY */
972 set_bit(MPTCP_DATA_READY
, &msk
->flags
);
979 static void mptcp_retransmit_handler(struct sock
*sk
)
981 struct mptcp_sock
*msk
= mptcp_sk(sk
);
983 if (atomic64_read(&msk
->snd_una
) == msk
->write_seq
) {
984 mptcp_stop_timer(sk
);
986 set_bit(MPTCP_WORK_RTX
, &msk
->flags
);
987 if (schedule_work(&msk
->work
))
992 static void mptcp_retransmit_timer(struct timer_list
*t
)
994 struct inet_connection_sock
*icsk
= from_timer(icsk
, t
,
995 icsk_retransmit_timer
);
996 struct sock
*sk
= &icsk
->icsk_inet
.sk
;
999 if (!sock_owned_by_user(sk
)) {
1000 mptcp_retransmit_handler(sk
);
1002 /* delegate our work to tcp_release_cb() */
1003 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED
,
1011 /* Find an idle subflow. Return NULL if there is unacked data at tcp
1014 * A backup subflow is returned only if that is the only kind available.
1016 static struct sock
*mptcp_subflow_get_retrans(const struct mptcp_sock
*msk
)
1018 struct mptcp_subflow_context
*subflow
;
1019 struct sock
*backup
= NULL
;
1021 sock_owned_by_me((const struct sock
*)msk
);
1023 mptcp_for_each_subflow(msk
, subflow
) {
1024 struct sock
*ssk
= mptcp_subflow_tcp_sock(subflow
);
1026 /* still data outstanding at TCP level? Don't retransmit. */
1027 if (!tcp_write_queue_empty(ssk
))
1030 if (subflow
->backup
) {
1042 /* subflow sockets can be either outgoing (connect) or incoming
1045 * Outgoing subflows use in-kernel sockets.
1046 * Incoming subflows do not have their own 'struct socket' allocated,
1047 * so we need to use tcp_close() after detaching them from the mptcp
1050 static void __mptcp_close_ssk(struct sock
*sk
, struct sock
*ssk
,
1051 struct mptcp_subflow_context
*subflow
,
1054 struct socket
*sock
= READ_ONCE(ssk
->sk_socket
);
1056 list_del(&subflow
->node
);
1058 if (sock
&& sock
!= sk
->sk_socket
) {
1059 /* outgoing subflow */
1062 /* incoming subflow */
1063 tcp_close(ssk
, timeout
);
1067 static unsigned int mptcp_sync_mss(struct sock
*sk
, u32 pmtu
)
1072 static void mptcp_check_for_eof(struct mptcp_sock
*msk
)
1074 struct mptcp_subflow_context
*subflow
;
1075 struct sock
*sk
= (struct sock
*)msk
;
1078 mptcp_for_each_subflow(msk
, subflow
)
1079 receivers
+= !subflow
->rx_eof
;
1081 if (!receivers
&& !(sk
->sk_shutdown
& RCV_SHUTDOWN
)) {
1082 /* hopefully temporary hack: propagate shutdown status
1083 * to msk, when all subflows agree on it
1085 sk
->sk_shutdown
|= RCV_SHUTDOWN
;
1087 smp_mb__before_atomic(); /* SHUTDOWN must be visible first */
1088 set_bit(MPTCP_DATA_READY
, &msk
->flags
);
1089 sk
->sk_data_ready(sk
);
1093 static void mptcp_worker(struct work_struct
*work
)
1095 struct mptcp_sock
*msk
= container_of(work
, struct mptcp_sock
, work
);
1096 struct sock
*ssk
, *sk
= &msk
->sk
.icsk_inet
.sk
;
1097 int orig_len
, orig_offset
, ret
, mss_now
= 0, size_goal
= 0;
1098 struct mptcp_data_frag
*dfrag
;
1105 mptcp_clean_una(sk
);
1106 __mptcp_flush_join_list(msk
);
1107 __mptcp_move_skbs(msk
);
1109 if (test_and_clear_bit(MPTCP_WORK_EOF
, &msk
->flags
))
1110 mptcp_check_for_eof(msk
);
1112 if (!test_and_clear_bit(MPTCP_WORK_RTX
, &msk
->flags
))
1115 dfrag
= mptcp_rtx_head(sk
);
1119 ssk
= mptcp_subflow_get_retrans(msk
);
1125 msg
.msg_flags
= MSG_DONTWAIT
;
1126 orig_len
= dfrag
->data_len
;
1127 orig_offset
= dfrag
->offset
;
1128 orig_write_seq
= dfrag
->data_seq
;
1129 while (dfrag
->data_len
> 0) {
1130 ret
= mptcp_sendmsg_frag(sk
, ssk
, &msg
, dfrag
, &timeo
, &mss_now
,
1135 MPTCP_INC_STATS(sock_net(sk
), MPTCP_MIB_RETRANSSEGS
);
1137 dfrag
->data_len
-= ret
;
1138 dfrag
->offset
+= ret
;
1141 tcp_push(ssk
, msg
.msg_flags
, mss_now
, tcp_sk(ssk
)->nonagle
,
1144 dfrag
->data_seq
= orig_write_seq
;
1145 dfrag
->offset
= orig_offset
;
1146 dfrag
->data_len
= orig_len
;
1148 mptcp_set_timeout(sk
, ssk
);
1152 if (!mptcp_timer_pending(sk
))
1153 mptcp_reset_timer(sk
);
1160 static int __mptcp_init_sock(struct sock
*sk
)
1162 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1164 spin_lock_init(&msk
->join_list_lock
);
1166 INIT_LIST_HEAD(&msk
->conn_list
);
1167 INIT_LIST_HEAD(&msk
->join_list
);
1168 INIT_LIST_HEAD(&msk
->rtx_queue
);
1169 __set_bit(MPTCP_SEND_SPACE
, &msk
->flags
);
1170 INIT_WORK(&msk
->work
, mptcp_worker
);
1173 inet_csk(sk
)->icsk_sync_mss
= mptcp_sync_mss
;
1175 mptcp_pm_data_init(msk
);
1177 /* re-use the csk retrans timer for MPTCP-level retrans */
1178 timer_setup(&msk
->sk
.icsk_retransmit_timer
, mptcp_retransmit_timer
, 0);
1183 static int mptcp_init_sock(struct sock
*sk
)
1185 struct net
*net
= sock_net(sk
);
1188 if (!mptcp_is_enabled(net
))
1189 return -ENOPROTOOPT
;
1191 if (unlikely(!net
->mib
.mptcp_statistics
) && !mptcp_mib_alloc(net
))
1194 ret
= __mptcp_init_sock(sk
);
1198 sk_sockets_allocated_inc(sk
);
1199 sk
->sk_sndbuf
= sock_net(sk
)->ipv4
.sysctl_tcp_wmem
[2];
1204 static void __mptcp_clear_xmit(struct sock
*sk
)
1206 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1207 struct mptcp_data_frag
*dtmp
, *dfrag
;
1209 sk_stop_timer(sk
, &msk
->sk
.icsk_retransmit_timer
);
1211 list_for_each_entry_safe(dfrag
, dtmp
, &msk
->rtx_queue
, list
)
1212 dfrag_clear(sk
, dfrag
);
1215 static void mptcp_cancel_work(struct sock
*sk
)
1217 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1219 if (cancel_work_sync(&msk
->work
))
1223 static void mptcp_subflow_shutdown(struct sock
*ssk
, int how
,
1224 bool data_fin_tx_enable
, u64 data_fin_tx_seq
)
1228 switch (ssk
->sk_state
) {
1230 if (!(how
& RCV_SHUTDOWN
))
1234 tcp_disconnect(ssk
, O_NONBLOCK
);
1237 if (data_fin_tx_enable
) {
1238 struct mptcp_subflow_context
*subflow
;
1240 subflow
= mptcp_subflow_ctx(ssk
);
1241 subflow
->data_fin_tx_seq
= data_fin_tx_seq
;
1242 subflow
->data_fin_tx_enable
= 1;
1245 ssk
->sk_shutdown
|= how
;
1246 tcp_shutdown(ssk
, how
);
1250 /* Wake up anyone sleeping in poll. */
1251 ssk
->sk_state_change(ssk
);
1255 /* Called with msk lock held, releases such lock before returning */
1256 static void mptcp_close(struct sock
*sk
, long timeout
)
1258 struct mptcp_subflow_context
*subflow
, *tmp
;
1259 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1260 LIST_HEAD(conn_list
);
1261 u64 data_fin_tx_seq
;
1265 mptcp_token_destroy(msk
->token
);
1266 inet_sk_state_store(sk
, TCP_CLOSE
);
1268 __mptcp_flush_join_list(msk
);
1270 list_splice_init(&msk
->conn_list
, &conn_list
);
1272 data_fin_tx_seq
= msk
->write_seq
;
1274 __mptcp_clear_xmit(sk
);
1278 list_for_each_entry_safe(subflow
, tmp
, &conn_list
, node
) {
1279 struct sock
*ssk
= mptcp_subflow_tcp_sock(subflow
);
1281 subflow
->data_fin_tx_seq
= data_fin_tx_seq
;
1282 subflow
->data_fin_tx_enable
= 1;
1283 __mptcp_close_ssk(sk
, ssk
, subflow
, timeout
);
1286 mptcp_cancel_work(sk
);
1287 mptcp_pm_close(msk
);
1289 __skb_queue_purge(&sk
->sk_receive_queue
);
1291 sk_common_release(sk
);
1294 static void mptcp_copy_inaddrs(struct sock
*msk
, const struct sock
*ssk
)
1296 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1297 const struct ipv6_pinfo
*ssk6
= inet6_sk(ssk
);
1298 struct ipv6_pinfo
*msk6
= inet6_sk(msk
);
1300 msk
->sk_v6_daddr
= ssk
->sk_v6_daddr
;
1301 msk
->sk_v6_rcv_saddr
= ssk
->sk_v6_rcv_saddr
;
1304 msk6
->saddr
= ssk6
->saddr
;
1305 msk6
->flow_label
= ssk6
->flow_label
;
1309 inet_sk(msk
)->inet_num
= inet_sk(ssk
)->inet_num
;
1310 inet_sk(msk
)->inet_dport
= inet_sk(ssk
)->inet_dport
;
1311 inet_sk(msk
)->inet_sport
= inet_sk(ssk
)->inet_sport
;
1312 inet_sk(msk
)->inet_daddr
= inet_sk(ssk
)->inet_daddr
;
1313 inet_sk(msk
)->inet_saddr
= inet_sk(ssk
)->inet_saddr
;
1314 inet_sk(msk
)->inet_rcv_saddr
= inet_sk(ssk
)->inet_rcv_saddr
;
1317 static int mptcp_disconnect(struct sock
*sk
, int flags
)
1319 /* Should never be called.
1320 * inet_stream_connect() calls ->disconnect, but that
1321 * refers to the subflow socket, not the mptcp one.
1327 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1328 static struct ipv6_pinfo
*mptcp_inet6_sk(const struct sock
*sk
)
1330 unsigned int offset
= sizeof(struct mptcp6_sock
) - sizeof(struct ipv6_pinfo
);
1332 return (struct ipv6_pinfo
*)(((u8
*)sk
) + offset
);
1336 struct sock
*mptcp_sk_clone(const struct sock
*sk
,
1337 const struct mptcp_options_received
*mp_opt
,
1338 struct request_sock
*req
)
1340 struct mptcp_subflow_request_sock
*subflow_req
= mptcp_subflow_rsk(req
);
1341 struct sock
*nsk
= sk_clone_lock(sk
, GFP_ATOMIC
);
1342 struct mptcp_sock
*msk
;
1348 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1349 if (nsk
->sk_family
== AF_INET6
)
1350 inet_sk(nsk
)->pinet6
= mptcp_inet6_sk(nsk
);
1353 __mptcp_init_sock(nsk
);
1355 msk
= mptcp_sk(nsk
);
1356 msk
->local_key
= subflow_req
->local_key
;
1357 msk
->token
= subflow_req
->token
;
1358 msk
->subflow
= NULL
;
1360 if (unlikely(mptcp_token_new_accept(subflow_req
->token
, nsk
))) {
1361 nsk
->sk_state
= TCP_CLOSE
;
1362 bh_unlock_sock(nsk
);
1364 /* we can't call into mptcp_close() here - possible BH context
1365 * free the sock directly.
1366 * sk_clone_lock() sets nsk refcnt to two, hence call sk_free()
1369 sk_common_release(nsk
);
1374 msk
->write_seq
= subflow_req
->idsn
+ 1;
1375 atomic64_set(&msk
->snd_una
, msk
->write_seq
);
1376 if (mp_opt
->mp_capable
) {
1377 msk
->can_ack
= true;
1378 msk
->remote_key
= mp_opt
->sndr_key
;
1379 mptcp_crypto_key_sha(msk
->remote_key
, NULL
, &ack_seq
);
1381 msk
->ack_seq
= ack_seq
;
1384 sock_reset_flag(nsk
, SOCK_RCU_FREE
);
1385 /* will be fully established after successful MPC subflow creation */
1386 inet_sk_state_store(nsk
, TCP_SYN_RECV
);
1387 bh_unlock_sock(nsk
);
1389 /* keep a single reference */
1394 static struct sock
*mptcp_accept(struct sock
*sk
, int flags
, int *err
,
1397 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1398 struct socket
*listener
;
1401 listener
= __mptcp_nmpc_socket(msk
);
1402 if (WARN_ON_ONCE(!listener
)) {
1407 pr_debug("msk=%p, listener=%p", msk
, mptcp_subflow_ctx(listener
->sk
));
1408 newsk
= inet_csk_accept(listener
->sk
, flags
, err
, kern
);
1412 pr_debug("msk=%p, subflow is mptcp=%d", msk
, sk_is_mptcp(newsk
));
1414 if (sk_is_mptcp(newsk
)) {
1415 struct mptcp_subflow_context
*subflow
;
1416 struct sock
*new_mptcp_sock
;
1417 struct sock
*ssk
= newsk
;
1419 subflow
= mptcp_subflow_ctx(newsk
);
1420 new_mptcp_sock
= subflow
->conn
;
1422 /* is_mptcp should be false if subflow->conn is missing, see
1423 * subflow_syn_recv_sock()
1425 if (WARN_ON_ONCE(!new_mptcp_sock
)) {
1426 tcp_sk(newsk
)->is_mptcp
= 0;
1430 /* acquire the 2nd reference for the owning socket */
1431 sock_hold(new_mptcp_sock
);
1434 bh_lock_sock(new_mptcp_sock
);
1435 msk
= mptcp_sk(new_mptcp_sock
);
1438 newsk
= new_mptcp_sock
;
1439 mptcp_copy_inaddrs(newsk
, ssk
);
1440 list_add(&subflow
->node
, &msk
->conn_list
);
1441 inet_sk_state_store(newsk
, TCP_ESTABLISHED
);
1443 bh_unlock_sock(new_mptcp_sock
);
1445 __MPTCP_INC_STATS(sock_net(sk
), MPTCP_MIB_MPCAPABLEPASSIVEACK
);
1448 MPTCP_INC_STATS(sock_net(sk
),
1449 MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK
);
1455 static void mptcp_destroy(struct sock
*sk
)
1457 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1459 if (msk
->cached_ext
)
1460 __skb_ext_put(msk
->cached_ext
);
1462 sk_sockets_allocated_dec(sk
);
1465 static int mptcp_setsockopt(struct sock
*sk
, int level
, int optname
,
1466 char __user
*optval
, unsigned int optlen
)
1468 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1469 struct socket
*ssock
;
1471 pr_debug("msk=%p", msk
);
1473 /* @@ the meaning of setsockopt() when the socket is connected and
1474 * there are multiple subflows is not yet defined. It is up to the
1475 * MPTCP-level socket to configure the subflows until the subflow
1476 * is in TCP fallback, when TCP socket options are passed through
1477 * to the one remaining subflow.
1480 ssock
= __mptcp_tcp_fallback(msk
);
1483 return tcp_setsockopt(ssock
->sk
, level
, optname
, optval
,
1489 static int mptcp_getsockopt(struct sock
*sk
, int level
, int optname
,
1490 char __user
*optval
, int __user
*option
)
1492 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1493 struct socket
*ssock
;
1495 pr_debug("msk=%p", msk
);
1497 /* @@ the meaning of setsockopt() when the socket is connected and
1498 * there are multiple subflows is not yet defined. It is up to the
1499 * MPTCP-level socket to configure the subflows until the subflow
1500 * is in TCP fallback, when socket options are passed through
1501 * to the one remaining subflow.
1504 ssock
= __mptcp_tcp_fallback(msk
);
1507 return tcp_getsockopt(ssock
->sk
, level
, optname
, optval
,
1513 #define MPTCP_DEFERRED_ALL (TCPF_DELACK_TIMER_DEFERRED | \
1514 TCPF_WRITE_TIMER_DEFERRED)
1516 /* this is very alike tcp_release_cb() but we must handle differently a
1517 * different set of events
1519 static void mptcp_release_cb(struct sock
*sk
)
1521 unsigned long flags
, nflags
;
1524 flags
= sk
->sk_tsq_flags
;
1525 if (!(flags
& MPTCP_DEFERRED_ALL
))
1527 nflags
= flags
& ~MPTCP_DEFERRED_ALL
;
1528 } while (cmpxchg(&sk
->sk_tsq_flags
, flags
, nflags
) != flags
);
1530 sock_release_ownership(sk
);
1532 if (flags
& TCPF_DELACK_TIMER_DEFERRED
) {
1533 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1536 ssk
= mptcp_subflow_recv_lookup(msk
);
1537 if (!ssk
|| !schedule_work(&msk
->work
))
1541 if (flags
& TCPF_WRITE_TIMER_DEFERRED
) {
1542 mptcp_retransmit_handler(sk
);
1547 static int mptcp_get_port(struct sock
*sk
, unsigned short snum
)
1549 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1550 struct socket
*ssock
;
1552 ssock
= __mptcp_nmpc_socket(msk
);
1553 pr_debug("msk=%p, subflow=%p", msk
, ssock
);
1554 if (WARN_ON_ONCE(!ssock
))
1557 return inet_csk_get_port(ssock
->sk
, snum
);
1560 void mptcp_finish_connect(struct sock
*ssk
)
1562 struct mptcp_subflow_context
*subflow
;
1563 struct mptcp_sock
*msk
;
1567 subflow
= mptcp_subflow_ctx(ssk
);
1571 if (!subflow
->mp_capable
) {
1572 MPTCP_INC_STATS(sock_net(sk
),
1573 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK
);
1577 pr_debug("msk=%p, token=%u", sk
, subflow
->token
);
1579 mptcp_crypto_key_sha(subflow
->remote_key
, NULL
, &ack_seq
);
1581 subflow
->map_seq
= ack_seq
;
1582 subflow
->map_subflow_seq
= 1;
1583 subflow
->rel_write_seq
= 1;
1585 /* the socket is not connected yet, no msk/subflow ops can access/race
1586 * accessing the field below
1588 WRITE_ONCE(msk
->remote_key
, subflow
->remote_key
);
1589 WRITE_ONCE(msk
->local_key
, subflow
->local_key
);
1590 WRITE_ONCE(msk
->token
, subflow
->token
);
1591 WRITE_ONCE(msk
->write_seq
, subflow
->idsn
+ 1);
1592 WRITE_ONCE(msk
->ack_seq
, ack_seq
);
1593 WRITE_ONCE(msk
->can_ack
, 1);
1594 atomic64_set(&msk
->snd_una
, msk
->write_seq
);
1596 mptcp_pm_new_connection(msk
, 0);
1599 static void mptcp_sock_graft(struct sock
*sk
, struct socket
*parent
)
1601 write_lock_bh(&sk
->sk_callback_lock
);
1602 rcu_assign_pointer(sk
->sk_wq
, &parent
->wq
);
1603 sk_set_socket(sk
, parent
);
1604 sk
->sk_uid
= SOCK_INODE(parent
)->i_uid
;
1605 write_unlock_bh(&sk
->sk_callback_lock
);
1608 bool mptcp_finish_join(struct sock
*sk
)
1610 struct mptcp_subflow_context
*subflow
= mptcp_subflow_ctx(sk
);
1611 struct mptcp_sock
*msk
= mptcp_sk(subflow
->conn
);
1612 struct sock
*parent
= (void *)msk
;
1613 struct socket
*parent_sock
;
1616 pr_debug("msk=%p, subflow=%p", msk
, subflow
);
1618 /* mptcp socket already closing? */
1619 if (inet_sk_state_load(parent
) != TCP_ESTABLISHED
)
1622 if (!msk
->pm
.server_side
)
1625 /* passive connection, attach to msk socket */
1626 parent_sock
= READ_ONCE(parent
->sk_socket
);
1627 if (parent_sock
&& !sk
->sk_socket
)
1628 mptcp_sock_graft(sk
, parent_sock
);
1630 ret
= mptcp_pm_allow_new_subflow(msk
);
1632 /* active connections are already on conn_list */
1633 spin_lock_bh(&msk
->join_list_lock
);
1634 if (!WARN_ON_ONCE(!list_empty(&subflow
->node
)))
1635 list_add_tail(&subflow
->node
, &msk
->join_list
);
1636 spin_unlock_bh(&msk
->join_list_lock
);
1641 bool mptcp_sk_is_subflow(const struct sock
*sk
)
1643 struct mptcp_subflow_context
*subflow
= mptcp_subflow_ctx(sk
);
1645 return subflow
->mp_join
== 1;
1648 static bool mptcp_memory_free(const struct sock
*sk
, int wake
)
1650 struct mptcp_sock
*msk
= mptcp_sk(sk
);
1652 return wake
? test_bit(MPTCP_SEND_SPACE
, &msk
->flags
) : true;
1655 static struct proto mptcp_prot
= {
1657 .owner
= THIS_MODULE
,
1658 .init
= mptcp_init_sock
,
1659 .disconnect
= mptcp_disconnect
,
1660 .close
= mptcp_close
,
1661 .accept
= mptcp_accept
,
1662 .setsockopt
= mptcp_setsockopt
,
1663 .getsockopt
= mptcp_getsockopt
,
1664 .shutdown
= tcp_shutdown
,
1665 .destroy
= mptcp_destroy
,
1666 .sendmsg
= mptcp_sendmsg
,
1667 .recvmsg
= mptcp_recvmsg
,
1668 .release_cb
= mptcp_release_cb
,
1670 .unhash
= inet_unhash
,
1671 .get_port
= mptcp_get_port
,
1672 .sockets_allocated
= &mptcp_sockets_allocated
,
1673 .memory_allocated
= &tcp_memory_allocated
,
1674 .memory_pressure
= &tcp_memory_pressure
,
1675 .stream_memory_free
= mptcp_memory_free
,
1676 .sysctl_wmem_offset
= offsetof(struct net
, ipv4
.sysctl_tcp_wmem
),
1677 .sysctl_mem
= sysctl_tcp_mem
,
1678 .obj_size
= sizeof(struct mptcp_sock
),
1679 .no_autobind
= true,
1682 static int mptcp_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
1684 struct mptcp_sock
*msk
= mptcp_sk(sock
->sk
);
1685 struct socket
*ssock
;
1688 lock_sock(sock
->sk
);
1689 ssock
= __mptcp_socket_create(msk
, MPTCP_SAME_STATE
);
1690 if (IS_ERR(ssock
)) {
1691 err
= PTR_ERR(ssock
);
1695 err
= ssock
->ops
->bind(ssock
, uaddr
, addr_len
);
1697 mptcp_copy_inaddrs(sock
->sk
, ssock
->sk
);
1700 release_sock(sock
->sk
);
1704 static int mptcp_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
1705 int addr_len
, int flags
)
1707 struct mptcp_sock
*msk
= mptcp_sk(sock
->sk
);
1708 struct socket
*ssock
;
1711 lock_sock(sock
->sk
);
1712 ssock
= __mptcp_socket_create(msk
, TCP_SYN_SENT
);
1713 if (IS_ERR(ssock
)) {
1714 err
= PTR_ERR(ssock
);
1718 #ifdef CONFIG_TCP_MD5SIG
1719 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
1722 if (rcu_access_pointer(tcp_sk(ssock
->sk
)->md5sig_info
))
1723 mptcp_subflow_ctx(ssock
->sk
)->request_mptcp
= 0;
1726 err
= ssock
->ops
->connect(ssock
, uaddr
, addr_len
, flags
);
1727 inet_sk_state_store(sock
->sk
, inet_sk_state_load(ssock
->sk
));
1728 mptcp_copy_inaddrs(sock
->sk
, ssock
->sk
);
1731 release_sock(sock
->sk
);
1735 static int mptcp_v4_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
1738 if (sock
->sk
->sk_prot
== &tcp_prot
) {
1739 /* we are being invoked from __sys_accept4, after
1740 * mptcp_accept() has just accepted a non-mp-capable
1741 * flow: sk is a tcp_sk, not an mptcp one.
1743 * Hand the socket over to tcp so all further socket ops
1746 sock
->ops
= &inet_stream_ops
;
1749 return inet_getname(sock
, uaddr
, peer
);
1752 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1753 static int mptcp_v6_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
1756 if (sock
->sk
->sk_prot
== &tcpv6_prot
) {
1757 /* we are being invoked from __sys_accept4 after
1758 * mptcp_accept() has accepted a non-mp-capable
1759 * subflow: sk is a tcp_sk, not mptcp.
1761 * Hand the socket over to tcp so all further
1762 * socket ops bypass mptcp.
1764 sock
->ops
= &inet6_stream_ops
;
1767 return inet6_getname(sock
, uaddr
, peer
);
1771 static int mptcp_listen(struct socket
*sock
, int backlog
)
1773 struct mptcp_sock
*msk
= mptcp_sk(sock
->sk
);
1774 struct socket
*ssock
;
1777 pr_debug("msk=%p", msk
);
1779 lock_sock(sock
->sk
);
1780 ssock
= __mptcp_socket_create(msk
, TCP_LISTEN
);
1781 if (IS_ERR(ssock
)) {
1782 err
= PTR_ERR(ssock
);
1786 sock_set_flag(sock
->sk
, SOCK_RCU_FREE
);
1788 err
= ssock
->ops
->listen(ssock
, backlog
);
1789 inet_sk_state_store(sock
->sk
, inet_sk_state_load(ssock
->sk
));
1791 mptcp_copy_inaddrs(sock
->sk
, ssock
->sk
);
1794 release_sock(sock
->sk
);
1798 static bool is_tcp_proto(const struct proto
*p
)
1800 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1801 return p
== &tcp_prot
|| p
== &tcpv6_prot
;
1803 return p
== &tcp_prot
;
1807 static int mptcp_stream_accept(struct socket
*sock
, struct socket
*newsock
,
1808 int flags
, bool kern
)
1810 struct mptcp_sock
*msk
= mptcp_sk(sock
->sk
);
1811 struct socket
*ssock
;
1814 pr_debug("msk=%p", msk
);
1816 lock_sock(sock
->sk
);
1817 if (sock
->sk
->sk_state
!= TCP_LISTEN
)
1820 ssock
= __mptcp_nmpc_socket(msk
);
1824 sock_hold(ssock
->sk
);
1825 release_sock(sock
->sk
);
1827 err
= ssock
->ops
->accept(sock
, newsock
, flags
, kern
);
1828 if (err
== 0 && !is_tcp_proto(newsock
->sk
->sk_prot
)) {
1829 struct mptcp_sock
*msk
= mptcp_sk(newsock
->sk
);
1830 struct mptcp_subflow_context
*subflow
;
1832 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
1833 * This is needed so NOSPACE flag can be set from tcp stack.
1835 __mptcp_flush_join_list(msk
);
1836 list_for_each_entry(subflow
, &msk
->conn_list
, node
) {
1837 struct sock
*ssk
= mptcp_subflow_tcp_sock(subflow
);
1839 if (!ssk
->sk_socket
)
1840 mptcp_sock_graft(ssk
, newsock
);
1844 sock_put(ssock
->sk
);
1848 release_sock(sock
->sk
);
1852 static __poll_t
mptcp_poll(struct file
*file
, struct socket
*sock
,
1853 struct poll_table_struct
*wait
)
1855 struct sock
*sk
= sock
->sk
;
1856 struct mptcp_sock
*msk
;
1857 struct socket
*ssock
;
1862 ssock
= __mptcp_tcp_fallback(msk
);
1864 ssock
= __mptcp_nmpc_socket(msk
);
1866 mask
= ssock
->ops
->poll(file
, ssock
, wait
);
1872 sock_poll_wait(file
, sock
, wait
);
1875 if (test_bit(MPTCP_DATA_READY
, &msk
->flags
))
1876 mask
= EPOLLIN
| EPOLLRDNORM
;
1877 if (sk_stream_is_writeable(sk
) &&
1878 test_bit(MPTCP_SEND_SPACE
, &msk
->flags
))
1879 mask
|= EPOLLOUT
| EPOLLWRNORM
;
1880 if (sk
->sk_shutdown
& RCV_SHUTDOWN
)
1881 mask
|= EPOLLIN
| EPOLLRDNORM
| EPOLLRDHUP
;
1888 static int mptcp_shutdown(struct socket
*sock
, int how
)
1890 struct mptcp_sock
*msk
= mptcp_sk(sock
->sk
);
1891 struct mptcp_subflow_context
*subflow
;
1892 struct socket
*ssock
;
1895 pr_debug("sk=%p, how=%d", msk
, how
);
1897 lock_sock(sock
->sk
);
1898 ssock
= __mptcp_tcp_fallback(msk
);
1900 release_sock(sock
->sk
);
1901 return inet_shutdown(ssock
, how
);
1904 if (how
== SHUT_WR
|| how
== SHUT_RDWR
)
1905 inet_sk_state_store(sock
->sk
, TCP_FIN_WAIT1
);
1909 if ((how
& ~SHUTDOWN_MASK
) || !how
) {
1914 if (sock
->state
== SS_CONNECTING
) {
1915 if ((1 << sock
->sk
->sk_state
) &
1916 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
1917 sock
->state
= SS_DISCONNECTING
;
1919 sock
->state
= SS_CONNECTED
;
1922 __mptcp_flush_join_list(msk
);
1923 mptcp_for_each_subflow(msk
, subflow
) {
1924 struct sock
*tcp_sk
= mptcp_subflow_tcp_sock(subflow
);
1926 mptcp_subflow_shutdown(tcp_sk
, how
, 1, msk
->write_seq
);
1930 release_sock(sock
->sk
);
1935 static const struct proto_ops mptcp_stream_ops
= {
1937 .owner
= THIS_MODULE
,
1938 .release
= inet_release
,
1940 .connect
= mptcp_stream_connect
,
1941 .socketpair
= sock_no_socketpair
,
1942 .accept
= mptcp_stream_accept
,
1943 .getname
= mptcp_v4_getname
,
1945 .ioctl
= inet_ioctl
,
1946 .gettstamp
= sock_gettstamp
,
1947 .listen
= mptcp_listen
,
1948 .shutdown
= mptcp_shutdown
,
1949 .setsockopt
= sock_common_setsockopt
,
1950 .getsockopt
= sock_common_getsockopt
,
1951 .sendmsg
= inet_sendmsg
,
1952 .recvmsg
= inet_recvmsg
,
1953 .mmap
= sock_no_mmap
,
1954 .sendpage
= inet_sendpage
,
1955 #ifdef CONFIG_COMPAT
1956 .compat_setsockopt
= compat_sock_common_setsockopt
,
1957 .compat_getsockopt
= compat_sock_common_getsockopt
,
1961 static struct inet_protosw mptcp_protosw
= {
1962 .type
= SOCK_STREAM
,
1963 .protocol
= IPPROTO_MPTCP
,
1964 .prot
= &mptcp_prot
,
1965 .ops
= &mptcp_stream_ops
,
1966 .flags
= INET_PROTOSW_ICSK
,
1969 void mptcp_proto_init(void)
1971 mptcp_prot
.h
.hashinfo
= tcp_prot
.h
.hashinfo
;
1973 if (percpu_counter_init(&mptcp_sockets_allocated
, 0, GFP_KERNEL
))
1974 panic("Failed to allocate MPTCP pcpu counter\n");
1976 mptcp_subflow_init();
1979 if (proto_register(&mptcp_prot
, 1) != 0)
1980 panic("Failed to register MPTCP proto.\n");
1982 inet_register_protosw(&mptcp_protosw
);
1984 BUILD_BUG_ON(sizeof(struct mptcp_skb_cb
) > sizeof_field(struct sk_buff
, cb
));
1987 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1988 static const struct proto_ops mptcp_v6_stream_ops
= {
1990 .owner
= THIS_MODULE
,
1991 .release
= inet6_release
,
1993 .connect
= mptcp_stream_connect
,
1994 .socketpair
= sock_no_socketpair
,
1995 .accept
= mptcp_stream_accept
,
1996 .getname
= mptcp_v6_getname
,
1998 .ioctl
= inet6_ioctl
,
1999 .gettstamp
= sock_gettstamp
,
2000 .listen
= mptcp_listen
,
2001 .shutdown
= mptcp_shutdown
,
2002 .setsockopt
= sock_common_setsockopt
,
2003 .getsockopt
= sock_common_getsockopt
,
2004 .sendmsg
= inet6_sendmsg
,
2005 .recvmsg
= inet6_recvmsg
,
2006 .mmap
= sock_no_mmap
,
2007 .sendpage
= inet_sendpage
,
2008 #ifdef CONFIG_COMPAT
2009 .compat_setsockopt
= compat_sock_common_setsockopt
,
2010 .compat_getsockopt
= compat_sock_common_getsockopt
,
2014 static struct proto mptcp_v6_prot
;
2016 static void mptcp_v6_destroy(struct sock
*sk
)
2019 inet6_destroy_sock(sk
);
2022 static struct inet_protosw mptcp_v6_protosw
= {
2023 .type
= SOCK_STREAM
,
2024 .protocol
= IPPROTO_MPTCP
,
2025 .prot
= &mptcp_v6_prot
,
2026 .ops
= &mptcp_v6_stream_ops
,
2027 .flags
= INET_PROTOSW_ICSK
,
2030 int mptcp_proto_v6_init(void)
2034 mptcp_v6_prot
= mptcp_prot
;
2035 strcpy(mptcp_v6_prot
.name
, "MPTCPv6");
2036 mptcp_v6_prot
.slab
= NULL
;
2037 mptcp_v6_prot
.destroy
= mptcp_v6_destroy
;
2038 mptcp_v6_prot
.obj_size
= sizeof(struct mptcp6_sock
);
2040 err
= proto_register(&mptcp_v6_prot
, 1);
2044 err
= inet6_register_protosw(&mptcp_v6_protosw
);
2046 proto_unregister(&mptcp_v6_prot
);