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>
25 #define MPTCP_SAME_STATE TCP_MAX_STATES
27 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
29 struct mptcp_sock msk;
34 /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
35 * completed yet or has failed, return the subflow socket.
36 * Otherwise return NULL.
38 static struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
40 if (!msk->subflow || READ_ONCE(msk->can_ack))
46 static bool __mptcp_needs_tcp_fallback(const struct mptcp_sock *msk)
48 return msk->first && !sk_is_mptcp(msk->first);
51 static struct socket *__mptcp_tcp_fallback(struct mptcp_sock *msk)
53 sock_owned_by_me((const struct sock *)msk);
55 if (likely(!__mptcp_needs_tcp_fallback(msk)))
59 release_sock((struct sock *)msk);
66 static bool __mptcp_can_create_subflow(const struct mptcp_sock *msk)
71 static struct socket *__mptcp_socket_create(struct mptcp_sock *msk, int state)
73 struct mptcp_subflow_context *subflow;
74 struct sock *sk = (struct sock *)msk;
78 ssock = __mptcp_nmpc_socket(msk);
82 if (!__mptcp_can_create_subflow(msk))
83 return ERR_PTR(-EINVAL);
85 err = mptcp_subflow_create_socket(sk, &ssock);
89 msk->first = ssock->sk;
91 subflow = mptcp_subflow_ctx(ssock->sk);
92 list_add(&subflow->node, &msk->conn_list);
93 subflow->request_mptcp = 1;
96 if (state != MPTCP_SAME_STATE)
97 inet_sk_state_store(sk, state);
101 static struct sock *mptcp_subflow_get(const struct mptcp_sock *msk)
103 struct mptcp_subflow_context *subflow;
105 sock_owned_by_me((const struct sock *)msk);
107 mptcp_for_each_subflow(msk, subflow) {
108 return mptcp_subflow_tcp_sock(subflow);
114 void mptcp_data_ready(struct sock *sk)
116 struct mptcp_sock *msk = mptcp_sk(sk);
118 set_bit(MPTCP_DATA_READY, &msk->flags);
119 sk->sk_data_ready(sk);
122 static bool mptcp_ext_cache_refill(struct mptcp_sock *msk)
124 if (!msk->cached_ext)
125 msk->cached_ext = __skb_ext_alloc();
127 return !!msk->cached_ext;
130 static struct sock *mptcp_subflow_recv_lookup(const struct mptcp_sock *msk)
132 struct mptcp_subflow_context *subflow;
133 struct sock *sk = (struct sock *)msk;
135 sock_owned_by_me(sk);
137 mptcp_for_each_subflow(msk, subflow) {
138 if (subflow->data_avail)
139 return mptcp_subflow_tcp_sock(subflow);
145 static inline bool mptcp_skb_can_collapse_to(const struct mptcp_sock *msk,
146 const struct sk_buff *skb,
147 const struct mptcp_ext *mpext)
149 if (!tcp_skb_can_collapse_to(skb))
152 /* can collapse only if MPTCP level sequence is in order */
153 return mpext && mpext->data_seq + mpext->data_len == msk->write_seq;
156 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
157 struct msghdr *msg, long *timeo, int *pmss_now,
160 int mss_now, avail_size, size_goal, ret;
161 struct mptcp_sock *msk = mptcp_sk(sk);
162 struct mptcp_ext *mpext = NULL;
163 struct sk_buff *skb, *tail;
164 bool can_collapse = false;
165 struct page_frag *pfrag;
168 /* use the mptcp page cache so that we can easily move the data
169 * from one substream to another, but do per subflow memory accounting
171 pfrag = sk_page_frag(sk);
172 while (!sk_page_frag_refill(ssk, pfrag) ||
173 !mptcp_ext_cache_refill(msk)) {
174 ret = sk_stream_wait_memory(ssk, timeo);
177 if (unlikely(__mptcp_needs_tcp_fallback(msk)))
181 /* compute copy limit */
182 mss_now = tcp_send_mss(ssk, &size_goal, msg->msg_flags);
184 *ps_goal = size_goal;
185 avail_size = size_goal;
186 skb = tcp_write_queue_tail(ssk);
188 mpext = skb_ext_find(skb, SKB_EXT_MPTCP);
190 /* Limit the write to the size available in the
191 * current skb, if any, so that we create at most a new skb.
192 * Explicitly tells TCP internals to avoid collapsing on later
193 * queue management operation, to avoid breaking the ext <->
194 * SSN association set here
196 can_collapse = (size_goal - skb->len > 0) &&
197 mptcp_skb_can_collapse_to(msk, skb, mpext);
199 TCP_SKB_CB(skb)->eor = 1;
201 avail_size = size_goal - skb->len;
203 psize = min_t(size_t, pfrag->size - pfrag->offset, avail_size);
206 pr_debug("left=%zu", msg_data_left(msg));
207 psize = copy_page_from_iter(pfrag->page, pfrag->offset,
208 min_t(size_t, msg_data_left(msg), psize),
210 pr_debug("left=%zu", msg_data_left(msg));
214 /* tell the TCP stack to delay the push so that we can safely
215 * access the skb after the sendpages call
217 ret = do_tcp_sendpages(ssk, pfrag->page, pfrag->offset, psize,
218 msg->msg_flags | MSG_SENDPAGE_NOTLAST);
221 if (unlikely(ret < psize))
222 iov_iter_revert(&msg->msg_iter, psize - ret);
224 /* if the tail skb extension is still the cached one, collapsing
225 * really happened. Note: we can't check for 'same skb' as the sk_buff
226 * hdr on tail can be transmitted, freed and re-allocated by the
227 * do_tcp_sendpages() call
229 tail = tcp_write_queue_tail(ssk);
230 if (mpext && tail && mpext == skb_ext_find(tail, SKB_EXT_MPTCP)) {
231 WARN_ON_ONCE(!can_collapse);
232 mpext->data_len += ret;
236 skb = tcp_write_queue_tail(ssk);
237 mpext = __skb_ext_set(skb, SKB_EXT_MPTCP, msk->cached_ext);
238 msk->cached_ext = NULL;
240 memset(mpext, 0, sizeof(*mpext));
241 mpext->data_seq = msk->write_seq;
242 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
243 mpext->data_len = ret;
247 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
248 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
252 pfrag->offset += ret;
253 msk->write_seq += ret;
254 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
259 static void ssk_check_wmem(struct mptcp_sock *msk, struct sock *ssk)
263 if (likely(sk_stream_is_writeable(ssk)))
266 sock = READ_ONCE(ssk->sk_socket);
269 clear_bit(MPTCP_SEND_SPACE, &msk->flags);
270 smp_mb__after_atomic();
271 /* set NOSPACE only after clearing SEND_SPACE flag */
272 set_bit(SOCK_NOSPACE, &sock->flags);
276 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
278 int mss_now = 0, size_goal = 0, ret = 0;
279 struct mptcp_sock *msk = mptcp_sk(sk);
280 struct socket *ssock;
285 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
289 ssock = __mptcp_tcp_fallback(msk);
290 if (unlikely(ssock)) {
292 pr_debug("fallback passthrough");
293 ret = sock_sendmsg(ssock, msg);
294 return ret >= 0 ? ret + copied : (copied ? copied : ret);
297 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
299 ssk = mptcp_subflow_get(msk);
305 pr_debug("conn_list->subflow=%p", ssk);
308 while (msg_data_left(msg)) {
309 ret = mptcp_sendmsg_frag(sk, ssk, msg, &timeo, &mss_now,
313 if (ret == 0 && unlikely(__mptcp_needs_tcp_fallback(msk))) {
315 ssock = __mptcp_tcp_fallback(msk);
324 tcp_push(ssk, msg->msg_flags, mss_now, tcp_sk(ssk)->nonagle,
328 ssk_check_wmem(msk, ssk);
334 int mptcp_read_actor(read_descriptor_t *desc, struct sk_buff *skb,
335 unsigned int offset, size_t len)
337 struct mptcp_read_arg *arg = desc->arg.data;
340 copy_len = min(desc->count, len);
342 if (likely(arg->msg)) {
345 err = skb_copy_datagram_msg(skb, offset, arg->msg, copy_len);
347 pr_debug("error path");
352 pr_debug("Flushing skb payload");
355 desc->count -= copy_len;
357 pr_debug("consumed %zu bytes, %zu left", copy_len, desc->count);
361 static void mptcp_wait_data(struct sock *sk, long *timeo)
363 DEFINE_WAIT_FUNC(wait, woken_wake_function);
364 struct mptcp_sock *msk = mptcp_sk(sk);
366 add_wait_queue(sk_sleep(sk), &wait);
367 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
369 sk_wait_event(sk, timeo,
370 test_and_clear_bit(MPTCP_DATA_READY, &msk->flags), &wait);
372 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
373 remove_wait_queue(sk_sleep(sk), &wait);
376 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
377 int nonblock, int flags, int *addr_len)
379 struct mptcp_sock *msk = mptcp_sk(sk);
380 struct mptcp_subflow_context *subflow;
381 bool more_data_avail = false;
382 struct mptcp_read_arg arg;
383 read_descriptor_t desc;
384 bool wait_data = false;
385 struct socket *ssock;
393 if (msg->msg_flags & ~(MSG_WAITALL | MSG_DONTWAIT))
397 ssock = __mptcp_tcp_fallback(msk);
398 if (unlikely(ssock)) {
400 pr_debug("fallback-read subflow=%p",
401 mptcp_subflow_ctx(ssock->sk));
402 copied = sock_recvmsg(ssock, msg, flags);
407 desc.arg.data = &arg;
410 timeo = sock_rcvtimeo(sk, nonblock);
412 len = min_t(size_t, len, INT_MAX);
413 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
419 ssk = mptcp_subflow_recv_lookup(msk);
420 pr_debug("msk=%p ssk=%p", msk, ssk);
424 subflow = mptcp_subflow_ctx(ssk);
429 /* try to read as much data as available */
430 map_remaining = subflow->map_data_len -
431 mptcp_subflow_get_map_offset(subflow);
432 desc.count = min_t(size_t, len - copied, map_remaining);
433 pr_debug("reading %zu bytes, copied %d", desc.count,
435 bytes_read = tcp_read_sock(ssk, &desc,
437 if (bytes_read < 0) {
444 pr_debug("msk ack_seq=%llx -> %llx", msk->ack_seq,
445 msk->ack_seq + bytes_read);
446 msk->ack_seq += bytes_read;
447 copied += bytes_read;
452 if (tp->urg_data && tp->urg_seq == tp->copied_seq) {
453 pr_err("Urgent data present, cannot proceed");
458 more_data_avail = mptcp_subflow_data_available(ssk);
459 } while (more_data_avail && !done);
464 more_data_avail = false;
466 /* only the master socket status is relevant here. The exit
467 * conditions mirror closely tcp_recvmsg()
469 if (copied >= target)
474 sk->sk_state == TCP_CLOSE ||
475 (sk->sk_shutdown & RCV_SHUTDOWN) ||
477 signal_pending(current))
481 copied = sock_error(sk);
485 if (sk->sk_shutdown & RCV_SHUTDOWN)
488 if (sk->sk_state == TCP_CLOSE) {
498 if (signal_pending(current)) {
499 copied = sock_intr_errno(timeo);
504 pr_debug("block timeout %ld", timeo);
506 mptcp_wait_data(sk, &timeo);
507 if (unlikely(__mptcp_tcp_fallback(msk)))
511 if (more_data_avail) {
512 if (!test_bit(MPTCP_DATA_READY, &msk->flags))
513 set_bit(MPTCP_DATA_READY, &msk->flags);
514 } else if (!wait_data) {
515 clear_bit(MPTCP_DATA_READY, &msk->flags);
517 /* .. race-breaker: ssk might get new data after last
518 * data_available() returns false.
520 ssk = mptcp_subflow_recv_lookup(msk);
522 set_bit(MPTCP_DATA_READY, &msk->flags);
529 /* subflow sockets can be either outgoing (connect) or incoming
532 * Outgoing subflows use in-kernel sockets.
533 * Incoming subflows do not have their own 'struct socket' allocated,
534 * so we need to use tcp_close() after detaching them from the mptcp
537 static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
538 struct mptcp_subflow_context *subflow,
541 struct socket *sock = READ_ONCE(ssk->sk_socket);
543 list_del(&subflow->node);
545 if (sock && sock != sk->sk_socket) {
546 /* outgoing subflow */
549 /* incoming subflow */
550 tcp_close(ssk, timeout);
554 static void mptcp_worker(struct work_struct *work)
556 struct mptcp_sock *msk = container_of(work, struct mptcp_sock, work);
557 struct sock *sk = &msk->sk.icsk_inet.sk;
565 static int __mptcp_init_sock(struct sock *sk)
567 struct mptcp_sock *msk = mptcp_sk(sk);
569 INIT_LIST_HEAD(&msk->conn_list);
570 __set_bit(MPTCP_SEND_SPACE, &msk->flags);
571 INIT_WORK(&msk->work, mptcp_worker);
578 static int mptcp_init_sock(struct sock *sk)
580 if (!mptcp_is_enabled(sock_net(sk)))
583 return __mptcp_init_sock(sk);
586 static void mptcp_cancel_work(struct sock *sk)
588 struct mptcp_sock *msk = mptcp_sk(sk);
590 if (cancel_work_sync(&msk->work))
594 static void mptcp_subflow_shutdown(struct sock *ssk, int how)
598 switch (ssk->sk_state) {
600 if (!(how & RCV_SHUTDOWN))
604 tcp_disconnect(ssk, O_NONBLOCK);
607 ssk->sk_shutdown |= how;
608 tcp_shutdown(ssk, how);
612 /* Wake up anyone sleeping in poll. */
613 ssk->sk_state_change(ssk);
617 /* Called with msk lock held, releases such lock before returning */
618 static void mptcp_close(struct sock *sk, long timeout)
620 struct mptcp_subflow_context *subflow, *tmp;
621 struct mptcp_sock *msk = mptcp_sk(sk);
622 LIST_HEAD(conn_list);
626 mptcp_token_destroy(msk->token);
627 inet_sk_state_store(sk, TCP_CLOSE);
629 list_splice_init(&msk->conn_list, &conn_list);
633 list_for_each_entry_safe(subflow, tmp, &conn_list, node) {
634 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
636 __mptcp_close_ssk(sk, ssk, subflow, timeout);
639 mptcp_cancel_work(sk);
641 sk_common_release(sk);
644 static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
646 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
647 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
648 struct ipv6_pinfo *msk6 = inet6_sk(msk);
650 msk->sk_v6_daddr = ssk->sk_v6_daddr;
651 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
654 msk6->saddr = ssk6->saddr;
655 msk6->flow_label = ssk6->flow_label;
659 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
660 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
661 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
662 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
663 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
664 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
667 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
668 static struct ipv6_pinfo *mptcp_inet6_sk(const struct sock *sk)
670 unsigned int offset = sizeof(struct mptcp6_sock) - sizeof(struct ipv6_pinfo);
672 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
676 static struct sock *mptcp_sk_clone_lock(const struct sock *sk)
678 struct sock *nsk = sk_clone_lock(sk, GFP_ATOMIC);
683 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
684 if (nsk->sk_family == AF_INET6)
685 inet_sk(nsk)->pinet6 = mptcp_inet6_sk(nsk);
691 static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
694 struct mptcp_sock *msk = mptcp_sk(sk);
695 struct socket *listener;
698 listener = __mptcp_nmpc_socket(msk);
699 if (WARN_ON_ONCE(!listener)) {
704 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
705 newsk = inet_csk_accept(listener->sk, flags, err, kern);
709 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
711 if (sk_is_mptcp(newsk)) {
712 struct mptcp_subflow_context *subflow;
713 struct sock *new_mptcp_sock;
714 struct sock *ssk = newsk;
717 subflow = mptcp_subflow_ctx(newsk);
721 new_mptcp_sock = mptcp_sk_clone_lock(sk);
722 if (!new_mptcp_sock) {
726 mptcp_subflow_shutdown(newsk, SHUT_RDWR + 1);
731 __mptcp_init_sock(new_mptcp_sock);
733 msk = mptcp_sk(new_mptcp_sock);
734 msk->local_key = subflow->local_key;
735 msk->token = subflow->token;
739 mptcp_token_update_accept(newsk, new_mptcp_sock);
741 msk->write_seq = subflow->idsn + 1;
742 if (subflow->can_ack) {
744 msk->remote_key = subflow->remote_key;
745 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
747 msk->ack_seq = ack_seq;
749 newsk = new_mptcp_sock;
750 mptcp_copy_inaddrs(newsk, ssk);
751 list_add(&subflow->node, &msk->conn_list);
753 /* will be fully established at mptcp_stream_accept()
756 inet_sk_state_store(new_mptcp_sock, TCP_SYN_RECV);
757 bh_unlock_sock(new_mptcp_sock);
761 /* the subflow can already receive packet, avoid racing with
762 * the receive path and process the pending ones
765 subflow->rel_write_seq = 1;
766 subflow->tcp_sock = ssk;
767 subflow->conn = new_mptcp_sock;
768 if (unlikely(!skb_queue_empty(&ssk->sk_receive_queue)))
769 mptcp_subflow_data_available(ssk);
776 static void mptcp_destroy(struct sock *sk)
778 struct mptcp_sock *msk = mptcp_sk(sk);
781 __skb_ext_put(msk->cached_ext);
784 static int mptcp_setsockopt(struct sock *sk, int level, int optname,
785 char __user *optval, unsigned int optlen)
787 struct mptcp_sock *msk = mptcp_sk(sk);
788 struct socket *ssock;
790 pr_debug("msk=%p", msk);
792 /* @@ the meaning of setsockopt() when the socket is connected and
793 * there are multiple subflows is not yet defined. It is up to the
794 * MPTCP-level socket to configure the subflows until the subflow
795 * is in TCP fallback, when TCP socket options are passed through
796 * to the one remaining subflow.
799 ssock = __mptcp_tcp_fallback(msk);
801 return tcp_setsockopt(ssock->sk, level, optname, optval,
809 static int mptcp_getsockopt(struct sock *sk, int level, int optname,
810 char __user *optval, int __user *option)
812 struct mptcp_sock *msk = mptcp_sk(sk);
813 struct socket *ssock;
815 pr_debug("msk=%p", msk);
817 /* @@ the meaning of setsockopt() when the socket is connected and
818 * there are multiple subflows is not yet defined. It is up to the
819 * MPTCP-level socket to configure the subflows until the subflow
820 * is in TCP fallback, when socket options are passed through
821 * to the one remaining subflow.
824 ssock = __mptcp_tcp_fallback(msk);
826 return tcp_getsockopt(ssock->sk, level, optname, optval,
834 static int mptcp_get_port(struct sock *sk, unsigned short snum)
836 struct mptcp_sock *msk = mptcp_sk(sk);
837 struct socket *ssock;
839 ssock = __mptcp_nmpc_socket(msk);
840 pr_debug("msk=%p, subflow=%p", msk, ssock);
841 if (WARN_ON_ONCE(!ssock))
844 return inet_csk_get_port(ssock->sk, snum);
847 void mptcp_finish_connect(struct sock *ssk)
849 struct mptcp_subflow_context *subflow;
850 struct mptcp_sock *msk;
854 subflow = mptcp_subflow_ctx(ssk);
856 if (!subflow->mp_capable)
862 pr_debug("msk=%p, token=%u", sk, subflow->token);
864 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
866 subflow->map_seq = ack_seq;
867 subflow->map_subflow_seq = 1;
868 subflow->rel_write_seq = 1;
870 /* the socket is not connected yet, no msk/subflow ops can access/race
871 * accessing the field below
873 WRITE_ONCE(msk->remote_key, subflow->remote_key);
874 WRITE_ONCE(msk->local_key, subflow->local_key);
875 WRITE_ONCE(msk->token, subflow->token);
876 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
877 WRITE_ONCE(msk->ack_seq, ack_seq);
878 WRITE_ONCE(msk->can_ack, 1);
881 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)
883 write_lock_bh(&sk->sk_callback_lock);
884 rcu_assign_pointer(sk->sk_wq, &parent->wq);
885 sk_set_socket(sk, parent);
886 sk->sk_uid = SOCK_INODE(parent)->i_uid;
887 write_unlock_bh(&sk->sk_callback_lock);
890 static bool mptcp_memory_free(const struct sock *sk, int wake)
892 struct mptcp_sock *msk = mptcp_sk(sk);
894 return wake ? test_bit(MPTCP_SEND_SPACE, &msk->flags) : true;
897 static struct proto mptcp_prot = {
899 .owner = THIS_MODULE,
900 .init = mptcp_init_sock,
901 .close = mptcp_close,
902 .accept = mptcp_accept,
903 .setsockopt = mptcp_setsockopt,
904 .getsockopt = mptcp_getsockopt,
905 .shutdown = tcp_shutdown,
906 .destroy = mptcp_destroy,
907 .sendmsg = mptcp_sendmsg,
908 .recvmsg = mptcp_recvmsg,
910 .unhash = inet_unhash,
911 .get_port = mptcp_get_port,
912 .stream_memory_free = mptcp_memory_free,
913 .obj_size = sizeof(struct mptcp_sock),
917 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
919 struct mptcp_sock *msk = mptcp_sk(sock->sk);
920 struct socket *ssock;
924 ssock = __mptcp_socket_create(msk, MPTCP_SAME_STATE);
926 err = PTR_ERR(ssock);
930 err = ssock->ops->bind(ssock, uaddr, addr_len);
932 mptcp_copy_inaddrs(sock->sk, ssock->sk);
935 release_sock(sock->sk);
939 static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
940 int addr_len, int flags)
942 struct mptcp_sock *msk = mptcp_sk(sock->sk);
943 struct socket *ssock;
947 ssock = __mptcp_socket_create(msk, TCP_SYN_SENT);
949 err = PTR_ERR(ssock);
953 #ifdef CONFIG_TCP_MD5SIG
954 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
957 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
958 mptcp_subflow_ctx(ssock->sk)->request_mptcp = 0;
961 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
962 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
963 mptcp_copy_inaddrs(sock->sk, ssock->sk);
966 release_sock(sock->sk);
970 static int mptcp_v4_getname(struct socket *sock, struct sockaddr *uaddr,
973 if (sock->sk->sk_prot == &tcp_prot) {
974 /* we are being invoked from __sys_accept4, after
975 * mptcp_accept() has just accepted a non-mp-capable
976 * flow: sk is a tcp_sk, not an mptcp one.
978 * Hand the socket over to tcp so all further socket ops
981 sock->ops = &inet_stream_ops;
984 return inet_getname(sock, uaddr, peer);
987 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
988 static int mptcp_v6_getname(struct socket *sock, struct sockaddr *uaddr,
991 if (sock->sk->sk_prot == &tcpv6_prot) {
992 /* we are being invoked from __sys_accept4 after
993 * mptcp_accept() has accepted a non-mp-capable
994 * subflow: sk is a tcp_sk, not mptcp.
996 * Hand the socket over to tcp so all further
997 * socket ops bypass mptcp.
999 sock->ops = &inet6_stream_ops;
1002 return inet6_getname(sock, uaddr, peer);
1006 static int mptcp_listen(struct socket *sock, int backlog)
1008 struct mptcp_sock *msk = mptcp_sk(sock->sk);
1009 struct socket *ssock;
1012 pr_debug("msk=%p", msk);
1014 lock_sock(sock->sk);
1015 ssock = __mptcp_socket_create(msk, TCP_LISTEN);
1016 if (IS_ERR(ssock)) {
1017 err = PTR_ERR(ssock);
1021 err = ssock->ops->listen(ssock, backlog);
1022 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
1024 mptcp_copy_inaddrs(sock->sk, ssock->sk);
1027 release_sock(sock->sk);
1031 static bool is_tcp_proto(const struct proto *p)
1033 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1034 return p == &tcp_prot || p == &tcpv6_prot;
1036 return p == &tcp_prot;
1040 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
1041 int flags, bool kern)
1043 struct mptcp_sock *msk = mptcp_sk(sock->sk);
1044 struct socket *ssock;
1047 pr_debug("msk=%p", msk);
1049 lock_sock(sock->sk);
1050 if (sock->sk->sk_state != TCP_LISTEN)
1053 ssock = __mptcp_nmpc_socket(msk);
1057 sock_hold(ssock->sk);
1058 release_sock(sock->sk);
1060 err = ssock->ops->accept(sock, newsock, flags, kern);
1061 if (err == 0 && !is_tcp_proto(newsock->sk->sk_prot)) {
1062 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
1063 struct mptcp_subflow_context *subflow;
1065 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
1066 * This is needed so NOSPACE flag can be set from tcp stack.
1068 list_for_each_entry(subflow, &msk->conn_list, node) {
1069 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1071 if (!ssk->sk_socket)
1072 mptcp_sock_graft(ssk, newsock);
1075 inet_sk_state_store(newsock->sk, TCP_ESTABLISHED);
1078 sock_put(ssock->sk);
1082 release_sock(sock->sk);
1086 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
1087 struct poll_table_struct *wait)
1089 struct sock *sk = sock->sk;
1090 struct mptcp_sock *msk;
1091 struct socket *ssock;
1096 ssock = __mptcp_nmpc_socket(msk);
1098 mask = ssock->ops->poll(file, ssock, wait);
1104 sock_poll_wait(file, sock, wait);
1106 ssock = __mptcp_tcp_fallback(msk);
1107 if (unlikely(ssock))
1108 return ssock->ops->poll(file, ssock, NULL);
1110 if (test_bit(MPTCP_DATA_READY, &msk->flags))
1111 mask = EPOLLIN | EPOLLRDNORM;
1112 if (sk_stream_is_writeable(sk) &&
1113 test_bit(MPTCP_SEND_SPACE, &msk->flags))
1114 mask |= EPOLLOUT | EPOLLWRNORM;
1115 if (sk->sk_shutdown & RCV_SHUTDOWN)
1116 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1123 static int mptcp_shutdown(struct socket *sock, int how)
1125 struct mptcp_sock *msk = mptcp_sk(sock->sk);
1126 struct mptcp_subflow_context *subflow;
1129 pr_debug("sk=%p, how=%d", msk, how);
1131 lock_sock(sock->sk);
1133 if (how == SHUT_WR || how == SHUT_RDWR)
1134 inet_sk_state_store(sock->sk, TCP_FIN_WAIT1);
1138 if ((how & ~SHUTDOWN_MASK) || !how) {
1143 if (sock->state == SS_CONNECTING) {
1144 if ((1 << sock->sk->sk_state) &
1145 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
1146 sock->state = SS_DISCONNECTING;
1148 sock->state = SS_CONNECTED;
1151 mptcp_for_each_subflow(msk, subflow) {
1152 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
1154 mptcp_subflow_shutdown(tcp_sk, how);
1158 release_sock(sock->sk);
1163 static const struct proto_ops mptcp_stream_ops = {
1165 .owner = THIS_MODULE,
1166 .release = inet_release,
1168 .connect = mptcp_stream_connect,
1169 .socketpair = sock_no_socketpair,
1170 .accept = mptcp_stream_accept,
1171 .getname = mptcp_v4_getname,
1173 .ioctl = inet_ioctl,
1174 .gettstamp = sock_gettstamp,
1175 .listen = mptcp_listen,
1176 .shutdown = mptcp_shutdown,
1177 .setsockopt = sock_common_setsockopt,
1178 .getsockopt = sock_common_getsockopt,
1179 .sendmsg = inet_sendmsg,
1180 .recvmsg = inet_recvmsg,
1181 .mmap = sock_no_mmap,
1182 .sendpage = inet_sendpage,
1183 #ifdef CONFIG_COMPAT
1184 .compat_setsockopt = compat_sock_common_setsockopt,
1185 .compat_getsockopt = compat_sock_common_getsockopt,
1189 static struct inet_protosw mptcp_protosw = {
1190 .type = SOCK_STREAM,
1191 .protocol = IPPROTO_MPTCP,
1192 .prot = &mptcp_prot,
1193 .ops = &mptcp_stream_ops,
1194 .flags = INET_PROTOSW_ICSK,
1197 void mptcp_proto_init(void)
1199 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
1201 mptcp_subflow_init();
1203 if (proto_register(&mptcp_prot, 1) != 0)
1204 panic("Failed to register MPTCP proto.\n");
1206 inet_register_protosw(&mptcp_protosw);
1209 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1210 static const struct proto_ops mptcp_v6_stream_ops = {
1212 .owner = THIS_MODULE,
1213 .release = inet6_release,
1215 .connect = mptcp_stream_connect,
1216 .socketpair = sock_no_socketpair,
1217 .accept = mptcp_stream_accept,
1218 .getname = mptcp_v6_getname,
1220 .ioctl = inet6_ioctl,
1221 .gettstamp = sock_gettstamp,
1222 .listen = mptcp_listen,
1223 .shutdown = mptcp_shutdown,
1224 .setsockopt = sock_common_setsockopt,
1225 .getsockopt = sock_common_getsockopt,
1226 .sendmsg = inet6_sendmsg,
1227 .recvmsg = inet6_recvmsg,
1228 .mmap = sock_no_mmap,
1229 .sendpage = inet_sendpage,
1230 #ifdef CONFIG_COMPAT
1231 .compat_setsockopt = compat_sock_common_setsockopt,
1232 .compat_getsockopt = compat_sock_common_getsockopt,
1236 static struct proto mptcp_v6_prot;
1238 static void mptcp_v6_destroy(struct sock *sk)
1241 inet6_destroy_sock(sk);
1244 static struct inet_protosw mptcp_v6_protosw = {
1245 .type = SOCK_STREAM,
1246 .protocol = IPPROTO_MPTCP,
1247 .prot = &mptcp_v6_prot,
1248 .ops = &mptcp_v6_stream_ops,
1249 .flags = INET_PROTOSW_ICSK,
1252 int mptcp_proto_v6_init(void)
1256 mptcp_v6_prot = mptcp_prot;
1257 strcpy(mptcp_v6_prot.name, "MPTCPv6");
1258 mptcp_v6_prot.slab = NULL;
1259 mptcp_v6_prot.destroy = mptcp_v6_destroy;
1260 mptcp_v6_prot.obj_size = sizeof(struct mptcp6_sock);
1262 err = proto_register(&mptcp_v6_prot, 1);
1266 err = inet6_register_protosw(&mptcp_v6_protosw);
1268 proto_unregister(&mptcp_v6_prot);