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>
13 #include <net/inet_common.h>
14 #include <net/inet_hashtables.h>
15 #include <net/protocol.h>
17 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
18 #include <net/transp_v6.h>
20 #include <net/mptcp.h>
23 #define MPTCP_SAME_STATE TCP_MAX_STATES
25 /* If msk has an initial subflow socket, and the MP_CAPABLE handshake has not
26 * completed yet or has failed, return the subflow socket.
27 * Otherwise return NULL.
29 static struct socket *__mptcp_nmpc_socket(const struct mptcp_sock *msk)
31 if (!msk->subflow || mptcp_subflow_ctx(msk->subflow->sk)->fourth_ack)
37 /* if msk has a single subflow, and the mp_capable handshake is failed,
39 * Otherwise returns NULL
41 static struct socket *__mptcp_tcp_fallback(const struct mptcp_sock *msk)
43 struct socket *ssock = __mptcp_nmpc_socket(msk);
45 sock_owned_by_me((const struct sock *)msk);
47 if (!ssock || sk_is_mptcp(ssock->sk))
53 static bool __mptcp_can_create_subflow(const struct mptcp_sock *msk)
55 return ((struct sock *)msk)->sk_state == TCP_CLOSE;
58 static struct socket *__mptcp_socket_create(struct mptcp_sock *msk, int state)
60 struct mptcp_subflow_context *subflow;
61 struct sock *sk = (struct sock *)msk;
65 ssock = __mptcp_nmpc_socket(msk);
69 if (!__mptcp_can_create_subflow(msk))
70 return ERR_PTR(-EINVAL);
72 err = mptcp_subflow_create_socket(sk, &ssock);
77 subflow = mptcp_subflow_ctx(ssock->sk);
78 list_add(&subflow->node, &msk->conn_list);
79 subflow->request_mptcp = 1;
82 if (state != MPTCP_SAME_STATE)
83 inet_sk_state_store(sk, state);
87 static struct sock *mptcp_subflow_get(const struct mptcp_sock *msk)
89 struct mptcp_subflow_context *subflow;
91 sock_owned_by_me((const struct sock *)msk);
93 mptcp_for_each_subflow(msk, subflow) {
94 return mptcp_subflow_tcp_sock(subflow);
100 static bool mptcp_ext_cache_refill(struct mptcp_sock *msk)
102 if (!msk->cached_ext)
103 msk->cached_ext = __skb_ext_alloc();
105 return !!msk->cached_ext;
108 static int mptcp_sendmsg_frag(struct sock *sk, struct sock *ssk,
109 struct msghdr *msg, long *timeo)
111 int mss_now = 0, size_goal = 0, ret = 0;
112 struct mptcp_sock *msk = mptcp_sk(sk);
113 struct mptcp_ext *mpext = NULL;
114 struct page_frag *pfrag;
118 /* use the mptcp page cache so that we can easily move the data
119 * from one substream to another, but do per subflow memory accounting
121 pfrag = sk_page_frag(sk);
122 while (!sk_page_frag_refill(ssk, pfrag) ||
123 !mptcp_ext_cache_refill(msk)) {
124 ret = sk_stream_wait_memory(ssk, timeo);
129 /* compute copy limit */
130 mss_now = tcp_send_mss(ssk, &size_goal, msg->msg_flags);
131 psize = min_t(int, pfrag->size - pfrag->offset, size_goal);
133 pr_debug("left=%zu", msg_data_left(msg));
134 psize = copy_page_from_iter(pfrag->page, pfrag->offset,
135 min_t(size_t, msg_data_left(msg), psize),
137 pr_debug("left=%zu", msg_data_left(msg));
141 /* Mark the end of the previous write so the beginning of the
142 * next write (with its own mptcp skb extension data) is not
145 skb = tcp_write_queue_tail(ssk);
147 TCP_SKB_CB(skb)->eor = 1;
149 ret = do_tcp_sendpages(ssk, pfrag->page, pfrag->offset, psize,
150 msg->msg_flags | MSG_SENDPAGE_NOTLAST);
153 if (unlikely(ret < psize))
154 iov_iter_revert(&msg->msg_iter, psize - ret);
156 skb = tcp_write_queue_tail(ssk);
157 mpext = __skb_ext_set(skb, SKB_EXT_MPTCP, msk->cached_ext);
158 msk->cached_ext = NULL;
160 memset(mpext, 0, sizeof(*mpext));
161 mpext->data_seq = msk->write_seq;
162 mpext->subflow_seq = mptcp_subflow_ctx(ssk)->rel_write_seq;
163 mpext->data_len = ret;
167 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u dsn64=%d",
168 mpext->data_seq, mpext->subflow_seq, mpext->data_len,
171 pfrag->offset += ret;
172 msk->write_seq += ret;
173 mptcp_subflow_ctx(ssk)->rel_write_seq += ret;
175 tcp_push(ssk, msg->msg_flags, mss_now, tcp_sk(ssk)->nonagle, size_goal);
179 static int mptcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
181 struct mptcp_sock *msk = mptcp_sk(sk);
182 struct socket *ssock;
188 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
192 ssock = __mptcp_tcp_fallback(msk);
194 pr_debug("fallback passthrough");
195 ret = sock_sendmsg(ssock, msg);
200 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
202 ssk = mptcp_subflow_get(msk);
208 pr_debug("conn_list->subflow=%p", ssk);
211 while (msg_data_left(msg)) {
212 ret = mptcp_sendmsg_frag(sk, ssk, msg, &timeo);
227 int mptcp_read_actor(read_descriptor_t *desc, struct sk_buff *skb,
228 unsigned int offset, size_t len)
230 struct mptcp_read_arg *arg = desc->arg.data;
233 copy_len = min(desc->count, len);
235 if (likely(arg->msg)) {
238 err = skb_copy_datagram_msg(skb, offset, arg->msg, copy_len);
240 pr_debug("error path");
245 pr_debug("Flushing skb payload");
248 desc->count -= copy_len;
250 pr_debug("consumed %zu bytes, %zu left", copy_len, desc->count);
254 static int mptcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
255 int nonblock, int flags, int *addr_len)
257 struct mptcp_sock *msk = mptcp_sk(sk);
258 struct socket *ssock;
262 if (msg->msg_flags & ~(MSG_WAITALL | MSG_DONTWAIT))
266 ssock = __mptcp_tcp_fallback(msk);
268 pr_debug("fallback-read subflow=%p",
269 mptcp_subflow_ctx(ssock->sk));
270 copied = sock_recvmsg(ssock, msg, flags);
275 ssk = mptcp_subflow_get(msk);
281 copied = sock_recvmsg(ssk->sk_socket, msg, flags);
288 /* subflow sockets can be either outgoing (connect) or incoming
291 * Outgoing subflows use in-kernel sockets.
292 * Incoming subflows do not have their own 'struct socket' allocated,
293 * so we need to use tcp_close() after detaching them from the mptcp
296 static void __mptcp_close_ssk(struct sock *sk, struct sock *ssk,
297 struct mptcp_subflow_context *subflow,
300 struct socket *sock = READ_ONCE(ssk->sk_socket);
302 list_del(&subflow->node);
304 if (sock && sock != sk->sk_socket) {
305 /* outgoing subflow */
308 /* incoming subflow */
309 tcp_close(ssk, timeout);
313 static int mptcp_init_sock(struct sock *sk)
315 struct mptcp_sock *msk = mptcp_sk(sk);
317 INIT_LIST_HEAD(&msk->conn_list);
322 static void mptcp_subflow_shutdown(struct sock *ssk, int how)
326 switch (ssk->sk_state) {
328 if (!(how & RCV_SHUTDOWN))
332 tcp_disconnect(ssk, O_NONBLOCK);
335 ssk->sk_shutdown |= how;
336 tcp_shutdown(ssk, how);
340 /* Wake up anyone sleeping in poll. */
341 ssk->sk_state_change(ssk);
345 static void mptcp_close(struct sock *sk, long timeout)
347 struct mptcp_subflow_context *subflow, *tmp;
348 struct mptcp_sock *msk = mptcp_sk(sk);
350 mptcp_token_destroy(msk->token);
351 inet_sk_state_store(sk, TCP_CLOSE);
355 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
356 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
358 __mptcp_close_ssk(sk, ssk, subflow, timeout);
362 __skb_ext_put(msk->cached_ext);
364 sk_common_release(sk);
367 static void mptcp_copy_inaddrs(struct sock *msk, const struct sock *ssk)
369 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
370 const struct ipv6_pinfo *ssk6 = inet6_sk(ssk);
371 struct ipv6_pinfo *msk6 = inet6_sk(msk);
373 msk->sk_v6_daddr = ssk->sk_v6_daddr;
374 msk->sk_v6_rcv_saddr = ssk->sk_v6_rcv_saddr;
377 msk6->saddr = ssk6->saddr;
378 msk6->flow_label = ssk6->flow_label;
382 inet_sk(msk)->inet_num = inet_sk(ssk)->inet_num;
383 inet_sk(msk)->inet_dport = inet_sk(ssk)->inet_dport;
384 inet_sk(msk)->inet_sport = inet_sk(ssk)->inet_sport;
385 inet_sk(msk)->inet_daddr = inet_sk(ssk)->inet_daddr;
386 inet_sk(msk)->inet_saddr = inet_sk(ssk)->inet_saddr;
387 inet_sk(msk)->inet_rcv_saddr = inet_sk(ssk)->inet_rcv_saddr;
390 static struct sock *mptcp_accept(struct sock *sk, int flags, int *err,
393 struct mptcp_sock *msk = mptcp_sk(sk);
394 struct socket *listener;
397 listener = __mptcp_nmpc_socket(msk);
398 if (WARN_ON_ONCE(!listener)) {
403 pr_debug("msk=%p, listener=%p", msk, mptcp_subflow_ctx(listener->sk));
404 newsk = inet_csk_accept(listener->sk, flags, err, kern);
408 pr_debug("msk=%p, subflow is mptcp=%d", msk, sk_is_mptcp(newsk));
410 if (sk_is_mptcp(newsk)) {
411 struct mptcp_subflow_context *subflow;
412 struct sock *new_mptcp_sock;
413 struct sock *ssk = newsk;
416 subflow = mptcp_subflow_ctx(newsk);
420 new_mptcp_sock = sk_clone_lock(sk, GFP_ATOMIC);
421 if (!new_mptcp_sock) {
425 mptcp_subflow_shutdown(newsk, SHUT_RDWR + 1);
430 mptcp_init_sock(new_mptcp_sock);
432 msk = mptcp_sk(new_mptcp_sock);
433 msk->remote_key = subflow->remote_key;
434 msk->local_key = subflow->local_key;
435 msk->token = subflow->token;
438 mptcp_token_update_accept(newsk, new_mptcp_sock);
440 mptcp_crypto_key_sha(msk->remote_key, NULL, &ack_seq);
441 msk->write_seq = subflow->idsn + 1;
443 msk->ack_seq = ack_seq;
444 subflow->map_seq = ack_seq;
445 subflow->map_subflow_seq = 1;
446 subflow->rel_write_seq = 1;
447 subflow->tcp_sock = ssk;
448 newsk = new_mptcp_sock;
449 mptcp_copy_inaddrs(newsk, ssk);
450 list_add(&subflow->node, &msk->conn_list);
452 /* will be fully established at mptcp_stream_accept()
455 inet_sk_state_store(new_mptcp_sock, TCP_SYN_RECV);
456 bh_unlock_sock(new_mptcp_sock);
464 static void mptcp_destroy(struct sock *sk)
468 static int mptcp_setsockopt(struct sock *sk, int level, int optname,
469 char __user *uoptval, unsigned int optlen)
471 struct mptcp_sock *msk = mptcp_sk(sk);
472 char __kernel *optval;
473 int ret = -EOPNOTSUPP;
474 struct socket *ssock;
476 /* will be treated as __user in tcp_setsockopt */
477 optval = (char __kernel __force *)uoptval;
479 pr_debug("msk=%p", msk);
481 /* @@ the meaning of setsockopt() when the socket is connected and
482 * there are multiple subflows is not defined.
485 ssock = __mptcp_socket_create(msk, MPTCP_SAME_STATE);
486 if (!IS_ERR(ssock)) {
487 pr_debug("subflow=%p", ssock->sk);
488 ret = kernel_setsockopt(ssock, level, optname, optval, optlen);
495 static int mptcp_getsockopt(struct sock *sk, int level, int optname,
496 char __user *uoptval, int __user *uoption)
498 struct mptcp_sock *msk = mptcp_sk(sk);
499 char __kernel *optval;
500 int ret = -EOPNOTSUPP;
501 int __kernel *option;
502 struct socket *ssock;
504 /* will be treated as __user in tcp_getsockopt */
505 optval = (char __kernel __force *)uoptval;
506 option = (int __kernel __force *)uoption;
508 pr_debug("msk=%p", msk);
510 /* @@ the meaning of getsockopt() when the socket is connected and
511 * there are multiple subflows is not defined.
514 ssock = __mptcp_socket_create(msk, MPTCP_SAME_STATE);
515 if (!IS_ERR(ssock)) {
516 pr_debug("subflow=%p", ssock->sk);
517 ret = kernel_getsockopt(ssock, level, optname, optval, option);
524 static int mptcp_get_port(struct sock *sk, unsigned short snum)
526 struct mptcp_sock *msk = mptcp_sk(sk);
527 struct socket *ssock;
529 ssock = __mptcp_nmpc_socket(msk);
530 pr_debug("msk=%p, subflow=%p", msk, ssock);
531 if (WARN_ON_ONCE(!ssock))
534 return inet_csk_get_port(ssock->sk, snum);
537 void mptcp_finish_connect(struct sock *ssk)
539 struct mptcp_subflow_context *subflow;
540 struct mptcp_sock *msk;
544 subflow = mptcp_subflow_ctx(ssk);
546 if (!subflow->mp_capable)
552 pr_debug("msk=%p, token=%u", sk, subflow->token);
554 mptcp_crypto_key_sha(subflow->remote_key, NULL, &ack_seq);
556 subflow->map_seq = ack_seq;
557 subflow->map_subflow_seq = 1;
558 subflow->rel_write_seq = 1;
560 /* the socket is not connected yet, no msk/subflow ops can access/race
561 * accessing the field below
563 WRITE_ONCE(msk->remote_key, subflow->remote_key);
564 WRITE_ONCE(msk->local_key, subflow->local_key);
565 WRITE_ONCE(msk->token, subflow->token);
566 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
567 WRITE_ONCE(msk->ack_seq, ack_seq);
570 static void mptcp_sock_graft(struct sock *sk, struct socket *parent)
572 write_lock_bh(&sk->sk_callback_lock);
573 rcu_assign_pointer(sk->sk_wq, &parent->wq);
574 sk_set_socket(sk, parent);
575 sk->sk_uid = SOCK_INODE(parent)->i_uid;
576 write_unlock_bh(&sk->sk_callback_lock);
579 static struct proto mptcp_prot = {
581 .owner = THIS_MODULE,
582 .init = mptcp_init_sock,
583 .close = mptcp_close,
584 .accept = mptcp_accept,
585 .setsockopt = mptcp_setsockopt,
586 .getsockopt = mptcp_getsockopt,
587 .shutdown = tcp_shutdown,
588 .destroy = mptcp_destroy,
589 .sendmsg = mptcp_sendmsg,
590 .recvmsg = mptcp_recvmsg,
592 .unhash = inet_unhash,
593 .get_port = mptcp_get_port,
594 .obj_size = sizeof(struct mptcp_sock),
598 static int mptcp_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
600 struct mptcp_sock *msk = mptcp_sk(sock->sk);
601 struct socket *ssock;
605 ssock = __mptcp_socket_create(msk, MPTCP_SAME_STATE);
607 err = PTR_ERR(ssock);
611 err = ssock->ops->bind(ssock, uaddr, addr_len);
613 mptcp_copy_inaddrs(sock->sk, ssock->sk);
616 release_sock(sock->sk);
620 static int mptcp_stream_connect(struct socket *sock, struct sockaddr *uaddr,
621 int addr_len, int flags)
623 struct mptcp_sock *msk = mptcp_sk(sock->sk);
624 struct socket *ssock;
628 ssock = __mptcp_socket_create(msk, TCP_SYN_SENT);
630 err = PTR_ERR(ssock);
634 #ifdef CONFIG_TCP_MD5SIG
635 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
638 if (rcu_access_pointer(tcp_sk(ssock->sk)->md5sig_info))
639 mptcp_subflow_ctx(ssock->sk)->request_mptcp = 0;
642 err = ssock->ops->connect(ssock, uaddr, addr_len, flags);
643 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
644 mptcp_copy_inaddrs(sock->sk, ssock->sk);
647 release_sock(sock->sk);
651 static int mptcp_v4_getname(struct socket *sock, struct sockaddr *uaddr,
654 if (sock->sk->sk_prot == &tcp_prot) {
655 /* we are being invoked from __sys_accept4, after
656 * mptcp_accept() has just accepted a non-mp-capable
657 * flow: sk is a tcp_sk, not an mptcp one.
659 * Hand the socket over to tcp so all further socket ops
662 sock->ops = &inet_stream_ops;
665 return inet_getname(sock, uaddr, peer);
668 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
669 static int mptcp_v6_getname(struct socket *sock, struct sockaddr *uaddr,
672 if (sock->sk->sk_prot == &tcpv6_prot) {
673 /* we are being invoked from __sys_accept4 after
674 * mptcp_accept() has accepted a non-mp-capable
675 * subflow: sk is a tcp_sk, not mptcp.
677 * Hand the socket over to tcp so all further
678 * socket ops bypass mptcp.
680 sock->ops = &inet6_stream_ops;
683 return inet6_getname(sock, uaddr, peer);
687 static int mptcp_listen(struct socket *sock, int backlog)
689 struct mptcp_sock *msk = mptcp_sk(sock->sk);
690 struct socket *ssock;
693 pr_debug("msk=%p", msk);
696 ssock = __mptcp_socket_create(msk, TCP_LISTEN);
698 err = PTR_ERR(ssock);
702 err = ssock->ops->listen(ssock, backlog);
703 inet_sk_state_store(sock->sk, inet_sk_state_load(ssock->sk));
705 mptcp_copy_inaddrs(sock->sk, ssock->sk);
708 release_sock(sock->sk);
712 static bool is_tcp_proto(const struct proto *p)
714 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
715 return p == &tcp_prot || p == &tcpv6_prot;
717 return p == &tcp_prot;
721 static int mptcp_stream_accept(struct socket *sock, struct socket *newsock,
722 int flags, bool kern)
724 struct mptcp_sock *msk = mptcp_sk(sock->sk);
725 struct socket *ssock;
728 pr_debug("msk=%p", msk);
731 if (sock->sk->sk_state != TCP_LISTEN)
734 ssock = __mptcp_nmpc_socket(msk);
738 sock_hold(ssock->sk);
739 release_sock(sock->sk);
741 err = ssock->ops->accept(sock, newsock, flags, kern);
742 if (err == 0 && !is_tcp_proto(newsock->sk->sk_prot)) {
743 struct mptcp_sock *msk = mptcp_sk(newsock->sk);
744 struct mptcp_subflow_context *subflow;
746 /* set ssk->sk_socket of accept()ed flows to mptcp socket.
747 * This is needed so NOSPACE flag can be set from tcp stack.
749 list_for_each_entry(subflow, &msk->conn_list, node) {
750 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
753 mptcp_sock_graft(ssk, newsock);
756 inet_sk_state_store(newsock->sk, TCP_ESTABLISHED);
763 release_sock(sock->sk);
767 static __poll_t mptcp_poll(struct file *file, struct socket *sock,
768 struct poll_table_struct *wait)
775 static int mptcp_shutdown(struct socket *sock, int how)
777 struct mptcp_sock *msk = mptcp_sk(sock->sk);
778 struct mptcp_subflow_context *subflow;
781 pr_debug("sk=%p, how=%d", msk, how);
785 if (how == SHUT_WR || how == SHUT_RDWR)
786 inet_sk_state_store(sock->sk, TCP_FIN_WAIT1);
790 if ((how & ~SHUTDOWN_MASK) || !how) {
795 if (sock->state == SS_CONNECTING) {
796 if ((1 << sock->sk->sk_state) &
797 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
798 sock->state = SS_DISCONNECTING;
800 sock->state = SS_CONNECTED;
803 mptcp_for_each_subflow(msk, subflow) {
804 struct sock *tcp_sk = mptcp_subflow_tcp_sock(subflow);
806 mptcp_subflow_shutdown(tcp_sk, how);
810 release_sock(sock->sk);
815 static struct proto_ops mptcp_stream_ops;
817 static struct inet_protosw mptcp_protosw = {
819 .protocol = IPPROTO_MPTCP,
821 .ops = &mptcp_stream_ops,
822 .flags = INET_PROTOSW_ICSK,
825 void __init mptcp_init(void)
827 mptcp_prot.h.hashinfo = tcp_prot.h.hashinfo;
828 mptcp_stream_ops = inet_stream_ops;
829 mptcp_stream_ops.bind = mptcp_bind;
830 mptcp_stream_ops.connect = mptcp_stream_connect;
831 mptcp_stream_ops.poll = mptcp_poll;
832 mptcp_stream_ops.accept = mptcp_stream_accept;
833 mptcp_stream_ops.getname = mptcp_v4_getname;
834 mptcp_stream_ops.listen = mptcp_listen;
835 mptcp_stream_ops.shutdown = mptcp_shutdown;
837 mptcp_subflow_init();
839 if (proto_register(&mptcp_prot, 1) != 0)
840 panic("Failed to register MPTCP proto.\n");
842 inet_register_protosw(&mptcp_protosw);
845 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
846 static struct proto_ops mptcp_v6_stream_ops;
847 static struct proto mptcp_v6_prot;
849 static void mptcp_v6_destroy(struct sock *sk)
852 inet6_destroy_sock(sk);
855 static struct inet_protosw mptcp_v6_protosw = {
857 .protocol = IPPROTO_MPTCP,
858 .prot = &mptcp_v6_prot,
859 .ops = &mptcp_v6_stream_ops,
860 .flags = INET_PROTOSW_ICSK,
863 int mptcpv6_init(void)
867 mptcp_v6_prot = mptcp_prot;
868 strcpy(mptcp_v6_prot.name, "MPTCPv6");
869 mptcp_v6_prot.slab = NULL;
870 mptcp_v6_prot.destroy = mptcp_v6_destroy;
871 mptcp_v6_prot.obj_size = sizeof(struct mptcp_sock) +
872 sizeof(struct ipv6_pinfo);
874 err = proto_register(&mptcp_v6_prot, 1);
878 mptcp_v6_stream_ops = inet6_stream_ops;
879 mptcp_v6_stream_ops.bind = mptcp_bind;
880 mptcp_v6_stream_ops.connect = mptcp_stream_connect;
881 mptcp_v6_stream_ops.poll = mptcp_poll;
882 mptcp_v6_stream_ops.accept = mptcp_stream_accept;
883 mptcp_v6_stream_ops.getname = mptcp_v6_getname;
884 mptcp_v6_stream_ops.listen = mptcp_listen;
885 mptcp_v6_stream_ops.shutdown = mptcp_shutdown;
887 err = inet6_register_protosw(&mptcp_v6_protosw);
889 proto_unregister(&mptcp_v6_prot);