1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PF_INET protocol family socket handler.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Florian La Roche, <flla@stud.uni-sb.de>
12 * Alan Cox, <A.Cox@swansea.ac.uk>
14 * Changes (see also sock.c)
17 * Karl Knutson : Socket protocol table
18 * A.N.Kuznetsov : Socket death error in accept().
19 * John Richardson : Fix non blocking error in connect()
20 * so sockets that fail to connect
21 * don't return -EINPROGRESS.
22 * Alan Cox : Asynchronous I/O support
23 * Alan Cox : Keep correct socket pointer on sock
26 * Alan Cox : Semantics of SO_LINGER aren't state
27 * moved to close when you look carefully.
28 * With this fixed and the accept bug fixed
29 * some RPC stuff seems happier.
30 * Niibe Yutaka : 4.4BSD style write async I/O
32 * Tony Gale : Fixed reuse semantics.
33 * Alan Cox : bind() shouldn't abort existing but dead
34 * sockets. Stops FTP netin:.. I hope.
35 * Alan Cox : bind() works correctly for RAW sockets.
36 * Note that FreeBSD at least was broken
37 * in this respect so be careful with
38 * compatibility tests...
39 * Alan Cox : routing cache support
40 * Alan Cox : memzero the socket structure for
42 * Matt Day : nonblock connect error handler
43 * Alan Cox : Allow large numbers of pending sockets
44 * (eg for big web sites), but only if
45 * specifically application requested.
46 * Alan Cox : New buffering throughout IP. Used
48 * Alan Cox : New buffering now used smartly.
49 * Alan Cox : BSD rather than common sense
50 * interpretation of listen.
51 * Germano Caronni : Assorted small races.
52 * Alan Cox : sendmsg/recvmsg basic support.
53 * Alan Cox : Only sendmsg/recvmsg now supported.
54 * Alan Cox : Locked down bind (see security list).
55 * Alan Cox : Loosened bind a little.
56 * Mike McLagan : ADD/DEL DLCI Ioctls
57 * Willy Konynenberg : Transparent proxying support.
58 * David S. Miller : New socket lookup architecture.
59 * Some other random speedups.
60 * Cyrus Durgin : Cleaned up file for kmod hacks.
61 * Andi Kleen : Fix inet_stream_connect TCP race.
64 #define pr_fmt(fmt) "IPv4: " fmt
66 #include <linux/err.h>
67 #include <linux/errno.h>
68 #include <linux/types.h>
69 #include <linux/socket.h>
71 #include <linux/kernel.h>
72 #include <linux/kmod.h>
73 #include <linux/sched.h>
74 #include <linux/timer.h>
75 #include <linux/string.h>
76 #include <linux/sockios.h>
77 #include <linux/net.h>
78 #include <linux/capability.h>
79 #include <linux/fcntl.h>
81 #include <linux/interrupt.h>
82 #include <linux/stat.h>
83 #include <linux/init.h>
84 #include <linux/poll.h>
85 #include <linux/netfilter_ipv4.h>
86 #include <linux/random.h>
87 #include <linux/slab.h>
89 #include <linux/uaccess.h>
91 #include <linux/inet.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/netdevice.h>
95 #include <net/checksum.h>
97 #include <net/protocol.h>
99 #include <net/route.h>
100 #include <net/ip_fib.h>
101 #include <net/inet_connection_sock.h>
104 #include <net/udplite.h>
105 #include <net/ping.h>
106 #include <linux/skbuff.h>
107 #include <net/sock.h>
109 #include <net/icmp.h>
110 #include <net/inet_common.h>
111 #include <net/ip_tunnels.h>
112 #include <net/xfrm.h>
113 #include <net/net_namespace.h>
114 #include <net/secure_seq.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
118 #include <net/l3mdev.h>
119 #include <net/compat.h>
121 #include <trace/events/sock.h>
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
129 /* New destruction routine */
131 void inet_sock_destruct(struct sock *sk)
133 struct inet_sock *inet = inet_sk(sk);
135 __skb_queue_purge(&sk->sk_receive_queue);
136 if (sk->sk_rx_skb_cache) {
137 __kfree_skb(sk->sk_rx_skb_cache);
138 sk->sk_rx_skb_cache = NULL;
140 __skb_queue_purge(&sk->sk_error_queue);
144 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 pr_err("Attempt to release TCP socket in state %d %p\n",
149 if (!sock_flag(sk, SOCK_DEAD)) {
150 pr_err("Attempt to release alive inet socket %p\n", sk);
154 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
155 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
156 WARN_ON(sk->sk_wmem_queued);
157 WARN_ON(sk->sk_forward_alloc);
159 kfree(rcu_dereference_protected(inet->inet_opt, 1));
160 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
161 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
162 sk_refcnt_debug_dec(sk);
164 EXPORT_SYMBOL(inet_sock_destruct);
167 * The routines beyond this point handle the behaviour of an AF_INET
168 * socket object. Mostly it punts to the subprotocols of IP to do
173 * Automatically bind an unbound socket.
176 static int inet_autobind(struct sock *sk)
178 struct inet_sock *inet;
179 /* We may need to bind the socket. */
182 if (!inet->inet_num) {
183 if (sk->sk_prot->get_port(sk, 0)) {
187 inet->inet_sport = htons(inet->inet_num);
194 * Move a socket into listening state.
196 int inet_listen(struct socket *sock, int backlog)
198 struct sock *sk = sock->sk;
199 unsigned char old_state;
200 int err, tcp_fastopen;
205 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
208 old_state = sk->sk_state;
209 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
212 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
213 /* Really, if the socket is already in listen state
214 * we can only allow the backlog to be adjusted.
216 if (old_state != TCP_LISTEN) {
217 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
218 * Note that only TCP sockets (SOCK_STREAM) will reach here.
219 * Also fastopen backlog may already been set via the option
220 * because the socket was in TCP_LISTEN state previously but
221 * was shutdown() rather than close().
223 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
224 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
225 (tcp_fastopen & TFO_SERVER_ENABLE) &&
226 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
227 fastopen_queue_tune(sk, backlog);
228 tcp_fastopen_init_key_once(sock_net(sk));
231 err = inet_csk_listen_start(sk, backlog);
234 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
242 EXPORT_SYMBOL(inet_listen);
245 * Create an inet socket.
248 static int inet_create(struct net *net, struct socket *sock, int protocol,
252 struct inet_protosw *answer;
253 struct inet_sock *inet;
254 struct proto *answer_prot;
255 unsigned char answer_flags;
256 int try_loading_module = 0;
259 if (protocol < 0 || protocol >= IPPROTO_MAX)
262 sock->state = SS_UNCONNECTED;
264 /* Look for the requested type/protocol pair. */
266 err = -ESOCKTNOSUPPORT;
268 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
271 /* Check the non-wild match. */
272 if (protocol == answer->protocol) {
273 if (protocol != IPPROTO_IP)
276 /* Check for the two wild cases. */
277 if (IPPROTO_IP == protocol) {
278 protocol = answer->protocol;
281 if (IPPROTO_IP == answer->protocol)
284 err = -EPROTONOSUPPORT;
288 if (try_loading_module < 2) {
291 * Be more specific, e.g. net-pf-2-proto-132-type-1
292 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
294 if (++try_loading_module == 1)
295 request_module("net-pf-%d-proto-%d-type-%d",
296 PF_INET, protocol, sock->type);
298 * Fall back to generic, e.g. net-pf-2-proto-132
299 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
302 request_module("net-pf-%d-proto-%d",
304 goto lookup_protocol;
310 if (sock->type == SOCK_RAW && !kern &&
311 !ns_capable(net->user_ns, CAP_NET_RAW))
314 sock->ops = answer->ops;
315 answer_prot = answer->prot;
316 answer_flags = answer->flags;
319 WARN_ON(!answer_prot->slab);
322 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
327 if (INET_PROTOSW_REUSE & answer_flags)
328 sk->sk_reuse = SK_CAN_REUSE;
330 if (INET_PROTOSW_ICSK & answer_flags)
331 inet_init_csk_locks(sk);
334 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
338 if (SOCK_RAW == sock->type) {
339 inet->inet_num = protocol;
340 if (IPPROTO_RAW == protocol)
344 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
345 inet->pmtudisc = IP_PMTUDISC_DONT;
347 inet->pmtudisc = IP_PMTUDISC_WANT;
351 sock_init_data(sock, sk);
353 sk->sk_destruct = inet_sock_destruct;
354 sk->sk_protocol = protocol;
355 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
362 inet->mc_list = NULL;
365 sk_refcnt_debug_inc(sk);
367 if (inet->inet_num) {
368 /* It assumes that any protocol which allows
369 * the user to assign a number at socket
370 * creation time automatically
373 inet->inet_sport = htons(inet->inet_num);
374 /* Add to protocol hash chains. */
375 err = sk->sk_prot->hash(sk);
377 sk_common_release(sk);
382 if (sk->sk_prot->init) {
383 err = sk->sk_prot->init(sk);
385 sk_common_release(sk);
391 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
393 sk_common_release(sk);
406 * The peer socket should always be NULL (or else). When we call this
407 * function we are destroying the object and from then on nobody
408 * should refer to it.
410 int inet_release(struct socket *sock)
412 struct sock *sk = sock->sk;
417 if (!sk->sk_kern_sock)
418 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
420 /* Applications forget to leave groups before exiting */
421 ip_mc_drop_socket(sk);
423 /* If linger is set, we don't return until the close
424 * is complete. Otherwise we return immediately. The
425 * actually closing is done the same either way.
427 * If the close is due to the process exiting, we never
431 if (sock_flag(sk, SOCK_LINGER) &&
432 !(current->flags & PF_EXITING))
433 timeout = sk->sk_lingertime;
434 sk->sk_prot->close(sk, timeout);
439 EXPORT_SYMBOL(inet_release);
441 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
443 struct sock *sk = sock->sk;
446 /* If the socket has its own bind function then use it. (RAW) */
447 if (sk->sk_prot->bind) {
448 return sk->sk_prot->bind(sk, uaddr, addr_len);
450 if (addr_len < sizeof(struct sockaddr_in))
453 /* BPF prog is run before any checks are done so that if the prog
454 * changes context in a wrong way it will be caught.
456 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
460 return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
462 EXPORT_SYMBOL(inet_bind);
464 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
467 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
468 struct inet_sock *inet = inet_sk(sk);
469 struct net *net = sock_net(sk);
472 u32 tb_id = RT_TABLE_LOCAL;
475 if (addr->sin_family != AF_INET) {
476 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
477 * only if s_addr is INADDR_ANY.
480 if (addr->sin_family != AF_UNSPEC ||
481 addr->sin_addr.s_addr != htonl(INADDR_ANY))
485 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
486 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
488 /* Not specified by any standard per-se, however it breaks too
489 * many applications when removed. It is unfortunate since
490 * allowing applications to make a non-local bind solves
491 * several problems with systems using dynamic addressing.
492 * (ie. your servers still start up even if your ISDN link
493 * is temporarily down)
495 err = -EADDRNOTAVAIL;
496 if (!inet_can_nonlocal_bind(net, inet) &&
497 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
498 chk_addr_ret != RTN_LOCAL &&
499 chk_addr_ret != RTN_MULTICAST &&
500 chk_addr_ret != RTN_BROADCAST)
503 snum = ntohs(addr->sin_port);
505 if (snum && inet_port_requires_bind_service(net, snum) &&
506 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
509 /* We keep a pair of addresses. rcv_saddr is the one
510 * used by hash lookups, and saddr is used for transmit.
512 * In the BSD API these are the same except where it
513 * would be illegal to use them (multicast/broadcast) in
514 * which case the sending device address is used.
516 if (flags & BIND_WITH_LOCK)
519 /* Check these errors (active socket, double bind). */
521 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
522 goto out_release_sock;
524 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
525 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
526 inet->inet_saddr = 0; /* Use device */
528 /* Make sure we are allowed to bind here. */
529 if (snum || !(inet->bind_address_no_port ||
530 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
531 if (sk->sk_prot->get_port(sk, snum)) {
532 inet->inet_saddr = inet->inet_rcv_saddr = 0;
534 goto out_release_sock;
536 if (!(flags & BIND_FROM_BPF)) {
537 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
539 inet->inet_saddr = inet->inet_rcv_saddr = 0;
540 goto out_release_sock;
545 if (inet->inet_rcv_saddr)
546 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
548 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
549 inet->inet_sport = htons(inet->inet_num);
550 inet->inet_daddr = 0;
551 inet->inet_dport = 0;
555 if (flags & BIND_WITH_LOCK)
561 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
562 int addr_len, int flags)
564 struct sock *sk = sock->sk;
567 if (addr_len < sizeof(uaddr->sa_family))
569 if (uaddr->sa_family == AF_UNSPEC)
570 return sk->sk_prot->disconnect(sk, flags);
572 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
573 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
578 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
580 return sk->sk_prot->connect(sk, uaddr, addr_len);
582 EXPORT_SYMBOL(inet_dgram_connect);
584 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
586 DEFINE_WAIT_FUNC(wait, woken_wake_function);
588 add_wait_queue(sk_sleep(sk), &wait);
589 sk->sk_write_pending += writebias;
590 sk->sk_wait_pending++;
592 /* Basic assumption: if someone sets sk->sk_err, he _must_
593 * change state of the socket from TCP_SYN_*.
594 * Connect() does not allow to get error notifications
595 * without closing the socket.
597 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
599 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
601 if (signal_pending(current) || !timeo)
604 remove_wait_queue(sk_sleep(sk), &wait);
605 sk->sk_write_pending -= writebias;
606 sk->sk_wait_pending--;
611 * Connect to a remote host. There is regrettably still a little
612 * TCP 'magic' in here.
614 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
615 int addr_len, int flags, int is_sendmsg)
617 struct sock *sk = sock->sk;
622 * uaddr can be NULL and addr_len can be 0 if:
623 * sk is a TCP fastopen active socket and
624 * TCP_FASTOPEN_CONNECT sockopt is set and
625 * we already have a valid cookie for this socket.
626 * In this case, user can call write() after connect().
627 * write() will invoke tcp_sendmsg_fastopen() which calls
628 * __inet_stream_connect().
631 if (addr_len < sizeof(uaddr->sa_family))
634 if (uaddr->sa_family == AF_UNSPEC) {
635 err = sk->sk_prot->disconnect(sk, flags);
636 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
641 switch (sock->state) {
649 if (inet_sk(sk)->defer_connect)
650 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
653 /* Fall out of switch with err, set for this state */
657 if (sk->sk_state != TCP_CLOSE)
660 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
661 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
666 err = sk->sk_prot->connect(sk, uaddr, addr_len);
670 sock->state = SS_CONNECTING;
672 if (!err && inet_sk(sk)->defer_connect)
675 /* Just entered SS_CONNECTING state; the only
676 * difference is that return value in non-blocking
677 * case is EINPROGRESS, rather than EALREADY.
683 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
685 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
686 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
687 tcp_sk(sk)->fastopen_req &&
688 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
690 /* Error code is set above */
691 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
694 err = sock_intr_errno(timeo);
695 if (signal_pending(current))
699 /* Connection was closed by RST, timeout, ICMP error
700 * or another process disconnected us.
702 if (sk->sk_state == TCP_CLOSE)
705 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
706 * and error was received after socket entered established state.
707 * Hence, it is handled normally after connect() return successfully.
710 sock->state = SS_CONNECTED;
716 err = sock_error(sk) ? : -ECONNABORTED;
717 sock->state = SS_UNCONNECTED;
718 if (sk->sk_prot->disconnect(sk, flags))
719 sock->state = SS_DISCONNECTING;
722 EXPORT_SYMBOL(__inet_stream_connect);
724 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
725 int addr_len, int flags)
730 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
731 release_sock(sock->sk);
734 EXPORT_SYMBOL(inet_stream_connect);
737 * Accept a pending connection. The TCP layer now gives BSD semantics.
740 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
743 struct sock *sk1 = sock->sk;
745 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
752 sock_rps_record_flow(sk2);
753 WARN_ON(!((1 << sk2->sk_state) &
754 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
755 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
757 sock_graft(sk2, newsock);
759 newsock->state = SS_CONNECTED;
765 EXPORT_SYMBOL(inet_accept);
768 * This does both peername and sockname.
770 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
773 struct sock *sk = sock->sk;
774 struct inet_sock *inet = inet_sk(sk);
775 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
777 sin->sin_family = AF_INET;
779 if (!inet->inet_dport ||
780 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
783 sin->sin_port = inet->inet_dport;
784 sin->sin_addr.s_addr = inet->inet_daddr;
786 __be32 addr = inet->inet_rcv_saddr;
788 addr = inet->inet_saddr;
789 sin->sin_port = inet->inet_sport;
790 sin->sin_addr.s_addr = addr;
792 if (cgroup_bpf_enabled)
793 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
794 peer ? BPF_CGROUP_INET4_GETPEERNAME :
795 BPF_CGROUP_INET4_GETSOCKNAME,
797 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
800 EXPORT_SYMBOL(inet_getname);
802 int inet_send_prepare(struct sock *sk)
804 sock_rps_record_flow(sk);
806 /* We may need to bind the socket. */
807 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
813 EXPORT_SYMBOL_GPL(inet_send_prepare);
815 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
817 struct sock *sk = sock->sk;
819 if (unlikely(inet_send_prepare(sk)))
822 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
825 EXPORT_SYMBOL(inet_sendmsg);
827 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
828 size_t size, int flags)
830 struct sock *sk = sock->sk;
832 if (unlikely(inet_send_prepare(sk)))
835 if (sk->sk_prot->sendpage)
836 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
837 return sock_no_sendpage(sock, page, offset, size, flags);
839 EXPORT_SYMBOL(inet_sendpage);
841 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
842 size_t, int, int, int *));
843 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
846 struct sock *sk = sock->sk;
850 if (likely(!(flags & MSG_ERRQUEUE)))
851 sock_rps_record_flow(sk);
853 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
854 sk, msg, size, flags & MSG_DONTWAIT,
855 flags & ~MSG_DONTWAIT, &addr_len);
857 msg->msg_namelen = addr_len;
860 EXPORT_SYMBOL(inet_recvmsg);
862 int inet_shutdown(struct socket *sock, int how)
864 struct sock *sk = sock->sk;
867 /* This should really check to make sure
868 * the socket is a TCP socket. (WHY AC...)
870 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
873 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
877 if (sock->state == SS_CONNECTING) {
878 if ((1 << sk->sk_state) &
879 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
880 sock->state = SS_DISCONNECTING;
882 sock->state = SS_CONNECTED;
885 switch (sk->sk_state) {
888 /* Hack to wake up other listeners, who can poll for
889 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
892 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
893 if (sk->sk_prot->shutdown)
894 sk->sk_prot->shutdown(sk, how);
897 /* Remaining two branches are temporary solution for missing
898 * close() in multithreaded environment. It is _not_ a good idea,
899 * but we have no choice until close() is repaired at VFS level.
902 if (!(how & RCV_SHUTDOWN))
906 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
907 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
911 /* Wake up anyone sleeping in poll. */
912 sk->sk_state_change(sk);
916 EXPORT_SYMBOL(inet_shutdown);
919 * ioctl() calls you can issue on an INET socket. Most of these are
920 * device configuration and stuff and very rarely used. Some ioctls
921 * pass on to the socket itself.
923 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
924 * loads the devconfigure module does its configuring and unloads it.
925 * There's a good 20K of config code hanging around the kernel.
928 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
930 struct sock *sk = sock->sk;
932 struct net *net = sock_net(sk);
933 void __user *p = (void __user *)arg;
940 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
942 err = ip_rt_ioctl(net, cmd, &rt);
950 err = arp_ioctl(net, cmd, (void __user *)arg);
957 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
959 err = devinet_ioctl(net, cmd, &ifr);
960 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
970 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
972 err = devinet_ioctl(net, cmd, &ifr);
975 if (sk->sk_prot->ioctl)
976 err = sk->sk_prot->ioctl(sk, cmd, arg);
983 EXPORT_SYMBOL(inet_ioctl);
986 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
987 struct compat_rtentry __user *ur)
992 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
993 3 * sizeof(struct sockaddr)) ||
994 get_user(rt.rt_flags, &ur->rt_flags) ||
995 get_user(rt.rt_metric, &ur->rt_metric) ||
996 get_user(rt.rt_mtu, &ur->rt_mtu) ||
997 get_user(rt.rt_window, &ur->rt_window) ||
998 get_user(rt.rt_irtt, &ur->rt_irtt) ||
999 get_user(rtdev, &ur->rt_dev))
1002 rt.rt_dev = compat_ptr(rtdev);
1003 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1006 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1008 void __user *argp = compat_ptr(arg);
1009 struct sock *sk = sock->sk;
1014 return inet_compat_routing_ioctl(sk, cmd, argp);
1016 if (!sk->sk_prot->compat_ioctl)
1017 return -ENOIOCTLCMD;
1018 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1021 #endif /* CONFIG_COMPAT */
1023 const struct proto_ops inet_stream_ops = {
1025 .owner = THIS_MODULE,
1026 .release = inet_release,
1028 .connect = inet_stream_connect,
1029 .socketpair = sock_no_socketpair,
1030 .accept = inet_accept,
1031 .getname = inet_getname,
1033 .ioctl = inet_ioctl,
1034 .gettstamp = sock_gettstamp,
1035 .listen = inet_listen,
1036 .shutdown = inet_shutdown,
1037 .setsockopt = sock_common_setsockopt,
1038 .getsockopt = sock_common_getsockopt,
1039 .sendmsg = inet_sendmsg,
1040 .recvmsg = inet_recvmsg,
1044 .sendpage = inet_sendpage,
1045 .splice_read = tcp_splice_read,
1046 .read_sock = tcp_read_sock,
1047 .sendmsg_locked = tcp_sendmsg_locked,
1048 .sendpage_locked = tcp_sendpage_locked,
1049 .peek_len = tcp_peek_len,
1050 #ifdef CONFIG_COMPAT
1051 .compat_ioctl = inet_compat_ioctl,
1053 .set_rcvlowat = tcp_set_rcvlowat,
1055 EXPORT_SYMBOL(inet_stream_ops);
1057 const struct proto_ops inet_dgram_ops = {
1059 .owner = THIS_MODULE,
1060 .release = inet_release,
1062 .connect = inet_dgram_connect,
1063 .socketpair = sock_no_socketpair,
1064 .accept = sock_no_accept,
1065 .getname = inet_getname,
1067 .ioctl = inet_ioctl,
1068 .gettstamp = sock_gettstamp,
1069 .listen = sock_no_listen,
1070 .shutdown = inet_shutdown,
1071 .setsockopt = sock_common_setsockopt,
1072 .getsockopt = sock_common_getsockopt,
1073 .sendmsg = inet_sendmsg,
1074 .recvmsg = inet_recvmsg,
1075 .mmap = sock_no_mmap,
1076 .sendpage = inet_sendpage,
1077 .set_peek_off = sk_set_peek_off,
1078 #ifdef CONFIG_COMPAT
1079 .compat_ioctl = inet_compat_ioctl,
1082 EXPORT_SYMBOL(inet_dgram_ops);
1085 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1088 static const struct proto_ops inet_sockraw_ops = {
1090 .owner = THIS_MODULE,
1091 .release = inet_release,
1093 .connect = inet_dgram_connect,
1094 .socketpair = sock_no_socketpair,
1095 .accept = sock_no_accept,
1096 .getname = inet_getname,
1097 .poll = datagram_poll,
1098 .ioctl = inet_ioctl,
1099 .gettstamp = sock_gettstamp,
1100 .listen = sock_no_listen,
1101 .shutdown = inet_shutdown,
1102 .setsockopt = sock_common_setsockopt,
1103 .getsockopt = sock_common_getsockopt,
1104 .sendmsg = inet_sendmsg,
1105 .recvmsg = inet_recvmsg,
1106 .mmap = sock_no_mmap,
1107 .sendpage = inet_sendpage,
1108 #ifdef CONFIG_COMPAT
1109 .compat_ioctl = inet_compat_ioctl,
1113 static const struct net_proto_family inet_family_ops = {
1115 .create = inet_create,
1116 .owner = THIS_MODULE,
1119 /* Upon startup we insert all the elements in inetsw_array[] into
1120 * the linked list inetsw.
1122 static struct inet_protosw inetsw_array[] =
1125 .type = SOCK_STREAM,
1126 .protocol = IPPROTO_TCP,
1128 .ops = &inet_stream_ops,
1129 .flags = INET_PROTOSW_PERMANENT |
1135 .protocol = IPPROTO_UDP,
1137 .ops = &inet_dgram_ops,
1138 .flags = INET_PROTOSW_PERMANENT,
1143 .protocol = IPPROTO_ICMP,
1145 .ops = &inet_sockraw_ops,
1146 .flags = INET_PROTOSW_REUSE,
1151 .protocol = IPPROTO_IP, /* wild card */
1153 .ops = &inet_sockraw_ops,
1154 .flags = INET_PROTOSW_REUSE,
1158 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1160 void inet_register_protosw(struct inet_protosw *p)
1162 struct list_head *lh;
1163 struct inet_protosw *answer;
1164 int protocol = p->protocol;
1165 struct list_head *last_perm;
1167 spin_lock_bh(&inetsw_lock);
1169 if (p->type >= SOCK_MAX)
1172 /* If we are trying to override a permanent protocol, bail. */
1173 last_perm = &inetsw[p->type];
1174 list_for_each(lh, &inetsw[p->type]) {
1175 answer = list_entry(lh, struct inet_protosw, list);
1176 /* Check only the non-wild match. */
1177 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1179 if (protocol == answer->protocol)
1184 /* Add the new entry after the last permanent entry if any, so that
1185 * the new entry does not override a permanent entry when matched with
1186 * a wild-card protocol. But it is allowed to override any existing
1187 * non-permanent entry. This means that when we remove this entry, the
1188 * system automatically returns to the old behavior.
1190 list_add_rcu(&p->list, last_perm);
1192 spin_unlock_bh(&inetsw_lock);
1197 pr_err("Attempt to override permanent protocol %d\n", protocol);
1201 pr_err("Ignoring attempt to register invalid socket type %d\n",
1205 EXPORT_SYMBOL(inet_register_protosw);
1207 void inet_unregister_protosw(struct inet_protosw *p)
1209 if (INET_PROTOSW_PERMANENT & p->flags) {
1210 pr_err("Attempt to unregister permanent protocol %d\n",
1213 spin_lock_bh(&inetsw_lock);
1214 list_del_rcu(&p->list);
1215 spin_unlock_bh(&inetsw_lock);
1220 EXPORT_SYMBOL(inet_unregister_protosw);
1222 static int inet_sk_reselect_saddr(struct sock *sk)
1224 struct inet_sock *inet = inet_sk(sk);
1225 __be32 old_saddr = inet->inet_saddr;
1226 __be32 daddr = inet->inet_daddr;
1230 struct ip_options_rcu *inet_opt;
1232 inet_opt = rcu_dereference_protected(inet->inet_opt,
1233 lockdep_sock_is_held(sk));
1234 if (inet_opt && inet_opt->opt.srr)
1235 daddr = inet_opt->opt.faddr;
1237 /* Query new route. */
1238 fl4 = &inet->cork.fl.u.ip4;
1239 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1240 sk->sk_bound_dev_if, sk->sk_protocol,
1241 inet->inet_sport, inet->inet_dport, sk);
1245 sk_setup_caps(sk, &rt->dst);
1247 new_saddr = fl4->saddr;
1249 if (new_saddr == old_saddr)
1252 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1253 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1254 __func__, &old_saddr, &new_saddr);
1257 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1260 * XXX The only one ugly spot where we need to
1261 * XXX really change the sockets identity after
1262 * XXX it has entered the hashes. -DaveM
1264 * Besides that, it does not check for connection
1265 * uniqueness. Wait for troubles.
1267 return __sk_prot_rehash(sk);
1270 int inet_sk_rebuild_header(struct sock *sk)
1272 struct inet_sock *inet = inet_sk(sk);
1273 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1275 struct ip_options_rcu *inet_opt;
1279 /* Route is OK, nothing to do. */
1285 inet_opt = rcu_dereference(inet->inet_opt);
1286 daddr = inet->inet_daddr;
1287 if (inet_opt && inet_opt->opt.srr)
1288 daddr = inet_opt->opt.faddr;
1290 fl4 = &inet->cork.fl.u.ip4;
1291 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1292 inet->inet_dport, inet->inet_sport,
1293 sk->sk_protocol, RT_CONN_FLAGS(sk),
1294 sk->sk_bound_dev_if);
1297 sk_setup_caps(sk, &rt->dst);
1301 /* Routing failed... */
1302 sk->sk_route_caps = 0;
1304 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1305 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1307 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1308 sk->sk_state != TCP_SYN_SENT ||
1309 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1310 (err = inet_sk_reselect_saddr(sk)) != 0)
1311 sk->sk_err_soft = -err;
1316 EXPORT_SYMBOL(inet_sk_rebuild_header);
1318 void inet_sk_set_state(struct sock *sk, int state)
1320 trace_inet_sock_set_state(sk, sk->sk_state, state);
1321 sk->sk_state = state;
1323 EXPORT_SYMBOL(inet_sk_set_state);
1325 void inet_sk_state_store(struct sock *sk, int newstate)
1327 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1328 smp_store_release(&sk->sk_state, newstate);
1331 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1332 netdev_features_t features)
1334 bool udpfrag = false, fixedid = false, gso_partial, encap;
1335 struct sk_buff *segs = ERR_PTR(-EINVAL);
1336 const struct net_offload *ops;
1337 unsigned int offset = 0;
1344 skb_reset_network_header(skb);
1345 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1346 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1351 if (ihl < sizeof(*iph))
1354 id = ntohs(iph->id);
1355 proto = iph->protocol;
1357 /* Warning: after this point, iph might be no longer valid */
1358 if (unlikely(!pskb_may_pull(skb, ihl)))
1360 __skb_pull(skb, ihl);
1362 encap = SKB_GSO_CB(skb)->encap_level > 0;
1364 features &= skb->dev->hw_enc_features;
1365 SKB_GSO_CB(skb)->encap_level += ihl;
1367 skb_reset_transport_header(skb);
1369 segs = ERR_PTR(-EPROTONOSUPPORT);
1371 if (!skb->encapsulation || encap) {
1372 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1373 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1375 /* fixed ID is invalid if DF bit is not set */
1376 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1380 ops = rcu_dereference(inet_offloads[proto]);
1381 if (likely(ops && ops->callbacks.gso_segment)) {
1382 segs = ops->callbacks.gso_segment(skb, features);
1384 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1387 if (IS_ERR_OR_NULL(segs))
1390 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1394 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1396 iph->frag_off = htons(offset >> 3);
1398 iph->frag_off |= htons(IP_MF);
1399 offset += skb->len - nhoff - ihl;
1400 tot_len = skb->len - nhoff;
1401 } else if (skb_is_gso(skb)) {
1403 iph->id = htons(id);
1404 id += skb_shinfo(skb)->gso_segs;
1408 tot_len = skb_shinfo(skb)->gso_size +
1409 SKB_GSO_CB(skb)->data_offset +
1410 skb->head - (unsigned char *)iph;
1412 tot_len = skb->len - nhoff;
1415 iph->id = htons(id++);
1416 tot_len = skb->len - nhoff;
1418 iph->tot_len = htons(tot_len);
1421 skb_reset_inner_headers(skb);
1422 skb->network_header = (u8 *)iph - skb->head;
1423 skb_reset_mac_len(skb);
1424 } while ((skb = skb->next));
1429 EXPORT_SYMBOL(inet_gso_segment);
1431 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1432 netdev_features_t features)
1434 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1435 return ERR_PTR(-EINVAL);
1437 return inet_gso_segment(skb, features);
1440 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1442 const struct net_offload *ops;
1443 struct sk_buff *pp = NULL;
1444 const struct iphdr *iph;
1452 off = skb_gro_offset(skb);
1453 hlen = off + sizeof(*iph);
1454 iph = skb_gro_header_fast(skb, off);
1455 if (skb_gro_header_hard(skb, hlen)) {
1456 iph = skb_gro_header_slow(skb, hlen, off);
1461 proto = iph->protocol;
1464 ops = rcu_dereference(inet_offloads[proto]);
1465 if (!ops || !ops->callbacks.gro_receive)
1468 if (*(u8 *)iph != 0x45)
1471 if (ip_is_fragment(iph))
1474 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1477 id = ntohl(*(__be32 *)&iph->id);
1478 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1481 list_for_each_entry(p, head, list) {
1485 if (!NAPI_GRO_CB(p)->same_flow)
1488 iph2 = (struct iphdr *)(p->data + off);
1489 /* The above works because, with the exception of the top
1490 * (inner most) layer, we only aggregate pkts with the same
1491 * hdr length so all the hdrs we'll need to verify will start
1492 * at the same offset.
1494 if ((iph->protocol ^ iph2->protocol) |
1495 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1496 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1497 NAPI_GRO_CB(p)->same_flow = 0;
1501 /* All fields must match except length and checksum. */
1502 NAPI_GRO_CB(p)->flush |=
1503 (iph->ttl ^ iph2->ttl) |
1504 (iph->tos ^ iph2->tos) |
1505 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1507 NAPI_GRO_CB(p)->flush |= flush;
1509 /* We need to store of the IP ID check to be included later
1510 * when we can verify that this packet does in fact belong
1513 flush_id = (u16)(id - ntohs(iph2->id));
1515 /* This bit of code makes it much easier for us to identify
1516 * the cases where we are doing atomic vs non-atomic IP ID
1517 * checks. Specifically an atomic check can return IP ID
1518 * values 0 - 0xFFFF, while a non-atomic check can only
1519 * return 0 or 0xFFFF.
1521 if (!NAPI_GRO_CB(p)->is_atomic ||
1522 !(iph->frag_off & htons(IP_DF))) {
1523 flush_id ^= NAPI_GRO_CB(p)->count;
1524 flush_id = flush_id ? 0xFFFF : 0;
1527 /* If the previous IP ID value was based on an atomic
1528 * datagram we can overwrite the value and ignore it.
1530 if (NAPI_GRO_CB(skb)->is_atomic)
1531 NAPI_GRO_CB(p)->flush_id = flush_id;
1533 NAPI_GRO_CB(p)->flush_id |= flush_id;
1536 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1537 NAPI_GRO_CB(skb)->flush |= flush;
1538 skb_set_network_header(skb, off);
1539 /* The above will be needed by the transport layer if there is one
1540 * immediately following this IP hdr.
1543 /* Note : No need to call skb_gro_postpull_rcsum() here,
1544 * as we already checked checksum over ipv4 header was 0
1546 skb_gro_pull(skb, sizeof(*iph));
1547 skb_set_transport_header(skb, skb_gro_offset(skb));
1549 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1550 ops->callbacks.gro_receive, head, skb);
1556 skb_gro_flush_final(skb, pp, flush);
1560 EXPORT_SYMBOL(inet_gro_receive);
1562 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1563 struct sk_buff *skb)
1565 if (NAPI_GRO_CB(skb)->encap_mark) {
1566 NAPI_GRO_CB(skb)->flush = 1;
1570 NAPI_GRO_CB(skb)->encap_mark = 1;
1572 return inet_gro_receive(head, skb);
1575 #define SECONDS_PER_DAY 86400
1577 /* inet_current_timestamp - Return IP network timestamp
1579 * Return milliseconds since midnight in network byte order.
1581 __be32 inet_current_timestamp(void)
1585 struct timespec64 ts;
1587 ktime_get_real_ts64(&ts);
1589 /* Get secs since midnight. */
1590 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1591 /* Convert to msecs. */
1592 msecs = secs * MSEC_PER_SEC;
1593 /* Convert nsec to msec. */
1594 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1596 /* Convert to network byte order. */
1597 return htonl(msecs);
1599 EXPORT_SYMBOL(inet_current_timestamp);
1601 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1603 unsigned int family = READ_ONCE(sk->sk_family);
1605 if (family == AF_INET)
1606 return ip_recv_error(sk, msg, len, addr_len);
1607 #if IS_ENABLED(CONFIG_IPV6)
1608 if (family == AF_INET6)
1609 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1614 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1616 __be16 newlen = htons(skb->len - nhoff);
1617 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1618 const struct net_offload *ops;
1619 int proto = iph->protocol;
1622 if (skb->encapsulation) {
1623 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1624 skb_set_inner_network_header(skb, nhoff);
1627 csum_replace2(&iph->check, iph->tot_len, newlen);
1628 iph->tot_len = newlen;
1631 ops = rcu_dereference(inet_offloads[proto]);
1632 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1635 /* Only need to add sizeof(*iph) to get to the next hdr below
1636 * because any hdr with option will have been flushed in
1637 * inet_gro_receive().
1639 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1640 tcp4_gro_complete, udp4_gro_complete,
1641 skb, nhoff + sizeof(*iph));
1648 EXPORT_SYMBOL(inet_gro_complete);
1650 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1652 skb->encapsulation = 1;
1653 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1654 return inet_gro_complete(skb, nhoff);
1657 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1658 unsigned short type, unsigned char protocol,
1661 struct socket *sock;
1662 int rc = sock_create_kern(net, family, type, protocol, &sock);
1666 (*sk)->sk_allocation = GFP_ATOMIC;
1668 * Unhash it so that IP input processing does not even see it,
1669 * we do not wish this socket to see incoming packets.
1671 (*sk)->sk_prot->unhash(*sk);
1675 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1677 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1679 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1681 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1683 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1685 unsigned long res = 0;
1688 for_each_possible_cpu(i)
1689 res += snmp_get_cpu_field(mib, i, offt);
1692 EXPORT_SYMBOL_GPL(snmp_fold_field);
1694 #if BITS_PER_LONG==32
1696 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1697 size_t syncp_offset)
1700 struct u64_stats_sync *syncp;
1704 bhptr = per_cpu_ptr(mib, cpu);
1705 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1707 start = u64_stats_fetch_begin_irq(syncp);
1708 v = *(((u64 *)bhptr) + offt);
1709 } while (u64_stats_fetch_retry_irq(syncp, start));
1713 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1715 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1720 for_each_possible_cpu(cpu) {
1721 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1725 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1728 #ifdef CONFIG_IP_MULTICAST
1729 static const struct net_protocol igmp_protocol = {
1730 .handler = igmp_rcv,
1735 static const struct net_protocol tcp_protocol = {
1736 .handler = tcp_v4_rcv,
1737 .err_handler = tcp_v4_err,
1740 .icmp_strict_tag_validation = 1,
1743 static const struct net_protocol udp_protocol = {
1745 .err_handler = udp_err,
1750 static const struct net_protocol icmp_protocol = {
1751 .handler = icmp_rcv,
1752 .err_handler = icmp_err,
1757 static __net_init int ipv4_mib_init_net(struct net *net)
1761 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1762 if (!net->mib.tcp_statistics)
1764 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1765 if (!net->mib.ip_statistics)
1768 for_each_possible_cpu(i) {
1769 struct ipstats_mib *af_inet_stats;
1770 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1771 u64_stats_init(&af_inet_stats->syncp);
1774 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1775 if (!net->mib.net_statistics)
1777 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1778 if (!net->mib.udp_statistics)
1780 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1781 if (!net->mib.udplite_statistics)
1782 goto err_udplite_mib;
1783 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1784 if (!net->mib.icmp_statistics)
1786 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1788 if (!net->mib.icmpmsg_statistics)
1789 goto err_icmpmsg_mib;
1795 free_percpu(net->mib.icmp_statistics);
1797 free_percpu(net->mib.udplite_statistics);
1799 free_percpu(net->mib.udp_statistics);
1801 free_percpu(net->mib.net_statistics);
1803 free_percpu(net->mib.ip_statistics);
1805 free_percpu(net->mib.tcp_statistics);
1810 static __net_exit void ipv4_mib_exit_net(struct net *net)
1812 kfree(net->mib.icmpmsg_statistics);
1813 free_percpu(net->mib.icmp_statistics);
1814 free_percpu(net->mib.udplite_statistics);
1815 free_percpu(net->mib.udp_statistics);
1816 free_percpu(net->mib.net_statistics);
1817 free_percpu(net->mib.ip_statistics);
1818 free_percpu(net->mib.tcp_statistics);
1820 /* allocated on demand, see mptcp_init_sock() */
1821 free_percpu(net->mib.mptcp_statistics);
1825 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1826 .init = ipv4_mib_init_net,
1827 .exit = ipv4_mib_exit_net,
1830 static int __init init_ipv4_mibs(void)
1832 return register_pernet_subsys(&ipv4_mib_ops);
1835 static __net_init int inet_init_net(struct net *net)
1838 * Set defaults for local port range
1840 seqlock_init(&net->ipv4.ip_local_ports.lock);
1841 net->ipv4.ip_local_ports.range[0] = 32768;
1842 net->ipv4.ip_local_ports.range[1] = 60999;
1844 seqlock_init(&net->ipv4.ping_group_range.lock);
1846 * Sane defaults - nobody may create ping sockets.
1847 * Boot scripts should set this to distro-specific group.
1849 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1850 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1852 /* Default values for sysctl-controlled parameters.
1853 * We set them here, in case sysctl is not compiled.
1855 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1856 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1857 net->ipv4.sysctl_ip_dynaddr = 0;
1858 net->ipv4.sysctl_ip_early_demux = 1;
1859 net->ipv4.sysctl_udp_early_demux = 1;
1860 net->ipv4.sysctl_tcp_early_demux = 1;
1861 net->ipv4.sysctl_nexthop_compat_mode = 1;
1862 #ifdef CONFIG_SYSCTL
1863 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1866 /* Some igmp sysctl, whose values are always used */
1867 net->ipv4.sysctl_igmp_max_memberships = 20;
1868 net->ipv4.sysctl_igmp_max_msf = 10;
1869 /* IGMP reports for link-local multicast groups are enabled by default */
1870 net->ipv4.sysctl_igmp_llm_reports = 1;
1871 net->ipv4.sysctl_igmp_qrv = 2;
1876 static __net_initdata struct pernet_operations af_inet_ops = {
1877 .init = inet_init_net,
1880 static int __init init_inet_pernet_ops(void)
1882 return register_pernet_subsys(&af_inet_ops);
1885 static int ipv4_proc_init(void);
1888 * IP protocol layer initialiser
1891 static struct packet_offload ip_packet_offload __read_mostly = {
1892 .type = cpu_to_be16(ETH_P_IP),
1894 .gso_segment = inet_gso_segment,
1895 .gro_receive = inet_gro_receive,
1896 .gro_complete = inet_gro_complete,
1900 static const struct net_offload ipip_offload = {
1902 .gso_segment = ipip_gso_segment,
1903 .gro_receive = ipip_gro_receive,
1904 .gro_complete = ipip_gro_complete,
1908 static int __init ipip_offload_init(void)
1910 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1913 static int __init ipv4_offload_init(void)
1918 if (udpv4_offload_init() < 0)
1919 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1920 if (tcpv4_offload_init() < 0)
1921 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1922 if (ipip_offload_init() < 0)
1923 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1925 dev_add_offload(&ip_packet_offload);
1929 fs_initcall(ipv4_offload_init);
1931 static struct packet_type ip_packet_type __read_mostly = {
1932 .type = cpu_to_be16(ETH_P_IP),
1934 .list_func = ip_list_rcv,
1937 static int __init inet_init(void)
1939 struct inet_protosw *q;
1940 struct list_head *r;
1943 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1945 rc = proto_register(&tcp_prot, 1);
1949 rc = proto_register(&udp_prot, 1);
1951 goto out_unregister_tcp_proto;
1953 rc = proto_register(&raw_prot, 1);
1955 goto out_unregister_udp_proto;
1957 rc = proto_register(&ping_prot, 1);
1959 goto out_unregister_raw_proto;
1962 * Tell SOCKET that we are alive...
1965 (void)sock_register(&inet_family_ops);
1967 #ifdef CONFIG_SYSCTL
1968 ip_static_sysctl_init();
1972 * Add all the base protocols.
1975 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1976 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1977 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1978 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1979 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1980 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1981 #ifdef CONFIG_IP_MULTICAST
1982 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1983 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1986 /* Register the socket-side information for inet_create. */
1987 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1990 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1991 inet_register_protosw(q);
1994 * Set the ARP module up
2000 * Set the IP module up
2005 /* Initialise per-cpu ipv4 mibs */
2006 if (init_ipv4_mibs())
2007 panic("%s: Cannot init ipv4 mibs\n", __func__);
2009 /* Setup TCP slab cache for open requests. */
2012 /* Setup UDP memory threshold */
2015 /* Add UDP-Lite (RFC 3828) */
2016 udplite4_register();
2023 * Set the ICMP layer up
2026 if (icmp_init() < 0)
2027 panic("Failed to create the ICMP control socket.\n");
2030 * Initialise the multicast router
2032 #if defined(CONFIG_IP_MROUTE)
2034 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2037 if (init_inet_pernet_ops())
2038 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2044 dev_add_pack(&ip_packet_type);
2046 ip_tunnel_core_init();
2051 out_unregister_raw_proto:
2052 proto_unregister(&raw_prot);
2053 out_unregister_udp_proto:
2054 proto_unregister(&udp_prot);
2055 out_unregister_tcp_proto:
2056 proto_unregister(&tcp_prot);
2060 fs_initcall(inet_init);
2062 /* ------------------------------------------------------------------------ */
2064 #ifdef CONFIG_PROC_FS
2065 static int __init ipv4_proc_init(void)
2069 if (raw_proc_init())
2071 if (tcp4_proc_init())
2073 if (udp4_proc_init())
2075 if (ping_proc_init())
2077 if (ip_misc_proc_init())
2094 #else /* CONFIG_PROC_FS */
2095 static int __init ipv4_proc_init(void)
2099 #endif /* CONFIG_PROC_FS */