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>
120 #include <trace/events/sock.h>
122 /* The inetsw table contains everything that inet_create needs to
123 * build a new socket.
125 static struct list_head inetsw[SOCK_MAX];
126 static DEFINE_SPINLOCK(inetsw_lock);
128 /* New destruction routine */
130 void inet_sock_destruct(struct sock *sk)
132 struct inet_sock *inet = inet_sk(sk);
134 __skb_queue_purge(&sk->sk_receive_queue);
135 if (sk->sk_rx_skb_cache) {
136 __kfree_skb(sk->sk_rx_skb_cache);
137 sk->sk_rx_skb_cache = NULL;
139 __skb_queue_purge(&sk->sk_error_queue);
143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144 pr_err("Attempt to release TCP socket in state %d %p\n",
148 if (!sock_flag(sk, SOCK_DEAD)) {
149 pr_err("Attempt to release alive inet socket %p\n", sk);
153 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
155 WARN_ON(sk->sk_wmem_queued);
156 WARN_ON(sk->sk_forward_alloc);
158 kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
160 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
161 sk_refcnt_debug_dec(sk);
163 EXPORT_SYMBOL(inet_sock_destruct);
166 * The routines beyond this point handle the behaviour of an AF_INET
167 * socket object. Mostly it punts to the subprotocols of IP to do
172 * Automatically bind an unbound socket.
175 static int inet_autobind(struct sock *sk)
177 struct inet_sock *inet;
178 /* We may need to bind the socket. */
181 if (!inet->inet_num) {
182 if (sk->sk_prot->get_port(sk, 0)) {
186 inet->inet_sport = htons(inet->inet_num);
193 * Move a socket into listening state.
195 int inet_listen(struct socket *sock, int backlog)
197 struct sock *sk = sock->sk;
198 unsigned char old_state;
199 int err, tcp_fastopen;
204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
207 old_state = sk->sk_state;
208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
211 sk->sk_max_ack_backlog = backlog;
212 /* Really, if the socket is already in listen state
213 * we can only allow the backlog to be adjusted.
215 if (old_state != TCP_LISTEN) {
216 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
217 * Note that only TCP sockets (SOCK_STREAM) will reach here.
218 * Also fastopen backlog may already been set via the option
219 * because the socket was in TCP_LISTEN state previously but
220 * was shutdown() rather than close().
222 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
223 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
224 (tcp_fastopen & TFO_SERVER_ENABLE) &&
225 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
226 fastopen_queue_tune(sk, backlog);
227 tcp_fastopen_init_key_once(sock_net(sk));
230 err = inet_csk_listen_start(sk, backlog);
233 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
241 EXPORT_SYMBOL(inet_listen);
244 * Create an inet socket.
247 static int inet_create(struct net *net, struct socket *sock, int protocol,
251 struct inet_protosw *answer;
252 struct inet_sock *inet;
253 struct proto *answer_prot;
254 unsigned char answer_flags;
255 int try_loading_module = 0;
258 if (protocol < 0 || protocol >= IPPROTO_MAX)
261 sock->state = SS_UNCONNECTED;
263 /* Look for the requested type/protocol pair. */
265 err = -ESOCKTNOSUPPORT;
267 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
270 /* Check the non-wild match. */
271 if (protocol == answer->protocol) {
272 if (protocol != IPPROTO_IP)
275 /* Check for the two wild cases. */
276 if (IPPROTO_IP == protocol) {
277 protocol = answer->protocol;
280 if (IPPROTO_IP == answer->protocol)
283 err = -EPROTONOSUPPORT;
287 if (try_loading_module < 2) {
290 * Be more specific, e.g. net-pf-2-proto-132-type-1
291 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
293 if (++try_loading_module == 1)
294 request_module("net-pf-%d-proto-%d-type-%d",
295 PF_INET, protocol, sock->type);
297 * Fall back to generic, e.g. net-pf-2-proto-132
298 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
301 request_module("net-pf-%d-proto-%d",
303 goto lookup_protocol;
309 if (sock->type == SOCK_RAW && !kern &&
310 !ns_capable(net->user_ns, CAP_NET_RAW))
313 sock->ops = answer->ops;
314 answer_prot = answer->prot;
315 answer_flags = answer->flags;
318 WARN_ON(!answer_prot->slab);
321 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
326 if (INET_PROTOSW_REUSE & answer_flags)
327 sk->sk_reuse = SK_CAN_REUSE;
330 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
334 if (SOCK_RAW == sock->type) {
335 inet->inet_num = protocol;
336 if (IPPROTO_RAW == protocol)
340 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
341 inet->pmtudisc = IP_PMTUDISC_DONT;
343 inet->pmtudisc = IP_PMTUDISC_WANT;
347 sock_init_data(sock, sk);
349 sk->sk_destruct = inet_sock_destruct;
350 sk->sk_protocol = protocol;
351 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
358 inet->mc_list = NULL;
361 sk_refcnt_debug_inc(sk);
363 if (inet->inet_num) {
364 /* It assumes that any protocol which allows
365 * the user to assign a number at socket
366 * creation time automatically
369 inet->inet_sport = htons(inet->inet_num);
370 /* Add to protocol hash chains. */
371 err = sk->sk_prot->hash(sk);
373 sk_common_release(sk);
378 if (sk->sk_prot->init) {
379 err = sk->sk_prot->init(sk);
381 sk_common_release(sk);
387 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
389 sk_common_release(sk);
402 * The peer socket should always be NULL (or else). When we call this
403 * function we are destroying the object and from then on nobody
404 * should refer to it.
406 int inet_release(struct socket *sock)
408 struct sock *sk = sock->sk;
413 /* Applications forget to leave groups before exiting */
414 ip_mc_drop_socket(sk);
416 /* If linger is set, we don't return until the close
417 * is complete. Otherwise we return immediately. The
418 * actually closing is done the same either way.
420 * If the close is due to the process exiting, we never
424 if (sock_flag(sk, SOCK_LINGER) &&
425 !(current->flags & PF_EXITING))
426 timeout = sk->sk_lingertime;
427 sk->sk_prot->close(sk, timeout);
432 EXPORT_SYMBOL(inet_release);
434 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
436 struct sock *sk = sock->sk;
439 /* If the socket has its own bind function then use it. (RAW) */
440 if (sk->sk_prot->bind) {
441 return sk->sk_prot->bind(sk, uaddr, addr_len);
443 if (addr_len < sizeof(struct sockaddr_in))
446 /* BPF prog is run before any checks are done so that if the prog
447 * changes context in a wrong way it will be caught.
449 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
453 return __inet_bind(sk, uaddr, addr_len, false, true);
455 EXPORT_SYMBOL(inet_bind);
457 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
458 bool force_bind_address_no_port, bool with_lock)
460 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
461 struct inet_sock *inet = inet_sk(sk);
462 struct net *net = sock_net(sk);
465 u32 tb_id = RT_TABLE_LOCAL;
468 if (addr->sin_family != AF_INET) {
469 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
470 * only if s_addr is INADDR_ANY.
473 if (addr->sin_family != AF_UNSPEC ||
474 addr->sin_addr.s_addr != htonl(INADDR_ANY))
478 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
479 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
481 /* Not specified by any standard per-se, however it breaks too
482 * many applications when removed. It is unfortunate since
483 * allowing applications to make a non-local bind solves
484 * several problems with systems using dynamic addressing.
485 * (ie. your servers still start up even if your ISDN link
486 * is temporarily down)
488 err = -EADDRNOTAVAIL;
489 if (!inet_can_nonlocal_bind(net, inet) &&
490 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
491 chk_addr_ret != RTN_LOCAL &&
492 chk_addr_ret != RTN_MULTICAST &&
493 chk_addr_ret != RTN_BROADCAST)
496 snum = ntohs(addr->sin_port);
498 if (snum && snum < inet_prot_sock(net) &&
499 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
502 /* We keep a pair of addresses. rcv_saddr is the one
503 * used by hash lookups, and saddr is used for transmit.
505 * In the BSD API these are the same except where it
506 * would be illegal to use them (multicast/broadcast) in
507 * which case the sending device address is used.
512 /* Check these errors (active socket, double bind). */
514 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
515 goto out_release_sock;
517 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
518 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
519 inet->inet_saddr = 0; /* Use device */
521 /* Make sure we are allowed to bind here. */
522 if (snum || !(inet->bind_address_no_port ||
523 force_bind_address_no_port)) {
524 if (sk->sk_prot->get_port(sk, snum)) {
525 inet->inet_saddr = inet->inet_rcv_saddr = 0;
527 goto out_release_sock;
529 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
531 inet->inet_saddr = inet->inet_rcv_saddr = 0;
532 goto out_release_sock;
536 if (inet->inet_rcv_saddr)
537 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
539 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
540 inet->inet_sport = htons(inet->inet_num);
541 inet->inet_daddr = 0;
542 inet->inet_dport = 0;
552 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
553 int addr_len, int flags)
555 struct sock *sk = sock->sk;
558 if (addr_len < sizeof(uaddr->sa_family))
560 if (uaddr->sa_family == AF_UNSPEC)
561 return sk->sk_prot->disconnect(sk, flags);
563 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
564 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
569 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
571 return sk->sk_prot->connect(sk, uaddr, addr_len);
573 EXPORT_SYMBOL(inet_dgram_connect);
575 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
577 DEFINE_WAIT_FUNC(wait, woken_wake_function);
579 add_wait_queue(sk_sleep(sk), &wait);
580 sk->sk_write_pending += writebias;
581 sk->sk_wait_pending++;
583 /* Basic assumption: if someone sets sk->sk_err, he _must_
584 * change state of the socket from TCP_SYN_*.
585 * Connect() does not allow to get error notifications
586 * without closing the socket.
588 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
590 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
592 if (signal_pending(current) || !timeo)
595 remove_wait_queue(sk_sleep(sk), &wait);
596 sk->sk_write_pending -= writebias;
597 sk->sk_wait_pending--;
602 * Connect to a remote host. There is regrettably still a little
603 * TCP 'magic' in here.
605 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
606 int addr_len, int flags, int is_sendmsg)
608 struct sock *sk = sock->sk;
613 * uaddr can be NULL and addr_len can be 0 if:
614 * sk is a TCP fastopen active socket and
615 * TCP_FASTOPEN_CONNECT sockopt is set and
616 * we already have a valid cookie for this socket.
617 * In this case, user can call write() after connect().
618 * write() will invoke tcp_sendmsg_fastopen() which calls
619 * __inet_stream_connect().
622 if (addr_len < sizeof(uaddr->sa_family))
625 if (uaddr->sa_family == AF_UNSPEC) {
626 err = sk->sk_prot->disconnect(sk, flags);
627 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
632 switch (sock->state) {
640 if (inet_sk(sk)->defer_connect)
641 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
644 /* Fall out of switch with err, set for this state */
648 if (sk->sk_state != TCP_CLOSE)
651 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
652 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
657 err = sk->sk_prot->connect(sk, uaddr, addr_len);
661 sock->state = SS_CONNECTING;
663 if (!err && inet_sk(sk)->defer_connect)
666 /* Just entered SS_CONNECTING state; the only
667 * difference is that return value in non-blocking
668 * case is EINPROGRESS, rather than EALREADY.
674 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
676 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
677 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
678 tcp_sk(sk)->fastopen_req &&
679 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
681 /* Error code is set above */
682 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
685 err = sock_intr_errno(timeo);
686 if (signal_pending(current))
690 /* Connection was closed by RST, timeout, ICMP error
691 * or another process disconnected us.
693 if (sk->sk_state == TCP_CLOSE)
696 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
697 * and error was received after socket entered established state.
698 * Hence, it is handled normally after connect() return successfully.
701 sock->state = SS_CONNECTED;
707 err = sock_error(sk) ? : -ECONNABORTED;
708 sock->state = SS_UNCONNECTED;
709 if (sk->sk_prot->disconnect(sk, flags))
710 sock->state = SS_DISCONNECTING;
713 EXPORT_SYMBOL(__inet_stream_connect);
715 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
716 int addr_len, int flags)
721 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
722 release_sock(sock->sk);
725 EXPORT_SYMBOL(inet_stream_connect);
728 * Accept a pending connection. The TCP layer now gives BSD semantics.
731 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
734 struct sock *sk1 = sock->sk;
736 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
743 sock_rps_record_flow(sk2);
744 WARN_ON(!((1 << sk2->sk_state) &
745 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
746 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
748 sock_graft(sk2, newsock);
750 newsock->state = SS_CONNECTED;
756 EXPORT_SYMBOL(inet_accept);
760 * This does both peername and sockname.
762 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
765 struct sock *sk = sock->sk;
766 struct inet_sock *inet = inet_sk(sk);
767 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
769 sin->sin_family = AF_INET;
771 if (!inet->inet_dport ||
772 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
775 sin->sin_port = inet->inet_dport;
776 sin->sin_addr.s_addr = inet->inet_daddr;
778 __be32 addr = inet->inet_rcv_saddr;
780 addr = inet->inet_saddr;
781 sin->sin_port = inet->inet_sport;
782 sin->sin_addr.s_addr = addr;
784 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
787 EXPORT_SYMBOL(inet_getname);
789 int inet_send_prepare(struct sock *sk)
791 sock_rps_record_flow(sk);
793 /* We may need to bind the socket. */
794 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
800 EXPORT_SYMBOL_GPL(inet_send_prepare);
802 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
804 struct sock *sk = sock->sk;
806 if (unlikely(inet_send_prepare(sk)))
809 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
812 EXPORT_SYMBOL(inet_sendmsg);
814 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
815 size_t size, int flags)
817 struct sock *sk = sock->sk;
819 if (unlikely(inet_send_prepare(sk)))
822 if (sk->sk_prot->sendpage)
823 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
824 return sock_no_sendpage(sock, page, offset, size, flags);
826 EXPORT_SYMBOL(inet_sendpage);
828 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
829 size_t, int, int, int *));
830 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
833 struct sock *sk = sock->sk;
837 if (likely(!(flags & MSG_ERRQUEUE)))
838 sock_rps_record_flow(sk);
840 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
841 sk, msg, size, flags & MSG_DONTWAIT,
842 flags & ~MSG_DONTWAIT, &addr_len);
844 msg->msg_namelen = addr_len;
847 EXPORT_SYMBOL(inet_recvmsg);
849 int inet_shutdown(struct socket *sock, int how)
851 struct sock *sk = sock->sk;
854 /* This should really check to make sure
855 * the socket is a TCP socket. (WHY AC...)
857 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
860 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
864 if (sock->state == SS_CONNECTING) {
865 if ((1 << sk->sk_state) &
866 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
867 sock->state = SS_DISCONNECTING;
869 sock->state = SS_CONNECTED;
872 switch (sk->sk_state) {
875 /* Hack to wake up other listeners, who can poll for
876 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
879 sk->sk_shutdown |= how;
880 if (sk->sk_prot->shutdown)
881 sk->sk_prot->shutdown(sk, how);
884 /* Remaining two branches are temporary solution for missing
885 * close() in multithreaded environment. It is _not_ a good idea,
886 * but we have no choice until close() is repaired at VFS level.
889 if (!(how & RCV_SHUTDOWN))
893 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
894 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
898 /* Wake up anyone sleeping in poll. */
899 sk->sk_state_change(sk);
903 EXPORT_SYMBOL(inet_shutdown);
906 * ioctl() calls you can issue on an INET socket. Most of these are
907 * device configuration and stuff and very rarely used. Some ioctls
908 * pass on to the socket itself.
910 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
911 * loads the devconfigure module does its configuring and unloads it.
912 * There's a good 20K of config code hanging around the kernel.
915 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
917 struct sock *sk = sock->sk;
919 struct net *net = sock_net(sk);
920 void __user *p = (void __user *)arg;
927 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
929 err = ip_rt_ioctl(net, cmd, &rt);
937 err = arp_ioctl(net, cmd, (void __user *)arg);
944 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
946 err = devinet_ioctl(net, cmd, &ifr);
947 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
957 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
959 err = devinet_ioctl(net, cmd, &ifr);
962 if (sk->sk_prot->ioctl)
963 err = sk->sk_prot->ioctl(sk, cmd, arg);
970 EXPORT_SYMBOL(inet_ioctl);
973 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
975 struct sock *sk = sock->sk;
976 int err = -ENOIOCTLCMD;
978 if (sk->sk_prot->compat_ioctl)
979 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
985 const struct proto_ops inet_stream_ops = {
987 .owner = THIS_MODULE,
988 .release = inet_release,
990 .connect = inet_stream_connect,
991 .socketpair = sock_no_socketpair,
992 .accept = inet_accept,
993 .getname = inet_getname,
996 .gettstamp = sock_gettstamp,
997 .listen = inet_listen,
998 .shutdown = inet_shutdown,
999 .setsockopt = sock_common_setsockopt,
1000 .getsockopt = sock_common_getsockopt,
1001 .sendmsg = inet_sendmsg,
1002 .recvmsg = inet_recvmsg,
1006 .sendpage = inet_sendpage,
1007 .splice_read = tcp_splice_read,
1008 .read_sock = tcp_read_sock,
1009 .sendmsg_locked = tcp_sendmsg_locked,
1010 .sendpage_locked = tcp_sendpage_locked,
1011 .peek_len = tcp_peek_len,
1012 #ifdef CONFIG_COMPAT
1013 .compat_setsockopt = compat_sock_common_setsockopt,
1014 .compat_getsockopt = compat_sock_common_getsockopt,
1015 .compat_ioctl = inet_compat_ioctl,
1017 .set_rcvlowat = tcp_set_rcvlowat,
1019 EXPORT_SYMBOL(inet_stream_ops);
1021 const struct proto_ops inet_dgram_ops = {
1023 .owner = THIS_MODULE,
1024 .release = inet_release,
1026 .connect = inet_dgram_connect,
1027 .socketpair = sock_no_socketpair,
1028 .accept = sock_no_accept,
1029 .getname = inet_getname,
1031 .ioctl = inet_ioctl,
1032 .gettstamp = sock_gettstamp,
1033 .listen = sock_no_listen,
1034 .shutdown = inet_shutdown,
1035 .setsockopt = sock_common_setsockopt,
1036 .getsockopt = sock_common_getsockopt,
1037 .sendmsg = inet_sendmsg,
1038 .recvmsg = inet_recvmsg,
1039 .mmap = sock_no_mmap,
1040 .sendpage = inet_sendpage,
1041 .set_peek_off = sk_set_peek_off,
1042 #ifdef CONFIG_COMPAT
1043 .compat_setsockopt = compat_sock_common_setsockopt,
1044 .compat_getsockopt = compat_sock_common_getsockopt,
1045 .compat_ioctl = inet_compat_ioctl,
1048 EXPORT_SYMBOL(inet_dgram_ops);
1051 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1054 static const struct proto_ops inet_sockraw_ops = {
1056 .owner = THIS_MODULE,
1057 .release = inet_release,
1059 .connect = inet_dgram_connect,
1060 .socketpair = sock_no_socketpair,
1061 .accept = sock_no_accept,
1062 .getname = inet_getname,
1063 .poll = datagram_poll,
1064 .ioctl = inet_ioctl,
1065 .gettstamp = sock_gettstamp,
1066 .listen = sock_no_listen,
1067 .shutdown = inet_shutdown,
1068 .setsockopt = sock_common_setsockopt,
1069 .getsockopt = sock_common_getsockopt,
1070 .sendmsg = inet_sendmsg,
1071 .recvmsg = inet_recvmsg,
1072 .mmap = sock_no_mmap,
1073 .sendpage = inet_sendpage,
1074 #ifdef CONFIG_COMPAT
1075 .compat_setsockopt = compat_sock_common_setsockopt,
1076 .compat_getsockopt = compat_sock_common_getsockopt,
1077 .compat_ioctl = inet_compat_ioctl,
1081 static const struct net_proto_family inet_family_ops = {
1083 .create = inet_create,
1084 .owner = THIS_MODULE,
1087 /* Upon startup we insert all the elements in inetsw_array[] into
1088 * the linked list inetsw.
1090 static struct inet_protosw inetsw_array[] =
1093 .type = SOCK_STREAM,
1094 .protocol = IPPROTO_TCP,
1096 .ops = &inet_stream_ops,
1097 .flags = INET_PROTOSW_PERMANENT |
1103 .protocol = IPPROTO_UDP,
1105 .ops = &inet_dgram_ops,
1106 .flags = INET_PROTOSW_PERMANENT,
1111 .protocol = IPPROTO_ICMP,
1113 .ops = &inet_sockraw_ops,
1114 .flags = INET_PROTOSW_REUSE,
1119 .protocol = IPPROTO_IP, /* wild card */
1121 .ops = &inet_sockraw_ops,
1122 .flags = INET_PROTOSW_REUSE,
1126 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1128 void inet_register_protosw(struct inet_protosw *p)
1130 struct list_head *lh;
1131 struct inet_protosw *answer;
1132 int protocol = p->protocol;
1133 struct list_head *last_perm;
1135 spin_lock_bh(&inetsw_lock);
1137 if (p->type >= SOCK_MAX)
1140 /* If we are trying to override a permanent protocol, bail. */
1141 last_perm = &inetsw[p->type];
1142 list_for_each(lh, &inetsw[p->type]) {
1143 answer = list_entry(lh, struct inet_protosw, list);
1144 /* Check only the non-wild match. */
1145 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1147 if (protocol == answer->protocol)
1152 /* Add the new entry after the last permanent entry if any, so that
1153 * the new entry does not override a permanent entry when matched with
1154 * a wild-card protocol. But it is allowed to override any existing
1155 * non-permanent entry. This means that when we remove this entry, the
1156 * system automatically returns to the old behavior.
1158 list_add_rcu(&p->list, last_perm);
1160 spin_unlock_bh(&inetsw_lock);
1165 pr_err("Attempt to override permanent protocol %d\n", protocol);
1169 pr_err("Ignoring attempt to register invalid socket type %d\n",
1173 EXPORT_SYMBOL(inet_register_protosw);
1175 void inet_unregister_protosw(struct inet_protosw *p)
1177 if (INET_PROTOSW_PERMANENT & p->flags) {
1178 pr_err("Attempt to unregister permanent protocol %d\n",
1181 spin_lock_bh(&inetsw_lock);
1182 list_del_rcu(&p->list);
1183 spin_unlock_bh(&inetsw_lock);
1188 EXPORT_SYMBOL(inet_unregister_protosw);
1190 static int inet_sk_reselect_saddr(struct sock *sk)
1192 struct inet_sock *inet = inet_sk(sk);
1193 __be32 old_saddr = inet->inet_saddr;
1194 __be32 daddr = inet->inet_daddr;
1198 struct ip_options_rcu *inet_opt;
1200 inet_opt = rcu_dereference_protected(inet->inet_opt,
1201 lockdep_sock_is_held(sk));
1202 if (inet_opt && inet_opt->opt.srr)
1203 daddr = inet_opt->opt.faddr;
1205 /* Query new route. */
1206 fl4 = &inet->cork.fl.u.ip4;
1207 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1208 sk->sk_bound_dev_if, sk->sk_protocol,
1209 inet->inet_sport, inet->inet_dport, sk);
1213 sk_setup_caps(sk, &rt->dst);
1215 new_saddr = fl4->saddr;
1217 if (new_saddr == old_saddr)
1220 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1221 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1222 __func__, &old_saddr, &new_saddr);
1225 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1228 * XXX The only one ugly spot where we need to
1229 * XXX really change the sockets identity after
1230 * XXX it has entered the hashes. -DaveM
1232 * Besides that, it does not check for connection
1233 * uniqueness. Wait for troubles.
1235 return __sk_prot_rehash(sk);
1238 int inet_sk_rebuild_header(struct sock *sk)
1240 struct inet_sock *inet = inet_sk(sk);
1241 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1243 struct ip_options_rcu *inet_opt;
1247 /* Route is OK, nothing to do. */
1253 inet_opt = rcu_dereference(inet->inet_opt);
1254 daddr = inet->inet_daddr;
1255 if (inet_opt && inet_opt->opt.srr)
1256 daddr = inet_opt->opt.faddr;
1258 fl4 = &inet->cork.fl.u.ip4;
1259 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1260 inet->inet_dport, inet->inet_sport,
1261 sk->sk_protocol, RT_CONN_FLAGS(sk),
1262 sk->sk_bound_dev_if);
1265 sk_setup_caps(sk, &rt->dst);
1269 /* Routing failed... */
1270 sk->sk_route_caps = 0;
1272 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1273 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1275 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1276 sk->sk_state != TCP_SYN_SENT ||
1277 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1278 (err = inet_sk_reselect_saddr(sk)) != 0)
1279 sk->sk_err_soft = -err;
1284 EXPORT_SYMBOL(inet_sk_rebuild_header);
1286 void inet_sk_set_state(struct sock *sk, int state)
1288 trace_inet_sock_set_state(sk, sk->sk_state, state);
1289 sk->sk_state = state;
1291 EXPORT_SYMBOL(inet_sk_set_state);
1293 void inet_sk_state_store(struct sock *sk, int newstate)
1295 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1296 smp_store_release(&sk->sk_state, newstate);
1299 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1300 netdev_features_t features)
1302 bool udpfrag = false, fixedid = false, gso_partial, encap;
1303 struct sk_buff *segs = ERR_PTR(-EINVAL);
1304 const struct net_offload *ops;
1305 unsigned int offset = 0;
1312 skb_reset_network_header(skb);
1313 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1314 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1319 if (ihl < sizeof(*iph))
1322 id = ntohs(iph->id);
1323 proto = iph->protocol;
1325 /* Warning: after this point, iph might be no longer valid */
1326 if (unlikely(!pskb_may_pull(skb, ihl)))
1328 __skb_pull(skb, ihl);
1330 encap = SKB_GSO_CB(skb)->encap_level > 0;
1332 features &= skb->dev->hw_enc_features;
1333 SKB_GSO_CB(skb)->encap_level += ihl;
1335 skb_reset_transport_header(skb);
1337 segs = ERR_PTR(-EPROTONOSUPPORT);
1339 if (!skb->encapsulation || encap) {
1340 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1341 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1343 /* fixed ID is invalid if DF bit is not set */
1344 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1348 ops = rcu_dereference(inet_offloads[proto]);
1349 if (likely(ops && ops->callbacks.gso_segment)) {
1350 segs = ops->callbacks.gso_segment(skb, features);
1352 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1355 if (IS_ERR_OR_NULL(segs))
1358 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1362 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1364 iph->frag_off = htons(offset >> 3);
1366 iph->frag_off |= htons(IP_MF);
1367 offset += skb->len - nhoff - ihl;
1368 tot_len = skb->len - nhoff;
1369 } else if (skb_is_gso(skb)) {
1371 iph->id = htons(id);
1372 id += skb_shinfo(skb)->gso_segs;
1376 tot_len = skb_shinfo(skb)->gso_size +
1377 SKB_GSO_CB(skb)->data_offset +
1378 skb->head - (unsigned char *)iph;
1380 tot_len = skb->len - nhoff;
1383 iph->id = htons(id++);
1384 tot_len = skb->len - nhoff;
1386 iph->tot_len = htons(tot_len);
1389 skb_reset_inner_headers(skb);
1390 skb->network_header = (u8 *)iph - skb->head;
1391 skb_reset_mac_len(skb);
1392 } while ((skb = skb->next));
1397 EXPORT_SYMBOL(inet_gso_segment);
1399 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1400 netdev_features_t features)
1402 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1403 return ERR_PTR(-EINVAL);
1405 return inet_gso_segment(skb, features);
1408 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *,
1410 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *,
1412 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1414 const struct net_offload *ops;
1415 struct sk_buff *pp = NULL;
1416 const struct iphdr *iph;
1424 off = skb_gro_offset(skb);
1425 hlen = off + sizeof(*iph);
1426 iph = skb_gro_header_fast(skb, off);
1427 if (skb_gro_header_hard(skb, hlen)) {
1428 iph = skb_gro_header_slow(skb, hlen, off);
1433 proto = iph->protocol;
1436 ops = rcu_dereference(inet_offloads[proto]);
1437 if (!ops || !ops->callbacks.gro_receive)
1440 if (*(u8 *)iph != 0x45)
1443 if (ip_is_fragment(iph))
1446 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1449 id = ntohl(*(__be32 *)&iph->id);
1450 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1453 list_for_each_entry(p, head, list) {
1457 if (!NAPI_GRO_CB(p)->same_flow)
1460 iph2 = (struct iphdr *)(p->data + off);
1461 /* The above works because, with the exception of the top
1462 * (inner most) layer, we only aggregate pkts with the same
1463 * hdr length so all the hdrs we'll need to verify will start
1464 * at the same offset.
1466 if ((iph->protocol ^ iph2->protocol) |
1467 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1468 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1469 NAPI_GRO_CB(p)->same_flow = 0;
1473 /* All fields must match except length and checksum. */
1474 NAPI_GRO_CB(p)->flush |=
1475 (iph->ttl ^ iph2->ttl) |
1476 (iph->tos ^ iph2->tos) |
1477 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1479 NAPI_GRO_CB(p)->flush |= flush;
1481 /* We need to store of the IP ID check to be included later
1482 * when we can verify that this packet does in fact belong
1485 flush_id = (u16)(id - ntohs(iph2->id));
1487 /* This bit of code makes it much easier for us to identify
1488 * the cases where we are doing atomic vs non-atomic IP ID
1489 * checks. Specifically an atomic check can return IP ID
1490 * values 0 - 0xFFFF, while a non-atomic check can only
1491 * return 0 or 0xFFFF.
1493 if (!NAPI_GRO_CB(p)->is_atomic ||
1494 !(iph->frag_off & htons(IP_DF))) {
1495 flush_id ^= NAPI_GRO_CB(p)->count;
1496 flush_id = flush_id ? 0xFFFF : 0;
1499 /* If the previous IP ID value was based on an atomic
1500 * datagram we can overwrite the value and ignore it.
1502 if (NAPI_GRO_CB(skb)->is_atomic)
1503 NAPI_GRO_CB(p)->flush_id = flush_id;
1505 NAPI_GRO_CB(p)->flush_id |= flush_id;
1508 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1509 NAPI_GRO_CB(skb)->flush |= flush;
1510 skb_set_network_header(skb, off);
1511 /* The above will be needed by the transport layer if there is one
1512 * immediately following this IP hdr.
1515 /* Note : No need to call skb_gro_postpull_rcsum() here,
1516 * as we already checked checksum over ipv4 header was 0
1518 skb_gro_pull(skb, sizeof(*iph));
1519 skb_set_transport_header(skb, skb_gro_offset(skb));
1521 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1522 ops->callbacks.gro_receive, head, skb);
1528 skb_gro_flush_final(skb, pp, flush);
1532 EXPORT_SYMBOL(inet_gro_receive);
1534 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1535 struct sk_buff *skb)
1537 if (NAPI_GRO_CB(skb)->encap_mark) {
1538 NAPI_GRO_CB(skb)->flush = 1;
1542 NAPI_GRO_CB(skb)->encap_mark = 1;
1544 return inet_gro_receive(head, skb);
1547 #define SECONDS_PER_DAY 86400
1549 /* inet_current_timestamp - Return IP network timestamp
1551 * Return milliseconds since midnight in network byte order.
1553 __be32 inet_current_timestamp(void)
1557 struct timespec64 ts;
1559 ktime_get_real_ts64(&ts);
1561 /* Get secs since midnight. */
1562 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1563 /* Convert to msecs. */
1564 msecs = secs * MSEC_PER_SEC;
1565 /* Convert nsec to msec. */
1566 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1568 /* Convert to network byte order. */
1569 return htonl(msecs);
1571 EXPORT_SYMBOL(inet_current_timestamp);
1573 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1575 if (sk->sk_family == AF_INET)
1576 return ip_recv_error(sk, msg, len, addr_len);
1577 #if IS_ENABLED(CONFIG_IPV6)
1578 if (sk->sk_family == AF_INET6)
1579 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1584 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *, int));
1585 INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int));
1586 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1588 __be16 newlen = htons(skb->len - nhoff);
1589 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1590 const struct net_offload *ops;
1591 int proto = iph->protocol;
1594 if (skb->encapsulation) {
1595 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1596 skb_set_inner_network_header(skb, nhoff);
1599 csum_replace2(&iph->check, iph->tot_len, newlen);
1600 iph->tot_len = newlen;
1603 ops = rcu_dereference(inet_offloads[proto]);
1604 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1607 /* Only need to add sizeof(*iph) to get to the next hdr below
1608 * because any hdr with option will have been flushed in
1609 * inet_gro_receive().
1611 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1612 tcp4_gro_complete, udp4_gro_complete,
1613 skb, nhoff + sizeof(*iph));
1620 EXPORT_SYMBOL(inet_gro_complete);
1622 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1624 skb->encapsulation = 1;
1625 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1626 return inet_gro_complete(skb, nhoff);
1629 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1630 unsigned short type, unsigned char protocol,
1633 struct socket *sock;
1634 int rc = sock_create_kern(net, family, type, protocol, &sock);
1638 (*sk)->sk_allocation = GFP_ATOMIC;
1640 * Unhash it so that IP input processing does not even see it,
1641 * we do not wish this socket to see incoming packets.
1643 (*sk)->sk_prot->unhash(*sk);
1647 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1649 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1651 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1653 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1655 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1657 unsigned long res = 0;
1660 for_each_possible_cpu(i)
1661 res += snmp_get_cpu_field(mib, i, offt);
1664 EXPORT_SYMBOL_GPL(snmp_fold_field);
1666 #if BITS_PER_LONG==32
1668 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1669 size_t syncp_offset)
1672 struct u64_stats_sync *syncp;
1676 bhptr = per_cpu_ptr(mib, cpu);
1677 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1679 start = u64_stats_fetch_begin_irq(syncp);
1680 v = *(((u64 *)bhptr) + offt);
1681 } while (u64_stats_fetch_retry_irq(syncp, start));
1685 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1687 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1692 for_each_possible_cpu(cpu) {
1693 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1697 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1700 #ifdef CONFIG_IP_MULTICAST
1701 static const struct net_protocol igmp_protocol = {
1702 .handler = igmp_rcv,
1707 static const struct net_protocol tcp_protocol = {
1708 .handler = tcp_v4_rcv,
1709 .err_handler = tcp_v4_err,
1712 .icmp_strict_tag_validation = 1,
1715 static const struct net_protocol udp_protocol = {
1717 .err_handler = udp_err,
1722 static const struct net_protocol icmp_protocol = {
1723 .handler = icmp_rcv,
1724 .err_handler = icmp_err,
1729 static __net_init int ipv4_mib_init_net(struct net *net)
1733 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1734 if (!net->mib.tcp_statistics)
1736 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1737 if (!net->mib.ip_statistics)
1740 for_each_possible_cpu(i) {
1741 struct ipstats_mib *af_inet_stats;
1742 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1743 u64_stats_init(&af_inet_stats->syncp);
1746 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1747 if (!net->mib.net_statistics)
1749 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1750 if (!net->mib.udp_statistics)
1752 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1753 if (!net->mib.udplite_statistics)
1754 goto err_udplite_mib;
1755 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1756 if (!net->mib.icmp_statistics)
1758 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1760 if (!net->mib.icmpmsg_statistics)
1761 goto err_icmpmsg_mib;
1767 free_percpu(net->mib.icmp_statistics);
1769 free_percpu(net->mib.udplite_statistics);
1771 free_percpu(net->mib.udp_statistics);
1773 free_percpu(net->mib.net_statistics);
1775 free_percpu(net->mib.ip_statistics);
1777 free_percpu(net->mib.tcp_statistics);
1782 static __net_exit void ipv4_mib_exit_net(struct net *net)
1784 kfree(net->mib.icmpmsg_statistics);
1785 free_percpu(net->mib.icmp_statistics);
1786 free_percpu(net->mib.udplite_statistics);
1787 free_percpu(net->mib.udp_statistics);
1788 free_percpu(net->mib.net_statistics);
1789 free_percpu(net->mib.ip_statistics);
1790 free_percpu(net->mib.tcp_statistics);
1793 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1794 .init = ipv4_mib_init_net,
1795 .exit = ipv4_mib_exit_net,
1798 static int __init init_ipv4_mibs(void)
1800 return register_pernet_subsys(&ipv4_mib_ops);
1803 static __net_init int inet_init_net(struct net *net)
1806 * Set defaults for local port range
1808 seqlock_init(&net->ipv4.ip_local_ports.lock);
1809 net->ipv4.ip_local_ports.range[0] = 32768;
1810 net->ipv4.ip_local_ports.range[1] = 60999;
1812 seqlock_init(&net->ipv4.ping_group_range.lock);
1814 * Sane defaults - nobody may create ping sockets.
1815 * Boot scripts should set this to distro-specific group.
1817 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1818 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1820 /* Default values for sysctl-controlled parameters.
1821 * We set them here, in case sysctl is not compiled.
1823 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1824 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1825 net->ipv4.sysctl_ip_dynaddr = 0;
1826 net->ipv4.sysctl_ip_early_demux = 1;
1827 net->ipv4.sysctl_udp_early_demux = 1;
1828 net->ipv4.sysctl_tcp_early_demux = 1;
1829 #ifdef CONFIG_SYSCTL
1830 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1833 /* Some igmp sysctl, whose values are always used */
1834 net->ipv4.sysctl_igmp_max_memberships = 20;
1835 net->ipv4.sysctl_igmp_max_msf = 10;
1836 /* IGMP reports for link-local multicast groups are enabled by default */
1837 net->ipv4.sysctl_igmp_llm_reports = 1;
1838 net->ipv4.sysctl_igmp_qrv = 2;
1843 static __net_initdata struct pernet_operations af_inet_ops = {
1844 .init = inet_init_net,
1847 static int __init init_inet_pernet_ops(void)
1849 return register_pernet_subsys(&af_inet_ops);
1852 static int ipv4_proc_init(void);
1855 * IP protocol layer initialiser
1858 static struct packet_offload ip_packet_offload __read_mostly = {
1859 .type = cpu_to_be16(ETH_P_IP),
1861 .gso_segment = inet_gso_segment,
1862 .gro_receive = inet_gro_receive,
1863 .gro_complete = inet_gro_complete,
1867 static const struct net_offload ipip_offload = {
1869 .gso_segment = ipip_gso_segment,
1870 .gro_receive = ipip_gro_receive,
1871 .gro_complete = ipip_gro_complete,
1875 static int __init ipip_offload_init(void)
1877 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1880 static int __init ipv4_offload_init(void)
1885 if (udpv4_offload_init() < 0)
1886 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1887 if (tcpv4_offload_init() < 0)
1888 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1889 if (ipip_offload_init() < 0)
1890 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1892 dev_add_offload(&ip_packet_offload);
1896 fs_initcall(ipv4_offload_init);
1898 static struct packet_type ip_packet_type __read_mostly = {
1899 .type = cpu_to_be16(ETH_P_IP),
1901 .list_func = ip_list_rcv,
1904 static int __init inet_init(void)
1906 struct inet_protosw *q;
1907 struct list_head *r;
1910 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1912 rc = proto_register(&tcp_prot, 1);
1916 rc = proto_register(&udp_prot, 1);
1918 goto out_unregister_tcp_proto;
1920 rc = proto_register(&raw_prot, 1);
1922 goto out_unregister_udp_proto;
1924 rc = proto_register(&ping_prot, 1);
1926 goto out_unregister_raw_proto;
1929 * Tell SOCKET that we are alive...
1932 (void)sock_register(&inet_family_ops);
1934 #ifdef CONFIG_SYSCTL
1935 ip_static_sysctl_init();
1939 * Add all the base protocols.
1942 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1943 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1944 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1945 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1946 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1947 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1948 #ifdef CONFIG_IP_MULTICAST
1949 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1950 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1953 /* Register the socket-side information for inet_create. */
1954 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1957 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1958 inet_register_protosw(q);
1961 * Set the ARP module up
1967 * Set the IP module up
1972 /* Initialise per-cpu ipv4 mibs */
1973 if (init_ipv4_mibs())
1974 panic("%s: Cannot init ipv4 mibs\n", __func__);
1976 /* Setup TCP slab cache for open requests. */
1979 /* Setup UDP memory threshold */
1982 /* Add UDP-Lite (RFC 3828) */
1983 udplite4_register();
1990 * Set the ICMP layer up
1993 if (icmp_init() < 0)
1994 panic("Failed to create the ICMP control socket.\n");
1997 * Initialise the multicast router
1999 #if defined(CONFIG_IP_MROUTE)
2001 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2004 if (init_inet_pernet_ops())
2005 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2011 dev_add_pack(&ip_packet_type);
2013 ip_tunnel_core_init();
2018 out_unregister_raw_proto:
2019 proto_unregister(&raw_prot);
2020 out_unregister_udp_proto:
2021 proto_unregister(&udp_prot);
2022 out_unregister_tcp_proto:
2023 proto_unregister(&tcp_prot);
2027 fs_initcall(inet_init);
2029 /* ------------------------------------------------------------------------ */
2031 #ifdef CONFIG_PROC_FS
2032 static int __init ipv4_proc_init(void)
2036 if (raw_proc_init())
2038 if (tcp4_proc_init())
2040 if (udp4_proc_init())
2042 if (ping_proc_init())
2044 if (ip_misc_proc_init())
2061 #else /* CONFIG_PROC_FS */
2062 static int __init ipv4_proc_init(void)
2066 #endif /* CONFIG_PROC_FS */