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>
106 #include <net/udplite.h>
107 #include <net/ping.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
111 #include <net/icmp.h>
112 #include <net/inet_common.h>
113 #include <net/ip_tunnels.h>
114 #include <net/xfrm.h>
115 #include <net/net_namespace.h>
116 #include <net/secure_seq.h>
117 #ifdef CONFIG_IP_MROUTE
118 #include <linux/mroute.h>
120 #include <net/l3mdev.h>
121 #include <net/compat.h>
123 #include <trace/events/sock.h>
125 /* The inetsw table contains everything that inet_create needs to
126 * build a new socket.
128 static struct list_head inetsw[SOCK_MAX];
129 static DEFINE_SPINLOCK(inetsw_lock);
131 /* New destruction routine */
133 void inet_sock_destruct(struct sock *sk)
135 struct inet_sock *inet = inet_sk(sk);
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
140 sk_mem_reclaim_final(sk);
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 pr_err("Attempt to release TCP socket in state %d %p\n",
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 pr_err("Attempt to release alive inet socket %p\n", sk);
152 WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
153 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
154 WARN_ON_ONCE(sk->sk_wmem_queued);
155 WARN_ON_ONCE(sk_forward_alloc_get(sk));
157 kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
159 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
161 EXPORT_SYMBOL(inet_sock_destruct);
164 * The routines beyond this point handle the behaviour of an AF_INET
165 * socket object. Mostly it punts to the subprotocols of IP to do
170 * Automatically bind an unbound socket.
173 static int inet_autobind(struct sock *sk)
175 struct inet_sock *inet;
176 /* We may need to bind the socket. */
179 if (!inet->inet_num) {
180 if (sk->sk_prot->get_port(sk, 0)) {
184 inet->inet_sport = htons(inet->inet_num);
191 * Move a socket into listening state.
193 int inet_listen(struct socket *sock, int backlog)
195 struct sock *sk = sock->sk;
196 unsigned char old_state;
197 int err, tcp_fastopen;
202 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205 old_state = sk->sk_state;
206 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
213 if (old_state != TCP_LISTEN) {
214 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
215 * Note that only TCP sockets (SOCK_STREAM) will reach here.
216 * Also fastopen backlog may already been set via the option
217 * because the socket was in TCP_LISTEN state previously but
218 * was shutdown() rather than close().
220 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
221 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
222 (tcp_fastopen & TFO_SERVER_ENABLE) &&
223 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
224 fastopen_queue_tune(sk, backlog);
225 tcp_fastopen_init_key_once(sock_net(sk));
228 err = inet_csk_listen_start(sk);
231 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
239 EXPORT_SYMBOL(inet_listen);
242 * Create an inet socket.
245 static int inet_create(struct net *net, struct socket *sock, int protocol,
249 struct inet_protosw *answer;
250 struct inet_sock *inet;
251 struct proto *answer_prot;
252 unsigned char answer_flags;
253 int try_loading_module = 0;
256 if (protocol < 0 || protocol >= IPPROTO_MAX)
259 sock->state = SS_UNCONNECTED;
261 /* Look for the requested type/protocol pair. */
263 err = -ESOCKTNOSUPPORT;
265 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
268 /* Check the non-wild match. */
269 if (protocol == answer->protocol) {
270 if (protocol != IPPROTO_IP)
273 /* Check for the two wild cases. */
274 if (IPPROTO_IP == protocol) {
275 protocol = answer->protocol;
278 if (IPPROTO_IP == answer->protocol)
281 err = -EPROTONOSUPPORT;
285 if (try_loading_module < 2) {
288 * Be more specific, e.g. net-pf-2-proto-132-type-1
289 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
291 if (++try_loading_module == 1)
292 request_module("net-pf-%d-proto-%d-type-%d",
293 PF_INET, protocol, sock->type);
295 * Fall back to generic, e.g. net-pf-2-proto-132
296 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
299 request_module("net-pf-%d-proto-%d",
301 goto lookup_protocol;
307 if (sock->type == SOCK_RAW && !kern &&
308 !ns_capable(net->user_ns, CAP_NET_RAW))
311 sock->ops = answer->ops;
312 answer_prot = answer->prot;
313 answer_flags = answer->flags;
316 WARN_ON(!answer_prot->slab);
319 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
324 if (INET_PROTOSW_REUSE & answer_flags)
325 sk->sk_reuse = SK_CAN_REUSE;
328 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
332 if (SOCK_RAW == sock->type) {
333 inet->inet_num = protocol;
334 if (IPPROTO_RAW == protocol)
338 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
339 inet->pmtudisc = IP_PMTUDISC_DONT;
341 inet->pmtudisc = IP_PMTUDISC_WANT;
345 sock_init_data(sock, sk);
347 sk->sk_destruct = inet_sock_destruct;
348 sk->sk_protocol = protocol;
349 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
350 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
357 inet->mc_list = NULL;
360 if (inet->inet_num) {
361 /* It assumes that any protocol which allows
362 * the user to assign a number at socket
363 * creation time automatically
366 inet->inet_sport = htons(inet->inet_num);
367 /* Add to protocol hash chains. */
368 err = sk->sk_prot->hash(sk);
370 sk_common_release(sk);
375 if (sk->sk_prot->init) {
376 err = sk->sk_prot->init(sk);
378 sk_common_release(sk);
384 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
386 sk_common_release(sk);
399 * The peer socket should always be NULL (or else). When we call this
400 * function we are destroying the object and from then on nobody
401 * should refer to it.
403 int inet_release(struct socket *sock)
405 struct sock *sk = sock->sk;
410 if (!sk->sk_kern_sock)
411 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(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_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
436 u32 flags = BIND_WITH_LOCK;
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_INET_BIND_LOCK(sk, uaddr,
450 CGROUP_INET4_BIND, &flags);
454 return __inet_bind(sk, uaddr, addr_len, flags);
457 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
459 return inet_bind_sk(sock->sk, uaddr, addr_len);
461 EXPORT_SYMBOL(inet_bind);
463 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
466 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
467 struct inet_sock *inet = inet_sk(sk);
468 struct net *net = sock_net(sk);
471 u32 tb_id = RT_TABLE_LOCAL;
474 if (addr->sin_family != AF_INET) {
475 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
476 * only if s_addr is INADDR_ANY.
479 if (addr->sin_family != AF_UNSPEC ||
480 addr->sin_addr.s_addr != htonl(INADDR_ANY))
484 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
485 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
487 /* Not specified by any standard per-se, however it breaks too
488 * many applications when removed. It is unfortunate since
489 * allowing applications to make a non-local bind solves
490 * several problems with systems using dynamic addressing.
491 * (ie. your servers still start up even if your ISDN link
492 * is temporarily down)
494 err = -EADDRNOTAVAIL;
495 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
499 snum = ntohs(addr->sin_port);
501 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
502 snum && inet_port_requires_bind_service(net, snum) &&
503 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
506 /* We keep a pair of addresses. rcv_saddr is the one
507 * used by hash lookups, and saddr is used for transmit.
509 * In the BSD API these are the same except where it
510 * would be illegal to use them (multicast/broadcast) in
511 * which case the sending device address is used.
513 if (flags & BIND_WITH_LOCK)
516 /* Check these errors (active socket, double bind). */
518 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
519 goto out_release_sock;
521 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
522 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
523 inet->inet_saddr = 0; /* Use device */
525 /* Make sure we are allowed to bind here. */
526 if (snum || !(inet->bind_address_no_port ||
527 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
528 err = sk->sk_prot->get_port(sk, snum);
530 inet->inet_saddr = inet->inet_rcv_saddr = 0;
531 goto out_release_sock;
533 if (!(flags & BIND_FROM_BPF)) {
534 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
536 inet->inet_saddr = inet->inet_rcv_saddr = 0;
537 if (sk->sk_prot->put_port)
538 sk->sk_prot->put_port(sk);
539 goto out_release_sock;
544 if (inet->inet_rcv_saddr)
545 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
547 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
548 inet->inet_sport = htons(inet->inet_num);
549 inet->inet_daddr = 0;
550 inet->inet_dport = 0;
554 if (flags & BIND_WITH_LOCK)
560 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
561 int addr_len, int flags)
563 struct sock *sk = sock->sk;
564 const struct proto *prot;
567 if (addr_len < sizeof(uaddr->sa_family))
570 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
571 prot = READ_ONCE(sk->sk_prot);
573 if (uaddr->sa_family == AF_UNSPEC)
574 return prot->disconnect(sk, flags);
576 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
577 err = prot->pre_connect(sk, uaddr, addr_len);
582 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
584 return prot->connect(sk, uaddr, addr_len);
586 EXPORT_SYMBOL(inet_dgram_connect);
588 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
590 DEFINE_WAIT_FUNC(wait, woken_wake_function);
592 add_wait_queue(sk_sleep(sk), &wait);
593 sk->sk_write_pending += writebias;
594 sk->sk_wait_pending++;
596 /* Basic assumption: if someone sets sk->sk_err, he _must_
597 * change state of the socket from TCP_SYN_*.
598 * Connect() does not allow to get error notifications
599 * without closing the socket.
601 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
603 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
605 if (signal_pending(current) || !timeo)
608 remove_wait_queue(sk_sleep(sk), &wait);
609 sk->sk_write_pending -= writebias;
610 sk->sk_wait_pending--;
615 * Connect to a remote host. There is regrettably still a little
616 * TCP 'magic' in here.
618 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
619 int addr_len, int flags, int is_sendmsg)
621 struct sock *sk = sock->sk;
626 * uaddr can be NULL and addr_len can be 0 if:
627 * sk is a TCP fastopen active socket and
628 * TCP_FASTOPEN_CONNECT sockopt is set and
629 * we already have a valid cookie for this socket.
630 * In this case, user can call write() after connect().
631 * write() will invoke tcp_sendmsg_fastopen() which calls
632 * __inet_stream_connect().
635 if (addr_len < sizeof(uaddr->sa_family))
638 if (uaddr->sa_family == AF_UNSPEC) {
639 err = sk->sk_prot->disconnect(sk, flags);
640 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
645 switch (sock->state) {
653 if (inet_sk(sk)->defer_connect)
654 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
657 /* Fall out of switch with err, set for this state */
661 if (sk->sk_state != TCP_CLOSE)
664 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
665 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
670 err = sk->sk_prot->connect(sk, uaddr, addr_len);
674 sock->state = SS_CONNECTING;
676 if (!err && inet_sk(sk)->defer_connect)
679 /* Just entered SS_CONNECTING state; the only
680 * difference is that return value in non-blocking
681 * case is EINPROGRESS, rather than EALREADY.
687 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
689 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
690 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
691 tcp_sk(sk)->fastopen_req &&
692 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
694 /* Error code is set above */
695 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
698 err = sock_intr_errno(timeo);
699 if (signal_pending(current))
703 /* Connection was closed by RST, timeout, ICMP error
704 * or another process disconnected us.
706 if (sk->sk_state == TCP_CLOSE)
709 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
710 * and error was received after socket entered established state.
711 * Hence, it is handled normally after connect() return successfully.
714 sock->state = SS_CONNECTED;
720 err = sock_error(sk) ? : -ECONNABORTED;
721 sock->state = SS_UNCONNECTED;
722 if (sk->sk_prot->disconnect(sk, flags))
723 sock->state = SS_DISCONNECTING;
726 EXPORT_SYMBOL(__inet_stream_connect);
728 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
729 int addr_len, int flags)
734 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
735 release_sock(sock->sk);
738 EXPORT_SYMBOL(inet_stream_connect);
740 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
742 sock_rps_record_flow(newsk);
743 WARN_ON(!((1 << newsk->sk_state) &
744 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
745 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
747 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
748 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
749 sock_graft(newsk, newsock);
751 newsock->state = SS_CONNECTED;
755 * Accept a pending connection. The TCP layer now gives BSD semantics.
758 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
761 struct sock *sk1 = sock->sk, *sk2;
764 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
765 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
770 __inet_accept(sock, newsock, sk2);
774 EXPORT_SYMBOL(inet_accept);
777 * This does both peername and sockname.
779 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
782 struct sock *sk = sock->sk;
783 struct inet_sock *inet = inet_sk(sk);
784 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
786 sin->sin_family = AF_INET;
789 if (!inet->inet_dport ||
790 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
795 sin->sin_port = inet->inet_dport;
796 sin->sin_addr.s_addr = inet->inet_daddr;
797 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
798 CGROUP_INET4_GETPEERNAME);
800 __be32 addr = inet->inet_rcv_saddr;
802 addr = inet->inet_saddr;
803 sin->sin_port = inet->inet_sport;
804 sin->sin_addr.s_addr = addr;
805 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
806 CGROUP_INET4_GETSOCKNAME);
809 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
812 EXPORT_SYMBOL(inet_getname);
814 int inet_send_prepare(struct sock *sk)
816 sock_rps_record_flow(sk);
818 /* We may need to bind the socket. */
819 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
825 EXPORT_SYMBOL_GPL(inet_send_prepare);
827 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
829 struct sock *sk = sock->sk;
831 if (unlikely(inet_send_prepare(sk)))
834 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
837 EXPORT_SYMBOL(inet_sendmsg);
839 void inet_splice_eof(struct socket *sock)
841 const struct proto *prot;
842 struct sock *sk = sock->sk;
844 if (unlikely(inet_send_prepare(sk)))
847 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
848 prot = READ_ONCE(sk->sk_prot);
849 if (prot->splice_eof)
850 prot->splice_eof(sock);
852 EXPORT_SYMBOL_GPL(inet_splice_eof);
854 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
855 size_t, int, int *));
856 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
859 struct sock *sk = sock->sk;
863 if (likely(!(flags & MSG_ERRQUEUE)))
864 sock_rps_record_flow(sk);
866 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
867 sk, msg, size, flags, &addr_len);
869 msg->msg_namelen = addr_len;
872 EXPORT_SYMBOL(inet_recvmsg);
874 int inet_shutdown(struct socket *sock, int how)
876 struct sock *sk = sock->sk;
879 /* This should really check to make sure
880 * the socket is a TCP socket. (WHY AC...)
882 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
885 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
889 if (sock->state == SS_CONNECTING) {
890 if ((1 << sk->sk_state) &
891 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
892 sock->state = SS_DISCONNECTING;
894 sock->state = SS_CONNECTED;
897 switch (sk->sk_state) {
900 /* Hack to wake up other listeners, who can poll for
901 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
904 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
905 if (sk->sk_prot->shutdown)
906 sk->sk_prot->shutdown(sk, how);
909 /* Remaining two branches are temporary solution for missing
910 * close() in multithreaded environment. It is _not_ a good idea,
911 * but we have no choice until close() is repaired at VFS level.
914 if (!(how & RCV_SHUTDOWN))
918 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
919 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
923 /* Wake up anyone sleeping in poll. */
924 sk->sk_state_change(sk);
928 EXPORT_SYMBOL(inet_shutdown);
931 * ioctl() calls you can issue on an INET socket. Most of these are
932 * device configuration and stuff and very rarely used. Some ioctls
933 * pass on to the socket itself.
935 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
936 * loads the devconfigure module does its configuring and unloads it.
937 * There's a good 20K of config code hanging around the kernel.
940 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
942 struct sock *sk = sock->sk;
944 struct net *net = sock_net(sk);
945 void __user *p = (void __user *)arg;
952 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
954 err = ip_rt_ioctl(net, cmd, &rt);
962 err = arp_ioctl(net, cmd, (void __user *)arg);
969 if (get_user_ifreq(&ifr, NULL, p))
971 err = devinet_ioctl(net, cmd, &ifr);
972 if (!err && put_user_ifreq(&ifr, p))
982 if (get_user_ifreq(&ifr, NULL, p))
984 err = devinet_ioctl(net, cmd, &ifr);
987 if (sk->sk_prot->ioctl)
988 err = sk_ioctl(sk, cmd, (void __user *)arg);
995 EXPORT_SYMBOL(inet_ioctl);
998 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
999 struct compat_rtentry __user *ur)
1001 compat_uptr_t rtdev;
1004 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1005 3 * sizeof(struct sockaddr)) ||
1006 get_user(rt.rt_flags, &ur->rt_flags) ||
1007 get_user(rt.rt_metric, &ur->rt_metric) ||
1008 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1009 get_user(rt.rt_window, &ur->rt_window) ||
1010 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1011 get_user(rtdev, &ur->rt_dev))
1014 rt.rt_dev = compat_ptr(rtdev);
1015 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1018 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1020 void __user *argp = compat_ptr(arg);
1021 struct sock *sk = sock->sk;
1026 return inet_compat_routing_ioctl(sk, cmd, argp);
1028 if (!sk->sk_prot->compat_ioctl)
1029 return -ENOIOCTLCMD;
1030 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1033 #endif /* CONFIG_COMPAT */
1035 const struct proto_ops inet_stream_ops = {
1037 .owner = THIS_MODULE,
1038 .release = inet_release,
1040 .connect = inet_stream_connect,
1041 .socketpair = sock_no_socketpair,
1042 .accept = inet_accept,
1043 .getname = inet_getname,
1045 .ioctl = inet_ioctl,
1046 .gettstamp = sock_gettstamp,
1047 .listen = inet_listen,
1048 .shutdown = inet_shutdown,
1049 .setsockopt = sock_common_setsockopt,
1050 .getsockopt = sock_common_getsockopt,
1051 .sendmsg = inet_sendmsg,
1052 .recvmsg = inet_recvmsg,
1056 .splice_eof = inet_splice_eof,
1057 .splice_read = tcp_splice_read,
1058 .read_sock = tcp_read_sock,
1059 .read_skb = tcp_read_skb,
1060 .sendmsg_locked = tcp_sendmsg_locked,
1061 .peek_len = tcp_peek_len,
1062 #ifdef CONFIG_COMPAT
1063 .compat_ioctl = inet_compat_ioctl,
1065 .set_rcvlowat = tcp_set_rcvlowat,
1067 EXPORT_SYMBOL(inet_stream_ops);
1069 const struct proto_ops inet_dgram_ops = {
1071 .owner = THIS_MODULE,
1072 .release = inet_release,
1074 .connect = inet_dgram_connect,
1075 .socketpair = sock_no_socketpair,
1076 .accept = sock_no_accept,
1077 .getname = inet_getname,
1079 .ioctl = inet_ioctl,
1080 .gettstamp = sock_gettstamp,
1081 .listen = sock_no_listen,
1082 .shutdown = inet_shutdown,
1083 .setsockopt = sock_common_setsockopt,
1084 .getsockopt = sock_common_getsockopt,
1085 .sendmsg = inet_sendmsg,
1086 .read_skb = udp_read_skb,
1087 .recvmsg = inet_recvmsg,
1088 .mmap = sock_no_mmap,
1089 .splice_eof = inet_splice_eof,
1090 .set_peek_off = sk_set_peek_off,
1091 #ifdef CONFIG_COMPAT
1092 .compat_ioctl = inet_compat_ioctl,
1095 EXPORT_SYMBOL(inet_dgram_ops);
1098 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1101 static const struct proto_ops inet_sockraw_ops = {
1103 .owner = THIS_MODULE,
1104 .release = inet_release,
1106 .connect = inet_dgram_connect,
1107 .socketpair = sock_no_socketpair,
1108 .accept = sock_no_accept,
1109 .getname = inet_getname,
1110 .poll = datagram_poll,
1111 .ioctl = inet_ioctl,
1112 .gettstamp = sock_gettstamp,
1113 .listen = sock_no_listen,
1114 .shutdown = inet_shutdown,
1115 .setsockopt = sock_common_setsockopt,
1116 .getsockopt = sock_common_getsockopt,
1117 .sendmsg = inet_sendmsg,
1118 .recvmsg = inet_recvmsg,
1119 .mmap = sock_no_mmap,
1120 .splice_eof = inet_splice_eof,
1121 #ifdef CONFIG_COMPAT
1122 .compat_ioctl = inet_compat_ioctl,
1126 static const struct net_proto_family inet_family_ops = {
1128 .create = inet_create,
1129 .owner = THIS_MODULE,
1132 /* Upon startup we insert all the elements in inetsw_array[] into
1133 * the linked list inetsw.
1135 static struct inet_protosw inetsw_array[] =
1138 .type = SOCK_STREAM,
1139 .protocol = IPPROTO_TCP,
1141 .ops = &inet_stream_ops,
1142 .flags = INET_PROTOSW_PERMANENT |
1148 .protocol = IPPROTO_UDP,
1150 .ops = &inet_dgram_ops,
1151 .flags = INET_PROTOSW_PERMANENT,
1156 .protocol = IPPROTO_ICMP,
1158 .ops = &inet_sockraw_ops,
1159 .flags = INET_PROTOSW_REUSE,
1164 .protocol = IPPROTO_IP, /* wild card */
1166 .ops = &inet_sockraw_ops,
1167 .flags = INET_PROTOSW_REUSE,
1171 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1173 void inet_register_protosw(struct inet_protosw *p)
1175 struct list_head *lh;
1176 struct inet_protosw *answer;
1177 int protocol = p->protocol;
1178 struct list_head *last_perm;
1180 spin_lock_bh(&inetsw_lock);
1182 if (p->type >= SOCK_MAX)
1185 /* If we are trying to override a permanent protocol, bail. */
1186 last_perm = &inetsw[p->type];
1187 list_for_each(lh, &inetsw[p->type]) {
1188 answer = list_entry(lh, struct inet_protosw, list);
1189 /* Check only the non-wild match. */
1190 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1192 if (protocol == answer->protocol)
1197 /* Add the new entry after the last permanent entry if any, so that
1198 * the new entry does not override a permanent entry when matched with
1199 * a wild-card protocol. But it is allowed to override any existing
1200 * non-permanent entry. This means that when we remove this entry, the
1201 * system automatically returns to the old behavior.
1203 list_add_rcu(&p->list, last_perm);
1205 spin_unlock_bh(&inetsw_lock);
1210 pr_err("Attempt to override permanent protocol %d\n", protocol);
1214 pr_err("Ignoring attempt to register invalid socket type %d\n",
1218 EXPORT_SYMBOL(inet_register_protosw);
1220 void inet_unregister_protosw(struct inet_protosw *p)
1222 if (INET_PROTOSW_PERMANENT & p->flags) {
1223 pr_err("Attempt to unregister permanent protocol %d\n",
1226 spin_lock_bh(&inetsw_lock);
1227 list_del_rcu(&p->list);
1228 spin_unlock_bh(&inetsw_lock);
1233 EXPORT_SYMBOL(inet_unregister_protosw);
1235 static int inet_sk_reselect_saddr(struct sock *sk)
1237 struct inet_sock *inet = inet_sk(sk);
1238 __be32 old_saddr = inet->inet_saddr;
1239 __be32 daddr = inet->inet_daddr;
1243 struct ip_options_rcu *inet_opt;
1246 inet_opt = rcu_dereference_protected(inet->inet_opt,
1247 lockdep_sock_is_held(sk));
1248 if (inet_opt && inet_opt->opt.srr)
1249 daddr = inet_opt->opt.faddr;
1251 /* Query new route. */
1252 fl4 = &inet->cork.fl.u.ip4;
1253 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1254 sk->sk_protocol, inet->inet_sport,
1255 inet->inet_dport, sk);
1259 new_saddr = fl4->saddr;
1261 if (new_saddr == old_saddr) {
1262 sk_setup_caps(sk, &rt->dst);
1266 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1272 sk_setup_caps(sk, &rt->dst);
1274 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1275 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1276 __func__, &old_saddr, &new_saddr);
1280 * XXX The only one ugly spot where we need to
1281 * XXX really change the sockets identity after
1282 * XXX it has entered the hashes. -DaveM
1284 * Besides that, it does not check for connection
1285 * uniqueness. Wait for troubles.
1287 return __sk_prot_rehash(sk);
1290 int inet_sk_rebuild_header(struct sock *sk)
1292 struct inet_sock *inet = inet_sk(sk);
1293 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1295 struct ip_options_rcu *inet_opt;
1299 /* Route is OK, nothing to do. */
1305 inet_opt = rcu_dereference(inet->inet_opt);
1306 daddr = inet->inet_daddr;
1307 if (inet_opt && inet_opt->opt.srr)
1308 daddr = inet_opt->opt.faddr;
1310 fl4 = &inet->cork.fl.u.ip4;
1311 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1312 inet->inet_dport, inet->inet_sport,
1313 sk->sk_protocol, RT_CONN_FLAGS(sk),
1314 sk->sk_bound_dev_if);
1317 sk_setup_caps(sk, &rt->dst);
1321 /* Routing failed... */
1322 sk->sk_route_caps = 0;
1324 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1325 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1327 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1328 sk->sk_state != TCP_SYN_SENT ||
1329 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1330 (err = inet_sk_reselect_saddr(sk)) != 0)
1331 WRITE_ONCE(sk->sk_err_soft, -err);
1336 EXPORT_SYMBOL(inet_sk_rebuild_header);
1338 void inet_sk_set_state(struct sock *sk, int state)
1340 trace_inet_sock_set_state(sk, sk->sk_state, state);
1341 sk->sk_state = state;
1343 EXPORT_SYMBOL(inet_sk_set_state);
1345 void inet_sk_state_store(struct sock *sk, int newstate)
1347 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1348 smp_store_release(&sk->sk_state, newstate);
1351 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1352 netdev_features_t features)
1354 bool udpfrag = false, fixedid = false, gso_partial, encap;
1355 struct sk_buff *segs = ERR_PTR(-EINVAL);
1356 const struct net_offload *ops;
1357 unsigned int offset = 0;
1364 skb_reset_network_header(skb);
1365 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1366 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1371 if (ihl < sizeof(*iph))
1374 id = ntohs(iph->id);
1375 proto = iph->protocol;
1377 /* Warning: after this point, iph might be no longer valid */
1378 if (unlikely(!pskb_may_pull(skb, ihl)))
1380 __skb_pull(skb, ihl);
1382 encap = SKB_GSO_CB(skb)->encap_level > 0;
1384 features &= skb->dev->hw_enc_features;
1385 SKB_GSO_CB(skb)->encap_level += ihl;
1387 skb_reset_transport_header(skb);
1389 segs = ERR_PTR(-EPROTONOSUPPORT);
1391 if (!skb->encapsulation || encap) {
1392 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1393 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1395 /* fixed ID is invalid if DF bit is not set */
1396 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1400 ops = rcu_dereference(inet_offloads[proto]);
1401 if (likely(ops && ops->callbacks.gso_segment)) {
1402 segs = ops->callbacks.gso_segment(skb, features);
1404 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1407 if (IS_ERR_OR_NULL(segs))
1410 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1414 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1416 iph->frag_off = htons(offset >> 3);
1418 iph->frag_off |= htons(IP_MF);
1419 offset += skb->len - nhoff - ihl;
1420 tot_len = skb->len - nhoff;
1421 } else if (skb_is_gso(skb)) {
1423 iph->id = htons(id);
1424 id += skb_shinfo(skb)->gso_segs;
1428 tot_len = skb_shinfo(skb)->gso_size +
1429 SKB_GSO_CB(skb)->data_offset +
1430 skb->head - (unsigned char *)iph;
1432 tot_len = skb->len - nhoff;
1435 iph->id = htons(id++);
1436 tot_len = skb->len - nhoff;
1438 iph->tot_len = htons(tot_len);
1441 skb_reset_inner_headers(skb);
1442 skb->network_header = (u8 *)iph - skb->head;
1443 skb_reset_mac_len(skb);
1444 } while ((skb = skb->next));
1450 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1451 netdev_features_t features)
1453 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1454 return ERR_PTR(-EINVAL);
1456 return inet_gso_segment(skb, features);
1459 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1461 const struct net_offload *ops;
1462 struct sk_buff *pp = NULL;
1463 const struct iphdr *iph;
1471 off = skb_gro_offset(skb);
1472 hlen = off + sizeof(*iph);
1473 iph = skb_gro_header(skb, hlen, off);
1477 proto = iph->protocol;
1479 ops = rcu_dereference(inet_offloads[proto]);
1480 if (!ops || !ops->callbacks.gro_receive)
1483 if (*(u8 *)iph != 0x45)
1486 if (ip_is_fragment(iph))
1489 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1492 NAPI_GRO_CB(skb)->proto = proto;
1493 id = ntohl(*(__be32 *)&iph->id);
1494 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1497 list_for_each_entry(p, head, list) {
1501 if (!NAPI_GRO_CB(p)->same_flow)
1504 iph2 = (struct iphdr *)(p->data + off);
1505 /* The above works because, with the exception of the top
1506 * (inner most) layer, we only aggregate pkts with the same
1507 * hdr length so all the hdrs we'll need to verify will start
1508 * at the same offset.
1510 if ((iph->protocol ^ iph2->protocol) |
1511 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1512 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1513 NAPI_GRO_CB(p)->same_flow = 0;
1517 /* All fields must match except length and checksum. */
1518 NAPI_GRO_CB(p)->flush |=
1519 (iph->ttl ^ iph2->ttl) |
1520 (iph->tos ^ iph2->tos) |
1521 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1523 NAPI_GRO_CB(p)->flush |= flush;
1525 /* We need to store of the IP ID check to be included later
1526 * when we can verify that this packet does in fact belong
1529 flush_id = (u16)(id - ntohs(iph2->id));
1531 /* This bit of code makes it much easier for us to identify
1532 * the cases where we are doing atomic vs non-atomic IP ID
1533 * checks. Specifically an atomic check can return IP ID
1534 * values 0 - 0xFFFF, while a non-atomic check can only
1535 * return 0 or 0xFFFF.
1537 if (!NAPI_GRO_CB(p)->is_atomic ||
1538 !(iph->frag_off & htons(IP_DF))) {
1539 flush_id ^= NAPI_GRO_CB(p)->count;
1540 flush_id = flush_id ? 0xFFFF : 0;
1543 /* If the previous IP ID value was based on an atomic
1544 * datagram we can overwrite the value and ignore it.
1546 if (NAPI_GRO_CB(skb)->is_atomic)
1547 NAPI_GRO_CB(p)->flush_id = flush_id;
1549 NAPI_GRO_CB(p)->flush_id |= flush_id;
1552 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1553 NAPI_GRO_CB(skb)->flush |= flush;
1554 skb_set_network_header(skb, off);
1555 /* The above will be needed by the transport layer if there is one
1556 * immediately following this IP hdr.
1559 /* Note : No need to call skb_gro_postpull_rcsum() here,
1560 * as we already checked checksum over ipv4 header was 0
1562 skb_gro_pull(skb, sizeof(*iph));
1563 skb_set_transport_header(skb, skb_gro_offset(skb));
1565 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1566 ops->callbacks.gro_receive, head, skb);
1569 skb_gro_flush_final(skb, pp, flush);
1574 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1575 struct sk_buff *skb)
1577 if (NAPI_GRO_CB(skb)->encap_mark) {
1578 NAPI_GRO_CB(skb)->flush = 1;
1582 NAPI_GRO_CB(skb)->encap_mark = 1;
1584 return inet_gro_receive(head, skb);
1587 #define SECONDS_PER_DAY 86400
1589 /* inet_current_timestamp - Return IP network timestamp
1591 * Return milliseconds since midnight in network byte order.
1593 __be32 inet_current_timestamp(void)
1597 struct timespec64 ts;
1599 ktime_get_real_ts64(&ts);
1601 /* Get secs since midnight. */
1602 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1603 /* Convert to msecs. */
1604 msecs = secs * MSEC_PER_SEC;
1605 /* Convert nsec to msec. */
1606 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1608 /* Convert to network byte order. */
1609 return htonl(msecs);
1611 EXPORT_SYMBOL(inet_current_timestamp);
1613 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1615 if (sk->sk_family == AF_INET)
1616 return ip_recv_error(sk, msg, len, addr_len);
1617 #if IS_ENABLED(CONFIG_IPV6)
1618 if (sk->sk_family == AF_INET6)
1619 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1624 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1626 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1627 const struct net_offload *ops;
1628 __be16 totlen = iph->tot_len;
1629 int proto = iph->protocol;
1632 if (skb->encapsulation) {
1633 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1634 skb_set_inner_network_header(skb, nhoff);
1637 iph_set_totlen(iph, skb->len - nhoff);
1638 csum_replace2(&iph->check, totlen, iph->tot_len);
1640 ops = rcu_dereference(inet_offloads[proto]);
1641 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1644 /* Only need to add sizeof(*iph) to get to the next hdr below
1645 * because any hdr with option will have been flushed in
1646 * inet_gro_receive().
1648 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1649 tcp4_gro_complete, udp4_gro_complete,
1650 skb, nhoff + sizeof(*iph));
1656 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1658 skb->encapsulation = 1;
1659 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1660 return inet_gro_complete(skb, nhoff);
1663 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1664 unsigned short type, unsigned char protocol,
1667 struct socket *sock;
1668 int rc = sock_create_kern(net, family, type, protocol, &sock);
1672 (*sk)->sk_allocation = GFP_ATOMIC;
1673 (*sk)->sk_use_task_frag = false;
1675 * Unhash it so that IP input processing does not even see it,
1676 * we do not wish this socket to see incoming packets.
1678 (*sk)->sk_prot->unhash(*sk);
1682 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1684 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1686 unsigned long res = 0;
1689 for_each_possible_cpu(i)
1690 res += snmp_get_cpu_field(mib, i, offt);
1693 EXPORT_SYMBOL_GPL(snmp_fold_field);
1695 #if BITS_PER_LONG==32
1697 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1698 size_t syncp_offset)
1701 struct u64_stats_sync *syncp;
1705 bhptr = per_cpu_ptr(mib, cpu);
1706 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1708 start = u64_stats_fetch_begin(syncp);
1709 v = *(((u64 *)bhptr) + offt);
1710 } while (u64_stats_fetch_retry(syncp, start));
1714 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1716 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1721 for_each_possible_cpu(cpu) {
1722 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1726 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1729 #ifdef CONFIG_IP_MULTICAST
1730 static const struct net_protocol igmp_protocol = {
1731 .handler = igmp_rcv,
1735 static const struct net_protocol tcp_protocol = {
1736 .handler = tcp_v4_rcv,
1737 .err_handler = tcp_v4_err,
1739 .icmp_strict_tag_validation = 1,
1742 static const struct net_protocol udp_protocol = {
1744 .err_handler = udp_err,
1748 static const struct net_protocol icmp_protocol = {
1749 .handler = icmp_rcv,
1750 .err_handler = icmp_err,
1754 static __net_init int ipv4_mib_init_net(struct net *net)
1758 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1759 if (!net->mib.tcp_statistics)
1761 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1762 if (!net->mib.ip_statistics)
1765 for_each_possible_cpu(i) {
1766 struct ipstats_mib *af_inet_stats;
1767 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1768 u64_stats_init(&af_inet_stats->syncp);
1771 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1772 if (!net->mib.net_statistics)
1774 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1775 if (!net->mib.udp_statistics)
1777 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1778 if (!net->mib.udplite_statistics)
1779 goto err_udplite_mib;
1780 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1781 if (!net->mib.icmp_statistics)
1783 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1785 if (!net->mib.icmpmsg_statistics)
1786 goto err_icmpmsg_mib;
1792 free_percpu(net->mib.icmp_statistics);
1794 free_percpu(net->mib.udplite_statistics);
1796 free_percpu(net->mib.udp_statistics);
1798 free_percpu(net->mib.net_statistics);
1800 free_percpu(net->mib.ip_statistics);
1802 free_percpu(net->mib.tcp_statistics);
1807 static __net_exit void ipv4_mib_exit_net(struct net *net)
1809 kfree(net->mib.icmpmsg_statistics);
1810 free_percpu(net->mib.icmp_statistics);
1811 free_percpu(net->mib.udplite_statistics);
1812 free_percpu(net->mib.udp_statistics);
1813 free_percpu(net->mib.net_statistics);
1814 free_percpu(net->mib.ip_statistics);
1815 free_percpu(net->mib.tcp_statistics);
1817 /* allocated on demand, see mptcp_init_sock() */
1818 free_percpu(net->mib.mptcp_statistics);
1822 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1823 .init = ipv4_mib_init_net,
1824 .exit = ipv4_mib_exit_net,
1827 static int __init init_ipv4_mibs(void)
1829 return register_pernet_subsys(&ipv4_mib_ops);
1832 static __net_init int inet_init_net(struct net *net)
1835 * Set defaults for local port range
1837 seqlock_init(&net->ipv4.ip_local_ports.lock);
1838 net->ipv4.ip_local_ports.range[0] = 32768;
1839 net->ipv4.ip_local_ports.range[1] = 60999;
1841 seqlock_init(&net->ipv4.ping_group_range.lock);
1843 * Sane defaults - nobody may create ping sockets.
1844 * Boot scripts should set this to distro-specific group.
1846 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1847 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1849 /* Default values for sysctl-controlled parameters.
1850 * We set them here, in case sysctl is not compiled.
1852 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1853 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1854 net->ipv4.sysctl_ip_dynaddr = 0;
1855 net->ipv4.sysctl_ip_early_demux = 1;
1856 net->ipv4.sysctl_udp_early_demux = 1;
1857 net->ipv4.sysctl_tcp_early_demux = 1;
1858 net->ipv4.sysctl_nexthop_compat_mode = 1;
1859 #ifdef CONFIG_SYSCTL
1860 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1863 /* Some igmp sysctl, whose values are always used */
1864 net->ipv4.sysctl_igmp_max_memberships = 20;
1865 net->ipv4.sysctl_igmp_max_msf = 10;
1866 /* IGMP reports for link-local multicast groups are enabled by default */
1867 net->ipv4.sysctl_igmp_llm_reports = 1;
1868 net->ipv4.sysctl_igmp_qrv = 2;
1870 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1875 static __net_initdata struct pernet_operations af_inet_ops = {
1876 .init = inet_init_net,
1879 static int __init init_inet_pernet_ops(void)
1881 return register_pernet_subsys(&af_inet_ops);
1884 static int ipv4_proc_init(void);
1887 * IP protocol layer initialiser
1890 static struct packet_offload ip_packet_offload __read_mostly = {
1891 .type = cpu_to_be16(ETH_P_IP),
1893 .gso_segment = inet_gso_segment,
1894 .gro_receive = inet_gro_receive,
1895 .gro_complete = inet_gro_complete,
1899 static const struct net_offload ipip_offload = {
1901 .gso_segment = ipip_gso_segment,
1902 .gro_receive = ipip_gro_receive,
1903 .gro_complete = ipip_gro_complete,
1907 static int __init ipip_offload_init(void)
1909 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1912 static int __init ipv4_offload_init(void)
1917 if (udpv4_offload_init() < 0)
1918 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1919 if (tcpv4_offload_init() < 0)
1920 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1921 if (ipip_offload_init() < 0)
1922 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1924 dev_add_offload(&ip_packet_offload);
1928 fs_initcall(ipv4_offload_init);
1930 static struct packet_type ip_packet_type __read_mostly = {
1931 .type = cpu_to_be16(ETH_P_IP),
1933 .list_func = ip_list_rcv,
1936 static int __init inet_init(void)
1938 struct inet_protosw *q;
1939 struct list_head *r;
1942 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1944 raw_hashinfo_init(&raw_v4_hashinfo);
1946 rc = proto_register(&tcp_prot, 1);
1950 rc = proto_register(&udp_prot, 1);
1952 goto out_unregister_tcp_proto;
1954 rc = proto_register(&raw_prot, 1);
1956 goto out_unregister_udp_proto;
1958 rc = proto_register(&ping_prot, 1);
1960 goto out_unregister_raw_proto;
1963 * Tell SOCKET that we are alive...
1966 (void)sock_register(&inet_family_ops);
1968 #ifdef CONFIG_SYSCTL
1969 ip_static_sysctl_init();
1973 * Add all the base protocols.
1976 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1977 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1978 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1979 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1980 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1981 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1982 #ifdef CONFIG_IP_MULTICAST
1983 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1984 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1987 /* Register the socket-side information for inet_create. */
1988 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1991 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1992 inet_register_protosw(q);
1995 * Set the ARP module up
2001 * Set the IP module up
2006 /* Initialise per-cpu ipv4 mibs */
2007 if (init_ipv4_mibs())
2008 panic("%s: Cannot init ipv4 mibs\n", __func__);
2010 /* Setup TCP slab cache for open requests. */
2013 /* Setup UDP memory threshold */
2016 /* Add UDP-Lite (RFC 3828) */
2017 udplite4_register();
2024 * Set the ICMP layer up
2027 if (icmp_init() < 0)
2028 panic("Failed to create the ICMP control socket.\n");
2031 * Initialise the multicast router
2033 #if defined(CONFIG_IP_MROUTE)
2035 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2038 if (init_inet_pernet_ops())
2039 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2045 dev_add_pack(&ip_packet_type);
2047 ip_tunnel_core_init();
2052 out_unregister_raw_proto:
2053 proto_unregister(&raw_prot);
2054 out_unregister_udp_proto:
2055 proto_unregister(&udp_prot);
2056 out_unregister_tcp_proto:
2057 proto_unregister(&tcp_prot);
2061 fs_initcall(inet_init);
2063 /* ------------------------------------------------------------------------ */
2065 #ifdef CONFIG_PROC_FS
2066 static int __init ipv4_proc_init(void)
2070 if (raw_proc_init())
2072 if (tcp4_proc_init())
2074 if (udp4_proc_init())
2076 if (ping_proc_init())
2078 if (ip_misc_proc_init())
2095 #else /* CONFIG_PROC_FS */
2096 static int __init ipv4_proc_init(void)
2100 #endif /* CONFIG_PROC_FS */