1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PF_INET protocol family socket handler.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Florian La Roche, <flla@stud.uni-sb.de>
12 * Alan Cox, <A.Cox@swansea.ac.uk>
14 * Changes (see also sock.c)
17 * Karl Knutson : Socket protocol table
18 * A.N.Kuznetsov : Socket death error in accept().
19 * John Richardson : Fix non blocking error in connect()
20 * so sockets that fail to connect
21 * don't return -EINPROGRESS.
22 * Alan Cox : Asynchronous I/O support
23 * Alan Cox : Keep correct socket pointer on sock
26 * Alan Cox : Semantics of SO_LINGER aren't state
27 * moved to close when you look carefully.
28 * With this fixed and the accept bug fixed
29 * some RPC stuff seems happier.
30 * Niibe Yutaka : 4.4BSD style write async I/O
32 * Tony Gale : Fixed reuse semantics.
33 * Alan Cox : bind() shouldn't abort existing but dead
34 * sockets. Stops FTP netin:.. I hope.
35 * Alan Cox : bind() works correctly for RAW sockets.
36 * Note that FreeBSD at least was broken
37 * in this respect so be careful with
38 * compatibility tests...
39 * Alan Cox : routing cache support
40 * Alan Cox : memzero the socket structure for
42 * Matt Day : nonblock connect error handler
43 * Alan Cox : Allow large numbers of pending sockets
44 * (eg for big web sites), but only if
45 * specifically application requested.
46 * Alan Cox : New buffering throughout IP. Used
48 * Alan Cox : New buffering now used smartly.
49 * Alan Cox : BSD rather than common sense
50 * interpretation of listen.
51 * Germano Caronni : Assorted small races.
52 * Alan Cox : sendmsg/recvmsg basic support.
53 * Alan Cox : Only sendmsg/recvmsg now supported.
54 * Alan Cox : Locked down bind (see security list).
55 * Alan Cox : Loosened bind a little.
56 * Mike McLagan : ADD/DEL DLCI Ioctls
57 * Willy Konynenberg : Transparent proxying support.
58 * David S. Miller : New socket lookup architecture.
59 * Some other random speedups.
60 * Cyrus Durgin : Cleaned up file for kmod hacks.
61 * Andi Kleen : Fix inet_stream_connect TCP race.
64 #define pr_fmt(fmt) "IPv4: " fmt
66 #include <linux/err.h>
67 #include <linux/errno.h>
68 #include <linux/types.h>
69 #include <linux/socket.h>
71 #include <linux/kernel.h>
72 #include <linux/kmod.h>
73 #include <linux/sched.h>
74 #include <linux/timer.h>
75 #include <linux/string.h>
76 #include <linux/sockios.h>
77 #include <linux/net.h>
78 #include <linux/capability.h>
79 #include <linux/fcntl.h>
81 #include <linux/interrupt.h>
82 #include <linux/stat.h>
83 #include <linux/init.h>
84 #include <linux/poll.h>
85 #include <linux/netfilter_ipv4.h>
86 #include <linux/random.h>
87 #include <linux/slab.h>
89 #include <linux/uaccess.h>
91 #include <linux/inet.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/netdevice.h>
95 #include <net/checksum.h>
97 #include <net/protocol.h>
99 #include <net/route.h>
100 #include <net/ip_fib.h>
101 #include <net/inet_connection_sock.h>
104 #include <net/udplite.h>
105 #include <net/ping.h>
106 #include <linux/skbuff.h>
107 #include <net/sock.h>
109 #include <net/icmp.h>
110 #include <net/inet_common.h>
111 #include <net/ip_tunnels.h>
112 #include <net/xfrm.h>
113 #include <net/net_namespace.h>
114 #include <net/secure_seq.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
118 #include <net/l3mdev.h>
119 #include <net/compat.h>
121 #include <trace/events/sock.h>
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
126 static struct list_head inetsw
[SOCK_MAX
];
127 static DEFINE_SPINLOCK(inetsw_lock
);
129 /* New destruction routine */
131 void inet_sock_destruct(struct sock
*sk
)
133 struct inet_sock
*inet
= inet_sk(sk
);
135 __skb_queue_purge(&sk
->sk_receive_queue
);
136 if (sk
->sk_rx_skb_cache
) {
137 __kfree_skb(sk
->sk_rx_skb_cache
);
138 sk
->sk_rx_skb_cache
= NULL
;
140 __skb_queue_purge(&sk
->sk_error_queue
);
144 if (sk
->sk_type
== SOCK_STREAM
&& sk
->sk_state
!= TCP_CLOSE
) {
145 pr_err("Attempt to release TCP socket in state %d %p\n",
149 if (!sock_flag(sk
, SOCK_DEAD
)) {
150 pr_err("Attempt to release alive inet socket %p\n", sk
);
154 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
155 WARN_ON(refcount_read(&sk
->sk_wmem_alloc
));
156 WARN_ON(sk
->sk_wmem_queued
);
157 WARN_ON(sk
->sk_forward_alloc
);
159 kfree(rcu_dereference_protected(inet
->inet_opt
, 1));
160 dst_release(rcu_dereference_protected(sk
->sk_dst_cache
, 1));
161 dst_release(sk
->sk_rx_dst
);
162 sk_refcnt_debug_dec(sk
);
164 EXPORT_SYMBOL(inet_sock_destruct
);
167 * The routines beyond this point handle the behaviour of an AF_INET
168 * socket object. Mostly it punts to the subprotocols of IP to do
173 * Automatically bind an unbound socket.
176 static int inet_autobind(struct sock
*sk
)
178 struct inet_sock
*inet
;
179 /* We may need to bind the socket. */
182 if (!inet
->inet_num
) {
183 if (sk
->sk_prot
->get_port(sk
, 0)) {
187 inet
->inet_sport
= htons(inet
->inet_num
);
194 * Move a socket into listening state.
196 int inet_listen(struct socket
*sock
, int backlog
)
198 struct sock
*sk
= sock
->sk
;
199 unsigned char old_state
;
200 int err
, tcp_fastopen
;
205 if (sock
->state
!= SS_UNCONNECTED
|| sock
->type
!= SOCK_STREAM
)
208 old_state
= sk
->sk_state
;
209 if (!((1 << old_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)))
212 WRITE_ONCE(sk
->sk_max_ack_backlog
, backlog
);
213 /* Really, if the socket is already in listen state
214 * we can only allow the backlog to be adjusted.
216 if (old_state
!= TCP_LISTEN
) {
217 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
218 * Note that only TCP sockets (SOCK_STREAM) will reach here.
219 * Also fastopen backlog may already been set via the option
220 * because the socket was in TCP_LISTEN state previously but
221 * was shutdown() rather than close().
223 tcp_fastopen
= sock_net(sk
)->ipv4
.sysctl_tcp_fastopen
;
224 if ((tcp_fastopen
& TFO_SERVER_WO_SOCKOPT1
) &&
225 (tcp_fastopen
& TFO_SERVER_ENABLE
) &&
226 !inet_csk(sk
)->icsk_accept_queue
.fastopenq
.max_qlen
) {
227 fastopen_queue_tune(sk
, backlog
);
228 tcp_fastopen_init_key_once(sock_net(sk
));
231 err
= inet_csk_listen_start(sk
, backlog
);
234 tcp_call_bpf(sk
, BPF_SOCK_OPS_TCP_LISTEN_CB
, 0, NULL
);
242 EXPORT_SYMBOL(inet_listen
);
245 * Create an inet socket.
248 static int inet_create(struct net
*net
, struct socket
*sock
, int protocol
,
252 struct inet_protosw
*answer
;
253 struct inet_sock
*inet
;
254 struct proto
*answer_prot
;
255 unsigned char answer_flags
;
256 int try_loading_module
= 0;
259 if (protocol
< 0 || protocol
>= IPPROTO_MAX
)
262 sock
->state
= SS_UNCONNECTED
;
264 /* Look for the requested type/protocol pair. */
266 err
= -ESOCKTNOSUPPORT
;
268 list_for_each_entry_rcu(answer
, &inetsw
[sock
->type
], list
) {
271 /* Check the non-wild match. */
272 if (protocol
== answer
->protocol
) {
273 if (protocol
!= IPPROTO_IP
)
276 /* Check for the two wild cases. */
277 if (IPPROTO_IP
== protocol
) {
278 protocol
= answer
->protocol
;
281 if (IPPROTO_IP
== answer
->protocol
)
284 err
= -EPROTONOSUPPORT
;
288 if (try_loading_module
< 2) {
291 * Be more specific, e.g. net-pf-2-proto-132-type-1
292 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
294 if (++try_loading_module
== 1)
295 request_module("net-pf-%d-proto-%d-type-%d",
296 PF_INET
, protocol
, sock
->type
);
298 * Fall back to generic, e.g. net-pf-2-proto-132
299 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
302 request_module("net-pf-%d-proto-%d",
304 goto lookup_protocol
;
310 if (sock
->type
== SOCK_RAW
&& !kern
&&
311 !ns_capable(net
->user_ns
, CAP_NET_RAW
))
314 sock
->ops
= answer
->ops
;
315 answer_prot
= answer
->prot
;
316 answer_flags
= answer
->flags
;
319 WARN_ON(!answer_prot
->slab
);
322 sk
= sk_alloc(net
, PF_INET
, GFP_KERNEL
, answer_prot
, kern
);
327 if (INET_PROTOSW_REUSE
& answer_flags
)
328 sk
->sk_reuse
= SK_CAN_REUSE
;
331 inet
->is_icsk
= (INET_PROTOSW_ICSK
& answer_flags
) != 0;
335 if (SOCK_RAW
== sock
->type
) {
336 inet
->inet_num
= protocol
;
337 if (IPPROTO_RAW
== protocol
)
341 if (net
->ipv4
.sysctl_ip_no_pmtu_disc
)
342 inet
->pmtudisc
= IP_PMTUDISC_DONT
;
344 inet
->pmtudisc
= IP_PMTUDISC_WANT
;
348 sock_init_data(sock
, sk
);
350 sk
->sk_destruct
= inet_sock_destruct
;
351 sk
->sk_protocol
= protocol
;
352 sk
->sk_backlog_rcv
= sk
->sk_prot
->backlog_rcv
;
359 inet
->mc_list
= NULL
;
362 sk_refcnt_debug_inc(sk
);
364 if (inet
->inet_num
) {
365 /* It assumes that any protocol which allows
366 * the user to assign a number at socket
367 * creation time automatically
370 inet
->inet_sport
= htons(inet
->inet_num
);
371 /* Add to protocol hash chains. */
372 err
= sk
->sk_prot
->hash(sk
);
374 sk_common_release(sk
);
379 if (sk
->sk_prot
->init
) {
380 err
= sk
->sk_prot
->init(sk
);
382 sk_common_release(sk
);
388 err
= BPF_CGROUP_RUN_PROG_INET_SOCK(sk
);
390 sk_common_release(sk
);
403 * The peer socket should always be NULL (or else). When we call this
404 * function we are destroying the object and from then on nobody
405 * should refer to it.
407 int inet_release(struct socket
*sock
)
409 struct sock
*sk
= sock
->sk
;
414 if (!sk
->sk_kern_sock
)
415 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk
);
417 /* Applications forget to leave groups before exiting */
418 ip_mc_drop_socket(sk
);
420 /* If linger is set, we don't return until the close
421 * is complete. Otherwise we return immediately. The
422 * actually closing is done the same either way.
424 * If the close is due to the process exiting, we never
428 if (sock_flag(sk
, SOCK_LINGER
) &&
429 !(current
->flags
& PF_EXITING
))
430 timeout
= sk
->sk_lingertime
;
431 sk
->sk_prot
->close(sk
, timeout
);
436 EXPORT_SYMBOL(inet_release
);
438 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
440 struct sock
*sk
= sock
->sk
;
443 /* If the socket has its own bind function then use it. (RAW) */
444 if (sk
->sk_prot
->bind
) {
445 return sk
->sk_prot
->bind(sk
, uaddr
, addr_len
);
447 if (addr_len
< sizeof(struct sockaddr_in
))
450 /* BPF prog is run before any checks are done so that if the prog
451 * changes context in a wrong way it will be caught.
453 err
= BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK(sk
, uaddr
);
457 return __inet_bind(sk
, uaddr
, addr_len
, BIND_WITH_LOCK
);
459 EXPORT_SYMBOL(inet_bind
);
461 int __inet_bind(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
,
464 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
465 struct inet_sock
*inet
= inet_sk(sk
);
466 struct net
*net
= sock_net(sk
);
469 u32 tb_id
= RT_TABLE_LOCAL
;
472 if (addr
->sin_family
!= AF_INET
) {
473 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
474 * only if s_addr is INADDR_ANY.
477 if (addr
->sin_family
!= AF_UNSPEC
||
478 addr
->sin_addr
.s_addr
!= htonl(INADDR_ANY
))
482 tb_id
= l3mdev_fib_table_by_index(net
, sk
->sk_bound_dev_if
) ? : tb_id
;
483 chk_addr_ret
= inet_addr_type_table(net
, addr
->sin_addr
.s_addr
, tb_id
);
485 /* Not specified by any standard per-se, however it breaks too
486 * many applications when removed. It is unfortunate since
487 * allowing applications to make a non-local bind solves
488 * several problems with systems using dynamic addressing.
489 * (ie. your servers still start up even if your ISDN link
490 * is temporarily down)
492 err
= -EADDRNOTAVAIL
;
493 if (!inet_can_nonlocal_bind(net
, inet
) &&
494 addr
->sin_addr
.s_addr
!= htonl(INADDR_ANY
) &&
495 chk_addr_ret
!= RTN_LOCAL
&&
496 chk_addr_ret
!= RTN_MULTICAST
&&
497 chk_addr_ret
!= RTN_BROADCAST
)
500 snum
= ntohs(addr
->sin_port
);
502 if (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 if (sk
->sk_prot
->get_port(sk
, snum
)) {
529 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 goto out_release_sock
;
542 if (inet
->inet_rcv_saddr
)
543 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
545 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
546 inet
->inet_sport
= htons(inet
->inet_num
);
547 inet
->inet_daddr
= 0;
548 inet
->inet_dport
= 0;
552 if (flags
& BIND_WITH_LOCK
)
558 int inet_dgram_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
559 int addr_len
, int flags
)
561 struct sock
*sk
= sock
->sk
;
564 if (addr_len
< sizeof(uaddr
->sa_family
))
566 if (uaddr
->sa_family
== AF_UNSPEC
)
567 return sk
->sk_prot
->disconnect(sk
, flags
);
569 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk
)) {
570 err
= sk
->sk_prot
->pre_connect(sk
, uaddr
, addr_len
);
575 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
577 return sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
579 EXPORT_SYMBOL(inet_dgram_connect
);
581 static long inet_wait_for_connect(struct sock
*sk
, long timeo
, int writebias
)
583 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
585 add_wait_queue(sk_sleep(sk
), &wait
);
586 sk
->sk_write_pending
+= writebias
;
588 /* Basic assumption: if someone sets sk->sk_err, he _must_
589 * change state of the socket from TCP_SYN_*.
590 * Connect() does not allow to get error notifications
591 * without closing the socket.
593 while ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
595 timeo
= wait_woken(&wait
, TASK_INTERRUPTIBLE
, timeo
);
597 if (signal_pending(current
) || !timeo
)
600 remove_wait_queue(sk_sleep(sk
), &wait
);
601 sk
->sk_write_pending
-= writebias
;
606 * Connect to a remote host. There is regrettably still a little
607 * TCP 'magic' in here.
609 int __inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
610 int addr_len
, int flags
, int is_sendmsg
)
612 struct sock
*sk
= sock
->sk
;
617 * uaddr can be NULL and addr_len can be 0 if:
618 * sk is a TCP fastopen active socket and
619 * TCP_FASTOPEN_CONNECT sockopt is set and
620 * we already have a valid cookie for this socket.
621 * In this case, user can call write() after connect().
622 * write() will invoke tcp_sendmsg_fastopen() which calls
623 * __inet_stream_connect().
626 if (addr_len
< sizeof(uaddr
->sa_family
))
629 if (uaddr
->sa_family
== AF_UNSPEC
) {
630 err
= sk
->sk_prot
->disconnect(sk
, flags
);
631 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
636 switch (sock
->state
) {
644 if (inet_sk(sk
)->defer_connect
)
645 err
= is_sendmsg
? -EINPROGRESS
: -EISCONN
;
648 /* Fall out of switch with err, set for this state */
652 if (sk
->sk_state
!= TCP_CLOSE
)
655 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk
)) {
656 err
= sk
->sk_prot
->pre_connect(sk
, uaddr
, addr_len
);
661 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
665 sock
->state
= SS_CONNECTING
;
667 if (!err
&& inet_sk(sk
)->defer_connect
)
670 /* Just entered SS_CONNECTING state; the only
671 * difference is that return value in non-blocking
672 * case is EINPROGRESS, rather than EALREADY.
678 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
680 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
681 int writebias
= (sk
->sk_protocol
== IPPROTO_TCP
) &&
682 tcp_sk(sk
)->fastopen_req
&&
683 tcp_sk(sk
)->fastopen_req
->data
? 1 : 0;
685 /* Error code is set above */
686 if (!timeo
|| !inet_wait_for_connect(sk
, timeo
, writebias
))
689 err
= sock_intr_errno(timeo
);
690 if (signal_pending(current
))
694 /* Connection was closed by RST, timeout, ICMP error
695 * or another process disconnected us.
697 if (sk
->sk_state
== TCP_CLOSE
)
700 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
701 * and error was received after socket entered established state.
702 * Hence, it is handled normally after connect() return successfully.
705 sock
->state
= SS_CONNECTED
;
711 err
= sock_error(sk
) ? : -ECONNABORTED
;
712 sock
->state
= SS_UNCONNECTED
;
713 if (sk
->sk_prot
->disconnect(sk
, flags
))
714 sock
->state
= SS_DISCONNECTING
;
717 EXPORT_SYMBOL(__inet_stream_connect
);
719 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
720 int addr_len
, int flags
)
725 err
= __inet_stream_connect(sock
, uaddr
, addr_len
, flags
, 0);
726 release_sock(sock
->sk
);
729 EXPORT_SYMBOL(inet_stream_connect
);
732 * Accept a pending connection. The TCP layer now gives BSD semantics.
735 int inet_accept(struct socket
*sock
, struct socket
*newsock
, int flags
,
738 struct sock
*sk1
= sock
->sk
;
740 struct sock
*sk2
= sk1
->sk_prot
->accept(sk1
, flags
, &err
, kern
);
747 sock_rps_record_flow(sk2
);
748 WARN_ON(!((1 << sk2
->sk_state
) &
749 (TCPF_ESTABLISHED
| TCPF_SYN_RECV
|
750 TCPF_CLOSE_WAIT
| TCPF_CLOSE
)));
752 sock_graft(sk2
, newsock
);
754 newsock
->state
= SS_CONNECTED
;
760 EXPORT_SYMBOL(inet_accept
);
763 * This does both peername and sockname.
765 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
768 struct sock
*sk
= sock
->sk
;
769 struct inet_sock
*inet
= inet_sk(sk
);
770 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, uaddr
);
772 sin
->sin_family
= AF_INET
;
774 if (!inet
->inet_dport
||
775 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
778 sin
->sin_port
= inet
->inet_dport
;
779 sin
->sin_addr
.s_addr
= inet
->inet_daddr
;
781 __be32 addr
= inet
->inet_rcv_saddr
;
783 addr
= inet
->inet_saddr
;
784 sin
->sin_port
= inet
->inet_sport
;
785 sin
->sin_addr
.s_addr
= addr
;
787 if (cgroup_bpf_enabled
)
788 BPF_CGROUP_RUN_SA_PROG_LOCK(sk
, (struct sockaddr
*)sin
,
789 peer
? BPF_CGROUP_INET4_GETPEERNAME
:
790 BPF_CGROUP_INET4_GETSOCKNAME
,
792 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
795 EXPORT_SYMBOL(inet_getname
);
797 int inet_send_prepare(struct sock
*sk
)
799 sock_rps_record_flow(sk
);
801 /* We may need to bind the socket. */
802 if (!inet_sk(sk
)->inet_num
&& !sk
->sk_prot
->no_autobind
&&
808 EXPORT_SYMBOL_GPL(inet_send_prepare
);
810 int inet_sendmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
)
812 struct sock
*sk
= sock
->sk
;
814 if (unlikely(inet_send_prepare(sk
)))
817 return INDIRECT_CALL_2(sk
->sk_prot
->sendmsg
, tcp_sendmsg
, udp_sendmsg
,
820 EXPORT_SYMBOL(inet_sendmsg
);
822 ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
823 size_t size
, int flags
)
825 struct sock
*sk
= sock
->sk
;
827 if (unlikely(inet_send_prepare(sk
)))
830 if (sk
->sk_prot
->sendpage
)
831 return sk
->sk_prot
->sendpage(sk
, page
, offset
, size
, flags
);
832 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
834 EXPORT_SYMBOL(inet_sendpage
);
836 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock
*, struct msghdr
*,
837 size_t, int, int, int *));
838 int inet_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t size
,
841 struct sock
*sk
= sock
->sk
;
845 if (likely(!(flags
& MSG_ERRQUEUE
)))
846 sock_rps_record_flow(sk
);
848 err
= INDIRECT_CALL_2(sk
->sk_prot
->recvmsg
, tcp_recvmsg
, udp_recvmsg
,
849 sk
, msg
, size
, flags
& MSG_DONTWAIT
,
850 flags
& ~MSG_DONTWAIT
, &addr_len
);
852 msg
->msg_namelen
= addr_len
;
855 EXPORT_SYMBOL(inet_recvmsg
);
857 int inet_shutdown(struct socket
*sock
, int how
)
859 struct sock
*sk
= sock
->sk
;
862 /* This should really check to make sure
863 * the socket is a TCP socket. (WHY AC...)
865 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
868 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
872 if (sock
->state
== SS_CONNECTING
) {
873 if ((1 << sk
->sk_state
) &
874 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
875 sock
->state
= SS_DISCONNECTING
;
877 sock
->state
= SS_CONNECTED
;
880 switch (sk
->sk_state
) {
883 /* Hack to wake up other listeners, who can poll for
884 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
887 sk
->sk_shutdown
|= how
;
888 if (sk
->sk_prot
->shutdown
)
889 sk
->sk_prot
->shutdown(sk
, how
);
892 /* Remaining two branches are temporary solution for missing
893 * close() in multithreaded environment. It is _not_ a good idea,
894 * but we have no choice until close() is repaired at VFS level.
897 if (!(how
& RCV_SHUTDOWN
))
901 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
902 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
906 /* Wake up anyone sleeping in poll. */
907 sk
->sk_state_change(sk
);
911 EXPORT_SYMBOL(inet_shutdown
);
914 * ioctl() calls you can issue on an INET socket. Most of these are
915 * device configuration and stuff and very rarely used. Some ioctls
916 * pass on to the socket itself.
918 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
919 * loads the devconfigure module does its configuring and unloads it.
920 * There's a good 20K of config code hanging around the kernel.
923 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
925 struct sock
*sk
= sock
->sk
;
927 struct net
*net
= sock_net(sk
);
928 void __user
*p
= (void __user
*)arg
;
935 if (copy_from_user(&rt
, p
, sizeof(struct rtentry
)))
937 err
= ip_rt_ioctl(net
, cmd
, &rt
);
945 err
= arp_ioctl(net
, cmd
, (void __user
*)arg
);
952 if (copy_from_user(&ifr
, p
, sizeof(struct ifreq
)))
954 err
= devinet_ioctl(net
, cmd
, &ifr
);
955 if (!err
&& copy_to_user(p
, &ifr
, sizeof(struct ifreq
)))
965 if (copy_from_user(&ifr
, p
, sizeof(struct ifreq
)))
967 err
= devinet_ioctl(net
, cmd
, &ifr
);
970 if (sk
->sk_prot
->ioctl
)
971 err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
);
978 EXPORT_SYMBOL(inet_ioctl
);
981 static int inet_compat_routing_ioctl(struct sock
*sk
, unsigned int cmd
,
982 struct compat_rtentry __user
*ur
)
987 if (copy_from_user(&rt
.rt_dst
, &ur
->rt_dst
,
988 3 * sizeof(struct sockaddr
)) ||
989 get_user(rt
.rt_flags
, &ur
->rt_flags
) ||
990 get_user(rt
.rt_metric
, &ur
->rt_metric
) ||
991 get_user(rt
.rt_mtu
, &ur
->rt_mtu
) ||
992 get_user(rt
.rt_window
, &ur
->rt_window
) ||
993 get_user(rt
.rt_irtt
, &ur
->rt_irtt
) ||
994 get_user(rtdev
, &ur
->rt_dev
))
997 rt
.rt_dev
= compat_ptr(rtdev
);
998 return ip_rt_ioctl(sock_net(sk
), cmd
, &rt
);
1001 static int inet_compat_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
1003 void __user
*argp
= compat_ptr(arg
);
1004 struct sock
*sk
= sock
->sk
;
1009 return inet_compat_routing_ioctl(sk
, cmd
, argp
);
1011 if (!sk
->sk_prot
->compat_ioctl
)
1012 return -ENOIOCTLCMD
;
1013 return sk
->sk_prot
->compat_ioctl(sk
, cmd
, arg
);
1016 #endif /* CONFIG_COMPAT */
1018 const struct proto_ops inet_stream_ops
= {
1020 .flags
= PROTO_CMSG_DATA_ONLY
,
1021 .owner
= THIS_MODULE
,
1022 .release
= inet_release
,
1024 .connect
= inet_stream_connect
,
1025 .socketpair
= sock_no_socketpair
,
1026 .accept
= inet_accept
,
1027 .getname
= inet_getname
,
1029 .ioctl
= inet_ioctl
,
1030 .gettstamp
= sock_gettstamp
,
1031 .listen
= inet_listen
,
1032 .shutdown
= inet_shutdown
,
1033 .setsockopt
= sock_common_setsockopt
,
1034 .getsockopt
= sock_common_getsockopt
,
1035 .sendmsg
= inet_sendmsg
,
1036 .recvmsg
= inet_recvmsg
,
1040 .sendpage
= inet_sendpage
,
1041 .splice_read
= tcp_splice_read
,
1042 .read_sock
= tcp_read_sock
,
1043 .sendmsg_locked
= tcp_sendmsg_locked
,
1044 .sendpage_locked
= tcp_sendpage_locked
,
1045 .peek_len
= tcp_peek_len
,
1046 #ifdef CONFIG_COMPAT
1047 .compat_ioctl
= inet_compat_ioctl
,
1049 .set_rcvlowat
= tcp_set_rcvlowat
,
1051 EXPORT_SYMBOL(inet_stream_ops
);
1053 const struct proto_ops inet_dgram_ops
= {
1055 .owner
= THIS_MODULE
,
1056 .release
= inet_release
,
1058 .connect
= inet_dgram_connect
,
1059 .socketpair
= sock_no_socketpair
,
1060 .accept
= sock_no_accept
,
1061 .getname
= inet_getname
,
1063 .ioctl
= inet_ioctl
,
1064 .gettstamp
= sock_gettstamp
,
1065 .listen
= sock_no_listen
,
1066 .shutdown
= inet_shutdown
,
1067 .setsockopt
= sock_common_setsockopt
,
1068 .getsockopt
= sock_common_getsockopt
,
1069 .sendmsg
= inet_sendmsg
,
1070 .recvmsg
= inet_recvmsg
,
1071 .mmap
= sock_no_mmap
,
1072 .sendpage
= inet_sendpage
,
1073 .set_peek_off
= sk_set_peek_off
,
1074 #ifdef CONFIG_COMPAT
1075 .compat_ioctl
= inet_compat_ioctl
,
1078 EXPORT_SYMBOL(inet_dgram_ops
);
1081 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1084 static const struct proto_ops inet_sockraw_ops
= {
1086 .owner
= THIS_MODULE
,
1087 .release
= inet_release
,
1089 .connect
= inet_dgram_connect
,
1090 .socketpair
= sock_no_socketpair
,
1091 .accept
= sock_no_accept
,
1092 .getname
= inet_getname
,
1093 .poll
= datagram_poll
,
1094 .ioctl
= inet_ioctl
,
1095 .gettstamp
= sock_gettstamp
,
1096 .listen
= sock_no_listen
,
1097 .shutdown
= inet_shutdown
,
1098 .setsockopt
= sock_common_setsockopt
,
1099 .getsockopt
= sock_common_getsockopt
,
1100 .sendmsg
= inet_sendmsg
,
1101 .recvmsg
= inet_recvmsg
,
1102 .mmap
= sock_no_mmap
,
1103 .sendpage
= inet_sendpage
,
1104 #ifdef CONFIG_COMPAT
1105 .compat_ioctl
= inet_compat_ioctl
,
1109 static const struct net_proto_family inet_family_ops
= {
1111 .create
= inet_create
,
1112 .owner
= THIS_MODULE
,
1115 /* Upon startup we insert all the elements in inetsw_array[] into
1116 * the linked list inetsw.
1118 static struct inet_protosw inetsw_array
[] =
1121 .type
= SOCK_STREAM
,
1122 .protocol
= IPPROTO_TCP
,
1124 .ops
= &inet_stream_ops
,
1125 .flags
= INET_PROTOSW_PERMANENT
|
1131 .protocol
= IPPROTO_UDP
,
1133 .ops
= &inet_dgram_ops
,
1134 .flags
= INET_PROTOSW_PERMANENT
,
1139 .protocol
= IPPROTO_ICMP
,
1141 .ops
= &inet_sockraw_ops
,
1142 .flags
= INET_PROTOSW_REUSE
,
1147 .protocol
= IPPROTO_IP
, /* wild card */
1149 .ops
= &inet_sockraw_ops
,
1150 .flags
= INET_PROTOSW_REUSE
,
1154 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1156 void inet_register_protosw(struct inet_protosw
*p
)
1158 struct list_head
*lh
;
1159 struct inet_protosw
*answer
;
1160 int protocol
= p
->protocol
;
1161 struct list_head
*last_perm
;
1163 spin_lock_bh(&inetsw_lock
);
1165 if (p
->type
>= SOCK_MAX
)
1168 /* If we are trying to override a permanent protocol, bail. */
1169 last_perm
= &inetsw
[p
->type
];
1170 list_for_each(lh
, &inetsw
[p
->type
]) {
1171 answer
= list_entry(lh
, struct inet_protosw
, list
);
1172 /* Check only the non-wild match. */
1173 if ((INET_PROTOSW_PERMANENT
& answer
->flags
) == 0)
1175 if (protocol
== answer
->protocol
)
1180 /* Add the new entry after the last permanent entry if any, so that
1181 * the new entry does not override a permanent entry when matched with
1182 * a wild-card protocol. But it is allowed to override any existing
1183 * non-permanent entry. This means that when we remove this entry, the
1184 * system automatically returns to the old behavior.
1186 list_add_rcu(&p
->list
, last_perm
);
1188 spin_unlock_bh(&inetsw_lock
);
1193 pr_err("Attempt to override permanent protocol %d\n", protocol
);
1197 pr_err("Ignoring attempt to register invalid socket type %d\n",
1201 EXPORT_SYMBOL(inet_register_protosw
);
1203 void inet_unregister_protosw(struct inet_protosw
*p
)
1205 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
1206 pr_err("Attempt to unregister permanent protocol %d\n",
1209 spin_lock_bh(&inetsw_lock
);
1210 list_del_rcu(&p
->list
);
1211 spin_unlock_bh(&inetsw_lock
);
1216 EXPORT_SYMBOL(inet_unregister_protosw
);
1218 static int inet_sk_reselect_saddr(struct sock
*sk
)
1220 struct inet_sock
*inet
= inet_sk(sk
);
1221 __be32 old_saddr
= inet
->inet_saddr
;
1222 __be32 daddr
= inet
->inet_daddr
;
1226 struct ip_options_rcu
*inet_opt
;
1228 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
1229 lockdep_sock_is_held(sk
));
1230 if (inet_opt
&& inet_opt
->opt
.srr
)
1231 daddr
= inet_opt
->opt
.faddr
;
1233 /* Query new route. */
1234 fl4
= &inet
->cork
.fl
.u
.ip4
;
1235 rt
= ip_route_connect(fl4
, daddr
, 0, RT_CONN_FLAGS(sk
),
1236 sk
->sk_bound_dev_if
, sk
->sk_protocol
,
1237 inet
->inet_sport
, inet
->inet_dport
, sk
);
1241 sk_setup_caps(sk
, &rt
->dst
);
1243 new_saddr
= fl4
->saddr
;
1245 if (new_saddr
== old_saddr
)
1248 if (sock_net(sk
)->ipv4
.sysctl_ip_dynaddr
> 1) {
1249 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1250 __func__
, &old_saddr
, &new_saddr
);
1253 inet
->inet_saddr
= inet
->inet_rcv_saddr
= new_saddr
;
1256 * XXX The only one ugly spot where we need to
1257 * XXX really change the sockets identity after
1258 * XXX it has entered the hashes. -DaveM
1260 * Besides that, it does not check for connection
1261 * uniqueness. Wait for troubles.
1263 return __sk_prot_rehash(sk
);
1266 int inet_sk_rebuild_header(struct sock
*sk
)
1268 struct inet_sock
*inet
= inet_sk(sk
);
1269 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1271 struct ip_options_rcu
*inet_opt
;
1275 /* Route is OK, nothing to do. */
1281 inet_opt
= rcu_dereference(inet
->inet_opt
);
1282 daddr
= inet
->inet_daddr
;
1283 if (inet_opt
&& inet_opt
->opt
.srr
)
1284 daddr
= inet_opt
->opt
.faddr
;
1286 fl4
= &inet
->cork
.fl
.u
.ip4
;
1287 rt
= ip_route_output_ports(sock_net(sk
), fl4
, sk
, daddr
, inet
->inet_saddr
,
1288 inet
->inet_dport
, inet
->inet_sport
,
1289 sk
->sk_protocol
, RT_CONN_FLAGS(sk
),
1290 sk
->sk_bound_dev_if
);
1293 sk_setup_caps(sk
, &rt
->dst
);
1297 /* Routing failed... */
1298 sk
->sk_route_caps
= 0;
1300 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1301 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1303 if (!sock_net(sk
)->ipv4
.sysctl_ip_dynaddr
||
1304 sk
->sk_state
!= TCP_SYN_SENT
||
1305 (sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
) ||
1306 (err
= inet_sk_reselect_saddr(sk
)) != 0)
1307 sk
->sk_err_soft
= -err
;
1312 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1314 void inet_sk_set_state(struct sock
*sk
, int state
)
1316 trace_inet_sock_set_state(sk
, sk
->sk_state
, state
);
1317 sk
->sk_state
= state
;
1319 EXPORT_SYMBOL(inet_sk_set_state
);
1321 void inet_sk_state_store(struct sock
*sk
, int newstate
)
1323 trace_inet_sock_set_state(sk
, sk
->sk_state
, newstate
);
1324 smp_store_release(&sk
->sk_state
, newstate
);
1327 struct sk_buff
*inet_gso_segment(struct sk_buff
*skb
,
1328 netdev_features_t features
)
1330 bool udpfrag
= false, fixedid
= false, gso_partial
, encap
;
1331 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1332 const struct net_offload
*ops
;
1333 unsigned int offset
= 0;
1340 skb_reset_network_header(skb
);
1341 nhoff
= skb_network_header(skb
) - skb_mac_header(skb
);
1342 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1347 if (ihl
< sizeof(*iph
))
1350 id
= ntohs(iph
->id
);
1351 proto
= iph
->protocol
;
1353 /* Warning: after this point, iph might be no longer valid */
1354 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1356 __skb_pull(skb
, ihl
);
1358 encap
= SKB_GSO_CB(skb
)->encap_level
> 0;
1360 features
&= skb
->dev
->hw_enc_features
;
1361 SKB_GSO_CB(skb
)->encap_level
+= ihl
;
1363 skb_reset_transport_header(skb
);
1365 segs
= ERR_PTR(-EPROTONOSUPPORT
);
1367 if (!skb
->encapsulation
|| encap
) {
1368 udpfrag
= !!(skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
);
1369 fixedid
= !!(skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_FIXEDID
);
1371 /* fixed ID is invalid if DF bit is not set */
1372 if (fixedid
&& !(ip_hdr(skb
)->frag_off
& htons(IP_DF
)))
1376 ops
= rcu_dereference(inet_offloads
[proto
]);
1377 if (likely(ops
&& ops
->callbacks
.gso_segment
))
1378 segs
= ops
->callbacks
.gso_segment(skb
, features
);
1380 if (IS_ERR_OR_NULL(segs
))
1383 gso_partial
= !!(skb_shinfo(segs
)->gso_type
& SKB_GSO_PARTIAL
);
1387 iph
= (struct iphdr
*)(skb_mac_header(skb
) + nhoff
);
1389 iph
->frag_off
= htons(offset
>> 3);
1391 iph
->frag_off
|= htons(IP_MF
);
1392 offset
+= skb
->len
- nhoff
- ihl
;
1393 tot_len
= skb
->len
- nhoff
;
1394 } else if (skb_is_gso(skb
)) {
1396 iph
->id
= htons(id
);
1397 id
+= skb_shinfo(skb
)->gso_segs
;
1401 tot_len
= skb_shinfo(skb
)->gso_size
+
1402 SKB_GSO_CB(skb
)->data_offset
+
1403 skb
->head
- (unsigned char *)iph
;
1405 tot_len
= skb
->len
- nhoff
;
1408 iph
->id
= htons(id
++);
1409 tot_len
= skb
->len
- nhoff
;
1411 iph
->tot_len
= htons(tot_len
);
1414 skb_reset_inner_headers(skb
);
1415 skb
->network_header
= (u8
*)iph
- skb
->head
;
1416 skb_reset_mac_len(skb
);
1417 } while ((skb
= skb
->next
));
1422 EXPORT_SYMBOL(inet_gso_segment
);
1424 static struct sk_buff
*ipip_gso_segment(struct sk_buff
*skb
,
1425 netdev_features_t features
)
1427 if (!(skb_shinfo(skb
)->gso_type
& SKB_GSO_IPXIP4
))
1428 return ERR_PTR(-EINVAL
);
1430 return inet_gso_segment(skb
, features
);
1433 struct sk_buff
*inet_gro_receive(struct list_head
*head
, struct sk_buff
*skb
)
1435 const struct net_offload
*ops
;
1436 struct sk_buff
*pp
= NULL
;
1437 const struct iphdr
*iph
;
1445 off
= skb_gro_offset(skb
);
1446 hlen
= off
+ sizeof(*iph
);
1447 iph
= skb_gro_header_fast(skb
, off
);
1448 if (skb_gro_header_hard(skb
, hlen
)) {
1449 iph
= skb_gro_header_slow(skb
, hlen
, off
);
1454 proto
= iph
->protocol
;
1457 ops
= rcu_dereference(inet_offloads
[proto
]);
1458 if (!ops
|| !ops
->callbacks
.gro_receive
)
1461 if (*(u8
*)iph
!= 0x45)
1464 if (ip_is_fragment(iph
))
1467 if (unlikely(ip_fast_csum((u8
*)iph
, 5)))
1470 id
= ntohl(*(__be32
*)&iph
->id
);
1471 flush
= (u16
)((ntohl(*(__be32
*)iph
) ^ skb_gro_len(skb
)) | (id
& ~IP_DF
));
1474 list_for_each_entry(p
, head
, list
) {
1478 if (!NAPI_GRO_CB(p
)->same_flow
)
1481 iph2
= (struct iphdr
*)(p
->data
+ off
);
1482 /* The above works because, with the exception of the top
1483 * (inner most) layer, we only aggregate pkts with the same
1484 * hdr length so all the hdrs we'll need to verify will start
1485 * at the same offset.
1487 if ((iph
->protocol
^ iph2
->protocol
) |
1488 ((__force u32
)iph
->saddr
^ (__force u32
)iph2
->saddr
) |
1489 ((__force u32
)iph
->daddr
^ (__force u32
)iph2
->daddr
)) {
1490 NAPI_GRO_CB(p
)->same_flow
= 0;
1494 /* All fields must match except length and checksum. */
1495 NAPI_GRO_CB(p
)->flush
|=
1496 (iph
->ttl
^ iph2
->ttl
) |
1497 (iph
->tos
^ iph2
->tos
) |
1498 ((iph
->frag_off
^ iph2
->frag_off
) & htons(IP_DF
));
1500 NAPI_GRO_CB(p
)->flush
|= flush
;
1502 /* We need to store of the IP ID check to be included later
1503 * when we can verify that this packet does in fact belong
1506 flush_id
= (u16
)(id
- ntohs(iph2
->id
));
1508 /* This bit of code makes it much easier for us to identify
1509 * the cases where we are doing atomic vs non-atomic IP ID
1510 * checks. Specifically an atomic check can return IP ID
1511 * values 0 - 0xFFFF, while a non-atomic check can only
1512 * return 0 or 0xFFFF.
1514 if (!NAPI_GRO_CB(p
)->is_atomic
||
1515 !(iph
->frag_off
& htons(IP_DF
))) {
1516 flush_id
^= NAPI_GRO_CB(p
)->count
;
1517 flush_id
= flush_id
? 0xFFFF : 0;
1520 /* If the previous IP ID value was based on an atomic
1521 * datagram we can overwrite the value and ignore it.
1523 if (NAPI_GRO_CB(skb
)->is_atomic
)
1524 NAPI_GRO_CB(p
)->flush_id
= flush_id
;
1526 NAPI_GRO_CB(p
)->flush_id
|= flush_id
;
1529 NAPI_GRO_CB(skb
)->is_atomic
= !!(iph
->frag_off
& htons(IP_DF
));
1530 NAPI_GRO_CB(skb
)->flush
|= flush
;
1531 skb_set_network_header(skb
, off
);
1532 /* The above will be needed by the transport layer if there is one
1533 * immediately following this IP hdr.
1536 /* Note : No need to call skb_gro_postpull_rcsum() here,
1537 * as we already checked checksum over ipv4 header was 0
1539 skb_gro_pull(skb
, sizeof(*iph
));
1540 skb_set_transport_header(skb
, skb_gro_offset(skb
));
1542 pp
= indirect_call_gro_receive(tcp4_gro_receive
, udp4_gro_receive
,
1543 ops
->callbacks
.gro_receive
, head
, skb
);
1549 skb_gro_flush_final(skb
, pp
, flush
);
1553 EXPORT_SYMBOL(inet_gro_receive
);
1555 static struct sk_buff
*ipip_gro_receive(struct list_head
*head
,
1556 struct sk_buff
*skb
)
1558 if (NAPI_GRO_CB(skb
)->encap_mark
) {
1559 NAPI_GRO_CB(skb
)->flush
= 1;
1563 NAPI_GRO_CB(skb
)->encap_mark
= 1;
1565 return inet_gro_receive(head
, skb
);
1568 #define SECONDS_PER_DAY 86400
1570 /* inet_current_timestamp - Return IP network timestamp
1572 * Return milliseconds since midnight in network byte order.
1574 __be32
inet_current_timestamp(void)
1578 struct timespec64 ts
;
1580 ktime_get_real_ts64(&ts
);
1582 /* Get secs since midnight. */
1583 (void)div_u64_rem(ts
.tv_sec
, SECONDS_PER_DAY
, &secs
);
1584 /* Convert to msecs. */
1585 msecs
= secs
* MSEC_PER_SEC
;
1586 /* Convert nsec to msec. */
1587 msecs
+= (u32
)ts
.tv_nsec
/ NSEC_PER_MSEC
;
1589 /* Convert to network byte order. */
1590 return htonl(msecs
);
1592 EXPORT_SYMBOL(inet_current_timestamp
);
1594 int inet_recv_error(struct sock
*sk
, struct msghdr
*msg
, int len
, int *addr_len
)
1596 if (sk
->sk_family
== AF_INET
)
1597 return ip_recv_error(sk
, msg
, len
, addr_len
);
1598 #if IS_ENABLED(CONFIG_IPV6)
1599 if (sk
->sk_family
== AF_INET6
)
1600 return pingv6_ops
.ipv6_recv_error(sk
, msg
, len
, addr_len
);
1605 int inet_gro_complete(struct sk_buff
*skb
, int nhoff
)
1607 __be16 newlen
= htons(skb
->len
- nhoff
);
1608 struct iphdr
*iph
= (struct iphdr
*)(skb
->data
+ nhoff
);
1609 const struct net_offload
*ops
;
1610 int proto
= iph
->protocol
;
1613 if (skb
->encapsulation
) {
1614 skb_set_inner_protocol(skb
, cpu_to_be16(ETH_P_IP
));
1615 skb_set_inner_network_header(skb
, nhoff
);
1618 csum_replace2(&iph
->check
, iph
->tot_len
, newlen
);
1619 iph
->tot_len
= newlen
;
1622 ops
= rcu_dereference(inet_offloads
[proto
]);
1623 if (WARN_ON(!ops
|| !ops
->callbacks
.gro_complete
))
1626 /* Only need to add sizeof(*iph) to get to the next hdr below
1627 * because any hdr with option will have been flushed in
1628 * inet_gro_receive().
1630 err
= INDIRECT_CALL_2(ops
->callbacks
.gro_complete
,
1631 tcp4_gro_complete
, udp4_gro_complete
,
1632 skb
, nhoff
+ sizeof(*iph
));
1639 EXPORT_SYMBOL(inet_gro_complete
);
1641 static int ipip_gro_complete(struct sk_buff
*skb
, int nhoff
)
1643 skb
->encapsulation
= 1;
1644 skb_shinfo(skb
)->gso_type
|= SKB_GSO_IPXIP4
;
1645 return inet_gro_complete(skb
, nhoff
);
1648 int inet_ctl_sock_create(struct sock
**sk
, unsigned short family
,
1649 unsigned short type
, unsigned char protocol
,
1652 struct socket
*sock
;
1653 int rc
= sock_create_kern(net
, family
, type
, protocol
, &sock
);
1657 (*sk
)->sk_allocation
= GFP_ATOMIC
;
1659 * Unhash it so that IP input processing does not even see it,
1660 * we do not wish this socket to see incoming packets.
1662 (*sk
)->sk_prot
->unhash(*sk
);
1666 EXPORT_SYMBOL_GPL(inet_ctl_sock_create
);
1668 u64
snmp_get_cpu_field(void __percpu
*mib
, int cpu
, int offt
)
1670 return *(((unsigned long *)per_cpu_ptr(mib
, cpu
)) + offt
);
1672 EXPORT_SYMBOL_GPL(snmp_get_cpu_field
);
1674 unsigned long snmp_fold_field(void __percpu
*mib
, int offt
)
1676 unsigned long res
= 0;
1679 for_each_possible_cpu(i
)
1680 res
+= snmp_get_cpu_field(mib
, i
, offt
);
1683 EXPORT_SYMBOL_GPL(snmp_fold_field
);
1685 #if BITS_PER_LONG==32
1687 u64
snmp_get_cpu_field64(void __percpu
*mib
, int cpu
, int offt
,
1688 size_t syncp_offset
)
1691 struct u64_stats_sync
*syncp
;
1695 bhptr
= per_cpu_ptr(mib
, cpu
);
1696 syncp
= (struct u64_stats_sync
*)(bhptr
+ syncp_offset
);
1698 start
= u64_stats_fetch_begin_irq(syncp
);
1699 v
= *(((u64
*)bhptr
) + offt
);
1700 } while (u64_stats_fetch_retry_irq(syncp
, start
));
1704 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64
);
1706 u64
snmp_fold_field64(void __percpu
*mib
, int offt
, size_t syncp_offset
)
1711 for_each_possible_cpu(cpu
) {
1712 res
+= snmp_get_cpu_field64(mib
, cpu
, offt
, syncp_offset
);
1716 EXPORT_SYMBOL_GPL(snmp_fold_field64
);
1719 #ifdef CONFIG_IP_MULTICAST
1720 static const struct net_protocol igmp_protocol
= {
1721 .handler
= igmp_rcv
,
1726 /* thinking of making this const? Don't.
1727 * early_demux can change based on sysctl.
1729 static struct net_protocol tcp_protocol
= {
1730 .early_demux
= tcp_v4_early_demux
,
1731 .early_demux_handler
= tcp_v4_early_demux
,
1732 .handler
= tcp_v4_rcv
,
1733 .err_handler
= tcp_v4_err
,
1736 .icmp_strict_tag_validation
= 1,
1739 /* thinking of making this const? Don't.
1740 * early_demux can change based on sysctl.
1742 static struct net_protocol udp_protocol
= {
1743 .early_demux
= udp_v4_early_demux
,
1744 .early_demux_handler
= udp_v4_early_demux
,
1746 .err_handler
= udp_err
,
1751 static const struct net_protocol icmp_protocol
= {
1752 .handler
= icmp_rcv
,
1753 .err_handler
= icmp_err
,
1758 static __net_init
int ipv4_mib_init_net(struct net
*net
)
1762 net
->mib
.tcp_statistics
= alloc_percpu(struct tcp_mib
);
1763 if (!net
->mib
.tcp_statistics
)
1765 net
->mib
.ip_statistics
= alloc_percpu(struct ipstats_mib
);
1766 if (!net
->mib
.ip_statistics
)
1769 for_each_possible_cpu(i
) {
1770 struct ipstats_mib
*af_inet_stats
;
1771 af_inet_stats
= per_cpu_ptr(net
->mib
.ip_statistics
, i
);
1772 u64_stats_init(&af_inet_stats
->syncp
);
1775 net
->mib
.net_statistics
= alloc_percpu(struct linux_mib
);
1776 if (!net
->mib
.net_statistics
)
1778 net
->mib
.udp_statistics
= alloc_percpu(struct udp_mib
);
1779 if (!net
->mib
.udp_statistics
)
1781 net
->mib
.udplite_statistics
= alloc_percpu(struct udp_mib
);
1782 if (!net
->mib
.udplite_statistics
)
1783 goto err_udplite_mib
;
1784 net
->mib
.icmp_statistics
= alloc_percpu(struct icmp_mib
);
1785 if (!net
->mib
.icmp_statistics
)
1787 net
->mib
.icmpmsg_statistics
= kzalloc(sizeof(struct icmpmsg_mib
),
1789 if (!net
->mib
.icmpmsg_statistics
)
1790 goto err_icmpmsg_mib
;
1796 free_percpu(net
->mib
.icmp_statistics
);
1798 free_percpu(net
->mib
.udplite_statistics
);
1800 free_percpu(net
->mib
.udp_statistics
);
1802 free_percpu(net
->mib
.net_statistics
);
1804 free_percpu(net
->mib
.ip_statistics
);
1806 free_percpu(net
->mib
.tcp_statistics
);
1811 static __net_exit
void ipv4_mib_exit_net(struct net
*net
)
1813 kfree(net
->mib
.icmpmsg_statistics
);
1814 free_percpu(net
->mib
.icmp_statistics
);
1815 free_percpu(net
->mib
.udplite_statistics
);
1816 free_percpu(net
->mib
.udp_statistics
);
1817 free_percpu(net
->mib
.net_statistics
);
1818 free_percpu(net
->mib
.ip_statistics
);
1819 free_percpu(net
->mib
.tcp_statistics
);
1821 /* allocated on demand, see mptcp_init_sock() */
1822 free_percpu(net
->mib
.mptcp_statistics
);
1826 static __net_initdata
struct pernet_operations ipv4_mib_ops
= {
1827 .init
= ipv4_mib_init_net
,
1828 .exit
= ipv4_mib_exit_net
,
1831 static int __init
init_ipv4_mibs(void)
1833 return register_pernet_subsys(&ipv4_mib_ops
);
1836 static __net_init
int inet_init_net(struct net
*net
)
1839 * Set defaults for local port range
1841 seqlock_init(&net
->ipv4
.ip_local_ports
.lock
);
1842 net
->ipv4
.ip_local_ports
.range
[0] = 32768;
1843 net
->ipv4
.ip_local_ports
.range
[1] = 60999;
1845 seqlock_init(&net
->ipv4
.ping_group_range
.lock
);
1847 * Sane defaults - nobody may create ping sockets.
1848 * Boot scripts should set this to distro-specific group.
1850 net
->ipv4
.ping_group_range
.range
[0] = make_kgid(&init_user_ns
, 1);
1851 net
->ipv4
.ping_group_range
.range
[1] = make_kgid(&init_user_ns
, 0);
1853 /* Default values for sysctl-controlled parameters.
1854 * We set them here, in case sysctl is not compiled.
1856 net
->ipv4
.sysctl_ip_default_ttl
= IPDEFTTL
;
1857 net
->ipv4
.sysctl_ip_fwd_update_priority
= 1;
1858 net
->ipv4
.sysctl_ip_dynaddr
= 0;
1859 net
->ipv4
.sysctl_ip_early_demux
= 1;
1860 net
->ipv4
.sysctl_udp_early_demux
= 1;
1861 net
->ipv4
.sysctl_tcp_early_demux
= 1;
1862 net
->ipv4
.sysctl_nexthop_compat_mode
= 1;
1863 #ifdef CONFIG_SYSCTL
1864 net
->ipv4
.sysctl_ip_prot_sock
= PROT_SOCK
;
1867 /* Some igmp sysctl, whose values are always used */
1868 net
->ipv4
.sysctl_igmp_max_memberships
= 20;
1869 net
->ipv4
.sysctl_igmp_max_msf
= 10;
1870 /* IGMP reports for link-local multicast groups are enabled by default */
1871 net
->ipv4
.sysctl_igmp_llm_reports
= 1;
1872 net
->ipv4
.sysctl_igmp_qrv
= 2;
1877 static __net_initdata
struct pernet_operations af_inet_ops
= {
1878 .init
= inet_init_net
,
1881 static int __init
init_inet_pernet_ops(void)
1883 return register_pernet_subsys(&af_inet_ops
);
1886 static int ipv4_proc_init(void);
1889 * IP protocol layer initialiser
1892 static struct packet_offload ip_packet_offload __read_mostly
= {
1893 .type
= cpu_to_be16(ETH_P_IP
),
1895 .gso_segment
= inet_gso_segment
,
1896 .gro_receive
= inet_gro_receive
,
1897 .gro_complete
= inet_gro_complete
,
1901 static const struct net_offload ipip_offload
= {
1903 .gso_segment
= ipip_gso_segment
,
1904 .gro_receive
= ipip_gro_receive
,
1905 .gro_complete
= ipip_gro_complete
,
1909 static int __init
ipip_offload_init(void)
1911 return inet_add_offload(&ipip_offload
, IPPROTO_IPIP
);
1914 static int __init
ipv4_offload_init(void)
1919 if (udpv4_offload_init() < 0)
1920 pr_crit("%s: Cannot add UDP protocol offload\n", __func__
);
1921 if (tcpv4_offload_init() < 0)
1922 pr_crit("%s: Cannot add TCP protocol offload\n", __func__
);
1923 if (ipip_offload_init() < 0)
1924 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__
);
1926 dev_add_offload(&ip_packet_offload
);
1930 fs_initcall(ipv4_offload_init
);
1932 static struct packet_type ip_packet_type __read_mostly
= {
1933 .type
= cpu_to_be16(ETH_P_IP
),
1935 .list_func
= ip_list_rcv
,
1938 static int __init
inet_init(void)
1940 struct inet_protosw
*q
;
1941 struct list_head
*r
;
1944 sock_skb_cb_check_size(sizeof(struct inet_skb_parm
));
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 /* Setup TCP slab cache for open requests. */
2009 /* Setup UDP memory threshold */
2012 /* Add UDP-Lite (RFC 3828) */
2013 udplite4_register();
2020 * Set the ICMP layer up
2023 if (icmp_init() < 0)
2024 panic("Failed to create the ICMP control socket.\n");
2027 * Initialise the multicast router
2029 #if defined(CONFIG_IP_MROUTE)
2031 pr_crit("%s: Cannot init ipv4 mroute\n", __func__
);
2034 if (init_inet_pernet_ops())
2035 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__
);
2037 * Initialise per-cpu ipv4 mibs
2040 if (init_ipv4_mibs())
2041 pr_crit("%s: Cannot init ipv4 mibs\n", __func__
);
2047 dev_add_pack(&ip_packet_type
);
2049 ip_tunnel_core_init();
2054 out_unregister_raw_proto
:
2055 proto_unregister(&raw_prot
);
2056 out_unregister_udp_proto
:
2057 proto_unregister(&udp_prot
);
2058 out_unregister_tcp_proto
:
2059 proto_unregister(&tcp_prot
);
2063 fs_initcall(inet_init
);
2065 /* ------------------------------------------------------------------------ */
2067 #ifdef CONFIG_PROC_FS
2068 static int __init
ipv4_proc_init(void)
2072 if (raw_proc_init())
2074 if (tcp4_proc_init())
2076 if (udp4_proc_init())
2078 if (ping_proc_init())
2080 if (ip_misc_proc_init())
2097 #else /* CONFIG_PROC_FS */
2098 static int __init
ipv4_proc_init(void)
2102 #endif /* CONFIG_PROC_FS */