1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
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.
6 *
7 * PF_INET protocol family socket handler.
8 *
9 * Authors: Ross Biro
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>
13 *
14 * Changes (see also sock.c)
15 *
16 * piggy,
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
24 * structures
25 * when accept() ed
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
31 * Alan Cox,
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
41 * compactness.
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
47 * dumbly.
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.
62 */
63
64#define pr_fmt(fmt) "IPv4: " fmt
65
66#include <linux/err.h>
67#include <linux/errno.h>
68#include <linux/types.h>
69#include <linux/socket.h>
70#include <linux/in.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>
80#include <linux/mm.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>
88
89#include <linux/uaccess.h>
90
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>
96#include <net/ip.h>
97#include <net/protocol.h>
98#include <net/arp.h>
99#include <net/route.h>
100#include <net/ip_fib.h>
101#include <net/inet_connection_sock.h>
102#include <net/gro.h>
103#include <net/gso.h>
104#include <net/tcp.h>
105#include <net/udp.h>
106#include <net/udplite.h>
107#include <net/ping.h>
108#include <linux/skbuff.h>
109#include <net/sock.h>
110#include <net/raw.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>
119#endif
120#include <net/l3mdev.h>
121#include <net/compat.h>
122
123#include <trace/events/sock.h>
124
125/* The inetsw table contains everything that inet_create needs to
126 * build a new socket.
127 */
128static struct list_head inetsw[SOCK_MAX];
129static DEFINE_SPINLOCK(inetsw_lock);
130
131/* New destruction routine */
132
133void inet_sock_destruct(struct sock *sk)
134{
135 struct inet_sock *inet = inet_sk(sk);
136
137 __skb_queue_purge(list: &sk->sk_receive_queue);
138 __skb_queue_purge(list: &sk->sk_error_queue);
139
140 sk_mem_reclaim_final(sk);
141
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",
144 sk->sk_state, sk);
145 return;
146 }
147 if (!sock_flag(sk, flag: SOCK_DEAD)) {
148 pr_err("Attempt to release alive inet socket %p\n", sk);
149 return;
150 }
151
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));
156
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));
160}
161EXPORT_SYMBOL(inet_sock_destruct);
162
163/*
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
166 * the work.
167 */
168
169/*
170 * Automatically bind an unbound socket.
171 */
172
173static int inet_autobind(struct sock *sk)
174{
175 struct inet_sock *inet;
176 /* We may need to bind the socket. */
177 lock_sock(sk);
178 inet = inet_sk(sk);
179 if (!inet->inet_num) {
180 if (sk->sk_prot->get_port(sk, 0)) {
181 release_sock(sk);
182 return -EAGAIN;
183 }
184 inet->inet_sport = htons(inet->inet_num);
185 }
186 release_sock(sk);
187 return 0;
188}
189
190int __inet_listen_sk(struct sock *sk, int backlog)
191{
192 unsigned char old_state = sk->sk_state;
193 int err, tcp_fastopen;
194
195 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
196 return -EINVAL;
197
198 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
199 /* Really, if the socket is already in listen state
200 * we can only allow the backlog to be adjusted.
201 */
202 if (old_state != TCP_LISTEN) {
203 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
204 * Note that only TCP sockets (SOCK_STREAM) will reach here.
205 * Also fastopen backlog may already been set via the option
206 * because the socket was in TCP_LISTEN state previously but
207 * was shutdown() rather than close().
208 */
209 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
210 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
211 (tcp_fastopen & TFO_SERVER_ENABLE) &&
212 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
213 fastopen_queue_tune(sk, backlog);
214 tcp_fastopen_init_key_once(net: sock_net(sk));
215 }
216
217 err = inet_csk_listen_start(sk);
218 if (err)
219 return err;
220
221 tcp_call_bpf(sk, op: BPF_SOCK_OPS_TCP_LISTEN_CB, nargs: 0, NULL);
222 }
223 return 0;
224}
225
226/*
227 * Move a socket into listening state.
228 */
229int inet_listen(struct socket *sock, int backlog)
230{
231 struct sock *sk = sock->sk;
232 int err = -EINVAL;
233
234 lock_sock(sk);
235
236 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
237 goto out;
238
239 err = __inet_listen_sk(sk, backlog);
240
241out:
242 release_sock(sk);
243 return err;
244}
245EXPORT_SYMBOL(inet_listen);
246
247/*
248 * Create an inet socket.
249 */
250
251static int inet_create(struct net *net, struct socket *sock, int protocol,
252 int kern)
253{
254 struct sock *sk;
255 struct inet_protosw *answer;
256 struct inet_sock *inet;
257 struct proto *answer_prot;
258 unsigned char answer_flags;
259 int try_loading_module = 0;
260 int err;
261
262 if (protocol < 0 || protocol >= IPPROTO_MAX)
263 return -EINVAL;
264
265 sock->state = SS_UNCONNECTED;
266
267 /* Look for the requested type/protocol pair. */
268lookup_protocol:
269 err = -ESOCKTNOSUPPORT;
270 rcu_read_lock();
271 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
272
273 err = 0;
274 /* Check the non-wild match. */
275 if (protocol == answer->protocol) {
276 if (protocol != IPPROTO_IP)
277 break;
278 } else {
279 /* Check for the two wild cases. */
280 if (IPPROTO_IP == protocol) {
281 protocol = answer->protocol;
282 break;
283 }
284 if (IPPROTO_IP == answer->protocol)
285 break;
286 }
287 err = -EPROTONOSUPPORT;
288 }
289
290 if (unlikely(err)) {
291 if (try_loading_module < 2) {
292 rcu_read_unlock();
293 /*
294 * Be more specific, e.g. net-pf-2-proto-132-type-1
295 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
296 */
297 if (++try_loading_module == 1)
298 request_module("net-pf-%d-proto-%d-type-%d",
299 PF_INET, protocol, sock->type);
300 /*
301 * Fall back to generic, e.g. net-pf-2-proto-132
302 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
303 */
304 else
305 request_module("net-pf-%d-proto-%d",
306 PF_INET, protocol);
307 goto lookup_protocol;
308 } else
309 goto out_rcu_unlock;
310 }
311
312 err = -EPERM;
313 if (sock->type == SOCK_RAW && !kern &&
314 !ns_capable(ns: net->user_ns, CAP_NET_RAW))
315 goto out_rcu_unlock;
316
317 sock->ops = answer->ops;
318 answer_prot = answer->prot;
319 answer_flags = answer->flags;
320 rcu_read_unlock();
321
322 WARN_ON(!answer_prot->slab);
323
324 err = -ENOMEM;
325 sk = sk_alloc(net, PF_INET, GFP_KERNEL, prot: answer_prot, kern);
326 if (!sk)
327 goto out;
328
329 err = 0;
330 if (INET_PROTOSW_REUSE & answer_flags)
331 sk->sk_reuse = SK_CAN_REUSE;
332
333 inet = inet_sk(sk);
334 inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
335
336 inet_clear_bit(NODEFRAG, sk);
337
338 if (SOCK_RAW == sock->type) {
339 inet->inet_num = protocol;
340 if (IPPROTO_RAW == protocol)
341 inet_set_bit(HDRINCL, sk);
342 }
343
344 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
345 inet->pmtudisc = IP_PMTUDISC_DONT;
346 else
347 inet->pmtudisc = IP_PMTUDISC_WANT;
348
349 atomic_set(v: &inet->inet_id, i: 0);
350
351 sock_init_data(sock, sk);
352
353 sk->sk_destruct = inet_sock_destruct;
354 sk->sk_protocol = protocol;
355 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
356 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
357
358 inet->uc_ttl = -1;
359 inet_set_bit(MC_LOOP, sk);
360 inet->mc_ttl = 1;
361 inet_set_bit(MC_ALL, sk);
362 inet->mc_index = 0;
363 inet->mc_list = NULL;
364 inet->rcv_tos = 0;
365
366 if (inet->inet_num) {
367 /* It assumes that any protocol which allows
368 * the user to assign a number at socket
369 * creation time automatically
370 * shares.
371 */
372 inet->inet_sport = htons(inet->inet_num);
373 /* Add to protocol hash chains. */
374 err = sk->sk_prot->hash(sk);
375 if (err) {
376 sk_common_release(sk);
377 goto out;
378 }
379 }
380
381 if (sk->sk_prot->init) {
382 err = sk->sk_prot->init(sk);
383 if (err) {
384 sk_common_release(sk);
385 goto out;
386 }
387 }
388
389 if (!kern) {
390 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
391 if (err) {
392 sk_common_release(sk);
393 goto out;
394 }
395 }
396out:
397 return err;
398out_rcu_unlock:
399 rcu_read_unlock();
400 goto out;
401}
402
403
404/*
405 * The peer socket should always be NULL (or else). When we call this
406 * function we are destroying the object and from then on nobody
407 * should refer to it.
408 */
409int inet_release(struct socket *sock)
410{
411 struct sock *sk = sock->sk;
412
413 if (sk) {
414 long timeout;
415
416 if (!sk->sk_kern_sock)
417 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
418
419 /* Applications forget to leave groups before exiting */
420 ip_mc_drop_socket(sk);
421
422 /* If linger is set, we don't return until the close
423 * is complete. Otherwise we return immediately. The
424 * actually closing is done the same either way.
425 *
426 * If the close is due to the process exiting, we never
427 * linger..
428 */
429 timeout = 0;
430 if (sock_flag(sk, flag: SOCK_LINGER) &&
431 !(current->flags & PF_EXITING))
432 timeout = sk->sk_lingertime;
433 sk->sk_prot->close(sk, timeout);
434 sock->sk = NULL;
435 }
436 return 0;
437}
438EXPORT_SYMBOL(inet_release);
439
440int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
441{
442 u32 flags = BIND_WITH_LOCK;
443 int err;
444
445 /* If the socket has its own bind function then use it. (RAW) */
446 if (sk->sk_prot->bind) {
447 return sk->sk_prot->bind(sk, uaddr, addr_len);
448 }
449 if (addr_len < sizeof(struct sockaddr_in))
450 return -EINVAL;
451
452 /* BPF prog is run before any checks are done so that if the prog
453 * changes context in a wrong way it will be caught.
454 */
455 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
456 CGROUP_INET4_BIND, &flags);
457 if (err)
458 return err;
459
460 return __inet_bind(sk, uaddr, addr_len, flags);
461}
462
463int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
464{
465 return inet_bind_sk(sk: sock->sk, uaddr, addr_len);
466}
467EXPORT_SYMBOL(inet_bind);
468
469int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
470 u32 flags)
471{
472 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
473 struct inet_sock *inet = inet_sk(sk);
474 struct net *net = sock_net(sk);
475 unsigned short snum;
476 int chk_addr_ret;
477 u32 tb_id = RT_TABLE_LOCAL;
478 int err;
479
480 if (addr->sin_family != AF_INET) {
481 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
482 * only if s_addr is INADDR_ANY.
483 */
484 err = -EAFNOSUPPORT;
485 if (addr->sin_family != AF_UNSPEC ||
486 addr->sin_addr.s_addr != htonl(INADDR_ANY))
487 goto out;
488 }
489
490 tb_id = l3mdev_fib_table_by_index(net, ifindex: sk->sk_bound_dev_if) ? : tb_id;
491 chk_addr_ret = inet_addr_type_table(net, addr: addr->sin_addr.s_addr, tb_id);
492
493 /* Not specified by any standard per-se, however it breaks too
494 * many applications when removed. It is unfortunate since
495 * allowing applications to make a non-local bind solves
496 * several problems with systems using dynamic addressing.
497 * (ie. your servers still start up even if your ISDN link
498 * is temporarily down)
499 */
500 err = -EADDRNOTAVAIL;
501 if (!inet_addr_valid_or_nonlocal(net, inet, addr: addr->sin_addr.s_addr,
502 addr_type: chk_addr_ret))
503 goto out;
504
505 snum = ntohs(addr->sin_port);
506 err = -EACCES;
507 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
508 snum && inet_port_requires_bind_service(net, port: snum) &&
509 !ns_capable(ns: net->user_ns, CAP_NET_BIND_SERVICE))
510 goto out;
511
512 /* We keep a pair of addresses. rcv_saddr is the one
513 * used by hash lookups, and saddr is used for transmit.
514 *
515 * In the BSD API these are the same except where it
516 * would be illegal to use them (multicast/broadcast) in
517 * which case the sending device address is used.
518 */
519 if (flags & BIND_WITH_LOCK)
520 lock_sock(sk);
521
522 /* Check these errors (active socket, double bind). */
523 err = -EINVAL;
524 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
525 goto out_release_sock;
526
527 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
528 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
529 inet->inet_saddr = 0; /* Use device */
530
531 /* Make sure we are allowed to bind here. */
532 if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
533 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
534 err = sk->sk_prot->get_port(sk, snum);
535 if (err) {
536 inet->inet_saddr = inet->inet_rcv_saddr = 0;
537 goto out_release_sock;
538 }
539 if (!(flags & BIND_FROM_BPF)) {
540 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
541 if (err) {
542 inet->inet_saddr = inet->inet_rcv_saddr = 0;
543 if (sk->sk_prot->put_port)
544 sk->sk_prot->put_port(sk);
545 goto out_release_sock;
546 }
547 }
548 }
549
550 if (inet->inet_rcv_saddr)
551 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
552 if (snum)
553 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
554 inet->inet_sport = htons(inet->inet_num);
555 inet->inet_daddr = 0;
556 inet->inet_dport = 0;
557 sk_dst_reset(sk);
558 err = 0;
559out_release_sock:
560 if (flags & BIND_WITH_LOCK)
561 release_sock(sk);
562out:
563 return err;
564}
565
566int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
567 int addr_len, int flags)
568{
569 struct sock *sk = sock->sk;
570 const struct proto *prot;
571 int err;
572
573 if (addr_len < sizeof(uaddr->sa_family))
574 return -EINVAL;
575
576 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
577 prot = READ_ONCE(sk->sk_prot);
578
579 if (uaddr->sa_family == AF_UNSPEC)
580 return prot->disconnect(sk, flags);
581
582 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
583 err = prot->pre_connect(sk, uaddr, addr_len);
584 if (err)
585 return err;
586 }
587
588 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
589 return -EAGAIN;
590 return prot->connect(sk, uaddr, addr_len);
591}
592EXPORT_SYMBOL(inet_dgram_connect);
593
594static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
595{
596 DEFINE_WAIT_FUNC(wait, woken_wake_function);
597
598 add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
599 sk->sk_write_pending += writebias;
600
601 /* Basic assumption: if someone sets sk->sk_err, he _must_
602 * change state of the socket from TCP_SYN_*.
603 * Connect() does not allow to get error notifications
604 * without closing the socket.
605 */
606 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
607 release_sock(sk);
608 timeo = wait_woken(wq_entry: &wait, TASK_INTERRUPTIBLE, timeout: timeo);
609 lock_sock(sk);
610 if (signal_pending(current) || !timeo)
611 break;
612 }
613 remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
614 sk->sk_write_pending -= writebias;
615 return timeo;
616}
617
618/*
619 * Connect to a remote host. There is regrettably still a little
620 * TCP 'magic' in here.
621 */
622int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
623 int addr_len, int flags, int is_sendmsg)
624{
625 struct sock *sk = sock->sk;
626 int err;
627 long timeo;
628
629 /*
630 * uaddr can be NULL and addr_len can be 0 if:
631 * sk is a TCP fastopen active socket and
632 * TCP_FASTOPEN_CONNECT sockopt is set and
633 * we already have a valid cookie for this socket.
634 * In this case, user can call write() after connect().
635 * write() will invoke tcp_sendmsg_fastopen() which calls
636 * __inet_stream_connect().
637 */
638 if (uaddr) {
639 if (addr_len < sizeof(uaddr->sa_family))
640 return -EINVAL;
641
642 if (uaddr->sa_family == AF_UNSPEC) {
643 sk->sk_disconnects++;
644 err = sk->sk_prot->disconnect(sk, flags);
645 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
646 goto out;
647 }
648 }
649
650 switch (sock->state) {
651 default:
652 err = -EINVAL;
653 goto out;
654 case SS_CONNECTED:
655 err = -EISCONN;
656 goto out;
657 case SS_CONNECTING:
658 if (inet_test_bit(DEFER_CONNECT, sk))
659 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
660 else
661 err = -EALREADY;
662 /* Fall out of switch with err, set for this state */
663 break;
664 case SS_UNCONNECTED:
665 err = -EISCONN;
666 if (sk->sk_state != TCP_CLOSE)
667 goto out;
668
669 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
670 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
671 if (err)
672 goto out;
673 }
674
675 err = sk->sk_prot->connect(sk, uaddr, addr_len);
676 if (err < 0)
677 goto out;
678
679 sock->state = SS_CONNECTING;
680
681 if (!err && inet_test_bit(DEFER_CONNECT, sk))
682 goto out;
683
684 /* Just entered SS_CONNECTING state; the only
685 * difference is that return value in non-blocking
686 * case is EINPROGRESS, rather than EALREADY.
687 */
688 err = -EINPROGRESS;
689 break;
690 }
691
692 timeo = sock_sndtimeo(sk, noblock: flags & O_NONBLOCK);
693
694 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
695 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
696 tcp_sk(sk)->fastopen_req &&
697 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
698 int dis = sk->sk_disconnects;
699
700 /* Error code is set above */
701 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
702 goto out;
703
704 err = sock_intr_errno(timeo);
705 if (signal_pending(current))
706 goto out;
707
708 if (dis != sk->sk_disconnects) {
709 err = -EPIPE;
710 goto out;
711 }
712 }
713
714 /* Connection was closed by RST, timeout, ICMP error
715 * or another process disconnected us.
716 */
717 if (sk->sk_state == TCP_CLOSE)
718 goto sock_error;
719
720 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
721 * and error was received after socket entered established state.
722 * Hence, it is handled normally after connect() return successfully.
723 */
724
725 sock->state = SS_CONNECTED;
726 err = 0;
727out:
728 return err;
729
730sock_error:
731 err = sock_error(sk) ? : -ECONNABORTED;
732 sock->state = SS_UNCONNECTED;
733 sk->sk_disconnects++;
734 if (sk->sk_prot->disconnect(sk, flags))
735 sock->state = SS_DISCONNECTING;
736 goto out;
737}
738EXPORT_SYMBOL(__inet_stream_connect);
739
740int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
741 int addr_len, int flags)
742{
743 int err;
744
745 lock_sock(sk: sock->sk);
746 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
747 release_sock(sk: sock->sk);
748 return err;
749}
750EXPORT_SYMBOL(inet_stream_connect);
751
752void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
753{
754 sock_rps_record_flow(sk: newsk);
755 WARN_ON(!((1 << newsk->sk_state) &
756 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
757 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
758
759 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
760 set_bit(SOCK_SUPPORT_ZC, addr: &newsock->flags);
761 sock_graft(sk: newsk, parent: newsock);
762
763 newsock->state = SS_CONNECTED;
764}
765
766/*
767 * Accept a pending connection. The TCP layer now gives BSD semantics.
768 */
769
770int inet_accept(struct socket *sock, struct socket *newsock, int flags,
771 bool kern)
772{
773 struct sock *sk1 = sock->sk, *sk2;
774 int err = -EINVAL;
775
776 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
777 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
778 if (!sk2)
779 return err;
780
781 lock_sock(sk: sk2);
782 __inet_accept(sock, newsock, newsk: sk2);
783 release_sock(sk: sk2);
784 return 0;
785}
786EXPORT_SYMBOL(inet_accept);
787
788/*
789 * This does both peername and sockname.
790 */
791int inet_getname(struct socket *sock, struct sockaddr *uaddr,
792 int peer)
793{
794 struct sock *sk = sock->sk;
795 struct inet_sock *inet = inet_sk(sk);
796 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
797 int sin_addr_len = sizeof(*sin);
798
799 sin->sin_family = AF_INET;
800 lock_sock(sk);
801 if (peer) {
802 if (!inet->inet_dport ||
803 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
804 peer == 1)) {
805 release_sock(sk);
806 return -ENOTCONN;
807 }
808 sin->sin_port = inet->inet_dport;
809 sin->sin_addr.s_addr = inet->inet_daddr;
810 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
811 CGROUP_INET4_GETPEERNAME);
812 } else {
813 __be32 addr = inet->inet_rcv_saddr;
814 if (!addr)
815 addr = inet->inet_saddr;
816 sin->sin_port = inet->inet_sport;
817 sin->sin_addr.s_addr = addr;
818 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
819 CGROUP_INET4_GETSOCKNAME);
820 }
821 release_sock(sk);
822 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
823 return sin_addr_len;
824}
825EXPORT_SYMBOL(inet_getname);
826
827int inet_send_prepare(struct sock *sk)
828{
829 sock_rps_record_flow(sk);
830
831 /* We may need to bind the socket. */
832 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
833 inet_autobind(sk))
834 return -EAGAIN;
835
836 return 0;
837}
838EXPORT_SYMBOL_GPL(inet_send_prepare);
839
840int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
841{
842 struct sock *sk = sock->sk;
843
844 if (unlikely(inet_send_prepare(sk)))
845 return -EAGAIN;
846
847 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
848 sk, msg, size);
849}
850EXPORT_SYMBOL(inet_sendmsg);
851
852void inet_splice_eof(struct socket *sock)
853{
854 const struct proto *prot;
855 struct sock *sk = sock->sk;
856
857 if (unlikely(inet_send_prepare(sk)))
858 return;
859
860 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
861 prot = READ_ONCE(sk->sk_prot);
862 if (prot->splice_eof)
863 prot->splice_eof(sock);
864}
865EXPORT_SYMBOL_GPL(inet_splice_eof);
866
867INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
868 size_t, int, int *));
869int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
870 int flags)
871{
872 struct sock *sk = sock->sk;
873 int addr_len = 0;
874 int err;
875
876 if (likely(!(flags & MSG_ERRQUEUE)))
877 sock_rps_record_flow(sk);
878
879 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
880 sk, msg, size, flags, &addr_len);
881 if (err >= 0)
882 msg->msg_namelen = addr_len;
883 return err;
884}
885EXPORT_SYMBOL(inet_recvmsg);
886
887int inet_shutdown(struct socket *sock, int how)
888{
889 struct sock *sk = sock->sk;
890 int err = 0;
891
892 /* This should really check to make sure
893 * the socket is a TCP socket. (WHY AC...)
894 */
895 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
896 1->2 bit 2 snds.
897 2->3 */
898 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
899 return -EINVAL;
900
901 lock_sock(sk);
902 if (sock->state == SS_CONNECTING) {
903 if ((1 << sk->sk_state) &
904 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
905 sock->state = SS_DISCONNECTING;
906 else
907 sock->state = SS_CONNECTED;
908 }
909
910 switch (sk->sk_state) {
911 case TCP_CLOSE:
912 err = -ENOTCONN;
913 /* Hack to wake up other listeners, who can poll for
914 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
915 fallthrough;
916 default:
917 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
918 if (sk->sk_prot->shutdown)
919 sk->sk_prot->shutdown(sk, how);
920 break;
921
922 /* Remaining two branches are temporary solution for missing
923 * close() in multithreaded environment. It is _not_ a good idea,
924 * but we have no choice until close() is repaired at VFS level.
925 */
926 case TCP_LISTEN:
927 if (!(how & RCV_SHUTDOWN))
928 break;
929 fallthrough;
930 case TCP_SYN_SENT:
931 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
932 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
933 break;
934 }
935
936 /* Wake up anyone sleeping in poll. */
937 sk->sk_state_change(sk);
938 release_sock(sk);
939 return err;
940}
941EXPORT_SYMBOL(inet_shutdown);
942
943/*
944 * ioctl() calls you can issue on an INET socket. Most of these are
945 * device configuration and stuff and very rarely used. Some ioctls
946 * pass on to the socket itself.
947 *
948 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
949 * loads the devconfigure module does its configuring and unloads it.
950 * There's a good 20K of config code hanging around the kernel.
951 */
952
953int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
954{
955 struct sock *sk = sock->sk;
956 int err = 0;
957 struct net *net = sock_net(sk);
958 void __user *p = (void __user *)arg;
959 struct ifreq ifr;
960 struct rtentry rt;
961
962 switch (cmd) {
963 case SIOCADDRT:
964 case SIOCDELRT:
965 if (copy_from_user(to: &rt, from: p, n: sizeof(struct rtentry)))
966 return -EFAULT;
967 err = ip_rt_ioctl(net, cmd, rt: &rt);
968 break;
969 case SIOCRTMSG:
970 err = -EINVAL;
971 break;
972 case SIOCDARP:
973 case SIOCGARP:
974 case SIOCSARP:
975 err = arp_ioctl(net, cmd, arg: (void __user *)arg);
976 break;
977 case SIOCGIFADDR:
978 case SIOCGIFBRDADDR:
979 case SIOCGIFNETMASK:
980 case SIOCGIFDSTADDR:
981 case SIOCGIFPFLAGS:
982 if (get_user_ifreq(ifr: &ifr, NULL, arg: p))
983 return -EFAULT;
984 err = devinet_ioctl(net, cmd, &ifr);
985 if (!err && put_user_ifreq(ifr: &ifr, arg: p))
986 err = -EFAULT;
987 break;
988
989 case SIOCSIFADDR:
990 case SIOCSIFBRDADDR:
991 case SIOCSIFNETMASK:
992 case SIOCSIFDSTADDR:
993 case SIOCSIFPFLAGS:
994 case SIOCSIFFLAGS:
995 if (get_user_ifreq(ifr: &ifr, NULL, arg: p))
996 return -EFAULT;
997 err = devinet_ioctl(net, cmd, &ifr);
998 break;
999 default:
1000 if (sk->sk_prot->ioctl)
1001 err = sk_ioctl(sk, cmd, arg: (void __user *)arg);
1002 else
1003 err = -ENOIOCTLCMD;
1004 break;
1005 }
1006 return err;
1007}
1008EXPORT_SYMBOL(inet_ioctl);
1009
1010#ifdef CONFIG_COMPAT
1011static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1012 struct compat_rtentry __user *ur)
1013{
1014 compat_uptr_t rtdev;
1015 struct rtentry rt;
1016
1017 if (copy_from_user(to: &rt.rt_dst, from: &ur->rt_dst,
1018 n: 3 * sizeof(struct sockaddr)) ||
1019 get_user(rt.rt_flags, &ur->rt_flags) ||
1020 get_user(rt.rt_metric, &ur->rt_metric) ||
1021 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1022 get_user(rt.rt_window, &ur->rt_window) ||
1023 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1024 get_user(rtdev, &ur->rt_dev))
1025 return -EFAULT;
1026
1027 rt.rt_dev = compat_ptr(uptr: rtdev);
1028 return ip_rt_ioctl(sock_net(sk), cmd, rt: &rt);
1029}
1030
1031static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1032{
1033 void __user *argp = compat_ptr(uptr: arg);
1034 struct sock *sk = sock->sk;
1035
1036 switch (cmd) {
1037 case SIOCADDRT:
1038 case SIOCDELRT:
1039 return inet_compat_routing_ioctl(sk, cmd, ur: argp);
1040 default:
1041 if (!sk->sk_prot->compat_ioctl)
1042 return -ENOIOCTLCMD;
1043 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1044 }
1045}
1046#endif /* CONFIG_COMPAT */
1047
1048const struct proto_ops inet_stream_ops = {
1049 .family = PF_INET,
1050 .owner = THIS_MODULE,
1051 .release = inet_release,
1052 .bind = inet_bind,
1053 .connect = inet_stream_connect,
1054 .socketpair = sock_no_socketpair,
1055 .accept = inet_accept,
1056 .getname = inet_getname,
1057 .poll = tcp_poll,
1058 .ioctl = inet_ioctl,
1059 .gettstamp = sock_gettstamp,
1060 .listen = inet_listen,
1061 .shutdown = inet_shutdown,
1062 .setsockopt = sock_common_setsockopt,
1063 .getsockopt = sock_common_getsockopt,
1064 .sendmsg = inet_sendmsg,
1065 .recvmsg = inet_recvmsg,
1066#ifdef CONFIG_MMU
1067 .mmap = tcp_mmap,
1068#endif
1069 .splice_eof = inet_splice_eof,
1070 .splice_read = tcp_splice_read,
1071 .read_sock = tcp_read_sock,
1072 .read_skb = tcp_read_skb,
1073 .sendmsg_locked = tcp_sendmsg_locked,
1074 .peek_len = tcp_peek_len,
1075#ifdef CONFIG_COMPAT
1076 .compat_ioctl = inet_compat_ioctl,
1077#endif
1078 .set_rcvlowat = tcp_set_rcvlowat,
1079};
1080EXPORT_SYMBOL(inet_stream_ops);
1081
1082const struct proto_ops inet_dgram_ops = {
1083 .family = PF_INET,
1084 .owner = THIS_MODULE,
1085 .release = inet_release,
1086 .bind = inet_bind,
1087 .connect = inet_dgram_connect,
1088 .socketpair = sock_no_socketpair,
1089 .accept = sock_no_accept,
1090 .getname = inet_getname,
1091 .poll = udp_poll,
1092 .ioctl = inet_ioctl,
1093 .gettstamp = sock_gettstamp,
1094 .listen = sock_no_listen,
1095 .shutdown = inet_shutdown,
1096 .setsockopt = sock_common_setsockopt,
1097 .getsockopt = sock_common_getsockopt,
1098 .sendmsg = inet_sendmsg,
1099 .read_skb = udp_read_skb,
1100 .recvmsg = inet_recvmsg,
1101 .mmap = sock_no_mmap,
1102 .splice_eof = inet_splice_eof,
1103 .set_peek_off = sk_set_peek_off,
1104#ifdef CONFIG_COMPAT
1105 .compat_ioctl = inet_compat_ioctl,
1106#endif
1107};
1108EXPORT_SYMBOL(inet_dgram_ops);
1109
1110/*
1111 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1112 * udp_poll
1113 */
1114static const struct proto_ops inet_sockraw_ops = {
1115 .family = PF_INET,
1116 .owner = THIS_MODULE,
1117 .release = inet_release,
1118 .bind = inet_bind,
1119 .connect = inet_dgram_connect,
1120 .socketpair = sock_no_socketpair,
1121 .accept = sock_no_accept,
1122 .getname = inet_getname,
1123 .poll = datagram_poll,
1124 .ioctl = inet_ioctl,
1125 .gettstamp = sock_gettstamp,
1126 .listen = sock_no_listen,
1127 .shutdown = inet_shutdown,
1128 .setsockopt = sock_common_setsockopt,
1129 .getsockopt = sock_common_getsockopt,
1130 .sendmsg = inet_sendmsg,
1131 .recvmsg = inet_recvmsg,
1132 .mmap = sock_no_mmap,
1133 .splice_eof = inet_splice_eof,
1134#ifdef CONFIG_COMPAT
1135 .compat_ioctl = inet_compat_ioctl,
1136#endif
1137};
1138
1139static const struct net_proto_family inet_family_ops = {
1140 .family = PF_INET,
1141 .create = inet_create,
1142 .owner = THIS_MODULE,
1143};
1144
1145/* Upon startup we insert all the elements in inetsw_array[] into
1146 * the linked list inetsw.
1147 */
1148static struct inet_protosw inetsw_array[] =
1149{
1150 {
1151 .type = SOCK_STREAM,
1152 .protocol = IPPROTO_TCP,
1153 .prot = &tcp_prot,
1154 .ops = &inet_stream_ops,
1155 .flags = INET_PROTOSW_PERMANENT |
1156 INET_PROTOSW_ICSK,
1157 },
1158
1159 {
1160 .type = SOCK_DGRAM,
1161 .protocol = IPPROTO_UDP,
1162 .prot = &udp_prot,
1163 .ops = &inet_dgram_ops,
1164 .flags = INET_PROTOSW_PERMANENT,
1165 },
1166
1167 {
1168 .type = SOCK_DGRAM,
1169 .protocol = IPPROTO_ICMP,
1170 .prot = &ping_prot,
1171 .ops = &inet_sockraw_ops,
1172 .flags = INET_PROTOSW_REUSE,
1173 },
1174
1175 {
1176 .type = SOCK_RAW,
1177 .protocol = IPPROTO_IP, /* wild card */
1178 .prot = &raw_prot,
1179 .ops = &inet_sockraw_ops,
1180 .flags = INET_PROTOSW_REUSE,
1181 }
1182};
1183
1184#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1185
1186void inet_register_protosw(struct inet_protosw *p)
1187{
1188 struct list_head *lh;
1189 struct inet_protosw *answer;
1190 int protocol = p->protocol;
1191 struct list_head *last_perm;
1192
1193 spin_lock_bh(lock: &inetsw_lock);
1194
1195 if (p->type >= SOCK_MAX)
1196 goto out_illegal;
1197
1198 /* If we are trying to override a permanent protocol, bail. */
1199 last_perm = &inetsw[p->type];
1200 list_for_each(lh, &inetsw[p->type]) {
1201 answer = list_entry(lh, struct inet_protosw, list);
1202 /* Check only the non-wild match. */
1203 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1204 break;
1205 if (protocol == answer->protocol)
1206 goto out_permanent;
1207 last_perm = lh;
1208 }
1209
1210 /* Add the new entry after the last permanent entry if any, so that
1211 * the new entry does not override a permanent entry when matched with
1212 * a wild-card protocol. But it is allowed to override any existing
1213 * non-permanent entry. This means that when we remove this entry, the
1214 * system automatically returns to the old behavior.
1215 */
1216 list_add_rcu(new: &p->list, head: last_perm);
1217out:
1218 spin_unlock_bh(lock: &inetsw_lock);
1219
1220 return;
1221
1222out_permanent:
1223 pr_err("Attempt to override permanent protocol %d\n", protocol);
1224 goto out;
1225
1226out_illegal:
1227 pr_err("Ignoring attempt to register invalid socket type %d\n",
1228 p->type);
1229 goto out;
1230}
1231EXPORT_SYMBOL(inet_register_protosw);
1232
1233void inet_unregister_protosw(struct inet_protosw *p)
1234{
1235 if (INET_PROTOSW_PERMANENT & p->flags) {
1236 pr_err("Attempt to unregister permanent protocol %d\n",
1237 p->protocol);
1238 } else {
1239 spin_lock_bh(lock: &inetsw_lock);
1240 list_del_rcu(entry: &p->list);
1241 spin_unlock_bh(lock: &inetsw_lock);
1242
1243 synchronize_net();
1244 }
1245}
1246EXPORT_SYMBOL(inet_unregister_protosw);
1247
1248static int inet_sk_reselect_saddr(struct sock *sk)
1249{
1250 struct inet_sock *inet = inet_sk(sk);
1251 __be32 old_saddr = inet->inet_saddr;
1252 __be32 daddr = inet->inet_daddr;
1253 struct flowi4 *fl4;
1254 struct rtable *rt;
1255 __be32 new_saddr;
1256 struct ip_options_rcu *inet_opt;
1257 int err;
1258
1259 inet_opt = rcu_dereference_protected(inet->inet_opt,
1260 lockdep_sock_is_held(sk));
1261 if (inet_opt && inet_opt->opt.srr)
1262 daddr = inet_opt->opt.faddr;
1263
1264 /* Query new route. */
1265 fl4 = &inet->cork.fl.u.ip4;
1266 rt = ip_route_connect(fl4, dst: daddr, src: 0, oif: sk->sk_bound_dev_if,
1267 protocol: sk->sk_protocol, sport: inet->inet_sport,
1268 dport: inet->inet_dport, sk);
1269 if (IS_ERR(ptr: rt))
1270 return PTR_ERR(ptr: rt);
1271
1272 new_saddr = fl4->saddr;
1273
1274 if (new_saddr == old_saddr) {
1275 sk_setup_caps(sk, dst: &rt->dst);
1276 return 0;
1277 }
1278
1279 err = inet_bhash2_update_saddr(sk, saddr: &new_saddr, AF_INET);
1280 if (err) {
1281 ip_rt_put(rt);
1282 return err;
1283 }
1284
1285 sk_setup_caps(sk, dst: &rt->dst);
1286
1287 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1288 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1289 __func__, &old_saddr, &new_saddr);
1290 }
1291
1292 /*
1293 * XXX The only one ugly spot where we need to
1294 * XXX really change the sockets identity after
1295 * XXX it has entered the hashes. -DaveM
1296 *
1297 * Besides that, it does not check for connection
1298 * uniqueness. Wait for troubles.
1299 */
1300 return __sk_prot_rehash(sk);
1301}
1302
1303int inet_sk_rebuild_header(struct sock *sk)
1304{
1305 struct inet_sock *inet = inet_sk(sk);
1306 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, cookie: 0);
1307 __be32 daddr;
1308 struct ip_options_rcu *inet_opt;
1309 struct flowi4 *fl4;
1310 int err;
1311
1312 /* Route is OK, nothing to do. */
1313 if (rt)
1314 return 0;
1315
1316 /* Reroute. */
1317 rcu_read_lock();
1318 inet_opt = rcu_dereference(inet->inet_opt);
1319 daddr = inet->inet_daddr;
1320 if (inet_opt && inet_opt->opt.srr)
1321 daddr = inet_opt->opt.faddr;
1322 rcu_read_unlock();
1323 fl4 = &inet->cork.fl.u.ip4;
1324 rt = ip_route_output_ports(net: sock_net(sk), fl4, sk, daddr, saddr: inet->inet_saddr,
1325 dport: inet->inet_dport, sport: inet->inet_sport,
1326 proto: sk->sk_protocol, RT_CONN_FLAGS(sk),
1327 oif: sk->sk_bound_dev_if);
1328 if (!IS_ERR(ptr: rt)) {
1329 err = 0;
1330 sk_setup_caps(sk, dst: &rt->dst);
1331 } else {
1332 err = PTR_ERR(ptr: rt);
1333
1334 /* Routing failed... */
1335 sk->sk_route_caps = 0;
1336 /*
1337 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1338 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1339 */
1340 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1341 sk->sk_state != TCP_SYN_SENT ||
1342 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1343 (err = inet_sk_reselect_saddr(sk)) != 0)
1344 WRITE_ONCE(sk->sk_err_soft, -err);
1345 }
1346
1347 return err;
1348}
1349EXPORT_SYMBOL(inet_sk_rebuild_header);
1350
1351void inet_sk_set_state(struct sock *sk, int state)
1352{
1353 trace_inet_sock_set_state(sk, oldstate: sk->sk_state, newstate: state);
1354 sk->sk_state = state;
1355}
1356EXPORT_SYMBOL(inet_sk_set_state);
1357
1358void inet_sk_state_store(struct sock *sk, int newstate)
1359{
1360 trace_inet_sock_set_state(sk, oldstate: sk->sk_state, newstate);
1361 smp_store_release(&sk->sk_state, newstate);
1362}
1363
1364struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1365 netdev_features_t features)
1366{
1367 bool udpfrag = false, fixedid = false, gso_partial, encap;
1368 struct sk_buff *segs = ERR_PTR(error: -EINVAL);
1369 const struct net_offload *ops;
1370 unsigned int offset = 0;
1371 struct iphdr *iph;
1372 int proto, tot_len;
1373 int nhoff;
1374 int ihl;
1375 int id;
1376
1377 skb_reset_network_header(skb);
1378 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1379 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1380 goto out;
1381
1382 iph = ip_hdr(skb);
1383 ihl = iph->ihl * 4;
1384 if (ihl < sizeof(*iph))
1385 goto out;
1386
1387 id = ntohs(iph->id);
1388 proto = iph->protocol;
1389
1390 /* Warning: after this point, iph might be no longer valid */
1391 if (unlikely(!pskb_may_pull(skb, ihl)))
1392 goto out;
1393 __skb_pull(skb, len: ihl);
1394
1395 encap = SKB_GSO_CB(skb)->encap_level > 0;
1396 if (encap)
1397 features &= skb->dev->hw_enc_features;
1398 SKB_GSO_CB(skb)->encap_level += ihl;
1399
1400 skb_reset_transport_header(skb);
1401
1402 segs = ERR_PTR(error: -EPROTONOSUPPORT);
1403
1404 if (!skb->encapsulation || encap) {
1405 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1406 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1407
1408 /* fixed ID is invalid if DF bit is not set */
1409 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1410 goto out;
1411 }
1412
1413 ops = rcu_dereference(inet_offloads[proto]);
1414 if (likely(ops && ops->callbacks.gso_segment)) {
1415 segs = ops->callbacks.gso_segment(skb, features);
1416 if (!segs)
1417 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1418 }
1419
1420 if (IS_ERR_OR_NULL(ptr: segs))
1421 goto out;
1422
1423 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1424
1425 skb = segs;
1426 do {
1427 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1428 if (udpfrag) {
1429 iph->frag_off = htons(offset >> 3);
1430 if (skb->next)
1431 iph->frag_off |= htons(IP_MF);
1432 offset += skb->len - nhoff - ihl;
1433 tot_len = skb->len - nhoff;
1434 } else if (skb_is_gso(skb)) {
1435 if (!fixedid) {
1436 iph->id = htons(id);
1437 id += skb_shinfo(skb)->gso_segs;
1438 }
1439
1440 if (gso_partial)
1441 tot_len = skb_shinfo(skb)->gso_size +
1442 SKB_GSO_CB(skb)->data_offset +
1443 skb->head - (unsigned char *)iph;
1444 else
1445 tot_len = skb->len - nhoff;
1446 } else {
1447 if (!fixedid)
1448 iph->id = htons(id++);
1449 tot_len = skb->len - nhoff;
1450 }
1451 iph->tot_len = htons(tot_len);
1452 ip_send_check(ip: iph);
1453 if (encap)
1454 skb_reset_inner_headers(skb);
1455 skb->network_header = (u8 *)iph - skb->head;
1456 skb_reset_mac_len(skb);
1457 } while ((skb = skb->next));
1458
1459out:
1460 return segs;
1461}
1462
1463static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1464 netdev_features_t features)
1465{
1466 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1467 return ERR_PTR(error: -EINVAL);
1468
1469 return inet_gso_segment(skb, features);
1470}
1471
1472struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1473{
1474 const struct net_offload *ops;
1475 struct sk_buff *pp = NULL;
1476 const struct iphdr *iph;
1477 struct sk_buff *p;
1478 unsigned int hlen;
1479 unsigned int off;
1480 unsigned int id;
1481 int flush = 1;
1482 int proto;
1483
1484 off = skb_gro_offset(skb);
1485 hlen = off + sizeof(*iph);
1486 iph = skb_gro_header(skb, hlen, offset: off);
1487 if (unlikely(!iph))
1488 goto out;
1489
1490 proto = iph->protocol;
1491
1492 ops = rcu_dereference(inet_offloads[proto]);
1493 if (!ops || !ops->callbacks.gro_receive)
1494 goto out;
1495
1496 if (*(u8 *)iph != 0x45)
1497 goto out;
1498
1499 if (ip_is_fragment(iph))
1500 goto out;
1501
1502 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1503 goto out;
1504
1505 NAPI_GRO_CB(skb)->proto = proto;
1506 id = ntohl(*(__be32 *)&iph->id);
1507 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1508 id >>= 16;
1509
1510 list_for_each_entry(p, head, list) {
1511 struct iphdr *iph2;
1512 u16 flush_id;
1513
1514 if (!NAPI_GRO_CB(p)->same_flow)
1515 continue;
1516
1517 iph2 = (struct iphdr *)(p->data + off);
1518 /* The above works because, with the exception of the top
1519 * (inner most) layer, we only aggregate pkts with the same
1520 * hdr length so all the hdrs we'll need to verify will start
1521 * at the same offset.
1522 */
1523 if ((iph->protocol ^ iph2->protocol) |
1524 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1525 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1526 NAPI_GRO_CB(p)->same_flow = 0;
1527 continue;
1528 }
1529
1530 /* All fields must match except length and checksum. */
1531 NAPI_GRO_CB(p)->flush |=
1532 (iph->ttl ^ iph2->ttl) |
1533 (iph->tos ^ iph2->tos) |
1534 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1535
1536 NAPI_GRO_CB(p)->flush |= flush;
1537
1538 /* We need to store of the IP ID check to be included later
1539 * when we can verify that this packet does in fact belong
1540 * to a given flow.
1541 */
1542 flush_id = (u16)(id - ntohs(iph2->id));
1543
1544 /* This bit of code makes it much easier for us to identify
1545 * the cases where we are doing atomic vs non-atomic IP ID
1546 * checks. Specifically an atomic check can return IP ID
1547 * values 0 - 0xFFFF, while a non-atomic check can only
1548 * return 0 or 0xFFFF.
1549 */
1550 if (!NAPI_GRO_CB(p)->is_atomic ||
1551 !(iph->frag_off & htons(IP_DF))) {
1552 flush_id ^= NAPI_GRO_CB(p)->count;
1553 flush_id = flush_id ? 0xFFFF : 0;
1554 }
1555
1556 /* If the previous IP ID value was based on an atomic
1557 * datagram we can overwrite the value and ignore it.
1558 */
1559 if (NAPI_GRO_CB(skb)->is_atomic)
1560 NAPI_GRO_CB(p)->flush_id = flush_id;
1561 else
1562 NAPI_GRO_CB(p)->flush_id |= flush_id;
1563 }
1564
1565 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1566 NAPI_GRO_CB(skb)->flush |= flush;
1567 skb_set_network_header(skb, offset: off);
1568 /* The above will be needed by the transport layer if there is one
1569 * immediately following this IP hdr.
1570 */
1571
1572 /* Note : No need to call skb_gro_postpull_rcsum() here,
1573 * as we already checked checksum over ipv4 header was 0
1574 */
1575 skb_gro_pull(skb, len: sizeof(*iph));
1576 skb_set_transport_header(skb, offset: skb_gro_offset(skb));
1577
1578 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1579 ops->callbacks.gro_receive, head, skb);
1580
1581out:
1582 skb_gro_flush_final(skb, pp, flush);
1583
1584 return pp;
1585}
1586
1587static struct sk_buff *ipip_gro_receive(struct list_head *head,
1588 struct sk_buff *skb)
1589{
1590 if (NAPI_GRO_CB(skb)->encap_mark) {
1591 NAPI_GRO_CB(skb)->flush = 1;
1592 return NULL;
1593 }
1594
1595 NAPI_GRO_CB(skb)->encap_mark = 1;
1596
1597 return inet_gro_receive(head, skb);
1598}
1599
1600#define SECONDS_PER_DAY 86400
1601
1602/* inet_current_timestamp - Return IP network timestamp
1603 *
1604 * Return milliseconds since midnight in network byte order.
1605 */
1606__be32 inet_current_timestamp(void)
1607{
1608 u32 secs;
1609 u32 msecs;
1610 struct timespec64 ts;
1611
1612 ktime_get_real_ts64(tv: &ts);
1613
1614 /* Get secs since midnight. */
1615 (void)div_u64_rem(dividend: ts.tv_sec, SECONDS_PER_DAY, remainder: &secs);
1616 /* Convert to msecs. */
1617 msecs = secs * MSEC_PER_SEC;
1618 /* Convert nsec to msec. */
1619 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1620
1621 /* Convert to network byte order. */
1622 return htonl(msecs);
1623}
1624EXPORT_SYMBOL(inet_current_timestamp);
1625
1626int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1627{
1628 if (sk->sk_family == AF_INET)
1629 return ip_recv_error(sk, msg, len, addr_len);
1630#if IS_ENABLED(CONFIG_IPV6)
1631 if (sk->sk_family == AF_INET6)
1632 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1633#endif
1634 return -EINVAL;
1635}
1636
1637int inet_gro_complete(struct sk_buff *skb, int nhoff)
1638{
1639 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1640 const struct net_offload *ops;
1641 __be16 totlen = iph->tot_len;
1642 int proto = iph->protocol;
1643 int err = -ENOSYS;
1644
1645 if (skb->encapsulation) {
1646 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1647 skb_set_inner_network_header(skb, offset: nhoff);
1648 }
1649
1650 iph_set_totlen(iph, len: skb->len - nhoff);
1651 csum_replace2(sum: &iph->check, old: totlen, new: iph->tot_len);
1652
1653 ops = rcu_dereference(inet_offloads[proto]);
1654 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1655 goto out;
1656
1657 /* Only need to add sizeof(*iph) to get to the next hdr below
1658 * because any hdr with option will have been flushed in
1659 * inet_gro_receive().
1660 */
1661 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1662 tcp4_gro_complete, udp4_gro_complete,
1663 skb, nhoff + sizeof(*iph));
1664
1665out:
1666 return err;
1667}
1668
1669static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1670{
1671 skb->encapsulation = 1;
1672 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1673 return inet_gro_complete(skb, nhoff);
1674}
1675
1676int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1677 unsigned short type, unsigned char protocol,
1678 struct net *net)
1679{
1680 struct socket *sock;
1681 int rc = sock_create_kern(net, family, type, proto: protocol, res: &sock);
1682
1683 if (rc == 0) {
1684 *sk = sock->sk;
1685 (*sk)->sk_allocation = GFP_ATOMIC;
1686 (*sk)->sk_use_task_frag = false;
1687 /*
1688 * Unhash it so that IP input processing does not even see it,
1689 * we do not wish this socket to see incoming packets.
1690 */
1691 (*sk)->sk_prot->unhash(*sk);
1692 }
1693 return rc;
1694}
1695EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1696
1697unsigned long snmp_fold_field(void __percpu *mib, int offt)
1698{
1699 unsigned long res = 0;
1700 int i;
1701
1702 for_each_possible_cpu(i)
1703 res += snmp_get_cpu_field(mib, cpu: i, offt);
1704 return res;
1705}
1706EXPORT_SYMBOL_GPL(snmp_fold_field);
1707
1708#if BITS_PER_LONG==32
1709
1710u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1711 size_t syncp_offset)
1712{
1713 void *bhptr;
1714 struct u64_stats_sync *syncp;
1715 u64 v;
1716 unsigned int start;
1717
1718 bhptr = per_cpu_ptr(mib, cpu);
1719 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1720 do {
1721 start = u64_stats_fetch_begin(syncp);
1722 v = *(((u64 *)bhptr) + offt);
1723 } while (u64_stats_fetch_retry(syncp, start));
1724
1725 return v;
1726}
1727EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1728
1729u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1730{
1731 u64 res = 0;
1732 int cpu;
1733
1734 for_each_possible_cpu(cpu) {
1735 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1736 }
1737 return res;
1738}
1739EXPORT_SYMBOL_GPL(snmp_fold_field64);
1740#endif
1741
1742#ifdef CONFIG_IP_MULTICAST
1743static const struct net_protocol igmp_protocol = {
1744 .handler = igmp_rcv,
1745};
1746#endif
1747
1748static const struct net_protocol tcp_protocol = {
1749 .handler = tcp_v4_rcv,
1750 .err_handler = tcp_v4_err,
1751 .no_policy = 1,
1752 .icmp_strict_tag_validation = 1,
1753};
1754
1755static const struct net_protocol udp_protocol = {
1756 .handler = udp_rcv,
1757 .err_handler = udp_err,
1758 .no_policy = 1,
1759};
1760
1761static const struct net_protocol icmp_protocol = {
1762 .handler = icmp_rcv,
1763 .err_handler = icmp_err,
1764 .no_policy = 1,
1765};
1766
1767static __net_init int ipv4_mib_init_net(struct net *net)
1768{
1769 int i;
1770
1771 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1772 if (!net->mib.tcp_statistics)
1773 goto err_tcp_mib;
1774 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1775 if (!net->mib.ip_statistics)
1776 goto err_ip_mib;
1777
1778 for_each_possible_cpu(i) {
1779 struct ipstats_mib *af_inet_stats;
1780 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1781 u64_stats_init(syncp: &af_inet_stats->syncp);
1782 }
1783
1784 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1785 if (!net->mib.net_statistics)
1786 goto err_net_mib;
1787 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1788 if (!net->mib.udp_statistics)
1789 goto err_udp_mib;
1790 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1791 if (!net->mib.udplite_statistics)
1792 goto err_udplite_mib;
1793 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1794 if (!net->mib.icmp_statistics)
1795 goto err_icmp_mib;
1796 net->mib.icmpmsg_statistics = kzalloc(size: sizeof(struct icmpmsg_mib),
1797 GFP_KERNEL);
1798 if (!net->mib.icmpmsg_statistics)
1799 goto err_icmpmsg_mib;
1800
1801 tcp_mib_init(net);
1802 return 0;
1803
1804err_icmpmsg_mib:
1805 free_percpu(pdata: net->mib.icmp_statistics);
1806err_icmp_mib:
1807 free_percpu(pdata: net->mib.udplite_statistics);
1808err_udplite_mib:
1809 free_percpu(pdata: net->mib.udp_statistics);
1810err_udp_mib:
1811 free_percpu(pdata: net->mib.net_statistics);
1812err_net_mib:
1813 free_percpu(pdata: net->mib.ip_statistics);
1814err_ip_mib:
1815 free_percpu(pdata: net->mib.tcp_statistics);
1816err_tcp_mib:
1817 return -ENOMEM;
1818}
1819
1820static __net_exit void ipv4_mib_exit_net(struct net *net)
1821{
1822 kfree(objp: net->mib.icmpmsg_statistics);
1823 free_percpu(pdata: net->mib.icmp_statistics);
1824 free_percpu(pdata: net->mib.udplite_statistics);
1825 free_percpu(pdata: net->mib.udp_statistics);
1826 free_percpu(pdata: net->mib.net_statistics);
1827 free_percpu(pdata: net->mib.ip_statistics);
1828 free_percpu(pdata: net->mib.tcp_statistics);
1829#ifdef CONFIG_MPTCP
1830 /* allocated on demand, see mptcp_init_sock() */
1831 free_percpu(pdata: net->mib.mptcp_statistics);
1832#endif
1833}
1834
1835static __net_initdata struct pernet_operations ipv4_mib_ops = {
1836 .init = ipv4_mib_init_net,
1837 .exit = ipv4_mib_exit_net,
1838};
1839
1840static int __init init_ipv4_mibs(void)
1841{
1842 return register_pernet_subsys(&ipv4_mib_ops);
1843}
1844
1845static __net_init int inet_init_net(struct net *net)
1846{
1847 /*
1848 * Set defaults for local port range
1849 */
1850 seqlock_init(&net->ipv4.ip_local_ports.lock);
1851 net->ipv4.ip_local_ports.range[0] = 32768;
1852 net->ipv4.ip_local_ports.range[1] = 60999;
1853
1854 seqlock_init(&net->ipv4.ping_group_range.lock);
1855 /*
1856 * Sane defaults - nobody may create ping sockets.
1857 * Boot scripts should set this to distro-specific group.
1858 */
1859 net->ipv4.ping_group_range.range[0] = make_kgid(from: &init_user_ns, gid: 1);
1860 net->ipv4.ping_group_range.range[1] = make_kgid(from: &init_user_ns, gid: 0);
1861
1862 /* Default values for sysctl-controlled parameters.
1863 * We set them here, in case sysctl is not compiled.
1864 */
1865 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1866 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1867 net->ipv4.sysctl_ip_dynaddr = 0;
1868 net->ipv4.sysctl_ip_early_demux = 1;
1869 net->ipv4.sysctl_udp_early_demux = 1;
1870 net->ipv4.sysctl_tcp_early_demux = 1;
1871 net->ipv4.sysctl_nexthop_compat_mode = 1;
1872#ifdef CONFIG_SYSCTL
1873 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1874#endif
1875
1876 /* Some igmp sysctl, whose values are always used */
1877 net->ipv4.sysctl_igmp_max_memberships = 20;
1878 net->ipv4.sysctl_igmp_max_msf = 10;
1879 /* IGMP reports for link-local multicast groups are enabled by default */
1880 net->ipv4.sysctl_igmp_llm_reports = 1;
1881 net->ipv4.sysctl_igmp_qrv = 2;
1882
1883 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1884
1885 return 0;
1886}
1887
1888static __net_initdata struct pernet_operations af_inet_ops = {
1889 .init = inet_init_net,
1890};
1891
1892static int __init init_inet_pernet_ops(void)
1893{
1894 return register_pernet_subsys(&af_inet_ops);
1895}
1896
1897static int ipv4_proc_init(void);
1898
1899/*
1900 * IP protocol layer initialiser
1901 */
1902
1903static struct packet_offload ip_packet_offload __read_mostly = {
1904 .type = cpu_to_be16(ETH_P_IP),
1905 .callbacks = {
1906 .gso_segment = inet_gso_segment,
1907 .gro_receive = inet_gro_receive,
1908 .gro_complete = inet_gro_complete,
1909 },
1910};
1911
1912static const struct net_offload ipip_offload = {
1913 .callbacks = {
1914 .gso_segment = ipip_gso_segment,
1915 .gro_receive = ipip_gro_receive,
1916 .gro_complete = ipip_gro_complete,
1917 },
1918};
1919
1920static int __init ipip_offload_init(void)
1921{
1922 return inet_add_offload(prot: &ipip_offload, IPPROTO_IPIP);
1923}
1924
1925static int __init ipv4_offload_init(void)
1926{
1927 /*
1928 * Add offloads
1929 */
1930 if (udpv4_offload_init() < 0)
1931 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1932 if (tcpv4_offload_init() < 0)
1933 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1934 if (ipip_offload_init() < 0)
1935 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1936
1937 dev_add_offload(po: &ip_packet_offload);
1938 return 0;
1939}
1940
1941fs_initcall(ipv4_offload_init);
1942
1943static struct packet_type ip_packet_type __read_mostly = {
1944 .type = cpu_to_be16(ETH_P_IP),
1945 .func = ip_rcv,
1946 .list_func = ip_list_rcv,
1947};
1948
1949static int __init inet_init(void)
1950{
1951 struct inet_protosw *q;
1952 struct list_head *r;
1953 int rc;
1954
1955 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1956
1957 raw_hashinfo_init(hashinfo: &raw_v4_hashinfo);
1958
1959 rc = proto_register(prot: &tcp_prot, alloc_slab: 1);
1960 if (rc)
1961 goto out;
1962
1963 rc = proto_register(prot: &udp_prot, alloc_slab: 1);
1964 if (rc)
1965 goto out_unregister_tcp_proto;
1966
1967 rc = proto_register(prot: &raw_prot, alloc_slab: 1);
1968 if (rc)
1969 goto out_unregister_udp_proto;
1970
1971 rc = proto_register(prot: &ping_prot, alloc_slab: 1);
1972 if (rc)
1973 goto out_unregister_raw_proto;
1974
1975 /*
1976 * Tell SOCKET that we are alive...
1977 */
1978
1979 (void)sock_register(fam: &inet_family_ops);
1980
1981#ifdef CONFIG_SYSCTL
1982 ip_static_sysctl_init();
1983#endif
1984
1985 /*
1986 * Add all the base protocols.
1987 */
1988
1989 if (inet_add_protocol(prot: &icmp_protocol, IPPROTO_ICMP) < 0)
1990 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1991 if (inet_add_protocol(prot: &udp_protocol, IPPROTO_UDP) < 0)
1992 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1993 if (inet_add_protocol(prot: &tcp_protocol, IPPROTO_TCP) < 0)
1994 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1995#ifdef CONFIG_IP_MULTICAST
1996 if (inet_add_protocol(prot: &igmp_protocol, IPPROTO_IGMP) < 0)
1997 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1998#endif
1999
2000 /* Register the socket-side information for inet_create. */
2001 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2002 INIT_LIST_HEAD(list: r);
2003
2004 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2005 inet_register_protosw(q);
2006
2007 /*
2008 * Set the ARP module up
2009 */
2010
2011 arp_init();
2012
2013 /*
2014 * Set the IP module up
2015 */
2016
2017 ip_init();
2018
2019 /* Initialise per-cpu ipv4 mibs */
2020 if (init_ipv4_mibs())
2021 panic(fmt: "%s: Cannot init ipv4 mibs\n", __func__);
2022
2023 /* Setup TCP slab cache for open requests. */
2024 tcp_init();
2025
2026 /* Setup UDP memory threshold */
2027 udp_init();
2028
2029 /* Add UDP-Lite (RFC 3828) */
2030 udplite4_register();
2031
2032 raw_init();
2033
2034 ping_init();
2035
2036 /*
2037 * Set the ICMP layer up
2038 */
2039
2040 if (icmp_init() < 0)
2041 panic(fmt: "Failed to create the ICMP control socket.\n");
2042
2043 /*
2044 * Initialise the multicast router
2045 */
2046#if defined(CONFIG_IP_MROUTE)
2047 if (ip_mr_init())
2048 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2049#endif
2050
2051 if (init_inet_pernet_ops())
2052 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2053
2054 ipv4_proc_init();
2055
2056 ipfrag_init();
2057
2058 dev_add_pack(pt: &ip_packet_type);
2059
2060 ip_tunnel_core_init();
2061
2062 rc = 0;
2063out:
2064 return rc;
2065out_unregister_raw_proto:
2066 proto_unregister(prot: &raw_prot);
2067out_unregister_udp_proto:
2068 proto_unregister(prot: &udp_prot);
2069out_unregister_tcp_proto:
2070 proto_unregister(prot: &tcp_prot);
2071 goto out;
2072}
2073
2074fs_initcall(inet_init);
2075
2076/* ------------------------------------------------------------------------ */
2077
2078#ifdef CONFIG_PROC_FS
2079static int __init ipv4_proc_init(void)
2080{
2081 int rc = 0;
2082
2083 if (raw_proc_init())
2084 goto out_raw;
2085 if (tcp4_proc_init())
2086 goto out_tcp;
2087 if (udp4_proc_init())
2088 goto out_udp;
2089 if (ping_proc_init())
2090 goto out_ping;
2091 if (ip_misc_proc_init())
2092 goto out_misc;
2093out:
2094 return rc;
2095out_misc:
2096 ping_proc_exit();
2097out_ping:
2098 udp4_proc_exit();
2099out_udp:
2100 tcp4_proc_exit();
2101out_tcp:
2102 raw_proc_exit();
2103out_raw:
2104 rc = -ENOMEM;
2105 goto out;
2106}
2107
2108#else /* CONFIG_PROC_FS */
2109static int __init ipv4_proc_init(void)
2110{
2111 return 0;
2112}
2113#endif /* CONFIG_PROC_FS */
2114

source code of linux/net/ipv4/af_inet.c