Loading...
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/tcp.h>
104#include <net/udp.h>
105#include <net/udplite.h>
106#include <net/ping.h>
107#include <linux/skbuff.h>
108#include <net/sock.h>
109#include <net/raw.h>
110#include <net/icmp.h>
111#include <net/inet_common.h>
112#include <net/ip_tunnels.h>
113#include <net/xfrm.h>
114#include <net/net_namespace.h>
115#include <net/secure_seq.h>
116#ifdef CONFIG_IP_MROUTE
117#include <linux/mroute.h>
118#endif
119#include <net/l3mdev.h>
120#include <net/compat.h>
121
122#include <trace/events/sock.h>
123
124/* The inetsw table contains everything that inet_create needs to
125 * build a new socket.
126 */
127static struct list_head inetsw[SOCK_MAX];
128static DEFINE_SPINLOCK(inetsw_lock);
129
130/* New destruction routine */
131
132void inet_sock_destruct(struct sock *sk)
133{
134 struct inet_sock *inet = inet_sk(sk);
135
136 __skb_queue_purge(&sk->sk_receive_queue);
137 __skb_queue_purge(&sk->sk_error_queue);
138
139 sk_mem_reclaim_final(sk);
140
141 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
142 pr_err("Attempt to release TCP socket in state %d %p\n",
143 sk->sk_state, sk);
144 return;
145 }
146 if (!sock_flag(sk, SOCK_DEAD)) {
147 pr_err("Attempt to release alive inet socket %p\n", sk);
148 return;
149 }
150
151 WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
152 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
153 WARN_ON_ONCE(sk->sk_wmem_queued);
154 WARN_ON_ONCE(sk_forward_alloc_get(sk));
155
156 kfree(rcu_dereference_protected(inet->inet_opt, 1));
157 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
158 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
159 sk_refcnt_debug_dec(sk);
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
190/*
191 * Move a socket into listening state.
192 */
193int inet_listen(struct socket *sock, int backlog)
194{
195 struct sock *sk = sock->sk;
196 unsigned char old_state;
197 int err, tcp_fastopen;
198
199 lock_sock(sk);
200
201 err = -EINVAL;
202 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
203 goto out;
204
205 old_state = sk->sk_state;
206 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
207 goto out;
208
209 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
212 */
213 if (old_state != TCP_LISTEN) {
214 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
215 * Note that only TCP sockets (SOCK_STREAM) will reach here.
216 * Also fastopen backlog may already been set via the option
217 * because the socket was in TCP_LISTEN state previously but
218 * was shutdown() rather than close().
219 */
220 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
221 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
222 (tcp_fastopen & TFO_SERVER_ENABLE) &&
223 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
224 fastopen_queue_tune(sk, backlog);
225 tcp_fastopen_init_key_once(sock_net(sk));
226 }
227
228 err = inet_csk_listen_start(sk);
229 if (err)
230 goto out;
231 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
232 }
233 err = 0;
234
235out:
236 release_sock(sk);
237 return err;
238}
239EXPORT_SYMBOL(inet_listen);
240
241/*
242 * Create an inet socket.
243 */
244
245static int inet_create(struct net *net, struct socket *sock, int protocol,
246 int kern)
247{
248 struct sock *sk;
249 struct inet_protosw *answer;
250 struct inet_sock *inet;
251 struct proto *answer_prot;
252 unsigned char answer_flags;
253 int try_loading_module = 0;
254 int err;
255
256 if (protocol < 0 || protocol >= IPPROTO_MAX)
257 return -EINVAL;
258
259 sock->state = SS_UNCONNECTED;
260
261 /* Look for the requested type/protocol pair. */
262lookup_protocol:
263 err = -ESOCKTNOSUPPORT;
264 rcu_read_lock();
265 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
266
267 err = 0;
268 /* Check the non-wild match. */
269 if (protocol == answer->protocol) {
270 if (protocol != IPPROTO_IP)
271 break;
272 } else {
273 /* Check for the two wild cases. */
274 if (IPPROTO_IP == protocol) {
275 protocol = answer->protocol;
276 break;
277 }
278 if (IPPROTO_IP == answer->protocol)
279 break;
280 }
281 err = -EPROTONOSUPPORT;
282 }
283
284 if (unlikely(err)) {
285 if (try_loading_module < 2) {
286 rcu_read_unlock();
287 /*
288 * Be more specific, e.g. net-pf-2-proto-132-type-1
289 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
290 */
291 if (++try_loading_module == 1)
292 request_module("net-pf-%d-proto-%d-type-%d",
293 PF_INET, protocol, sock->type);
294 /*
295 * Fall back to generic, e.g. net-pf-2-proto-132
296 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
297 */
298 else
299 request_module("net-pf-%d-proto-%d",
300 PF_INET, protocol);
301 goto lookup_protocol;
302 } else
303 goto out_rcu_unlock;
304 }
305
306 err = -EPERM;
307 if (sock->type == SOCK_RAW && !kern &&
308 !ns_capable(net->user_ns, CAP_NET_RAW))
309 goto out_rcu_unlock;
310
311 sock->ops = answer->ops;
312 answer_prot = answer->prot;
313 answer_flags = answer->flags;
314 rcu_read_unlock();
315
316 WARN_ON(!answer_prot->slab);
317
318 err = -ENOMEM;
319 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
320 if (!sk)
321 goto out;
322
323 err = 0;
324 if (INET_PROTOSW_REUSE & answer_flags)
325 sk->sk_reuse = SK_CAN_REUSE;
326
327 inet = inet_sk(sk);
328 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
329
330 inet->nodefrag = 0;
331
332 if (SOCK_RAW == sock->type) {
333 inet->inet_num = protocol;
334 if (IPPROTO_RAW == protocol)
335 inet->hdrincl = 1;
336 }
337
338 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
339 inet->pmtudisc = IP_PMTUDISC_DONT;
340 else
341 inet->pmtudisc = IP_PMTUDISC_WANT;
342
343 inet->inet_id = 0;
344
345 sock_init_data(sock, sk);
346
347 sk->sk_destruct = inet_sock_destruct;
348 sk->sk_protocol = protocol;
349 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
350 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
351
352 inet->uc_ttl = -1;
353 inet->mc_loop = 1;
354 inet->mc_ttl = 1;
355 inet->mc_all = 1;
356 inet->mc_index = 0;
357 inet->mc_list = NULL;
358 inet->rcv_tos = 0;
359
360 sk_refcnt_debug_inc(sk);
361
362 if (inet->inet_num) {
363 /* It assumes that any protocol which allows
364 * the user to assign a number at socket
365 * creation time automatically
366 * shares.
367 */
368 inet->inet_sport = htons(inet->inet_num);
369 /* Add to protocol hash chains. */
370 err = sk->sk_prot->hash(sk);
371 if (err) {
372 sk_common_release(sk);
373 goto out;
374 }
375 }
376
377 if (sk->sk_prot->init) {
378 err = sk->sk_prot->init(sk);
379 if (err) {
380 sk_common_release(sk);
381 goto out;
382 }
383 }
384
385 if (!kern) {
386 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
387 if (err) {
388 sk_common_release(sk);
389 goto out;
390 }
391 }
392out:
393 return err;
394out_rcu_unlock:
395 rcu_read_unlock();
396 goto out;
397}
398
399
400/*
401 * The peer socket should always be NULL (or else). When we call this
402 * function we are destroying the object and from then on nobody
403 * should refer to it.
404 */
405int inet_release(struct socket *sock)
406{
407 struct sock *sk = sock->sk;
408
409 if (sk) {
410 long timeout;
411
412 if (!sk->sk_kern_sock)
413 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
414
415 /* Applications forget to leave groups before exiting */
416 ip_mc_drop_socket(sk);
417
418 /* If linger is set, we don't return until the close
419 * is complete. Otherwise we return immediately. The
420 * actually closing is done the same either way.
421 *
422 * If the close is due to the process exiting, we never
423 * linger..
424 */
425 timeout = 0;
426 if (sock_flag(sk, SOCK_LINGER) &&
427 !(current->flags & PF_EXITING))
428 timeout = sk->sk_lingertime;
429 sk->sk_prot->close(sk, timeout);
430 sock->sk = NULL;
431 }
432 return 0;
433}
434EXPORT_SYMBOL(inet_release);
435
436int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
437{
438 struct sock *sk = sock->sk;
439 u32 flags = BIND_WITH_LOCK;
440 int err;
441
442 /* If the socket has its own bind function then use it. (RAW) */
443 if (sk->sk_prot->bind) {
444 return sk->sk_prot->bind(sk, uaddr, addr_len);
445 }
446 if (addr_len < sizeof(struct sockaddr_in))
447 return -EINVAL;
448
449 /* BPF prog is run before any checks are done so that if the prog
450 * changes context in a wrong way it will be caught.
451 */
452 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
453 CGROUP_INET4_BIND, &flags);
454 if (err)
455 return err;
456
457 return __inet_bind(sk, uaddr, addr_len, flags);
458}
459EXPORT_SYMBOL(inet_bind);
460
461int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
462 u32 flags)
463{
464 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
465 struct inet_sock *inet = inet_sk(sk);
466 struct net *net = sock_net(sk);
467 unsigned short snum;
468 int chk_addr_ret;
469 u32 tb_id = RT_TABLE_LOCAL;
470 int err;
471
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.
475 */
476 err = -EAFNOSUPPORT;
477 if (addr->sin_family != AF_UNSPEC ||
478 addr->sin_addr.s_addr != htonl(INADDR_ANY))
479 goto out;
480 }
481
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);
484
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)
491 */
492 err = -EADDRNOTAVAIL;
493 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
494 chk_addr_ret))
495 goto out;
496
497 snum = ntohs(addr->sin_port);
498 err = -EACCES;
499 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
500 snum && inet_port_requires_bind_service(net, snum) &&
501 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
502 goto out;
503
504 /* We keep a pair of addresses. rcv_saddr is the one
505 * used by hash lookups, and saddr is used for transmit.
506 *
507 * In the BSD API these are the same except where it
508 * would be illegal to use them (multicast/broadcast) in
509 * which case the sending device address is used.
510 */
511 if (flags & BIND_WITH_LOCK)
512 lock_sock(sk);
513
514 /* Check these errors (active socket, double bind). */
515 err = -EINVAL;
516 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
517 goto out_release_sock;
518
519 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
520 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
521 inet->inet_saddr = 0; /* Use device */
522
523 /* Make sure we are allowed to bind here. */
524 if (snum || !(inet->bind_address_no_port ||
525 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
526 err = sk->sk_prot->get_port(sk, snum);
527 if (err) {
528 inet->inet_saddr = inet->inet_rcv_saddr = 0;
529 goto out_release_sock;
530 }
531 if (!(flags & BIND_FROM_BPF)) {
532 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
533 if (err) {
534 inet->inet_saddr = inet->inet_rcv_saddr = 0;
535 if (sk->sk_prot->put_port)
536 sk->sk_prot->put_port(sk);
537 goto out_release_sock;
538 }
539 }
540 }
541
542 if (inet->inet_rcv_saddr)
543 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
544 if (snum)
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;
549 sk_dst_reset(sk);
550 err = 0;
551out_release_sock:
552 if (flags & BIND_WITH_LOCK)
553 release_sock(sk);
554out:
555 return err;
556}
557
558int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
559 int addr_len, int flags)
560{
561 struct sock *sk = sock->sk;
562 const struct proto *prot;
563 int err;
564
565 if (addr_len < sizeof(uaddr->sa_family))
566 return -EINVAL;
567
568 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
569 prot = READ_ONCE(sk->sk_prot);
570
571 if (uaddr->sa_family == AF_UNSPEC)
572 return prot->disconnect(sk, flags);
573
574 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
575 err = prot->pre_connect(sk, uaddr, addr_len);
576 if (err)
577 return err;
578 }
579
580 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
581 return -EAGAIN;
582 return prot->connect(sk, uaddr, addr_len);
583}
584EXPORT_SYMBOL(inet_dgram_connect);
585
586static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
587{
588 DEFINE_WAIT_FUNC(wait, woken_wake_function);
589
590 add_wait_queue(sk_sleep(sk), &wait);
591 sk->sk_write_pending += writebias;
592
593 /* Basic assumption: if someone sets sk->sk_err, he _must_
594 * change state of the socket from TCP_SYN_*.
595 * Connect() does not allow to get error notifications
596 * without closing the socket.
597 */
598 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
599 release_sock(sk);
600 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
601 lock_sock(sk);
602 if (signal_pending(current) || !timeo)
603 break;
604 }
605 remove_wait_queue(sk_sleep(sk), &wait);
606 sk->sk_write_pending -= writebias;
607 return timeo;
608}
609
610/*
611 * Connect to a remote host. There is regrettably still a little
612 * TCP 'magic' in here.
613 */
614int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
615 int addr_len, int flags, int is_sendmsg)
616{
617 struct sock *sk = sock->sk;
618 int err;
619 long timeo;
620
621 /*
622 * uaddr can be NULL and addr_len can be 0 if:
623 * sk is a TCP fastopen active socket and
624 * TCP_FASTOPEN_CONNECT sockopt is set and
625 * we already have a valid cookie for this socket.
626 * In this case, user can call write() after connect().
627 * write() will invoke tcp_sendmsg_fastopen() which calls
628 * __inet_stream_connect().
629 */
630 if (uaddr) {
631 if (addr_len < sizeof(uaddr->sa_family))
632 return -EINVAL;
633
634 if (uaddr->sa_family == AF_UNSPEC) {
635 err = sk->sk_prot->disconnect(sk, flags);
636 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
637 goto out;
638 }
639 }
640
641 switch (sock->state) {
642 default:
643 err = -EINVAL;
644 goto out;
645 case SS_CONNECTED:
646 err = -EISCONN;
647 goto out;
648 case SS_CONNECTING:
649 if (inet_sk(sk)->defer_connect)
650 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
651 else
652 err = -EALREADY;
653 /* Fall out of switch with err, set for this state */
654 break;
655 case SS_UNCONNECTED:
656 err = -EISCONN;
657 if (sk->sk_state != TCP_CLOSE)
658 goto out;
659
660 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
661 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
662 if (err)
663 goto out;
664 }
665
666 err = sk->sk_prot->connect(sk, uaddr, addr_len);
667 if (err < 0)
668 goto out;
669
670 sock->state = SS_CONNECTING;
671
672 if (!err && inet_sk(sk)->defer_connect)
673 goto out;
674
675 /* Just entered SS_CONNECTING state; the only
676 * difference is that return value in non-blocking
677 * case is EINPROGRESS, rather than EALREADY.
678 */
679 err = -EINPROGRESS;
680 break;
681 }
682
683 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
684
685 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
686 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
687 tcp_sk(sk)->fastopen_req &&
688 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
689
690 /* Error code is set above */
691 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
692 goto out;
693
694 err = sock_intr_errno(timeo);
695 if (signal_pending(current))
696 goto out;
697 }
698
699 /* Connection was closed by RST, timeout, ICMP error
700 * or another process disconnected us.
701 */
702 if (sk->sk_state == TCP_CLOSE)
703 goto sock_error;
704
705 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
706 * and error was received after socket entered established state.
707 * Hence, it is handled normally after connect() return successfully.
708 */
709
710 sock->state = SS_CONNECTED;
711 err = 0;
712out:
713 return err;
714
715sock_error:
716 err = sock_error(sk) ? : -ECONNABORTED;
717 sock->state = SS_UNCONNECTED;
718 if (sk->sk_prot->disconnect(sk, flags))
719 sock->state = SS_DISCONNECTING;
720 goto out;
721}
722EXPORT_SYMBOL(__inet_stream_connect);
723
724int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
725 int addr_len, int flags)
726{
727 int err;
728
729 lock_sock(sock->sk);
730 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
731 release_sock(sock->sk);
732 return err;
733}
734EXPORT_SYMBOL(inet_stream_connect);
735
736/*
737 * Accept a pending connection. The TCP layer now gives BSD semantics.
738 */
739
740int inet_accept(struct socket *sock, struct socket *newsock, int flags,
741 bool kern)
742{
743 struct sock *sk1 = sock->sk, *sk2;
744 int err = -EINVAL;
745
746 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
747 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
748 if (!sk2)
749 goto do_err;
750
751 lock_sock(sk2);
752
753 sock_rps_record_flow(sk2);
754 WARN_ON(!((1 << sk2->sk_state) &
755 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
756 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
757
758 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
759 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
760 sock_graft(sk2, newsock);
761
762 newsock->state = SS_CONNECTED;
763 err = 0;
764 release_sock(sk2);
765do_err:
766 return err;
767}
768EXPORT_SYMBOL(inet_accept);
769
770/*
771 * This does both peername and sockname.
772 */
773int inet_getname(struct socket *sock, struct sockaddr *uaddr,
774 int peer)
775{
776 struct sock *sk = sock->sk;
777 struct inet_sock *inet = inet_sk(sk);
778 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
779
780 sin->sin_family = AF_INET;
781 lock_sock(sk);
782 if (peer) {
783 if (!inet->inet_dport ||
784 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
785 peer == 1)) {
786 release_sock(sk);
787 return -ENOTCONN;
788 }
789 sin->sin_port = inet->inet_dport;
790 sin->sin_addr.s_addr = inet->inet_daddr;
791 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
792 CGROUP_INET4_GETPEERNAME);
793 } else {
794 __be32 addr = inet->inet_rcv_saddr;
795 if (!addr)
796 addr = inet->inet_saddr;
797 sin->sin_port = inet->inet_sport;
798 sin->sin_addr.s_addr = addr;
799 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
800 CGROUP_INET4_GETSOCKNAME);
801 }
802 release_sock(sk);
803 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
804 return sizeof(*sin);
805}
806EXPORT_SYMBOL(inet_getname);
807
808int inet_send_prepare(struct sock *sk)
809{
810 sock_rps_record_flow(sk);
811
812 /* We may need to bind the socket. */
813 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
814 inet_autobind(sk))
815 return -EAGAIN;
816
817 return 0;
818}
819EXPORT_SYMBOL_GPL(inet_send_prepare);
820
821int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
822{
823 struct sock *sk = sock->sk;
824
825 if (unlikely(inet_send_prepare(sk)))
826 return -EAGAIN;
827
828 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
829 sk, msg, size);
830}
831EXPORT_SYMBOL(inet_sendmsg);
832
833ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
834 size_t size, int flags)
835{
836 struct sock *sk = sock->sk;
837 const struct proto *prot;
838
839 if (unlikely(inet_send_prepare(sk)))
840 return -EAGAIN;
841
842 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
843 prot = READ_ONCE(sk->sk_prot);
844 if (prot->sendpage)
845 return prot->sendpage(sk, page, offset, size, flags);
846 return sock_no_sendpage(sock, page, offset, size, flags);
847}
848EXPORT_SYMBOL(inet_sendpage);
849
850INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
851 size_t, int, int *));
852int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
853 int flags)
854{
855 struct sock *sk = sock->sk;
856 int addr_len = 0;
857 int err;
858
859 if (likely(!(flags & MSG_ERRQUEUE)))
860 sock_rps_record_flow(sk);
861
862 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
863 sk, msg, size, flags, &addr_len);
864 if (err >= 0)
865 msg->msg_namelen = addr_len;
866 return err;
867}
868EXPORT_SYMBOL(inet_recvmsg);
869
870int inet_shutdown(struct socket *sock, int how)
871{
872 struct sock *sk = sock->sk;
873 int err = 0;
874
875 /* This should really check to make sure
876 * the socket is a TCP socket. (WHY AC...)
877 */
878 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
879 1->2 bit 2 snds.
880 2->3 */
881 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
882 return -EINVAL;
883
884 lock_sock(sk);
885 if (sock->state == SS_CONNECTING) {
886 if ((1 << sk->sk_state) &
887 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
888 sock->state = SS_DISCONNECTING;
889 else
890 sock->state = SS_CONNECTED;
891 }
892
893 switch (sk->sk_state) {
894 case TCP_CLOSE:
895 err = -ENOTCONN;
896 /* Hack to wake up other listeners, who can poll for
897 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
898 fallthrough;
899 default:
900 sk->sk_shutdown |= how;
901 if (sk->sk_prot->shutdown)
902 sk->sk_prot->shutdown(sk, how);
903 break;
904
905 /* Remaining two branches are temporary solution for missing
906 * close() in multithreaded environment. It is _not_ a good idea,
907 * but we have no choice until close() is repaired at VFS level.
908 */
909 case TCP_LISTEN:
910 if (!(how & RCV_SHUTDOWN))
911 break;
912 fallthrough;
913 case TCP_SYN_SENT:
914 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
915 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
916 break;
917 }
918
919 /* Wake up anyone sleeping in poll. */
920 sk->sk_state_change(sk);
921 release_sock(sk);
922 return err;
923}
924EXPORT_SYMBOL(inet_shutdown);
925
926/*
927 * ioctl() calls you can issue on an INET socket. Most of these are
928 * device configuration and stuff and very rarely used. Some ioctls
929 * pass on to the socket itself.
930 *
931 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
932 * loads the devconfigure module does its configuring and unloads it.
933 * There's a good 20K of config code hanging around the kernel.
934 */
935
936int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
937{
938 struct sock *sk = sock->sk;
939 int err = 0;
940 struct net *net = sock_net(sk);
941 void __user *p = (void __user *)arg;
942 struct ifreq ifr;
943 struct rtentry rt;
944
945 switch (cmd) {
946 case SIOCADDRT:
947 case SIOCDELRT:
948 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
949 return -EFAULT;
950 err = ip_rt_ioctl(net, cmd, &rt);
951 break;
952 case SIOCRTMSG:
953 err = -EINVAL;
954 break;
955 case SIOCDARP:
956 case SIOCGARP:
957 case SIOCSARP:
958 err = arp_ioctl(net, cmd, (void __user *)arg);
959 break;
960 case SIOCGIFADDR:
961 case SIOCGIFBRDADDR:
962 case SIOCGIFNETMASK:
963 case SIOCGIFDSTADDR:
964 case SIOCGIFPFLAGS:
965 if (get_user_ifreq(&ifr, NULL, p))
966 return -EFAULT;
967 err = devinet_ioctl(net, cmd, &ifr);
968 if (!err && put_user_ifreq(&ifr, p))
969 err = -EFAULT;
970 break;
971
972 case SIOCSIFADDR:
973 case SIOCSIFBRDADDR:
974 case SIOCSIFNETMASK:
975 case SIOCSIFDSTADDR:
976 case SIOCSIFPFLAGS:
977 case SIOCSIFFLAGS:
978 if (get_user_ifreq(&ifr, NULL, p))
979 return -EFAULT;
980 err = devinet_ioctl(net, cmd, &ifr);
981 break;
982 default:
983 if (sk->sk_prot->ioctl)
984 err = sk->sk_prot->ioctl(sk, cmd, arg);
985 else
986 err = -ENOIOCTLCMD;
987 break;
988 }
989 return err;
990}
991EXPORT_SYMBOL(inet_ioctl);
992
993#ifdef CONFIG_COMPAT
994static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
995 struct compat_rtentry __user *ur)
996{
997 compat_uptr_t rtdev;
998 struct rtentry rt;
999
1000 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1001 3 * sizeof(struct sockaddr)) ||
1002 get_user(rt.rt_flags, &ur->rt_flags) ||
1003 get_user(rt.rt_metric, &ur->rt_metric) ||
1004 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1005 get_user(rt.rt_window, &ur->rt_window) ||
1006 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1007 get_user(rtdev, &ur->rt_dev))
1008 return -EFAULT;
1009
1010 rt.rt_dev = compat_ptr(rtdev);
1011 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1012}
1013
1014static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1015{
1016 void __user *argp = compat_ptr(arg);
1017 struct sock *sk = sock->sk;
1018
1019 switch (cmd) {
1020 case SIOCADDRT:
1021 case SIOCDELRT:
1022 return inet_compat_routing_ioctl(sk, cmd, argp);
1023 default:
1024 if (!sk->sk_prot->compat_ioctl)
1025 return -ENOIOCTLCMD;
1026 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1027 }
1028}
1029#endif /* CONFIG_COMPAT */
1030
1031const struct proto_ops inet_stream_ops = {
1032 .family = PF_INET,
1033 .owner = THIS_MODULE,
1034 .release = inet_release,
1035 .bind = inet_bind,
1036 .connect = inet_stream_connect,
1037 .socketpair = sock_no_socketpair,
1038 .accept = inet_accept,
1039 .getname = inet_getname,
1040 .poll = tcp_poll,
1041 .ioctl = inet_ioctl,
1042 .gettstamp = sock_gettstamp,
1043 .listen = inet_listen,
1044 .shutdown = inet_shutdown,
1045 .setsockopt = sock_common_setsockopt,
1046 .getsockopt = sock_common_getsockopt,
1047 .sendmsg = inet_sendmsg,
1048 .recvmsg = inet_recvmsg,
1049#ifdef CONFIG_MMU
1050 .mmap = tcp_mmap,
1051#endif
1052 .sendpage = inet_sendpage,
1053 .splice_read = tcp_splice_read,
1054 .read_sock = tcp_read_sock,
1055 .read_skb = tcp_read_skb,
1056 .sendmsg_locked = tcp_sendmsg_locked,
1057 .sendpage_locked = tcp_sendpage_locked,
1058 .peek_len = tcp_peek_len,
1059#ifdef CONFIG_COMPAT
1060 .compat_ioctl = inet_compat_ioctl,
1061#endif
1062 .set_rcvlowat = tcp_set_rcvlowat,
1063};
1064EXPORT_SYMBOL(inet_stream_ops);
1065
1066const struct proto_ops inet_dgram_ops = {
1067 .family = PF_INET,
1068 .owner = THIS_MODULE,
1069 .release = inet_release,
1070 .bind = inet_bind,
1071 .connect = inet_dgram_connect,
1072 .socketpair = sock_no_socketpair,
1073 .accept = sock_no_accept,
1074 .getname = inet_getname,
1075 .poll = udp_poll,
1076 .ioctl = inet_ioctl,
1077 .gettstamp = sock_gettstamp,
1078 .listen = sock_no_listen,
1079 .shutdown = inet_shutdown,
1080 .setsockopt = sock_common_setsockopt,
1081 .getsockopt = sock_common_getsockopt,
1082 .sendmsg = inet_sendmsg,
1083 .read_skb = udp_read_skb,
1084 .recvmsg = inet_recvmsg,
1085 .mmap = sock_no_mmap,
1086 .sendpage = inet_sendpage,
1087 .set_peek_off = sk_set_peek_off,
1088#ifdef CONFIG_COMPAT
1089 .compat_ioctl = inet_compat_ioctl,
1090#endif
1091};
1092EXPORT_SYMBOL(inet_dgram_ops);
1093
1094/*
1095 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1096 * udp_poll
1097 */
1098static const struct proto_ops inet_sockraw_ops = {
1099 .family = PF_INET,
1100 .owner = THIS_MODULE,
1101 .release = inet_release,
1102 .bind = inet_bind,
1103 .connect = inet_dgram_connect,
1104 .socketpair = sock_no_socketpair,
1105 .accept = sock_no_accept,
1106 .getname = inet_getname,
1107 .poll = datagram_poll,
1108 .ioctl = inet_ioctl,
1109 .gettstamp = sock_gettstamp,
1110 .listen = sock_no_listen,
1111 .shutdown = inet_shutdown,
1112 .setsockopt = sock_common_setsockopt,
1113 .getsockopt = sock_common_getsockopt,
1114 .sendmsg = inet_sendmsg,
1115 .recvmsg = inet_recvmsg,
1116 .mmap = sock_no_mmap,
1117 .sendpage = inet_sendpage,
1118#ifdef CONFIG_COMPAT
1119 .compat_ioctl = inet_compat_ioctl,
1120#endif
1121};
1122
1123static const struct net_proto_family inet_family_ops = {
1124 .family = PF_INET,
1125 .create = inet_create,
1126 .owner = THIS_MODULE,
1127};
1128
1129/* Upon startup we insert all the elements in inetsw_array[] into
1130 * the linked list inetsw.
1131 */
1132static struct inet_protosw inetsw_array[] =
1133{
1134 {
1135 .type = SOCK_STREAM,
1136 .protocol = IPPROTO_TCP,
1137 .prot = &tcp_prot,
1138 .ops = &inet_stream_ops,
1139 .flags = INET_PROTOSW_PERMANENT |
1140 INET_PROTOSW_ICSK,
1141 },
1142
1143 {
1144 .type = SOCK_DGRAM,
1145 .protocol = IPPROTO_UDP,
1146 .prot = &udp_prot,
1147 .ops = &inet_dgram_ops,
1148 .flags = INET_PROTOSW_PERMANENT,
1149 },
1150
1151 {
1152 .type = SOCK_DGRAM,
1153 .protocol = IPPROTO_ICMP,
1154 .prot = &ping_prot,
1155 .ops = &inet_sockraw_ops,
1156 .flags = INET_PROTOSW_REUSE,
1157 },
1158
1159 {
1160 .type = SOCK_RAW,
1161 .protocol = IPPROTO_IP, /* wild card */
1162 .prot = &raw_prot,
1163 .ops = &inet_sockraw_ops,
1164 .flags = INET_PROTOSW_REUSE,
1165 }
1166};
1167
1168#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1169
1170void inet_register_protosw(struct inet_protosw *p)
1171{
1172 struct list_head *lh;
1173 struct inet_protosw *answer;
1174 int protocol = p->protocol;
1175 struct list_head *last_perm;
1176
1177 spin_lock_bh(&inetsw_lock);
1178
1179 if (p->type >= SOCK_MAX)
1180 goto out_illegal;
1181
1182 /* If we are trying to override a permanent protocol, bail. */
1183 last_perm = &inetsw[p->type];
1184 list_for_each(lh, &inetsw[p->type]) {
1185 answer = list_entry(lh, struct inet_protosw, list);
1186 /* Check only the non-wild match. */
1187 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1188 break;
1189 if (protocol == answer->protocol)
1190 goto out_permanent;
1191 last_perm = lh;
1192 }
1193
1194 /* Add the new entry after the last permanent entry if any, so that
1195 * the new entry does not override a permanent entry when matched with
1196 * a wild-card protocol. But it is allowed to override any existing
1197 * non-permanent entry. This means that when we remove this entry, the
1198 * system automatically returns to the old behavior.
1199 */
1200 list_add_rcu(&p->list, last_perm);
1201out:
1202 spin_unlock_bh(&inetsw_lock);
1203
1204 return;
1205
1206out_permanent:
1207 pr_err("Attempt to override permanent protocol %d\n", protocol);
1208 goto out;
1209
1210out_illegal:
1211 pr_err("Ignoring attempt to register invalid socket type %d\n",
1212 p->type);
1213 goto out;
1214}
1215EXPORT_SYMBOL(inet_register_protosw);
1216
1217void inet_unregister_protosw(struct inet_protosw *p)
1218{
1219 if (INET_PROTOSW_PERMANENT & p->flags) {
1220 pr_err("Attempt to unregister permanent protocol %d\n",
1221 p->protocol);
1222 } else {
1223 spin_lock_bh(&inetsw_lock);
1224 list_del_rcu(&p->list);
1225 spin_unlock_bh(&inetsw_lock);
1226
1227 synchronize_net();
1228 }
1229}
1230EXPORT_SYMBOL(inet_unregister_protosw);
1231
1232static int inet_sk_reselect_saddr(struct sock *sk)
1233{
1234 struct inet_sock *inet = inet_sk(sk);
1235 __be32 old_saddr = inet->inet_saddr;
1236 __be32 daddr = inet->inet_daddr;
1237 struct flowi4 *fl4;
1238 struct rtable *rt;
1239 __be32 new_saddr;
1240 struct ip_options_rcu *inet_opt;
1241 int err;
1242
1243 inet_opt = rcu_dereference_protected(inet->inet_opt,
1244 lockdep_sock_is_held(sk));
1245 if (inet_opt && inet_opt->opt.srr)
1246 daddr = inet_opt->opt.faddr;
1247
1248 /* Query new route. */
1249 fl4 = &inet->cork.fl.u.ip4;
1250 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1251 sk->sk_protocol, inet->inet_sport,
1252 inet->inet_dport, sk);
1253 if (IS_ERR(rt))
1254 return PTR_ERR(rt);
1255
1256 new_saddr = fl4->saddr;
1257
1258 if (new_saddr == old_saddr) {
1259 sk_setup_caps(sk, &rt->dst);
1260 return 0;
1261 }
1262
1263 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1264 if (err) {
1265 ip_rt_put(rt);
1266 return err;
1267 }
1268
1269 sk_setup_caps(sk, &rt->dst);
1270
1271 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1272 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1273 __func__, &old_saddr, &new_saddr);
1274 }
1275
1276 /*
1277 * XXX The only one ugly spot where we need to
1278 * XXX really change the sockets identity after
1279 * XXX it has entered the hashes. -DaveM
1280 *
1281 * Besides that, it does not check for connection
1282 * uniqueness. Wait for troubles.
1283 */
1284 return __sk_prot_rehash(sk);
1285}
1286
1287int inet_sk_rebuild_header(struct sock *sk)
1288{
1289 struct inet_sock *inet = inet_sk(sk);
1290 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1291 __be32 daddr;
1292 struct ip_options_rcu *inet_opt;
1293 struct flowi4 *fl4;
1294 int err;
1295
1296 /* Route is OK, nothing to do. */
1297 if (rt)
1298 return 0;
1299
1300 /* Reroute. */
1301 rcu_read_lock();
1302 inet_opt = rcu_dereference(inet->inet_opt);
1303 daddr = inet->inet_daddr;
1304 if (inet_opt && inet_opt->opt.srr)
1305 daddr = inet_opt->opt.faddr;
1306 rcu_read_unlock();
1307 fl4 = &inet->cork.fl.u.ip4;
1308 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1309 inet->inet_dport, inet->inet_sport,
1310 sk->sk_protocol, RT_CONN_FLAGS(sk),
1311 sk->sk_bound_dev_if);
1312 if (!IS_ERR(rt)) {
1313 err = 0;
1314 sk_setup_caps(sk, &rt->dst);
1315 } else {
1316 err = PTR_ERR(rt);
1317
1318 /* Routing failed... */
1319 sk->sk_route_caps = 0;
1320 /*
1321 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1322 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1323 */
1324 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1325 sk->sk_state != TCP_SYN_SENT ||
1326 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1327 (err = inet_sk_reselect_saddr(sk)) != 0)
1328 sk->sk_err_soft = -err;
1329 }
1330
1331 return err;
1332}
1333EXPORT_SYMBOL(inet_sk_rebuild_header);
1334
1335void inet_sk_set_state(struct sock *sk, int state)
1336{
1337 trace_inet_sock_set_state(sk, sk->sk_state, state);
1338 sk->sk_state = state;
1339}
1340EXPORT_SYMBOL(inet_sk_set_state);
1341
1342void inet_sk_state_store(struct sock *sk, int newstate)
1343{
1344 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1345 smp_store_release(&sk->sk_state, newstate);
1346}
1347
1348struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1349 netdev_features_t features)
1350{
1351 bool udpfrag = false, fixedid = false, gso_partial, encap;
1352 struct sk_buff *segs = ERR_PTR(-EINVAL);
1353 const struct net_offload *ops;
1354 unsigned int offset = 0;
1355 struct iphdr *iph;
1356 int proto, tot_len;
1357 int nhoff;
1358 int ihl;
1359 int id;
1360
1361 skb_reset_network_header(skb);
1362 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1363 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1364 goto out;
1365
1366 iph = ip_hdr(skb);
1367 ihl = iph->ihl * 4;
1368 if (ihl < sizeof(*iph))
1369 goto out;
1370
1371 id = ntohs(iph->id);
1372 proto = iph->protocol;
1373
1374 /* Warning: after this point, iph might be no longer valid */
1375 if (unlikely(!pskb_may_pull(skb, ihl)))
1376 goto out;
1377 __skb_pull(skb, ihl);
1378
1379 encap = SKB_GSO_CB(skb)->encap_level > 0;
1380 if (encap)
1381 features &= skb->dev->hw_enc_features;
1382 SKB_GSO_CB(skb)->encap_level += ihl;
1383
1384 skb_reset_transport_header(skb);
1385
1386 segs = ERR_PTR(-EPROTONOSUPPORT);
1387
1388 if (!skb->encapsulation || encap) {
1389 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1390 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1391
1392 /* fixed ID is invalid if DF bit is not set */
1393 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1394 goto out;
1395 }
1396
1397 ops = rcu_dereference(inet_offloads[proto]);
1398 if (likely(ops && ops->callbacks.gso_segment)) {
1399 segs = ops->callbacks.gso_segment(skb, features);
1400 if (!segs)
1401 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1402 }
1403
1404 if (IS_ERR_OR_NULL(segs))
1405 goto out;
1406
1407 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1408
1409 skb = segs;
1410 do {
1411 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1412 if (udpfrag) {
1413 iph->frag_off = htons(offset >> 3);
1414 if (skb->next)
1415 iph->frag_off |= htons(IP_MF);
1416 offset += skb->len - nhoff - ihl;
1417 tot_len = skb->len - nhoff;
1418 } else if (skb_is_gso(skb)) {
1419 if (!fixedid) {
1420 iph->id = htons(id);
1421 id += skb_shinfo(skb)->gso_segs;
1422 }
1423
1424 if (gso_partial)
1425 tot_len = skb_shinfo(skb)->gso_size +
1426 SKB_GSO_CB(skb)->data_offset +
1427 skb->head - (unsigned char *)iph;
1428 else
1429 tot_len = skb->len - nhoff;
1430 } else {
1431 if (!fixedid)
1432 iph->id = htons(id++);
1433 tot_len = skb->len - nhoff;
1434 }
1435 iph->tot_len = htons(tot_len);
1436 ip_send_check(iph);
1437 if (encap)
1438 skb_reset_inner_headers(skb);
1439 skb->network_header = (u8 *)iph - skb->head;
1440 skb_reset_mac_len(skb);
1441 } while ((skb = skb->next));
1442
1443out:
1444 return segs;
1445}
1446
1447static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1448 netdev_features_t features)
1449{
1450 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1451 return ERR_PTR(-EINVAL);
1452
1453 return inet_gso_segment(skb, features);
1454}
1455
1456struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1457{
1458 const struct net_offload *ops;
1459 struct sk_buff *pp = NULL;
1460 const struct iphdr *iph;
1461 struct sk_buff *p;
1462 unsigned int hlen;
1463 unsigned int off;
1464 unsigned int id;
1465 int flush = 1;
1466 int proto;
1467
1468 off = skb_gro_offset(skb);
1469 hlen = off + sizeof(*iph);
1470 iph = skb_gro_header(skb, hlen, off);
1471 if (unlikely(!iph))
1472 goto out;
1473
1474 proto = iph->protocol;
1475
1476 ops = rcu_dereference(inet_offloads[proto]);
1477 if (!ops || !ops->callbacks.gro_receive)
1478 goto out;
1479
1480 if (*(u8 *)iph != 0x45)
1481 goto out;
1482
1483 if (ip_is_fragment(iph))
1484 goto out;
1485
1486 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1487 goto out;
1488
1489 id = ntohl(*(__be32 *)&iph->id);
1490 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1491 id >>= 16;
1492
1493 list_for_each_entry(p, head, list) {
1494 struct iphdr *iph2;
1495 u16 flush_id;
1496
1497 if (!NAPI_GRO_CB(p)->same_flow)
1498 continue;
1499
1500 iph2 = (struct iphdr *)(p->data + off);
1501 /* The above works because, with the exception of the top
1502 * (inner most) layer, we only aggregate pkts with the same
1503 * hdr length so all the hdrs we'll need to verify will start
1504 * at the same offset.
1505 */
1506 if ((iph->protocol ^ iph2->protocol) |
1507 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1508 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1509 NAPI_GRO_CB(p)->same_flow = 0;
1510 continue;
1511 }
1512
1513 /* All fields must match except length and checksum. */
1514 NAPI_GRO_CB(p)->flush |=
1515 (iph->ttl ^ iph2->ttl) |
1516 (iph->tos ^ iph2->tos) |
1517 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1518
1519 NAPI_GRO_CB(p)->flush |= flush;
1520
1521 /* We need to store of the IP ID check to be included later
1522 * when we can verify that this packet does in fact belong
1523 * to a given flow.
1524 */
1525 flush_id = (u16)(id - ntohs(iph2->id));
1526
1527 /* This bit of code makes it much easier for us to identify
1528 * the cases where we are doing atomic vs non-atomic IP ID
1529 * checks. Specifically an atomic check can return IP ID
1530 * values 0 - 0xFFFF, while a non-atomic check can only
1531 * return 0 or 0xFFFF.
1532 */
1533 if (!NAPI_GRO_CB(p)->is_atomic ||
1534 !(iph->frag_off & htons(IP_DF))) {
1535 flush_id ^= NAPI_GRO_CB(p)->count;
1536 flush_id = flush_id ? 0xFFFF : 0;
1537 }
1538
1539 /* If the previous IP ID value was based on an atomic
1540 * datagram we can overwrite the value and ignore it.
1541 */
1542 if (NAPI_GRO_CB(skb)->is_atomic)
1543 NAPI_GRO_CB(p)->flush_id = flush_id;
1544 else
1545 NAPI_GRO_CB(p)->flush_id |= flush_id;
1546 }
1547
1548 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1549 NAPI_GRO_CB(skb)->flush |= flush;
1550 skb_set_network_header(skb, off);
1551 /* The above will be needed by the transport layer if there is one
1552 * immediately following this IP hdr.
1553 */
1554
1555 /* Note : No need to call skb_gro_postpull_rcsum() here,
1556 * as we already checked checksum over ipv4 header was 0
1557 */
1558 skb_gro_pull(skb, sizeof(*iph));
1559 skb_set_transport_header(skb, skb_gro_offset(skb));
1560
1561 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1562 ops->callbacks.gro_receive, head, skb);
1563
1564out:
1565 skb_gro_flush_final(skb, pp, flush);
1566
1567 return pp;
1568}
1569
1570static struct sk_buff *ipip_gro_receive(struct list_head *head,
1571 struct sk_buff *skb)
1572{
1573 if (NAPI_GRO_CB(skb)->encap_mark) {
1574 NAPI_GRO_CB(skb)->flush = 1;
1575 return NULL;
1576 }
1577
1578 NAPI_GRO_CB(skb)->encap_mark = 1;
1579
1580 return inet_gro_receive(head, skb);
1581}
1582
1583#define SECONDS_PER_DAY 86400
1584
1585/* inet_current_timestamp - Return IP network timestamp
1586 *
1587 * Return milliseconds since midnight in network byte order.
1588 */
1589__be32 inet_current_timestamp(void)
1590{
1591 u32 secs;
1592 u32 msecs;
1593 struct timespec64 ts;
1594
1595 ktime_get_real_ts64(&ts);
1596
1597 /* Get secs since midnight. */
1598 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1599 /* Convert to msecs. */
1600 msecs = secs * MSEC_PER_SEC;
1601 /* Convert nsec to msec. */
1602 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1603
1604 /* Convert to network byte order. */
1605 return htonl(msecs);
1606}
1607EXPORT_SYMBOL(inet_current_timestamp);
1608
1609int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1610{
1611 if (sk->sk_family == AF_INET)
1612 return ip_recv_error(sk, msg, len, addr_len);
1613#if IS_ENABLED(CONFIG_IPV6)
1614 if (sk->sk_family == AF_INET6)
1615 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1616#endif
1617 return -EINVAL;
1618}
1619
1620int inet_gro_complete(struct sk_buff *skb, int nhoff)
1621{
1622 __be16 newlen = htons(skb->len - nhoff);
1623 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1624 const struct net_offload *ops;
1625 int proto = iph->protocol;
1626 int err = -ENOSYS;
1627
1628 if (skb->encapsulation) {
1629 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1630 skb_set_inner_network_header(skb, nhoff);
1631 }
1632
1633 csum_replace2(&iph->check, iph->tot_len, newlen);
1634 iph->tot_len = newlen;
1635
1636 ops = rcu_dereference(inet_offloads[proto]);
1637 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1638 goto out;
1639
1640 /* Only need to add sizeof(*iph) to get to the next hdr below
1641 * because any hdr with option will have been flushed in
1642 * inet_gro_receive().
1643 */
1644 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1645 tcp4_gro_complete, udp4_gro_complete,
1646 skb, nhoff + sizeof(*iph));
1647
1648out:
1649 return err;
1650}
1651
1652static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1653{
1654 skb->encapsulation = 1;
1655 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1656 return inet_gro_complete(skb, nhoff);
1657}
1658
1659int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1660 unsigned short type, unsigned char protocol,
1661 struct net *net)
1662{
1663 struct socket *sock;
1664 int rc = sock_create_kern(net, family, type, protocol, &sock);
1665
1666 if (rc == 0) {
1667 *sk = sock->sk;
1668 (*sk)->sk_allocation = GFP_ATOMIC;
1669 (*sk)->sk_use_task_frag = false;
1670 /*
1671 * Unhash it so that IP input processing does not even see it,
1672 * we do not wish this socket to see incoming packets.
1673 */
1674 (*sk)->sk_prot->unhash(*sk);
1675 }
1676 return rc;
1677}
1678EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1679
1680unsigned long snmp_fold_field(void __percpu *mib, int offt)
1681{
1682 unsigned long res = 0;
1683 int i;
1684
1685 for_each_possible_cpu(i)
1686 res += snmp_get_cpu_field(mib, i, offt);
1687 return res;
1688}
1689EXPORT_SYMBOL_GPL(snmp_fold_field);
1690
1691#if BITS_PER_LONG==32
1692
1693u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1694 size_t syncp_offset)
1695{
1696 void *bhptr;
1697 struct u64_stats_sync *syncp;
1698 u64 v;
1699 unsigned int start;
1700
1701 bhptr = per_cpu_ptr(mib, cpu);
1702 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1703 do {
1704 start = u64_stats_fetch_begin(syncp);
1705 v = *(((u64 *)bhptr) + offt);
1706 } while (u64_stats_fetch_retry(syncp, start));
1707
1708 return v;
1709}
1710EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1711
1712u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1713{
1714 u64 res = 0;
1715 int cpu;
1716
1717 for_each_possible_cpu(cpu) {
1718 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1719 }
1720 return res;
1721}
1722EXPORT_SYMBOL_GPL(snmp_fold_field64);
1723#endif
1724
1725#ifdef CONFIG_IP_MULTICAST
1726static const struct net_protocol igmp_protocol = {
1727 .handler = igmp_rcv,
1728};
1729#endif
1730
1731static const struct net_protocol tcp_protocol = {
1732 .handler = tcp_v4_rcv,
1733 .err_handler = tcp_v4_err,
1734 .no_policy = 1,
1735 .icmp_strict_tag_validation = 1,
1736};
1737
1738static const struct net_protocol udp_protocol = {
1739 .handler = udp_rcv,
1740 .err_handler = udp_err,
1741 .no_policy = 1,
1742};
1743
1744static const struct net_protocol icmp_protocol = {
1745 .handler = icmp_rcv,
1746 .err_handler = icmp_err,
1747 .no_policy = 1,
1748};
1749
1750static __net_init int ipv4_mib_init_net(struct net *net)
1751{
1752 int i;
1753
1754 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1755 if (!net->mib.tcp_statistics)
1756 goto err_tcp_mib;
1757 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1758 if (!net->mib.ip_statistics)
1759 goto err_ip_mib;
1760
1761 for_each_possible_cpu(i) {
1762 struct ipstats_mib *af_inet_stats;
1763 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1764 u64_stats_init(&af_inet_stats->syncp);
1765 }
1766
1767 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1768 if (!net->mib.net_statistics)
1769 goto err_net_mib;
1770 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1771 if (!net->mib.udp_statistics)
1772 goto err_udp_mib;
1773 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1774 if (!net->mib.udplite_statistics)
1775 goto err_udplite_mib;
1776 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1777 if (!net->mib.icmp_statistics)
1778 goto err_icmp_mib;
1779 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1780 GFP_KERNEL);
1781 if (!net->mib.icmpmsg_statistics)
1782 goto err_icmpmsg_mib;
1783
1784 tcp_mib_init(net);
1785 return 0;
1786
1787err_icmpmsg_mib:
1788 free_percpu(net->mib.icmp_statistics);
1789err_icmp_mib:
1790 free_percpu(net->mib.udplite_statistics);
1791err_udplite_mib:
1792 free_percpu(net->mib.udp_statistics);
1793err_udp_mib:
1794 free_percpu(net->mib.net_statistics);
1795err_net_mib:
1796 free_percpu(net->mib.ip_statistics);
1797err_ip_mib:
1798 free_percpu(net->mib.tcp_statistics);
1799err_tcp_mib:
1800 return -ENOMEM;
1801}
1802
1803static __net_exit void ipv4_mib_exit_net(struct net *net)
1804{
1805 kfree(net->mib.icmpmsg_statistics);
1806 free_percpu(net->mib.icmp_statistics);
1807 free_percpu(net->mib.udplite_statistics);
1808 free_percpu(net->mib.udp_statistics);
1809 free_percpu(net->mib.net_statistics);
1810 free_percpu(net->mib.ip_statistics);
1811 free_percpu(net->mib.tcp_statistics);
1812#ifdef CONFIG_MPTCP
1813 /* allocated on demand, see mptcp_init_sock() */
1814 free_percpu(net->mib.mptcp_statistics);
1815#endif
1816}
1817
1818static __net_initdata struct pernet_operations ipv4_mib_ops = {
1819 .init = ipv4_mib_init_net,
1820 .exit = ipv4_mib_exit_net,
1821};
1822
1823static int __init init_ipv4_mibs(void)
1824{
1825 return register_pernet_subsys(&ipv4_mib_ops);
1826}
1827
1828static __net_init int inet_init_net(struct net *net)
1829{
1830 /*
1831 * Set defaults for local port range
1832 */
1833 seqlock_init(&net->ipv4.ip_local_ports.lock);
1834 net->ipv4.ip_local_ports.range[0] = 32768;
1835 net->ipv4.ip_local_ports.range[1] = 60999;
1836
1837 seqlock_init(&net->ipv4.ping_group_range.lock);
1838 /*
1839 * Sane defaults - nobody may create ping sockets.
1840 * Boot scripts should set this to distro-specific group.
1841 */
1842 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1843 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1844
1845 /* Default values for sysctl-controlled parameters.
1846 * We set them here, in case sysctl is not compiled.
1847 */
1848 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1849 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1850 net->ipv4.sysctl_ip_dynaddr = 0;
1851 net->ipv4.sysctl_ip_early_demux = 1;
1852 net->ipv4.sysctl_udp_early_demux = 1;
1853 net->ipv4.sysctl_tcp_early_demux = 1;
1854 net->ipv4.sysctl_nexthop_compat_mode = 1;
1855#ifdef CONFIG_SYSCTL
1856 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1857#endif
1858
1859 /* Some igmp sysctl, whose values are always used */
1860 net->ipv4.sysctl_igmp_max_memberships = 20;
1861 net->ipv4.sysctl_igmp_max_msf = 10;
1862 /* IGMP reports for link-local multicast groups are enabled by default */
1863 net->ipv4.sysctl_igmp_llm_reports = 1;
1864 net->ipv4.sysctl_igmp_qrv = 2;
1865
1866 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1867
1868 return 0;
1869}
1870
1871static __net_initdata struct pernet_operations af_inet_ops = {
1872 .init = inet_init_net,
1873};
1874
1875static int __init init_inet_pernet_ops(void)
1876{
1877 return register_pernet_subsys(&af_inet_ops);
1878}
1879
1880static int ipv4_proc_init(void);
1881
1882/*
1883 * IP protocol layer initialiser
1884 */
1885
1886static struct packet_offload ip_packet_offload __read_mostly = {
1887 .type = cpu_to_be16(ETH_P_IP),
1888 .callbacks = {
1889 .gso_segment = inet_gso_segment,
1890 .gro_receive = inet_gro_receive,
1891 .gro_complete = inet_gro_complete,
1892 },
1893};
1894
1895static const struct net_offload ipip_offload = {
1896 .callbacks = {
1897 .gso_segment = ipip_gso_segment,
1898 .gro_receive = ipip_gro_receive,
1899 .gro_complete = ipip_gro_complete,
1900 },
1901};
1902
1903static int __init ipip_offload_init(void)
1904{
1905 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1906}
1907
1908static int __init ipv4_offload_init(void)
1909{
1910 /*
1911 * Add offloads
1912 */
1913 if (udpv4_offload_init() < 0)
1914 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1915 if (tcpv4_offload_init() < 0)
1916 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1917 if (ipip_offload_init() < 0)
1918 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1919
1920 dev_add_offload(&ip_packet_offload);
1921 return 0;
1922}
1923
1924fs_initcall(ipv4_offload_init);
1925
1926static struct packet_type ip_packet_type __read_mostly = {
1927 .type = cpu_to_be16(ETH_P_IP),
1928 .func = ip_rcv,
1929 .list_func = ip_list_rcv,
1930};
1931
1932static int __init inet_init(void)
1933{
1934 struct inet_protosw *q;
1935 struct list_head *r;
1936 int rc;
1937
1938 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1939
1940 raw_hashinfo_init(&raw_v4_hashinfo);
1941
1942 rc = proto_register(&tcp_prot, 1);
1943 if (rc)
1944 goto out;
1945
1946 rc = proto_register(&udp_prot, 1);
1947 if (rc)
1948 goto out_unregister_tcp_proto;
1949
1950 rc = proto_register(&raw_prot, 1);
1951 if (rc)
1952 goto out_unregister_udp_proto;
1953
1954 rc = proto_register(&ping_prot, 1);
1955 if (rc)
1956 goto out_unregister_raw_proto;
1957
1958 /*
1959 * Tell SOCKET that we are alive...
1960 */
1961
1962 (void)sock_register(&inet_family_ops);
1963
1964#ifdef CONFIG_SYSCTL
1965 ip_static_sysctl_init();
1966#endif
1967
1968 /*
1969 * Add all the base protocols.
1970 */
1971
1972 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1973 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1974 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1975 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1976 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1977 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1978#ifdef CONFIG_IP_MULTICAST
1979 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1980 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1981#endif
1982
1983 /* Register the socket-side information for inet_create. */
1984 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1985 INIT_LIST_HEAD(r);
1986
1987 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1988 inet_register_protosw(q);
1989
1990 /*
1991 * Set the ARP module up
1992 */
1993
1994 arp_init();
1995
1996 /*
1997 * Set the IP module up
1998 */
1999
2000 ip_init();
2001
2002 /* Initialise per-cpu ipv4 mibs */
2003 if (init_ipv4_mibs())
2004 panic("%s: Cannot init ipv4 mibs\n", __func__);
2005
2006 /* Setup TCP slab cache for open requests. */
2007 tcp_init();
2008
2009 /* Setup UDP memory threshold */
2010 udp_init();
2011
2012 /* Add UDP-Lite (RFC 3828) */
2013 udplite4_register();
2014
2015 raw_init();
2016
2017 ping_init();
2018
2019 /*
2020 * Set the ICMP layer up
2021 */
2022
2023 if (icmp_init() < 0)
2024 panic("Failed to create the ICMP control socket.\n");
2025
2026 /*
2027 * Initialise the multicast router
2028 */
2029#if defined(CONFIG_IP_MROUTE)
2030 if (ip_mr_init())
2031 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2032#endif
2033
2034 if (init_inet_pernet_ops())
2035 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2036
2037 ipv4_proc_init();
2038
2039 ipfrag_init();
2040
2041 dev_add_pack(&ip_packet_type);
2042
2043 ip_tunnel_core_init();
2044
2045 rc = 0;
2046out:
2047 return rc;
2048out_unregister_raw_proto:
2049 proto_unregister(&raw_prot);
2050out_unregister_udp_proto:
2051 proto_unregister(&udp_prot);
2052out_unregister_tcp_proto:
2053 proto_unregister(&tcp_prot);
2054 goto out;
2055}
2056
2057fs_initcall(inet_init);
2058
2059/* ------------------------------------------------------------------------ */
2060
2061#ifdef CONFIG_PROC_FS
2062static int __init ipv4_proc_init(void)
2063{
2064 int rc = 0;
2065
2066 if (raw_proc_init())
2067 goto out_raw;
2068 if (tcp4_proc_init())
2069 goto out_tcp;
2070 if (udp4_proc_init())
2071 goto out_udp;
2072 if (ping_proc_init())
2073 goto out_ping;
2074 if (ip_misc_proc_init())
2075 goto out_misc;
2076out:
2077 return rc;
2078out_misc:
2079 ping_proc_exit();
2080out_ping:
2081 udp4_proc_exit();
2082out_udp:
2083 tcp4_proc_exit();
2084out_tcp:
2085 raw_proc_exit();
2086out_raw:
2087 rc = -ENOMEM;
2088 goto out;
2089}
2090
2091#else /* CONFIG_PROC_FS */
2092static int __init ipv4_proc_init(void)
2093{
2094 return 0;
2095}
2096#endif /* CONFIG_PROC_FS */
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/tcp.h>
103#include <net/udp.h>
104#include <net/udplite.h>
105#include <net/ping.h>
106#include <linux/skbuff.h>
107#include <net/sock.h>
108#include <net/raw.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>
117#endif
118#include <net/l3mdev.h>
119#include <net/compat.h>
120
121#include <trace/events/sock.h>
122
123/* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
125 */
126static struct list_head inetsw[SOCK_MAX];
127static DEFINE_SPINLOCK(inetsw_lock);
128
129/* New destruction routine */
130
131void inet_sock_destruct(struct sock *sk)
132{
133 struct inet_sock *inet = inet_sk(sk);
134
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;
139 }
140 __skb_queue_purge(&sk->sk_error_queue);
141
142 sk_mem_reclaim(sk);
143
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",
146 sk->sk_state, sk);
147 return;
148 }
149 if (!sock_flag(sk, SOCK_DEAD)) {
150 pr_err("Attempt to release alive inet socket %p\n", sk);
151 return;
152 }
153
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);
158
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);
163}
164EXPORT_SYMBOL(inet_sock_destruct);
165
166/*
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
169 * the work.
170 */
171
172/*
173 * Automatically bind an unbound socket.
174 */
175
176static int inet_autobind(struct sock *sk)
177{
178 struct inet_sock *inet;
179 /* We may need to bind the socket. */
180 lock_sock(sk);
181 inet = inet_sk(sk);
182 if (!inet->inet_num) {
183 if (sk->sk_prot->get_port(sk, 0)) {
184 release_sock(sk);
185 return -EAGAIN;
186 }
187 inet->inet_sport = htons(inet->inet_num);
188 }
189 release_sock(sk);
190 return 0;
191}
192
193/*
194 * Move a socket into listening state.
195 */
196int inet_listen(struct socket *sock, int backlog)
197{
198 struct sock *sk = sock->sk;
199 unsigned char old_state;
200 int err, tcp_fastopen;
201
202 lock_sock(sk);
203
204 err = -EINVAL;
205 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
206 goto out;
207
208 old_state = sk->sk_state;
209 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
210 goto out;
211
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.
215 */
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().
222 */
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));
229 }
230
231 err = inet_csk_listen_start(sk, backlog);
232 if (err)
233 goto out;
234 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
235 }
236 err = 0;
237
238out:
239 release_sock(sk);
240 return err;
241}
242EXPORT_SYMBOL(inet_listen);
243
244/*
245 * Create an inet socket.
246 */
247
248static int inet_create(struct net *net, struct socket *sock, int protocol,
249 int kern)
250{
251 struct sock *sk;
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;
257 int err;
258
259 if (protocol < 0 || protocol >= IPPROTO_MAX)
260 return -EINVAL;
261
262 sock->state = SS_UNCONNECTED;
263
264 /* Look for the requested type/protocol pair. */
265lookup_protocol:
266 err = -ESOCKTNOSUPPORT;
267 rcu_read_lock();
268 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
269
270 err = 0;
271 /* Check the non-wild match. */
272 if (protocol == answer->protocol) {
273 if (protocol != IPPROTO_IP)
274 break;
275 } else {
276 /* Check for the two wild cases. */
277 if (IPPROTO_IP == protocol) {
278 protocol = answer->protocol;
279 break;
280 }
281 if (IPPROTO_IP == answer->protocol)
282 break;
283 }
284 err = -EPROTONOSUPPORT;
285 }
286
287 if (unlikely(err)) {
288 if (try_loading_module < 2) {
289 rcu_read_unlock();
290 /*
291 * Be more specific, e.g. net-pf-2-proto-132-type-1
292 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
293 */
294 if (++try_loading_module == 1)
295 request_module("net-pf-%d-proto-%d-type-%d",
296 PF_INET, protocol, sock->type);
297 /*
298 * Fall back to generic, e.g. net-pf-2-proto-132
299 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
300 */
301 else
302 request_module("net-pf-%d-proto-%d",
303 PF_INET, protocol);
304 goto lookup_protocol;
305 } else
306 goto out_rcu_unlock;
307 }
308
309 err = -EPERM;
310 if (sock->type == SOCK_RAW && !kern &&
311 !ns_capable(net->user_ns, CAP_NET_RAW))
312 goto out_rcu_unlock;
313
314 sock->ops = answer->ops;
315 answer_prot = answer->prot;
316 answer_flags = answer->flags;
317 rcu_read_unlock();
318
319 WARN_ON(!answer_prot->slab);
320
321 err = -ENOMEM;
322 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
323 if (!sk)
324 goto out;
325
326 err = 0;
327 if (INET_PROTOSW_REUSE & answer_flags)
328 sk->sk_reuse = SK_CAN_REUSE;
329
330 inet = inet_sk(sk);
331 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
332
333 inet->nodefrag = 0;
334
335 if (SOCK_RAW == sock->type) {
336 inet->inet_num = protocol;
337 if (IPPROTO_RAW == protocol)
338 inet->hdrincl = 1;
339 }
340
341 if (net->ipv4.sysctl_ip_no_pmtu_disc)
342 inet->pmtudisc = IP_PMTUDISC_DONT;
343 else
344 inet->pmtudisc = IP_PMTUDISC_WANT;
345
346 inet->inet_id = 0;
347
348 sock_init_data(sock, sk);
349
350 sk->sk_destruct = inet_sock_destruct;
351 sk->sk_protocol = protocol;
352 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
353
354 inet->uc_ttl = -1;
355 inet->mc_loop = 1;
356 inet->mc_ttl = 1;
357 inet->mc_all = 1;
358 inet->mc_index = 0;
359 inet->mc_list = NULL;
360 inet->rcv_tos = 0;
361
362 sk_refcnt_debug_inc(sk);
363
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
368 * shares.
369 */
370 inet->inet_sport = htons(inet->inet_num);
371 /* Add to protocol hash chains. */
372 err = sk->sk_prot->hash(sk);
373 if (err) {
374 sk_common_release(sk);
375 goto out;
376 }
377 }
378
379 if (sk->sk_prot->init) {
380 err = sk->sk_prot->init(sk);
381 if (err) {
382 sk_common_release(sk);
383 goto out;
384 }
385 }
386
387 if (!kern) {
388 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
389 if (err) {
390 sk_common_release(sk);
391 goto out;
392 }
393 }
394out:
395 return err;
396out_rcu_unlock:
397 rcu_read_unlock();
398 goto out;
399}
400
401
402/*
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.
406 */
407int inet_release(struct socket *sock)
408{
409 struct sock *sk = sock->sk;
410
411 if (sk) {
412 long timeout;
413
414 if (!sk->sk_kern_sock)
415 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
416
417 /* Applications forget to leave groups before exiting */
418 ip_mc_drop_socket(sk);
419
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.
423 *
424 * If the close is due to the process exiting, we never
425 * linger..
426 */
427 timeout = 0;
428 if (sock_flag(sk, SOCK_LINGER) &&
429 !(current->flags & PF_EXITING))
430 timeout = sk->sk_lingertime;
431 sk->sk_prot->close(sk, timeout);
432 sock->sk = NULL;
433 }
434 return 0;
435}
436EXPORT_SYMBOL(inet_release);
437
438int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
439{
440 struct sock *sk = sock->sk;
441 u32 flags = BIND_WITH_LOCK;
442 int err;
443
444 /* If the socket has its own bind function then use it. (RAW) */
445 if (sk->sk_prot->bind) {
446 return sk->sk_prot->bind(sk, uaddr, addr_len);
447 }
448 if (addr_len < sizeof(struct sockaddr_in))
449 return -EINVAL;
450
451 /* BPF prog is run before any checks are done so that if the prog
452 * changes context in a wrong way it will be caught.
453 */
454 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
455 BPF_CGROUP_INET4_BIND, &flags);
456 if (err)
457 return err;
458
459 return __inet_bind(sk, uaddr, addr_len, flags);
460}
461EXPORT_SYMBOL(inet_bind);
462
463int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
464 u32 flags)
465{
466 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
467 struct inet_sock *inet = inet_sk(sk);
468 struct net *net = sock_net(sk);
469 unsigned short snum;
470 int chk_addr_ret;
471 u32 tb_id = RT_TABLE_LOCAL;
472 int err;
473
474 if (addr->sin_family != AF_INET) {
475 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
476 * only if s_addr is INADDR_ANY.
477 */
478 err = -EAFNOSUPPORT;
479 if (addr->sin_family != AF_UNSPEC ||
480 addr->sin_addr.s_addr != htonl(INADDR_ANY))
481 goto out;
482 }
483
484 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
485 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
486
487 /* Not specified by any standard per-se, however it breaks too
488 * many applications when removed. It is unfortunate since
489 * allowing applications to make a non-local bind solves
490 * several problems with systems using dynamic addressing.
491 * (ie. your servers still start up even if your ISDN link
492 * is temporarily down)
493 */
494 err = -EADDRNOTAVAIL;
495 if (!inet_can_nonlocal_bind(net, inet) &&
496 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
497 chk_addr_ret != RTN_LOCAL &&
498 chk_addr_ret != RTN_MULTICAST &&
499 chk_addr_ret != RTN_BROADCAST)
500 goto out;
501
502 snum = ntohs(addr->sin_port);
503 err = -EACCES;
504 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
505 snum && inet_port_requires_bind_service(net, snum) &&
506 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
507 goto out;
508
509 /* We keep a pair of addresses. rcv_saddr is the one
510 * used by hash lookups, and saddr is used for transmit.
511 *
512 * In the BSD API these are the same except where it
513 * would be illegal to use them (multicast/broadcast) in
514 * which case the sending device address is used.
515 */
516 if (flags & BIND_WITH_LOCK)
517 lock_sock(sk);
518
519 /* Check these errors (active socket, double bind). */
520 err = -EINVAL;
521 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
522 goto out_release_sock;
523
524 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
525 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
526 inet->inet_saddr = 0; /* Use device */
527
528 /* Make sure we are allowed to bind here. */
529 if (snum || !(inet->bind_address_no_port ||
530 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
531 if (sk->sk_prot->get_port(sk, snum)) {
532 inet->inet_saddr = inet->inet_rcv_saddr = 0;
533 err = -EADDRINUSE;
534 goto out_release_sock;
535 }
536 if (!(flags & BIND_FROM_BPF)) {
537 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
538 if (err) {
539 inet->inet_saddr = inet->inet_rcv_saddr = 0;
540 goto out_release_sock;
541 }
542 }
543 }
544
545 if (inet->inet_rcv_saddr)
546 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
547 if (snum)
548 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
549 inet->inet_sport = htons(inet->inet_num);
550 inet->inet_daddr = 0;
551 inet->inet_dport = 0;
552 sk_dst_reset(sk);
553 err = 0;
554out_release_sock:
555 if (flags & BIND_WITH_LOCK)
556 release_sock(sk);
557out:
558 return err;
559}
560
561int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
562 int addr_len, int flags)
563{
564 struct sock *sk = sock->sk;
565 int err;
566
567 if (addr_len < sizeof(uaddr->sa_family))
568 return -EINVAL;
569 if (uaddr->sa_family == AF_UNSPEC)
570 return sk->sk_prot->disconnect(sk, flags);
571
572 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
573 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
574 if (err)
575 return err;
576 }
577
578 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
579 return -EAGAIN;
580 return sk->sk_prot->connect(sk, uaddr, addr_len);
581}
582EXPORT_SYMBOL(inet_dgram_connect);
583
584static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
585{
586 DEFINE_WAIT_FUNC(wait, woken_wake_function);
587
588 add_wait_queue(sk_sleep(sk), &wait);
589 sk->sk_write_pending += writebias;
590
591 /* Basic assumption: if someone sets sk->sk_err, he _must_
592 * change state of the socket from TCP_SYN_*.
593 * Connect() does not allow to get error notifications
594 * without closing the socket.
595 */
596 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
597 release_sock(sk);
598 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
599 lock_sock(sk);
600 if (signal_pending(current) || !timeo)
601 break;
602 }
603 remove_wait_queue(sk_sleep(sk), &wait);
604 sk->sk_write_pending -= writebias;
605 return timeo;
606}
607
608/*
609 * Connect to a remote host. There is regrettably still a little
610 * TCP 'magic' in here.
611 */
612int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
613 int addr_len, int flags, int is_sendmsg)
614{
615 struct sock *sk = sock->sk;
616 int err;
617 long timeo;
618
619 /*
620 * uaddr can be NULL and addr_len can be 0 if:
621 * sk is a TCP fastopen active socket and
622 * TCP_FASTOPEN_CONNECT sockopt is set and
623 * we already have a valid cookie for this socket.
624 * In this case, user can call write() after connect().
625 * write() will invoke tcp_sendmsg_fastopen() which calls
626 * __inet_stream_connect().
627 */
628 if (uaddr) {
629 if (addr_len < sizeof(uaddr->sa_family))
630 return -EINVAL;
631
632 if (uaddr->sa_family == AF_UNSPEC) {
633 err = sk->sk_prot->disconnect(sk, flags);
634 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
635 goto out;
636 }
637 }
638
639 switch (sock->state) {
640 default:
641 err = -EINVAL;
642 goto out;
643 case SS_CONNECTED:
644 err = -EISCONN;
645 goto out;
646 case SS_CONNECTING:
647 if (inet_sk(sk)->defer_connect)
648 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
649 else
650 err = -EALREADY;
651 /* Fall out of switch with err, set for this state */
652 break;
653 case SS_UNCONNECTED:
654 err = -EISCONN;
655 if (sk->sk_state != TCP_CLOSE)
656 goto out;
657
658 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
659 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
660 if (err)
661 goto out;
662 }
663
664 err = sk->sk_prot->connect(sk, uaddr, addr_len);
665 if (err < 0)
666 goto out;
667
668 sock->state = SS_CONNECTING;
669
670 if (!err && inet_sk(sk)->defer_connect)
671 goto out;
672
673 /* Just entered SS_CONNECTING state; the only
674 * difference is that return value in non-blocking
675 * case is EINPROGRESS, rather than EALREADY.
676 */
677 err = -EINPROGRESS;
678 break;
679 }
680
681 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
682
683 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
684 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
685 tcp_sk(sk)->fastopen_req &&
686 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
687
688 /* Error code is set above */
689 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
690 goto out;
691
692 err = sock_intr_errno(timeo);
693 if (signal_pending(current))
694 goto out;
695 }
696
697 /* Connection was closed by RST, timeout, ICMP error
698 * or another process disconnected us.
699 */
700 if (sk->sk_state == TCP_CLOSE)
701 goto sock_error;
702
703 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
704 * and error was received after socket entered established state.
705 * Hence, it is handled normally after connect() return successfully.
706 */
707
708 sock->state = SS_CONNECTED;
709 err = 0;
710out:
711 return err;
712
713sock_error:
714 err = sock_error(sk) ? : -ECONNABORTED;
715 sock->state = SS_UNCONNECTED;
716 if (sk->sk_prot->disconnect(sk, flags))
717 sock->state = SS_DISCONNECTING;
718 goto out;
719}
720EXPORT_SYMBOL(__inet_stream_connect);
721
722int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
723 int addr_len, int flags)
724{
725 int err;
726
727 lock_sock(sock->sk);
728 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
729 release_sock(sock->sk);
730 return err;
731}
732EXPORT_SYMBOL(inet_stream_connect);
733
734/*
735 * Accept a pending connection. The TCP layer now gives BSD semantics.
736 */
737
738int inet_accept(struct socket *sock, struct socket *newsock, int flags,
739 bool kern)
740{
741 struct sock *sk1 = sock->sk;
742 int err = -EINVAL;
743 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
744
745 if (!sk2)
746 goto do_err;
747
748 lock_sock(sk2);
749
750 sock_rps_record_flow(sk2);
751 WARN_ON(!((1 << sk2->sk_state) &
752 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
753 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
754
755 sock_graft(sk2, newsock);
756
757 newsock->state = SS_CONNECTED;
758 err = 0;
759 release_sock(sk2);
760do_err:
761 return err;
762}
763EXPORT_SYMBOL(inet_accept);
764
765/*
766 * This does both peername and sockname.
767 */
768int inet_getname(struct socket *sock, struct sockaddr *uaddr,
769 int peer)
770{
771 struct sock *sk = sock->sk;
772 struct inet_sock *inet = inet_sk(sk);
773 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
774
775 sin->sin_family = AF_INET;
776 if (peer) {
777 if (!inet->inet_dport ||
778 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
779 peer == 1))
780 return -ENOTCONN;
781 sin->sin_port = inet->inet_dport;
782 sin->sin_addr.s_addr = inet->inet_daddr;
783 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
784 BPF_CGROUP_INET4_GETPEERNAME,
785 NULL);
786 } else {
787 __be32 addr = inet->inet_rcv_saddr;
788 if (!addr)
789 addr = inet->inet_saddr;
790 sin->sin_port = inet->inet_sport;
791 sin->sin_addr.s_addr = addr;
792 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
793 BPF_CGROUP_INET4_GETSOCKNAME,
794 NULL);
795 }
796 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
797 return sizeof(*sin);
798}
799EXPORT_SYMBOL(inet_getname);
800
801int inet_send_prepare(struct sock *sk)
802{
803 sock_rps_record_flow(sk);
804
805 /* We may need to bind the socket. */
806 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
807 inet_autobind(sk))
808 return -EAGAIN;
809
810 return 0;
811}
812EXPORT_SYMBOL_GPL(inet_send_prepare);
813
814int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
815{
816 struct sock *sk = sock->sk;
817
818 if (unlikely(inet_send_prepare(sk)))
819 return -EAGAIN;
820
821 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
822 sk, msg, size);
823}
824EXPORT_SYMBOL(inet_sendmsg);
825
826ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
827 size_t size, int flags)
828{
829 struct sock *sk = sock->sk;
830
831 if (unlikely(inet_send_prepare(sk)))
832 return -EAGAIN;
833
834 if (sk->sk_prot->sendpage)
835 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
836 return sock_no_sendpage(sock, page, offset, size, flags);
837}
838EXPORT_SYMBOL(inet_sendpage);
839
840INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
841 size_t, int, int, int *));
842int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
843 int flags)
844{
845 struct sock *sk = sock->sk;
846 int addr_len = 0;
847 int err;
848
849 if (likely(!(flags & MSG_ERRQUEUE)))
850 sock_rps_record_flow(sk);
851
852 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
853 sk, msg, size, flags & MSG_DONTWAIT,
854 flags & ~MSG_DONTWAIT, &addr_len);
855 if (err >= 0)
856 msg->msg_namelen = addr_len;
857 return err;
858}
859EXPORT_SYMBOL(inet_recvmsg);
860
861int inet_shutdown(struct socket *sock, int how)
862{
863 struct sock *sk = sock->sk;
864 int err = 0;
865
866 /* This should really check to make sure
867 * the socket is a TCP socket. (WHY AC...)
868 */
869 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
870 1->2 bit 2 snds.
871 2->3 */
872 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
873 return -EINVAL;
874
875 lock_sock(sk);
876 if (sock->state == SS_CONNECTING) {
877 if ((1 << sk->sk_state) &
878 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
879 sock->state = SS_DISCONNECTING;
880 else
881 sock->state = SS_CONNECTED;
882 }
883
884 switch (sk->sk_state) {
885 case TCP_CLOSE:
886 err = -ENOTCONN;
887 /* Hack to wake up other listeners, who can poll for
888 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
889 fallthrough;
890 default:
891 sk->sk_shutdown |= how;
892 if (sk->sk_prot->shutdown)
893 sk->sk_prot->shutdown(sk, how);
894 break;
895
896 /* Remaining two branches are temporary solution for missing
897 * close() in multithreaded environment. It is _not_ a good idea,
898 * but we have no choice until close() is repaired at VFS level.
899 */
900 case TCP_LISTEN:
901 if (!(how & RCV_SHUTDOWN))
902 break;
903 fallthrough;
904 case TCP_SYN_SENT:
905 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
906 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
907 break;
908 }
909
910 /* Wake up anyone sleeping in poll. */
911 sk->sk_state_change(sk);
912 release_sock(sk);
913 return err;
914}
915EXPORT_SYMBOL(inet_shutdown);
916
917/*
918 * ioctl() calls you can issue on an INET socket. Most of these are
919 * device configuration and stuff and very rarely used. Some ioctls
920 * pass on to the socket itself.
921 *
922 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
923 * loads the devconfigure module does its configuring and unloads it.
924 * There's a good 20K of config code hanging around the kernel.
925 */
926
927int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
928{
929 struct sock *sk = sock->sk;
930 int err = 0;
931 struct net *net = sock_net(sk);
932 void __user *p = (void __user *)arg;
933 struct ifreq ifr;
934 struct rtentry rt;
935
936 switch (cmd) {
937 case SIOCADDRT:
938 case SIOCDELRT:
939 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
940 return -EFAULT;
941 err = ip_rt_ioctl(net, cmd, &rt);
942 break;
943 case SIOCRTMSG:
944 err = -EINVAL;
945 break;
946 case SIOCDARP:
947 case SIOCGARP:
948 case SIOCSARP:
949 err = arp_ioctl(net, cmd, (void __user *)arg);
950 break;
951 case SIOCGIFADDR:
952 case SIOCGIFBRDADDR:
953 case SIOCGIFNETMASK:
954 case SIOCGIFDSTADDR:
955 case SIOCGIFPFLAGS:
956 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
957 return -EFAULT;
958 err = devinet_ioctl(net, cmd, &ifr);
959 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
960 err = -EFAULT;
961 break;
962
963 case SIOCSIFADDR:
964 case SIOCSIFBRDADDR:
965 case SIOCSIFNETMASK:
966 case SIOCSIFDSTADDR:
967 case SIOCSIFPFLAGS:
968 case SIOCSIFFLAGS:
969 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
970 return -EFAULT;
971 err = devinet_ioctl(net, cmd, &ifr);
972 break;
973 default:
974 if (sk->sk_prot->ioctl)
975 err = sk->sk_prot->ioctl(sk, cmd, arg);
976 else
977 err = -ENOIOCTLCMD;
978 break;
979 }
980 return err;
981}
982EXPORT_SYMBOL(inet_ioctl);
983
984#ifdef CONFIG_COMPAT
985static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
986 struct compat_rtentry __user *ur)
987{
988 compat_uptr_t rtdev;
989 struct rtentry rt;
990
991 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
992 3 * sizeof(struct sockaddr)) ||
993 get_user(rt.rt_flags, &ur->rt_flags) ||
994 get_user(rt.rt_metric, &ur->rt_metric) ||
995 get_user(rt.rt_mtu, &ur->rt_mtu) ||
996 get_user(rt.rt_window, &ur->rt_window) ||
997 get_user(rt.rt_irtt, &ur->rt_irtt) ||
998 get_user(rtdev, &ur->rt_dev))
999 return -EFAULT;
1000
1001 rt.rt_dev = compat_ptr(rtdev);
1002 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1003}
1004
1005static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1006{
1007 void __user *argp = compat_ptr(arg);
1008 struct sock *sk = sock->sk;
1009
1010 switch (cmd) {
1011 case SIOCADDRT:
1012 case SIOCDELRT:
1013 return inet_compat_routing_ioctl(sk, cmd, argp);
1014 default:
1015 if (!sk->sk_prot->compat_ioctl)
1016 return -ENOIOCTLCMD;
1017 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1018 }
1019}
1020#endif /* CONFIG_COMPAT */
1021
1022const struct proto_ops inet_stream_ops = {
1023 .family = PF_INET,
1024 .owner = THIS_MODULE,
1025 .release = inet_release,
1026 .bind = inet_bind,
1027 .connect = inet_stream_connect,
1028 .socketpair = sock_no_socketpair,
1029 .accept = inet_accept,
1030 .getname = inet_getname,
1031 .poll = tcp_poll,
1032 .ioctl = inet_ioctl,
1033 .gettstamp = sock_gettstamp,
1034 .listen = inet_listen,
1035 .shutdown = inet_shutdown,
1036 .setsockopt = sock_common_setsockopt,
1037 .getsockopt = sock_common_getsockopt,
1038 .sendmsg = inet_sendmsg,
1039 .recvmsg = inet_recvmsg,
1040#ifdef CONFIG_MMU
1041 .mmap = tcp_mmap,
1042#endif
1043 .sendpage = inet_sendpage,
1044 .splice_read = tcp_splice_read,
1045 .read_sock = tcp_read_sock,
1046 .sendmsg_locked = tcp_sendmsg_locked,
1047 .sendpage_locked = tcp_sendpage_locked,
1048 .peek_len = tcp_peek_len,
1049#ifdef CONFIG_COMPAT
1050 .compat_ioctl = inet_compat_ioctl,
1051#endif
1052 .set_rcvlowat = tcp_set_rcvlowat,
1053};
1054EXPORT_SYMBOL(inet_stream_ops);
1055
1056const struct proto_ops inet_dgram_ops = {
1057 .family = PF_INET,
1058 .owner = THIS_MODULE,
1059 .release = inet_release,
1060 .bind = inet_bind,
1061 .connect = inet_dgram_connect,
1062 .socketpair = sock_no_socketpair,
1063 .accept = sock_no_accept,
1064 .getname = inet_getname,
1065 .poll = udp_poll,
1066 .ioctl = inet_ioctl,
1067 .gettstamp = sock_gettstamp,
1068 .listen = sock_no_listen,
1069 .shutdown = inet_shutdown,
1070 .setsockopt = sock_common_setsockopt,
1071 .getsockopt = sock_common_getsockopt,
1072 .sendmsg = inet_sendmsg,
1073 .read_sock = udp_read_sock,
1074 .recvmsg = inet_recvmsg,
1075 .mmap = sock_no_mmap,
1076 .sendpage = inet_sendpage,
1077 .set_peek_off = sk_set_peek_off,
1078#ifdef CONFIG_COMPAT
1079 .compat_ioctl = inet_compat_ioctl,
1080#endif
1081};
1082EXPORT_SYMBOL(inet_dgram_ops);
1083
1084/*
1085 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1086 * udp_poll
1087 */
1088static const struct proto_ops inet_sockraw_ops = {
1089 .family = PF_INET,
1090 .owner = THIS_MODULE,
1091 .release = inet_release,
1092 .bind = inet_bind,
1093 .connect = inet_dgram_connect,
1094 .socketpair = sock_no_socketpair,
1095 .accept = sock_no_accept,
1096 .getname = inet_getname,
1097 .poll = datagram_poll,
1098 .ioctl = inet_ioctl,
1099 .gettstamp = sock_gettstamp,
1100 .listen = sock_no_listen,
1101 .shutdown = inet_shutdown,
1102 .setsockopt = sock_common_setsockopt,
1103 .getsockopt = sock_common_getsockopt,
1104 .sendmsg = inet_sendmsg,
1105 .recvmsg = inet_recvmsg,
1106 .mmap = sock_no_mmap,
1107 .sendpage = inet_sendpage,
1108#ifdef CONFIG_COMPAT
1109 .compat_ioctl = inet_compat_ioctl,
1110#endif
1111};
1112
1113static const struct net_proto_family inet_family_ops = {
1114 .family = PF_INET,
1115 .create = inet_create,
1116 .owner = THIS_MODULE,
1117};
1118
1119/* Upon startup we insert all the elements in inetsw_array[] into
1120 * the linked list inetsw.
1121 */
1122static struct inet_protosw inetsw_array[] =
1123{
1124 {
1125 .type = SOCK_STREAM,
1126 .protocol = IPPROTO_TCP,
1127 .prot = &tcp_prot,
1128 .ops = &inet_stream_ops,
1129 .flags = INET_PROTOSW_PERMANENT |
1130 INET_PROTOSW_ICSK,
1131 },
1132
1133 {
1134 .type = SOCK_DGRAM,
1135 .protocol = IPPROTO_UDP,
1136 .prot = &udp_prot,
1137 .ops = &inet_dgram_ops,
1138 .flags = INET_PROTOSW_PERMANENT,
1139 },
1140
1141 {
1142 .type = SOCK_DGRAM,
1143 .protocol = IPPROTO_ICMP,
1144 .prot = &ping_prot,
1145 .ops = &inet_sockraw_ops,
1146 .flags = INET_PROTOSW_REUSE,
1147 },
1148
1149 {
1150 .type = SOCK_RAW,
1151 .protocol = IPPROTO_IP, /* wild card */
1152 .prot = &raw_prot,
1153 .ops = &inet_sockraw_ops,
1154 .flags = INET_PROTOSW_REUSE,
1155 }
1156};
1157
1158#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1159
1160void inet_register_protosw(struct inet_protosw *p)
1161{
1162 struct list_head *lh;
1163 struct inet_protosw *answer;
1164 int protocol = p->protocol;
1165 struct list_head *last_perm;
1166
1167 spin_lock_bh(&inetsw_lock);
1168
1169 if (p->type >= SOCK_MAX)
1170 goto out_illegal;
1171
1172 /* If we are trying to override a permanent protocol, bail. */
1173 last_perm = &inetsw[p->type];
1174 list_for_each(lh, &inetsw[p->type]) {
1175 answer = list_entry(lh, struct inet_protosw, list);
1176 /* Check only the non-wild match. */
1177 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1178 break;
1179 if (protocol == answer->protocol)
1180 goto out_permanent;
1181 last_perm = lh;
1182 }
1183
1184 /* Add the new entry after the last permanent entry if any, so that
1185 * the new entry does not override a permanent entry when matched with
1186 * a wild-card protocol. But it is allowed to override any existing
1187 * non-permanent entry. This means that when we remove this entry, the
1188 * system automatically returns to the old behavior.
1189 */
1190 list_add_rcu(&p->list, last_perm);
1191out:
1192 spin_unlock_bh(&inetsw_lock);
1193
1194 return;
1195
1196out_permanent:
1197 pr_err("Attempt to override permanent protocol %d\n", protocol);
1198 goto out;
1199
1200out_illegal:
1201 pr_err("Ignoring attempt to register invalid socket type %d\n",
1202 p->type);
1203 goto out;
1204}
1205EXPORT_SYMBOL(inet_register_protosw);
1206
1207void inet_unregister_protosw(struct inet_protosw *p)
1208{
1209 if (INET_PROTOSW_PERMANENT & p->flags) {
1210 pr_err("Attempt to unregister permanent protocol %d\n",
1211 p->protocol);
1212 } else {
1213 spin_lock_bh(&inetsw_lock);
1214 list_del_rcu(&p->list);
1215 spin_unlock_bh(&inetsw_lock);
1216
1217 synchronize_net();
1218 }
1219}
1220EXPORT_SYMBOL(inet_unregister_protosw);
1221
1222static int inet_sk_reselect_saddr(struct sock *sk)
1223{
1224 struct inet_sock *inet = inet_sk(sk);
1225 __be32 old_saddr = inet->inet_saddr;
1226 __be32 daddr = inet->inet_daddr;
1227 struct flowi4 *fl4;
1228 struct rtable *rt;
1229 __be32 new_saddr;
1230 struct ip_options_rcu *inet_opt;
1231
1232 inet_opt = rcu_dereference_protected(inet->inet_opt,
1233 lockdep_sock_is_held(sk));
1234 if (inet_opt && inet_opt->opt.srr)
1235 daddr = inet_opt->opt.faddr;
1236
1237 /* Query new route. */
1238 fl4 = &inet->cork.fl.u.ip4;
1239 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1240 sk->sk_bound_dev_if, sk->sk_protocol,
1241 inet->inet_sport, inet->inet_dport, sk);
1242 if (IS_ERR(rt))
1243 return PTR_ERR(rt);
1244
1245 sk_setup_caps(sk, &rt->dst);
1246
1247 new_saddr = fl4->saddr;
1248
1249 if (new_saddr == old_saddr)
1250 return 0;
1251
1252 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1253 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1254 __func__, &old_saddr, &new_saddr);
1255 }
1256
1257 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1258
1259 /*
1260 * XXX The only one ugly spot where we need to
1261 * XXX really change the sockets identity after
1262 * XXX it has entered the hashes. -DaveM
1263 *
1264 * Besides that, it does not check for connection
1265 * uniqueness. Wait for troubles.
1266 */
1267 return __sk_prot_rehash(sk);
1268}
1269
1270int inet_sk_rebuild_header(struct sock *sk)
1271{
1272 struct inet_sock *inet = inet_sk(sk);
1273 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1274 __be32 daddr;
1275 struct ip_options_rcu *inet_opt;
1276 struct flowi4 *fl4;
1277 int err;
1278
1279 /* Route is OK, nothing to do. */
1280 if (rt)
1281 return 0;
1282
1283 /* Reroute. */
1284 rcu_read_lock();
1285 inet_opt = rcu_dereference(inet->inet_opt);
1286 daddr = inet->inet_daddr;
1287 if (inet_opt && inet_opt->opt.srr)
1288 daddr = inet_opt->opt.faddr;
1289 rcu_read_unlock();
1290 fl4 = &inet->cork.fl.u.ip4;
1291 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1292 inet->inet_dport, inet->inet_sport,
1293 sk->sk_protocol, RT_CONN_FLAGS(sk),
1294 sk->sk_bound_dev_if);
1295 if (!IS_ERR(rt)) {
1296 err = 0;
1297 sk_setup_caps(sk, &rt->dst);
1298 } else {
1299 err = PTR_ERR(rt);
1300
1301 /* Routing failed... */
1302 sk->sk_route_caps = 0;
1303 /*
1304 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1305 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1306 */
1307 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1308 sk->sk_state != TCP_SYN_SENT ||
1309 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1310 (err = inet_sk_reselect_saddr(sk)) != 0)
1311 sk->sk_err_soft = -err;
1312 }
1313
1314 return err;
1315}
1316EXPORT_SYMBOL(inet_sk_rebuild_header);
1317
1318void inet_sk_set_state(struct sock *sk, int state)
1319{
1320 trace_inet_sock_set_state(sk, sk->sk_state, state);
1321 sk->sk_state = state;
1322}
1323EXPORT_SYMBOL(inet_sk_set_state);
1324
1325void inet_sk_state_store(struct sock *sk, int newstate)
1326{
1327 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1328 smp_store_release(&sk->sk_state, newstate);
1329}
1330
1331struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1332 netdev_features_t features)
1333{
1334 bool udpfrag = false, fixedid = false, gso_partial, encap;
1335 struct sk_buff *segs = ERR_PTR(-EINVAL);
1336 const struct net_offload *ops;
1337 unsigned int offset = 0;
1338 struct iphdr *iph;
1339 int proto, tot_len;
1340 int nhoff;
1341 int ihl;
1342 int id;
1343
1344 skb_reset_network_header(skb);
1345 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1346 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1347 goto out;
1348
1349 iph = ip_hdr(skb);
1350 ihl = iph->ihl * 4;
1351 if (ihl < sizeof(*iph))
1352 goto out;
1353
1354 id = ntohs(iph->id);
1355 proto = iph->protocol;
1356
1357 /* Warning: after this point, iph might be no longer valid */
1358 if (unlikely(!pskb_may_pull(skb, ihl)))
1359 goto out;
1360 __skb_pull(skb, ihl);
1361
1362 encap = SKB_GSO_CB(skb)->encap_level > 0;
1363 if (encap)
1364 features &= skb->dev->hw_enc_features;
1365 SKB_GSO_CB(skb)->encap_level += ihl;
1366
1367 skb_reset_transport_header(skb);
1368
1369 segs = ERR_PTR(-EPROTONOSUPPORT);
1370
1371 if (!skb->encapsulation || encap) {
1372 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1373 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1374
1375 /* fixed ID is invalid if DF bit is not set */
1376 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1377 goto out;
1378 }
1379
1380 ops = rcu_dereference(inet_offloads[proto]);
1381 if (likely(ops && ops->callbacks.gso_segment))
1382 segs = ops->callbacks.gso_segment(skb, features);
1383
1384 if (IS_ERR_OR_NULL(segs))
1385 goto out;
1386
1387 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1388
1389 skb = segs;
1390 do {
1391 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1392 if (udpfrag) {
1393 iph->frag_off = htons(offset >> 3);
1394 if (skb->next)
1395 iph->frag_off |= htons(IP_MF);
1396 offset += skb->len - nhoff - ihl;
1397 tot_len = skb->len - nhoff;
1398 } else if (skb_is_gso(skb)) {
1399 if (!fixedid) {
1400 iph->id = htons(id);
1401 id += skb_shinfo(skb)->gso_segs;
1402 }
1403
1404 if (gso_partial)
1405 tot_len = skb_shinfo(skb)->gso_size +
1406 SKB_GSO_CB(skb)->data_offset +
1407 skb->head - (unsigned char *)iph;
1408 else
1409 tot_len = skb->len - nhoff;
1410 } else {
1411 if (!fixedid)
1412 iph->id = htons(id++);
1413 tot_len = skb->len - nhoff;
1414 }
1415 iph->tot_len = htons(tot_len);
1416 ip_send_check(iph);
1417 if (encap)
1418 skb_reset_inner_headers(skb);
1419 skb->network_header = (u8 *)iph - skb->head;
1420 skb_reset_mac_len(skb);
1421 } while ((skb = skb->next));
1422
1423out:
1424 return segs;
1425}
1426
1427static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1428 netdev_features_t features)
1429{
1430 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1431 return ERR_PTR(-EINVAL);
1432
1433 return inet_gso_segment(skb, features);
1434}
1435
1436struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1437{
1438 const struct net_offload *ops;
1439 struct sk_buff *pp = NULL;
1440 const struct iphdr *iph;
1441 struct sk_buff *p;
1442 unsigned int hlen;
1443 unsigned int off;
1444 unsigned int id;
1445 int flush = 1;
1446 int proto;
1447
1448 off = skb_gro_offset(skb);
1449 hlen = off + sizeof(*iph);
1450 iph = skb_gro_header_fast(skb, off);
1451 if (skb_gro_header_hard(skb, hlen)) {
1452 iph = skb_gro_header_slow(skb, hlen, off);
1453 if (unlikely(!iph))
1454 goto out;
1455 }
1456
1457 proto = iph->protocol;
1458
1459 rcu_read_lock();
1460 ops = rcu_dereference(inet_offloads[proto]);
1461 if (!ops || !ops->callbacks.gro_receive)
1462 goto out_unlock;
1463
1464 if (*(u8 *)iph != 0x45)
1465 goto out_unlock;
1466
1467 if (ip_is_fragment(iph))
1468 goto out_unlock;
1469
1470 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1471 goto out_unlock;
1472
1473 id = ntohl(*(__be32 *)&iph->id);
1474 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1475 id >>= 16;
1476
1477 list_for_each_entry(p, head, list) {
1478 struct iphdr *iph2;
1479 u16 flush_id;
1480
1481 if (!NAPI_GRO_CB(p)->same_flow)
1482 continue;
1483
1484 iph2 = (struct iphdr *)(p->data + off);
1485 /* The above works because, with the exception of the top
1486 * (inner most) layer, we only aggregate pkts with the same
1487 * hdr length so all the hdrs we'll need to verify will start
1488 * at the same offset.
1489 */
1490 if ((iph->protocol ^ iph2->protocol) |
1491 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1492 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1493 NAPI_GRO_CB(p)->same_flow = 0;
1494 continue;
1495 }
1496
1497 /* All fields must match except length and checksum. */
1498 NAPI_GRO_CB(p)->flush |=
1499 (iph->ttl ^ iph2->ttl) |
1500 (iph->tos ^ iph2->tos) |
1501 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1502
1503 NAPI_GRO_CB(p)->flush |= flush;
1504
1505 /* We need to store of the IP ID check to be included later
1506 * when we can verify that this packet does in fact belong
1507 * to a given flow.
1508 */
1509 flush_id = (u16)(id - ntohs(iph2->id));
1510
1511 /* This bit of code makes it much easier for us to identify
1512 * the cases where we are doing atomic vs non-atomic IP ID
1513 * checks. Specifically an atomic check can return IP ID
1514 * values 0 - 0xFFFF, while a non-atomic check can only
1515 * return 0 or 0xFFFF.
1516 */
1517 if (!NAPI_GRO_CB(p)->is_atomic ||
1518 !(iph->frag_off & htons(IP_DF))) {
1519 flush_id ^= NAPI_GRO_CB(p)->count;
1520 flush_id = flush_id ? 0xFFFF : 0;
1521 }
1522
1523 /* If the previous IP ID value was based on an atomic
1524 * datagram we can overwrite the value and ignore it.
1525 */
1526 if (NAPI_GRO_CB(skb)->is_atomic)
1527 NAPI_GRO_CB(p)->flush_id = flush_id;
1528 else
1529 NAPI_GRO_CB(p)->flush_id |= flush_id;
1530 }
1531
1532 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1533 NAPI_GRO_CB(skb)->flush |= flush;
1534 skb_set_network_header(skb, off);
1535 /* The above will be needed by the transport layer if there is one
1536 * immediately following this IP hdr.
1537 */
1538
1539 /* Note : No need to call skb_gro_postpull_rcsum() here,
1540 * as we already checked checksum over ipv4 header was 0
1541 */
1542 skb_gro_pull(skb, sizeof(*iph));
1543 skb_set_transport_header(skb, skb_gro_offset(skb));
1544
1545 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1546 ops->callbacks.gro_receive, head, skb);
1547
1548out_unlock:
1549 rcu_read_unlock();
1550
1551out:
1552 skb_gro_flush_final(skb, pp, flush);
1553
1554 return pp;
1555}
1556
1557static struct sk_buff *ipip_gro_receive(struct list_head *head,
1558 struct sk_buff *skb)
1559{
1560 if (NAPI_GRO_CB(skb)->encap_mark) {
1561 NAPI_GRO_CB(skb)->flush = 1;
1562 return NULL;
1563 }
1564
1565 NAPI_GRO_CB(skb)->encap_mark = 1;
1566
1567 return inet_gro_receive(head, skb);
1568}
1569
1570#define SECONDS_PER_DAY 86400
1571
1572/* inet_current_timestamp - Return IP network timestamp
1573 *
1574 * Return milliseconds since midnight in network byte order.
1575 */
1576__be32 inet_current_timestamp(void)
1577{
1578 u32 secs;
1579 u32 msecs;
1580 struct timespec64 ts;
1581
1582 ktime_get_real_ts64(&ts);
1583
1584 /* Get secs since midnight. */
1585 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1586 /* Convert to msecs. */
1587 msecs = secs * MSEC_PER_SEC;
1588 /* Convert nsec to msec. */
1589 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1590
1591 /* Convert to network byte order. */
1592 return htonl(msecs);
1593}
1594EXPORT_SYMBOL(inet_current_timestamp);
1595
1596int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1597{
1598 if (sk->sk_family == AF_INET)
1599 return ip_recv_error(sk, msg, len, addr_len);
1600#if IS_ENABLED(CONFIG_IPV6)
1601 if (sk->sk_family == AF_INET6)
1602 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1603#endif
1604 return -EINVAL;
1605}
1606
1607int inet_gro_complete(struct sk_buff *skb, int nhoff)
1608{
1609 __be16 newlen = htons(skb->len - nhoff);
1610 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1611 const struct net_offload *ops;
1612 int proto = iph->protocol;
1613 int err = -ENOSYS;
1614
1615 if (skb->encapsulation) {
1616 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1617 skb_set_inner_network_header(skb, nhoff);
1618 }
1619
1620 csum_replace2(&iph->check, iph->tot_len, newlen);
1621 iph->tot_len = newlen;
1622
1623 rcu_read_lock();
1624 ops = rcu_dereference(inet_offloads[proto]);
1625 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1626 goto out_unlock;
1627
1628 /* Only need to add sizeof(*iph) to get to the next hdr below
1629 * because any hdr with option will have been flushed in
1630 * inet_gro_receive().
1631 */
1632 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1633 tcp4_gro_complete, udp4_gro_complete,
1634 skb, nhoff + sizeof(*iph));
1635
1636out_unlock:
1637 rcu_read_unlock();
1638
1639 return err;
1640}
1641
1642static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1643{
1644 skb->encapsulation = 1;
1645 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1646 return inet_gro_complete(skb, nhoff);
1647}
1648
1649int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1650 unsigned short type, unsigned char protocol,
1651 struct net *net)
1652{
1653 struct socket *sock;
1654 int rc = sock_create_kern(net, family, type, protocol, &sock);
1655
1656 if (rc == 0) {
1657 *sk = sock->sk;
1658 (*sk)->sk_allocation = GFP_ATOMIC;
1659 /*
1660 * Unhash it so that IP input processing does not even see it,
1661 * we do not wish this socket to see incoming packets.
1662 */
1663 (*sk)->sk_prot->unhash(*sk);
1664 }
1665 return rc;
1666}
1667EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1668
1669u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1670{
1671 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1672}
1673EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1674
1675unsigned long snmp_fold_field(void __percpu *mib, int offt)
1676{
1677 unsigned long res = 0;
1678 int i;
1679
1680 for_each_possible_cpu(i)
1681 res += snmp_get_cpu_field(mib, i, offt);
1682 return res;
1683}
1684EXPORT_SYMBOL_GPL(snmp_fold_field);
1685
1686#if BITS_PER_LONG==32
1687
1688u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1689 size_t syncp_offset)
1690{
1691 void *bhptr;
1692 struct u64_stats_sync *syncp;
1693 u64 v;
1694 unsigned int start;
1695
1696 bhptr = per_cpu_ptr(mib, cpu);
1697 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1698 do {
1699 start = u64_stats_fetch_begin_irq(syncp);
1700 v = *(((u64 *)bhptr) + offt);
1701 } while (u64_stats_fetch_retry_irq(syncp, start));
1702
1703 return v;
1704}
1705EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1706
1707u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1708{
1709 u64 res = 0;
1710 int cpu;
1711
1712 for_each_possible_cpu(cpu) {
1713 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1714 }
1715 return res;
1716}
1717EXPORT_SYMBOL_GPL(snmp_fold_field64);
1718#endif
1719
1720#ifdef CONFIG_IP_MULTICAST
1721static const struct net_protocol igmp_protocol = {
1722 .handler = igmp_rcv,
1723};
1724#endif
1725
1726/* thinking of making this const? Don't.
1727 * early_demux can change based on sysctl.
1728 */
1729static 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,
1734 .no_policy = 1,
1735 .icmp_strict_tag_validation = 1,
1736};
1737
1738/* thinking of making this const? Don't.
1739 * early_demux can change based on sysctl.
1740 */
1741static struct net_protocol udp_protocol = {
1742 .early_demux = udp_v4_early_demux,
1743 .early_demux_handler = udp_v4_early_demux,
1744 .handler = udp_rcv,
1745 .err_handler = udp_err,
1746 .no_policy = 1,
1747};
1748
1749static const struct net_protocol icmp_protocol = {
1750 .handler = icmp_rcv,
1751 .err_handler = icmp_err,
1752 .no_policy = 1,
1753};
1754
1755static __net_init int ipv4_mib_init_net(struct net *net)
1756{
1757 int i;
1758
1759 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1760 if (!net->mib.tcp_statistics)
1761 goto err_tcp_mib;
1762 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1763 if (!net->mib.ip_statistics)
1764 goto err_ip_mib;
1765
1766 for_each_possible_cpu(i) {
1767 struct ipstats_mib *af_inet_stats;
1768 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1769 u64_stats_init(&af_inet_stats->syncp);
1770 }
1771
1772 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1773 if (!net->mib.net_statistics)
1774 goto err_net_mib;
1775 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1776 if (!net->mib.udp_statistics)
1777 goto err_udp_mib;
1778 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1779 if (!net->mib.udplite_statistics)
1780 goto err_udplite_mib;
1781 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1782 if (!net->mib.icmp_statistics)
1783 goto err_icmp_mib;
1784 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1785 GFP_KERNEL);
1786 if (!net->mib.icmpmsg_statistics)
1787 goto err_icmpmsg_mib;
1788
1789 tcp_mib_init(net);
1790 return 0;
1791
1792err_icmpmsg_mib:
1793 free_percpu(net->mib.icmp_statistics);
1794err_icmp_mib:
1795 free_percpu(net->mib.udplite_statistics);
1796err_udplite_mib:
1797 free_percpu(net->mib.udp_statistics);
1798err_udp_mib:
1799 free_percpu(net->mib.net_statistics);
1800err_net_mib:
1801 free_percpu(net->mib.ip_statistics);
1802err_ip_mib:
1803 free_percpu(net->mib.tcp_statistics);
1804err_tcp_mib:
1805 return -ENOMEM;
1806}
1807
1808static __net_exit void ipv4_mib_exit_net(struct net *net)
1809{
1810 kfree(net->mib.icmpmsg_statistics);
1811 free_percpu(net->mib.icmp_statistics);
1812 free_percpu(net->mib.udplite_statistics);
1813 free_percpu(net->mib.udp_statistics);
1814 free_percpu(net->mib.net_statistics);
1815 free_percpu(net->mib.ip_statistics);
1816 free_percpu(net->mib.tcp_statistics);
1817#ifdef CONFIG_MPTCP
1818 /* allocated on demand, see mptcp_init_sock() */
1819 free_percpu(net->mib.mptcp_statistics);
1820#endif
1821}
1822
1823static __net_initdata struct pernet_operations ipv4_mib_ops = {
1824 .init = ipv4_mib_init_net,
1825 .exit = ipv4_mib_exit_net,
1826};
1827
1828static int __init init_ipv4_mibs(void)
1829{
1830 return register_pernet_subsys(&ipv4_mib_ops);
1831}
1832
1833static __net_init int inet_init_net(struct net *net)
1834{
1835 /*
1836 * Set defaults for local port range
1837 */
1838 seqlock_init(&net->ipv4.ip_local_ports.lock);
1839 net->ipv4.ip_local_ports.range[0] = 32768;
1840 net->ipv4.ip_local_ports.range[1] = 60999;
1841
1842 seqlock_init(&net->ipv4.ping_group_range.lock);
1843 /*
1844 * Sane defaults - nobody may create ping sockets.
1845 * Boot scripts should set this to distro-specific group.
1846 */
1847 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1848 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1849
1850 /* Default values for sysctl-controlled parameters.
1851 * We set them here, in case sysctl is not compiled.
1852 */
1853 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1854 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1855 net->ipv4.sysctl_ip_dynaddr = 0;
1856 net->ipv4.sysctl_ip_early_demux = 1;
1857 net->ipv4.sysctl_udp_early_demux = 1;
1858 net->ipv4.sysctl_tcp_early_demux = 1;
1859 net->ipv4.sysctl_nexthop_compat_mode = 1;
1860#ifdef CONFIG_SYSCTL
1861 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1862#endif
1863
1864 /* Some igmp sysctl, whose values are always used */
1865 net->ipv4.sysctl_igmp_max_memberships = 20;
1866 net->ipv4.sysctl_igmp_max_msf = 10;
1867 /* IGMP reports for link-local multicast groups are enabled by default */
1868 net->ipv4.sysctl_igmp_llm_reports = 1;
1869 net->ipv4.sysctl_igmp_qrv = 2;
1870
1871 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1872
1873 return 0;
1874}
1875
1876static __net_initdata struct pernet_operations af_inet_ops = {
1877 .init = inet_init_net,
1878};
1879
1880static int __init init_inet_pernet_ops(void)
1881{
1882 return register_pernet_subsys(&af_inet_ops);
1883}
1884
1885static int ipv4_proc_init(void);
1886
1887/*
1888 * IP protocol layer initialiser
1889 */
1890
1891static struct packet_offload ip_packet_offload __read_mostly = {
1892 .type = cpu_to_be16(ETH_P_IP),
1893 .callbacks = {
1894 .gso_segment = inet_gso_segment,
1895 .gro_receive = inet_gro_receive,
1896 .gro_complete = inet_gro_complete,
1897 },
1898};
1899
1900static const struct net_offload ipip_offload = {
1901 .callbacks = {
1902 .gso_segment = ipip_gso_segment,
1903 .gro_receive = ipip_gro_receive,
1904 .gro_complete = ipip_gro_complete,
1905 },
1906};
1907
1908static int __init ipip_offload_init(void)
1909{
1910 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1911}
1912
1913static int __init ipv4_offload_init(void)
1914{
1915 /*
1916 * Add offloads
1917 */
1918 if (udpv4_offload_init() < 0)
1919 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1920 if (tcpv4_offload_init() < 0)
1921 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1922 if (ipip_offload_init() < 0)
1923 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1924
1925 dev_add_offload(&ip_packet_offload);
1926 return 0;
1927}
1928
1929fs_initcall(ipv4_offload_init);
1930
1931static struct packet_type ip_packet_type __read_mostly = {
1932 .type = cpu_to_be16(ETH_P_IP),
1933 .func = ip_rcv,
1934 .list_func = ip_list_rcv,
1935};
1936
1937static int __init inet_init(void)
1938{
1939 struct inet_protosw *q;
1940 struct list_head *r;
1941 int rc;
1942
1943 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1944
1945 rc = proto_register(&tcp_prot, 1);
1946 if (rc)
1947 goto out;
1948
1949 rc = proto_register(&udp_prot, 1);
1950 if (rc)
1951 goto out_unregister_tcp_proto;
1952
1953 rc = proto_register(&raw_prot, 1);
1954 if (rc)
1955 goto out_unregister_udp_proto;
1956
1957 rc = proto_register(&ping_prot, 1);
1958 if (rc)
1959 goto out_unregister_raw_proto;
1960
1961 /*
1962 * Tell SOCKET that we are alive...
1963 */
1964
1965 (void)sock_register(&inet_family_ops);
1966
1967#ifdef CONFIG_SYSCTL
1968 ip_static_sysctl_init();
1969#endif
1970
1971 /*
1972 * Add all the base protocols.
1973 */
1974
1975 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1976 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1977 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1978 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1979 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1980 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1981#ifdef CONFIG_IP_MULTICAST
1982 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1983 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1984#endif
1985
1986 /* Register the socket-side information for inet_create. */
1987 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1988 INIT_LIST_HEAD(r);
1989
1990 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1991 inet_register_protosw(q);
1992
1993 /*
1994 * Set the ARP module up
1995 */
1996
1997 arp_init();
1998
1999 /*
2000 * Set the IP module up
2001 */
2002
2003 ip_init();
2004
2005 /* Setup TCP slab cache for open requests. */
2006 tcp_init();
2007
2008 /* Setup UDP memory threshold */
2009 udp_init();
2010
2011 /* Add UDP-Lite (RFC 3828) */
2012 udplite4_register();
2013
2014 raw_init();
2015
2016 ping_init();
2017
2018 /*
2019 * Set the ICMP layer up
2020 */
2021
2022 if (icmp_init() < 0)
2023 panic("Failed to create the ICMP control socket.\n");
2024
2025 /*
2026 * Initialise the multicast router
2027 */
2028#if defined(CONFIG_IP_MROUTE)
2029 if (ip_mr_init())
2030 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2031#endif
2032
2033 if (init_inet_pernet_ops())
2034 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2035 /*
2036 * Initialise per-cpu ipv4 mibs
2037 */
2038
2039 if (init_ipv4_mibs())
2040 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2041
2042 ipv4_proc_init();
2043
2044 ipfrag_init();
2045
2046 dev_add_pack(&ip_packet_type);
2047
2048 ip_tunnel_core_init();
2049
2050 rc = 0;
2051out:
2052 return rc;
2053out_unregister_raw_proto:
2054 proto_unregister(&raw_prot);
2055out_unregister_udp_proto:
2056 proto_unregister(&udp_prot);
2057out_unregister_tcp_proto:
2058 proto_unregister(&tcp_prot);
2059 goto out;
2060}
2061
2062fs_initcall(inet_init);
2063
2064/* ------------------------------------------------------------------------ */
2065
2066#ifdef CONFIG_PROC_FS
2067static int __init ipv4_proc_init(void)
2068{
2069 int rc = 0;
2070
2071 if (raw_proc_init())
2072 goto out_raw;
2073 if (tcp4_proc_init())
2074 goto out_tcp;
2075 if (udp4_proc_init())
2076 goto out_udp;
2077 if (ping_proc_init())
2078 goto out_ping;
2079 if (ip_misc_proc_init())
2080 goto out_misc;
2081out:
2082 return rc;
2083out_misc:
2084 ping_proc_exit();
2085out_ping:
2086 udp4_proc_exit();
2087out_udp:
2088 tcp4_proc_exit();
2089out_tcp:
2090 raw_proc_exit();
2091out_raw:
2092 rc = -ENOMEM;
2093 goto out;
2094}
2095
2096#else /* CONFIG_PROC_FS */
2097static int __init ipv4_proc_init(void)
2098{
2099 return 0;
2100}
2101#endif /* CONFIG_PROC_FS */