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
120 #include <trace/events/sock.h>
121
122 /* The inetsw table contains everything that inet_create needs to
123 * build a new socket.
124 */
125 static struct list_head inetsw[SOCK_MAX];
126 static DEFINE_SPINLOCK(inetsw_lock);
127
128 /* New destruction routine */
129
inet_sock_destruct(struct sock * sk)130 void inet_sock_destruct(struct sock *sk)
131 {
132 struct inet_sock *inet = inet_sk(sk);
133
134 __skb_queue_purge(&sk->sk_receive_queue);
135 if (sk->sk_rx_skb_cache) {
136 __kfree_skb(sk->sk_rx_skb_cache);
137 sk->sk_rx_skb_cache = NULL;
138 }
139 __skb_queue_purge(&sk->sk_error_queue);
140
141 sk_mem_reclaim(sk);
142
143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144 pr_err("Attempt to release TCP socket in state %d %p\n",
145 sk->sk_state, sk);
146 return;
147 }
148 if (!sock_flag(sk, SOCK_DEAD)) {
149 pr_err("Attempt to release alive inet socket %p\n", sk);
150 return;
151 }
152
153 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
155 WARN_ON(sk->sk_wmem_queued);
156 WARN_ON(sk->sk_forward_alloc);
157
158 kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
160 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
161 sk_refcnt_debug_dec(sk);
162 }
163 EXPORT_SYMBOL(inet_sock_destruct);
164
165 /*
166 * The routines beyond this point handle the behaviour of an AF_INET
167 * socket object. Mostly it punts to the subprotocols of IP to do
168 * the work.
169 */
170
171 /*
172 * Automatically bind an unbound socket.
173 */
174
inet_autobind(struct sock * sk)175 static int inet_autobind(struct sock *sk)
176 {
177 struct inet_sock *inet;
178 /* We may need to bind the socket. */
179 lock_sock(sk);
180 inet = inet_sk(sk);
181 if (!inet->inet_num) {
182 if (sk->sk_prot->get_port(sk, 0)) {
183 release_sock(sk);
184 return -EAGAIN;
185 }
186 inet->inet_sport = htons(inet->inet_num);
187 }
188 release_sock(sk);
189 return 0;
190 }
191
192 /*
193 * Move a socket into listening state.
194 */
inet_listen(struct socket * sock,int backlog)195 int inet_listen(struct socket *sock, int backlog)
196 {
197 struct sock *sk = sock->sk;
198 unsigned char old_state;
199 int err, tcp_fastopen;
200
201 lock_sock(sk);
202
203 err = -EINVAL;
204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205 goto out;
206
207 old_state = sk->sk_state;
208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209 goto out;
210
211 sk->sk_max_ack_backlog = backlog;
212 /* Really, if the socket is already in listen state
213 * we can only allow the backlog to be adjusted.
214 */
215 if (old_state != TCP_LISTEN) {
216 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
217 * Note that only TCP sockets (SOCK_STREAM) will reach here.
218 * Also fastopen backlog may already been set via the option
219 * because the socket was in TCP_LISTEN state previously but
220 * was shutdown() rather than close().
221 */
222 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
223 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
224 (tcp_fastopen & TFO_SERVER_ENABLE) &&
225 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
226 fastopen_queue_tune(sk, backlog);
227 tcp_fastopen_init_key_once(sock_net(sk));
228 }
229
230 err = inet_csk_listen_start(sk, backlog);
231 if (err)
232 goto out;
233 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
234 }
235 err = 0;
236
237 out:
238 release_sock(sk);
239 return err;
240 }
241 EXPORT_SYMBOL(inet_listen);
242
243 /*
244 * Create an inet socket.
245 */
246
inet_create(struct net * net,struct socket * sock,int protocol,int kern)247 static int inet_create(struct net *net, struct socket *sock, int protocol,
248 int kern)
249 {
250 struct sock *sk;
251 struct inet_protosw *answer;
252 struct inet_sock *inet;
253 struct proto *answer_prot;
254 unsigned char answer_flags;
255 int try_loading_module = 0;
256 int err;
257
258 if (protocol < 0 || protocol >= IPPROTO_MAX)
259 return -EINVAL;
260
261 sock->state = SS_UNCONNECTED;
262
263 /* Look for the requested type/protocol pair. */
264 lookup_protocol:
265 err = -ESOCKTNOSUPPORT;
266 rcu_read_lock();
267 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
268
269 err = 0;
270 /* Check the non-wild match. */
271 if (protocol == answer->protocol) {
272 if (protocol != IPPROTO_IP)
273 break;
274 } else {
275 /* Check for the two wild cases. */
276 if (IPPROTO_IP == protocol) {
277 protocol = answer->protocol;
278 break;
279 }
280 if (IPPROTO_IP == answer->protocol)
281 break;
282 }
283 err = -EPROTONOSUPPORT;
284 }
285
286 if (unlikely(err)) {
287 if (try_loading_module < 2) {
288 rcu_read_unlock();
289 /*
290 * Be more specific, e.g. net-pf-2-proto-132-type-1
291 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
292 */
293 if (++try_loading_module == 1)
294 request_module("net-pf-%d-proto-%d-type-%d",
295 PF_INET, protocol, sock->type);
296 /*
297 * Fall back to generic, e.g. net-pf-2-proto-132
298 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
299 */
300 else
301 request_module("net-pf-%d-proto-%d",
302 PF_INET, protocol);
303 goto lookup_protocol;
304 } else
305 goto out_rcu_unlock;
306 }
307
308 err = -EPERM;
309 if (sock->type == SOCK_RAW && !kern &&
310 !ns_capable(net->user_ns, CAP_NET_RAW))
311 goto out_rcu_unlock;
312
313 sock->ops = answer->ops;
314 answer_prot = answer->prot;
315 answer_flags = answer->flags;
316 rcu_read_unlock();
317
318 WARN_ON(!answer_prot->slab);
319
320 err = -ENOBUFS;
321 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
322 if (!sk)
323 goto out;
324
325 err = 0;
326 if (INET_PROTOSW_REUSE & answer_flags)
327 sk->sk_reuse = SK_CAN_REUSE;
328
329 inet = inet_sk(sk);
330 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
331
332 inet->nodefrag = 0;
333
334 if (SOCK_RAW == sock->type) {
335 inet->inet_num = protocol;
336 if (IPPROTO_RAW == protocol)
337 inet->hdrincl = 1;
338 }
339
340 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
341 inet->pmtudisc = IP_PMTUDISC_DONT;
342 else
343 inet->pmtudisc = IP_PMTUDISC_WANT;
344
345 inet->inet_id = 0;
346
347 sock_init_data(sock, sk);
348
349 sk->sk_destruct = inet_sock_destruct;
350 sk->sk_protocol = protocol;
351 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
352
353 inet->uc_ttl = -1;
354 inet->mc_loop = 1;
355 inet->mc_ttl = 1;
356 inet->mc_all = 1;
357 inet->mc_index = 0;
358 inet->mc_list = NULL;
359 inet->rcv_tos = 0;
360
361 sk_refcnt_debug_inc(sk);
362
363 if (inet->inet_num) {
364 /* It assumes that any protocol which allows
365 * the user to assign a number at socket
366 * creation time automatically
367 * shares.
368 */
369 inet->inet_sport = htons(inet->inet_num);
370 /* Add to protocol hash chains. */
371 err = sk->sk_prot->hash(sk);
372 if (err) {
373 sk_common_release(sk);
374 goto out;
375 }
376 }
377
378 if (sk->sk_prot->init) {
379 err = sk->sk_prot->init(sk);
380 if (err) {
381 sk_common_release(sk);
382 goto out;
383 }
384 }
385
386 if (!kern) {
387 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
388 if (err) {
389 sk_common_release(sk);
390 goto out;
391 }
392 }
393 out:
394 return err;
395 out_rcu_unlock:
396 rcu_read_unlock();
397 goto out;
398 }
399
400
401 /*
402 * The peer socket should always be NULL (or else). When we call this
403 * function we are destroying the object and from then on nobody
404 * should refer to it.
405 */
inet_release(struct socket * sock)406 int inet_release(struct socket *sock)
407 {
408 struct sock *sk = sock->sk;
409
410 if (sk) {
411 long timeout;
412
413 /* Applications forget to leave groups before exiting */
414 ip_mc_drop_socket(sk);
415
416 /* If linger is set, we don't return until the close
417 * is complete. Otherwise we return immediately. The
418 * actually closing is done the same either way.
419 *
420 * If the close is due to the process exiting, we never
421 * linger..
422 */
423 timeout = 0;
424 if (sock_flag(sk, SOCK_LINGER) &&
425 !(current->flags & PF_EXITING))
426 timeout = sk->sk_lingertime;
427 sk->sk_prot->close(sk, timeout);
428 sock->sk = NULL;
429 }
430 return 0;
431 }
432 EXPORT_SYMBOL(inet_release);
433
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)434 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
435 {
436 struct sock *sk = sock->sk;
437 int err;
438
439 /* If the socket has its own bind function then use it. (RAW) */
440 if (sk->sk_prot->bind) {
441 return sk->sk_prot->bind(sk, uaddr, addr_len);
442 }
443 if (addr_len < sizeof(struct sockaddr_in))
444 return -EINVAL;
445
446 /* BPF prog is run before any checks are done so that if the prog
447 * changes context in a wrong way it will be caught.
448 */
449 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
450 if (err)
451 return err;
452
453 return __inet_bind(sk, uaddr, addr_len, false, true);
454 }
455 EXPORT_SYMBOL(inet_bind);
456
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,bool force_bind_address_no_port,bool with_lock)457 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
458 bool force_bind_address_no_port, bool with_lock)
459 {
460 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
461 struct inet_sock *inet = inet_sk(sk);
462 struct net *net = sock_net(sk);
463 unsigned short snum;
464 int chk_addr_ret;
465 u32 tb_id = RT_TABLE_LOCAL;
466 int err;
467
468 if (addr->sin_family != AF_INET) {
469 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
470 * only if s_addr is INADDR_ANY.
471 */
472 err = -EAFNOSUPPORT;
473 if (addr->sin_family != AF_UNSPEC ||
474 addr->sin_addr.s_addr != htonl(INADDR_ANY))
475 goto out;
476 }
477
478 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
479 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
480
481 /* Not specified by any standard per-se, however it breaks too
482 * many applications when removed. It is unfortunate since
483 * allowing applications to make a non-local bind solves
484 * several problems with systems using dynamic addressing.
485 * (ie. your servers still start up even if your ISDN link
486 * is temporarily down)
487 */
488 err = -EADDRNOTAVAIL;
489 if (!inet_can_nonlocal_bind(net, inet) &&
490 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
491 chk_addr_ret != RTN_LOCAL &&
492 chk_addr_ret != RTN_MULTICAST &&
493 chk_addr_ret != RTN_BROADCAST)
494 goto out;
495
496 snum = ntohs(addr->sin_port);
497 err = -EPERM;
498 if (snum && inet_is_local_unbindable_port(net, snum))
499 goto out;
500
501 err = -EACCES;
502 if (snum && snum < inet_prot_sock(net) &&
503 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
504 goto out;
505
506 /* We keep a pair of addresses. rcv_saddr is the one
507 * used by hash lookups, and saddr is used for transmit.
508 *
509 * In the BSD API these are the same except where it
510 * would be illegal to use them (multicast/broadcast) in
511 * which case the sending device address is used.
512 */
513 if (with_lock)
514 lock_sock(sk);
515
516 /* Check these errors (active socket, double bind). */
517 err = -EINVAL;
518 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
519 goto out_release_sock;
520
521 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
522 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
523 inet->inet_saddr = 0; /* Use device */
524
525 /* Make sure we are allowed to bind here. */
526 if (snum || !(inet->bind_address_no_port ||
527 force_bind_address_no_port)) {
528 if (sk->sk_prot->get_port(sk, snum)) {
529 inet->inet_saddr = inet->inet_rcv_saddr = 0;
530 err = -EADDRINUSE;
531 goto out_release_sock;
532 }
533 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
534 if (err) {
535 inet->inet_saddr = inet->inet_rcv_saddr = 0;
536 goto out_release_sock;
537 }
538 }
539
540 if (inet->inet_rcv_saddr)
541 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
542 if (snum)
543 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
544 inet->inet_sport = htons(inet->inet_num);
545 inet->inet_daddr = 0;
546 inet->inet_dport = 0;
547 sk_dst_reset(sk);
548 err = 0;
549 out_release_sock:
550 if (with_lock)
551 release_sock(sk);
552 out:
553 return err;
554 }
555
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)556 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
557 int addr_len, int flags)
558 {
559 struct sock *sk = sock->sk;
560 int err;
561
562 if (addr_len < sizeof(uaddr->sa_family))
563 return -EINVAL;
564 if (uaddr->sa_family == AF_UNSPEC)
565 return sk->sk_prot->disconnect(sk, flags);
566
567 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
568 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
569 if (err)
570 return err;
571 }
572
573 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
574 return -EAGAIN;
575 return sk->sk_prot->connect(sk, uaddr, addr_len);
576 }
577 EXPORT_SYMBOL(inet_dgram_connect);
578
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)579 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
580 {
581 DEFINE_WAIT_FUNC(wait, woken_wake_function);
582
583 add_wait_queue(sk_sleep(sk), &wait);
584 sk->sk_write_pending += writebias;
585
586 /* Basic assumption: if someone sets sk->sk_err, he _must_
587 * change state of the socket from TCP_SYN_*.
588 * Connect() does not allow to get error notifications
589 * without closing the socket.
590 */
591 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
592 release_sock(sk);
593 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
594 lock_sock(sk);
595 if (signal_pending(current) || !timeo)
596 break;
597 }
598 remove_wait_queue(sk_sleep(sk), &wait);
599 sk->sk_write_pending -= writebias;
600 return timeo;
601 }
602
603 /*
604 * Connect to a remote host. There is regrettably still a little
605 * TCP 'magic' in here.
606 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)607 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
608 int addr_len, int flags, int is_sendmsg)
609 {
610 struct sock *sk = sock->sk;
611 int err;
612 long timeo;
613
614 /*
615 * uaddr can be NULL and addr_len can be 0 if:
616 * sk is a TCP fastopen active socket and
617 * TCP_FASTOPEN_CONNECT sockopt is set and
618 * we already have a valid cookie for this socket.
619 * In this case, user can call write() after connect().
620 * write() will invoke tcp_sendmsg_fastopen() which calls
621 * __inet_stream_connect().
622 */
623 if (uaddr) {
624 if (addr_len < sizeof(uaddr->sa_family))
625 return -EINVAL;
626
627 if (uaddr->sa_family == AF_UNSPEC) {
628 err = sk->sk_prot->disconnect(sk, flags);
629 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
630 goto out;
631 }
632 }
633
634 switch (sock->state) {
635 default:
636 err = -EINVAL;
637 goto out;
638 case SS_CONNECTED:
639 err = -EISCONN;
640 goto out;
641 case SS_CONNECTING:
642 if (inet_sk(sk)->defer_connect)
643 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
644 else
645 err = -EALREADY;
646 /* Fall out of switch with err, set for this state */
647 break;
648 case SS_UNCONNECTED:
649 err = -EISCONN;
650 if (sk->sk_state != TCP_CLOSE)
651 goto out;
652
653 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
654 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
655 if (err)
656 goto out;
657 }
658
659 err = sk->sk_prot->connect(sk, uaddr, addr_len);
660 if (err < 0)
661 goto out;
662
663 sock->state = SS_CONNECTING;
664
665 if (!err && inet_sk(sk)->defer_connect)
666 goto out;
667
668 /* Just entered SS_CONNECTING state; the only
669 * difference is that return value in non-blocking
670 * case is EINPROGRESS, rather than EALREADY.
671 */
672 err = -EINPROGRESS;
673 break;
674 }
675
676 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
677
678 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
679 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
680 tcp_sk(sk)->fastopen_req &&
681 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
682
683 /* Error code is set above */
684 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
685 goto out;
686
687 err = sock_intr_errno(timeo);
688 if (signal_pending(current))
689 goto out;
690 }
691
692 /* Connection was closed by RST, timeout, ICMP error
693 * or another process disconnected us.
694 */
695 if (sk->sk_state == TCP_CLOSE)
696 goto sock_error;
697
698 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
699 * and error was received after socket entered established state.
700 * Hence, it is handled normally after connect() return successfully.
701 */
702
703 sock->state = SS_CONNECTED;
704 err = 0;
705 out:
706 return err;
707
708 sock_error:
709 err = sock_error(sk) ? : -ECONNABORTED;
710 sock->state = SS_UNCONNECTED;
711 if (sk->sk_prot->disconnect(sk, flags))
712 sock->state = SS_DISCONNECTING;
713 goto out;
714 }
715 EXPORT_SYMBOL(__inet_stream_connect);
716
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)717 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
718 int addr_len, int flags)
719 {
720 int err;
721
722 lock_sock(sock->sk);
723 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
724 release_sock(sock->sk);
725 return err;
726 }
727 EXPORT_SYMBOL(inet_stream_connect);
728
729 /*
730 * Accept a pending connection. The TCP layer now gives BSD semantics.
731 */
732
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)733 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
734 bool kern)
735 {
736 struct sock *sk1 = sock->sk;
737 int err = -EINVAL;
738 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
739
740 if (!sk2)
741 goto do_err;
742
743 lock_sock(sk2);
744
745 sock_rps_record_flow(sk2);
746 WARN_ON(!((1 << sk2->sk_state) &
747 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
748 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
749
750 sock_graft(sk2, newsock);
751
752 newsock->state = SS_CONNECTED;
753 err = 0;
754 release_sock(sk2);
755 do_err:
756 return err;
757 }
758 EXPORT_SYMBOL(inet_accept);
759
760
761 /*
762 * This does both peername and sockname.
763 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)764 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
765 int peer)
766 {
767 struct sock *sk = sock->sk;
768 struct inet_sock *inet = inet_sk(sk);
769 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
770
771 sin->sin_family = AF_INET;
772 if (peer) {
773 if (!inet->inet_dport ||
774 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
775 peer == 1))
776 return -ENOTCONN;
777 sin->sin_port = inet->inet_dport;
778 sin->sin_addr.s_addr = inet->inet_daddr;
779 } else {
780 __be32 addr = inet->inet_rcv_saddr;
781 if (!addr)
782 addr = inet->inet_saddr;
783 sin->sin_port = inet->inet_sport;
784 sin->sin_addr.s_addr = addr;
785 }
786 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
787 return sizeof(*sin);
788 }
789 EXPORT_SYMBOL(inet_getname);
790
inet_send_prepare(struct sock * sk)791 int inet_send_prepare(struct sock *sk)
792 {
793 sock_rps_record_flow(sk);
794
795 /* We may need to bind the socket. */
796 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
797 inet_autobind(sk))
798 return -EAGAIN;
799
800 return 0;
801 }
802 EXPORT_SYMBOL_GPL(inet_send_prepare);
803
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)804 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
805 {
806 struct sock *sk = sock->sk;
807
808 if (unlikely(inet_send_prepare(sk)))
809 return -EAGAIN;
810
811 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
812 sk, msg, size);
813 }
814 EXPORT_SYMBOL(inet_sendmsg);
815
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)816 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
817 size_t size, int flags)
818 {
819 struct sock *sk = sock->sk;
820
821 if (unlikely(inet_send_prepare(sk)))
822 return -EAGAIN;
823
824 if (sk->sk_prot->sendpage)
825 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
826 return sock_no_sendpage(sock, page, offset, size, flags);
827 }
828 EXPORT_SYMBOL(inet_sendpage);
829
830 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
831 size_t, int, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)832 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
833 int flags)
834 {
835 struct sock *sk = sock->sk;
836 int addr_len = 0;
837 int err;
838
839 if (likely(!(flags & MSG_ERRQUEUE)))
840 sock_rps_record_flow(sk);
841
842 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
843 sk, msg, size, flags & MSG_DONTWAIT,
844 flags & ~MSG_DONTWAIT, &addr_len);
845 if (err >= 0)
846 msg->msg_namelen = addr_len;
847 return err;
848 }
849 EXPORT_SYMBOL(inet_recvmsg);
850
inet_shutdown(struct socket * sock,int how)851 int inet_shutdown(struct socket *sock, int how)
852 {
853 struct sock *sk = sock->sk;
854 int err = 0;
855
856 /* This should really check to make sure
857 * the socket is a TCP socket. (WHY AC...)
858 */
859 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
860 1->2 bit 2 snds.
861 2->3 */
862 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
863 return -EINVAL;
864
865 lock_sock(sk);
866 if (sock->state == SS_CONNECTING) {
867 if ((1 << sk->sk_state) &
868 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
869 sock->state = SS_DISCONNECTING;
870 else
871 sock->state = SS_CONNECTED;
872 }
873
874 switch (sk->sk_state) {
875 case TCP_CLOSE:
876 err = -ENOTCONN;
877 /* Hack to wake up other listeners, who can poll for
878 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
879 /* fall through */
880 default:
881 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
882 if (sk->sk_prot->shutdown)
883 sk->sk_prot->shutdown(sk, how);
884 break;
885
886 /* Remaining two branches are temporary solution for missing
887 * close() in multithreaded environment. It is _not_ a good idea,
888 * but we have no choice until close() is repaired at VFS level.
889 */
890 case TCP_LISTEN:
891 if (!(how & RCV_SHUTDOWN))
892 break;
893 /* fall through */
894 case TCP_SYN_SENT:
895 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
896 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
897 break;
898 }
899
900 /* Wake up anyone sleeping in poll. */
901 sk->sk_state_change(sk);
902 release_sock(sk);
903 return err;
904 }
905 EXPORT_SYMBOL(inet_shutdown);
906
907 /*
908 * ioctl() calls you can issue on an INET socket. Most of these are
909 * device configuration and stuff and very rarely used. Some ioctls
910 * pass on to the socket itself.
911 *
912 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
913 * loads the devconfigure module does its configuring and unloads it.
914 * There's a good 20K of config code hanging around the kernel.
915 */
916
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)917 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
918 {
919 struct sock *sk = sock->sk;
920 int err = 0;
921 struct net *net = sock_net(sk);
922 void __user *p = (void __user *)arg;
923 struct ifreq ifr;
924 struct rtentry rt;
925
926 switch (cmd) {
927 case SIOCADDRT:
928 case SIOCDELRT:
929 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
930 return -EFAULT;
931 err = ip_rt_ioctl(net, cmd, &rt);
932 break;
933 case SIOCRTMSG:
934 err = -EINVAL;
935 break;
936 case SIOCDARP:
937 case SIOCGARP:
938 case SIOCSARP:
939 err = arp_ioctl(net, cmd, (void __user *)arg);
940 break;
941 case SIOCGIFADDR:
942 case SIOCGIFBRDADDR:
943 case SIOCGIFNETMASK:
944 case SIOCGIFDSTADDR:
945 case SIOCGIFPFLAGS:
946 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
947 return -EFAULT;
948 err = devinet_ioctl(net, cmd, &ifr);
949 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
950 err = -EFAULT;
951 break;
952
953 case SIOCSIFADDR:
954 case SIOCSIFBRDADDR:
955 case SIOCSIFNETMASK:
956 case SIOCSIFDSTADDR:
957 case SIOCSIFPFLAGS:
958 case SIOCSIFFLAGS:
959 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
960 return -EFAULT;
961 err = devinet_ioctl(net, cmd, &ifr);
962 break;
963 default:
964 if (sk->sk_prot->ioctl)
965 err = sk->sk_prot->ioctl(sk, cmd, arg);
966 else
967 err = -ENOIOCTLCMD;
968 break;
969 }
970 return err;
971 }
972 EXPORT_SYMBOL(inet_ioctl);
973
974 #ifdef CONFIG_COMPAT
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)975 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
976 {
977 struct sock *sk = sock->sk;
978 int err = -ENOIOCTLCMD;
979
980 if (sk->sk_prot->compat_ioctl)
981 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
982
983 return err;
984 }
985 #endif
986
987 const struct proto_ops inet_stream_ops = {
988 .family = PF_INET,
989 .owner = THIS_MODULE,
990 .release = inet_release,
991 .bind = inet_bind,
992 .connect = inet_stream_connect,
993 .socketpair = sock_no_socketpair,
994 .accept = inet_accept,
995 .getname = inet_getname,
996 .poll = tcp_poll,
997 .ioctl = inet_ioctl,
998 .gettstamp = sock_gettstamp,
999 .listen = inet_listen,
1000 .shutdown = inet_shutdown,
1001 .setsockopt = sock_common_setsockopt,
1002 .getsockopt = sock_common_getsockopt,
1003 .sendmsg = inet_sendmsg,
1004 .recvmsg = inet_recvmsg,
1005 #ifdef CONFIG_MMU
1006 .mmap = tcp_mmap,
1007 #endif
1008 .sendpage = inet_sendpage,
1009 .splice_read = tcp_splice_read,
1010 .read_sock = tcp_read_sock,
1011 .sendmsg_locked = tcp_sendmsg_locked,
1012 .sendpage_locked = tcp_sendpage_locked,
1013 .peek_len = tcp_peek_len,
1014 #ifdef CONFIG_COMPAT
1015 .compat_setsockopt = compat_sock_common_setsockopt,
1016 .compat_getsockopt = compat_sock_common_getsockopt,
1017 .compat_ioctl = inet_compat_ioctl,
1018 #endif
1019 .set_rcvlowat = tcp_set_rcvlowat,
1020 };
1021 EXPORT_SYMBOL(inet_stream_ops);
1022
1023 const struct proto_ops inet_dgram_ops = {
1024 .family = PF_INET,
1025 .owner = THIS_MODULE,
1026 .release = inet_release,
1027 .bind = inet_bind,
1028 .connect = inet_dgram_connect,
1029 .socketpair = sock_no_socketpair,
1030 .accept = sock_no_accept,
1031 .getname = inet_getname,
1032 .poll = udp_poll,
1033 .ioctl = inet_ioctl,
1034 .gettstamp = sock_gettstamp,
1035 .listen = sock_no_listen,
1036 .shutdown = inet_shutdown,
1037 .setsockopt = sock_common_setsockopt,
1038 .getsockopt = sock_common_getsockopt,
1039 .sendmsg = inet_sendmsg,
1040 .recvmsg = inet_recvmsg,
1041 .mmap = sock_no_mmap,
1042 .sendpage = inet_sendpage,
1043 .set_peek_off = sk_set_peek_off,
1044 #ifdef CONFIG_COMPAT
1045 .compat_setsockopt = compat_sock_common_setsockopt,
1046 .compat_getsockopt = compat_sock_common_getsockopt,
1047 .compat_ioctl = inet_compat_ioctl,
1048 #endif
1049 };
1050 EXPORT_SYMBOL(inet_dgram_ops);
1051
1052 /*
1053 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1054 * udp_poll
1055 */
1056 static const struct proto_ops inet_sockraw_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 = datagram_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 .recvmsg = inet_recvmsg,
1074 .mmap = sock_no_mmap,
1075 .sendpage = inet_sendpage,
1076 #ifdef CONFIG_COMPAT
1077 .compat_setsockopt = compat_sock_common_setsockopt,
1078 .compat_getsockopt = compat_sock_common_getsockopt,
1079 .compat_ioctl = inet_compat_ioctl,
1080 #endif
1081 };
1082
1083 static const struct net_proto_family inet_family_ops = {
1084 .family = PF_INET,
1085 .create = inet_create,
1086 .owner = THIS_MODULE,
1087 };
1088
1089 /* Upon startup we insert all the elements in inetsw_array[] into
1090 * the linked list inetsw.
1091 */
1092 static struct inet_protosw inetsw_array[] =
1093 {
1094 {
1095 .type = SOCK_STREAM,
1096 .protocol = IPPROTO_TCP,
1097 .prot = &tcp_prot,
1098 .ops = &inet_stream_ops,
1099 .flags = INET_PROTOSW_PERMANENT |
1100 INET_PROTOSW_ICSK,
1101 },
1102
1103 {
1104 .type = SOCK_DGRAM,
1105 .protocol = IPPROTO_UDP,
1106 .prot = &udp_prot,
1107 .ops = &inet_dgram_ops,
1108 .flags = INET_PROTOSW_PERMANENT,
1109 },
1110
1111 {
1112 .type = SOCK_DGRAM,
1113 .protocol = IPPROTO_ICMP,
1114 .prot = &ping_prot,
1115 .ops = &inet_sockraw_ops,
1116 .flags = INET_PROTOSW_REUSE,
1117 },
1118
1119 {
1120 .type = SOCK_RAW,
1121 .protocol = IPPROTO_IP, /* wild card */
1122 .prot = &raw_prot,
1123 .ops = &inet_sockraw_ops,
1124 .flags = INET_PROTOSW_REUSE,
1125 }
1126 };
1127
1128 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1129
inet_register_protosw(struct inet_protosw * p)1130 void inet_register_protosw(struct inet_protosw *p)
1131 {
1132 struct list_head *lh;
1133 struct inet_protosw *answer;
1134 int protocol = p->protocol;
1135 struct list_head *last_perm;
1136
1137 spin_lock_bh(&inetsw_lock);
1138
1139 if (p->type >= SOCK_MAX)
1140 goto out_illegal;
1141
1142 /* If we are trying to override a permanent protocol, bail. */
1143 last_perm = &inetsw[p->type];
1144 list_for_each(lh, &inetsw[p->type]) {
1145 answer = list_entry(lh, struct inet_protosw, list);
1146 /* Check only the non-wild match. */
1147 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1148 break;
1149 if (protocol == answer->protocol)
1150 goto out_permanent;
1151 last_perm = lh;
1152 }
1153
1154 /* Add the new entry after the last permanent entry if any, so that
1155 * the new entry does not override a permanent entry when matched with
1156 * a wild-card protocol. But it is allowed to override any existing
1157 * non-permanent entry. This means that when we remove this entry, the
1158 * system automatically returns to the old behavior.
1159 */
1160 list_add_rcu(&p->list, last_perm);
1161 out:
1162 spin_unlock_bh(&inetsw_lock);
1163
1164 return;
1165
1166 out_permanent:
1167 pr_err("Attempt to override permanent protocol %d\n", protocol);
1168 goto out;
1169
1170 out_illegal:
1171 pr_err("Ignoring attempt to register invalid socket type %d\n",
1172 p->type);
1173 goto out;
1174 }
1175 EXPORT_SYMBOL(inet_register_protosw);
1176
inet_unregister_protosw(struct inet_protosw * p)1177 void inet_unregister_protosw(struct inet_protosw *p)
1178 {
1179 if (INET_PROTOSW_PERMANENT & p->flags) {
1180 pr_err("Attempt to unregister permanent protocol %d\n",
1181 p->protocol);
1182 } else {
1183 spin_lock_bh(&inetsw_lock);
1184 list_del_rcu(&p->list);
1185 spin_unlock_bh(&inetsw_lock);
1186
1187 synchronize_net();
1188 }
1189 }
1190 EXPORT_SYMBOL(inet_unregister_protosw);
1191
inet_sk_reselect_saddr(struct sock * sk)1192 static int inet_sk_reselect_saddr(struct sock *sk)
1193 {
1194 struct inet_sock *inet = inet_sk(sk);
1195 __be32 old_saddr = inet->inet_saddr;
1196 __be32 daddr = inet->inet_daddr;
1197 struct flowi4 *fl4;
1198 struct rtable *rt;
1199 __be32 new_saddr;
1200 struct ip_options_rcu *inet_opt;
1201
1202 inet_opt = rcu_dereference_protected(inet->inet_opt,
1203 lockdep_sock_is_held(sk));
1204 if (inet_opt && inet_opt->opt.srr)
1205 daddr = inet_opt->opt.faddr;
1206
1207 /* Query new route. */
1208 fl4 = &inet->cork.fl.u.ip4;
1209 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1210 sk->sk_bound_dev_if, sk->sk_protocol,
1211 inet->inet_sport, inet->inet_dport, sk);
1212 if (IS_ERR(rt))
1213 return PTR_ERR(rt);
1214
1215 sk_setup_caps(sk, &rt->dst);
1216
1217 new_saddr = fl4->saddr;
1218
1219 if (new_saddr == old_saddr)
1220 return 0;
1221
1222 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1223 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1224 __func__, &old_saddr, &new_saddr);
1225 }
1226
1227 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1228
1229 /*
1230 * XXX The only one ugly spot where we need to
1231 * XXX really change the sockets identity after
1232 * XXX it has entered the hashes. -DaveM
1233 *
1234 * Besides that, it does not check for connection
1235 * uniqueness. Wait for troubles.
1236 */
1237 return __sk_prot_rehash(sk);
1238 }
1239
inet_sk_rebuild_header(struct sock * sk)1240 int inet_sk_rebuild_header(struct sock *sk)
1241 {
1242 struct inet_sock *inet = inet_sk(sk);
1243 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1244 __be32 daddr;
1245 struct ip_options_rcu *inet_opt;
1246 struct flowi4 *fl4;
1247 int err;
1248
1249 /* Route is OK, nothing to do. */
1250 if (rt)
1251 return 0;
1252
1253 /* Reroute. */
1254 rcu_read_lock();
1255 inet_opt = rcu_dereference(inet->inet_opt);
1256 daddr = inet->inet_daddr;
1257 if (inet_opt && inet_opt->opt.srr)
1258 daddr = inet_opt->opt.faddr;
1259 rcu_read_unlock();
1260 fl4 = &inet->cork.fl.u.ip4;
1261 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1262 inet->inet_dport, inet->inet_sport,
1263 sk->sk_protocol, RT_CONN_FLAGS(sk),
1264 sk->sk_bound_dev_if);
1265 if (!IS_ERR(rt)) {
1266 err = 0;
1267 sk_setup_caps(sk, &rt->dst);
1268 } else {
1269 err = PTR_ERR(rt);
1270
1271 /* Routing failed... */
1272 sk->sk_route_caps = 0;
1273 /*
1274 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1275 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1276 */
1277 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1278 sk->sk_state != TCP_SYN_SENT ||
1279 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1280 (err = inet_sk_reselect_saddr(sk)) != 0)
1281 sk->sk_err_soft = -err;
1282 }
1283
1284 return err;
1285 }
1286 EXPORT_SYMBOL(inet_sk_rebuild_header);
1287
inet_sk_set_state(struct sock * sk,int state)1288 void inet_sk_set_state(struct sock *sk, int state)
1289 {
1290 trace_inet_sock_set_state(sk, sk->sk_state, state);
1291 sk->sk_state = state;
1292 }
1293 EXPORT_SYMBOL(inet_sk_set_state);
1294
inet_sk_state_store(struct sock * sk,int newstate)1295 void inet_sk_state_store(struct sock *sk, int newstate)
1296 {
1297 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1298 smp_store_release(&sk->sk_state, newstate);
1299 }
1300
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1301 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1302 netdev_features_t features)
1303 {
1304 bool udpfrag = false, fixedid = false, gso_partial, encap;
1305 struct sk_buff *segs = ERR_PTR(-EINVAL);
1306 const struct net_offload *ops;
1307 unsigned int offset = 0;
1308 struct iphdr *iph;
1309 int proto, tot_len;
1310 int nhoff;
1311 int ihl;
1312 int id;
1313
1314 skb_reset_network_header(skb);
1315 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1316 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1317 goto out;
1318
1319 iph = ip_hdr(skb);
1320 ihl = iph->ihl * 4;
1321 if (ihl < sizeof(*iph))
1322 goto out;
1323
1324 id = ntohs(iph->id);
1325 proto = iph->protocol;
1326
1327 /* Warning: after this point, iph might be no longer valid */
1328 if (unlikely(!pskb_may_pull(skb, ihl)))
1329 goto out;
1330 __skb_pull(skb, ihl);
1331
1332 encap = SKB_GSO_CB(skb)->encap_level > 0;
1333 if (encap)
1334 features &= skb->dev->hw_enc_features;
1335 SKB_GSO_CB(skb)->encap_level += ihl;
1336
1337 skb_reset_transport_header(skb);
1338
1339 segs = ERR_PTR(-EPROTONOSUPPORT);
1340
1341 if (!skb->encapsulation || encap) {
1342 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1343 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1344
1345 /* fixed ID is invalid if DF bit is not set */
1346 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1347 goto out;
1348 }
1349
1350 ops = rcu_dereference(inet_offloads[proto]);
1351 if (likely(ops && ops->callbacks.gso_segment)) {
1352 segs = ops->callbacks.gso_segment(skb, features);
1353 if (!segs)
1354 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1355 }
1356
1357 if (IS_ERR_OR_NULL(segs))
1358 goto out;
1359
1360 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1361
1362 skb = segs;
1363 do {
1364 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1365 if (udpfrag) {
1366 iph->frag_off = htons(offset >> 3);
1367 if (skb->next)
1368 iph->frag_off |= htons(IP_MF);
1369 offset += skb->len - nhoff - ihl;
1370 tot_len = skb->len - nhoff;
1371 } else if (skb_is_gso(skb)) {
1372 if (!fixedid) {
1373 iph->id = htons(id);
1374 id += skb_shinfo(skb)->gso_segs;
1375 }
1376
1377 if (gso_partial)
1378 tot_len = skb_shinfo(skb)->gso_size +
1379 SKB_GSO_CB(skb)->data_offset +
1380 skb->head - (unsigned char *)iph;
1381 else
1382 tot_len = skb->len - nhoff;
1383 } else {
1384 if (!fixedid)
1385 iph->id = htons(id++);
1386 tot_len = skb->len - nhoff;
1387 }
1388 iph->tot_len = htons(tot_len);
1389 ip_send_check(iph);
1390 if (encap)
1391 skb_reset_inner_headers(skb);
1392 skb->network_header = (u8 *)iph - skb->head;
1393 skb_reset_mac_len(skb);
1394 } while ((skb = skb->next));
1395
1396 out:
1397 return segs;
1398 }
1399 EXPORT_SYMBOL(inet_gso_segment);
1400
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1401 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1402 netdev_features_t features)
1403 {
1404 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1405 return ERR_PTR(-EINVAL);
1406
1407 return inet_gso_segment(skb, features);
1408 }
1409
1410 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *,
1411 struct sk_buff *));
1412 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *,
1413 struct sk_buff *));
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1414 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1415 {
1416 const struct net_offload *ops;
1417 struct sk_buff *pp = NULL;
1418 const struct iphdr *iph;
1419 struct sk_buff *p;
1420 unsigned int hlen;
1421 unsigned int off;
1422 unsigned int id;
1423 int flush = 1;
1424 int proto;
1425
1426 off = skb_gro_offset(skb);
1427 hlen = off + sizeof(*iph);
1428 iph = skb_gro_header_fast(skb, off);
1429 if (skb_gro_header_hard(skb, hlen)) {
1430 iph = skb_gro_header_slow(skb, hlen, off);
1431 if (unlikely(!iph))
1432 goto out;
1433 }
1434
1435 proto = iph->protocol;
1436
1437 rcu_read_lock();
1438 ops = rcu_dereference(inet_offloads[proto]);
1439 if (!ops || !ops->callbacks.gro_receive)
1440 goto out_unlock;
1441
1442 if (*(u8 *)iph != 0x45)
1443 goto out_unlock;
1444
1445 if (ip_is_fragment(iph))
1446 goto out_unlock;
1447
1448 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1449 goto out_unlock;
1450
1451 id = ntohl(*(__be32 *)&iph->id);
1452 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1453 id >>= 16;
1454
1455 list_for_each_entry(p, head, list) {
1456 struct iphdr *iph2;
1457 u16 flush_id;
1458
1459 if (!NAPI_GRO_CB(p)->same_flow)
1460 continue;
1461
1462 iph2 = (struct iphdr *)(p->data + off);
1463 /* The above works because, with the exception of the top
1464 * (inner most) layer, we only aggregate pkts with the same
1465 * hdr length so all the hdrs we'll need to verify will start
1466 * at the same offset.
1467 */
1468 if ((iph->protocol ^ iph2->protocol) |
1469 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1470 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1471 NAPI_GRO_CB(p)->same_flow = 0;
1472 continue;
1473 }
1474
1475 /* All fields must match except length and checksum. */
1476 NAPI_GRO_CB(p)->flush |=
1477 (iph->ttl ^ iph2->ttl) |
1478 (iph->tos ^ iph2->tos) |
1479 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1480
1481 NAPI_GRO_CB(p)->flush |= flush;
1482
1483 /* We need to store of the IP ID check to be included later
1484 * when we can verify that this packet does in fact belong
1485 * to a given flow.
1486 */
1487 flush_id = (u16)(id - ntohs(iph2->id));
1488
1489 /* This bit of code makes it much easier for us to identify
1490 * the cases where we are doing atomic vs non-atomic IP ID
1491 * checks. Specifically an atomic check can return IP ID
1492 * values 0 - 0xFFFF, while a non-atomic check can only
1493 * return 0 or 0xFFFF.
1494 */
1495 if (!NAPI_GRO_CB(p)->is_atomic ||
1496 !(iph->frag_off & htons(IP_DF))) {
1497 flush_id ^= NAPI_GRO_CB(p)->count;
1498 flush_id = flush_id ? 0xFFFF : 0;
1499 }
1500
1501 /* If the previous IP ID value was based on an atomic
1502 * datagram we can overwrite the value and ignore it.
1503 */
1504 if (NAPI_GRO_CB(skb)->is_atomic)
1505 NAPI_GRO_CB(p)->flush_id = flush_id;
1506 else
1507 NAPI_GRO_CB(p)->flush_id |= flush_id;
1508 }
1509
1510 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1511 NAPI_GRO_CB(skb)->flush |= flush;
1512 skb_set_network_header(skb, off);
1513 /* The above will be needed by the transport layer if there is one
1514 * immediately following this IP hdr.
1515 */
1516
1517 /* Note : No need to call skb_gro_postpull_rcsum() here,
1518 * as we already checked checksum over ipv4 header was 0
1519 */
1520 skb_gro_pull(skb, sizeof(*iph));
1521 skb_set_transport_header(skb, skb_gro_offset(skb));
1522
1523 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1524 ops->callbacks.gro_receive, head, skb);
1525
1526 out_unlock:
1527 rcu_read_unlock();
1528
1529 out:
1530 skb_gro_flush_final(skb, pp, flush);
1531
1532 return pp;
1533 }
1534 EXPORT_SYMBOL(inet_gro_receive);
1535
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1536 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1537 struct sk_buff *skb)
1538 {
1539 if (NAPI_GRO_CB(skb)->encap_mark) {
1540 NAPI_GRO_CB(skb)->flush = 1;
1541 return NULL;
1542 }
1543
1544 NAPI_GRO_CB(skb)->encap_mark = 1;
1545
1546 return inet_gro_receive(head, skb);
1547 }
1548
1549 #define SECONDS_PER_DAY 86400
1550
1551 /* inet_current_timestamp - Return IP network timestamp
1552 *
1553 * Return milliseconds since midnight in network byte order.
1554 */
inet_current_timestamp(void)1555 __be32 inet_current_timestamp(void)
1556 {
1557 u32 secs;
1558 u32 msecs;
1559 struct timespec64 ts;
1560
1561 ktime_get_real_ts64(&ts);
1562
1563 /* Get secs since midnight. */
1564 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1565 /* Convert to msecs. */
1566 msecs = secs * MSEC_PER_SEC;
1567 /* Convert nsec to msec. */
1568 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1569
1570 /* Convert to network byte order. */
1571 return htonl(msecs);
1572 }
1573 EXPORT_SYMBOL(inet_current_timestamp);
1574
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1575 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1576 {
1577 unsigned int family = READ_ONCE(sk->sk_family);
1578
1579 if (family == AF_INET)
1580 return ip_recv_error(sk, msg, len, addr_len);
1581 #if IS_ENABLED(CONFIG_IPV6)
1582 if (family == AF_INET6)
1583 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1584 #endif
1585 return -EINVAL;
1586 }
1587
1588 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *, int));
1589 INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int));
inet_gro_complete(struct sk_buff * skb,int nhoff)1590 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1591 {
1592 __be16 newlen = htons(skb->len - nhoff);
1593 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1594 const struct net_offload *ops;
1595 int proto = iph->protocol;
1596 int err = -ENOSYS;
1597
1598 if (skb->encapsulation) {
1599 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1600 skb_set_inner_network_header(skb, nhoff);
1601 }
1602
1603 csum_replace2(&iph->check, iph->tot_len, newlen);
1604 iph->tot_len = newlen;
1605
1606 rcu_read_lock();
1607 ops = rcu_dereference(inet_offloads[proto]);
1608 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1609 goto out_unlock;
1610
1611 /* Only need to add sizeof(*iph) to get to the next hdr below
1612 * because any hdr with option will have been flushed in
1613 * inet_gro_receive().
1614 */
1615 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1616 tcp4_gro_complete, udp4_gro_complete,
1617 skb, nhoff + sizeof(*iph));
1618
1619 out_unlock:
1620 rcu_read_unlock();
1621
1622 return err;
1623 }
1624 EXPORT_SYMBOL(inet_gro_complete);
1625
ipip_gro_complete(struct sk_buff * skb,int nhoff)1626 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1627 {
1628 skb->encapsulation = 1;
1629 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1630 return inet_gro_complete(skb, nhoff);
1631 }
1632
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1633 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1634 unsigned short type, unsigned char protocol,
1635 struct net *net)
1636 {
1637 struct socket *sock;
1638 int rc = sock_create_kern(net, family, type, protocol, &sock);
1639
1640 if (rc == 0) {
1641 *sk = sock->sk;
1642 (*sk)->sk_allocation = GFP_ATOMIC;
1643 /*
1644 * Unhash it so that IP input processing does not even see it,
1645 * we do not wish this socket to see incoming packets.
1646 */
1647 (*sk)->sk_prot->unhash(*sk);
1648 }
1649 return rc;
1650 }
1651 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1652
snmp_get_cpu_field(void __percpu * mib,int cpu,int offt)1653 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1654 {
1655 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1656 }
1657 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1658
snmp_fold_field(void __percpu * mib,int offt)1659 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1660 {
1661 unsigned long res = 0;
1662 int i;
1663
1664 for_each_possible_cpu(i)
1665 res += snmp_get_cpu_field(mib, i, offt);
1666 return res;
1667 }
1668 EXPORT_SYMBOL_GPL(snmp_fold_field);
1669
1670 #if BITS_PER_LONG==32
1671
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1672 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1673 size_t syncp_offset)
1674 {
1675 void *bhptr;
1676 struct u64_stats_sync *syncp;
1677 u64 v;
1678 unsigned int start;
1679
1680 bhptr = per_cpu_ptr(mib, cpu);
1681 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1682 do {
1683 start = u64_stats_fetch_begin_irq(syncp);
1684 v = *(((u64 *)bhptr) + offt);
1685 } while (u64_stats_fetch_retry_irq(syncp, start));
1686
1687 return v;
1688 }
1689 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1690
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1691 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1692 {
1693 u64 res = 0;
1694 int cpu;
1695
1696 for_each_possible_cpu(cpu) {
1697 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1698 }
1699 return res;
1700 }
1701 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1702 #endif
1703
1704 #ifdef CONFIG_IP_MULTICAST
1705 static const struct net_protocol igmp_protocol = {
1706 .handler = igmp_rcv,
1707 .netns_ok = 1,
1708 };
1709 #endif
1710
1711 static const struct net_protocol tcp_protocol = {
1712 .handler = tcp_v4_rcv,
1713 .err_handler = tcp_v4_err,
1714 .no_policy = 1,
1715 .netns_ok = 1,
1716 .icmp_strict_tag_validation = 1,
1717 };
1718
1719 static const struct net_protocol udp_protocol = {
1720 .handler = udp_rcv,
1721 .err_handler = udp_err,
1722 .no_policy = 1,
1723 .netns_ok = 1,
1724 };
1725
1726 static const struct net_protocol icmp_protocol = {
1727 .handler = icmp_rcv,
1728 .err_handler = icmp_err,
1729 .no_policy = 1,
1730 .netns_ok = 1,
1731 };
1732
ipv4_mib_init_net(struct net * net)1733 static __net_init int ipv4_mib_init_net(struct net *net)
1734 {
1735 int i;
1736
1737 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1738 if (!net->mib.tcp_statistics)
1739 goto err_tcp_mib;
1740 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1741 if (!net->mib.ip_statistics)
1742 goto err_ip_mib;
1743
1744 for_each_possible_cpu(i) {
1745 struct ipstats_mib *af_inet_stats;
1746 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1747 u64_stats_init(&af_inet_stats->syncp);
1748 }
1749
1750 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1751 if (!net->mib.net_statistics)
1752 goto err_net_mib;
1753 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1754 if (!net->mib.udp_statistics)
1755 goto err_udp_mib;
1756 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1757 if (!net->mib.udplite_statistics)
1758 goto err_udplite_mib;
1759 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1760 if (!net->mib.icmp_statistics)
1761 goto err_icmp_mib;
1762 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1763 GFP_KERNEL);
1764 if (!net->mib.icmpmsg_statistics)
1765 goto err_icmpmsg_mib;
1766
1767 tcp_mib_init(net);
1768 return 0;
1769
1770 err_icmpmsg_mib:
1771 free_percpu(net->mib.icmp_statistics);
1772 err_icmp_mib:
1773 free_percpu(net->mib.udplite_statistics);
1774 err_udplite_mib:
1775 free_percpu(net->mib.udp_statistics);
1776 err_udp_mib:
1777 free_percpu(net->mib.net_statistics);
1778 err_net_mib:
1779 free_percpu(net->mib.ip_statistics);
1780 err_ip_mib:
1781 free_percpu(net->mib.tcp_statistics);
1782 err_tcp_mib:
1783 return -ENOMEM;
1784 }
1785
ipv4_mib_exit_net(struct net * net)1786 static __net_exit void ipv4_mib_exit_net(struct net *net)
1787 {
1788 kfree(net->mib.icmpmsg_statistics);
1789 free_percpu(net->mib.icmp_statistics);
1790 free_percpu(net->mib.udplite_statistics);
1791 free_percpu(net->mib.udp_statistics);
1792 free_percpu(net->mib.net_statistics);
1793 free_percpu(net->mib.ip_statistics);
1794 free_percpu(net->mib.tcp_statistics);
1795 }
1796
1797 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1798 .init = ipv4_mib_init_net,
1799 .exit = ipv4_mib_exit_net,
1800 };
1801
init_ipv4_mibs(void)1802 static int __init init_ipv4_mibs(void)
1803 {
1804 return register_pernet_subsys(&ipv4_mib_ops);
1805 }
1806
inet_init_net(struct net * net)1807 static __net_init int inet_init_net(struct net *net)
1808 {
1809 /*
1810 * Set defaults for local port range
1811 */
1812 seqlock_init(&net->ipv4.ip_local_ports.lock);
1813 net->ipv4.ip_local_ports.range[0] = 32768;
1814 net->ipv4.ip_local_ports.range[1] = 60999;
1815
1816 seqlock_init(&net->ipv4.ping_group_range.lock);
1817 /*
1818 * Sane defaults - nobody may create ping sockets.
1819 * Boot scripts should set this to distro-specific group.
1820 */
1821 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1822 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1823
1824 /* Default values for sysctl-controlled parameters.
1825 * We set them here, in case sysctl is not compiled.
1826 */
1827 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1828 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1829 net->ipv4.sysctl_ip_dynaddr = 0;
1830 net->ipv4.sysctl_ip_early_demux = 1;
1831 net->ipv4.sysctl_udp_early_demux = 1;
1832 net->ipv4.sysctl_tcp_early_demux = 1;
1833 #ifdef CONFIG_SYSCTL
1834 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1835 #endif
1836
1837 /* Some igmp sysctl, whose values are always used */
1838 net->ipv4.sysctl_igmp_max_memberships = 20;
1839 net->ipv4.sysctl_igmp_max_msf = 10;
1840 /* IGMP reports for link-local multicast groups are enabled by default */
1841 net->ipv4.sysctl_igmp_llm_reports = 1;
1842 net->ipv4.sysctl_igmp_qrv = 2;
1843
1844 return 0;
1845 }
1846
1847 static __net_initdata struct pernet_operations af_inet_ops = {
1848 .init = inet_init_net,
1849 };
1850
init_inet_pernet_ops(void)1851 static int __init init_inet_pernet_ops(void)
1852 {
1853 return register_pernet_subsys(&af_inet_ops);
1854 }
1855
1856 static int ipv4_proc_init(void);
1857
1858 /*
1859 * IP protocol layer initialiser
1860 */
1861
1862 static struct packet_offload ip_packet_offload __read_mostly = {
1863 .type = cpu_to_be16(ETH_P_IP),
1864 .callbacks = {
1865 .gso_segment = inet_gso_segment,
1866 .gro_receive = inet_gro_receive,
1867 .gro_complete = inet_gro_complete,
1868 },
1869 };
1870
1871 static const struct net_offload ipip_offload = {
1872 .callbacks = {
1873 .gso_segment = ipip_gso_segment,
1874 .gro_receive = ipip_gro_receive,
1875 .gro_complete = ipip_gro_complete,
1876 },
1877 };
1878
ipip_offload_init(void)1879 static int __init ipip_offload_init(void)
1880 {
1881 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1882 }
1883
ipv4_offload_init(void)1884 static int __init ipv4_offload_init(void)
1885 {
1886 /*
1887 * Add offloads
1888 */
1889 if (udpv4_offload_init() < 0)
1890 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1891 if (tcpv4_offload_init() < 0)
1892 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1893 if (ipip_offload_init() < 0)
1894 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1895
1896 dev_add_offload(&ip_packet_offload);
1897 return 0;
1898 }
1899
1900 fs_initcall(ipv4_offload_init);
1901
1902 static struct packet_type ip_packet_type __read_mostly = {
1903 .type = cpu_to_be16(ETH_P_IP),
1904 .func = ip_rcv,
1905 .list_func = ip_list_rcv,
1906 };
1907
inet_init(void)1908 static int __init inet_init(void)
1909 {
1910 struct inet_protosw *q;
1911 struct list_head *r;
1912 int rc = -EINVAL;
1913
1914 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1915
1916 rc = proto_register(&tcp_prot, 1);
1917 if (rc)
1918 goto out;
1919
1920 rc = proto_register(&udp_prot, 1);
1921 if (rc)
1922 goto out_unregister_tcp_proto;
1923
1924 rc = proto_register(&raw_prot, 1);
1925 if (rc)
1926 goto out_unregister_udp_proto;
1927
1928 rc = proto_register(&ping_prot, 1);
1929 if (rc)
1930 goto out_unregister_raw_proto;
1931
1932 /*
1933 * Tell SOCKET that we are alive...
1934 */
1935
1936 (void)sock_register(&inet_family_ops);
1937
1938 #ifdef CONFIG_SYSCTL
1939 ip_static_sysctl_init();
1940 #endif
1941
1942 /*
1943 * Add all the base protocols.
1944 */
1945
1946 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1947 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1948 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1949 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1950 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1951 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1952 #ifdef CONFIG_IP_MULTICAST
1953 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1954 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1955 #endif
1956
1957 /* Register the socket-side information for inet_create. */
1958 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1959 INIT_LIST_HEAD(r);
1960
1961 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1962 inet_register_protosw(q);
1963
1964 /*
1965 * Set the ARP module up
1966 */
1967
1968 arp_init();
1969
1970 /*
1971 * Set the IP module up
1972 */
1973
1974 ip_init();
1975
1976 /* Initialise per-cpu ipv4 mibs */
1977 if (init_ipv4_mibs())
1978 panic("%s: Cannot init ipv4 mibs\n", __func__);
1979
1980 /* Setup TCP slab cache for open requests. */
1981 tcp_init();
1982
1983 /* Setup UDP memory threshold */
1984 udp_init();
1985
1986 /* Add UDP-Lite (RFC 3828) */
1987 udplite4_register();
1988
1989 raw_init();
1990
1991 ping_init();
1992
1993 /*
1994 * Set the ICMP layer up
1995 */
1996
1997 if (icmp_init() < 0)
1998 panic("Failed to create the ICMP control socket.\n");
1999
2000 /*
2001 * Initialise the multicast router
2002 */
2003 #if defined(CONFIG_IP_MROUTE)
2004 if (ip_mr_init())
2005 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2006 #endif
2007
2008 if (init_inet_pernet_ops())
2009 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2010
2011 ipv4_proc_init();
2012
2013 ipfrag_init();
2014
2015 dev_add_pack(&ip_packet_type);
2016
2017 ip_tunnel_core_init();
2018
2019 rc = 0;
2020 out:
2021 return rc;
2022 out_unregister_raw_proto:
2023 proto_unregister(&raw_prot);
2024 out_unregister_udp_proto:
2025 proto_unregister(&udp_prot);
2026 out_unregister_tcp_proto:
2027 proto_unregister(&tcp_prot);
2028 goto out;
2029 }
2030
2031 fs_initcall(inet_init);
2032
2033 /* ------------------------------------------------------------------------ */
2034
2035 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2036 static int __init ipv4_proc_init(void)
2037 {
2038 int rc = 0;
2039
2040 if (raw_proc_init())
2041 goto out_raw;
2042 if (tcp4_proc_init())
2043 goto out_tcp;
2044 if (udp4_proc_init())
2045 goto out_udp;
2046 if (ping_proc_init())
2047 goto out_ping;
2048 if (ip_misc_proc_init())
2049 goto out_misc;
2050 out:
2051 return rc;
2052 out_misc:
2053 ping_proc_exit();
2054 out_ping:
2055 udp4_proc_exit();
2056 out_udp:
2057 tcp4_proc_exit();
2058 out_tcp:
2059 raw_proc_exit();
2060 out_raw:
2061 rc = -ENOMEM;
2062 goto out;
2063 }
2064
2065 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2066 static int __init ipv4_proc_init(void)
2067 {
2068 return 0;
2069 }
2070 #endif /* CONFIG_PROC_FS */
2071