1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PF_INET protocol family socket handler.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Florian La Roche, <flla@stud.uni-sb.de>
12 * Alan Cox, <A.Cox@swansea.ac.uk>
13 *
14 * Changes (see also sock.c)
15 *
16 * piggy,
17 * Karl Knutson : Socket protocol table
18 * A.N.Kuznetsov : Socket death error in accept().
19 * John Richardson : Fix non blocking error in connect()
20 * so sockets that fail to connect
21 * don't return -EINPROGRESS.
22 * Alan Cox : Asynchronous I/O support
23 * Alan Cox : Keep correct socket pointer on sock
24 * structures
25 * when accept() ed
26 * Alan Cox : Semantics of SO_LINGER aren't state
27 * moved to close when you look carefully.
28 * With this fixed and the accept bug fixed
29 * some RPC stuff seems happier.
30 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Alan Cox,
32 * Tony Gale : Fixed reuse semantics.
33 * Alan Cox : bind() shouldn't abort existing but dead
34 * sockets. Stops FTP netin:.. I hope.
35 * Alan Cox : bind() works correctly for RAW sockets.
36 * Note that FreeBSD at least was broken
37 * in this respect so be careful with
38 * compatibility tests...
39 * Alan Cox : routing cache support
40 * Alan Cox : memzero the socket structure for
41 * compactness.
42 * Matt Day : nonblock connect error handler
43 * Alan Cox : Allow large numbers of pending sockets
44 * (eg for big web sites), but only if
45 * specifically application requested.
46 * Alan Cox : New buffering throughout IP. Used
47 * dumbly.
48 * Alan Cox : New buffering now used smartly.
49 * Alan Cox : BSD rather than common sense
50 * interpretation of listen.
51 * Germano Caronni : Assorted small races.
52 * Alan Cox : sendmsg/recvmsg basic support.
53 * Alan Cox : Only sendmsg/recvmsg now supported.
54 * Alan Cox : Locked down bind (see security list).
55 * Alan Cox : Loosened bind a little.
56 * Mike McLagan : ADD/DEL DLCI Ioctls
57 * Willy Konynenberg : Transparent proxying support.
58 * David S. Miller : New socket lookup architecture.
59 * Some other random speedups.
60 * Cyrus Durgin : Cleaned up file for kmod hacks.
61 * Andi Kleen : Fix inet_stream_connect TCP race.
62 */
63
64 #define pr_fmt(fmt) "IPv4: " fmt
65
66 #include <linux/err.h>
67 #include <linux/errno.h>
68 #include <linux/types.h>
69 #include <linux/socket.h>
70 #include <linux/in.h>
71 #include <linux/kernel.h>
72 #include <linux/kmod.h>
73 #include <linux/sched.h>
74 #include <linux/timer.h>
75 #include <linux/string.h>
76 #include <linux/sockios.h>
77 #include <linux/net.h>
78 #include <linux/capability.h>
79 #include <linux/fcntl.h>
80 #include <linux/mm.h>
81 #include <linux/interrupt.h>
82 #include <linux/stat.h>
83 #include <linux/init.h>
84 #include <linux/poll.h>
85 #include <linux/netfilter_ipv4.h>
86 #include <linux/random.h>
87 #include <linux/slab.h>
88
89 #include <linux/uaccess.h>
90
91 #include <linux/inet.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/netdevice.h>
95 #include <net/checksum.h>
96 #include <net/ip.h>
97 #include <net/protocol.h>
98 #include <net/arp.h>
99 #include <net/route.h>
100 #include <net/ip_fib.h>
101 #include <net/inet_connection_sock.h>
102 #include <net/gro.h>
103 #include <net/gso.h>
104 #include <net/tcp.h>
105 #include <net/udp.h>
106 #include <net/udplite.h>
107 #include <net/ping.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
110 #include <net/raw.h>
111 #include <net/icmp.h>
112 #include <net/inet_common.h>
113 #include <net/ip_tunnels.h>
114 #include <net/xfrm.h>
115 #include <net/net_namespace.h>
116 #include <net/secure_seq.h>
117 #ifdef CONFIG_IP_MROUTE
118 #include <linux/mroute.h>
119 #endif
120 #include <net/l3mdev.h>
121 #include <net/compat.h>
122 #include <net/rps.h>
123
124 #include <trace/events/sock.h>
125
126 /* The inetsw table contains everything that inet_create needs to
127 * build a new socket.
128 */
129 static struct list_head inetsw[SOCK_MAX];
130 static DEFINE_SPINLOCK(inetsw_lock);
131
132 /* New destruction routine */
133
inet_sock_destruct(struct sock * sk)134 void inet_sock_destruct(struct sock *sk)
135 {
136 struct inet_sock *inet = inet_sk(sk);
137
138 __skb_queue_purge(&sk->sk_receive_queue);
139 __skb_queue_purge(&sk->sk_error_queue);
140
141 sk_mem_reclaim_final(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_ONCE(atomic_read(&sk->sk_rmem_alloc));
154 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
155 WARN_ON_ONCE(sk->sk_wmem_queued);
156 WARN_ON_ONCE(sk_forward_alloc_get(sk));
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 }
162 EXPORT_SYMBOL(inet_sock_destruct);
163
164 /*
165 * The routines beyond this point handle the behaviour of an AF_INET
166 * socket object. Mostly it punts to the subprotocols of IP to do
167 * the work.
168 */
169
170 /*
171 * Automatically bind an unbound socket.
172 */
173
inet_autobind(struct sock * sk)174 static int inet_autobind(struct sock *sk)
175 {
176 struct inet_sock *inet;
177 /* We may need to bind the socket. */
178 lock_sock(sk);
179 inet = inet_sk(sk);
180 if (!inet->inet_num) {
181 if (sk->sk_prot->get_port(sk, 0)) {
182 release_sock(sk);
183 return -EAGAIN;
184 }
185 inet->inet_sport = htons(inet->inet_num);
186 }
187 release_sock(sk);
188 return 0;
189 }
190
__inet_listen_sk(struct sock * sk,int backlog)191 int __inet_listen_sk(struct sock *sk, int backlog)
192 {
193 unsigned char old_state = sk->sk_state;
194 int err, tcp_fastopen;
195
196 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
197 return -EINVAL;
198
199 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
200 /* Really, if the socket is already in listen state
201 * we can only allow the backlog to be adjusted.
202 */
203 if (old_state != TCP_LISTEN) {
204 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
205 * Note that only TCP sockets (SOCK_STREAM) will reach here.
206 * Also fastopen backlog may already been set via the option
207 * because the socket was in TCP_LISTEN state previously but
208 * was shutdown() rather than close().
209 */
210 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
211 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
212 (tcp_fastopen & TFO_SERVER_ENABLE) &&
213 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
214 fastopen_queue_tune(sk, backlog);
215 tcp_fastopen_init_key_once(sock_net(sk));
216 }
217
218 err = inet_csk_listen_start(sk);
219 if (err)
220 return err;
221
222 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
223 }
224 return 0;
225 }
226
227 /*
228 * Move a socket into listening state.
229 */
inet_listen(struct socket * sock,int backlog)230 int inet_listen(struct socket *sock, int backlog)
231 {
232 struct sock *sk = sock->sk;
233 int err = -EINVAL;
234
235 lock_sock(sk);
236
237 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
238 goto out;
239
240 err = __inet_listen_sk(sk, backlog);
241
242 out:
243 release_sock(sk);
244 return err;
245 }
246 EXPORT_SYMBOL(inet_listen);
247
248 /*
249 * Create an inet socket.
250 */
251
inet_create(struct net * net,struct socket * sock,int protocol,int kern)252 static int inet_create(struct net *net, struct socket *sock, int protocol,
253 int kern)
254 {
255 struct sock *sk;
256 struct inet_protosw *answer;
257 struct inet_sock *inet;
258 struct proto *answer_prot;
259 unsigned char answer_flags;
260 int try_loading_module = 0;
261 int err;
262
263 if (protocol < 0 || protocol >= IPPROTO_MAX)
264 return -EINVAL;
265
266 sock->state = SS_UNCONNECTED;
267
268 /* Look for the requested type/protocol pair. */
269 lookup_protocol:
270 err = -ESOCKTNOSUPPORT;
271 rcu_read_lock();
272 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
273
274 err = 0;
275 /* Check the non-wild match. */
276 if (protocol == answer->protocol) {
277 if (protocol != IPPROTO_IP)
278 break;
279 } else {
280 /* Check for the two wild cases. */
281 if (IPPROTO_IP == protocol) {
282 protocol = answer->protocol;
283 break;
284 }
285 if (IPPROTO_IP == answer->protocol)
286 break;
287 }
288 err = -EPROTONOSUPPORT;
289 }
290
291 if (unlikely(err)) {
292 if (try_loading_module < 2) {
293 rcu_read_unlock();
294 /*
295 * Be more specific, e.g. net-pf-2-proto-132-type-1
296 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
297 */
298 if (++try_loading_module == 1)
299 request_module("net-pf-%d-proto-%d-type-%d",
300 PF_INET, protocol, sock->type);
301 /*
302 * Fall back to generic, e.g. net-pf-2-proto-132
303 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
304 */
305 else
306 request_module("net-pf-%d-proto-%d",
307 PF_INET, protocol);
308 goto lookup_protocol;
309 } else
310 goto out_rcu_unlock;
311 }
312
313 err = -EPERM;
314 if (sock->type == SOCK_RAW && !kern &&
315 !ns_capable(net->user_ns, CAP_NET_RAW))
316 goto out_rcu_unlock;
317
318 sock->ops = answer->ops;
319 answer_prot = answer->prot;
320 answer_flags = answer->flags;
321 rcu_read_unlock();
322
323 WARN_ON(!answer_prot->slab);
324
325 err = -ENOMEM;
326 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
327 if (!sk)
328 goto out;
329
330 err = 0;
331 if (INET_PROTOSW_REUSE & answer_flags)
332 sk->sk_reuse = SK_CAN_REUSE;
333
334 if (INET_PROTOSW_ICSK & answer_flags)
335 inet_init_csk_locks(sk);
336
337 inet = inet_sk(sk);
338 inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
339
340 inet_clear_bit(NODEFRAG, sk);
341
342 if (SOCK_RAW == sock->type) {
343 inet->inet_num = protocol;
344 if (IPPROTO_RAW == protocol)
345 inet_set_bit(HDRINCL, sk);
346 }
347
348 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
349 inet->pmtudisc = IP_PMTUDISC_DONT;
350 else
351 inet->pmtudisc = IP_PMTUDISC_WANT;
352
353 atomic_set(&inet->inet_id, 0);
354
355 sock_init_data(sock, sk);
356
357 sk->sk_destruct = inet_sock_destruct;
358 sk->sk_protocol = protocol;
359 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
360 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
361
362 inet->uc_ttl = -1;
363 inet_set_bit(MC_LOOP, sk);
364 inet->mc_ttl = 1;
365 inet_set_bit(MC_ALL, sk);
366 inet->mc_index = 0;
367 inet->mc_list = NULL;
368 inet->rcv_tos = 0;
369
370 if (inet->inet_num) {
371 /* It assumes that any protocol which allows
372 * the user to assign a number at socket
373 * creation time automatically
374 * shares.
375 */
376 inet->inet_sport = htons(inet->inet_num);
377 /* Add to protocol hash chains. */
378 err = sk->sk_prot->hash(sk);
379 if (err) {
380 sk_common_release(sk);
381 goto out;
382 }
383 }
384
385 if (sk->sk_prot->init) {
386 err = sk->sk_prot->init(sk);
387 if (err) {
388 sk_common_release(sk);
389 goto out;
390 }
391 }
392
393 if (!kern) {
394 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
395 if (err) {
396 sk_common_release(sk);
397 goto out;
398 }
399 }
400 out:
401 return err;
402 out_rcu_unlock:
403 rcu_read_unlock();
404 goto out;
405 }
406
407
408 /*
409 * The peer socket should always be NULL (or else). When we call this
410 * function we are destroying the object and from then on nobody
411 * should refer to it.
412 */
inet_release(struct socket * sock)413 int inet_release(struct socket *sock)
414 {
415 struct sock *sk = sock->sk;
416
417 if (sk) {
418 long timeout;
419
420 if (!sk->sk_kern_sock)
421 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
422
423 /* Applications forget to leave groups before exiting */
424 ip_mc_drop_socket(sk);
425
426 /* If linger is set, we don't return until the close
427 * is complete. Otherwise we return immediately. The
428 * actually closing is done the same either way.
429 *
430 * If the close is due to the process exiting, we never
431 * linger..
432 */
433 timeout = 0;
434 if (sock_flag(sk, SOCK_LINGER) &&
435 !(current->flags & PF_EXITING))
436 timeout = sk->sk_lingertime;
437 sk->sk_prot->close(sk, timeout);
438 sock->sk = NULL;
439 }
440 return 0;
441 }
442 EXPORT_SYMBOL(inet_release);
443
inet_bind_sk(struct sock * sk,struct sockaddr * uaddr,int addr_len)444 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
445 {
446 u32 flags = BIND_WITH_LOCK;
447 int err;
448
449 /* If the socket has its own bind function then use it. (RAW) */
450 if (sk->sk_prot->bind) {
451 return sk->sk_prot->bind(sk, uaddr, addr_len);
452 }
453 if (addr_len < sizeof(struct sockaddr_in))
454 return -EINVAL;
455
456 /* BPF prog is run before any checks are done so that if the prog
457 * changes context in a wrong way it will be caught.
458 */
459 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
460 CGROUP_INET4_BIND, &flags);
461 if (err)
462 return err;
463
464 return __inet_bind(sk, uaddr, addr_len, flags);
465 }
466
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)467 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
468 {
469 return inet_bind_sk(sock->sk, uaddr, addr_len);
470 }
471 EXPORT_SYMBOL(inet_bind);
472
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)473 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
474 u32 flags)
475 {
476 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
477 struct inet_sock *inet = inet_sk(sk);
478 struct net *net = sock_net(sk);
479 unsigned short snum;
480 int chk_addr_ret;
481 u32 tb_id = RT_TABLE_LOCAL;
482 int err;
483
484 if (addr->sin_family != AF_INET) {
485 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
486 * only if s_addr is INADDR_ANY.
487 */
488 err = -EAFNOSUPPORT;
489 if (addr->sin_family != AF_UNSPEC ||
490 addr->sin_addr.s_addr != htonl(INADDR_ANY))
491 goto out;
492 }
493
494 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
495 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
496
497 /* Not specified by any standard per-se, however it breaks too
498 * many applications when removed. It is unfortunate since
499 * allowing applications to make a non-local bind solves
500 * several problems with systems using dynamic addressing.
501 * (ie. your servers still start up even if your ISDN link
502 * is temporarily down)
503 */
504 err = -EADDRNOTAVAIL;
505 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
506 chk_addr_ret))
507 goto out;
508
509 snum = ntohs(addr->sin_port);
510 err = -EACCES;
511 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
512 snum && inet_port_requires_bind_service(net, snum) &&
513 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
514 goto out;
515
516 /* We keep a pair of addresses. rcv_saddr is the one
517 * used by hash lookups, and saddr is used for transmit.
518 *
519 * In the BSD API these are the same except where it
520 * would be illegal to use them (multicast/broadcast) in
521 * which case the sending device address is used.
522 */
523 if (flags & BIND_WITH_LOCK)
524 lock_sock(sk);
525
526 /* Check these errors (active socket, double bind). */
527 err = -EINVAL;
528 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
529 goto out_release_sock;
530
531 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
532 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
533 inet->inet_saddr = 0; /* Use device */
534
535 /* Make sure we are allowed to bind here. */
536 if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
537 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
538 err = sk->sk_prot->get_port(sk, snum);
539 if (err) {
540 inet->inet_saddr = inet->inet_rcv_saddr = 0;
541 goto out_release_sock;
542 }
543 if (!(flags & BIND_FROM_BPF)) {
544 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
545 if (err) {
546 inet->inet_saddr = inet->inet_rcv_saddr = 0;
547 if (sk->sk_prot->put_port)
548 sk->sk_prot->put_port(sk);
549 goto out_release_sock;
550 }
551 }
552 }
553
554 if (inet->inet_rcv_saddr)
555 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
556 if (snum)
557 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
558 inet->inet_sport = htons(inet->inet_num);
559 inet->inet_daddr = 0;
560 inet->inet_dport = 0;
561 sk_dst_reset(sk);
562 err = 0;
563 out_release_sock:
564 if (flags & BIND_WITH_LOCK)
565 release_sock(sk);
566 out:
567 return err;
568 }
569
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)570 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
571 int addr_len, int flags)
572 {
573 struct sock *sk = sock->sk;
574 const struct proto *prot;
575 int err;
576
577 if (addr_len < sizeof(uaddr->sa_family))
578 return -EINVAL;
579
580 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
581 prot = READ_ONCE(sk->sk_prot);
582
583 if (uaddr->sa_family == AF_UNSPEC)
584 return prot->disconnect(sk, flags);
585
586 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
587 err = prot->pre_connect(sk, uaddr, addr_len);
588 if (err)
589 return err;
590 }
591
592 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
593 return -EAGAIN;
594 return prot->connect(sk, uaddr, addr_len);
595 }
596 EXPORT_SYMBOL(inet_dgram_connect);
597
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)598 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
599 {
600 DEFINE_WAIT_FUNC(wait, woken_wake_function);
601
602 add_wait_queue(sk_sleep(sk), &wait);
603 sk->sk_write_pending += writebias;
604
605 /* Basic assumption: if someone sets sk->sk_err, he _must_
606 * change state of the socket from TCP_SYN_*.
607 * Connect() does not allow to get error notifications
608 * without closing the socket.
609 */
610 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
611 release_sock(sk);
612 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
613 lock_sock(sk);
614 if (signal_pending(current) || !timeo)
615 break;
616 }
617 remove_wait_queue(sk_sleep(sk), &wait);
618 sk->sk_write_pending -= writebias;
619 return timeo;
620 }
621
622 /*
623 * Connect to a remote host. There is regrettably still a little
624 * TCP 'magic' in here.
625 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)626 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
627 int addr_len, int flags, int is_sendmsg)
628 {
629 struct sock *sk = sock->sk;
630 int err;
631 long timeo;
632
633 /*
634 * uaddr can be NULL and addr_len can be 0 if:
635 * sk is a TCP fastopen active socket and
636 * TCP_FASTOPEN_CONNECT sockopt is set and
637 * we already have a valid cookie for this socket.
638 * In this case, user can call write() after connect().
639 * write() will invoke tcp_sendmsg_fastopen() which calls
640 * __inet_stream_connect().
641 */
642 if (uaddr) {
643 if (addr_len < sizeof(uaddr->sa_family))
644 return -EINVAL;
645
646 if (uaddr->sa_family == AF_UNSPEC) {
647 sk->sk_disconnects++;
648 err = sk->sk_prot->disconnect(sk, flags);
649 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
650 goto out;
651 }
652 }
653
654 switch (sock->state) {
655 default:
656 err = -EINVAL;
657 goto out;
658 case SS_CONNECTED:
659 err = -EISCONN;
660 goto out;
661 case SS_CONNECTING:
662 if (inet_test_bit(DEFER_CONNECT, sk))
663 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
664 else
665 err = -EALREADY;
666 /* Fall out of switch with err, set for this state */
667 break;
668 case SS_UNCONNECTED:
669 err = -EISCONN;
670 if (sk->sk_state != TCP_CLOSE)
671 goto out;
672
673 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
674 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
675 if (err)
676 goto out;
677 }
678
679 err = sk->sk_prot->connect(sk, uaddr, addr_len);
680 if (err < 0)
681 goto out;
682
683 sock->state = SS_CONNECTING;
684
685 if (!err && inet_test_bit(DEFER_CONNECT, sk))
686 goto out;
687
688 /* Just entered SS_CONNECTING state; the only
689 * difference is that return value in non-blocking
690 * case is EINPROGRESS, rather than EALREADY.
691 */
692 err = -EINPROGRESS;
693 break;
694 }
695
696 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
697
698 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
699 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
700 tcp_sk(sk)->fastopen_req &&
701 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
702 int dis = sk->sk_disconnects;
703
704 /* Error code is set above */
705 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
706 goto out;
707
708 err = sock_intr_errno(timeo);
709 if (signal_pending(current))
710 goto out;
711
712 if (dis != sk->sk_disconnects) {
713 err = -EPIPE;
714 goto out;
715 }
716 }
717
718 /* Connection was closed by RST, timeout, ICMP error
719 * or another process disconnected us.
720 */
721 if (sk->sk_state == TCP_CLOSE)
722 goto sock_error;
723
724 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
725 * and error was received after socket entered established state.
726 * Hence, it is handled normally after connect() return successfully.
727 */
728
729 sock->state = SS_CONNECTED;
730 err = 0;
731 out:
732 return err;
733
734 sock_error:
735 err = sock_error(sk) ? : -ECONNABORTED;
736 sock->state = SS_UNCONNECTED;
737 sk->sk_disconnects++;
738 if (sk->sk_prot->disconnect(sk, flags))
739 sock->state = SS_DISCONNECTING;
740 goto out;
741 }
742 EXPORT_SYMBOL(__inet_stream_connect);
743
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)744 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
745 int addr_len, int flags)
746 {
747 int err;
748
749 lock_sock(sock->sk);
750 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
751 release_sock(sock->sk);
752 return err;
753 }
754 EXPORT_SYMBOL(inet_stream_connect);
755
__inet_accept(struct socket * sock,struct socket * newsock,struct sock * newsk)756 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
757 {
758 sock_rps_record_flow(newsk);
759 WARN_ON(!((1 << newsk->sk_state) &
760 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
761 TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
762 TCPF_CLOSING | TCPF_CLOSE_WAIT |
763 TCPF_CLOSE)));
764
765 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
766 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
767 sock_graft(newsk, newsock);
768
769 newsock->state = SS_CONNECTED;
770 }
771
772 /*
773 * Accept a pending connection. The TCP layer now gives BSD semantics.
774 */
775
inet_accept(struct socket * sock,struct socket * newsock,struct proto_accept_arg * arg)776 int inet_accept(struct socket *sock, struct socket *newsock,
777 struct proto_accept_arg *arg)
778 {
779 struct sock *sk1 = sock->sk, *sk2;
780
781 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
782 arg->err = -EINVAL;
783 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, arg);
784 if (!sk2)
785 return arg->err;
786
787 lock_sock(sk2);
788 __inet_accept(sock, newsock, sk2);
789 release_sock(sk2);
790 return 0;
791 }
792 EXPORT_SYMBOL(inet_accept);
793
794 /*
795 * This does both peername and sockname.
796 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)797 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
798 int peer)
799 {
800 struct sock *sk = sock->sk;
801 struct inet_sock *inet = inet_sk(sk);
802 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
803 int sin_addr_len = sizeof(*sin);
804
805 sin->sin_family = AF_INET;
806 lock_sock(sk);
807 if (peer) {
808 if (!inet->inet_dport ||
809 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
810 peer == 1)) {
811 release_sock(sk);
812 return -ENOTCONN;
813 }
814 sin->sin_port = inet->inet_dport;
815 sin->sin_addr.s_addr = inet->inet_daddr;
816 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
817 CGROUP_INET4_GETPEERNAME);
818 } else {
819 __be32 addr = inet->inet_rcv_saddr;
820 if (!addr)
821 addr = inet->inet_saddr;
822 sin->sin_port = inet->inet_sport;
823 sin->sin_addr.s_addr = addr;
824 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
825 CGROUP_INET4_GETSOCKNAME);
826 }
827 release_sock(sk);
828 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
829 return sin_addr_len;
830 }
831 EXPORT_SYMBOL(inet_getname);
832
inet_send_prepare(struct sock * sk)833 int inet_send_prepare(struct sock *sk)
834 {
835 sock_rps_record_flow(sk);
836
837 /* We may need to bind the socket. */
838 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
839 inet_autobind(sk))
840 return -EAGAIN;
841
842 return 0;
843 }
844 EXPORT_SYMBOL_GPL(inet_send_prepare);
845
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)846 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
847 {
848 struct sock *sk = sock->sk;
849
850 if (unlikely(inet_send_prepare(sk)))
851 return -EAGAIN;
852
853 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
854 sk, msg, size);
855 }
856 EXPORT_SYMBOL(inet_sendmsg);
857
inet_splice_eof(struct socket * sock)858 void inet_splice_eof(struct socket *sock)
859 {
860 const struct proto *prot;
861 struct sock *sk = sock->sk;
862
863 if (unlikely(inet_send_prepare(sk)))
864 return;
865
866 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
867 prot = READ_ONCE(sk->sk_prot);
868 if (prot->splice_eof)
869 prot->splice_eof(sock);
870 }
871 EXPORT_SYMBOL_GPL(inet_splice_eof);
872
873 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
874 size_t, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)875 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
876 int flags)
877 {
878 struct sock *sk = sock->sk;
879 int addr_len = 0;
880 int err;
881
882 if (likely(!(flags & MSG_ERRQUEUE)))
883 sock_rps_record_flow(sk);
884
885 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
886 sk, msg, size, flags, &addr_len);
887 if (err >= 0)
888 msg->msg_namelen = addr_len;
889 return err;
890 }
891 EXPORT_SYMBOL(inet_recvmsg);
892
inet_shutdown(struct socket * sock,int how)893 int inet_shutdown(struct socket *sock, int how)
894 {
895 struct sock *sk = sock->sk;
896 int err = 0;
897
898 /* This should really check to make sure
899 * the socket is a TCP socket. (WHY AC...)
900 */
901 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
902 1->2 bit 2 snds.
903 2->3 */
904 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
905 return -EINVAL;
906
907 lock_sock(sk);
908 if (sock->state == SS_CONNECTING) {
909 if ((1 << sk->sk_state) &
910 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
911 sock->state = SS_DISCONNECTING;
912 else
913 sock->state = SS_CONNECTED;
914 }
915
916 switch (sk->sk_state) {
917 case TCP_CLOSE:
918 err = -ENOTCONN;
919 /* Hack to wake up other listeners, who can poll for
920 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
921 fallthrough;
922 default:
923 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
924 if (sk->sk_prot->shutdown)
925 sk->sk_prot->shutdown(sk, how);
926 break;
927
928 /* Remaining two branches are temporary solution for missing
929 * close() in multithreaded environment. It is _not_ a good idea,
930 * but we have no choice until close() is repaired at VFS level.
931 */
932 case TCP_LISTEN:
933 if (!(how & RCV_SHUTDOWN))
934 break;
935 fallthrough;
936 case TCP_SYN_SENT:
937 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
938 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
939 break;
940 }
941
942 /* Wake up anyone sleeping in poll. */
943 sk->sk_state_change(sk);
944 release_sock(sk);
945 return err;
946 }
947 EXPORT_SYMBOL(inet_shutdown);
948
949 /*
950 * ioctl() calls you can issue on an INET socket. Most of these are
951 * device configuration and stuff and very rarely used. Some ioctls
952 * pass on to the socket itself.
953 *
954 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
955 * loads the devconfigure module does its configuring and unloads it.
956 * There's a good 20K of config code hanging around the kernel.
957 */
958
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)959 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
960 {
961 struct sock *sk = sock->sk;
962 int err = 0;
963 struct net *net = sock_net(sk);
964 void __user *p = (void __user *)arg;
965 struct ifreq ifr;
966 struct rtentry rt;
967
968 switch (cmd) {
969 case SIOCADDRT:
970 case SIOCDELRT:
971 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
972 return -EFAULT;
973 err = ip_rt_ioctl(net, cmd, &rt);
974 break;
975 case SIOCRTMSG:
976 err = -EINVAL;
977 break;
978 case SIOCDARP:
979 case SIOCGARP:
980 case SIOCSARP:
981 err = arp_ioctl(net, cmd, (void __user *)arg);
982 break;
983 case SIOCGIFADDR:
984 case SIOCGIFBRDADDR:
985 case SIOCGIFNETMASK:
986 case SIOCGIFDSTADDR:
987 case SIOCGIFPFLAGS:
988 if (get_user_ifreq(&ifr, NULL, p))
989 return -EFAULT;
990 err = devinet_ioctl(net, cmd, &ifr);
991 if (!err && put_user_ifreq(&ifr, p))
992 err = -EFAULT;
993 break;
994
995 case SIOCSIFADDR:
996 case SIOCSIFBRDADDR:
997 case SIOCSIFNETMASK:
998 case SIOCSIFDSTADDR:
999 case SIOCSIFPFLAGS:
1000 case SIOCSIFFLAGS:
1001 if (get_user_ifreq(&ifr, NULL, p))
1002 return -EFAULT;
1003 err = devinet_ioctl(net, cmd, &ifr);
1004 break;
1005 default:
1006 if (sk->sk_prot->ioctl)
1007 err = sk_ioctl(sk, cmd, (void __user *)arg);
1008 else
1009 err = -ENOIOCTLCMD;
1010 break;
1011 }
1012 return err;
1013 }
1014 EXPORT_SYMBOL(inet_ioctl);
1015
1016 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)1017 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1018 struct compat_rtentry __user *ur)
1019 {
1020 compat_uptr_t rtdev;
1021 struct rtentry rt;
1022
1023 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1024 3 * sizeof(struct sockaddr)) ||
1025 get_user(rt.rt_flags, &ur->rt_flags) ||
1026 get_user(rt.rt_metric, &ur->rt_metric) ||
1027 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1028 get_user(rt.rt_window, &ur->rt_window) ||
1029 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1030 get_user(rtdev, &ur->rt_dev))
1031 return -EFAULT;
1032
1033 rt.rt_dev = compat_ptr(rtdev);
1034 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1035 }
1036
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1037 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1038 {
1039 void __user *argp = compat_ptr(arg);
1040 struct sock *sk = sock->sk;
1041
1042 switch (cmd) {
1043 case SIOCADDRT:
1044 case SIOCDELRT:
1045 return inet_compat_routing_ioctl(sk, cmd, argp);
1046 default:
1047 if (!sk->sk_prot->compat_ioctl)
1048 return -ENOIOCTLCMD;
1049 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1050 }
1051 }
1052 #endif /* CONFIG_COMPAT */
1053
1054 const struct proto_ops inet_stream_ops = {
1055 .family = PF_INET,
1056 .owner = THIS_MODULE,
1057 .release = inet_release,
1058 .bind = inet_bind,
1059 .connect = inet_stream_connect,
1060 .socketpair = sock_no_socketpair,
1061 .accept = inet_accept,
1062 .getname = inet_getname,
1063 .poll = tcp_poll,
1064 .ioctl = inet_ioctl,
1065 .gettstamp = sock_gettstamp,
1066 .listen = inet_listen,
1067 .shutdown = inet_shutdown,
1068 .setsockopt = sock_common_setsockopt,
1069 .getsockopt = sock_common_getsockopt,
1070 .sendmsg = inet_sendmsg,
1071 .recvmsg = inet_recvmsg,
1072 #ifdef CONFIG_MMU
1073 .mmap = tcp_mmap,
1074 #endif
1075 .splice_eof = inet_splice_eof,
1076 .splice_read = tcp_splice_read,
1077 .set_peek_off = sk_set_peek_off,
1078 .read_sock = tcp_read_sock,
1079 .read_skb = tcp_read_skb,
1080 .sendmsg_locked = tcp_sendmsg_locked,
1081 .peek_len = tcp_peek_len,
1082 #ifdef CONFIG_COMPAT
1083 .compat_ioctl = inet_compat_ioctl,
1084 #endif
1085 .set_rcvlowat = tcp_set_rcvlowat,
1086 };
1087 EXPORT_SYMBOL(inet_stream_ops);
1088
1089 const struct proto_ops inet_dgram_ops = {
1090 .family = PF_INET,
1091 .owner = THIS_MODULE,
1092 .release = inet_release,
1093 .bind = inet_bind,
1094 .connect = inet_dgram_connect,
1095 .socketpair = sock_no_socketpair,
1096 .accept = sock_no_accept,
1097 .getname = inet_getname,
1098 .poll = udp_poll,
1099 .ioctl = inet_ioctl,
1100 .gettstamp = sock_gettstamp,
1101 .listen = sock_no_listen,
1102 .shutdown = inet_shutdown,
1103 .setsockopt = sock_common_setsockopt,
1104 .getsockopt = sock_common_getsockopt,
1105 .sendmsg = inet_sendmsg,
1106 .read_skb = udp_read_skb,
1107 .recvmsg = inet_recvmsg,
1108 .mmap = sock_no_mmap,
1109 .splice_eof = inet_splice_eof,
1110 .set_peek_off = udp_set_peek_off,
1111 #ifdef CONFIG_COMPAT
1112 .compat_ioctl = inet_compat_ioctl,
1113 #endif
1114 };
1115 EXPORT_SYMBOL(inet_dgram_ops);
1116
1117 /*
1118 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1119 * udp_poll
1120 */
1121 static const struct proto_ops inet_sockraw_ops = {
1122 .family = PF_INET,
1123 .owner = THIS_MODULE,
1124 .release = inet_release,
1125 .bind = inet_bind,
1126 .connect = inet_dgram_connect,
1127 .socketpair = sock_no_socketpair,
1128 .accept = sock_no_accept,
1129 .getname = inet_getname,
1130 .poll = datagram_poll,
1131 .ioctl = inet_ioctl,
1132 .gettstamp = sock_gettstamp,
1133 .listen = sock_no_listen,
1134 .shutdown = inet_shutdown,
1135 .setsockopt = sock_common_setsockopt,
1136 .getsockopt = sock_common_getsockopt,
1137 .sendmsg = inet_sendmsg,
1138 .recvmsg = inet_recvmsg,
1139 .mmap = sock_no_mmap,
1140 .splice_eof = inet_splice_eof,
1141 #ifdef CONFIG_COMPAT
1142 .compat_ioctl = inet_compat_ioctl,
1143 #endif
1144 };
1145
1146 static const struct net_proto_family inet_family_ops = {
1147 .family = PF_INET,
1148 .create = inet_create,
1149 .owner = THIS_MODULE,
1150 };
1151
1152 /* Upon startup we insert all the elements in inetsw_array[] into
1153 * the linked list inetsw.
1154 */
1155 static struct inet_protosw inetsw_array[] =
1156 {
1157 {
1158 .type = SOCK_STREAM,
1159 .protocol = IPPROTO_TCP,
1160 .prot = &tcp_prot,
1161 .ops = &inet_stream_ops,
1162 .flags = INET_PROTOSW_PERMANENT |
1163 INET_PROTOSW_ICSK,
1164 },
1165
1166 {
1167 .type = SOCK_DGRAM,
1168 .protocol = IPPROTO_UDP,
1169 .prot = &udp_prot,
1170 .ops = &inet_dgram_ops,
1171 .flags = INET_PROTOSW_PERMANENT,
1172 },
1173
1174 {
1175 .type = SOCK_DGRAM,
1176 .protocol = IPPROTO_ICMP,
1177 .prot = &ping_prot,
1178 .ops = &inet_sockraw_ops,
1179 .flags = INET_PROTOSW_REUSE,
1180 },
1181
1182 {
1183 .type = SOCK_RAW,
1184 .protocol = IPPROTO_IP, /* wild card */
1185 .prot = &raw_prot,
1186 .ops = &inet_sockraw_ops,
1187 .flags = INET_PROTOSW_REUSE,
1188 }
1189 };
1190
1191 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1192
inet_register_protosw(struct inet_protosw * p)1193 void inet_register_protosw(struct inet_protosw *p)
1194 {
1195 struct list_head *lh;
1196 struct inet_protosw *answer;
1197 int protocol = p->protocol;
1198 struct list_head *last_perm;
1199
1200 spin_lock_bh(&inetsw_lock);
1201
1202 if (p->type >= SOCK_MAX)
1203 goto out_illegal;
1204
1205 /* If we are trying to override a permanent protocol, bail. */
1206 last_perm = &inetsw[p->type];
1207 list_for_each(lh, &inetsw[p->type]) {
1208 answer = list_entry(lh, struct inet_protosw, list);
1209 /* Check only the non-wild match. */
1210 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1211 break;
1212 if (protocol == answer->protocol)
1213 goto out_permanent;
1214 last_perm = lh;
1215 }
1216
1217 /* Add the new entry after the last permanent entry if any, so that
1218 * the new entry does not override a permanent entry when matched with
1219 * a wild-card protocol. But it is allowed to override any existing
1220 * non-permanent entry. This means that when we remove this entry, the
1221 * system automatically returns to the old behavior.
1222 */
1223 list_add_rcu(&p->list, last_perm);
1224 out:
1225 spin_unlock_bh(&inetsw_lock);
1226
1227 return;
1228
1229 out_permanent:
1230 pr_err("Attempt to override permanent protocol %d\n", protocol);
1231 goto out;
1232
1233 out_illegal:
1234 pr_err("Ignoring attempt to register invalid socket type %d\n",
1235 p->type);
1236 goto out;
1237 }
1238 EXPORT_SYMBOL(inet_register_protosw);
1239
inet_unregister_protosw(struct inet_protosw * p)1240 void inet_unregister_protosw(struct inet_protosw *p)
1241 {
1242 if (INET_PROTOSW_PERMANENT & p->flags) {
1243 pr_err("Attempt to unregister permanent protocol %d\n",
1244 p->protocol);
1245 } else {
1246 spin_lock_bh(&inetsw_lock);
1247 list_del_rcu(&p->list);
1248 spin_unlock_bh(&inetsw_lock);
1249
1250 synchronize_net();
1251 }
1252 }
1253 EXPORT_SYMBOL(inet_unregister_protosw);
1254
inet_sk_reselect_saddr(struct sock * sk)1255 static int inet_sk_reselect_saddr(struct sock *sk)
1256 {
1257 struct inet_sock *inet = inet_sk(sk);
1258 __be32 old_saddr = inet->inet_saddr;
1259 __be32 daddr = inet->inet_daddr;
1260 struct flowi4 *fl4;
1261 struct rtable *rt;
1262 __be32 new_saddr;
1263 struct ip_options_rcu *inet_opt;
1264 int err;
1265
1266 inet_opt = rcu_dereference_protected(inet->inet_opt,
1267 lockdep_sock_is_held(sk));
1268 if (inet_opt && inet_opt->opt.srr)
1269 daddr = inet_opt->opt.faddr;
1270
1271 /* Query new route. */
1272 fl4 = &inet->cork.fl.u.ip4;
1273 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1274 sk->sk_protocol, inet->inet_sport,
1275 inet->inet_dport, sk);
1276 if (IS_ERR(rt))
1277 return PTR_ERR(rt);
1278
1279 new_saddr = fl4->saddr;
1280
1281 if (new_saddr == old_saddr) {
1282 sk_setup_caps(sk, &rt->dst);
1283 return 0;
1284 }
1285
1286 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1287 if (err) {
1288 ip_rt_put(rt);
1289 return err;
1290 }
1291
1292 sk_setup_caps(sk, &rt->dst);
1293
1294 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1295 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1296 __func__, &old_saddr, &new_saddr);
1297 }
1298
1299 /*
1300 * XXX The only one ugly spot where we need to
1301 * XXX really change the sockets identity after
1302 * XXX it has entered the hashes. -DaveM
1303 *
1304 * Besides that, it does not check for connection
1305 * uniqueness. Wait for troubles.
1306 */
1307 return __sk_prot_rehash(sk);
1308 }
1309
inet_sk_rebuild_header(struct sock * sk)1310 int inet_sk_rebuild_header(struct sock *sk)
1311 {
1312 struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0));
1313 struct inet_sock *inet = inet_sk(sk);
1314 __be32 daddr;
1315 struct ip_options_rcu *inet_opt;
1316 struct flowi4 *fl4;
1317 int err;
1318
1319 /* Route is OK, nothing to do. */
1320 if (rt)
1321 return 0;
1322
1323 /* Reroute. */
1324 rcu_read_lock();
1325 inet_opt = rcu_dereference(inet->inet_opt);
1326 daddr = inet->inet_daddr;
1327 if (inet_opt && inet_opt->opt.srr)
1328 daddr = inet_opt->opt.faddr;
1329 rcu_read_unlock();
1330 fl4 = &inet->cork.fl.u.ip4;
1331 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1332 inet->inet_dport, inet->inet_sport,
1333 sk->sk_protocol, ip_sock_rt_tos(sk),
1334 sk->sk_bound_dev_if);
1335 if (!IS_ERR(rt)) {
1336 err = 0;
1337 sk_setup_caps(sk, &rt->dst);
1338 } else {
1339 err = PTR_ERR(rt);
1340
1341 /* Routing failed... */
1342 sk->sk_route_caps = 0;
1343 /*
1344 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1345 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1346 */
1347 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1348 sk->sk_state != TCP_SYN_SENT ||
1349 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1350 (err = inet_sk_reselect_saddr(sk)) != 0)
1351 WRITE_ONCE(sk->sk_err_soft, -err);
1352 }
1353
1354 return err;
1355 }
1356 EXPORT_SYMBOL(inet_sk_rebuild_header);
1357
inet_sk_set_state(struct sock * sk,int state)1358 void inet_sk_set_state(struct sock *sk, int state)
1359 {
1360 trace_inet_sock_set_state(sk, sk->sk_state, state);
1361 sk->sk_state = state;
1362 }
1363 EXPORT_SYMBOL(inet_sk_set_state);
1364
inet_sk_state_store(struct sock * sk,int newstate)1365 void inet_sk_state_store(struct sock *sk, int newstate)
1366 {
1367 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1368 smp_store_release(&sk->sk_state, newstate);
1369 }
1370
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1371 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1372 netdev_features_t features)
1373 {
1374 bool udpfrag = false, fixedid = false, gso_partial, encap;
1375 struct sk_buff *segs = ERR_PTR(-EINVAL);
1376 const struct net_offload *ops;
1377 unsigned int offset = 0;
1378 struct iphdr *iph;
1379 int proto, tot_len;
1380 int nhoff;
1381 int ihl;
1382 int id;
1383
1384 skb_reset_network_header(skb);
1385 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1386 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1387 goto out;
1388
1389 iph = ip_hdr(skb);
1390 ihl = iph->ihl * 4;
1391 if (ihl < sizeof(*iph))
1392 goto out;
1393
1394 id = ntohs(iph->id);
1395 proto = iph->protocol;
1396
1397 /* Warning: after this point, iph might be no longer valid */
1398 if (unlikely(!pskb_may_pull(skb, ihl)))
1399 goto out;
1400 __skb_pull(skb, ihl);
1401
1402 encap = SKB_GSO_CB(skb)->encap_level > 0;
1403 if (encap)
1404 features &= skb->dev->hw_enc_features;
1405 SKB_GSO_CB(skb)->encap_level += ihl;
1406
1407 skb_reset_transport_header(skb);
1408
1409 segs = ERR_PTR(-EPROTONOSUPPORT);
1410
1411 if (!skb->encapsulation || encap) {
1412 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1413 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1414
1415 /* fixed ID is invalid if DF bit is not set */
1416 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1417 goto out;
1418 }
1419
1420 ops = rcu_dereference(inet_offloads[proto]);
1421 if (likely(ops && ops->callbacks.gso_segment)) {
1422 segs = ops->callbacks.gso_segment(skb, features);
1423 if (!segs)
1424 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1425 }
1426
1427 if (IS_ERR_OR_NULL(segs))
1428 goto out;
1429
1430 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1431
1432 skb = segs;
1433 do {
1434 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1435 if (udpfrag) {
1436 iph->frag_off = htons(offset >> 3);
1437 if (skb->next)
1438 iph->frag_off |= htons(IP_MF);
1439 offset += skb->len - nhoff - ihl;
1440 tot_len = skb->len - nhoff;
1441 } else if (skb_is_gso(skb)) {
1442 if (!fixedid) {
1443 iph->id = htons(id);
1444 id += skb_shinfo(skb)->gso_segs;
1445 }
1446
1447 if (gso_partial)
1448 tot_len = skb_shinfo(skb)->gso_size +
1449 SKB_GSO_CB(skb)->data_offset +
1450 skb->head - (unsigned char *)iph;
1451 else
1452 tot_len = skb->len - nhoff;
1453 } else {
1454 if (!fixedid)
1455 iph->id = htons(id++);
1456 tot_len = skb->len - nhoff;
1457 }
1458 iph->tot_len = htons(tot_len);
1459 ip_send_check(iph);
1460 if (encap)
1461 skb_reset_inner_headers(skb);
1462 skb->network_header = (u8 *)iph - skb->head;
1463 skb_reset_mac_len(skb);
1464 } while ((skb = skb->next));
1465
1466 out:
1467 return segs;
1468 }
1469
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1470 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1471 netdev_features_t features)
1472 {
1473 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1474 return ERR_PTR(-EINVAL);
1475
1476 return inet_gso_segment(skb, features);
1477 }
1478
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1479 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1480 {
1481 const struct net_offload *ops;
1482 struct sk_buff *pp = NULL;
1483 const struct iphdr *iph;
1484 struct sk_buff *p;
1485 unsigned int hlen;
1486 unsigned int off;
1487 int flush = 1;
1488 int proto;
1489
1490 off = skb_gro_offset(skb);
1491 hlen = off + sizeof(*iph);
1492 iph = skb_gro_header(skb, hlen, off);
1493 if (unlikely(!iph))
1494 goto out;
1495
1496 proto = iph->protocol;
1497
1498 ops = rcu_dereference(inet_offloads[proto]);
1499 if (!ops || !ops->callbacks.gro_receive)
1500 goto out;
1501
1502 if (*(u8 *)iph != 0x45)
1503 goto out;
1504
1505 if (ip_is_fragment(iph))
1506 goto out;
1507
1508 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1509 goto out;
1510
1511 NAPI_GRO_CB(skb)->proto = proto;
1512 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (ntohl(*(__be32 *)&iph->id) & ~IP_DF));
1513
1514 list_for_each_entry(p, head, list) {
1515 struct iphdr *iph2;
1516
1517 if (!NAPI_GRO_CB(p)->same_flow)
1518 continue;
1519
1520 iph2 = (struct iphdr *)(p->data + off);
1521 /* The above works because, with the exception of the top
1522 * (inner most) layer, we only aggregate pkts with the same
1523 * hdr length so all the hdrs we'll need to verify will start
1524 * at the same offset.
1525 */
1526 if ((iph->protocol ^ iph2->protocol) |
1527 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1528 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1529 NAPI_GRO_CB(p)->same_flow = 0;
1530 continue;
1531 }
1532 }
1533
1534 NAPI_GRO_CB(skb)->flush |= flush;
1535 NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off;
1536
1537 /* Note : No need to call skb_gro_postpull_rcsum() here,
1538 * as we already checked checksum over ipv4 header was 0
1539 */
1540 skb_gro_pull(skb, sizeof(*iph));
1541 skb_set_transport_header(skb, skb_gro_offset(skb));
1542
1543 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1544 ops->callbacks.gro_receive, head, skb);
1545
1546 out:
1547 skb_gro_flush_final(skb, pp, flush);
1548
1549 return pp;
1550 }
1551
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1552 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1553 struct sk_buff *skb)
1554 {
1555 if (NAPI_GRO_CB(skb)->encap_mark) {
1556 NAPI_GRO_CB(skb)->flush = 1;
1557 return NULL;
1558 }
1559
1560 NAPI_GRO_CB(skb)->encap_mark = 1;
1561
1562 return inet_gro_receive(head, skb);
1563 }
1564
1565 #define SECONDS_PER_DAY 86400
1566
1567 /* inet_current_timestamp - Return IP network timestamp
1568 *
1569 * Return milliseconds since midnight in network byte order.
1570 */
inet_current_timestamp(void)1571 __be32 inet_current_timestamp(void)
1572 {
1573 u32 secs;
1574 u32 msecs;
1575 struct timespec64 ts;
1576
1577 ktime_get_real_ts64(&ts);
1578
1579 /* Get secs since midnight. */
1580 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1581 /* Convert to msecs. */
1582 msecs = secs * MSEC_PER_SEC;
1583 /* Convert nsec to msec. */
1584 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1585
1586 /* Convert to network byte order. */
1587 return htonl(msecs);
1588 }
1589 EXPORT_SYMBOL(inet_current_timestamp);
1590
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1591 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1592 {
1593 unsigned int family = READ_ONCE(sk->sk_family);
1594
1595 if (family == AF_INET)
1596 return ip_recv_error(sk, msg, len, addr_len);
1597 #if IS_ENABLED(CONFIG_IPV6)
1598 if (family == AF_INET6)
1599 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1600 #endif
1601 return -EINVAL;
1602 }
1603 EXPORT_SYMBOL(inet_recv_error);
1604
inet_gro_complete(struct sk_buff * skb,int nhoff)1605 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1606 {
1607 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1608 const struct net_offload *ops;
1609 __be16 totlen = iph->tot_len;
1610 int proto = iph->protocol;
1611 int err = -ENOSYS;
1612
1613 if (skb->encapsulation) {
1614 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1615 skb_set_inner_network_header(skb, nhoff);
1616 }
1617
1618 iph_set_totlen(iph, skb->len - nhoff);
1619 csum_replace2(&iph->check, totlen, iph->tot_len);
1620
1621 ops = rcu_dereference(inet_offloads[proto]);
1622 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1623 goto out;
1624
1625 /* Only need to add sizeof(*iph) to get to the next hdr below
1626 * because any hdr with option will have been flushed in
1627 * inet_gro_receive().
1628 */
1629 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1630 tcp4_gro_complete, udp4_gro_complete,
1631 skb, nhoff + sizeof(*iph));
1632
1633 out:
1634 return err;
1635 }
1636
ipip_gro_complete(struct sk_buff * skb,int nhoff)1637 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1638 {
1639 skb->encapsulation = 1;
1640 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1641 return inet_gro_complete(skb, nhoff);
1642 }
1643
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1644 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1645 unsigned short type, unsigned char protocol,
1646 struct net *net)
1647 {
1648 struct socket *sock;
1649 int rc = sock_create_kern(net, family, type, protocol, &sock);
1650
1651 if (rc == 0) {
1652 *sk = sock->sk;
1653 (*sk)->sk_allocation = GFP_ATOMIC;
1654 (*sk)->sk_use_task_frag = false;
1655 /*
1656 * Unhash it so that IP input processing does not even see it,
1657 * we do not wish this socket to see incoming packets.
1658 */
1659 (*sk)->sk_prot->unhash(*sk);
1660 }
1661 return rc;
1662 }
1663 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1664
snmp_fold_field(void __percpu * mib,int offt)1665 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1666 {
1667 unsigned long res = 0;
1668 int i;
1669
1670 for_each_possible_cpu(i)
1671 res += snmp_get_cpu_field(mib, i, offt);
1672 return res;
1673 }
1674 EXPORT_SYMBOL_GPL(snmp_fold_field);
1675
1676 #if BITS_PER_LONG==32
1677
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1678 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1679 size_t syncp_offset)
1680 {
1681 void *bhptr;
1682 struct u64_stats_sync *syncp;
1683 u64 v;
1684 unsigned int start;
1685
1686 bhptr = per_cpu_ptr(mib, cpu);
1687 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1688 do {
1689 start = u64_stats_fetch_begin(syncp);
1690 v = *(((u64 *)bhptr) + offt);
1691 } while (u64_stats_fetch_retry(syncp, start));
1692
1693 return v;
1694 }
1695 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1696
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1697 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1698 {
1699 u64 res = 0;
1700 int cpu;
1701
1702 for_each_possible_cpu(cpu) {
1703 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1704 }
1705 return res;
1706 }
1707 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1708 #endif
1709
1710 #ifdef CONFIG_IP_MULTICAST
1711 static const struct net_protocol igmp_protocol = {
1712 .handler = igmp_rcv,
1713 };
1714 #endif
1715
1716 static const struct net_protocol icmp_protocol = {
1717 .handler = icmp_rcv,
1718 .err_handler = icmp_err,
1719 .no_policy = 1,
1720 };
1721
ipv4_mib_init_net(struct net * net)1722 static __net_init int ipv4_mib_init_net(struct net *net)
1723 {
1724 int i;
1725
1726 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1727 if (!net->mib.tcp_statistics)
1728 goto err_tcp_mib;
1729 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1730 if (!net->mib.ip_statistics)
1731 goto err_ip_mib;
1732
1733 for_each_possible_cpu(i) {
1734 struct ipstats_mib *af_inet_stats;
1735 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1736 u64_stats_init(&af_inet_stats->syncp);
1737 }
1738
1739 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1740 if (!net->mib.net_statistics)
1741 goto err_net_mib;
1742 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1743 if (!net->mib.udp_statistics)
1744 goto err_udp_mib;
1745 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1746 if (!net->mib.udplite_statistics)
1747 goto err_udplite_mib;
1748 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1749 if (!net->mib.icmp_statistics)
1750 goto err_icmp_mib;
1751 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1752 GFP_KERNEL);
1753 if (!net->mib.icmpmsg_statistics)
1754 goto err_icmpmsg_mib;
1755
1756 tcp_mib_init(net);
1757 return 0;
1758
1759 err_icmpmsg_mib:
1760 free_percpu(net->mib.icmp_statistics);
1761 err_icmp_mib:
1762 free_percpu(net->mib.udplite_statistics);
1763 err_udplite_mib:
1764 free_percpu(net->mib.udp_statistics);
1765 err_udp_mib:
1766 free_percpu(net->mib.net_statistics);
1767 err_net_mib:
1768 free_percpu(net->mib.ip_statistics);
1769 err_ip_mib:
1770 free_percpu(net->mib.tcp_statistics);
1771 err_tcp_mib:
1772 return -ENOMEM;
1773 }
1774
ipv4_mib_exit_net(struct net * net)1775 static __net_exit void ipv4_mib_exit_net(struct net *net)
1776 {
1777 kfree(net->mib.icmpmsg_statistics);
1778 free_percpu(net->mib.icmp_statistics);
1779 free_percpu(net->mib.udplite_statistics);
1780 free_percpu(net->mib.udp_statistics);
1781 free_percpu(net->mib.net_statistics);
1782 free_percpu(net->mib.ip_statistics);
1783 free_percpu(net->mib.tcp_statistics);
1784 #ifdef CONFIG_MPTCP
1785 /* allocated on demand, see mptcp_init_sock() */
1786 free_percpu(net->mib.mptcp_statistics);
1787 #endif
1788 }
1789
1790 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1791 .init = ipv4_mib_init_net,
1792 .exit = ipv4_mib_exit_net,
1793 };
1794
init_ipv4_mibs(void)1795 static int __init init_ipv4_mibs(void)
1796 {
1797 return register_pernet_subsys(&ipv4_mib_ops);
1798 }
1799
inet_init_net(struct net * net)1800 static __net_init int inet_init_net(struct net *net)
1801 {
1802 /*
1803 * Set defaults for local port range
1804 */
1805 net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
1806
1807 seqlock_init(&net->ipv4.ping_group_range.lock);
1808 /*
1809 * Sane defaults - nobody may create ping sockets.
1810 * Boot scripts should set this to distro-specific group.
1811 */
1812 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1813 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1814
1815 /* Default values for sysctl-controlled parameters.
1816 * We set them here, in case sysctl is not compiled.
1817 */
1818 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1819 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1820 net->ipv4.sysctl_ip_dynaddr = 0;
1821 net->ipv4.sysctl_ip_early_demux = 1;
1822 net->ipv4.sysctl_udp_early_demux = 1;
1823 net->ipv4.sysctl_tcp_early_demux = 1;
1824 net->ipv4.sysctl_nexthop_compat_mode = 1;
1825 #ifdef CONFIG_SYSCTL
1826 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1827 #endif
1828
1829 /* Some igmp sysctl, whose values are always used */
1830 net->ipv4.sysctl_igmp_max_memberships = 20;
1831 net->ipv4.sysctl_igmp_max_msf = 10;
1832 /* IGMP reports for link-local multicast groups are enabled by default */
1833 net->ipv4.sysctl_igmp_llm_reports = 1;
1834 net->ipv4.sysctl_igmp_qrv = 2;
1835
1836 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1837
1838 return 0;
1839 }
1840
1841 static __net_initdata struct pernet_operations af_inet_ops = {
1842 .init = inet_init_net,
1843 };
1844
init_inet_pernet_ops(void)1845 static int __init init_inet_pernet_ops(void)
1846 {
1847 return register_pernet_subsys(&af_inet_ops);
1848 }
1849
1850 static int ipv4_proc_init(void);
1851
1852 /*
1853 * IP protocol layer initialiser
1854 */
1855
1856
1857 static const struct net_offload ipip_offload = {
1858 .callbacks = {
1859 .gso_segment = ipip_gso_segment,
1860 .gro_receive = ipip_gro_receive,
1861 .gro_complete = ipip_gro_complete,
1862 },
1863 };
1864
ipip_offload_init(void)1865 static int __init ipip_offload_init(void)
1866 {
1867 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1868 }
1869
ipv4_offload_init(void)1870 static int __init ipv4_offload_init(void)
1871 {
1872 /*
1873 * Add offloads
1874 */
1875 if (udpv4_offload_init() < 0)
1876 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1877 if (tcpv4_offload_init() < 0)
1878 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1879 if (ipip_offload_init() < 0)
1880 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1881
1882 net_hotdata.ip_packet_offload = (struct packet_offload) {
1883 .type = cpu_to_be16(ETH_P_IP),
1884 .callbacks = {
1885 .gso_segment = inet_gso_segment,
1886 .gro_receive = inet_gro_receive,
1887 .gro_complete = inet_gro_complete,
1888 },
1889 };
1890 dev_add_offload(&net_hotdata.ip_packet_offload);
1891 return 0;
1892 }
1893
1894 fs_initcall(ipv4_offload_init);
1895
1896 static struct packet_type ip_packet_type __read_mostly = {
1897 .type = cpu_to_be16(ETH_P_IP),
1898 .func = ip_rcv,
1899 .list_func = ip_list_rcv,
1900 };
1901
inet_init(void)1902 static int __init inet_init(void)
1903 {
1904 struct inet_protosw *q;
1905 struct list_head *r;
1906 int rc;
1907
1908 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1909
1910 raw_hashinfo_init(&raw_v4_hashinfo);
1911
1912 rc = proto_register(&tcp_prot, 1);
1913 if (rc)
1914 goto out;
1915
1916 rc = proto_register(&udp_prot, 1);
1917 if (rc)
1918 goto out_unregister_tcp_proto;
1919
1920 rc = proto_register(&raw_prot, 1);
1921 if (rc)
1922 goto out_unregister_udp_proto;
1923
1924 rc = proto_register(&ping_prot, 1);
1925 if (rc)
1926 goto out_unregister_raw_proto;
1927
1928 /*
1929 * Tell SOCKET that we are alive...
1930 */
1931
1932 (void)sock_register(&inet_family_ops);
1933
1934 #ifdef CONFIG_SYSCTL
1935 ip_static_sysctl_init();
1936 #endif
1937
1938 /*
1939 * Add all the base protocols.
1940 */
1941
1942 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1943 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1944
1945 net_hotdata.udp_protocol = (struct net_protocol) {
1946 .handler = udp_rcv,
1947 .err_handler = udp_err,
1948 .no_policy = 1,
1949 };
1950 if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
1951 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1952
1953 net_hotdata.tcp_protocol = (struct net_protocol) {
1954 .handler = tcp_v4_rcv,
1955 .err_handler = tcp_v4_err,
1956 .no_policy = 1,
1957 .icmp_strict_tag_validation = 1,
1958 };
1959 if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
1960 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1961 #ifdef CONFIG_IP_MULTICAST
1962 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1963 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1964 #endif
1965
1966 /* Register the socket-side information for inet_create. */
1967 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1968 INIT_LIST_HEAD(r);
1969
1970 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1971 inet_register_protosw(q);
1972
1973 /*
1974 * Set the ARP module up
1975 */
1976
1977 arp_init();
1978
1979 /*
1980 * Set the IP module up
1981 */
1982
1983 ip_init();
1984
1985 /* Initialise per-cpu ipv4 mibs */
1986 if (init_ipv4_mibs())
1987 panic("%s: Cannot init ipv4 mibs\n", __func__);
1988
1989 /* Setup TCP slab cache for open requests. */
1990 tcp_init();
1991
1992 /* Setup UDP memory threshold */
1993 udp_init();
1994
1995 /* Add UDP-Lite (RFC 3828) */
1996 udplite4_register();
1997
1998 raw_init();
1999
2000 ping_init();
2001
2002 /*
2003 * Set the ICMP layer up
2004 */
2005
2006 if (icmp_init() < 0)
2007 panic("Failed to create the ICMP control socket.\n");
2008
2009 /*
2010 * Initialise the multicast router
2011 */
2012 #if defined(CONFIG_IP_MROUTE)
2013 if (ip_mr_init())
2014 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2015 #endif
2016
2017 if (init_inet_pernet_ops())
2018 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2019
2020 ipv4_proc_init();
2021
2022 ipfrag_init();
2023
2024 dev_add_pack(&ip_packet_type);
2025
2026 ip_tunnel_core_init();
2027
2028 rc = 0;
2029 out:
2030 return rc;
2031 out_unregister_raw_proto:
2032 proto_unregister(&raw_prot);
2033 out_unregister_udp_proto:
2034 proto_unregister(&udp_prot);
2035 out_unregister_tcp_proto:
2036 proto_unregister(&tcp_prot);
2037 goto out;
2038 }
2039
2040 fs_initcall(inet_init);
2041
2042 /* ------------------------------------------------------------------------ */
2043
2044 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2045 static int __init ipv4_proc_init(void)
2046 {
2047 int rc = 0;
2048
2049 if (raw_proc_init())
2050 goto out_raw;
2051 if (tcp4_proc_init())
2052 goto out_tcp;
2053 if (udp4_proc_init())
2054 goto out_udp;
2055 if (ping_proc_init())
2056 goto out_ping;
2057 if (ip_misc_proc_init())
2058 goto out_misc;
2059 out:
2060 return rc;
2061 out_misc:
2062 ping_proc_exit();
2063 out_ping:
2064 udp4_proc_exit();
2065 out_udp:
2066 tcp4_proc_exit();
2067 out_tcp:
2068 raw_proc_exit();
2069 out_raw:
2070 rc = -ENOMEM;
2071 goto out;
2072 }
2073
2074 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2075 static int __init ipv4_proc_init(void)
2076 {
2077 return 0;
2078 }
2079 #endif /* CONFIG_PROC_FS */
2080