1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PF_INET protocol family socket handler.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Florian La Roche, <flla@stud.uni-sb.de>
12 * Alan Cox, <A.Cox@swansea.ac.uk>
13 *
14 * Changes (see also sock.c)
15 *
16 * piggy,
17 * Karl Knutson : Socket protocol table
18 * A.N.Kuznetsov : Socket death error in accept().
19 * John Richardson : Fix non blocking error in connect()
20 * so sockets that fail to connect
21 * don't return -EINPROGRESS.
22 * Alan Cox : Asynchronous I/O support
23 * Alan Cox : Keep correct socket pointer on sock
24 * structures
25 * when accept() ed
26 * Alan Cox : Semantics of SO_LINGER aren't state
27 * moved to close when you look carefully.
28 * With this fixed and the accept bug fixed
29 * some RPC stuff seems happier.
30 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Alan Cox,
32 * Tony Gale : Fixed reuse semantics.
33 * Alan Cox : bind() shouldn't abort existing but dead
34 * sockets. Stops FTP netin:.. I hope.
35 * Alan Cox : bind() works correctly for RAW sockets.
36 * Note that FreeBSD at least was broken
37 * in this respect so be careful with
38 * compatibility tests...
39 * Alan Cox : routing cache support
40 * Alan Cox : memzero the socket structure for
41 * compactness.
42 * Matt Day : nonblock connect error handler
43 * Alan Cox : Allow large numbers of pending sockets
44 * (eg for big web sites), but only if
45 * specifically application requested.
46 * Alan Cox : New buffering throughout IP. Used
47 * dumbly.
48 * Alan Cox : New buffering now used smartly.
49 * Alan Cox : BSD rather than common sense
50 * interpretation of listen.
51 * Germano Caronni : Assorted small races.
52 * Alan Cox : sendmsg/recvmsg basic support.
53 * Alan Cox : Only sendmsg/recvmsg now supported.
54 * Alan Cox : Locked down bind (see security list).
55 * Alan Cox : Loosened bind a little.
56 * Mike McLagan : ADD/DEL DLCI Ioctls
57 * Willy Konynenberg : Transparent proxying support.
58 * David S. Miller : New socket lookup architecture.
59 * Some other random speedups.
60 * Cyrus Durgin : Cleaned up file for kmod hacks.
61 * Andi Kleen : Fix inet_stream_connect TCP race.
62 */
63
64 #define pr_fmt(fmt) "IPv4: " fmt
65
66 #include <linux/err.h>
67 #include <linux/errno.h>
68 #include <linux/types.h>
69 #include <linux/socket.h>
70 #include <linux/in.h>
71 #include <linux/kernel.h>
72 #include <linux/kmod.h>
73 #include <linux/sched.h>
74 #include <linux/timer.h>
75 #include <linux/string.h>
76 #include <linux/sockios.h>
77 #include <linux/net.h>
78 #include <linux/capability.h>
79 #include <linux/fcntl.h>
80 #include <linux/mm.h>
81 #include <linux/interrupt.h>
82 #include <linux/stat.h>
83 #include <linux/init.h>
84 #include <linux/poll.h>
85 #include <linux/netfilter_ipv4.h>
86 #include <linux/random.h>
87 #include <linux/slab.h>
88
89 #include <linux/uaccess.h>
90
91 #include <linux/inet.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/netdevice.h>
95 #include <net/checksum.h>
96 #include <net/ip.h>
97 #include <net/protocol.h>
98 #include <net/arp.h>
99 #include <net/route.h>
100 #include <net/ip_fib.h>
101 #include <net/inet_connection_sock.h>
102 #include <net/tcp.h>
103 #include <net/udp.h>
104 #include <net/udplite.h>
105 #include <net/ping.h>
106 #include <linux/skbuff.h>
107 #include <net/sock.h>
108 #include <net/raw.h>
109 #include <net/icmp.h>
110 #include <net/inet_common.h>
111 #include <net/ip_tunnels.h>
112 #include <net/xfrm.h>
113 #include <net/net_namespace.h>
114 #include <net/secure_seq.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
117 #endif
118 #include <net/l3mdev.h>
119 #include <net/compat.h>
120
121 #include <trace/events/sock.h>
122
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
125 */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128
129 /* New destruction routine */
130
inet_sock_destruct(struct sock * sk)131 void inet_sock_destruct(struct sock *sk)
132 {
133 struct inet_sock *inet = inet_sk(sk);
134
135 __skb_queue_purge(&sk->sk_receive_queue);
136 if (sk->sk_rx_skb_cache) {
137 __kfree_skb(sk->sk_rx_skb_cache);
138 sk->sk_rx_skb_cache = NULL;
139 }
140 __skb_queue_purge(&sk->sk_error_queue);
141
142 sk_mem_reclaim(sk);
143
144 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 pr_err("Attempt to release TCP socket in state %d %p\n",
146 sk->sk_state, sk);
147 return;
148 }
149 if (!sock_flag(sk, SOCK_DEAD)) {
150 pr_err("Attempt to release alive inet socket %p\n", sk);
151 return;
152 }
153
154 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
155 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
156 WARN_ON(sk->sk_wmem_queued);
157 WARN_ON(sk->sk_forward_alloc);
158
159 kfree(rcu_dereference_protected(inet->inet_opt, 1));
160 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
161 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
162 sk_refcnt_debug_dec(sk);
163 }
164 EXPORT_SYMBOL(inet_sock_destruct);
165
166 /*
167 * The routines beyond this point handle the behaviour of an AF_INET
168 * socket object. Mostly it punts to the subprotocols of IP to do
169 * the work.
170 */
171
172 /*
173 * Automatically bind an unbound socket.
174 */
175
inet_autobind(struct sock * sk)176 static int inet_autobind(struct sock *sk)
177 {
178 struct inet_sock *inet;
179 /* We may need to bind the socket. */
180 lock_sock(sk);
181 inet = inet_sk(sk);
182 if (!inet->inet_num) {
183 if (sk->sk_prot->get_port(sk, 0)) {
184 release_sock(sk);
185 return -EAGAIN;
186 }
187 inet->inet_sport = htons(inet->inet_num);
188 }
189 release_sock(sk);
190 return 0;
191 }
192
193 /*
194 * Move a socket into listening state.
195 */
inet_listen(struct socket * sock,int backlog)196 int inet_listen(struct socket *sock, int backlog)
197 {
198 struct sock *sk = sock->sk;
199 unsigned char old_state;
200 int err, tcp_fastopen;
201
202 lock_sock(sk);
203
204 err = -EINVAL;
205 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
206 goto out;
207
208 old_state = sk->sk_state;
209 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
210 goto out;
211
212 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
213 /* Really, if the socket is already in listen state
214 * we can only allow the backlog to be adjusted.
215 */
216 if (old_state != TCP_LISTEN) {
217 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
218 * Note that only TCP sockets (SOCK_STREAM) will reach here.
219 * Also fastopen backlog may already been set via the option
220 * because the socket was in TCP_LISTEN state previously but
221 * was shutdown() rather than close().
222 */
223 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
224 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
225 (tcp_fastopen & TFO_SERVER_ENABLE) &&
226 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
227 fastopen_queue_tune(sk, backlog);
228 tcp_fastopen_init_key_once(sock_net(sk));
229 }
230
231 err = inet_csk_listen_start(sk, backlog);
232 if (err)
233 goto out;
234 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
235 }
236 err = 0;
237
238 out:
239 release_sock(sk);
240 return err;
241 }
242 EXPORT_SYMBOL(inet_listen);
243
244 /*
245 * Create an inet socket.
246 */
247
inet_create(struct net * net,struct socket * sock,int protocol,int kern)248 static int inet_create(struct net *net, struct socket *sock, int protocol,
249 int kern)
250 {
251 struct sock *sk;
252 struct inet_protosw *answer;
253 struct inet_sock *inet;
254 struct proto *answer_prot;
255 unsigned char answer_flags;
256 int try_loading_module = 0;
257 int err;
258
259 if (protocol < 0 || protocol >= IPPROTO_MAX)
260 return -EINVAL;
261
262 sock->state = SS_UNCONNECTED;
263
264 /* Look for the requested type/protocol pair. */
265 lookup_protocol:
266 err = -ESOCKTNOSUPPORT;
267 rcu_read_lock();
268 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
269
270 err = 0;
271 /* Check the non-wild match. */
272 if (protocol == answer->protocol) {
273 if (protocol != IPPROTO_IP)
274 break;
275 } else {
276 /* Check for the two wild cases. */
277 if (IPPROTO_IP == protocol) {
278 protocol = answer->protocol;
279 break;
280 }
281 if (IPPROTO_IP == answer->protocol)
282 break;
283 }
284 err = -EPROTONOSUPPORT;
285 }
286
287 if (unlikely(err)) {
288 if (try_loading_module < 2) {
289 rcu_read_unlock();
290 /*
291 * Be more specific, e.g. net-pf-2-proto-132-type-1
292 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
293 */
294 if (++try_loading_module == 1)
295 request_module("net-pf-%d-proto-%d-type-%d",
296 PF_INET, protocol, sock->type);
297 /*
298 * Fall back to generic, e.g. net-pf-2-proto-132
299 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
300 */
301 else
302 request_module("net-pf-%d-proto-%d",
303 PF_INET, protocol);
304 goto lookup_protocol;
305 } else
306 goto out_rcu_unlock;
307 }
308
309 err = -EPERM;
310 if (sock->type == SOCK_RAW && !kern &&
311 !ns_capable(net->user_ns, CAP_NET_RAW))
312 goto out_rcu_unlock;
313
314 sock->ops = answer->ops;
315 answer_prot = answer->prot;
316 answer_flags = answer->flags;
317 rcu_read_unlock();
318
319 WARN_ON(!answer_prot->slab);
320
321 err = -ENOMEM;
322 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
323 if (!sk)
324 goto out;
325
326 err = 0;
327 if (INET_PROTOSW_REUSE & answer_flags)
328 sk->sk_reuse = SK_CAN_REUSE;
329
330 if (INET_PROTOSW_ICSK & answer_flags)
331 inet_init_csk_locks(sk);
332
333 inet = inet_sk(sk);
334 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
335
336 inet->nodefrag = 0;
337
338 if (SOCK_RAW == sock->type) {
339 inet->inet_num = protocol;
340 if (IPPROTO_RAW == protocol)
341 inet->hdrincl = 1;
342 }
343
344 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
345 inet->pmtudisc = IP_PMTUDISC_DONT;
346 else
347 inet->pmtudisc = IP_PMTUDISC_WANT;
348
349 inet->inet_id = 0;
350
351 sock_init_data(sock, sk);
352
353 sk->sk_destruct = inet_sock_destruct;
354 sk->sk_protocol = protocol;
355 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
356
357 inet->uc_ttl = -1;
358 inet->mc_loop = 1;
359 inet->mc_ttl = 1;
360 inet->mc_all = 1;
361 inet->mc_index = 0;
362 inet->mc_list = NULL;
363 inet->rcv_tos = 0;
364
365 sk_refcnt_debug_inc(sk);
366
367 if (inet->inet_num) {
368 /* It assumes that any protocol which allows
369 * the user to assign a number at socket
370 * creation time automatically
371 * shares.
372 */
373 inet->inet_sport = htons(inet->inet_num);
374 /* Add to protocol hash chains. */
375 err = sk->sk_prot->hash(sk);
376 if (err) {
377 sk_common_release(sk);
378 goto out;
379 }
380 }
381
382 if (sk->sk_prot->init) {
383 err = sk->sk_prot->init(sk);
384 if (err) {
385 sk_common_release(sk);
386 goto out;
387 }
388 }
389
390 if (!kern) {
391 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
392 if (err) {
393 sk_common_release(sk);
394 goto out;
395 }
396 }
397 out:
398 return err;
399 out_rcu_unlock:
400 rcu_read_unlock();
401 goto out;
402 }
403
404
405 /*
406 * The peer socket should always be NULL (or else). When we call this
407 * function we are destroying the object and from then on nobody
408 * should refer to it.
409 */
inet_release(struct socket * sock)410 int inet_release(struct socket *sock)
411 {
412 struct sock *sk = sock->sk;
413
414 if (sk) {
415 long timeout;
416
417 if (!sk->sk_kern_sock)
418 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
419
420 /* Applications forget to leave groups before exiting */
421 ip_mc_drop_socket(sk);
422
423 /* If linger is set, we don't return until the close
424 * is complete. Otherwise we return immediately. The
425 * actually closing is done the same either way.
426 *
427 * If the close is due to the process exiting, we never
428 * linger..
429 */
430 timeout = 0;
431 if (sock_flag(sk, SOCK_LINGER) &&
432 !(current->flags & PF_EXITING))
433 timeout = sk->sk_lingertime;
434 sk->sk_prot->close(sk, timeout);
435 sock->sk = NULL;
436 }
437 return 0;
438 }
439 EXPORT_SYMBOL(inet_release);
440
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)441 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
442 {
443 struct sock *sk = sock->sk;
444 u32 flags = BIND_WITH_LOCK;
445 int err;
446
447 /* If the socket has its own bind function then use it. (RAW) */
448 if (sk->sk_prot->bind) {
449 return sk->sk_prot->bind(sk, uaddr, addr_len);
450 }
451 if (addr_len < sizeof(struct sockaddr_in))
452 return -EINVAL;
453
454 /* BPF prog is run before any checks are done so that if the prog
455 * changes context in a wrong way it will be caught.
456 */
457 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr,
458 CGROUP_INET4_BIND, &flags);
459 if (err)
460 return err;
461
462 return __inet_bind(sk, uaddr, addr_len, flags);
463 }
464 EXPORT_SYMBOL(inet_bind);
465
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)466 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
467 u32 flags)
468 {
469 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
470 struct inet_sock *inet = inet_sk(sk);
471 struct net *net = sock_net(sk);
472 unsigned short snum;
473 int chk_addr_ret;
474 u32 tb_id = RT_TABLE_LOCAL;
475 int err;
476
477 if (addr->sin_family != AF_INET) {
478 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
479 * only if s_addr is INADDR_ANY.
480 */
481 err = -EAFNOSUPPORT;
482 if (addr->sin_family != AF_UNSPEC ||
483 addr->sin_addr.s_addr != htonl(INADDR_ANY))
484 goto out;
485 }
486
487 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
488 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
489
490 /* Not specified by any standard per-se, however it breaks too
491 * many applications when removed. It is unfortunate since
492 * allowing applications to make a non-local bind solves
493 * several problems with systems using dynamic addressing.
494 * (ie. your servers still start up even if your ISDN link
495 * is temporarily down)
496 */
497 err = -EADDRNOTAVAIL;
498 if (!inet_can_nonlocal_bind(net, inet) &&
499 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
500 chk_addr_ret != RTN_LOCAL &&
501 chk_addr_ret != RTN_MULTICAST &&
502 chk_addr_ret != RTN_BROADCAST)
503 goto out;
504
505 snum = ntohs(addr->sin_port);
506 err = -EPERM;
507 if (snum && inet_is_local_unbindable_port(net, snum))
508 goto out;
509
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->bind_address_no_port ||
537 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
538 if (sk->sk_prot->get_port(sk, snum)) {
539 inet->inet_saddr = inet->inet_rcv_saddr = 0;
540 err = -EADDRINUSE;
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 goto out_release_sock;
548 }
549 }
550 }
551
552 if (inet->inet_rcv_saddr)
553 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
554 if (snum)
555 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
556 inet->inet_sport = htons(inet->inet_num);
557 inet->inet_daddr = 0;
558 inet->inet_dport = 0;
559 sk_dst_reset(sk);
560 err = 0;
561 out_release_sock:
562 if (flags & BIND_WITH_LOCK)
563 release_sock(sk);
564 out:
565 return err;
566 }
567
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)568 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
569 int addr_len, int flags)
570 {
571 struct sock *sk = sock->sk;
572 int err;
573
574 if (addr_len < sizeof(uaddr->sa_family))
575 return -EINVAL;
576 if (uaddr->sa_family == AF_UNSPEC)
577 return sk->sk_prot->disconnect(sk, flags);
578
579 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
580 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
581 if (err)
582 return err;
583 }
584
585 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
586 return -EAGAIN;
587 return sk->sk_prot->connect(sk, uaddr, addr_len);
588 }
589 EXPORT_SYMBOL(inet_dgram_connect);
590
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)591 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
592 {
593 DEFINE_WAIT_FUNC(wait, woken_wake_function);
594
595 add_wait_queue(sk_sleep(sk), &wait);
596 sk->sk_write_pending += writebias;
597 sk->sk_wait_pending++;
598
599 /* Basic assumption: if someone sets sk->sk_err, he _must_
600 * change state of the socket from TCP_SYN_*.
601 * Connect() does not allow to get error notifications
602 * without closing the socket.
603 */
604 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
605 release_sock(sk);
606 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
607 lock_sock(sk);
608 if (signal_pending(current) || !timeo)
609 break;
610 }
611 remove_wait_queue(sk_sleep(sk), &wait);
612 sk->sk_write_pending -= writebias;
613 sk->sk_wait_pending--;
614 return timeo;
615 }
616
617 /*
618 * Connect to a remote host. There is regrettably still a little
619 * TCP 'magic' in here.
620 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)621 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
622 int addr_len, int flags, int is_sendmsg)
623 {
624 struct sock *sk = sock->sk;
625 int err;
626 long timeo;
627
628 /*
629 * uaddr can be NULL and addr_len can be 0 if:
630 * sk is a TCP fastopen active socket and
631 * TCP_FASTOPEN_CONNECT sockopt is set and
632 * we already have a valid cookie for this socket.
633 * In this case, user can call write() after connect().
634 * write() will invoke tcp_sendmsg_fastopen() which calls
635 * __inet_stream_connect().
636 */
637 if (uaddr) {
638 if (addr_len < sizeof(uaddr->sa_family))
639 return -EINVAL;
640
641 if (uaddr->sa_family == AF_UNSPEC) {
642 err = sk->sk_prot->disconnect(sk, flags);
643 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
644 goto out;
645 }
646 }
647
648 switch (sock->state) {
649 default:
650 err = -EINVAL;
651 goto out;
652 case SS_CONNECTED:
653 err = -EISCONN;
654 goto out;
655 case SS_CONNECTING:
656 if (inet_sk(sk)->defer_connect)
657 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
658 else
659 err = -EALREADY;
660 /* Fall out of switch with err, set for this state */
661 break;
662 case SS_UNCONNECTED:
663 err = -EISCONN;
664 if (sk->sk_state != TCP_CLOSE)
665 goto out;
666
667 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
668 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
669 if (err)
670 goto out;
671 }
672
673 err = sk->sk_prot->connect(sk, uaddr, addr_len);
674 if (err < 0)
675 goto out;
676
677 sock->state = SS_CONNECTING;
678
679 if (!err && inet_sk(sk)->defer_connect)
680 goto out;
681
682 /* Just entered SS_CONNECTING state; the only
683 * difference is that return value in non-blocking
684 * case is EINPROGRESS, rather than EALREADY.
685 */
686 err = -EINPROGRESS;
687 break;
688 }
689
690 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
691
692 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
693 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
694 tcp_sk(sk)->fastopen_req &&
695 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
696
697 /* Error code is set above */
698 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
699 goto out;
700
701 err = sock_intr_errno(timeo);
702 if (signal_pending(current))
703 goto out;
704 }
705
706 /* Connection was closed by RST, timeout, ICMP error
707 * or another process disconnected us.
708 */
709 if (sk->sk_state == TCP_CLOSE)
710 goto sock_error;
711
712 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
713 * and error was received after socket entered established state.
714 * Hence, it is handled normally after connect() return successfully.
715 */
716
717 sock->state = SS_CONNECTED;
718 err = 0;
719 out:
720 return err;
721
722 sock_error:
723 err = sock_error(sk) ? : -ECONNABORTED;
724 sock->state = SS_UNCONNECTED;
725 if (sk->sk_prot->disconnect(sk, flags))
726 sock->state = SS_DISCONNECTING;
727 goto out;
728 }
729 EXPORT_SYMBOL(__inet_stream_connect);
730
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)731 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
732 int addr_len, int flags)
733 {
734 int err;
735
736 lock_sock(sock->sk);
737 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
738 release_sock(sock->sk);
739 return err;
740 }
741 EXPORT_SYMBOL(inet_stream_connect);
742
743 /*
744 * Accept a pending connection. The TCP layer now gives BSD semantics.
745 */
746
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)747 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
748 bool kern)
749 {
750 struct sock *sk1 = sock->sk;
751 int err = -EINVAL;
752 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
753
754 if (!sk2)
755 goto do_err;
756
757 lock_sock(sk2);
758
759 sock_rps_record_flow(sk2);
760 WARN_ON(!((1 << sk2->sk_state) &
761 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
762 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
763
764 sock_graft(sk2, newsock);
765
766 newsock->state = SS_CONNECTED;
767 err = 0;
768 release_sock(sk2);
769 do_err:
770 return err;
771 }
772 EXPORT_SYMBOL(inet_accept);
773
774 /*
775 * This does both peername and sockname.
776 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)777 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
778 int peer)
779 {
780 struct sock *sk = sock->sk;
781 struct inet_sock *inet = inet_sk(sk);
782 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
783
784 sin->sin_family = AF_INET;
785 lock_sock(sk);
786 if (peer) {
787 if (!inet->inet_dport ||
788 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
789 peer == 1)) {
790 release_sock(sk);
791 return -ENOTCONN;
792 }
793 sin->sin_port = inet->inet_dport;
794 sin->sin_addr.s_addr = inet->inet_daddr;
795 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
796 CGROUP_INET4_GETPEERNAME);
797 } else {
798 __be32 addr = inet->inet_rcv_saddr;
799 if (!addr)
800 addr = inet->inet_saddr;
801 sin->sin_port = inet->inet_sport;
802 sin->sin_addr.s_addr = addr;
803 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin,
804 CGROUP_INET4_GETSOCKNAME);
805 }
806 release_sock(sk);
807 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
808 return sizeof(*sin);
809 }
810 EXPORT_SYMBOL(inet_getname);
811
inet_send_prepare(struct sock * sk)812 int inet_send_prepare(struct sock *sk)
813 {
814 sock_rps_record_flow(sk);
815
816 /* We may need to bind the socket. */
817 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
818 inet_autobind(sk))
819 return -EAGAIN;
820
821 return 0;
822 }
823 EXPORT_SYMBOL_GPL(inet_send_prepare);
824
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)825 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
826 {
827 struct sock *sk = sock->sk;
828
829 if (unlikely(inet_send_prepare(sk)))
830 return -EAGAIN;
831
832 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
833 sk, msg, size);
834 }
835 EXPORT_SYMBOL(inet_sendmsg);
836
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)837 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
838 size_t size, int flags)
839 {
840 struct sock *sk = sock->sk;
841
842 if (unlikely(inet_send_prepare(sk)))
843 return -EAGAIN;
844
845 if (sk->sk_prot->sendpage)
846 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
847 return sock_no_sendpage(sock, page, offset, size, flags);
848 }
849 EXPORT_SYMBOL(inet_sendpage);
850
851 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
852 size_t, int, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)853 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
854 int flags)
855 {
856 struct sock *sk = sock->sk;
857 int addr_len = 0;
858 int err;
859
860 if (likely(!(flags & MSG_ERRQUEUE)))
861 sock_rps_record_flow(sk);
862
863 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
864 sk, msg, size, flags & MSG_DONTWAIT,
865 flags & ~MSG_DONTWAIT, &addr_len);
866 if (err >= 0)
867 msg->msg_namelen = addr_len;
868 return err;
869 }
870 EXPORT_SYMBOL(inet_recvmsg);
871
inet_shutdown(struct socket * sock,int how)872 int inet_shutdown(struct socket *sock, int how)
873 {
874 struct sock *sk = sock->sk;
875 int err = 0;
876
877 /* This should really check to make sure
878 * the socket is a TCP socket. (WHY AC...)
879 */
880 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
881 1->2 bit 2 snds.
882 2->3 */
883 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
884 return -EINVAL;
885
886 lock_sock(sk);
887 if (sock->state == SS_CONNECTING) {
888 if ((1 << sk->sk_state) &
889 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
890 sock->state = SS_DISCONNECTING;
891 else
892 sock->state = SS_CONNECTED;
893 }
894
895 switch (sk->sk_state) {
896 case TCP_CLOSE:
897 err = -ENOTCONN;
898 /* Hack to wake up other listeners, who can poll for
899 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
900 fallthrough;
901 default:
902 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
903 if (sk->sk_prot->shutdown)
904 sk->sk_prot->shutdown(sk, how);
905 break;
906
907 /* Remaining two branches are temporary solution for missing
908 * close() in multithreaded environment. It is _not_ a good idea,
909 * but we have no choice until close() is repaired at VFS level.
910 */
911 case TCP_LISTEN:
912 if (!(how & RCV_SHUTDOWN))
913 break;
914 fallthrough;
915 case TCP_SYN_SENT:
916 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
917 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
918 break;
919 }
920
921 /* Wake up anyone sleeping in poll. */
922 sk->sk_state_change(sk);
923 release_sock(sk);
924 return err;
925 }
926 EXPORT_SYMBOL(inet_shutdown);
927
928 /*
929 * ioctl() calls you can issue on an INET socket. Most of these are
930 * device configuration and stuff and very rarely used. Some ioctls
931 * pass on to the socket itself.
932 *
933 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
934 * loads the devconfigure module does its configuring and unloads it.
935 * There's a good 20K of config code hanging around the kernel.
936 */
937
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)938 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
939 {
940 struct sock *sk = sock->sk;
941 int err = 0;
942 struct net *net = sock_net(sk);
943 void __user *p = (void __user *)arg;
944 struct ifreq ifr;
945 struct rtentry rt;
946
947 switch (cmd) {
948 case SIOCADDRT:
949 case SIOCDELRT:
950 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
951 return -EFAULT;
952 err = ip_rt_ioctl(net, cmd, &rt);
953 break;
954 case SIOCRTMSG:
955 err = -EINVAL;
956 break;
957 case SIOCDARP:
958 case SIOCGARP:
959 case SIOCSARP:
960 err = arp_ioctl(net, cmd, (void __user *)arg);
961 break;
962 case SIOCGIFADDR:
963 case SIOCGIFBRDADDR:
964 case SIOCGIFNETMASK:
965 case SIOCGIFDSTADDR:
966 case SIOCGIFPFLAGS:
967 if (get_user_ifreq(&ifr, NULL, p))
968 return -EFAULT;
969 err = devinet_ioctl(net, cmd, &ifr);
970 if (!err && put_user_ifreq(&ifr, p))
971 err = -EFAULT;
972 break;
973
974 case SIOCSIFADDR:
975 case SIOCSIFBRDADDR:
976 case SIOCSIFNETMASK:
977 case SIOCSIFDSTADDR:
978 case SIOCSIFPFLAGS:
979 case SIOCSIFFLAGS:
980 if (get_user_ifreq(&ifr, NULL, p))
981 return -EFAULT;
982 err = devinet_ioctl(net, cmd, &ifr);
983 break;
984 default:
985 if (sk->sk_prot->ioctl)
986 err = sk->sk_prot->ioctl(sk, cmd, arg);
987 else
988 err = -ENOIOCTLCMD;
989 break;
990 }
991 return err;
992 }
993 EXPORT_SYMBOL(inet_ioctl);
994
995 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)996 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
997 struct compat_rtentry __user *ur)
998 {
999 compat_uptr_t rtdev;
1000 struct rtentry rt;
1001
1002 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1003 3 * sizeof(struct sockaddr)) ||
1004 get_user(rt.rt_flags, &ur->rt_flags) ||
1005 get_user(rt.rt_metric, &ur->rt_metric) ||
1006 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1007 get_user(rt.rt_window, &ur->rt_window) ||
1008 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1009 get_user(rtdev, &ur->rt_dev))
1010 return -EFAULT;
1011
1012 rt.rt_dev = compat_ptr(rtdev);
1013 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1014 }
1015
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1016 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1017 {
1018 void __user *argp = compat_ptr(arg);
1019 struct sock *sk = sock->sk;
1020
1021 switch (cmd) {
1022 case SIOCADDRT:
1023 case SIOCDELRT:
1024 return inet_compat_routing_ioctl(sk, cmd, argp);
1025 default:
1026 if (!sk->sk_prot->compat_ioctl)
1027 return -ENOIOCTLCMD;
1028 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1029 }
1030 }
1031 #endif /* CONFIG_COMPAT */
1032
1033 const struct proto_ops inet_stream_ops = {
1034 .family = PF_INET,
1035 .owner = THIS_MODULE,
1036 .release = inet_release,
1037 .bind = inet_bind,
1038 .connect = inet_stream_connect,
1039 .socketpair = sock_no_socketpair,
1040 .accept = inet_accept,
1041 .getname = inet_getname,
1042 .poll = tcp_poll,
1043 .ioctl = inet_ioctl,
1044 .gettstamp = sock_gettstamp,
1045 .listen = inet_listen,
1046 .shutdown = inet_shutdown,
1047 .setsockopt = sock_common_setsockopt,
1048 .getsockopt = sock_common_getsockopt,
1049 .sendmsg = inet_sendmsg,
1050 .recvmsg = inet_recvmsg,
1051 #ifdef CONFIG_MMU
1052 .mmap = tcp_mmap,
1053 #endif
1054 .sendpage = inet_sendpage,
1055 .splice_read = tcp_splice_read,
1056 .read_sock = tcp_read_sock,
1057 .sendmsg_locked = tcp_sendmsg_locked,
1058 .sendpage_locked = tcp_sendpage_locked,
1059 .peek_len = tcp_peek_len,
1060 #ifdef CONFIG_COMPAT
1061 .compat_ioctl = inet_compat_ioctl,
1062 #endif
1063 .set_rcvlowat = tcp_set_rcvlowat,
1064 };
1065 EXPORT_SYMBOL(inet_stream_ops);
1066
1067 const struct proto_ops inet_dgram_ops = {
1068 .family = PF_INET,
1069 .owner = THIS_MODULE,
1070 .release = inet_release,
1071 .bind = inet_bind,
1072 .connect = inet_dgram_connect,
1073 .socketpair = sock_no_socketpair,
1074 .accept = sock_no_accept,
1075 .getname = inet_getname,
1076 .poll = udp_poll,
1077 .ioctl = inet_ioctl,
1078 .gettstamp = sock_gettstamp,
1079 .listen = sock_no_listen,
1080 .shutdown = inet_shutdown,
1081 .setsockopt = sock_common_setsockopt,
1082 .getsockopt = sock_common_getsockopt,
1083 .sendmsg = inet_sendmsg,
1084 .read_sock = udp_read_sock,
1085 .recvmsg = inet_recvmsg,
1086 .mmap = sock_no_mmap,
1087 .sendpage = inet_sendpage,
1088 .set_peek_off = sk_set_peek_off,
1089 #ifdef CONFIG_COMPAT
1090 .compat_ioctl = inet_compat_ioctl,
1091 #endif
1092 };
1093 EXPORT_SYMBOL(inet_dgram_ops);
1094
1095 /*
1096 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1097 * udp_poll
1098 */
1099 static const struct proto_ops inet_sockraw_ops = {
1100 .family = PF_INET,
1101 .owner = THIS_MODULE,
1102 .release = inet_release,
1103 .bind = inet_bind,
1104 .connect = inet_dgram_connect,
1105 .socketpair = sock_no_socketpair,
1106 .accept = sock_no_accept,
1107 .getname = inet_getname,
1108 .poll = datagram_poll,
1109 .ioctl = inet_ioctl,
1110 .gettstamp = sock_gettstamp,
1111 .listen = sock_no_listen,
1112 .shutdown = inet_shutdown,
1113 .setsockopt = sock_common_setsockopt,
1114 .getsockopt = sock_common_getsockopt,
1115 .sendmsg = inet_sendmsg,
1116 .recvmsg = inet_recvmsg,
1117 .mmap = sock_no_mmap,
1118 .sendpage = inet_sendpage,
1119 #ifdef CONFIG_COMPAT
1120 .compat_ioctl = inet_compat_ioctl,
1121 #endif
1122 };
1123
1124 static const struct net_proto_family inet_family_ops = {
1125 .family = PF_INET,
1126 .create = inet_create,
1127 .owner = THIS_MODULE,
1128 };
1129
1130 /* Upon startup we insert all the elements in inetsw_array[] into
1131 * the linked list inetsw.
1132 */
1133 static struct inet_protosw inetsw_array[] =
1134 {
1135 {
1136 .type = SOCK_STREAM,
1137 .protocol = IPPROTO_TCP,
1138 .prot = &tcp_prot,
1139 .ops = &inet_stream_ops,
1140 .flags = INET_PROTOSW_PERMANENT |
1141 INET_PROTOSW_ICSK,
1142 },
1143
1144 {
1145 .type = SOCK_DGRAM,
1146 .protocol = IPPROTO_UDP,
1147 .prot = &udp_prot,
1148 .ops = &inet_dgram_ops,
1149 .flags = INET_PROTOSW_PERMANENT,
1150 },
1151
1152 {
1153 .type = SOCK_DGRAM,
1154 .protocol = IPPROTO_ICMP,
1155 .prot = &ping_prot,
1156 .ops = &inet_sockraw_ops,
1157 .flags = INET_PROTOSW_REUSE,
1158 },
1159
1160 {
1161 .type = SOCK_RAW,
1162 .protocol = IPPROTO_IP, /* wild card */
1163 .prot = &raw_prot,
1164 .ops = &inet_sockraw_ops,
1165 .flags = INET_PROTOSW_REUSE,
1166 }
1167 };
1168
1169 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1170
inet_register_protosw(struct inet_protosw * p)1171 void inet_register_protosw(struct inet_protosw *p)
1172 {
1173 struct list_head *lh;
1174 struct inet_protosw *answer;
1175 int protocol = p->protocol;
1176 struct list_head *last_perm;
1177
1178 spin_lock_bh(&inetsw_lock);
1179
1180 if (p->type >= SOCK_MAX)
1181 goto out_illegal;
1182
1183 /* If we are trying to override a permanent protocol, bail. */
1184 last_perm = &inetsw[p->type];
1185 list_for_each(lh, &inetsw[p->type]) {
1186 answer = list_entry(lh, struct inet_protosw, list);
1187 /* Check only the non-wild match. */
1188 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1189 break;
1190 if (protocol == answer->protocol)
1191 goto out_permanent;
1192 last_perm = lh;
1193 }
1194
1195 /* Add the new entry after the last permanent entry if any, so that
1196 * the new entry does not override a permanent entry when matched with
1197 * a wild-card protocol. But it is allowed to override any existing
1198 * non-permanent entry. This means that when we remove this entry, the
1199 * system automatically returns to the old behavior.
1200 */
1201 list_add_rcu(&p->list, last_perm);
1202 out:
1203 spin_unlock_bh(&inetsw_lock);
1204
1205 return;
1206
1207 out_permanent:
1208 pr_err("Attempt to override permanent protocol %d\n", protocol);
1209 goto out;
1210
1211 out_illegal:
1212 pr_err("Ignoring attempt to register invalid socket type %d\n",
1213 p->type);
1214 goto out;
1215 }
1216 EXPORT_SYMBOL(inet_register_protosw);
1217
inet_unregister_protosw(struct inet_protosw * p)1218 void inet_unregister_protosw(struct inet_protosw *p)
1219 {
1220 if (INET_PROTOSW_PERMANENT & p->flags) {
1221 pr_err("Attempt to unregister permanent protocol %d\n",
1222 p->protocol);
1223 } else {
1224 spin_lock_bh(&inetsw_lock);
1225 list_del_rcu(&p->list);
1226 spin_unlock_bh(&inetsw_lock);
1227
1228 synchronize_net();
1229 }
1230 }
1231 EXPORT_SYMBOL(inet_unregister_protosw);
1232
inet_sk_reselect_saddr(struct sock * sk)1233 static int inet_sk_reselect_saddr(struct sock *sk)
1234 {
1235 struct inet_sock *inet = inet_sk(sk);
1236 __be32 old_saddr = inet->inet_saddr;
1237 __be32 daddr = inet->inet_daddr;
1238 struct flowi4 *fl4;
1239 struct rtable *rt;
1240 __be32 new_saddr;
1241 struct ip_options_rcu *inet_opt;
1242
1243 inet_opt = rcu_dereference_protected(inet->inet_opt,
1244 lockdep_sock_is_held(sk));
1245 if (inet_opt && inet_opt->opt.srr)
1246 daddr = inet_opt->opt.faddr;
1247
1248 /* Query new route. */
1249 fl4 = &inet->cork.fl.u.ip4;
1250 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1251 sk->sk_bound_dev_if, sk->sk_protocol,
1252 inet->inet_sport, inet->inet_dport, sk);
1253 if (IS_ERR(rt))
1254 return PTR_ERR(rt);
1255
1256 sk_setup_caps(sk, &rt->dst);
1257
1258 new_saddr = fl4->saddr;
1259
1260 if (new_saddr == old_saddr)
1261 return 0;
1262
1263 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1264 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1265 __func__, &old_saddr, &new_saddr);
1266 }
1267
1268 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1269
1270 /*
1271 * XXX The only one ugly spot where we need to
1272 * XXX really change the sockets identity after
1273 * XXX it has entered the hashes. -DaveM
1274 *
1275 * Besides that, it does not check for connection
1276 * uniqueness. Wait for troubles.
1277 */
1278 return __sk_prot_rehash(sk);
1279 }
1280
inet_sk_rebuild_header(struct sock * sk)1281 int inet_sk_rebuild_header(struct sock *sk)
1282 {
1283 struct inet_sock *inet = inet_sk(sk);
1284 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1285 __be32 daddr;
1286 struct ip_options_rcu *inet_opt;
1287 struct flowi4 *fl4;
1288 int err;
1289
1290 /* Route is OK, nothing to do. */
1291 if (rt)
1292 return 0;
1293
1294 /* Reroute. */
1295 rcu_read_lock();
1296 inet_opt = rcu_dereference(inet->inet_opt);
1297 daddr = inet->inet_daddr;
1298 if (inet_opt && inet_opt->opt.srr)
1299 daddr = inet_opt->opt.faddr;
1300 rcu_read_unlock();
1301 fl4 = &inet->cork.fl.u.ip4;
1302 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1303 inet->inet_dport, inet->inet_sport,
1304 sk->sk_protocol, RT_CONN_FLAGS(sk),
1305 sk->sk_bound_dev_if);
1306 if (!IS_ERR(rt)) {
1307 err = 0;
1308 sk_setup_caps(sk, &rt->dst);
1309 } else {
1310 err = PTR_ERR(rt);
1311
1312 /* Routing failed... */
1313 sk->sk_route_caps = 0;
1314 /*
1315 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1316 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1317 */
1318 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1319 sk->sk_state != TCP_SYN_SENT ||
1320 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1321 (err = inet_sk_reselect_saddr(sk)) != 0)
1322 sk->sk_err_soft = -err;
1323 }
1324
1325 return err;
1326 }
1327 EXPORT_SYMBOL(inet_sk_rebuild_header);
1328
inet_sk_set_state(struct sock * sk,int state)1329 void inet_sk_set_state(struct sock *sk, int state)
1330 {
1331 trace_inet_sock_set_state(sk, sk->sk_state, state);
1332 sk->sk_state = state;
1333 }
1334 EXPORT_SYMBOL(inet_sk_set_state);
1335
inet_sk_state_store(struct sock * sk,int newstate)1336 void inet_sk_state_store(struct sock *sk, int newstate)
1337 {
1338 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1339 smp_store_release(&sk->sk_state, newstate);
1340 }
1341
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1342 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1343 netdev_features_t features)
1344 {
1345 bool udpfrag = false, fixedid = false, gso_partial, encap;
1346 struct sk_buff *segs = ERR_PTR(-EINVAL);
1347 const struct net_offload *ops;
1348 unsigned int offset = 0;
1349 struct iphdr *iph;
1350 int proto, tot_len;
1351 int nhoff;
1352 int ihl;
1353 int id;
1354
1355 skb_reset_network_header(skb);
1356 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1357 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1358 goto out;
1359
1360 iph = ip_hdr(skb);
1361 ihl = iph->ihl * 4;
1362 if (ihl < sizeof(*iph))
1363 goto out;
1364
1365 id = ntohs(iph->id);
1366 proto = iph->protocol;
1367
1368 /* Warning: after this point, iph might be no longer valid */
1369 if (unlikely(!pskb_may_pull(skb, ihl)))
1370 goto out;
1371 __skb_pull(skb, ihl);
1372
1373 encap = SKB_GSO_CB(skb)->encap_level > 0;
1374 if (encap)
1375 features &= skb->dev->hw_enc_features;
1376 SKB_GSO_CB(skb)->encap_level += ihl;
1377
1378 skb_reset_transport_header(skb);
1379
1380 segs = ERR_PTR(-EPROTONOSUPPORT);
1381
1382 if (!skb->encapsulation || encap) {
1383 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1384 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1385
1386 /* fixed ID is invalid if DF bit is not set */
1387 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1388 goto out;
1389 }
1390
1391 ops = rcu_dereference(inet_offloads[proto]);
1392 if (likely(ops && ops->callbacks.gso_segment)) {
1393 segs = ops->callbacks.gso_segment(skb, features);
1394 if (!segs)
1395 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1396 }
1397
1398 if (IS_ERR_OR_NULL(segs))
1399 goto out;
1400
1401 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1402
1403 skb = segs;
1404 do {
1405 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1406 if (udpfrag) {
1407 iph->frag_off = htons(offset >> 3);
1408 if (skb->next)
1409 iph->frag_off |= htons(IP_MF);
1410 offset += skb->len - nhoff - ihl;
1411 tot_len = skb->len - nhoff;
1412 } else if (skb_is_gso(skb)) {
1413 if (!fixedid) {
1414 iph->id = htons(id);
1415 id += skb_shinfo(skb)->gso_segs;
1416 }
1417
1418 if (gso_partial)
1419 tot_len = skb_shinfo(skb)->gso_size +
1420 SKB_GSO_CB(skb)->data_offset +
1421 skb->head - (unsigned char *)iph;
1422 else
1423 tot_len = skb->len - nhoff;
1424 } else {
1425 if (!fixedid)
1426 iph->id = htons(id++);
1427 tot_len = skb->len - nhoff;
1428 }
1429 iph->tot_len = htons(tot_len);
1430 ip_send_check(iph);
1431 if (encap)
1432 skb_reset_inner_headers(skb);
1433 skb->network_header = (u8 *)iph - skb->head;
1434 skb_reset_mac_len(skb);
1435 } while ((skb = skb->next));
1436
1437 out:
1438 return segs;
1439 }
1440
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1441 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1442 netdev_features_t features)
1443 {
1444 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1445 return ERR_PTR(-EINVAL);
1446
1447 return inet_gso_segment(skb, features);
1448 }
1449
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1450 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1451 {
1452 const struct net_offload *ops;
1453 struct sk_buff *pp = NULL;
1454 const struct iphdr *iph;
1455 struct sk_buff *p;
1456 unsigned int hlen;
1457 unsigned int off;
1458 unsigned int id;
1459 int flush = 1;
1460 int proto;
1461
1462 off = skb_gro_offset(skb);
1463 hlen = off + sizeof(*iph);
1464 iph = skb_gro_header_fast(skb, off);
1465 if (skb_gro_header_hard(skb, hlen)) {
1466 iph = skb_gro_header_slow(skb, hlen, off);
1467 if (unlikely(!iph))
1468 goto out;
1469 }
1470
1471 proto = iph->protocol;
1472
1473 rcu_read_lock();
1474 ops = rcu_dereference(inet_offloads[proto]);
1475 if (!ops || !ops->callbacks.gro_receive)
1476 goto out_unlock;
1477
1478 if (*(u8 *)iph != 0x45)
1479 goto out_unlock;
1480
1481 if (ip_is_fragment(iph))
1482 goto out_unlock;
1483
1484 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1485 goto out_unlock;
1486
1487 id = ntohl(*(__be32 *)&iph->id);
1488 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1489 id >>= 16;
1490
1491 list_for_each_entry(p, head, list) {
1492 struct iphdr *iph2;
1493 u16 flush_id;
1494
1495 if (!NAPI_GRO_CB(p)->same_flow)
1496 continue;
1497
1498 iph2 = (struct iphdr *)(p->data + off);
1499 /* The above works because, with the exception of the top
1500 * (inner most) layer, we only aggregate pkts with the same
1501 * hdr length so all the hdrs we'll need to verify will start
1502 * at the same offset.
1503 */
1504 if ((iph->protocol ^ iph2->protocol) |
1505 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1506 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1507 NAPI_GRO_CB(p)->same_flow = 0;
1508 continue;
1509 }
1510
1511 /* All fields must match except length and checksum. */
1512 NAPI_GRO_CB(p)->flush |=
1513 (iph->ttl ^ iph2->ttl) |
1514 (iph->tos ^ iph2->tos) |
1515 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1516
1517 NAPI_GRO_CB(p)->flush |= flush;
1518
1519 /* We need to store of the IP ID check to be included later
1520 * when we can verify that this packet does in fact belong
1521 * to a given flow.
1522 */
1523 flush_id = (u16)(id - ntohs(iph2->id));
1524
1525 /* This bit of code makes it much easier for us to identify
1526 * the cases where we are doing atomic vs non-atomic IP ID
1527 * checks. Specifically an atomic check can return IP ID
1528 * values 0 - 0xFFFF, while a non-atomic check can only
1529 * return 0 or 0xFFFF.
1530 */
1531 if (!NAPI_GRO_CB(p)->is_atomic ||
1532 !(iph->frag_off & htons(IP_DF))) {
1533 flush_id ^= NAPI_GRO_CB(p)->count;
1534 flush_id = flush_id ? 0xFFFF : 0;
1535 }
1536
1537 /* If the previous IP ID value was based on an atomic
1538 * datagram we can overwrite the value and ignore it.
1539 */
1540 if (NAPI_GRO_CB(skb)->is_atomic)
1541 NAPI_GRO_CB(p)->flush_id = flush_id;
1542 else
1543 NAPI_GRO_CB(p)->flush_id |= flush_id;
1544 }
1545
1546 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1547 NAPI_GRO_CB(skb)->flush |= flush;
1548 skb_set_network_header(skb, off);
1549 /* The above will be needed by the transport layer if there is one
1550 * immediately following this IP hdr.
1551 */
1552
1553 /* Note : No need to call skb_gro_postpull_rcsum() here,
1554 * as we already checked checksum over ipv4 header was 0
1555 */
1556 skb_gro_pull(skb, sizeof(*iph));
1557 skb_set_transport_header(skb, skb_gro_offset(skb));
1558
1559 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1560 ops->callbacks.gro_receive, head, skb);
1561
1562 out_unlock:
1563 rcu_read_unlock();
1564
1565 out:
1566 skb_gro_flush_final(skb, pp, flush);
1567
1568 return pp;
1569 }
1570
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1571 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1572 struct sk_buff *skb)
1573 {
1574 if (NAPI_GRO_CB(skb)->encap_mark) {
1575 NAPI_GRO_CB(skb)->flush = 1;
1576 return NULL;
1577 }
1578
1579 NAPI_GRO_CB(skb)->encap_mark = 1;
1580
1581 return inet_gro_receive(head, skb);
1582 }
1583
1584 #define SECONDS_PER_DAY 86400
1585
1586 /* inet_current_timestamp - Return IP network timestamp
1587 *
1588 * Return milliseconds since midnight in network byte order.
1589 */
inet_current_timestamp(void)1590 __be32 inet_current_timestamp(void)
1591 {
1592 u32 secs;
1593 u32 msecs;
1594 struct timespec64 ts;
1595
1596 ktime_get_real_ts64(&ts);
1597
1598 /* Get secs since midnight. */
1599 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1600 /* Convert to msecs. */
1601 msecs = secs * MSEC_PER_SEC;
1602 /* Convert nsec to msec. */
1603 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1604
1605 /* Convert to network byte order. */
1606 return htonl(msecs);
1607 }
1608 EXPORT_SYMBOL(inet_current_timestamp);
1609
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1610 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1611 {
1612 unsigned int family = READ_ONCE(sk->sk_family);
1613
1614 if (family == AF_INET)
1615 return ip_recv_error(sk, msg, len, addr_len);
1616 #if IS_ENABLED(CONFIG_IPV6)
1617 if (family == AF_INET6)
1618 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1619 #endif
1620 return -EINVAL;
1621 }
1622
inet_gro_complete(struct sk_buff * skb,int nhoff)1623 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1624 {
1625 __be16 newlen = htons(skb->len - nhoff);
1626 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1627 const struct net_offload *ops;
1628 int proto = iph->protocol;
1629 int err = -ENOSYS;
1630
1631 if (skb->encapsulation) {
1632 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1633 skb_set_inner_network_header(skb, nhoff);
1634 }
1635
1636 csum_replace2(&iph->check, iph->tot_len, newlen);
1637 iph->tot_len = newlen;
1638
1639 rcu_read_lock();
1640 ops = rcu_dereference(inet_offloads[proto]);
1641 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1642 goto out_unlock;
1643
1644 /* Only need to add sizeof(*iph) to get to the next hdr below
1645 * because any hdr with option will have been flushed in
1646 * inet_gro_receive().
1647 */
1648 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1649 tcp4_gro_complete, udp4_gro_complete,
1650 skb, nhoff + sizeof(*iph));
1651
1652 out_unlock:
1653 rcu_read_unlock();
1654
1655 return err;
1656 }
1657
ipip_gro_complete(struct sk_buff * skb,int nhoff)1658 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1659 {
1660 skb->encapsulation = 1;
1661 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1662 return inet_gro_complete(skb, nhoff);
1663 }
1664
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1665 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1666 unsigned short type, unsigned char protocol,
1667 struct net *net)
1668 {
1669 struct socket *sock;
1670 int rc = sock_create_kern(net, family, type, protocol, &sock);
1671
1672 if (rc == 0) {
1673 *sk = sock->sk;
1674 (*sk)->sk_allocation = GFP_ATOMIC;
1675 /*
1676 * Unhash it so that IP input processing does not even see it,
1677 * we do not wish this socket to see incoming packets.
1678 */
1679 (*sk)->sk_prot->unhash(*sk);
1680 }
1681 return rc;
1682 }
1683 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1684
snmp_get_cpu_field(void __percpu * mib,int cpu,int offt)1685 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1686 {
1687 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1688 }
1689 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1690
snmp_fold_field(void __percpu * mib,int offt)1691 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1692 {
1693 unsigned long res = 0;
1694 int i;
1695
1696 for_each_possible_cpu(i)
1697 res += snmp_get_cpu_field(mib, i, offt);
1698 return res;
1699 }
1700 EXPORT_SYMBOL_GPL(snmp_fold_field);
1701
1702 #if BITS_PER_LONG==32
1703
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1704 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1705 size_t syncp_offset)
1706 {
1707 void *bhptr;
1708 struct u64_stats_sync *syncp;
1709 u64 v;
1710 unsigned int start;
1711
1712 bhptr = per_cpu_ptr(mib, cpu);
1713 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1714 do {
1715 start = u64_stats_fetch_begin_irq(syncp);
1716 v = *(((u64 *)bhptr) + offt);
1717 } while (u64_stats_fetch_retry_irq(syncp, start));
1718
1719 return v;
1720 }
1721 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1722
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1723 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1724 {
1725 u64 res = 0;
1726 int cpu;
1727
1728 for_each_possible_cpu(cpu) {
1729 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1730 }
1731 return res;
1732 }
1733 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1734 #endif
1735
1736 #ifdef CONFIG_IP_MULTICAST
1737 static const struct net_protocol igmp_protocol = {
1738 .handler = igmp_rcv,
1739 };
1740 #endif
1741
1742 static const struct net_protocol tcp_protocol = {
1743 .handler = tcp_v4_rcv,
1744 .err_handler = tcp_v4_err,
1745 .no_policy = 1,
1746 .icmp_strict_tag_validation = 1,
1747 };
1748
1749 static const struct net_protocol udp_protocol = {
1750 .handler = udp_rcv,
1751 .err_handler = udp_err,
1752 .no_policy = 1,
1753 };
1754
1755 static const struct net_protocol icmp_protocol = {
1756 .handler = icmp_rcv,
1757 .err_handler = icmp_err,
1758 .no_policy = 1,
1759 };
1760
ipv4_mib_init_net(struct net * net)1761 static __net_init int ipv4_mib_init_net(struct net *net)
1762 {
1763 int i;
1764
1765 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1766 if (!net->mib.tcp_statistics)
1767 goto err_tcp_mib;
1768 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1769 if (!net->mib.ip_statistics)
1770 goto err_ip_mib;
1771
1772 for_each_possible_cpu(i) {
1773 struct ipstats_mib *af_inet_stats;
1774 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1775 u64_stats_init(&af_inet_stats->syncp);
1776 }
1777
1778 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1779 if (!net->mib.net_statistics)
1780 goto err_net_mib;
1781 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1782 if (!net->mib.udp_statistics)
1783 goto err_udp_mib;
1784 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1785 if (!net->mib.udplite_statistics)
1786 goto err_udplite_mib;
1787 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1788 if (!net->mib.icmp_statistics)
1789 goto err_icmp_mib;
1790 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1791 GFP_KERNEL);
1792 if (!net->mib.icmpmsg_statistics)
1793 goto err_icmpmsg_mib;
1794
1795 tcp_mib_init(net);
1796 return 0;
1797
1798 err_icmpmsg_mib:
1799 free_percpu(net->mib.icmp_statistics);
1800 err_icmp_mib:
1801 free_percpu(net->mib.udplite_statistics);
1802 err_udplite_mib:
1803 free_percpu(net->mib.udp_statistics);
1804 err_udp_mib:
1805 free_percpu(net->mib.net_statistics);
1806 err_net_mib:
1807 free_percpu(net->mib.ip_statistics);
1808 err_ip_mib:
1809 free_percpu(net->mib.tcp_statistics);
1810 err_tcp_mib:
1811 return -ENOMEM;
1812 }
1813
ipv4_mib_exit_net(struct net * net)1814 static __net_exit void ipv4_mib_exit_net(struct net *net)
1815 {
1816 kfree(net->mib.icmpmsg_statistics);
1817 free_percpu(net->mib.icmp_statistics);
1818 free_percpu(net->mib.udplite_statistics);
1819 free_percpu(net->mib.udp_statistics);
1820 free_percpu(net->mib.net_statistics);
1821 free_percpu(net->mib.ip_statistics);
1822 free_percpu(net->mib.tcp_statistics);
1823 #ifdef CONFIG_MPTCP
1824 /* allocated on demand, see mptcp_init_sock() */
1825 free_percpu(net->mib.mptcp_statistics);
1826 #endif
1827 }
1828
1829 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1830 .init = ipv4_mib_init_net,
1831 .exit = ipv4_mib_exit_net,
1832 };
1833
init_ipv4_mibs(void)1834 static int __init init_ipv4_mibs(void)
1835 {
1836 return register_pernet_subsys(&ipv4_mib_ops);
1837 }
1838
inet_init_net(struct net * net)1839 static __net_init int inet_init_net(struct net *net)
1840 {
1841 /*
1842 * Set defaults for local port range
1843 */
1844 seqlock_init(&net->ipv4.ip_local_ports.lock);
1845 net->ipv4.ip_local_ports.range[0] = 32768;
1846 net->ipv4.ip_local_ports.range[1] = 60999;
1847
1848 seqlock_init(&net->ipv4.ping_group_range.lock);
1849 /*
1850 * Sane defaults - nobody may create ping sockets.
1851 * Boot scripts should set this to distro-specific group.
1852 */
1853 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1854 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1855
1856 /* Default values for sysctl-controlled parameters.
1857 * We set them here, in case sysctl is not compiled.
1858 */
1859 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1860 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1861 net->ipv4.sysctl_ip_dynaddr = 0;
1862 net->ipv4.sysctl_ip_early_demux = 1;
1863 net->ipv4.sysctl_udp_early_demux = 1;
1864 net->ipv4.sysctl_tcp_early_demux = 1;
1865 net->ipv4.sysctl_nexthop_compat_mode = 1;
1866 #ifdef CONFIG_SYSCTL
1867 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1868 #endif
1869
1870 /* Some igmp sysctl, whose values are always used */
1871 net->ipv4.sysctl_igmp_max_memberships = 20;
1872 net->ipv4.sysctl_igmp_max_msf = 10;
1873 /* IGMP reports for link-local multicast groups are enabled by default */
1874 net->ipv4.sysctl_igmp_llm_reports = 1;
1875 net->ipv4.sysctl_igmp_qrv = 2;
1876
1877 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1878
1879 return 0;
1880 }
1881
1882 static __net_initdata struct pernet_operations af_inet_ops = {
1883 .init = inet_init_net,
1884 };
1885
init_inet_pernet_ops(void)1886 static int __init init_inet_pernet_ops(void)
1887 {
1888 return register_pernet_subsys(&af_inet_ops);
1889 }
1890
1891 static int ipv4_proc_init(void);
1892
1893 /*
1894 * IP protocol layer initialiser
1895 */
1896
1897 static struct packet_offload ip_packet_offload __read_mostly = {
1898 .type = cpu_to_be16(ETH_P_IP),
1899 .callbacks = {
1900 .gso_segment = inet_gso_segment,
1901 .gro_receive = inet_gro_receive,
1902 .gro_complete = inet_gro_complete,
1903 },
1904 };
1905
1906 static const struct net_offload ipip_offload = {
1907 .callbacks = {
1908 .gso_segment = ipip_gso_segment,
1909 .gro_receive = ipip_gro_receive,
1910 .gro_complete = ipip_gro_complete,
1911 },
1912 };
1913
ipip_offload_init(void)1914 static int __init ipip_offload_init(void)
1915 {
1916 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1917 }
1918
ipv4_offload_init(void)1919 static int __init ipv4_offload_init(void)
1920 {
1921 /*
1922 * Add offloads
1923 */
1924 if (udpv4_offload_init() < 0)
1925 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1926 if (tcpv4_offload_init() < 0)
1927 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1928 if (ipip_offload_init() < 0)
1929 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1930
1931 dev_add_offload(&ip_packet_offload);
1932 return 0;
1933 }
1934
1935 fs_initcall(ipv4_offload_init);
1936
1937 static struct packet_type ip_packet_type __read_mostly = {
1938 .type = cpu_to_be16(ETH_P_IP),
1939 .func = ip_rcv,
1940 .list_func = ip_list_rcv,
1941 };
1942
inet_init(void)1943 static int __init inet_init(void)
1944 {
1945 struct inet_protosw *q;
1946 struct list_head *r;
1947 int rc;
1948
1949 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1950
1951 rc = proto_register(&tcp_prot, 1);
1952 if (rc)
1953 goto out;
1954
1955 rc = proto_register(&udp_prot, 1);
1956 if (rc)
1957 goto out_unregister_tcp_proto;
1958
1959 rc = proto_register(&raw_prot, 1);
1960 if (rc)
1961 goto out_unregister_udp_proto;
1962
1963 rc = proto_register(&ping_prot, 1);
1964 if (rc)
1965 goto out_unregister_raw_proto;
1966
1967 /*
1968 * Tell SOCKET that we are alive...
1969 */
1970
1971 (void)sock_register(&inet_family_ops);
1972
1973 #ifdef CONFIG_SYSCTL
1974 ip_static_sysctl_init();
1975 #endif
1976
1977 /*
1978 * Add all the base protocols.
1979 */
1980
1981 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1982 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1983 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1984 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1985 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1986 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1987 #ifdef CONFIG_IP_MULTICAST
1988 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1989 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1990 #endif
1991
1992 /* Register the socket-side information for inet_create. */
1993 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1994 INIT_LIST_HEAD(r);
1995
1996 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1997 inet_register_protosw(q);
1998
1999 /*
2000 * Set the ARP module up
2001 */
2002
2003 arp_init();
2004
2005 /*
2006 * Set the IP module up
2007 */
2008
2009 ip_init();
2010
2011 /* Initialise per-cpu ipv4 mibs */
2012 if (init_ipv4_mibs())
2013 panic("%s: Cannot init ipv4 mibs\n", __func__);
2014
2015 /* Setup TCP slab cache for open requests. */
2016 tcp_init();
2017
2018 /* Setup UDP memory threshold */
2019 udp_init();
2020
2021 /* Add UDP-Lite (RFC 3828) */
2022 udplite4_register();
2023
2024 raw_init();
2025
2026 ping_init();
2027
2028 /*
2029 * Set the ICMP layer up
2030 */
2031
2032 if (icmp_init() < 0)
2033 panic("Failed to create the ICMP control socket.\n");
2034
2035 /*
2036 * Initialise the multicast router
2037 */
2038 #if defined(CONFIG_IP_MROUTE)
2039 if (ip_mr_init())
2040 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2041 #endif
2042
2043 if (init_inet_pernet_ops())
2044 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2045
2046 ipv4_proc_init();
2047
2048 ipfrag_init();
2049
2050 dev_add_pack(&ip_packet_type);
2051
2052 ip_tunnel_core_init();
2053
2054 rc = 0;
2055 out:
2056 return rc;
2057 out_unregister_raw_proto:
2058 proto_unregister(&raw_prot);
2059 out_unregister_udp_proto:
2060 proto_unregister(&udp_prot);
2061 out_unregister_tcp_proto:
2062 proto_unregister(&tcp_prot);
2063 goto out;
2064 }
2065
2066 fs_initcall(inet_init);
2067
2068 /* ------------------------------------------------------------------------ */
2069
2070 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2071 static int __init ipv4_proc_init(void)
2072 {
2073 int rc = 0;
2074
2075 if (raw_proc_init())
2076 goto out_raw;
2077 if (tcp4_proc_init())
2078 goto out_tcp;
2079 if (udp4_proc_init())
2080 goto out_udp;
2081 if (ping_proc_init())
2082 goto out_ping;
2083 if (ip_misc_proc_init())
2084 goto out_misc;
2085 out:
2086 return rc;
2087 out_misc:
2088 ping_proc_exit();
2089 out_ping:
2090 udp4_proc_exit();
2091 out_udp:
2092 tcp4_proc_exit();
2093 out_tcp:
2094 raw_proc_exit();
2095 out_raw:
2096 rc = -ENOMEM;
2097 goto out;
2098 }
2099
2100 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2101 static int __init ipv4_proc_init(void)
2102 {
2103 return 0;
2104 }
2105 #endif /* CONFIG_PROC_FS */
2106