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 = -ENOBUFS;
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 int err;
445
446 /* If the socket has its own bind function then use it. (RAW) */
447 if (sk->sk_prot->bind) {
448 return sk->sk_prot->bind(sk, uaddr, addr_len);
449 }
450 if (addr_len < sizeof(struct sockaddr_in))
451 return -EINVAL;
452
453 /* BPF prog is run before any checks are done so that if the prog
454 * changes context in a wrong way it will be caught.
455 */
456 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
457 if (err)
458 return err;
459
460 return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
461 }
462 EXPORT_SYMBOL(inet_bind);
463
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)464 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
465 u32 flags)
466 {
467 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
468 struct inet_sock *inet = inet_sk(sk);
469 struct net *net = sock_net(sk);
470 unsigned short snum;
471 int chk_addr_ret;
472 u32 tb_id = RT_TABLE_LOCAL;
473 int err;
474
475 if (addr->sin_family != AF_INET) {
476 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
477 * only if s_addr is INADDR_ANY.
478 */
479 err = -EAFNOSUPPORT;
480 if (addr->sin_family != AF_UNSPEC ||
481 addr->sin_addr.s_addr != htonl(INADDR_ANY))
482 goto out;
483 }
484
485 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
486 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
487
488 /* Not specified by any standard per-se, however it breaks too
489 * many applications when removed. It is unfortunate since
490 * allowing applications to make a non-local bind solves
491 * several problems with systems using dynamic addressing.
492 * (ie. your servers still start up even if your ISDN link
493 * is temporarily down)
494 */
495 err = -EADDRNOTAVAIL;
496 if (!inet_can_nonlocal_bind(net, inet) &&
497 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
498 chk_addr_ret != RTN_LOCAL &&
499 chk_addr_ret != RTN_MULTICAST &&
500 chk_addr_ret != RTN_BROADCAST)
501 goto out;
502
503 snum = ntohs(addr->sin_port);
504 err = -EPERM;
505 if (snum && inet_is_local_unbindable_port(net, snum))
506 goto out;
507
508 err = -EACCES;
509 if (snum && inet_port_requires_bind_service(net, snum) &&
510 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
511 goto out;
512
513 /* We keep a pair of addresses. rcv_saddr is the one
514 * used by hash lookups, and saddr is used for transmit.
515 *
516 * In the BSD API these are the same except where it
517 * would be illegal to use them (multicast/broadcast) in
518 * which case the sending device address is used.
519 */
520 if (flags & BIND_WITH_LOCK)
521 lock_sock(sk);
522
523 /* Check these errors (active socket, double bind). */
524 err = -EINVAL;
525 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
526 goto out_release_sock;
527
528 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
529 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
530 inet->inet_saddr = 0; /* Use device */
531
532 /* Make sure we are allowed to bind here. */
533 if (snum || !(inet->bind_address_no_port ||
534 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
535 if (sk->sk_prot->get_port(sk, snum)) {
536 inet->inet_saddr = inet->inet_rcv_saddr = 0;
537 err = -EADDRINUSE;
538 goto out_release_sock;
539 }
540 if (!(flags & BIND_FROM_BPF)) {
541 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
542 if (err) {
543 inet->inet_saddr = inet->inet_rcv_saddr = 0;
544 goto out_release_sock;
545 }
546 }
547 }
548
549 if (inet->inet_rcv_saddr)
550 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
551 if (snum)
552 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
553 inet->inet_sport = htons(inet->inet_num);
554 inet->inet_daddr = 0;
555 inet->inet_dport = 0;
556 sk_dst_reset(sk);
557 err = 0;
558 out_release_sock:
559 if (flags & BIND_WITH_LOCK)
560 release_sock(sk);
561 out:
562 return err;
563 }
564
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)565 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
566 int addr_len, int flags)
567 {
568 struct sock *sk = sock->sk;
569 int err;
570
571 if (addr_len < sizeof(uaddr->sa_family))
572 return -EINVAL;
573 if (uaddr->sa_family == AF_UNSPEC)
574 return sk->sk_prot->disconnect(sk, flags);
575
576 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
577 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
578 if (err)
579 return err;
580 }
581
582 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
583 return -EAGAIN;
584 return sk->sk_prot->connect(sk, uaddr, addr_len);
585 }
586 EXPORT_SYMBOL(inet_dgram_connect);
587
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)588 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
589 {
590 DEFINE_WAIT_FUNC(wait, woken_wake_function);
591
592 add_wait_queue(sk_sleep(sk), &wait);
593 sk->sk_write_pending += writebias;
594
595 /* Basic assumption: if someone sets sk->sk_err, he _must_
596 * change state of the socket from TCP_SYN_*.
597 * Connect() does not allow to get error notifications
598 * without closing the socket.
599 */
600 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
601 release_sock(sk);
602 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
603 lock_sock(sk);
604 if (signal_pending(current) || !timeo)
605 break;
606 }
607 remove_wait_queue(sk_sleep(sk), &wait);
608 sk->sk_write_pending -= writebias;
609 return timeo;
610 }
611
612 /*
613 * Connect to a remote host. There is regrettably still a little
614 * TCP 'magic' in here.
615 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)616 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
617 int addr_len, int flags, int is_sendmsg)
618 {
619 struct sock *sk = sock->sk;
620 int err;
621 long timeo;
622
623 /*
624 * uaddr can be NULL and addr_len can be 0 if:
625 * sk is a TCP fastopen active socket and
626 * TCP_FASTOPEN_CONNECT sockopt is set and
627 * we already have a valid cookie for this socket.
628 * In this case, user can call write() after connect().
629 * write() will invoke tcp_sendmsg_fastopen() which calls
630 * __inet_stream_connect().
631 */
632 if (uaddr) {
633 if (addr_len < sizeof(uaddr->sa_family))
634 return -EINVAL;
635
636 if (uaddr->sa_family == AF_UNSPEC) {
637 err = sk->sk_prot->disconnect(sk, flags);
638 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
639 goto out;
640 }
641 }
642
643 switch (sock->state) {
644 default:
645 err = -EINVAL;
646 goto out;
647 case SS_CONNECTED:
648 err = -EISCONN;
649 goto out;
650 case SS_CONNECTING:
651 if (inet_sk(sk)->defer_connect)
652 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
653 else
654 err = -EALREADY;
655 /* Fall out of switch with err, set for this state */
656 break;
657 case SS_UNCONNECTED:
658 err = -EISCONN;
659 if (sk->sk_state != TCP_CLOSE)
660 goto out;
661
662 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
663 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
664 if (err)
665 goto out;
666 }
667
668 err = sk->sk_prot->connect(sk, uaddr, addr_len);
669 if (err < 0)
670 goto out;
671
672 sock->state = SS_CONNECTING;
673
674 if (!err && inet_sk(sk)->defer_connect)
675 goto out;
676
677 /* Just entered SS_CONNECTING state; the only
678 * difference is that return value in non-blocking
679 * case is EINPROGRESS, rather than EALREADY.
680 */
681 err = -EINPROGRESS;
682 break;
683 }
684
685 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
686
687 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
688 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
689 tcp_sk(sk)->fastopen_req &&
690 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
691
692 /* Error code is set above */
693 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
694 goto out;
695
696 err = sock_intr_errno(timeo);
697 if (signal_pending(current))
698 goto out;
699 }
700
701 /* Connection was closed by RST, timeout, ICMP error
702 * or another process disconnected us.
703 */
704 if (sk->sk_state == TCP_CLOSE)
705 goto sock_error;
706
707 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
708 * and error was received after socket entered established state.
709 * Hence, it is handled normally after connect() return successfully.
710 */
711
712 sock->state = SS_CONNECTED;
713 err = 0;
714 out:
715 return err;
716
717 sock_error:
718 err = sock_error(sk) ? : -ECONNABORTED;
719 sock->state = SS_UNCONNECTED;
720 if (sk->sk_prot->disconnect(sk, flags))
721 sock->state = SS_DISCONNECTING;
722 goto out;
723 }
724 EXPORT_SYMBOL(__inet_stream_connect);
725
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)726 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
727 int addr_len, int flags)
728 {
729 int err;
730
731 lock_sock(sock->sk);
732 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
733 release_sock(sock->sk);
734 return err;
735 }
736 EXPORT_SYMBOL(inet_stream_connect);
737
738 /*
739 * Accept a pending connection. The TCP layer now gives BSD semantics.
740 */
741
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)742 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
743 bool kern)
744 {
745 struct sock *sk1 = sock->sk;
746 int err = -EINVAL;
747 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
748
749 if (!sk2)
750 goto do_err;
751
752 lock_sock(sk2);
753
754 sock_rps_record_flow(sk2);
755 WARN_ON(!((1 << sk2->sk_state) &
756 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
757 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
758
759 sock_graft(sk2, newsock);
760
761 newsock->state = SS_CONNECTED;
762 err = 0;
763 release_sock(sk2);
764 do_err:
765 return err;
766 }
767 EXPORT_SYMBOL(inet_accept);
768
769 /*
770 * This does both peername and sockname.
771 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)772 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
773 int peer)
774 {
775 struct sock *sk = sock->sk;
776 struct inet_sock *inet = inet_sk(sk);
777 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
778
779 sin->sin_family = AF_INET;
780 if (peer) {
781 if (!inet->inet_dport ||
782 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
783 peer == 1))
784 return -ENOTCONN;
785 sin->sin_port = inet->inet_dport;
786 sin->sin_addr.s_addr = inet->inet_daddr;
787 } else {
788 __be32 addr = inet->inet_rcv_saddr;
789 if (!addr)
790 addr = inet->inet_saddr;
791 sin->sin_port = inet->inet_sport;
792 sin->sin_addr.s_addr = addr;
793 }
794 if (cgroup_bpf_enabled)
795 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
796 peer ? BPF_CGROUP_INET4_GETPEERNAME :
797 BPF_CGROUP_INET4_GETSOCKNAME,
798 NULL);
799 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
800 return sizeof(*sin);
801 }
802 EXPORT_SYMBOL(inet_getname);
803
inet_send_prepare(struct sock * sk)804 int inet_send_prepare(struct sock *sk)
805 {
806 sock_rps_record_flow(sk);
807
808 /* We may need to bind the socket. */
809 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
810 inet_autobind(sk))
811 return -EAGAIN;
812
813 return 0;
814 }
815 EXPORT_SYMBOL_GPL(inet_send_prepare);
816
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)817 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
818 {
819 struct sock *sk = sock->sk;
820
821 if (unlikely(inet_send_prepare(sk)))
822 return -EAGAIN;
823
824 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
825 sk, msg, size);
826 }
827 EXPORT_SYMBOL(inet_sendmsg);
828
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)829 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
830 size_t size, int flags)
831 {
832 struct sock *sk = sock->sk;
833
834 if (unlikely(inet_send_prepare(sk)))
835 return -EAGAIN;
836
837 if (sk->sk_prot->sendpage)
838 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
839 return sock_no_sendpage(sock, page, offset, size, flags);
840 }
841 EXPORT_SYMBOL(inet_sendpage);
842
843 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
844 size_t, int, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)845 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
846 int flags)
847 {
848 struct sock *sk = sock->sk;
849 int addr_len = 0;
850 int err;
851
852 if (likely(!(flags & MSG_ERRQUEUE)))
853 sock_rps_record_flow(sk);
854
855 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
856 sk, msg, size, flags & MSG_DONTWAIT,
857 flags & ~MSG_DONTWAIT, &addr_len);
858 if (err >= 0)
859 msg->msg_namelen = addr_len;
860 return err;
861 }
862 EXPORT_SYMBOL(inet_recvmsg);
863
inet_shutdown(struct socket * sock,int how)864 int inet_shutdown(struct socket *sock, int how)
865 {
866 struct sock *sk = sock->sk;
867 int err = 0;
868
869 /* This should really check to make sure
870 * the socket is a TCP socket. (WHY AC...)
871 */
872 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
873 1->2 bit 2 snds.
874 2->3 */
875 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
876 return -EINVAL;
877
878 lock_sock(sk);
879 if (sock->state == SS_CONNECTING) {
880 if ((1 << sk->sk_state) &
881 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
882 sock->state = SS_DISCONNECTING;
883 else
884 sock->state = SS_CONNECTED;
885 }
886
887 switch (sk->sk_state) {
888 case TCP_CLOSE:
889 err = -ENOTCONN;
890 /* Hack to wake up other listeners, who can poll for
891 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
892 fallthrough;
893 default:
894 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
895 if (sk->sk_prot->shutdown)
896 sk->sk_prot->shutdown(sk, how);
897 break;
898
899 /* Remaining two branches are temporary solution for missing
900 * close() in multithreaded environment. It is _not_ a good idea,
901 * but we have no choice until close() is repaired at VFS level.
902 */
903 case TCP_LISTEN:
904 if (!(how & RCV_SHUTDOWN))
905 break;
906 fallthrough;
907 case TCP_SYN_SENT:
908 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
909 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
910 break;
911 }
912
913 /* Wake up anyone sleeping in poll. */
914 sk->sk_state_change(sk);
915 release_sock(sk);
916 return err;
917 }
918 EXPORT_SYMBOL(inet_shutdown);
919
920 /*
921 * ioctl() calls you can issue on an INET socket. Most of these are
922 * device configuration and stuff and very rarely used. Some ioctls
923 * pass on to the socket itself.
924 *
925 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
926 * loads the devconfigure module does its configuring and unloads it.
927 * There's a good 20K of config code hanging around the kernel.
928 */
929
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)930 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
931 {
932 struct sock *sk = sock->sk;
933 int err = 0;
934 struct net *net = sock_net(sk);
935 void __user *p = (void __user *)arg;
936 struct ifreq ifr;
937 struct rtentry rt;
938
939 switch (cmd) {
940 case SIOCADDRT:
941 case SIOCDELRT:
942 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
943 return -EFAULT;
944 err = ip_rt_ioctl(net, cmd, &rt);
945 break;
946 case SIOCRTMSG:
947 err = -EINVAL;
948 break;
949 case SIOCDARP:
950 case SIOCGARP:
951 case SIOCSARP:
952 err = arp_ioctl(net, cmd, (void __user *)arg);
953 break;
954 case SIOCGIFADDR:
955 case SIOCGIFBRDADDR:
956 case SIOCGIFNETMASK:
957 case SIOCGIFDSTADDR:
958 case SIOCGIFPFLAGS:
959 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
960 return -EFAULT;
961 err = devinet_ioctl(net, cmd, &ifr);
962 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
963 err = -EFAULT;
964 break;
965
966 case SIOCSIFADDR:
967 case SIOCSIFBRDADDR:
968 case SIOCSIFNETMASK:
969 case SIOCSIFDSTADDR:
970 case SIOCSIFPFLAGS:
971 case SIOCSIFFLAGS:
972 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
973 return -EFAULT;
974 err = devinet_ioctl(net, cmd, &ifr);
975 break;
976 default:
977 if (sk->sk_prot->ioctl)
978 err = sk->sk_prot->ioctl(sk, cmd, arg);
979 else
980 err = -ENOIOCTLCMD;
981 break;
982 }
983 return err;
984 }
985 EXPORT_SYMBOL(inet_ioctl);
986
987 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)988 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
989 struct compat_rtentry __user *ur)
990 {
991 compat_uptr_t rtdev;
992 struct rtentry rt;
993
994 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
995 3 * sizeof(struct sockaddr)) ||
996 get_user(rt.rt_flags, &ur->rt_flags) ||
997 get_user(rt.rt_metric, &ur->rt_metric) ||
998 get_user(rt.rt_mtu, &ur->rt_mtu) ||
999 get_user(rt.rt_window, &ur->rt_window) ||
1000 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1001 get_user(rtdev, &ur->rt_dev))
1002 return -EFAULT;
1003
1004 rt.rt_dev = compat_ptr(rtdev);
1005 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1006 }
1007
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1008 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1009 {
1010 void __user *argp = compat_ptr(arg);
1011 struct sock *sk = sock->sk;
1012
1013 switch (cmd) {
1014 case SIOCADDRT:
1015 case SIOCDELRT:
1016 return inet_compat_routing_ioctl(sk, cmd, argp);
1017 default:
1018 if (!sk->sk_prot->compat_ioctl)
1019 return -ENOIOCTLCMD;
1020 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1021 }
1022 }
1023 #endif /* CONFIG_COMPAT */
1024
1025 const struct proto_ops inet_stream_ops = {
1026 .family = PF_INET,
1027 .owner = THIS_MODULE,
1028 .release = inet_release,
1029 .bind = inet_bind,
1030 .connect = inet_stream_connect,
1031 .socketpair = sock_no_socketpair,
1032 .accept = inet_accept,
1033 .getname = inet_getname,
1034 .poll = tcp_poll,
1035 .ioctl = inet_ioctl,
1036 .gettstamp = sock_gettstamp,
1037 .listen = inet_listen,
1038 .shutdown = inet_shutdown,
1039 .setsockopt = sock_common_setsockopt,
1040 .getsockopt = sock_common_getsockopt,
1041 .sendmsg = inet_sendmsg,
1042 .recvmsg = inet_recvmsg,
1043 #ifdef CONFIG_MMU
1044 .mmap = tcp_mmap,
1045 #endif
1046 .sendpage = inet_sendpage,
1047 .splice_read = tcp_splice_read,
1048 .read_sock = tcp_read_sock,
1049 .sendmsg_locked = tcp_sendmsg_locked,
1050 .sendpage_locked = tcp_sendpage_locked,
1051 .peek_len = tcp_peek_len,
1052 #ifdef CONFIG_COMPAT
1053 .compat_ioctl = inet_compat_ioctl,
1054 #endif
1055 .set_rcvlowat = tcp_set_rcvlowat,
1056 };
1057 EXPORT_SYMBOL(inet_stream_ops);
1058
1059 const struct proto_ops inet_dgram_ops = {
1060 .family = PF_INET,
1061 .owner = THIS_MODULE,
1062 .release = inet_release,
1063 .bind = inet_bind,
1064 .connect = inet_dgram_connect,
1065 .socketpair = sock_no_socketpair,
1066 .accept = sock_no_accept,
1067 .getname = inet_getname,
1068 .poll = udp_poll,
1069 .ioctl = inet_ioctl,
1070 .gettstamp = sock_gettstamp,
1071 .listen = sock_no_listen,
1072 .shutdown = inet_shutdown,
1073 .setsockopt = sock_common_setsockopt,
1074 .getsockopt = sock_common_getsockopt,
1075 .sendmsg = inet_sendmsg,
1076 .recvmsg = inet_recvmsg,
1077 .mmap = sock_no_mmap,
1078 .sendpage = inet_sendpage,
1079 .set_peek_off = sk_set_peek_off,
1080 #ifdef CONFIG_COMPAT
1081 .compat_ioctl = inet_compat_ioctl,
1082 #endif
1083 };
1084 EXPORT_SYMBOL(inet_dgram_ops);
1085
1086 /*
1087 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1088 * udp_poll
1089 */
1090 static const struct proto_ops inet_sockraw_ops = {
1091 .family = PF_INET,
1092 .owner = THIS_MODULE,
1093 .release = inet_release,
1094 .bind = inet_bind,
1095 .connect = inet_dgram_connect,
1096 .socketpair = sock_no_socketpair,
1097 .accept = sock_no_accept,
1098 .getname = inet_getname,
1099 .poll = datagram_poll,
1100 .ioctl = inet_ioctl,
1101 .gettstamp = sock_gettstamp,
1102 .listen = sock_no_listen,
1103 .shutdown = inet_shutdown,
1104 .setsockopt = sock_common_setsockopt,
1105 .getsockopt = sock_common_getsockopt,
1106 .sendmsg = inet_sendmsg,
1107 .recvmsg = inet_recvmsg,
1108 .mmap = sock_no_mmap,
1109 .sendpage = inet_sendpage,
1110 #ifdef CONFIG_COMPAT
1111 .compat_ioctl = inet_compat_ioctl,
1112 #endif
1113 };
1114
1115 static const struct net_proto_family inet_family_ops = {
1116 .family = PF_INET,
1117 .create = inet_create,
1118 .owner = THIS_MODULE,
1119 };
1120
1121 /* Upon startup we insert all the elements in inetsw_array[] into
1122 * the linked list inetsw.
1123 */
1124 static struct inet_protosw inetsw_array[] =
1125 {
1126 {
1127 .type = SOCK_STREAM,
1128 .protocol = IPPROTO_TCP,
1129 .prot = &tcp_prot,
1130 .ops = &inet_stream_ops,
1131 .flags = INET_PROTOSW_PERMANENT |
1132 INET_PROTOSW_ICSK,
1133 },
1134
1135 {
1136 .type = SOCK_DGRAM,
1137 .protocol = IPPROTO_UDP,
1138 .prot = &udp_prot,
1139 .ops = &inet_dgram_ops,
1140 .flags = INET_PROTOSW_PERMANENT,
1141 },
1142
1143 {
1144 .type = SOCK_DGRAM,
1145 .protocol = IPPROTO_ICMP,
1146 .prot = &ping_prot,
1147 .ops = &inet_sockraw_ops,
1148 .flags = INET_PROTOSW_REUSE,
1149 },
1150
1151 {
1152 .type = SOCK_RAW,
1153 .protocol = IPPROTO_IP, /* wild card */
1154 .prot = &raw_prot,
1155 .ops = &inet_sockraw_ops,
1156 .flags = INET_PROTOSW_REUSE,
1157 }
1158 };
1159
1160 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1161
inet_register_protosw(struct inet_protosw * p)1162 void inet_register_protosw(struct inet_protosw *p)
1163 {
1164 struct list_head *lh;
1165 struct inet_protosw *answer;
1166 int protocol = p->protocol;
1167 struct list_head *last_perm;
1168
1169 spin_lock_bh(&inetsw_lock);
1170
1171 if (p->type >= SOCK_MAX)
1172 goto out_illegal;
1173
1174 /* If we are trying to override a permanent protocol, bail. */
1175 last_perm = &inetsw[p->type];
1176 list_for_each(lh, &inetsw[p->type]) {
1177 answer = list_entry(lh, struct inet_protosw, list);
1178 /* Check only the non-wild match. */
1179 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1180 break;
1181 if (protocol == answer->protocol)
1182 goto out_permanent;
1183 last_perm = lh;
1184 }
1185
1186 /* Add the new entry after the last permanent entry if any, so that
1187 * the new entry does not override a permanent entry when matched with
1188 * a wild-card protocol. But it is allowed to override any existing
1189 * non-permanent entry. This means that when we remove this entry, the
1190 * system automatically returns to the old behavior.
1191 */
1192 list_add_rcu(&p->list, last_perm);
1193 out:
1194 spin_unlock_bh(&inetsw_lock);
1195
1196 return;
1197
1198 out_permanent:
1199 pr_err("Attempt to override permanent protocol %d\n", protocol);
1200 goto out;
1201
1202 out_illegal:
1203 pr_err("Ignoring attempt to register invalid socket type %d\n",
1204 p->type);
1205 goto out;
1206 }
1207 EXPORT_SYMBOL(inet_register_protosw);
1208
inet_unregister_protosw(struct inet_protosw * p)1209 void inet_unregister_protosw(struct inet_protosw *p)
1210 {
1211 if (INET_PROTOSW_PERMANENT & p->flags) {
1212 pr_err("Attempt to unregister permanent protocol %d\n",
1213 p->protocol);
1214 } else {
1215 spin_lock_bh(&inetsw_lock);
1216 list_del_rcu(&p->list);
1217 spin_unlock_bh(&inetsw_lock);
1218
1219 synchronize_net();
1220 }
1221 }
1222 EXPORT_SYMBOL(inet_unregister_protosw);
1223
inet_sk_reselect_saddr(struct sock * sk)1224 static int inet_sk_reselect_saddr(struct sock *sk)
1225 {
1226 struct inet_sock *inet = inet_sk(sk);
1227 __be32 old_saddr = inet->inet_saddr;
1228 __be32 daddr = inet->inet_daddr;
1229 struct flowi4 *fl4;
1230 struct rtable *rt;
1231 __be32 new_saddr;
1232 struct ip_options_rcu *inet_opt;
1233
1234 inet_opt = rcu_dereference_protected(inet->inet_opt,
1235 lockdep_sock_is_held(sk));
1236 if (inet_opt && inet_opt->opt.srr)
1237 daddr = inet_opt->opt.faddr;
1238
1239 /* Query new route. */
1240 fl4 = &inet->cork.fl.u.ip4;
1241 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1242 sk->sk_bound_dev_if, sk->sk_protocol,
1243 inet->inet_sport, inet->inet_dport, sk);
1244 if (IS_ERR(rt))
1245 return PTR_ERR(rt);
1246
1247 sk_setup_caps(sk, &rt->dst);
1248
1249 new_saddr = fl4->saddr;
1250
1251 if (new_saddr == old_saddr)
1252 return 0;
1253
1254 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1255 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1256 __func__, &old_saddr, &new_saddr);
1257 }
1258
1259 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1260
1261 /*
1262 * XXX The only one ugly spot where we need to
1263 * XXX really change the sockets identity after
1264 * XXX it has entered the hashes. -DaveM
1265 *
1266 * Besides that, it does not check for connection
1267 * uniqueness. Wait for troubles.
1268 */
1269 return __sk_prot_rehash(sk);
1270 }
1271
inet_sk_rebuild_header(struct sock * sk)1272 int inet_sk_rebuild_header(struct sock *sk)
1273 {
1274 struct inet_sock *inet = inet_sk(sk);
1275 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1276 __be32 daddr;
1277 struct ip_options_rcu *inet_opt;
1278 struct flowi4 *fl4;
1279 int err;
1280
1281 /* Route is OK, nothing to do. */
1282 if (rt)
1283 return 0;
1284
1285 /* Reroute. */
1286 rcu_read_lock();
1287 inet_opt = rcu_dereference(inet->inet_opt);
1288 daddr = inet->inet_daddr;
1289 if (inet_opt && inet_opt->opt.srr)
1290 daddr = inet_opt->opt.faddr;
1291 rcu_read_unlock();
1292 fl4 = &inet->cork.fl.u.ip4;
1293 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1294 inet->inet_dport, inet->inet_sport,
1295 sk->sk_protocol, RT_CONN_FLAGS(sk),
1296 sk->sk_bound_dev_if);
1297 if (!IS_ERR(rt)) {
1298 err = 0;
1299 sk_setup_caps(sk, &rt->dst);
1300 } else {
1301 err = PTR_ERR(rt);
1302
1303 /* Routing failed... */
1304 sk->sk_route_caps = 0;
1305 /*
1306 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1307 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1308 */
1309 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1310 sk->sk_state != TCP_SYN_SENT ||
1311 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1312 (err = inet_sk_reselect_saddr(sk)) != 0)
1313 sk->sk_err_soft = -err;
1314 }
1315
1316 return err;
1317 }
1318 EXPORT_SYMBOL(inet_sk_rebuild_header);
1319
inet_sk_set_state(struct sock * sk,int state)1320 void inet_sk_set_state(struct sock *sk, int state)
1321 {
1322 trace_inet_sock_set_state(sk, sk->sk_state, state);
1323 sk->sk_state = state;
1324 }
1325 EXPORT_SYMBOL(inet_sk_set_state);
1326
inet_sk_state_store(struct sock * sk,int newstate)1327 void inet_sk_state_store(struct sock *sk, int newstate)
1328 {
1329 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1330 smp_store_release(&sk->sk_state, newstate);
1331 }
1332
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1333 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1334 netdev_features_t features)
1335 {
1336 bool udpfrag = false, fixedid = false, gso_partial, encap;
1337 struct sk_buff *segs = ERR_PTR(-EINVAL);
1338 const struct net_offload *ops;
1339 unsigned int offset = 0;
1340 struct iphdr *iph;
1341 int proto, tot_len;
1342 int nhoff;
1343 int ihl;
1344 int id;
1345
1346 skb_reset_network_header(skb);
1347 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1348 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1349 goto out;
1350
1351 iph = ip_hdr(skb);
1352 ihl = iph->ihl * 4;
1353 if (ihl < sizeof(*iph))
1354 goto out;
1355
1356 id = ntohs(iph->id);
1357 proto = iph->protocol;
1358
1359 /* Warning: after this point, iph might be no longer valid */
1360 if (unlikely(!pskb_may_pull(skb, ihl)))
1361 goto out;
1362 __skb_pull(skb, ihl);
1363
1364 encap = SKB_GSO_CB(skb)->encap_level > 0;
1365 if (encap)
1366 features &= skb->dev->hw_enc_features;
1367 SKB_GSO_CB(skb)->encap_level += ihl;
1368
1369 skb_reset_transport_header(skb);
1370
1371 segs = ERR_PTR(-EPROTONOSUPPORT);
1372
1373 if (!skb->encapsulation || encap) {
1374 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1375 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1376
1377 /* fixed ID is invalid if DF bit is not set */
1378 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1379 goto out;
1380 }
1381
1382 ops = rcu_dereference(inet_offloads[proto]);
1383 if (likely(ops && ops->callbacks.gso_segment)) {
1384 segs = ops->callbacks.gso_segment(skb, features);
1385 if (!segs)
1386 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1387 }
1388
1389 if (IS_ERR_OR_NULL(segs))
1390 goto out;
1391
1392 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1393
1394 skb = segs;
1395 do {
1396 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1397 if (udpfrag) {
1398 iph->frag_off = htons(offset >> 3);
1399 if (skb->next)
1400 iph->frag_off |= htons(IP_MF);
1401 offset += skb->len - nhoff - ihl;
1402 tot_len = skb->len - nhoff;
1403 } else if (skb_is_gso(skb)) {
1404 if (!fixedid) {
1405 iph->id = htons(id);
1406 id += skb_shinfo(skb)->gso_segs;
1407 }
1408
1409 if (gso_partial)
1410 tot_len = skb_shinfo(skb)->gso_size +
1411 SKB_GSO_CB(skb)->data_offset +
1412 skb->head - (unsigned char *)iph;
1413 else
1414 tot_len = skb->len - nhoff;
1415 } else {
1416 if (!fixedid)
1417 iph->id = htons(id++);
1418 tot_len = skb->len - nhoff;
1419 }
1420 iph->tot_len = htons(tot_len);
1421 ip_send_check(iph);
1422 if (encap)
1423 skb_reset_inner_headers(skb);
1424 skb->network_header = (u8 *)iph - skb->head;
1425 skb_reset_mac_len(skb);
1426 } while ((skb = skb->next));
1427
1428 out:
1429 return segs;
1430 }
1431 EXPORT_SYMBOL(inet_gso_segment);
1432
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1433 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1434 netdev_features_t features)
1435 {
1436 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1437 return ERR_PTR(-EINVAL);
1438
1439 return inet_gso_segment(skb, features);
1440 }
1441
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1442 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1443 {
1444 const struct net_offload *ops;
1445 struct sk_buff *pp = NULL;
1446 const struct iphdr *iph;
1447 struct sk_buff *p;
1448 unsigned int hlen;
1449 unsigned int off;
1450 unsigned int id;
1451 int flush = 1;
1452 int proto;
1453
1454 off = skb_gro_offset(skb);
1455 hlen = off + sizeof(*iph);
1456 iph = skb_gro_header_fast(skb, off);
1457 if (skb_gro_header_hard(skb, hlen)) {
1458 iph = skb_gro_header_slow(skb, hlen, off);
1459 if (unlikely(!iph))
1460 goto out;
1461 }
1462
1463 proto = iph->protocol;
1464
1465 rcu_read_lock();
1466 ops = rcu_dereference(inet_offloads[proto]);
1467 if (!ops || !ops->callbacks.gro_receive)
1468 goto out_unlock;
1469
1470 if (*(u8 *)iph != 0x45)
1471 goto out_unlock;
1472
1473 if (ip_is_fragment(iph))
1474 goto out_unlock;
1475
1476 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1477 goto out_unlock;
1478
1479 id = ntohl(*(__be32 *)&iph->id);
1480 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1481 id >>= 16;
1482
1483 list_for_each_entry(p, head, list) {
1484 struct iphdr *iph2;
1485 u16 flush_id;
1486
1487 if (!NAPI_GRO_CB(p)->same_flow)
1488 continue;
1489
1490 iph2 = (struct iphdr *)(p->data + off);
1491 /* The above works because, with the exception of the top
1492 * (inner most) layer, we only aggregate pkts with the same
1493 * hdr length so all the hdrs we'll need to verify will start
1494 * at the same offset.
1495 */
1496 if ((iph->protocol ^ iph2->protocol) |
1497 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1498 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1499 NAPI_GRO_CB(p)->same_flow = 0;
1500 continue;
1501 }
1502
1503 /* All fields must match except length and checksum. */
1504 NAPI_GRO_CB(p)->flush |=
1505 (iph->ttl ^ iph2->ttl) |
1506 (iph->tos ^ iph2->tos) |
1507 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1508
1509 NAPI_GRO_CB(p)->flush |= flush;
1510
1511 /* We need to store of the IP ID check to be included later
1512 * when we can verify that this packet does in fact belong
1513 * to a given flow.
1514 */
1515 flush_id = (u16)(id - ntohs(iph2->id));
1516
1517 /* This bit of code makes it much easier for us to identify
1518 * the cases where we are doing atomic vs non-atomic IP ID
1519 * checks. Specifically an atomic check can return IP ID
1520 * values 0 - 0xFFFF, while a non-atomic check can only
1521 * return 0 or 0xFFFF.
1522 */
1523 if (!NAPI_GRO_CB(p)->is_atomic ||
1524 !(iph->frag_off & htons(IP_DF))) {
1525 flush_id ^= NAPI_GRO_CB(p)->count;
1526 flush_id = flush_id ? 0xFFFF : 0;
1527 }
1528
1529 /* If the previous IP ID value was based on an atomic
1530 * datagram we can overwrite the value and ignore it.
1531 */
1532 if (NAPI_GRO_CB(skb)->is_atomic)
1533 NAPI_GRO_CB(p)->flush_id = flush_id;
1534 else
1535 NAPI_GRO_CB(p)->flush_id |= flush_id;
1536 }
1537
1538 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1539 NAPI_GRO_CB(skb)->flush |= flush;
1540 skb_set_network_header(skb, off);
1541 /* The above will be needed by the transport layer if there is one
1542 * immediately following this IP hdr.
1543 */
1544
1545 /* Note : No need to call skb_gro_postpull_rcsum() here,
1546 * as we already checked checksum over ipv4 header was 0
1547 */
1548 skb_gro_pull(skb, sizeof(*iph));
1549 skb_set_transport_header(skb, skb_gro_offset(skb));
1550
1551 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1552 ops->callbacks.gro_receive, head, skb);
1553
1554 out_unlock:
1555 rcu_read_unlock();
1556
1557 out:
1558 skb_gro_flush_final(skb, pp, flush);
1559
1560 return pp;
1561 }
1562 EXPORT_SYMBOL(inet_gro_receive);
1563
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1564 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1565 struct sk_buff *skb)
1566 {
1567 if (NAPI_GRO_CB(skb)->encap_mark) {
1568 NAPI_GRO_CB(skb)->flush = 1;
1569 return NULL;
1570 }
1571
1572 NAPI_GRO_CB(skb)->encap_mark = 1;
1573
1574 return inet_gro_receive(head, skb);
1575 }
1576
1577 #define SECONDS_PER_DAY 86400
1578
1579 /* inet_current_timestamp - Return IP network timestamp
1580 *
1581 * Return milliseconds since midnight in network byte order.
1582 */
inet_current_timestamp(void)1583 __be32 inet_current_timestamp(void)
1584 {
1585 u32 secs;
1586 u32 msecs;
1587 struct timespec64 ts;
1588
1589 ktime_get_real_ts64(&ts);
1590
1591 /* Get secs since midnight. */
1592 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1593 /* Convert to msecs. */
1594 msecs = secs * MSEC_PER_SEC;
1595 /* Convert nsec to msec. */
1596 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1597
1598 /* Convert to network byte order. */
1599 return htonl(msecs);
1600 }
1601 EXPORT_SYMBOL(inet_current_timestamp);
1602
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1603 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1604 {
1605 unsigned int family = READ_ONCE(sk->sk_family);
1606
1607 if (family == AF_INET)
1608 return ip_recv_error(sk, msg, len, addr_len);
1609 #if IS_ENABLED(CONFIG_IPV6)
1610 if (family == AF_INET6)
1611 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1612 #endif
1613 return -EINVAL;
1614 }
1615
inet_gro_complete(struct sk_buff * skb,int nhoff)1616 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1617 {
1618 __be16 newlen = htons(skb->len - nhoff);
1619 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1620 const struct net_offload *ops;
1621 int proto = iph->protocol;
1622 int err = -ENOSYS;
1623
1624 if (skb->encapsulation) {
1625 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1626 skb_set_inner_network_header(skb, nhoff);
1627 }
1628
1629 csum_replace2(&iph->check, iph->tot_len, newlen);
1630 iph->tot_len = newlen;
1631
1632 rcu_read_lock();
1633 ops = rcu_dereference(inet_offloads[proto]);
1634 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1635 goto out_unlock;
1636
1637 /* Only need to add sizeof(*iph) to get to the next hdr below
1638 * because any hdr with option will have been flushed in
1639 * inet_gro_receive().
1640 */
1641 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1642 tcp4_gro_complete, udp4_gro_complete,
1643 skb, nhoff + sizeof(*iph));
1644
1645 out_unlock:
1646 rcu_read_unlock();
1647
1648 return err;
1649 }
1650 EXPORT_SYMBOL(inet_gro_complete);
1651
ipip_gro_complete(struct sk_buff * skb,int nhoff)1652 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1653 {
1654 skb->encapsulation = 1;
1655 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1656 return inet_gro_complete(skb, nhoff);
1657 }
1658
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1659 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1660 unsigned short type, unsigned char protocol,
1661 struct net *net)
1662 {
1663 struct socket *sock;
1664 int rc = sock_create_kern(net, family, type, protocol, &sock);
1665
1666 if (rc == 0) {
1667 *sk = sock->sk;
1668 (*sk)->sk_allocation = GFP_ATOMIC;
1669 /*
1670 * Unhash it so that IP input processing does not even see it,
1671 * we do not wish this socket to see incoming packets.
1672 */
1673 (*sk)->sk_prot->unhash(*sk);
1674 }
1675 return rc;
1676 }
1677 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1678
snmp_get_cpu_field(void __percpu * mib,int cpu,int offt)1679 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1680 {
1681 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1682 }
1683 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1684
snmp_fold_field(void __percpu * mib,int offt)1685 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1686 {
1687 unsigned long res = 0;
1688 int i;
1689
1690 for_each_possible_cpu(i)
1691 res += snmp_get_cpu_field(mib, i, offt);
1692 return res;
1693 }
1694 EXPORT_SYMBOL_GPL(snmp_fold_field);
1695
1696 #if BITS_PER_LONG==32
1697
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1698 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1699 size_t syncp_offset)
1700 {
1701 void *bhptr;
1702 struct u64_stats_sync *syncp;
1703 u64 v;
1704 unsigned int start;
1705
1706 bhptr = per_cpu_ptr(mib, cpu);
1707 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1708 do {
1709 start = u64_stats_fetch_begin_irq(syncp);
1710 v = *(((u64 *)bhptr) + offt);
1711 } while (u64_stats_fetch_retry_irq(syncp, start));
1712
1713 return v;
1714 }
1715 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1716
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1717 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1718 {
1719 u64 res = 0;
1720 int cpu;
1721
1722 for_each_possible_cpu(cpu) {
1723 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1724 }
1725 return res;
1726 }
1727 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1728 #endif
1729
1730 #ifdef CONFIG_IP_MULTICAST
1731 static const struct net_protocol igmp_protocol = {
1732 .handler = igmp_rcv,
1733 .netns_ok = 1,
1734 };
1735 #endif
1736
1737 static const struct net_protocol tcp_protocol = {
1738 .handler = tcp_v4_rcv,
1739 .err_handler = tcp_v4_err,
1740 .no_policy = 1,
1741 .netns_ok = 1,
1742 .icmp_strict_tag_validation = 1,
1743 };
1744
1745 static const struct net_protocol udp_protocol = {
1746 .handler = udp_rcv,
1747 .err_handler = udp_err,
1748 .no_policy = 1,
1749 .netns_ok = 1,
1750 };
1751
1752 static const struct net_protocol icmp_protocol = {
1753 .handler = icmp_rcv,
1754 .err_handler = icmp_err,
1755 .no_policy = 1,
1756 .netns_ok = 1,
1757 };
1758
ipv4_mib_init_net(struct net * net)1759 static __net_init int ipv4_mib_init_net(struct net *net)
1760 {
1761 int i;
1762
1763 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1764 if (!net->mib.tcp_statistics)
1765 goto err_tcp_mib;
1766 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1767 if (!net->mib.ip_statistics)
1768 goto err_ip_mib;
1769
1770 for_each_possible_cpu(i) {
1771 struct ipstats_mib *af_inet_stats;
1772 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1773 u64_stats_init(&af_inet_stats->syncp);
1774 }
1775
1776 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1777 if (!net->mib.net_statistics)
1778 goto err_net_mib;
1779 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1780 if (!net->mib.udp_statistics)
1781 goto err_udp_mib;
1782 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1783 if (!net->mib.udplite_statistics)
1784 goto err_udplite_mib;
1785 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1786 if (!net->mib.icmp_statistics)
1787 goto err_icmp_mib;
1788 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1789 GFP_KERNEL);
1790 if (!net->mib.icmpmsg_statistics)
1791 goto err_icmpmsg_mib;
1792
1793 tcp_mib_init(net);
1794 return 0;
1795
1796 err_icmpmsg_mib:
1797 free_percpu(net->mib.icmp_statistics);
1798 err_icmp_mib:
1799 free_percpu(net->mib.udplite_statistics);
1800 err_udplite_mib:
1801 free_percpu(net->mib.udp_statistics);
1802 err_udp_mib:
1803 free_percpu(net->mib.net_statistics);
1804 err_net_mib:
1805 free_percpu(net->mib.ip_statistics);
1806 err_ip_mib:
1807 free_percpu(net->mib.tcp_statistics);
1808 err_tcp_mib:
1809 return -ENOMEM;
1810 }
1811
ipv4_mib_exit_net(struct net * net)1812 static __net_exit void ipv4_mib_exit_net(struct net *net)
1813 {
1814 kfree(net->mib.icmpmsg_statistics);
1815 free_percpu(net->mib.icmp_statistics);
1816 free_percpu(net->mib.udplite_statistics);
1817 free_percpu(net->mib.udp_statistics);
1818 free_percpu(net->mib.net_statistics);
1819 free_percpu(net->mib.ip_statistics);
1820 free_percpu(net->mib.tcp_statistics);
1821 #ifdef CONFIG_MPTCP
1822 /* allocated on demand, see mptcp_init_sock() */
1823 free_percpu(net->mib.mptcp_statistics);
1824 #endif
1825 }
1826
1827 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1828 .init = ipv4_mib_init_net,
1829 .exit = ipv4_mib_exit_net,
1830 };
1831
init_ipv4_mibs(void)1832 static int __init init_ipv4_mibs(void)
1833 {
1834 return register_pernet_subsys(&ipv4_mib_ops);
1835 }
1836
inet_init_net(struct net * net)1837 static __net_init int inet_init_net(struct net *net)
1838 {
1839 /*
1840 * Set defaults for local port range
1841 */
1842 seqlock_init(&net->ipv4.ip_local_ports.lock);
1843 net->ipv4.ip_local_ports.range[0] = 32768;
1844 net->ipv4.ip_local_ports.range[1] = 60999;
1845
1846 seqlock_init(&net->ipv4.ping_group_range.lock);
1847 /*
1848 * Sane defaults - nobody may create ping sockets.
1849 * Boot scripts should set this to distro-specific group.
1850 */
1851 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1852 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1853
1854 /* Default values for sysctl-controlled parameters.
1855 * We set them here, in case sysctl is not compiled.
1856 */
1857 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1858 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1859 net->ipv4.sysctl_ip_dynaddr = 0;
1860 net->ipv4.sysctl_ip_early_demux = 1;
1861 net->ipv4.sysctl_udp_early_demux = 1;
1862 net->ipv4.sysctl_tcp_early_demux = 1;
1863 net->ipv4.sysctl_nexthop_compat_mode = 1;
1864 #ifdef CONFIG_SYSCTL
1865 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1866 #endif
1867
1868 /* Some igmp sysctl, whose values are always used */
1869 net->ipv4.sysctl_igmp_max_memberships = 20;
1870 net->ipv4.sysctl_igmp_max_msf = 10;
1871 /* IGMP reports for link-local multicast groups are enabled by default */
1872 net->ipv4.sysctl_igmp_llm_reports = 1;
1873 net->ipv4.sysctl_igmp_qrv = 2;
1874
1875 return 0;
1876 }
1877
1878 static __net_initdata struct pernet_operations af_inet_ops = {
1879 .init = inet_init_net,
1880 };
1881
init_inet_pernet_ops(void)1882 static int __init init_inet_pernet_ops(void)
1883 {
1884 return register_pernet_subsys(&af_inet_ops);
1885 }
1886
1887 static int ipv4_proc_init(void);
1888
1889 /*
1890 * IP protocol layer initialiser
1891 */
1892
1893 static struct packet_offload ip_packet_offload __read_mostly = {
1894 .type = cpu_to_be16(ETH_P_IP),
1895 .callbacks = {
1896 .gso_segment = inet_gso_segment,
1897 .gro_receive = inet_gro_receive,
1898 .gro_complete = inet_gro_complete,
1899 },
1900 };
1901
1902 static const struct net_offload ipip_offload = {
1903 .callbacks = {
1904 .gso_segment = ipip_gso_segment,
1905 .gro_receive = ipip_gro_receive,
1906 .gro_complete = ipip_gro_complete,
1907 },
1908 };
1909
ipip_offload_init(void)1910 static int __init ipip_offload_init(void)
1911 {
1912 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1913 }
1914
ipv4_offload_init(void)1915 static int __init ipv4_offload_init(void)
1916 {
1917 /*
1918 * Add offloads
1919 */
1920 if (udpv4_offload_init() < 0)
1921 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1922 if (tcpv4_offload_init() < 0)
1923 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1924 if (ipip_offload_init() < 0)
1925 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1926
1927 dev_add_offload(&ip_packet_offload);
1928 return 0;
1929 }
1930
1931 fs_initcall(ipv4_offload_init);
1932
1933 static struct packet_type ip_packet_type __read_mostly = {
1934 .type = cpu_to_be16(ETH_P_IP),
1935 .func = ip_rcv,
1936 .list_func = ip_list_rcv,
1937 };
1938
inet_init(void)1939 static int __init inet_init(void)
1940 {
1941 struct inet_protosw *q;
1942 struct list_head *r;
1943 int rc;
1944
1945 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1946
1947 rc = proto_register(&tcp_prot, 1);
1948 if (rc)
1949 goto out;
1950
1951 rc = proto_register(&udp_prot, 1);
1952 if (rc)
1953 goto out_unregister_tcp_proto;
1954
1955 rc = proto_register(&raw_prot, 1);
1956 if (rc)
1957 goto out_unregister_udp_proto;
1958
1959 rc = proto_register(&ping_prot, 1);
1960 if (rc)
1961 goto out_unregister_raw_proto;
1962
1963 /*
1964 * Tell SOCKET that we are alive...
1965 */
1966
1967 (void)sock_register(&inet_family_ops);
1968
1969 #ifdef CONFIG_SYSCTL
1970 ip_static_sysctl_init();
1971 #endif
1972
1973 /*
1974 * Add all the base protocols.
1975 */
1976
1977 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1978 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1979 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1980 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1981 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1982 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1983 #ifdef CONFIG_IP_MULTICAST
1984 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1985 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1986 #endif
1987
1988 /* Register the socket-side information for inet_create. */
1989 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1990 INIT_LIST_HEAD(r);
1991
1992 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1993 inet_register_protosw(q);
1994
1995 /*
1996 * Set the ARP module up
1997 */
1998
1999 arp_init();
2000
2001 /*
2002 * Set the IP module up
2003 */
2004
2005 ip_init();
2006
2007 /* Initialise per-cpu ipv4 mibs */
2008 if (init_ipv4_mibs())
2009 panic("%s: Cannot init ipv4 mibs\n", __func__);
2010
2011 /* Setup TCP slab cache for open requests. */
2012 tcp_init();
2013
2014 /* Setup UDP memory threshold */
2015 udp_init();
2016
2017 /* Add UDP-Lite (RFC 3828) */
2018 udplite4_register();
2019
2020 raw_init();
2021
2022 ping_init();
2023
2024 /*
2025 * Set the ICMP layer up
2026 */
2027
2028 if (icmp_init() < 0)
2029 panic("Failed to create the ICMP control socket.\n");
2030
2031 /*
2032 * Initialise the multicast router
2033 */
2034 #if defined(CONFIG_IP_MROUTE)
2035 if (ip_mr_init())
2036 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2037 #endif
2038
2039 if (init_inet_pernet_ops())
2040 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2041
2042 ipv4_proc_init();
2043
2044 ipfrag_init();
2045
2046 dev_add_pack(&ip_packet_type);
2047
2048 ip_tunnel_core_init();
2049
2050 rc = 0;
2051 out:
2052 return rc;
2053 out_unregister_raw_proto:
2054 proto_unregister(&raw_prot);
2055 out_unregister_udp_proto:
2056 proto_unregister(&udp_prot);
2057 out_unregister_tcp_proto:
2058 proto_unregister(&tcp_prot);
2059 goto out;
2060 }
2061
2062 fs_initcall(inet_init);
2063
2064 /* ------------------------------------------------------------------------ */
2065
2066 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2067 static int __init ipv4_proc_init(void)
2068 {
2069 int rc = 0;
2070
2071 if (raw_proc_init())
2072 goto out_raw;
2073 if (tcp4_proc_init())
2074 goto out_tcp;
2075 if (udp4_proc_init())
2076 goto out_udp;
2077 if (ping_proc_init())
2078 goto out_ping;
2079 if (ip_misc_proc_init())
2080 goto out_misc;
2081 out:
2082 return rc;
2083 out_misc:
2084 ping_proc_exit();
2085 out_ping:
2086 udp4_proc_exit();
2087 out_udp:
2088 tcp4_proc_exit();
2089 out_tcp:
2090 raw_proc_exit();
2091 out_raw:
2092 rc = -ENOMEM;
2093 goto out;
2094 }
2095
2096 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2097 static int __init ipv4_proc_init(void)
2098 {
2099 return 0;
2100 }
2101 #endif /* CONFIG_PROC_FS */
2102