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 inet = inet_sk(sk);
331 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
332
333 inet->nodefrag = 0;
334
335 if (SOCK_RAW == sock->type) {
336 inet->inet_num = protocol;
337 if (IPPROTO_RAW == protocol)
338 inet->hdrincl = 1;
339 }
340
341 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
342 inet->pmtudisc = IP_PMTUDISC_DONT;
343 else
344 inet->pmtudisc = IP_PMTUDISC_WANT;
345
346 inet->inet_id = 0;
347
348 sock_init_data(sock, sk);
349
350 sk->sk_destruct = inet_sock_destruct;
351 sk->sk_protocol = protocol;
352 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
353
354 inet->uc_ttl = -1;
355 inet->mc_loop = 1;
356 inet->mc_ttl = 1;
357 inet->mc_all = 1;
358 inet->mc_index = 0;
359 inet->mc_list = NULL;
360 inet->rcv_tos = 0;
361
362 sk_refcnt_debug_inc(sk);
363
364 if (inet->inet_num) {
365 /* It assumes that any protocol which allows
366 * the user to assign a number at socket
367 * creation time automatically
368 * shares.
369 */
370 inet->inet_sport = htons(inet->inet_num);
371 /* Add to protocol hash chains. */
372 err = sk->sk_prot->hash(sk);
373 if (err) {
374 sk_common_release(sk);
375 goto out;
376 }
377 }
378
379 if (sk->sk_prot->init) {
380 err = sk->sk_prot->init(sk);
381 if (err) {
382 sk_common_release(sk);
383 goto out;
384 }
385 }
386
387 if (!kern) {
388 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
389 if (err) {
390 sk_common_release(sk);
391 goto out;
392 }
393 }
394 out:
395 return err;
396 out_rcu_unlock:
397 rcu_read_unlock();
398 goto out;
399 }
400
401
402 /*
403 * The peer socket should always be NULL (or else). When we call this
404 * function we are destroying the object and from then on nobody
405 * should refer to it.
406 */
inet_release(struct socket * sock)407 int inet_release(struct socket *sock)
408 {
409 struct sock *sk = sock->sk;
410
411 if (sk) {
412 long timeout;
413
414 if (!sk->sk_kern_sock)
415 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
416
417 /* Applications forget to leave groups before exiting */
418 ip_mc_drop_socket(sk);
419
420 /* If linger is set, we don't return until the close
421 * is complete. Otherwise we return immediately. The
422 * actually closing is done the same either way.
423 *
424 * If the close is due to the process exiting, we never
425 * linger..
426 */
427 timeout = 0;
428 if (sock_flag(sk, SOCK_LINGER) &&
429 !(current->flags & PF_EXITING))
430 timeout = sk->sk_lingertime;
431 sk->sk_prot->close(sk, timeout);
432 sock->sk = NULL;
433 }
434 return 0;
435 }
436 EXPORT_SYMBOL(inet_release);
437
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)438 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
439 {
440 struct sock *sk = sock->sk;
441 int err;
442
443 /* If the socket has its own bind function then use it. (RAW) */
444 if (sk->sk_prot->bind) {
445 return sk->sk_prot->bind(sk, uaddr, addr_len);
446 }
447 if (addr_len < sizeof(struct sockaddr_in))
448 return -EINVAL;
449
450 /* BPF prog is run before any checks are done so that if the prog
451 * changes context in a wrong way it will be caught.
452 */
453 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
454 if (err)
455 return err;
456
457 return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
458 }
459 EXPORT_SYMBOL(inet_bind);
460
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)461 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
462 u32 flags)
463 {
464 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
465 struct inet_sock *inet = inet_sk(sk);
466 struct net *net = sock_net(sk);
467 unsigned short snum;
468 int chk_addr_ret;
469 u32 tb_id = RT_TABLE_LOCAL;
470 int err;
471
472 if (addr->sin_family != AF_INET) {
473 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
474 * only if s_addr is INADDR_ANY.
475 */
476 err = -EAFNOSUPPORT;
477 if (addr->sin_family != AF_UNSPEC ||
478 addr->sin_addr.s_addr != htonl(INADDR_ANY))
479 goto out;
480 }
481
482 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
483 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
484
485 /* Not specified by any standard per-se, however it breaks too
486 * many applications when removed. It is unfortunate since
487 * allowing applications to make a non-local bind solves
488 * several problems with systems using dynamic addressing.
489 * (ie. your servers still start up even if your ISDN link
490 * is temporarily down)
491 */
492 err = -EADDRNOTAVAIL;
493 if (!inet_can_nonlocal_bind(net, inet) &&
494 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
495 chk_addr_ret != RTN_LOCAL &&
496 chk_addr_ret != RTN_MULTICAST &&
497 chk_addr_ret != RTN_BROADCAST)
498 goto out;
499
500 snum = ntohs(addr->sin_port);
501 err = -EPERM;
502 if (snum && inet_is_local_unbindable_port(net, snum))
503 goto out;
504
505 err = -EACCES;
506 if (snum && inet_port_requires_bind_service(net, snum) &&
507 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
508 goto out;
509
510 /* We keep a pair of addresses. rcv_saddr is the one
511 * used by hash lookups, and saddr is used for transmit.
512 *
513 * In the BSD API these are the same except where it
514 * would be illegal to use them (multicast/broadcast) in
515 * which case the sending device address is used.
516 */
517 if (flags & BIND_WITH_LOCK)
518 lock_sock(sk);
519
520 /* Check these errors (active socket, double bind). */
521 err = -EINVAL;
522 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
523 goto out_release_sock;
524
525 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
526 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
527 inet->inet_saddr = 0; /* Use device */
528
529 /* Make sure we are allowed to bind here. */
530 if (snum || !(inet->bind_address_no_port ||
531 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
532 if (sk->sk_prot->get_port(sk, snum)) {
533 inet->inet_saddr = inet->inet_rcv_saddr = 0;
534 err = -EADDRINUSE;
535 goto out_release_sock;
536 }
537 if (!(flags & BIND_FROM_BPF)) {
538 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
539 if (err) {
540 inet->inet_saddr = inet->inet_rcv_saddr = 0;
541 goto out_release_sock;
542 }
543 }
544 }
545
546 if (inet->inet_rcv_saddr)
547 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
548 if (snum)
549 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
550 inet->inet_sport = htons(inet->inet_num);
551 inet->inet_daddr = 0;
552 inet->inet_dport = 0;
553 sk_dst_reset(sk);
554 err = 0;
555 out_release_sock:
556 if (flags & BIND_WITH_LOCK)
557 release_sock(sk);
558 out:
559 return err;
560 }
561
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)562 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
563 int addr_len, int flags)
564 {
565 struct sock *sk = sock->sk;
566 int err;
567
568 if (addr_len < sizeof(uaddr->sa_family))
569 return -EINVAL;
570 if (uaddr->sa_family == AF_UNSPEC)
571 return sk->sk_prot->disconnect(sk, flags);
572
573 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
574 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
575 if (err)
576 return err;
577 }
578
579 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
580 return -EAGAIN;
581 return sk->sk_prot->connect(sk, uaddr, addr_len);
582 }
583 EXPORT_SYMBOL(inet_dgram_connect);
584
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)585 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
586 {
587 DEFINE_WAIT_FUNC(wait, woken_wake_function);
588
589 add_wait_queue(sk_sleep(sk), &wait);
590 sk->sk_write_pending += writebias;
591
592 /* Basic assumption: if someone sets sk->sk_err, he _must_
593 * change state of the socket from TCP_SYN_*.
594 * Connect() does not allow to get error notifications
595 * without closing the socket.
596 */
597 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
598 release_sock(sk);
599 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
600 lock_sock(sk);
601 if (signal_pending(current) || !timeo)
602 break;
603 }
604 remove_wait_queue(sk_sleep(sk), &wait);
605 sk->sk_write_pending -= writebias;
606 return timeo;
607 }
608
609 /*
610 * Connect to a remote host. There is regrettably still a little
611 * TCP 'magic' in here.
612 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)613 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
614 int addr_len, int flags, int is_sendmsg)
615 {
616 struct sock *sk = sock->sk;
617 int err;
618 long timeo;
619
620 /*
621 * uaddr can be NULL and addr_len can be 0 if:
622 * sk is a TCP fastopen active socket and
623 * TCP_FASTOPEN_CONNECT sockopt is set and
624 * we already have a valid cookie for this socket.
625 * In this case, user can call write() after connect().
626 * write() will invoke tcp_sendmsg_fastopen() which calls
627 * __inet_stream_connect().
628 */
629 if (uaddr) {
630 if (addr_len < sizeof(uaddr->sa_family))
631 return -EINVAL;
632
633 if (uaddr->sa_family == AF_UNSPEC) {
634 err = sk->sk_prot->disconnect(sk, flags);
635 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
636 goto out;
637 }
638 }
639
640 switch (sock->state) {
641 default:
642 err = -EINVAL;
643 goto out;
644 case SS_CONNECTED:
645 err = -EISCONN;
646 goto out;
647 case SS_CONNECTING:
648 if (inet_sk(sk)->defer_connect)
649 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
650 else
651 err = -EALREADY;
652 /* Fall out of switch with err, set for this state */
653 break;
654 case SS_UNCONNECTED:
655 err = -EISCONN;
656 if (sk->sk_state != TCP_CLOSE)
657 goto out;
658
659 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
660 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
661 if (err)
662 goto out;
663 }
664
665 err = sk->sk_prot->connect(sk, uaddr, addr_len);
666 if (err < 0)
667 goto out;
668
669 sock->state = SS_CONNECTING;
670
671 if (!err && inet_sk(sk)->defer_connect)
672 goto out;
673
674 /* Just entered SS_CONNECTING state; the only
675 * difference is that return value in non-blocking
676 * case is EINPROGRESS, rather than EALREADY.
677 */
678 err = -EINPROGRESS;
679 break;
680 }
681
682 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
683
684 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
685 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
686 tcp_sk(sk)->fastopen_req &&
687 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
688
689 /* Error code is set above */
690 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
691 goto out;
692
693 err = sock_intr_errno(timeo);
694 if (signal_pending(current))
695 goto out;
696 }
697
698 /* Connection was closed by RST, timeout, ICMP error
699 * or another process disconnected us.
700 */
701 if (sk->sk_state == TCP_CLOSE)
702 goto sock_error;
703
704 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
705 * and error was received after socket entered established state.
706 * Hence, it is handled normally after connect() return successfully.
707 */
708
709 sock->state = SS_CONNECTED;
710 err = 0;
711 out:
712 return err;
713
714 sock_error:
715 err = sock_error(sk) ? : -ECONNABORTED;
716 sock->state = SS_UNCONNECTED;
717 if (sk->sk_prot->disconnect(sk, flags))
718 sock->state = SS_DISCONNECTING;
719 goto out;
720 }
721 EXPORT_SYMBOL(__inet_stream_connect);
722
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)723 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
724 int addr_len, int flags)
725 {
726 int err;
727
728 lock_sock(sock->sk);
729 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
730 release_sock(sock->sk);
731 return err;
732 }
733 EXPORT_SYMBOL(inet_stream_connect);
734
735 /*
736 * Accept a pending connection. The TCP layer now gives BSD semantics.
737 */
738
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)739 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
740 bool kern)
741 {
742 struct sock *sk1 = sock->sk;
743 int err = -EINVAL;
744 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
745
746 if (!sk2)
747 goto do_err;
748
749 lock_sock(sk2);
750
751 sock_rps_record_flow(sk2);
752 WARN_ON(!((1 << sk2->sk_state) &
753 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
754 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
755
756 sock_graft(sk2, newsock);
757
758 newsock->state = SS_CONNECTED;
759 err = 0;
760 release_sock(sk2);
761 do_err:
762 return err;
763 }
764 EXPORT_SYMBOL(inet_accept);
765
766 /*
767 * This does both peername and sockname.
768 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)769 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
770 int peer)
771 {
772 struct sock *sk = sock->sk;
773 struct inet_sock *inet = inet_sk(sk);
774 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
775
776 sin->sin_family = AF_INET;
777 if (peer) {
778 if (!inet->inet_dport ||
779 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
780 peer == 1))
781 return -ENOTCONN;
782 sin->sin_port = inet->inet_dport;
783 sin->sin_addr.s_addr = inet->inet_daddr;
784 } else {
785 __be32 addr = inet->inet_rcv_saddr;
786 if (!addr)
787 addr = inet->inet_saddr;
788 sin->sin_port = inet->inet_sport;
789 sin->sin_addr.s_addr = addr;
790 }
791 if (cgroup_bpf_enabled)
792 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
793 peer ? BPF_CGROUP_INET4_GETPEERNAME :
794 BPF_CGROUP_INET4_GETSOCKNAME,
795 NULL);
796 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
797 return sizeof(*sin);
798 }
799 EXPORT_SYMBOL(inet_getname);
800
inet_send_prepare(struct sock * sk)801 int inet_send_prepare(struct sock *sk)
802 {
803 sock_rps_record_flow(sk);
804
805 /* We may need to bind the socket. */
806 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
807 inet_autobind(sk))
808 return -EAGAIN;
809
810 return 0;
811 }
812 EXPORT_SYMBOL_GPL(inet_send_prepare);
813
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)814 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
815 {
816 struct sock *sk = sock->sk;
817
818 if (unlikely(inet_send_prepare(sk)))
819 return -EAGAIN;
820
821 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
822 sk, msg, size);
823 }
824 EXPORT_SYMBOL(inet_sendmsg);
825
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)826 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
827 size_t size, int flags)
828 {
829 struct sock *sk = sock->sk;
830
831 if (unlikely(inet_send_prepare(sk)))
832 return -EAGAIN;
833
834 if (sk->sk_prot->sendpage)
835 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
836 return sock_no_sendpage(sock, page, offset, size, flags);
837 }
838 EXPORT_SYMBOL(inet_sendpage);
839
840 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
841 size_t, int, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)842 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
843 int flags)
844 {
845 struct sock *sk = sock->sk;
846 int addr_len = 0;
847 int err;
848
849 if (likely(!(flags & MSG_ERRQUEUE)))
850 sock_rps_record_flow(sk);
851
852 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
853 sk, msg, size, flags & MSG_DONTWAIT,
854 flags & ~MSG_DONTWAIT, &addr_len);
855 if (err >= 0)
856 msg->msg_namelen = addr_len;
857 return err;
858 }
859 EXPORT_SYMBOL(inet_recvmsg);
860
inet_shutdown(struct socket * sock,int how)861 int inet_shutdown(struct socket *sock, int how)
862 {
863 struct sock *sk = sock->sk;
864 int err = 0;
865
866 /* This should really check to make sure
867 * the socket is a TCP socket. (WHY AC...)
868 */
869 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
870 1->2 bit 2 snds.
871 2->3 */
872 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
873 return -EINVAL;
874
875 lock_sock(sk);
876 if (sock->state == SS_CONNECTING) {
877 if ((1 << sk->sk_state) &
878 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
879 sock->state = SS_DISCONNECTING;
880 else
881 sock->state = SS_CONNECTED;
882 }
883
884 switch (sk->sk_state) {
885 case TCP_CLOSE:
886 err = -ENOTCONN;
887 /* Hack to wake up other listeners, who can poll for
888 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
889 fallthrough;
890 default:
891 sk->sk_shutdown |= how;
892 if (sk->sk_prot->shutdown)
893 sk->sk_prot->shutdown(sk, how);
894 break;
895
896 /* Remaining two branches are temporary solution for missing
897 * close() in multithreaded environment. It is _not_ a good idea,
898 * but we have no choice until close() is repaired at VFS level.
899 */
900 case TCP_LISTEN:
901 if (!(how & RCV_SHUTDOWN))
902 break;
903 fallthrough;
904 case TCP_SYN_SENT:
905 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
906 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
907 break;
908 }
909
910 /* Wake up anyone sleeping in poll. */
911 sk->sk_state_change(sk);
912 release_sock(sk);
913 return err;
914 }
915 EXPORT_SYMBOL(inet_shutdown);
916
917 /*
918 * ioctl() calls you can issue on an INET socket. Most of these are
919 * device configuration and stuff and very rarely used. Some ioctls
920 * pass on to the socket itself.
921 *
922 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
923 * loads the devconfigure module does its configuring and unloads it.
924 * There's a good 20K of config code hanging around the kernel.
925 */
926
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)927 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
928 {
929 struct sock *sk = sock->sk;
930 int err = 0;
931 struct net *net = sock_net(sk);
932 void __user *p = (void __user *)arg;
933 struct ifreq ifr;
934 struct rtentry rt;
935
936 switch (cmd) {
937 case SIOCADDRT:
938 case SIOCDELRT:
939 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
940 return -EFAULT;
941 err = ip_rt_ioctl(net, cmd, &rt);
942 break;
943 case SIOCRTMSG:
944 err = -EINVAL;
945 break;
946 case SIOCDARP:
947 case SIOCGARP:
948 case SIOCSARP:
949 err = arp_ioctl(net, cmd, (void __user *)arg);
950 break;
951 case SIOCGIFADDR:
952 case SIOCGIFBRDADDR:
953 case SIOCGIFNETMASK:
954 case SIOCGIFDSTADDR:
955 case SIOCGIFPFLAGS:
956 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
957 return -EFAULT;
958 err = devinet_ioctl(net, cmd, &ifr);
959 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
960 err = -EFAULT;
961 break;
962
963 case SIOCSIFADDR:
964 case SIOCSIFBRDADDR:
965 case SIOCSIFNETMASK:
966 case SIOCSIFDSTADDR:
967 case SIOCSIFPFLAGS:
968 case SIOCSIFFLAGS:
969 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
970 return -EFAULT;
971 err = devinet_ioctl(net, cmd, &ifr);
972 break;
973 default:
974 if (sk->sk_prot->ioctl)
975 err = sk->sk_prot->ioctl(sk, cmd, arg);
976 else
977 err = -ENOIOCTLCMD;
978 break;
979 }
980 return err;
981 }
982 EXPORT_SYMBOL(inet_ioctl);
983
984 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)985 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
986 struct compat_rtentry __user *ur)
987 {
988 compat_uptr_t rtdev;
989 struct rtentry rt;
990
991 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
992 3 * sizeof(struct sockaddr)) ||
993 get_user(rt.rt_flags, &ur->rt_flags) ||
994 get_user(rt.rt_metric, &ur->rt_metric) ||
995 get_user(rt.rt_mtu, &ur->rt_mtu) ||
996 get_user(rt.rt_window, &ur->rt_window) ||
997 get_user(rt.rt_irtt, &ur->rt_irtt) ||
998 get_user(rtdev, &ur->rt_dev))
999 return -EFAULT;
1000
1001 rt.rt_dev = compat_ptr(rtdev);
1002 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1003 }
1004
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1005 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1006 {
1007 void __user *argp = compat_ptr(arg);
1008 struct sock *sk = sock->sk;
1009
1010 switch (cmd) {
1011 case SIOCADDRT:
1012 case SIOCDELRT:
1013 return inet_compat_routing_ioctl(sk, cmd, argp);
1014 default:
1015 if (!sk->sk_prot->compat_ioctl)
1016 return -ENOIOCTLCMD;
1017 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1018 }
1019 }
1020 #endif /* CONFIG_COMPAT */
1021
1022 const struct proto_ops inet_stream_ops = {
1023 .family = PF_INET,
1024 .owner = THIS_MODULE,
1025 .release = inet_release,
1026 .bind = inet_bind,
1027 .connect = inet_stream_connect,
1028 .socketpair = sock_no_socketpair,
1029 .accept = inet_accept,
1030 .getname = inet_getname,
1031 .poll = tcp_poll,
1032 .ioctl = inet_ioctl,
1033 .gettstamp = sock_gettstamp,
1034 .listen = inet_listen,
1035 .shutdown = inet_shutdown,
1036 .setsockopt = sock_common_setsockopt,
1037 .getsockopt = sock_common_getsockopt,
1038 .sendmsg = inet_sendmsg,
1039 .recvmsg = inet_recvmsg,
1040 #ifdef CONFIG_MMU
1041 .mmap = tcp_mmap,
1042 #endif
1043 .sendpage = inet_sendpage,
1044 .splice_read = tcp_splice_read,
1045 .read_sock = tcp_read_sock,
1046 .sendmsg_locked = tcp_sendmsg_locked,
1047 .sendpage_locked = tcp_sendpage_locked,
1048 .peek_len = tcp_peek_len,
1049 #ifdef CONFIG_COMPAT
1050 .compat_ioctl = inet_compat_ioctl,
1051 #endif
1052 .set_rcvlowat = tcp_set_rcvlowat,
1053 };
1054 EXPORT_SYMBOL(inet_stream_ops);
1055
1056 const struct proto_ops inet_dgram_ops = {
1057 .family = PF_INET,
1058 .owner = THIS_MODULE,
1059 .release = inet_release,
1060 .bind = inet_bind,
1061 .connect = inet_dgram_connect,
1062 .socketpair = sock_no_socketpair,
1063 .accept = sock_no_accept,
1064 .getname = inet_getname,
1065 .poll = udp_poll,
1066 .ioctl = inet_ioctl,
1067 .gettstamp = sock_gettstamp,
1068 .listen = sock_no_listen,
1069 .shutdown = inet_shutdown,
1070 .setsockopt = sock_common_setsockopt,
1071 .getsockopt = sock_common_getsockopt,
1072 .sendmsg = inet_sendmsg,
1073 .recvmsg = inet_recvmsg,
1074 .mmap = sock_no_mmap,
1075 .sendpage = inet_sendpage,
1076 .set_peek_off = sk_set_peek_off,
1077 #ifdef CONFIG_COMPAT
1078 .compat_ioctl = inet_compat_ioctl,
1079 #endif
1080 };
1081 EXPORT_SYMBOL(inet_dgram_ops);
1082
1083 /*
1084 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1085 * udp_poll
1086 */
1087 static const struct proto_ops inet_sockraw_ops = {
1088 .family = PF_INET,
1089 .owner = THIS_MODULE,
1090 .release = inet_release,
1091 .bind = inet_bind,
1092 .connect = inet_dgram_connect,
1093 .socketpair = sock_no_socketpair,
1094 .accept = sock_no_accept,
1095 .getname = inet_getname,
1096 .poll = datagram_poll,
1097 .ioctl = inet_ioctl,
1098 .gettstamp = sock_gettstamp,
1099 .listen = sock_no_listen,
1100 .shutdown = inet_shutdown,
1101 .setsockopt = sock_common_setsockopt,
1102 .getsockopt = sock_common_getsockopt,
1103 .sendmsg = inet_sendmsg,
1104 .recvmsg = inet_recvmsg,
1105 .mmap = sock_no_mmap,
1106 .sendpage = inet_sendpage,
1107 #ifdef CONFIG_COMPAT
1108 .compat_ioctl = inet_compat_ioctl,
1109 #endif
1110 };
1111
1112 static const struct net_proto_family inet_family_ops = {
1113 .family = PF_INET,
1114 .create = inet_create,
1115 .owner = THIS_MODULE,
1116 };
1117
1118 /* Upon startup we insert all the elements in inetsw_array[] into
1119 * the linked list inetsw.
1120 */
1121 static struct inet_protosw inetsw_array[] =
1122 {
1123 {
1124 .type = SOCK_STREAM,
1125 .protocol = IPPROTO_TCP,
1126 .prot = &tcp_prot,
1127 .ops = &inet_stream_ops,
1128 .flags = INET_PROTOSW_PERMANENT |
1129 INET_PROTOSW_ICSK,
1130 },
1131
1132 {
1133 .type = SOCK_DGRAM,
1134 .protocol = IPPROTO_UDP,
1135 .prot = &udp_prot,
1136 .ops = &inet_dgram_ops,
1137 .flags = INET_PROTOSW_PERMANENT,
1138 },
1139
1140 {
1141 .type = SOCK_DGRAM,
1142 .protocol = IPPROTO_ICMP,
1143 .prot = &ping_prot,
1144 .ops = &inet_sockraw_ops,
1145 .flags = INET_PROTOSW_REUSE,
1146 },
1147
1148 {
1149 .type = SOCK_RAW,
1150 .protocol = IPPROTO_IP, /* wild card */
1151 .prot = &raw_prot,
1152 .ops = &inet_sockraw_ops,
1153 .flags = INET_PROTOSW_REUSE,
1154 }
1155 };
1156
1157 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1158
inet_register_protosw(struct inet_protosw * p)1159 void inet_register_protosw(struct inet_protosw *p)
1160 {
1161 struct list_head *lh;
1162 struct inet_protosw *answer;
1163 int protocol = p->protocol;
1164 struct list_head *last_perm;
1165
1166 spin_lock_bh(&inetsw_lock);
1167
1168 if (p->type >= SOCK_MAX)
1169 goto out_illegal;
1170
1171 /* If we are trying to override a permanent protocol, bail. */
1172 last_perm = &inetsw[p->type];
1173 list_for_each(lh, &inetsw[p->type]) {
1174 answer = list_entry(lh, struct inet_protosw, list);
1175 /* Check only the non-wild match. */
1176 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1177 break;
1178 if (protocol == answer->protocol)
1179 goto out_permanent;
1180 last_perm = lh;
1181 }
1182
1183 /* Add the new entry after the last permanent entry if any, so that
1184 * the new entry does not override a permanent entry when matched with
1185 * a wild-card protocol. But it is allowed to override any existing
1186 * non-permanent entry. This means that when we remove this entry, the
1187 * system automatically returns to the old behavior.
1188 */
1189 list_add_rcu(&p->list, last_perm);
1190 out:
1191 spin_unlock_bh(&inetsw_lock);
1192
1193 return;
1194
1195 out_permanent:
1196 pr_err("Attempt to override permanent protocol %d\n", protocol);
1197 goto out;
1198
1199 out_illegal:
1200 pr_err("Ignoring attempt to register invalid socket type %d\n",
1201 p->type);
1202 goto out;
1203 }
1204 EXPORT_SYMBOL(inet_register_protosw);
1205
inet_unregister_protosw(struct inet_protosw * p)1206 void inet_unregister_protosw(struct inet_protosw *p)
1207 {
1208 if (INET_PROTOSW_PERMANENT & p->flags) {
1209 pr_err("Attempt to unregister permanent protocol %d\n",
1210 p->protocol);
1211 } else {
1212 spin_lock_bh(&inetsw_lock);
1213 list_del_rcu(&p->list);
1214 spin_unlock_bh(&inetsw_lock);
1215
1216 synchronize_net();
1217 }
1218 }
1219 EXPORT_SYMBOL(inet_unregister_protosw);
1220
inet_sk_reselect_saddr(struct sock * sk)1221 static int inet_sk_reselect_saddr(struct sock *sk)
1222 {
1223 struct inet_sock *inet = inet_sk(sk);
1224 __be32 old_saddr = inet->inet_saddr;
1225 __be32 daddr = inet->inet_daddr;
1226 struct flowi4 *fl4;
1227 struct rtable *rt;
1228 __be32 new_saddr;
1229 struct ip_options_rcu *inet_opt;
1230
1231 inet_opt = rcu_dereference_protected(inet->inet_opt,
1232 lockdep_sock_is_held(sk));
1233 if (inet_opt && inet_opt->opt.srr)
1234 daddr = inet_opt->opt.faddr;
1235
1236 /* Query new route. */
1237 fl4 = &inet->cork.fl.u.ip4;
1238 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1239 sk->sk_bound_dev_if, sk->sk_protocol,
1240 inet->inet_sport, inet->inet_dport, sk);
1241 if (IS_ERR(rt))
1242 return PTR_ERR(rt);
1243
1244 sk_setup_caps(sk, &rt->dst);
1245
1246 new_saddr = fl4->saddr;
1247
1248 if (new_saddr == old_saddr)
1249 return 0;
1250
1251 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1252 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1253 __func__, &old_saddr, &new_saddr);
1254 }
1255
1256 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1257
1258 /*
1259 * XXX The only one ugly spot where we need to
1260 * XXX really change the sockets identity after
1261 * XXX it has entered the hashes. -DaveM
1262 *
1263 * Besides that, it does not check for connection
1264 * uniqueness. Wait for troubles.
1265 */
1266 return __sk_prot_rehash(sk);
1267 }
1268
inet_sk_rebuild_header(struct sock * sk)1269 int inet_sk_rebuild_header(struct sock *sk)
1270 {
1271 struct inet_sock *inet = inet_sk(sk);
1272 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1273 __be32 daddr;
1274 struct ip_options_rcu *inet_opt;
1275 struct flowi4 *fl4;
1276 int err;
1277
1278 /* Route is OK, nothing to do. */
1279 if (rt)
1280 return 0;
1281
1282 /* Reroute. */
1283 rcu_read_lock();
1284 inet_opt = rcu_dereference(inet->inet_opt);
1285 daddr = inet->inet_daddr;
1286 if (inet_opt && inet_opt->opt.srr)
1287 daddr = inet_opt->opt.faddr;
1288 rcu_read_unlock();
1289 fl4 = &inet->cork.fl.u.ip4;
1290 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1291 inet->inet_dport, inet->inet_sport,
1292 sk->sk_protocol, RT_CONN_FLAGS(sk),
1293 sk->sk_bound_dev_if);
1294 if (!IS_ERR(rt)) {
1295 err = 0;
1296 sk_setup_caps(sk, &rt->dst);
1297 } else {
1298 err = PTR_ERR(rt);
1299
1300 /* Routing failed... */
1301 sk->sk_route_caps = 0;
1302 /*
1303 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1304 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1305 */
1306 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1307 sk->sk_state != TCP_SYN_SENT ||
1308 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1309 (err = inet_sk_reselect_saddr(sk)) != 0)
1310 sk->sk_err_soft = -err;
1311 }
1312
1313 return err;
1314 }
1315 EXPORT_SYMBOL(inet_sk_rebuild_header);
1316
inet_sk_set_state(struct sock * sk,int state)1317 void inet_sk_set_state(struct sock *sk, int state)
1318 {
1319 trace_inet_sock_set_state(sk, sk->sk_state, state);
1320 sk->sk_state = state;
1321 }
1322 EXPORT_SYMBOL(inet_sk_set_state);
1323
inet_sk_state_store(struct sock * sk,int newstate)1324 void inet_sk_state_store(struct sock *sk, int newstate)
1325 {
1326 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1327 smp_store_release(&sk->sk_state, newstate);
1328 }
1329
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1330 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1331 netdev_features_t features)
1332 {
1333 bool udpfrag = false, fixedid = false, gso_partial, encap;
1334 struct sk_buff *segs = ERR_PTR(-EINVAL);
1335 const struct net_offload *ops;
1336 unsigned int offset = 0;
1337 struct iphdr *iph;
1338 int proto, tot_len;
1339 int nhoff;
1340 int ihl;
1341 int id;
1342
1343 skb_reset_network_header(skb);
1344 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1345 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1346 goto out;
1347
1348 iph = ip_hdr(skb);
1349 ihl = iph->ihl * 4;
1350 if (ihl < sizeof(*iph))
1351 goto out;
1352
1353 id = ntohs(iph->id);
1354 proto = iph->protocol;
1355
1356 /* Warning: after this point, iph might be no longer valid */
1357 if (unlikely(!pskb_may_pull(skb, ihl)))
1358 goto out;
1359 __skb_pull(skb, ihl);
1360
1361 encap = SKB_GSO_CB(skb)->encap_level > 0;
1362 if (encap)
1363 features &= skb->dev->hw_enc_features;
1364 SKB_GSO_CB(skb)->encap_level += ihl;
1365
1366 skb_reset_transport_header(skb);
1367
1368 segs = ERR_PTR(-EPROTONOSUPPORT);
1369
1370 if (!skb->encapsulation || encap) {
1371 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1372 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1373
1374 /* fixed ID is invalid if DF bit is not set */
1375 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1376 goto out;
1377 }
1378
1379 ops = rcu_dereference(inet_offloads[proto]);
1380 if (likely(ops && ops->callbacks.gso_segment)) {
1381 segs = ops->callbacks.gso_segment(skb, features);
1382 if (!segs)
1383 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1384 }
1385
1386 if (IS_ERR_OR_NULL(segs))
1387 goto out;
1388
1389 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1390
1391 skb = segs;
1392 do {
1393 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1394 if (udpfrag) {
1395 iph->frag_off = htons(offset >> 3);
1396 if (skb->next)
1397 iph->frag_off |= htons(IP_MF);
1398 offset += skb->len - nhoff - ihl;
1399 tot_len = skb->len - nhoff;
1400 } else if (skb_is_gso(skb)) {
1401 if (!fixedid) {
1402 iph->id = htons(id);
1403 id += skb_shinfo(skb)->gso_segs;
1404 }
1405
1406 if (gso_partial)
1407 tot_len = skb_shinfo(skb)->gso_size +
1408 SKB_GSO_CB(skb)->data_offset +
1409 skb->head - (unsigned char *)iph;
1410 else
1411 tot_len = skb->len - nhoff;
1412 } else {
1413 if (!fixedid)
1414 iph->id = htons(id++);
1415 tot_len = skb->len - nhoff;
1416 }
1417 iph->tot_len = htons(tot_len);
1418 ip_send_check(iph);
1419 if (encap)
1420 skb_reset_inner_headers(skb);
1421 skb->network_header = (u8 *)iph - skb->head;
1422 skb_reset_mac_len(skb);
1423 } while ((skb = skb->next));
1424
1425 out:
1426 return segs;
1427 }
1428 EXPORT_SYMBOL(inet_gso_segment);
1429
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1430 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1431 netdev_features_t features)
1432 {
1433 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1434 return ERR_PTR(-EINVAL);
1435
1436 return inet_gso_segment(skb, features);
1437 }
1438
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1439 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1440 {
1441 const struct net_offload *ops;
1442 struct sk_buff *pp = NULL;
1443 const struct iphdr *iph;
1444 struct sk_buff *p;
1445 unsigned int hlen;
1446 unsigned int off;
1447 unsigned int id;
1448 int flush = 1;
1449 int proto;
1450
1451 off = skb_gro_offset(skb);
1452 hlen = off + sizeof(*iph);
1453 iph = skb_gro_header_fast(skb, off);
1454 if (skb_gro_header_hard(skb, hlen)) {
1455 iph = skb_gro_header_slow(skb, hlen, off);
1456 if (unlikely(!iph))
1457 goto out;
1458 }
1459
1460 proto = iph->protocol;
1461
1462 rcu_read_lock();
1463 ops = rcu_dereference(inet_offloads[proto]);
1464 if (!ops || !ops->callbacks.gro_receive)
1465 goto out_unlock;
1466
1467 if (*(u8 *)iph != 0x45)
1468 goto out_unlock;
1469
1470 if (ip_is_fragment(iph))
1471 goto out_unlock;
1472
1473 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1474 goto out_unlock;
1475
1476 id = ntohl(*(__be32 *)&iph->id);
1477 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1478 id >>= 16;
1479
1480 list_for_each_entry(p, head, list) {
1481 struct iphdr *iph2;
1482 u16 flush_id;
1483
1484 if (!NAPI_GRO_CB(p)->same_flow)
1485 continue;
1486
1487 iph2 = (struct iphdr *)(p->data + off);
1488 /* The above works because, with the exception of the top
1489 * (inner most) layer, we only aggregate pkts with the same
1490 * hdr length so all the hdrs we'll need to verify will start
1491 * at the same offset.
1492 */
1493 if ((iph->protocol ^ iph2->protocol) |
1494 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1495 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1496 NAPI_GRO_CB(p)->same_flow = 0;
1497 continue;
1498 }
1499
1500 /* All fields must match except length and checksum. */
1501 NAPI_GRO_CB(p)->flush |=
1502 (iph->ttl ^ iph2->ttl) |
1503 (iph->tos ^ iph2->tos) |
1504 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1505
1506 NAPI_GRO_CB(p)->flush |= flush;
1507
1508 /* We need to store of the IP ID check to be included later
1509 * when we can verify that this packet does in fact belong
1510 * to a given flow.
1511 */
1512 flush_id = (u16)(id - ntohs(iph2->id));
1513
1514 /* This bit of code makes it much easier for us to identify
1515 * the cases where we are doing atomic vs non-atomic IP ID
1516 * checks. Specifically an atomic check can return IP ID
1517 * values 0 - 0xFFFF, while a non-atomic check can only
1518 * return 0 or 0xFFFF.
1519 */
1520 if (!NAPI_GRO_CB(p)->is_atomic ||
1521 !(iph->frag_off & htons(IP_DF))) {
1522 flush_id ^= NAPI_GRO_CB(p)->count;
1523 flush_id = flush_id ? 0xFFFF : 0;
1524 }
1525
1526 /* If the previous IP ID value was based on an atomic
1527 * datagram we can overwrite the value and ignore it.
1528 */
1529 if (NAPI_GRO_CB(skb)->is_atomic)
1530 NAPI_GRO_CB(p)->flush_id = flush_id;
1531 else
1532 NAPI_GRO_CB(p)->flush_id |= flush_id;
1533 }
1534
1535 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1536 NAPI_GRO_CB(skb)->flush |= flush;
1537 skb_set_network_header(skb, off);
1538 /* The above will be needed by the transport layer if there is one
1539 * immediately following this IP hdr.
1540 */
1541
1542 /* Note : No need to call skb_gro_postpull_rcsum() here,
1543 * as we already checked checksum over ipv4 header was 0
1544 */
1545 skb_gro_pull(skb, sizeof(*iph));
1546 skb_set_transport_header(skb, skb_gro_offset(skb));
1547
1548 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1549 ops->callbacks.gro_receive, head, skb);
1550
1551 out_unlock:
1552 rcu_read_unlock();
1553
1554 out:
1555 skb_gro_flush_final(skb, pp, flush);
1556
1557 return pp;
1558 }
1559 EXPORT_SYMBOL(inet_gro_receive);
1560
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1561 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1562 struct sk_buff *skb)
1563 {
1564 if (NAPI_GRO_CB(skb)->encap_mark) {
1565 NAPI_GRO_CB(skb)->flush = 1;
1566 return NULL;
1567 }
1568
1569 NAPI_GRO_CB(skb)->encap_mark = 1;
1570
1571 return inet_gro_receive(head, skb);
1572 }
1573
1574 #define SECONDS_PER_DAY 86400
1575
1576 /* inet_current_timestamp - Return IP network timestamp
1577 *
1578 * Return milliseconds since midnight in network byte order.
1579 */
inet_current_timestamp(void)1580 __be32 inet_current_timestamp(void)
1581 {
1582 u32 secs;
1583 u32 msecs;
1584 struct timespec64 ts;
1585
1586 ktime_get_real_ts64(&ts);
1587
1588 /* Get secs since midnight. */
1589 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1590 /* Convert to msecs. */
1591 msecs = secs * MSEC_PER_SEC;
1592 /* Convert nsec to msec. */
1593 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1594
1595 /* Convert to network byte order. */
1596 return htonl(msecs);
1597 }
1598 EXPORT_SYMBOL(inet_current_timestamp);
1599
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1600 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1601 {
1602 if (sk->sk_family == AF_INET)
1603 return ip_recv_error(sk, msg, len, addr_len);
1604 #if IS_ENABLED(CONFIG_IPV6)
1605 if (sk->sk_family == AF_INET6)
1606 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1607 #endif
1608 return -EINVAL;
1609 }
1610
inet_gro_complete(struct sk_buff * skb,int nhoff)1611 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1612 {
1613 __be16 newlen = htons(skb->len - nhoff);
1614 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1615 const struct net_offload *ops;
1616 int proto = iph->protocol;
1617 int err = -ENOSYS;
1618
1619 if (skb->encapsulation) {
1620 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1621 skb_set_inner_network_header(skb, nhoff);
1622 }
1623
1624 csum_replace2(&iph->check, iph->tot_len, newlen);
1625 iph->tot_len = newlen;
1626
1627 rcu_read_lock();
1628 ops = rcu_dereference(inet_offloads[proto]);
1629 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1630 goto out_unlock;
1631
1632 /* Only need to add sizeof(*iph) to get to the next hdr below
1633 * because any hdr with option will have been flushed in
1634 * inet_gro_receive().
1635 */
1636 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1637 tcp4_gro_complete, udp4_gro_complete,
1638 skb, nhoff + sizeof(*iph));
1639
1640 out_unlock:
1641 rcu_read_unlock();
1642
1643 return err;
1644 }
1645 EXPORT_SYMBOL(inet_gro_complete);
1646
ipip_gro_complete(struct sk_buff * skb,int nhoff)1647 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1648 {
1649 skb->encapsulation = 1;
1650 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1651 return inet_gro_complete(skb, nhoff);
1652 }
1653
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1654 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1655 unsigned short type, unsigned char protocol,
1656 struct net *net)
1657 {
1658 struct socket *sock;
1659 int rc = sock_create_kern(net, family, type, protocol, &sock);
1660
1661 if (rc == 0) {
1662 *sk = sock->sk;
1663 (*sk)->sk_allocation = GFP_ATOMIC;
1664 /*
1665 * Unhash it so that IP input processing does not even see it,
1666 * we do not wish this socket to see incoming packets.
1667 */
1668 (*sk)->sk_prot->unhash(*sk);
1669 }
1670 return rc;
1671 }
1672 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1673
snmp_get_cpu_field(void __percpu * mib,int cpu,int offt)1674 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1675 {
1676 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1677 }
1678 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1679
snmp_fold_field(void __percpu * mib,int offt)1680 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1681 {
1682 unsigned long res = 0;
1683 int i;
1684
1685 for_each_possible_cpu(i)
1686 res += snmp_get_cpu_field(mib, i, offt);
1687 return res;
1688 }
1689 EXPORT_SYMBOL_GPL(snmp_fold_field);
1690
1691 #if BITS_PER_LONG==32
1692
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1693 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1694 size_t syncp_offset)
1695 {
1696 void *bhptr;
1697 struct u64_stats_sync *syncp;
1698 u64 v;
1699 unsigned int start;
1700
1701 bhptr = per_cpu_ptr(mib, cpu);
1702 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1703 do {
1704 start = u64_stats_fetch_begin_irq(syncp);
1705 v = *(((u64 *)bhptr) + offt);
1706 } while (u64_stats_fetch_retry_irq(syncp, start));
1707
1708 return v;
1709 }
1710 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1711
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1712 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1713 {
1714 u64 res = 0;
1715 int cpu;
1716
1717 for_each_possible_cpu(cpu) {
1718 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1719 }
1720 return res;
1721 }
1722 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1723 #endif
1724
1725 #ifdef CONFIG_IP_MULTICAST
1726 static const struct net_protocol igmp_protocol = {
1727 .handler = igmp_rcv,
1728 .netns_ok = 1,
1729 };
1730 #endif
1731
1732 static const struct net_protocol tcp_protocol = {
1733 .handler = tcp_v4_rcv,
1734 .err_handler = tcp_v4_err,
1735 .no_policy = 1,
1736 .netns_ok = 1,
1737 .icmp_strict_tag_validation = 1,
1738 };
1739
1740 static const struct net_protocol udp_protocol = {
1741 .handler = udp_rcv,
1742 .err_handler = udp_err,
1743 .no_policy = 1,
1744 .netns_ok = 1,
1745 };
1746
1747 static const struct net_protocol icmp_protocol = {
1748 .handler = icmp_rcv,
1749 .err_handler = icmp_err,
1750 .no_policy = 1,
1751 .netns_ok = 1,
1752 };
1753
ipv4_mib_init_net(struct net * net)1754 static __net_init int ipv4_mib_init_net(struct net *net)
1755 {
1756 int i;
1757
1758 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1759 if (!net->mib.tcp_statistics)
1760 goto err_tcp_mib;
1761 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1762 if (!net->mib.ip_statistics)
1763 goto err_ip_mib;
1764
1765 for_each_possible_cpu(i) {
1766 struct ipstats_mib *af_inet_stats;
1767 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1768 u64_stats_init(&af_inet_stats->syncp);
1769 }
1770
1771 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1772 if (!net->mib.net_statistics)
1773 goto err_net_mib;
1774 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1775 if (!net->mib.udp_statistics)
1776 goto err_udp_mib;
1777 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1778 if (!net->mib.udplite_statistics)
1779 goto err_udplite_mib;
1780 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1781 if (!net->mib.icmp_statistics)
1782 goto err_icmp_mib;
1783 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1784 GFP_KERNEL);
1785 if (!net->mib.icmpmsg_statistics)
1786 goto err_icmpmsg_mib;
1787
1788 tcp_mib_init(net);
1789 return 0;
1790
1791 err_icmpmsg_mib:
1792 free_percpu(net->mib.icmp_statistics);
1793 err_icmp_mib:
1794 free_percpu(net->mib.udplite_statistics);
1795 err_udplite_mib:
1796 free_percpu(net->mib.udp_statistics);
1797 err_udp_mib:
1798 free_percpu(net->mib.net_statistics);
1799 err_net_mib:
1800 free_percpu(net->mib.ip_statistics);
1801 err_ip_mib:
1802 free_percpu(net->mib.tcp_statistics);
1803 err_tcp_mib:
1804 return -ENOMEM;
1805 }
1806
ipv4_mib_exit_net(struct net * net)1807 static __net_exit void ipv4_mib_exit_net(struct net *net)
1808 {
1809 kfree(net->mib.icmpmsg_statistics);
1810 free_percpu(net->mib.icmp_statistics);
1811 free_percpu(net->mib.udplite_statistics);
1812 free_percpu(net->mib.udp_statistics);
1813 free_percpu(net->mib.net_statistics);
1814 free_percpu(net->mib.ip_statistics);
1815 free_percpu(net->mib.tcp_statistics);
1816 #ifdef CONFIG_MPTCP
1817 /* allocated on demand, see mptcp_init_sock() */
1818 free_percpu(net->mib.mptcp_statistics);
1819 #endif
1820 }
1821
1822 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1823 .init = ipv4_mib_init_net,
1824 .exit = ipv4_mib_exit_net,
1825 };
1826
init_ipv4_mibs(void)1827 static int __init init_ipv4_mibs(void)
1828 {
1829 return register_pernet_subsys(&ipv4_mib_ops);
1830 }
1831
inet_init_net(struct net * net)1832 static __net_init int inet_init_net(struct net *net)
1833 {
1834 /*
1835 * Set defaults for local port range
1836 */
1837 seqlock_init(&net->ipv4.ip_local_ports.lock);
1838 net->ipv4.ip_local_ports.range[0] = 32768;
1839 net->ipv4.ip_local_ports.range[1] = 60999;
1840
1841 seqlock_init(&net->ipv4.ping_group_range.lock);
1842 /*
1843 * Sane defaults - nobody may create ping sockets.
1844 * Boot scripts should set this to distro-specific group.
1845 */
1846 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1847 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1848
1849 /* Default values for sysctl-controlled parameters.
1850 * We set them here, in case sysctl is not compiled.
1851 */
1852 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1853 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1854 net->ipv4.sysctl_ip_dynaddr = 0;
1855 net->ipv4.sysctl_ip_early_demux = 1;
1856 net->ipv4.sysctl_udp_early_demux = 1;
1857 net->ipv4.sysctl_tcp_early_demux = 1;
1858 net->ipv4.sysctl_nexthop_compat_mode = 1;
1859 #ifdef CONFIG_SYSCTL
1860 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1861 #endif
1862
1863 /* Some igmp sysctl, whose values are always used */
1864 net->ipv4.sysctl_igmp_max_memberships = 20;
1865 net->ipv4.sysctl_igmp_max_msf = 10;
1866 /* IGMP reports for link-local multicast groups are enabled by default */
1867 net->ipv4.sysctl_igmp_llm_reports = 1;
1868 net->ipv4.sysctl_igmp_qrv = 2;
1869
1870 return 0;
1871 }
1872
1873 static __net_initdata struct pernet_operations af_inet_ops = {
1874 .init = inet_init_net,
1875 };
1876
init_inet_pernet_ops(void)1877 static int __init init_inet_pernet_ops(void)
1878 {
1879 return register_pernet_subsys(&af_inet_ops);
1880 }
1881
1882 static int ipv4_proc_init(void);
1883
1884 /*
1885 * IP protocol layer initialiser
1886 */
1887
1888 static struct packet_offload ip_packet_offload __read_mostly = {
1889 .type = cpu_to_be16(ETH_P_IP),
1890 .callbacks = {
1891 .gso_segment = inet_gso_segment,
1892 .gro_receive = inet_gro_receive,
1893 .gro_complete = inet_gro_complete,
1894 },
1895 };
1896
1897 static const struct net_offload ipip_offload = {
1898 .callbacks = {
1899 .gso_segment = ipip_gso_segment,
1900 .gro_receive = ipip_gro_receive,
1901 .gro_complete = ipip_gro_complete,
1902 },
1903 };
1904
ipip_offload_init(void)1905 static int __init ipip_offload_init(void)
1906 {
1907 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1908 }
1909
ipv4_offload_init(void)1910 static int __init ipv4_offload_init(void)
1911 {
1912 /*
1913 * Add offloads
1914 */
1915 if (udpv4_offload_init() < 0)
1916 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1917 if (tcpv4_offload_init() < 0)
1918 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1919 if (ipip_offload_init() < 0)
1920 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1921
1922 dev_add_offload(&ip_packet_offload);
1923 return 0;
1924 }
1925
1926 fs_initcall(ipv4_offload_init);
1927
1928 static struct packet_type ip_packet_type __read_mostly = {
1929 .type = cpu_to_be16(ETH_P_IP),
1930 .func = ip_rcv,
1931 .list_func = ip_list_rcv,
1932 };
1933
inet_init(void)1934 static int __init inet_init(void)
1935 {
1936 struct inet_protosw *q;
1937 struct list_head *r;
1938 int rc;
1939
1940 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1941
1942 rc = proto_register(&tcp_prot, 1);
1943 if (rc)
1944 goto out;
1945
1946 rc = proto_register(&udp_prot, 1);
1947 if (rc)
1948 goto out_unregister_tcp_proto;
1949
1950 rc = proto_register(&raw_prot, 1);
1951 if (rc)
1952 goto out_unregister_udp_proto;
1953
1954 rc = proto_register(&ping_prot, 1);
1955 if (rc)
1956 goto out_unregister_raw_proto;
1957
1958 /*
1959 * Tell SOCKET that we are alive...
1960 */
1961
1962 (void)sock_register(&inet_family_ops);
1963
1964 #ifdef CONFIG_SYSCTL
1965 ip_static_sysctl_init();
1966 #endif
1967
1968 /*
1969 * Add all the base protocols.
1970 */
1971
1972 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1973 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1974 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1975 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1976 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1977 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1978 #ifdef CONFIG_IP_MULTICAST
1979 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1980 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1981 #endif
1982
1983 /* Register the socket-side information for inet_create. */
1984 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1985 INIT_LIST_HEAD(r);
1986
1987 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1988 inet_register_protosw(q);
1989
1990 /*
1991 * Set the ARP module up
1992 */
1993
1994 arp_init();
1995
1996 /*
1997 * Set the IP module up
1998 */
1999
2000 ip_init();
2001
2002 /* Initialise per-cpu ipv4 mibs */
2003 if (init_ipv4_mibs())
2004 panic("%s: Cannot init ipv4 mibs\n", __func__);
2005
2006 /* Setup TCP slab cache for open requests. */
2007 tcp_init();
2008
2009 /* Setup UDP memory threshold */
2010 udp_init();
2011
2012 /* Add UDP-Lite (RFC 3828) */
2013 udplite4_register();
2014
2015 raw_init();
2016
2017 ping_init();
2018
2019 /*
2020 * Set the ICMP layer up
2021 */
2022
2023 if (icmp_init() < 0)
2024 panic("Failed to create the ICMP control socket.\n");
2025
2026 /*
2027 * Initialise the multicast router
2028 */
2029 #if defined(CONFIG_IP_MROUTE)
2030 if (ip_mr_init())
2031 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2032 #endif
2033
2034 if (init_inet_pernet_ops())
2035 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2036
2037 ipv4_proc_init();
2038
2039 ipfrag_init();
2040
2041 dev_add_pack(&ip_packet_type);
2042
2043 ip_tunnel_core_init();
2044
2045 rc = 0;
2046 out:
2047 return rc;
2048 out_unregister_raw_proto:
2049 proto_unregister(&raw_prot);
2050 out_unregister_udp_proto:
2051 proto_unregister(&udp_prot);
2052 out_unregister_tcp_proto:
2053 proto_unregister(&tcp_prot);
2054 goto out;
2055 }
2056
2057 fs_initcall(inet_init);
2058
2059 /* ------------------------------------------------------------------------ */
2060
2061 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2062 static int __init ipv4_proc_init(void)
2063 {
2064 int rc = 0;
2065
2066 if (raw_proc_init())
2067 goto out_raw;
2068 if (tcp4_proc_init())
2069 goto out_tcp;
2070 if (udp4_proc_init())
2071 goto out_udp;
2072 if (ping_proc_init())
2073 goto out_ping;
2074 if (ip_misc_proc_init())
2075 goto out_misc;
2076 out:
2077 return rc;
2078 out_misc:
2079 ping_proc_exit();
2080 out_ping:
2081 udp4_proc_exit();
2082 out_udp:
2083 tcp4_proc_exit();
2084 out_tcp:
2085 raw_proc_exit();
2086 out_raw:
2087 rc = -ENOMEM;
2088 goto out;
2089 }
2090
2091 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2092 static int __init ipv4_proc_init(void)
2093 {
2094 return 0;
2095 }
2096 #endif /* CONFIG_PROC_FS */
2097