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 #ifdef CONFIG_LOWPOWER_PROTOCOL
121 #include <net/lowpower_protocol.h>
122 #endif /* CONFIG_LOWPOWER_PROTOCOL */
123
124 #include <trace/events/sock.h>
125
126 /* The inetsw table contains everything that inet_create needs to
127 * build a new socket.
128 */
129 static struct list_head inetsw[SOCK_MAX];
130 static DEFINE_SPINLOCK(inetsw_lock);
131
132 /* New destruction routine */
133
inet_sock_destruct(struct sock * sk)134 void inet_sock_destruct(struct sock *sk)
135 {
136 struct inet_sock *inet = inet_sk(sk);
137
138 __skb_queue_purge(&sk->sk_receive_queue);
139 if (sk->sk_rx_skb_cache) {
140 __kfree_skb(sk->sk_rx_skb_cache);
141 sk->sk_rx_skb_cache = NULL;
142 }
143 __skb_queue_purge(&sk->sk_error_queue);
144
145 sk_mem_reclaim(sk);
146
147 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
148 pr_err("Attempt to release TCP socket in state %d %p\n",
149 sk->sk_state, sk);
150 return;
151 }
152 if (!sock_flag(sk, SOCK_DEAD)) {
153 pr_err("Attempt to release alive inet socket %p\n", sk);
154 return;
155 }
156
157 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
158 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
159 WARN_ON(sk->sk_wmem_queued);
160 WARN_ON(sk->sk_forward_alloc);
161
162 kfree(rcu_dereference_protected(inet->inet_opt, 1));
163 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
164 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
165 sk_refcnt_debug_dec(sk);
166 }
167 EXPORT_SYMBOL(inet_sock_destruct);
168
169 /*
170 * The routines beyond this point handle the behaviour of an AF_INET
171 * socket object. Mostly it punts to the subprotocols of IP to do
172 * the work.
173 */
174
175 /*
176 * Automatically bind an unbound socket.
177 */
178
inet_autobind(struct sock * sk)179 static int inet_autobind(struct sock *sk)
180 {
181 struct inet_sock *inet;
182 /* We may need to bind the socket. */
183 lock_sock(sk);
184 inet = inet_sk(sk);
185 if (!inet->inet_num) {
186 if (sk->sk_prot->get_port(sk, 0)) {
187 release_sock(sk);
188 return -EAGAIN;
189 }
190 inet->inet_sport = htons(inet->inet_num);
191 }
192 release_sock(sk);
193 return 0;
194 }
195
196 /*
197 * Move a socket into listening state.
198 */
inet_listen(struct socket * sock,int backlog)199 int inet_listen(struct socket *sock, int backlog)
200 {
201 struct sock *sk = sock->sk;
202 unsigned char old_state;
203 int err, tcp_fastopen;
204
205 lock_sock(sk);
206
207 err = -EINVAL;
208 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
209 goto out;
210
211 old_state = sk->sk_state;
212 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
213 goto out;
214
215 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
216 /* Really, if the socket is already in listen state
217 * we can only allow the backlog to be adjusted.
218 */
219 if (old_state != TCP_LISTEN) {
220 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
221 * Note that only TCP sockets (SOCK_STREAM) will reach here.
222 * Also fastopen backlog may already been set via the option
223 * because the socket was in TCP_LISTEN state previously but
224 * was shutdown() rather than close().
225 */
226 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
227 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
228 (tcp_fastopen & TFO_SERVER_ENABLE) &&
229 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
230 fastopen_queue_tune(sk, backlog);
231 tcp_fastopen_init_key_once(sock_net(sk));
232 }
233
234 err = inet_csk_listen_start(sk, backlog);
235 if (err)
236 goto out;
237 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
238 }
239 err = 0;
240
241 out:
242 release_sock(sk);
243 return err;
244 }
245 EXPORT_SYMBOL(inet_listen);
246
247 /*
248 * Create an inet socket.
249 */
250
inet_create(struct net * net,struct socket * sock,int protocol,int kern)251 static int inet_create(struct net *net, struct socket *sock, int protocol,
252 int kern)
253 {
254 struct sock *sk;
255 struct inet_protosw *answer;
256 struct inet_sock *inet;
257 struct proto *answer_prot;
258 unsigned char answer_flags;
259 int try_loading_module = 0;
260 int err;
261
262 if (protocol < 0 || protocol >= IPPROTO_MAX)
263 return -EINVAL;
264
265 sock->state = SS_UNCONNECTED;
266
267 /* Look for the requested type/protocol pair. */
268 lookup_protocol:
269 err = -ESOCKTNOSUPPORT;
270 rcu_read_lock();
271 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
272
273 err = 0;
274 /* Check the non-wild match. */
275 if (protocol == answer->protocol) {
276 if (protocol != IPPROTO_IP)
277 break;
278 } else {
279 /* Check for the two wild cases. */
280 if (IPPROTO_IP == protocol) {
281 protocol = answer->protocol;
282 break;
283 }
284 if (IPPROTO_IP == answer->protocol)
285 break;
286 }
287 err = -EPROTONOSUPPORT;
288 }
289
290 if (unlikely(err)) {
291 if (try_loading_module < 2) {
292 rcu_read_unlock();
293 /*
294 * Be more specific, e.g. net-pf-2-proto-132-type-1
295 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
296 */
297 if (++try_loading_module == 1)
298 request_module("net-pf-%d-proto-%d-type-%d",
299 PF_INET, protocol, sock->type);
300 /*
301 * Fall back to generic, e.g. net-pf-2-proto-132
302 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
303 */
304 else
305 request_module("net-pf-%d-proto-%d",
306 PF_INET, protocol);
307 goto lookup_protocol;
308 } else
309 goto out_rcu_unlock;
310 }
311
312 err = -EPERM;
313 if (sock->type == SOCK_RAW && !kern &&
314 !ns_capable(net->user_ns, CAP_NET_RAW))
315 goto out_rcu_unlock;
316
317 sock->ops = answer->ops;
318 answer_prot = answer->prot;
319 answer_flags = answer->flags;
320 rcu_read_unlock();
321
322 WARN_ON(!answer_prot->slab);
323
324 err = -ENOBUFS;
325 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
326 if (!sk)
327 goto out;
328
329 err = 0;
330 if (INET_PROTOSW_REUSE & answer_flags)
331 sk->sk_reuse = SK_CAN_REUSE;
332
333 if (INET_PROTOSW_ICSK & answer_flags)
334 inet_init_csk_locks(sk);
335
336 inet = inet_sk(sk);
337 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
338
339 inet->nodefrag = 0;
340
341 if (SOCK_RAW == sock->type) {
342 inet->inet_num = protocol;
343 if (IPPROTO_RAW == protocol)
344 inet->hdrincl = 1;
345 }
346
347 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
348 inet->pmtudisc = IP_PMTUDISC_DONT;
349 else
350 inet->pmtudisc = IP_PMTUDISC_WANT;
351
352 inet->inet_id = 0;
353
354 sock_init_data(sock, sk);
355
356 sk->sk_destruct = inet_sock_destruct;
357 sk->sk_protocol = protocol;
358 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
359
360 inet->uc_ttl = -1;
361 inet->mc_loop = 1;
362 inet->mc_ttl = 1;
363 inet->mc_all = 1;
364 inet->mc_index = 0;
365 inet->mc_list = NULL;
366 inet->rcv_tos = 0;
367
368 sk_refcnt_debug_inc(sk);
369
370 if (inet->inet_num) {
371 /* It assumes that any protocol which allows
372 * the user to assign a number at socket
373 * creation time automatically
374 * shares.
375 */
376 inet->inet_sport = htons(inet->inet_num);
377 /* Add to protocol hash chains. */
378 err = sk->sk_prot->hash(sk);
379 if (err) {
380 sk_common_release(sk);
381 goto out;
382 }
383 }
384
385 if (sk->sk_prot->init) {
386 err = sk->sk_prot->init(sk);
387 if (err) {
388 sk_common_release(sk);
389 goto out;
390 }
391 }
392
393 if (!kern) {
394 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
395 if (err) {
396 sk_common_release(sk);
397 goto out;
398 }
399 }
400 out:
401 return err;
402 out_rcu_unlock:
403 rcu_read_unlock();
404 goto out;
405 }
406
407
408 /*
409 * The peer socket should always be NULL (or else). When we call this
410 * function we are destroying the object and from then on nobody
411 * should refer to it.
412 */
inet_release(struct socket * sock)413 int inet_release(struct socket *sock)
414 {
415 struct sock *sk = sock->sk;
416
417 if (sk) {
418 long timeout;
419
420 if (!sk->sk_kern_sock)
421 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
422
423 /* Applications forget to leave groups before exiting */
424 ip_mc_drop_socket(sk);
425
426 /* If linger is set, we don't return until the close
427 * is complete. Otherwise we return immediately. The
428 * actually closing is done the same either way.
429 *
430 * If the close is due to the process exiting, we never
431 * linger..
432 */
433 timeout = 0;
434 if (sock_flag(sk, SOCK_LINGER) &&
435 !(current->flags & PF_EXITING))
436 timeout = sk->sk_lingertime;
437 sk->sk_prot->close(sk, timeout);
438 sock->sk = NULL;
439 }
440 return 0;
441 }
442 EXPORT_SYMBOL(inet_release);
443
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)444 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
445 {
446 struct sock *sk = sock->sk;
447 int err;
448
449 /* If the socket has its own bind function then use it. (RAW) */
450 if (sk->sk_prot->bind) {
451 return sk->sk_prot->bind(sk, uaddr, addr_len);
452 }
453 if (addr_len < sizeof(struct sockaddr_in))
454 return -EINVAL;
455
456 /* BPF prog is run before any checks are done so that if the prog
457 * changes context in a wrong way it will be caught.
458 */
459 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
460 if (err)
461 return err;
462
463 return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
464 }
465 EXPORT_SYMBOL(inet_bind);
466
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)467 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
468 u32 flags)
469 {
470 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
471 struct inet_sock *inet = inet_sk(sk);
472 struct net *net = sock_net(sk);
473 unsigned short snum;
474 int chk_addr_ret;
475 u32 tb_id = RT_TABLE_LOCAL;
476 int err;
477
478 if (addr->sin_family != AF_INET) {
479 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
480 * only if s_addr is INADDR_ANY.
481 */
482 err = -EAFNOSUPPORT;
483 if (addr->sin_family != AF_UNSPEC ||
484 addr->sin_addr.s_addr != htonl(INADDR_ANY))
485 goto out;
486 }
487
488 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
489 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
490
491 /* Not specified by any standard per-se, however it breaks too
492 * many applications when removed. It is unfortunate since
493 * allowing applications to make a non-local bind solves
494 * several problems with systems using dynamic addressing.
495 * (ie. your servers still start up even if your ISDN link
496 * is temporarily down)
497 */
498 err = -EADDRNOTAVAIL;
499 if (!inet_can_nonlocal_bind(net, inet) &&
500 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
501 chk_addr_ret != RTN_LOCAL &&
502 chk_addr_ret != RTN_MULTICAST &&
503 chk_addr_ret != RTN_BROADCAST)
504 goto out;
505
506 snum = ntohs(addr->sin_port);
507 err = -EACCES;
508 if (snum && inet_port_requires_bind_service(net, snum) &&
509 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
510 goto out;
511
512 /* We keep a pair of addresses. rcv_saddr is the one
513 * used by hash lookups, and saddr is used for transmit.
514 *
515 * In the BSD API these are the same except where it
516 * would be illegal to use them (multicast/broadcast) in
517 * which case the sending device address is used.
518 */
519 if (flags & BIND_WITH_LOCK)
520 lock_sock(sk);
521
522 /* Check these errors (active socket, double bind). */
523 err = -EINVAL;
524 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
525 goto out_release_sock;
526
527 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
528 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
529 inet->inet_saddr = 0; /* Use device */
530
531 /* Make sure we are allowed to bind here. */
532 if (snum || !(inet->bind_address_no_port ||
533 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
534 if (sk->sk_prot->get_port(sk, snum)) {
535 inet->inet_saddr = inet->inet_rcv_saddr = 0;
536 err = -EADDRINUSE;
537 goto out_release_sock;
538 }
539 if (!(flags & BIND_FROM_BPF)) {
540 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
541 if (err) {
542 inet->inet_saddr = inet->inet_rcv_saddr = 0;
543 goto out_release_sock;
544 }
545 }
546 }
547
548 if (inet->inet_rcv_saddr)
549 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
550 if (snum)
551 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
552 inet->inet_sport = htons(inet->inet_num);
553 inet->inet_daddr = 0;
554 inet->inet_dport = 0;
555 sk_dst_reset(sk);
556 err = 0;
557 out_release_sock:
558 if (flags & BIND_WITH_LOCK)
559 release_sock(sk);
560 out:
561 return err;
562 }
563
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)564 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
565 int addr_len, int flags)
566 {
567 struct sock *sk = sock->sk;
568 const struct proto *prot;
569 int err;
570
571 if (addr_len < sizeof(uaddr->sa_family))
572 return -EINVAL;
573
574 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
575 prot = READ_ONCE(sk->sk_prot);
576
577 if (uaddr->sa_family == AF_UNSPEC)
578 return prot->disconnect(sk, flags);
579
580 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
581 err = prot->pre_connect(sk, uaddr, addr_len);
582 if (err)
583 return err;
584 }
585
586 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
587 return -EAGAIN;
588 return prot->connect(sk, uaddr, addr_len);
589 }
590 EXPORT_SYMBOL(inet_dgram_connect);
591
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)592 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
593 {
594 DEFINE_WAIT_FUNC(wait, woken_wake_function);
595
596 add_wait_queue(sk_sleep(sk), &wait);
597 sk->sk_write_pending += writebias;
598 sk->sk_wait_pending++;
599
600 /* Basic assumption: if someone sets sk->sk_err, he _must_
601 * change state of the socket from TCP_SYN_*.
602 * Connect() does not allow to get error notifications
603 * without closing the socket.
604 */
605 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
606 release_sock(sk);
607 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
608 lock_sock(sk);
609 if (signal_pending(current) || !timeo)
610 break;
611 }
612 remove_wait_queue(sk_sleep(sk), &wait);
613 sk->sk_write_pending -= writebias;
614 sk->sk_wait_pending--;
615 return timeo;
616 }
617
618 /*
619 * Connect to a remote host. There is regrettably still a little
620 * TCP 'magic' in here.
621 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)622 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
623 int addr_len, int flags, int is_sendmsg)
624 {
625 struct sock *sk = sock->sk;
626 int err;
627 long timeo;
628
629 /*
630 * uaddr can be NULL and addr_len can be 0 if:
631 * sk is a TCP fastopen active socket and
632 * TCP_FASTOPEN_CONNECT sockopt is set and
633 * we already have a valid cookie for this socket.
634 * In this case, user can call write() after connect().
635 * write() will invoke tcp_sendmsg_fastopen() which calls
636 * __inet_stream_connect().
637 */
638 if (uaddr) {
639 if (addr_len < sizeof(uaddr->sa_family))
640 return -EINVAL;
641
642 if (uaddr->sa_family == AF_UNSPEC) {
643 err = sk->sk_prot->disconnect(sk, flags);
644 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
645 goto out;
646 }
647 }
648
649 switch (sock->state) {
650 default:
651 err = -EINVAL;
652 goto out;
653 case SS_CONNECTED:
654 err = -EISCONN;
655 goto out;
656 case SS_CONNECTING:
657 if (inet_sk(sk)->defer_connect)
658 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
659 else
660 err = -EALREADY;
661 /* Fall out of switch with err, set for this state */
662 break;
663 case SS_UNCONNECTED:
664 err = -EISCONN;
665 if (sk->sk_state != TCP_CLOSE)
666 goto out;
667
668 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
669 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
670 if (err)
671 goto out;
672 }
673
674 err = sk->sk_prot->connect(sk, uaddr, addr_len);
675 if (err < 0)
676 goto out;
677
678 sock->state = SS_CONNECTING;
679
680 if (!err && inet_sk(sk)->defer_connect)
681 goto out;
682
683 /* Just entered SS_CONNECTING state; the only
684 * difference is that return value in non-blocking
685 * case is EINPROGRESS, rather than EALREADY.
686 */
687 err = -EINPROGRESS;
688 break;
689 }
690
691 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
692
693 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
694 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
695 tcp_sk(sk)->fastopen_req &&
696 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
697
698 /* Error code is set above */
699 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
700 goto out;
701
702 err = sock_intr_errno(timeo);
703 if (signal_pending(current))
704 goto out;
705 }
706
707 /* Connection was closed by RST, timeout, ICMP error
708 * or another process disconnected us.
709 */
710 if (sk->sk_state == TCP_CLOSE)
711 goto sock_error;
712
713 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
714 * and error was received after socket entered established state.
715 * Hence, it is handled normally after connect() return successfully.
716 */
717
718 sock->state = SS_CONNECTED;
719 err = 0;
720 out:
721 return err;
722
723 sock_error:
724 err = sock_error(sk) ? : -ECONNABORTED;
725 sock->state = SS_UNCONNECTED;
726 if (sk->sk_prot->disconnect(sk, flags))
727 sock->state = SS_DISCONNECTING;
728 goto out;
729 }
730 EXPORT_SYMBOL(__inet_stream_connect);
731
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)732 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
733 int addr_len, int flags)
734 {
735 int err;
736
737 lock_sock(sock->sk);
738 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
739 release_sock(sock->sk);
740 return err;
741 }
742 EXPORT_SYMBOL(inet_stream_connect);
743
744 /*
745 * Accept a pending connection. The TCP layer now gives BSD semantics.
746 */
747
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)748 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
749 bool kern)
750 {
751 struct sock *sk1 = sock->sk, *sk2;
752 int err = -EINVAL;
753
754 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
755 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
756 if (!sk2)
757 goto do_err;
758
759 lock_sock(sk2);
760
761 sock_rps_record_flow(sk2);
762 WARN_ON(!((1 << sk2->sk_state) &
763 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
764 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
765
766 sock_graft(sk2, newsock);
767
768 newsock->state = SS_CONNECTED;
769 err = 0;
770 release_sock(sk2);
771 do_err:
772 return err;
773 }
774 EXPORT_SYMBOL(inet_accept);
775
776 /*
777 * This does both peername and sockname.
778 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)779 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
780 int peer)
781 {
782 struct sock *sk = sock->sk;
783 struct inet_sock *inet = inet_sk(sk);
784 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
785
786 sin->sin_family = AF_INET;
787 if (peer) {
788 if (!inet->inet_dport ||
789 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
790 peer == 1))
791 return -ENOTCONN;
792 sin->sin_port = inet->inet_dport;
793 sin->sin_addr.s_addr = inet->inet_daddr;
794 } else {
795 __be32 addr = inet->inet_rcv_saddr;
796 if (!addr)
797 addr = inet->inet_saddr;
798 sin->sin_port = inet->inet_sport;
799 sin->sin_addr.s_addr = addr;
800 }
801 if (cgroup_bpf_enabled)
802 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
803 peer ? BPF_CGROUP_INET4_GETPEERNAME :
804 BPF_CGROUP_INET4_GETSOCKNAME,
805 NULL);
806 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
807 return sizeof(*sin);
808 }
809 EXPORT_SYMBOL(inet_getname);
810
inet_send_prepare(struct sock * sk)811 int inet_send_prepare(struct sock *sk)
812 {
813 sock_rps_record_flow(sk);
814
815 /* We may need to bind the socket. */
816 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
817 inet_autobind(sk))
818 return -EAGAIN;
819
820 return 0;
821 }
822 EXPORT_SYMBOL_GPL(inet_send_prepare);
823
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)824 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
825 {
826 struct sock *sk = sock->sk;
827
828 if (unlikely(inet_send_prepare(sk)))
829 return -EAGAIN;
830
831 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
832 sk, msg, size);
833 }
834 EXPORT_SYMBOL(inet_sendmsg);
835
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)836 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
837 size_t size, int flags)
838 {
839 struct sock *sk = sock->sk;
840 const struct proto *prot;
841
842 if (unlikely(inet_send_prepare(sk)))
843 return -EAGAIN;
844
845 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
846 prot = READ_ONCE(sk->sk_prot);
847 if (prot->sendpage)
848 return prot->sendpage(sk, page, offset, size, flags);
849 return sock_no_sendpage(sock, page, offset, size, flags);
850 }
851 EXPORT_SYMBOL(inet_sendpage);
852
853 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
854 size_t, int, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)855 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
856 int flags)
857 {
858 struct sock *sk = sock->sk;
859 int addr_len = 0;
860 int err;
861
862 if (likely(!(flags & MSG_ERRQUEUE)))
863 sock_rps_record_flow(sk);
864
865 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
866 sk, msg, size, flags & MSG_DONTWAIT,
867 flags & ~MSG_DONTWAIT, &addr_len);
868 if (err >= 0)
869 msg->msg_namelen = addr_len;
870 return err;
871 }
872 EXPORT_SYMBOL(inet_recvmsg);
873
inet_shutdown(struct socket * sock,int how)874 int inet_shutdown(struct socket *sock, int how)
875 {
876 struct sock *sk = sock->sk;
877 int err = 0;
878
879 /* This should really check to make sure
880 * the socket is a TCP socket. (WHY AC...)
881 */
882 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
883 1->2 bit 2 snds.
884 2->3 */
885 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
886 return -EINVAL;
887
888 lock_sock(sk);
889 if (sock->state == SS_CONNECTING) {
890 if ((1 << sk->sk_state) &
891 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
892 sock->state = SS_DISCONNECTING;
893 else
894 sock->state = SS_CONNECTED;
895 }
896
897 switch (sk->sk_state) {
898 case TCP_CLOSE:
899 err = -ENOTCONN;
900 /* Hack to wake up other listeners, who can poll for
901 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
902 fallthrough;
903 default:
904 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
905 if (sk->sk_prot->shutdown)
906 sk->sk_prot->shutdown(sk, how);
907 break;
908
909 /* Remaining two branches are temporary solution for missing
910 * close() in multithreaded environment. It is _not_ a good idea,
911 * but we have no choice until close() is repaired at VFS level.
912 */
913 case TCP_LISTEN:
914 if (!(how & RCV_SHUTDOWN))
915 break;
916 fallthrough;
917 case TCP_SYN_SENT:
918 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
919 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
920 break;
921 }
922
923 /* Wake up anyone sleeping in poll. */
924 sk->sk_state_change(sk);
925 release_sock(sk);
926 return err;
927 }
928 EXPORT_SYMBOL(inet_shutdown);
929
930 /*
931 * ioctl() calls you can issue on an INET socket. Most of these are
932 * device configuration and stuff and very rarely used. Some ioctls
933 * pass on to the socket itself.
934 *
935 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
936 * loads the devconfigure module does its configuring and unloads it.
937 * There's a good 20K of config code hanging around the kernel.
938 */
939
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)940 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
941 {
942 struct sock *sk = sock->sk;
943 int err = 0;
944 struct net *net = sock_net(sk);
945 void __user *p = (void __user *)arg;
946 struct ifreq ifr;
947 struct rtentry rt;
948
949 switch (cmd) {
950 case SIOCADDRT:
951 case SIOCDELRT:
952 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
953 return -EFAULT;
954 err = ip_rt_ioctl(net, cmd, &rt);
955 break;
956 case SIOCRTMSG:
957 err = -EINVAL;
958 break;
959 case SIOCDARP:
960 case SIOCGARP:
961 case SIOCSARP:
962 err = arp_ioctl(net, cmd, (void __user *)arg);
963 break;
964 case SIOCGIFADDR:
965 case SIOCGIFBRDADDR:
966 case SIOCGIFNETMASK:
967 case SIOCGIFDSTADDR:
968 case SIOCGIFPFLAGS:
969 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
970 return -EFAULT;
971 err = devinet_ioctl(net, cmd, &ifr);
972 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
973 err = -EFAULT;
974 break;
975
976 case SIOCSIFADDR:
977 case SIOCSIFBRDADDR:
978 case SIOCSIFNETMASK:
979 case SIOCSIFDSTADDR:
980 case SIOCSIFPFLAGS:
981 case SIOCSIFFLAGS:
982 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
983 return -EFAULT;
984 err = devinet_ioctl(net, cmd, &ifr);
985 break;
986 default:
987 if (sk->sk_prot->ioctl)
988 err = sk->sk_prot->ioctl(sk, cmd, arg);
989 else
990 err = -ENOIOCTLCMD;
991 break;
992 }
993 return err;
994 }
995 EXPORT_SYMBOL(inet_ioctl);
996
997 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)998 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
999 struct compat_rtentry __user *ur)
1000 {
1001 compat_uptr_t rtdev;
1002 struct rtentry rt;
1003
1004 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1005 3 * sizeof(struct sockaddr)) ||
1006 get_user(rt.rt_flags, &ur->rt_flags) ||
1007 get_user(rt.rt_metric, &ur->rt_metric) ||
1008 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1009 get_user(rt.rt_window, &ur->rt_window) ||
1010 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1011 get_user(rtdev, &ur->rt_dev))
1012 return -EFAULT;
1013
1014 rt.rt_dev = compat_ptr(rtdev);
1015 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1016 }
1017
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1018 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1019 {
1020 void __user *argp = compat_ptr(arg);
1021 struct sock *sk = sock->sk;
1022
1023 switch (cmd) {
1024 case SIOCADDRT:
1025 case SIOCDELRT:
1026 return inet_compat_routing_ioctl(sk, cmd, argp);
1027 default:
1028 if (!sk->sk_prot->compat_ioctl)
1029 return -ENOIOCTLCMD;
1030 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1031 }
1032 }
1033 #endif /* CONFIG_COMPAT */
1034
1035 const struct proto_ops inet_stream_ops = {
1036 .family = PF_INET,
1037 .owner = THIS_MODULE,
1038 .release = inet_release,
1039 .bind = inet_bind,
1040 .connect = inet_stream_connect,
1041 .socketpair = sock_no_socketpair,
1042 .accept = inet_accept,
1043 .getname = inet_getname,
1044 .poll = tcp_poll,
1045 .ioctl = inet_ioctl,
1046 .gettstamp = sock_gettstamp,
1047 .listen = inet_listen,
1048 .shutdown = inet_shutdown,
1049 .setsockopt = sock_common_setsockopt,
1050 .getsockopt = sock_common_getsockopt,
1051 .sendmsg = inet_sendmsg,
1052 .recvmsg = inet_recvmsg,
1053 #ifdef CONFIG_MMU
1054 .mmap = tcp_mmap,
1055 #endif
1056 .sendpage = inet_sendpage,
1057 .splice_read = tcp_splice_read,
1058 .read_sock = tcp_read_sock,
1059 .sendmsg_locked = tcp_sendmsg_locked,
1060 .sendpage_locked = tcp_sendpage_locked,
1061 .peek_len = tcp_peek_len,
1062 #ifdef CONFIG_COMPAT
1063 .compat_ioctl = inet_compat_ioctl,
1064 #endif
1065 .set_rcvlowat = tcp_set_rcvlowat,
1066 };
1067 EXPORT_SYMBOL(inet_stream_ops);
1068
1069 const struct proto_ops inet_dgram_ops = {
1070 .family = PF_INET,
1071 .owner = THIS_MODULE,
1072 .release = inet_release,
1073 .bind = inet_bind,
1074 .connect = inet_dgram_connect,
1075 .socketpair = sock_no_socketpair,
1076 .accept = sock_no_accept,
1077 .getname = inet_getname,
1078 .poll = udp_poll,
1079 .ioctl = inet_ioctl,
1080 .gettstamp = sock_gettstamp,
1081 .listen = sock_no_listen,
1082 .shutdown = inet_shutdown,
1083 .setsockopt = sock_common_setsockopt,
1084 .getsockopt = sock_common_getsockopt,
1085 .sendmsg = inet_sendmsg,
1086 .recvmsg = inet_recvmsg,
1087 .mmap = sock_no_mmap,
1088 .sendpage = inet_sendpage,
1089 .set_peek_off = sk_set_peek_off,
1090 #ifdef CONFIG_COMPAT
1091 .compat_ioctl = inet_compat_ioctl,
1092 #endif
1093 };
1094 EXPORT_SYMBOL(inet_dgram_ops);
1095
1096 /*
1097 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1098 * udp_poll
1099 */
1100 static const struct proto_ops inet_sockraw_ops = {
1101 .family = PF_INET,
1102 .owner = THIS_MODULE,
1103 .release = inet_release,
1104 .bind = inet_bind,
1105 .connect = inet_dgram_connect,
1106 .socketpair = sock_no_socketpair,
1107 .accept = sock_no_accept,
1108 .getname = inet_getname,
1109 .poll = datagram_poll,
1110 .ioctl = inet_ioctl,
1111 .gettstamp = sock_gettstamp,
1112 .listen = sock_no_listen,
1113 .shutdown = inet_shutdown,
1114 .setsockopt = sock_common_setsockopt,
1115 .getsockopt = sock_common_getsockopt,
1116 .sendmsg = inet_sendmsg,
1117 .recvmsg = inet_recvmsg,
1118 .mmap = sock_no_mmap,
1119 .sendpage = inet_sendpage,
1120 #ifdef CONFIG_COMPAT
1121 .compat_ioctl = inet_compat_ioctl,
1122 #endif
1123 };
1124
1125 static const struct net_proto_family inet_family_ops = {
1126 .family = PF_INET,
1127 .create = inet_create,
1128 .owner = THIS_MODULE,
1129 };
1130
1131 /* Upon startup we insert all the elements in inetsw_array[] into
1132 * the linked list inetsw.
1133 */
1134 static struct inet_protosw inetsw_array[] =
1135 {
1136 {
1137 .type = SOCK_STREAM,
1138 .protocol = IPPROTO_TCP,
1139 .prot = &tcp_prot,
1140 .ops = &inet_stream_ops,
1141 .flags = INET_PROTOSW_PERMANENT |
1142 INET_PROTOSW_ICSK,
1143 },
1144
1145 {
1146 .type = SOCK_DGRAM,
1147 .protocol = IPPROTO_UDP,
1148 .prot = &udp_prot,
1149 .ops = &inet_dgram_ops,
1150 .flags = INET_PROTOSW_PERMANENT,
1151 },
1152
1153 {
1154 .type = SOCK_DGRAM,
1155 .protocol = IPPROTO_ICMP,
1156 .prot = &ping_prot,
1157 .ops = &inet_sockraw_ops,
1158 .flags = INET_PROTOSW_REUSE,
1159 },
1160
1161 {
1162 .type = SOCK_RAW,
1163 .protocol = IPPROTO_IP, /* wild card */
1164 .prot = &raw_prot,
1165 .ops = &inet_sockraw_ops,
1166 .flags = INET_PROTOSW_REUSE,
1167 }
1168 };
1169
1170 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1171
inet_register_protosw(struct inet_protosw * p)1172 void inet_register_protosw(struct inet_protosw *p)
1173 {
1174 struct list_head *lh;
1175 struct inet_protosw *answer;
1176 int protocol = p->protocol;
1177 struct list_head *last_perm;
1178
1179 spin_lock_bh(&inetsw_lock);
1180
1181 if (p->type >= SOCK_MAX)
1182 goto out_illegal;
1183
1184 /* If we are trying to override a permanent protocol, bail. */
1185 last_perm = &inetsw[p->type];
1186 list_for_each(lh, &inetsw[p->type]) {
1187 answer = list_entry(lh, struct inet_protosw, list);
1188 /* Check only the non-wild match. */
1189 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1190 break;
1191 if (protocol == answer->protocol)
1192 goto out_permanent;
1193 last_perm = lh;
1194 }
1195
1196 /* Add the new entry after the last permanent entry if any, so that
1197 * the new entry does not override a permanent entry when matched with
1198 * a wild-card protocol. But it is allowed to override any existing
1199 * non-permanent entry. This means that when we remove this entry, the
1200 * system automatically returns to the old behavior.
1201 */
1202 list_add_rcu(&p->list, last_perm);
1203 out:
1204 spin_unlock_bh(&inetsw_lock);
1205
1206 return;
1207
1208 out_permanent:
1209 pr_err("Attempt to override permanent protocol %d\n", protocol);
1210 goto out;
1211
1212 out_illegal:
1213 pr_err("Ignoring attempt to register invalid socket type %d\n",
1214 p->type);
1215 goto out;
1216 }
1217 EXPORT_SYMBOL(inet_register_protosw);
1218
inet_unregister_protosw(struct inet_protosw * p)1219 void inet_unregister_protosw(struct inet_protosw *p)
1220 {
1221 if (INET_PROTOSW_PERMANENT & p->flags) {
1222 pr_err("Attempt to unregister permanent protocol %d\n",
1223 p->protocol);
1224 } else {
1225 spin_lock_bh(&inetsw_lock);
1226 list_del_rcu(&p->list);
1227 spin_unlock_bh(&inetsw_lock);
1228
1229 synchronize_net();
1230 }
1231 }
1232 EXPORT_SYMBOL(inet_unregister_protosw);
1233
inet_sk_reselect_saddr(struct sock * sk)1234 static int inet_sk_reselect_saddr(struct sock *sk)
1235 {
1236 struct inet_sock *inet = inet_sk(sk);
1237 __be32 old_saddr = inet->inet_saddr;
1238 __be32 daddr = inet->inet_daddr;
1239 struct flowi4 *fl4;
1240 struct rtable *rt;
1241 __be32 new_saddr;
1242 struct ip_options_rcu *inet_opt;
1243
1244 inet_opt = rcu_dereference_protected(inet->inet_opt,
1245 lockdep_sock_is_held(sk));
1246 if (inet_opt && inet_opt->opt.srr)
1247 daddr = inet_opt->opt.faddr;
1248
1249 /* Query new route. */
1250 fl4 = &inet->cork.fl.u.ip4;
1251 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1252 sk->sk_bound_dev_if, sk->sk_protocol,
1253 inet->inet_sport, inet->inet_dport, sk);
1254 if (IS_ERR(rt))
1255 return PTR_ERR(rt);
1256
1257 sk_setup_caps(sk, &rt->dst);
1258
1259 new_saddr = fl4->saddr;
1260
1261 if (new_saddr == old_saddr)
1262 return 0;
1263
1264 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1265 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1266 __func__, &old_saddr, &new_saddr);
1267 }
1268
1269 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1270
1271 /*
1272 * XXX The only one ugly spot where we need to
1273 * XXX really change the sockets identity after
1274 * XXX it has entered the hashes. -DaveM
1275 *
1276 * Besides that, it does not check for connection
1277 * uniqueness. Wait for troubles.
1278 */
1279 return __sk_prot_rehash(sk);
1280 }
1281
inet_sk_rebuild_header(struct sock * sk)1282 int inet_sk_rebuild_header(struct sock *sk)
1283 {
1284 struct inet_sock *inet = inet_sk(sk);
1285 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1286 __be32 daddr;
1287 struct ip_options_rcu *inet_opt;
1288 struct flowi4 *fl4;
1289 int err;
1290
1291 /* Route is OK, nothing to do. */
1292 if (rt)
1293 return 0;
1294
1295 /* Reroute. */
1296 rcu_read_lock();
1297 inet_opt = rcu_dereference(inet->inet_opt);
1298 daddr = inet->inet_daddr;
1299 if (inet_opt && inet_opt->opt.srr)
1300 daddr = inet_opt->opt.faddr;
1301 rcu_read_unlock();
1302 fl4 = &inet->cork.fl.u.ip4;
1303 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1304 inet->inet_dport, inet->inet_sport,
1305 sk->sk_protocol, RT_CONN_FLAGS(sk),
1306 sk->sk_bound_dev_if);
1307 if (!IS_ERR(rt)) {
1308 err = 0;
1309 sk_setup_caps(sk, &rt->dst);
1310 } else {
1311 err = PTR_ERR(rt);
1312
1313 /* Routing failed... */
1314 sk->sk_route_caps = 0;
1315 /*
1316 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1317 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1318 */
1319 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1320 sk->sk_state != TCP_SYN_SENT ||
1321 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1322 (err = inet_sk_reselect_saddr(sk)) != 0)
1323 sk->sk_err_soft = -err;
1324 }
1325
1326 return err;
1327 }
1328 EXPORT_SYMBOL(inet_sk_rebuild_header);
1329
inet_sk_set_state(struct sock * sk,int state)1330 void inet_sk_set_state(struct sock *sk, int state)
1331 {
1332 trace_inet_sock_set_state(sk, sk->sk_state, state);
1333 sk->sk_state = state;
1334 }
1335 EXPORT_SYMBOL(inet_sk_set_state);
1336
inet_sk_state_store(struct sock * sk,int newstate)1337 void inet_sk_state_store(struct sock *sk, int newstate)
1338 {
1339 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1340 smp_store_release(&sk->sk_state, newstate);
1341 }
1342
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1343 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1344 netdev_features_t features)
1345 {
1346 bool udpfrag = false, fixedid = false, gso_partial, encap;
1347 struct sk_buff *segs = ERR_PTR(-EINVAL);
1348 const struct net_offload *ops;
1349 unsigned int offset = 0;
1350 struct iphdr *iph;
1351 int proto, tot_len;
1352 int nhoff;
1353 int ihl;
1354 int id;
1355
1356 skb_reset_network_header(skb);
1357 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1358 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1359 goto out;
1360
1361 iph = ip_hdr(skb);
1362 ihl = iph->ihl * 4;
1363 if (ihl < sizeof(*iph))
1364 goto out;
1365
1366 id = ntohs(iph->id);
1367 proto = iph->protocol;
1368
1369 /* Warning: after this point, iph might be no longer valid */
1370 if (unlikely(!pskb_may_pull(skb, ihl)))
1371 goto out;
1372 __skb_pull(skb, ihl);
1373
1374 encap = SKB_GSO_CB(skb)->encap_level > 0;
1375 if (encap)
1376 features &= skb->dev->hw_enc_features;
1377 SKB_GSO_CB(skb)->encap_level += ihl;
1378
1379 skb_reset_transport_header(skb);
1380
1381 segs = ERR_PTR(-EPROTONOSUPPORT);
1382
1383 if (!skb->encapsulation || encap) {
1384 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1385 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1386
1387 /* fixed ID is invalid if DF bit is not set */
1388 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1389 goto out;
1390 }
1391
1392 ops = rcu_dereference(inet_offloads[proto]);
1393 if (likely(ops && ops->callbacks.gso_segment)) {
1394 segs = ops->callbacks.gso_segment(skb, features);
1395 if (!segs)
1396 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1397 }
1398
1399 if (IS_ERR_OR_NULL(segs))
1400 goto out;
1401
1402 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1403
1404 skb = segs;
1405 do {
1406 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1407 if (udpfrag) {
1408 iph->frag_off = htons(offset >> 3);
1409 if (skb->next)
1410 iph->frag_off |= htons(IP_MF);
1411 offset += skb->len - nhoff - ihl;
1412 tot_len = skb->len - nhoff;
1413 } else if (skb_is_gso(skb)) {
1414 if (!fixedid) {
1415 iph->id = htons(id);
1416 id += skb_shinfo(skb)->gso_segs;
1417 }
1418
1419 if (gso_partial)
1420 tot_len = skb_shinfo(skb)->gso_size +
1421 SKB_GSO_CB(skb)->data_offset +
1422 skb->head - (unsigned char *)iph;
1423 else
1424 tot_len = skb->len - nhoff;
1425 } else {
1426 if (!fixedid)
1427 iph->id = htons(id++);
1428 tot_len = skb->len - nhoff;
1429 }
1430 iph->tot_len = htons(tot_len);
1431 ip_send_check(iph);
1432 if (encap)
1433 skb_reset_inner_headers(skb);
1434 skb->network_header = (u8 *)iph - skb->head;
1435 skb_reset_mac_len(skb);
1436 } while ((skb = skb->next));
1437
1438 out:
1439 return segs;
1440 }
1441 EXPORT_SYMBOL(inet_gso_segment);
1442
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1443 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1444 netdev_features_t features)
1445 {
1446 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1447 return ERR_PTR(-EINVAL);
1448
1449 return inet_gso_segment(skb, features);
1450 }
1451
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1452 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1453 {
1454 const struct net_offload *ops;
1455 struct sk_buff *pp = NULL;
1456 const struct iphdr *iph;
1457 struct sk_buff *p;
1458 unsigned int hlen;
1459 unsigned int off;
1460 unsigned int id;
1461 int flush = 1;
1462 int proto;
1463
1464 off = skb_gro_offset(skb);
1465 hlen = off + sizeof(*iph);
1466 iph = skb_gro_header_fast(skb, off);
1467 if (skb_gro_header_hard(skb, hlen)) {
1468 iph = skb_gro_header_slow(skb, hlen, off);
1469 if (unlikely(!iph))
1470 goto out;
1471 }
1472
1473 proto = iph->protocol;
1474
1475 rcu_read_lock();
1476 ops = rcu_dereference(inet_offloads[proto]);
1477 if (!ops || !ops->callbacks.gro_receive)
1478 goto out_unlock;
1479
1480 if (*(u8 *)iph != 0x45)
1481 goto out_unlock;
1482
1483 if (ip_is_fragment(iph))
1484 goto out_unlock;
1485
1486 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1487 goto out_unlock;
1488
1489 id = ntohl(*(__be32 *)&iph->id);
1490 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1491 id >>= 16;
1492
1493 list_for_each_entry(p, head, list) {
1494 struct iphdr *iph2;
1495 u16 flush_id;
1496
1497 if (!NAPI_GRO_CB(p)->same_flow)
1498 continue;
1499
1500 iph2 = (struct iphdr *)(p->data + off);
1501 /* The above works because, with the exception of the top
1502 * (inner most) layer, we only aggregate pkts with the same
1503 * hdr length so all the hdrs we'll need to verify will start
1504 * at the same offset.
1505 */
1506 if ((iph->protocol ^ iph2->protocol) |
1507 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1508 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1509 NAPI_GRO_CB(p)->same_flow = 0;
1510 continue;
1511 }
1512
1513 /* All fields must match except length and checksum. */
1514 NAPI_GRO_CB(p)->flush |=
1515 (iph->ttl ^ iph2->ttl) |
1516 (iph->tos ^ iph2->tos) |
1517 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1518
1519 NAPI_GRO_CB(p)->flush |= flush;
1520
1521 /* We need to store of the IP ID check to be included later
1522 * when we can verify that this packet does in fact belong
1523 * to a given flow.
1524 */
1525 flush_id = (u16)(id - ntohs(iph2->id));
1526
1527 /* This bit of code makes it much easier for us to identify
1528 * the cases where we are doing atomic vs non-atomic IP ID
1529 * checks. Specifically an atomic check can return IP ID
1530 * values 0 - 0xFFFF, while a non-atomic check can only
1531 * return 0 or 0xFFFF.
1532 */
1533 if (!NAPI_GRO_CB(p)->is_atomic ||
1534 !(iph->frag_off & htons(IP_DF))) {
1535 flush_id ^= NAPI_GRO_CB(p)->count;
1536 flush_id = flush_id ? 0xFFFF : 0;
1537 }
1538
1539 /* If the previous IP ID value was based on an atomic
1540 * datagram we can overwrite the value and ignore it.
1541 */
1542 if (NAPI_GRO_CB(skb)->is_atomic)
1543 NAPI_GRO_CB(p)->flush_id = flush_id;
1544 else
1545 NAPI_GRO_CB(p)->flush_id |= flush_id;
1546 }
1547
1548 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1549 NAPI_GRO_CB(skb)->flush |= flush;
1550 skb_set_network_header(skb, off);
1551 /* The above will be needed by the transport layer if there is one
1552 * immediately following this IP hdr.
1553 */
1554
1555 /* Note : No need to call skb_gro_postpull_rcsum() here,
1556 * as we already checked checksum over ipv4 header was 0
1557 */
1558 skb_gro_pull(skb, sizeof(*iph));
1559 skb_set_transport_header(skb, skb_gro_offset(skb));
1560
1561 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1562 ops->callbacks.gro_receive, head, skb);
1563
1564 out_unlock:
1565 rcu_read_unlock();
1566
1567 out:
1568 skb_gro_flush_final(skb, pp, flush);
1569
1570 return pp;
1571 }
1572 EXPORT_SYMBOL(inet_gro_receive);
1573
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1574 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1575 struct sk_buff *skb)
1576 {
1577 if (NAPI_GRO_CB(skb)->encap_mark) {
1578 NAPI_GRO_CB(skb)->flush = 1;
1579 return NULL;
1580 }
1581
1582 NAPI_GRO_CB(skb)->encap_mark = 1;
1583
1584 return inet_gro_receive(head, skb);
1585 }
1586
1587 #define SECONDS_PER_DAY 86400
1588
1589 /* inet_current_timestamp - Return IP network timestamp
1590 *
1591 * Return milliseconds since midnight in network byte order.
1592 */
inet_current_timestamp(void)1593 __be32 inet_current_timestamp(void)
1594 {
1595 u32 secs;
1596 u32 msecs;
1597 struct timespec64 ts;
1598
1599 ktime_get_real_ts64(&ts);
1600
1601 /* Get secs since midnight. */
1602 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1603 /* Convert to msecs. */
1604 msecs = secs * MSEC_PER_SEC;
1605 /* Convert nsec to msec. */
1606 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1607
1608 /* Convert to network byte order. */
1609 return htonl(msecs);
1610 }
1611 EXPORT_SYMBOL(inet_current_timestamp);
1612
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1613 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1614 {
1615 if (sk->sk_family == AF_INET)
1616 return ip_recv_error(sk, msg, len, addr_len);
1617 #if IS_ENABLED(CONFIG_IPV6)
1618 if (sk->sk_family == AF_INET6)
1619 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1620 #endif
1621 return -EINVAL;
1622 }
1623
inet_gro_complete(struct sk_buff * skb,int nhoff)1624 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1625 {
1626 __be16 newlen = htons(skb->len - nhoff);
1627 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1628 const struct net_offload *ops;
1629 int proto = iph->protocol;
1630 int err = -ENOSYS;
1631
1632 if (skb->encapsulation) {
1633 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1634 skb_set_inner_network_header(skb, nhoff);
1635 }
1636
1637 csum_replace2(&iph->check, iph->tot_len, newlen);
1638 iph->tot_len = newlen;
1639
1640 rcu_read_lock();
1641 ops = rcu_dereference(inet_offloads[proto]);
1642 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1643 goto out_unlock;
1644
1645 /* Only need to add sizeof(*iph) to get to the next hdr below
1646 * because any hdr with option will have been flushed in
1647 * inet_gro_receive().
1648 */
1649 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1650 tcp4_gro_complete, udp4_gro_complete,
1651 skb, nhoff + sizeof(*iph));
1652
1653 out_unlock:
1654 rcu_read_unlock();
1655
1656 return err;
1657 }
1658 EXPORT_SYMBOL(inet_gro_complete);
1659
ipip_gro_complete(struct sk_buff * skb,int nhoff)1660 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1661 {
1662 skb->encapsulation = 1;
1663 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1664 return inet_gro_complete(skb, nhoff);
1665 }
1666
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1667 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1668 unsigned short type, unsigned char protocol,
1669 struct net *net)
1670 {
1671 struct socket *sock;
1672 int rc = sock_create_kern(net, family, type, protocol, &sock);
1673
1674 if (rc == 0) {
1675 *sk = sock->sk;
1676 (*sk)->sk_allocation = GFP_ATOMIC;
1677 /*
1678 * Unhash it so that IP input processing does not even see it,
1679 * we do not wish this socket to see incoming packets.
1680 */
1681 (*sk)->sk_prot->unhash(*sk);
1682 }
1683 return rc;
1684 }
1685 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1686
snmp_get_cpu_field(void __percpu * mib,int cpu,int offt)1687 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1688 {
1689 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1690 }
1691 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1692
snmp_fold_field(void __percpu * mib,int offt)1693 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1694 {
1695 unsigned long res = 0;
1696 int i;
1697
1698 for_each_possible_cpu(i)
1699 res += snmp_get_cpu_field(mib, i, offt);
1700 return res;
1701 }
1702 EXPORT_SYMBOL_GPL(snmp_fold_field);
1703
1704 #if BITS_PER_LONG==32
1705
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1706 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1707 size_t syncp_offset)
1708 {
1709 void *bhptr;
1710 struct u64_stats_sync *syncp;
1711 u64 v;
1712 unsigned int start;
1713
1714 bhptr = per_cpu_ptr(mib, cpu);
1715 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1716 do {
1717 start = u64_stats_fetch_begin_irq(syncp);
1718 v = *(((u64 *)bhptr) + offt);
1719 } while (u64_stats_fetch_retry_irq(syncp, start));
1720
1721 return v;
1722 }
1723 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1724
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1725 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1726 {
1727 u64 res = 0;
1728 int cpu;
1729
1730 for_each_possible_cpu(cpu) {
1731 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1732 }
1733 return res;
1734 }
1735 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1736 #endif
1737
1738 #ifdef CONFIG_IP_MULTICAST
1739 static const struct net_protocol igmp_protocol = {
1740 .handler = igmp_rcv,
1741 .netns_ok = 1,
1742 };
1743 #endif
1744
1745 static const struct net_protocol tcp_protocol = {
1746 .handler = tcp_v4_rcv,
1747 .err_handler = tcp_v4_err,
1748 .no_policy = 1,
1749 .netns_ok = 1,
1750 .icmp_strict_tag_validation = 1,
1751 };
1752
1753 static const struct net_protocol udp_protocol = {
1754 .handler = udp_rcv,
1755 .err_handler = udp_err,
1756 .no_policy = 1,
1757 .netns_ok = 1,
1758 };
1759
1760 static const struct net_protocol icmp_protocol = {
1761 .handler = icmp_rcv,
1762 .err_handler = icmp_err,
1763 .no_policy = 1,
1764 .netns_ok = 1,
1765 };
1766
ipv4_mib_init_net(struct net * net)1767 static __net_init int ipv4_mib_init_net(struct net *net)
1768 {
1769 int i;
1770
1771 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1772 if (!net->mib.tcp_statistics)
1773 goto err_tcp_mib;
1774 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1775 if (!net->mib.ip_statistics)
1776 goto err_ip_mib;
1777
1778 for_each_possible_cpu(i) {
1779 struct ipstats_mib *af_inet_stats;
1780 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1781 u64_stats_init(&af_inet_stats->syncp);
1782 }
1783
1784 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1785 if (!net->mib.net_statistics)
1786 goto err_net_mib;
1787 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1788 if (!net->mib.udp_statistics)
1789 goto err_udp_mib;
1790 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1791 if (!net->mib.udplite_statistics)
1792 goto err_udplite_mib;
1793 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1794 if (!net->mib.icmp_statistics)
1795 goto err_icmp_mib;
1796 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1797 GFP_KERNEL);
1798 if (!net->mib.icmpmsg_statistics)
1799 goto err_icmpmsg_mib;
1800
1801 tcp_mib_init(net);
1802 return 0;
1803
1804 err_icmpmsg_mib:
1805 free_percpu(net->mib.icmp_statistics);
1806 err_icmp_mib:
1807 free_percpu(net->mib.udplite_statistics);
1808 err_udplite_mib:
1809 free_percpu(net->mib.udp_statistics);
1810 err_udp_mib:
1811 free_percpu(net->mib.net_statistics);
1812 err_net_mib:
1813 free_percpu(net->mib.ip_statistics);
1814 err_ip_mib:
1815 free_percpu(net->mib.tcp_statistics);
1816 err_tcp_mib:
1817 return -ENOMEM;
1818 }
1819
ipv4_mib_exit_net(struct net * net)1820 static __net_exit void ipv4_mib_exit_net(struct net *net)
1821 {
1822 kfree(net->mib.icmpmsg_statistics);
1823 free_percpu(net->mib.icmp_statistics);
1824 free_percpu(net->mib.udplite_statistics);
1825 free_percpu(net->mib.udp_statistics);
1826 free_percpu(net->mib.net_statistics);
1827 free_percpu(net->mib.ip_statistics);
1828 free_percpu(net->mib.tcp_statistics);
1829 #ifdef CONFIG_MPTCP
1830 /* allocated on demand, see mptcp_init_sock() */
1831 free_percpu(net->mib.mptcp_statistics);
1832 #endif
1833 }
1834
1835 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1836 .init = ipv4_mib_init_net,
1837 .exit = ipv4_mib_exit_net,
1838 };
1839
init_ipv4_mibs(void)1840 static int __init init_ipv4_mibs(void)
1841 {
1842 return register_pernet_subsys(&ipv4_mib_ops);
1843 }
1844
inet_init_net(struct net * net)1845 static __net_init int inet_init_net(struct net *net)
1846 {
1847 /*
1848 * Set defaults for local port range
1849 */
1850 seqlock_init(&net->ipv4.ip_local_ports.lock);
1851 net->ipv4.ip_local_ports.range[0] = 32768;
1852 net->ipv4.ip_local_ports.range[1] = 60999;
1853
1854 seqlock_init(&net->ipv4.ping_group_range.lock);
1855 /*
1856 * Sane defaults - nobody may create ping sockets.
1857 * Boot scripts should set this to distro-specific group.
1858 */
1859 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1860 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1861
1862 /* Default values for sysctl-controlled parameters.
1863 * We set them here, in case sysctl is not compiled.
1864 */
1865 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1866 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1867 net->ipv4.sysctl_ip_dynaddr = 0;
1868 net->ipv4.sysctl_ip_early_demux = 1;
1869 net->ipv4.sysctl_udp_early_demux = 1;
1870 net->ipv4.sysctl_tcp_early_demux = 1;
1871 net->ipv4.sysctl_nexthop_compat_mode = 1;
1872 #ifdef CONFIG_SYSCTL
1873 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1874 #endif
1875
1876 /* Some igmp sysctl, whose values are always used */
1877 net->ipv4.sysctl_igmp_max_memberships = 20;
1878 net->ipv4.sysctl_igmp_max_msf = 10;
1879 /* IGMP reports for link-local multicast groups are enabled by default */
1880 net->ipv4.sysctl_igmp_llm_reports = 1;
1881 net->ipv4.sysctl_igmp_qrv = 2;
1882
1883 #ifdef CONFIG_LOWPOWER_PROTOCOL
1884 lowpower_protocol_net_init(net);
1885 #endif /* CONFIG_LOWPOWER_PROTOCOL */
1886 return 0;
1887 }
1888
1889 static __net_initdata struct pernet_operations af_inet_ops = {
1890 .init = inet_init_net,
1891 };
1892
init_inet_pernet_ops(void)1893 static int __init init_inet_pernet_ops(void)
1894 {
1895 return register_pernet_subsys(&af_inet_ops);
1896 }
1897
1898 static int ipv4_proc_init(void);
1899
1900 /*
1901 * IP protocol layer initialiser
1902 */
1903
1904 static struct packet_offload ip_packet_offload __read_mostly = {
1905 .type = cpu_to_be16(ETH_P_IP),
1906 .callbacks = {
1907 .gso_segment = inet_gso_segment,
1908 .gro_receive = inet_gro_receive,
1909 .gro_complete = inet_gro_complete,
1910 },
1911 };
1912
1913 static const struct net_offload ipip_offload = {
1914 .callbacks = {
1915 .gso_segment = ipip_gso_segment,
1916 .gro_receive = ipip_gro_receive,
1917 .gro_complete = ipip_gro_complete,
1918 },
1919 };
1920
ipip_offload_init(void)1921 static int __init ipip_offload_init(void)
1922 {
1923 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1924 }
1925
ipv4_offload_init(void)1926 static int __init ipv4_offload_init(void)
1927 {
1928 /*
1929 * Add offloads
1930 */
1931 if (udpv4_offload_init() < 0)
1932 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1933 if (tcpv4_offload_init() < 0)
1934 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1935 if (ipip_offload_init() < 0)
1936 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1937
1938 dev_add_offload(&ip_packet_offload);
1939 return 0;
1940 }
1941
1942 fs_initcall(ipv4_offload_init);
1943
1944 static struct packet_type ip_packet_type __read_mostly = {
1945 .type = cpu_to_be16(ETH_P_IP),
1946 .func = ip_rcv,
1947 .list_func = ip_list_rcv,
1948 };
1949
inet_init(void)1950 static int __init inet_init(void)
1951 {
1952 struct inet_protosw *q;
1953 struct list_head *r;
1954 int rc;
1955
1956 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1957
1958 rc = proto_register(&tcp_prot, 1);
1959 if (rc)
1960 goto out;
1961
1962 rc = proto_register(&udp_prot, 1);
1963 if (rc)
1964 goto out_unregister_tcp_proto;
1965
1966 rc = proto_register(&raw_prot, 1);
1967 if (rc)
1968 goto out_unregister_udp_proto;
1969
1970 rc = proto_register(&ping_prot, 1);
1971 if (rc)
1972 goto out_unregister_raw_proto;
1973
1974 /*
1975 * Tell SOCKET that we are alive...
1976 */
1977
1978 (void)sock_register(&inet_family_ops);
1979
1980 #ifdef CONFIG_SYSCTL
1981 ip_static_sysctl_init();
1982 #endif
1983
1984 /*
1985 * Add all the base protocols.
1986 */
1987
1988 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1989 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1990 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1991 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1992 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1993 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1994 #ifdef CONFIG_IP_MULTICAST
1995 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1996 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1997 #endif
1998
1999 /* Register the socket-side information for inet_create. */
2000 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2001 INIT_LIST_HEAD(r);
2002
2003 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2004 inet_register_protosw(q);
2005
2006 /*
2007 * Set the ARP module up
2008 */
2009
2010 arp_init();
2011
2012 /*
2013 * Set the IP module up
2014 */
2015
2016 ip_init();
2017
2018 /* Initialise per-cpu ipv4 mibs */
2019 if (init_ipv4_mibs())
2020 panic("%s: Cannot init ipv4 mibs\n", __func__);
2021
2022 /* Setup TCP slab cache for open requests. */
2023 tcp_init();
2024
2025 /* Setup UDP memory threshold */
2026 udp_init();
2027
2028 /* Add UDP-Lite (RFC 3828) */
2029 udplite4_register();
2030
2031 raw_init();
2032
2033 ping_init();
2034
2035 /*
2036 * Set the ICMP layer up
2037 */
2038
2039 if (icmp_init() < 0)
2040 panic("Failed to create the ICMP control socket.\n");
2041
2042 /*
2043 * Initialise the multicast router
2044 */
2045 #if defined(CONFIG_IP_MROUTE)
2046 if (ip_mr_init())
2047 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2048 #endif
2049
2050 if (init_inet_pernet_ops())
2051 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2052
2053 ipv4_proc_init();
2054
2055 ipfrag_init();
2056
2057 dev_add_pack(&ip_packet_type);
2058
2059 ip_tunnel_core_init();
2060
2061 rc = 0;
2062 out:
2063 return rc;
2064 out_unregister_raw_proto:
2065 proto_unregister(&raw_prot);
2066 out_unregister_udp_proto:
2067 proto_unregister(&udp_prot);
2068 out_unregister_tcp_proto:
2069 proto_unregister(&tcp_prot);
2070 goto out;
2071 }
2072
2073 fs_initcall(inet_init);
2074
2075 /* ------------------------------------------------------------------------ */
2076
2077 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2078 static int __init ipv4_proc_init(void)
2079 {
2080 int rc = 0;
2081
2082 if (raw_proc_init())
2083 goto out_raw;
2084 if (tcp4_proc_init())
2085 goto out_tcp;
2086 if (udp4_proc_init())
2087 goto out_udp;
2088 if (ping_proc_init())
2089 goto out_ping;
2090 if (ip_misc_proc_init())
2091 goto out_misc;
2092 out:
2093 return rc;
2094 out_misc:
2095 ping_proc_exit();
2096 out_ping:
2097 udp4_proc_exit();
2098 out_udp:
2099 tcp4_proc_exit();
2100 out_tcp:
2101 raw_proc_exit();
2102 out_raw:
2103 rc = -ENOMEM;
2104 goto out;
2105 }
2106
2107 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2108 static int __init ipv4_proc_init(void)
2109 {
2110 return 0;
2111 }
2112 #endif /* CONFIG_PROC_FS */
2113