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 goto out_sk_release;
381 }
382
383 if (sk->sk_prot->init) {
384 err = sk->sk_prot->init(sk);
385 if (err)
386 goto out_sk_release;
387 }
388
389 if (!kern) {
390 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
391 if (err)
392 goto out_sk_release;
393 }
394 out:
395 return err;
396 out_rcu_unlock:
397 rcu_read_unlock();
398 goto out;
399 out_sk_release:
400 sk_common_release(sk);
401 sock->sk = NULL;
402 goto out;
403 }
404
405
406 /*
407 * The peer socket should always be NULL (or else). When we call this
408 * function we are destroying the object and from then on nobody
409 * should refer to it.
410 */
inet_release(struct socket * sock)411 int inet_release(struct socket *sock)
412 {
413 struct sock *sk = sock->sk;
414
415 if (sk) {
416 long timeout;
417
418 if (!sk->sk_kern_sock)
419 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
420
421 /* Applications forget to leave groups before exiting */
422 ip_mc_drop_socket(sk);
423
424 /* If linger is set, we don't return until the close
425 * is complete. Otherwise we return immediately. The
426 * actually closing is done the same either way.
427 *
428 * If the close is due to the process exiting, we never
429 * linger..
430 */
431 timeout = 0;
432 if (sock_flag(sk, SOCK_LINGER) &&
433 !(current->flags & PF_EXITING))
434 timeout = sk->sk_lingertime;
435 sk->sk_prot->close(sk, timeout);
436 sock->sk = NULL;
437 }
438 return 0;
439 }
440 EXPORT_SYMBOL(inet_release);
441
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)442 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
443 {
444 struct sock *sk = sock->sk;
445 int err;
446
447 /* If the socket has its own bind function then use it. (RAW) */
448 if (sk->sk_prot->bind) {
449 return sk->sk_prot->bind(sk, uaddr, addr_len);
450 }
451 if (addr_len < sizeof(struct sockaddr_in))
452 return -EINVAL;
453
454 /* BPF prog is run before any checks are done so that if the prog
455 * changes context in a wrong way it will be caught.
456 */
457 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
458 if (err)
459 return err;
460
461 return __inet_bind(sk, uaddr, addr_len, BIND_WITH_LOCK);
462 }
463 EXPORT_SYMBOL(inet_bind);
464
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)465 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
466 u32 flags)
467 {
468 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
469 struct inet_sock *inet = inet_sk(sk);
470 struct net *net = sock_net(sk);
471 unsigned short snum;
472 int chk_addr_ret;
473 u32 tb_id = RT_TABLE_LOCAL;
474 int err;
475
476 if (addr->sin_family != AF_INET) {
477 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
478 * only if s_addr is INADDR_ANY.
479 */
480 err = -EAFNOSUPPORT;
481 if (addr->sin_family != AF_UNSPEC ||
482 addr->sin_addr.s_addr != htonl(INADDR_ANY))
483 goto out;
484 }
485
486 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
487 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
488
489 /* Not specified by any standard per-se, however it breaks too
490 * many applications when removed. It is unfortunate since
491 * allowing applications to make a non-local bind solves
492 * several problems with systems using dynamic addressing.
493 * (ie. your servers still start up even if your ISDN link
494 * is temporarily down)
495 */
496 err = -EADDRNOTAVAIL;
497 if (!inet_can_nonlocal_bind(net, inet) &&
498 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
499 chk_addr_ret != RTN_LOCAL &&
500 chk_addr_ret != RTN_MULTICAST &&
501 chk_addr_ret != RTN_BROADCAST)
502 goto out;
503
504 snum = ntohs(addr->sin_port);
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 const struct proto *prot;
567 int err;
568
569 if (addr_len < sizeof(uaddr->sa_family))
570 return -EINVAL;
571
572 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
573 prot = READ_ONCE(sk->sk_prot);
574
575 if (uaddr->sa_family == AF_UNSPEC)
576 return prot->disconnect(sk, flags);
577
578 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
579 err = prot->pre_connect(sk, uaddr, addr_len);
580 if (err)
581 return err;
582 }
583
584 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
585 return -EAGAIN;
586 return prot->connect(sk, uaddr, addr_len);
587 }
588 EXPORT_SYMBOL(inet_dgram_connect);
589
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)590 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
591 {
592 DEFINE_WAIT_FUNC(wait, woken_wake_function);
593
594 add_wait_queue(sk_sleep(sk), &wait);
595 sk->sk_write_pending += writebias;
596 sk->sk_wait_pending++;
597
598 /* Basic assumption: if someone sets sk->sk_err, he _must_
599 * change state of the socket from TCP_SYN_*.
600 * Connect() does not allow to get error notifications
601 * without closing the socket.
602 */
603 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
604 release_sock(sk);
605 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
606 lock_sock(sk);
607 if (signal_pending(current) || !timeo)
608 break;
609 }
610 remove_wait_queue(sk_sleep(sk), &wait);
611 sk->sk_write_pending -= writebias;
612 sk->sk_wait_pending--;
613 return timeo;
614 }
615
616 /*
617 * Connect to a remote host. There is regrettably still a little
618 * TCP 'magic' in here.
619 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)620 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
621 int addr_len, int flags, int is_sendmsg)
622 {
623 struct sock *sk = sock->sk;
624 int err;
625 long timeo;
626
627 /*
628 * uaddr can be NULL and addr_len can be 0 if:
629 * sk is a TCP fastopen active socket and
630 * TCP_FASTOPEN_CONNECT sockopt is set and
631 * we already have a valid cookie for this socket.
632 * In this case, user can call write() after connect().
633 * write() will invoke tcp_sendmsg_fastopen() which calls
634 * __inet_stream_connect().
635 */
636 if (uaddr) {
637 if (addr_len < sizeof(uaddr->sa_family))
638 return -EINVAL;
639
640 if (uaddr->sa_family == AF_UNSPEC) {
641 err = sk->sk_prot->disconnect(sk, flags);
642 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
643 goto out;
644 }
645 }
646
647 switch (sock->state) {
648 default:
649 err = -EINVAL;
650 goto out;
651 case SS_CONNECTED:
652 err = -EISCONN;
653 goto out;
654 case SS_CONNECTING:
655 if (inet_sk(sk)->defer_connect)
656 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
657 else
658 err = -EALREADY;
659 /* Fall out of switch with err, set for this state */
660 break;
661 case SS_UNCONNECTED:
662 err = -EISCONN;
663 if (sk->sk_state != TCP_CLOSE)
664 goto out;
665
666 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
667 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
668 if (err)
669 goto out;
670 }
671
672 err = sk->sk_prot->connect(sk, uaddr, addr_len);
673 if (err < 0)
674 goto out;
675
676 sock->state = SS_CONNECTING;
677
678 if (!err && inet_sk(sk)->defer_connect)
679 goto out;
680
681 /* Just entered SS_CONNECTING state; the only
682 * difference is that return value in non-blocking
683 * case is EINPROGRESS, rather than EALREADY.
684 */
685 err = -EINPROGRESS;
686 break;
687 }
688
689 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
690
691 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
692 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
693 tcp_sk(sk)->fastopen_req &&
694 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
695
696 /* Error code is set above */
697 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
698 goto out;
699
700 err = sock_intr_errno(timeo);
701 if (signal_pending(current))
702 goto out;
703 }
704
705 /* Connection was closed by RST, timeout, ICMP error
706 * or another process disconnected us.
707 */
708 if (sk->sk_state == TCP_CLOSE)
709 goto sock_error;
710
711 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
712 * and error was received after socket entered established state.
713 * Hence, it is handled normally after connect() return successfully.
714 */
715
716 sock->state = SS_CONNECTED;
717 err = 0;
718 out:
719 return err;
720
721 sock_error:
722 err = sock_error(sk) ? : -ECONNABORTED;
723 sock->state = SS_UNCONNECTED;
724 if (sk->sk_prot->disconnect(sk, flags))
725 sock->state = SS_DISCONNECTING;
726 goto out;
727 }
728 EXPORT_SYMBOL(__inet_stream_connect);
729
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)730 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
731 int addr_len, int flags)
732 {
733 int err;
734
735 lock_sock(sock->sk);
736 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
737 release_sock(sock->sk);
738 return err;
739 }
740 EXPORT_SYMBOL(inet_stream_connect);
741
742 /*
743 * Accept a pending connection. The TCP layer now gives BSD semantics.
744 */
745
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)746 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
747 bool kern)
748 {
749 struct sock *sk1 = sock->sk, *sk2;
750 int err = -EINVAL;
751
752 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
753 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
754 if (!sk2)
755 goto do_err;
756
757 lock_sock(sk2);
758
759 sock_rps_record_flow(sk2);
760 WARN_ON(!((1 << sk2->sk_state) &
761 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
762 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
763
764 sock_graft(sk2, newsock);
765
766 newsock->state = SS_CONNECTED;
767 err = 0;
768 release_sock(sk2);
769 do_err:
770 return err;
771 }
772 EXPORT_SYMBOL(inet_accept);
773
774 /*
775 * This does both peername and sockname.
776 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)777 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
778 int peer)
779 {
780 struct sock *sk = sock->sk;
781 struct inet_sock *inet = inet_sk(sk);
782 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
783
784 sin->sin_family = AF_INET;
785 if (peer) {
786 if (!inet->inet_dport ||
787 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
788 peer == 1))
789 return -ENOTCONN;
790 sin->sin_port = inet->inet_dport;
791 sin->sin_addr.s_addr = inet->inet_daddr;
792 } else {
793 __be32 addr = inet->inet_rcv_saddr;
794 if (!addr)
795 addr = inet->inet_saddr;
796 sin->sin_port = inet->inet_sport;
797 sin->sin_addr.s_addr = addr;
798 }
799 if (cgroup_bpf_enabled)
800 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, (struct sockaddr *)sin,
801 peer ? BPF_CGROUP_INET4_GETPEERNAME :
802 BPF_CGROUP_INET4_GETSOCKNAME,
803 NULL);
804 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
805 return sizeof(*sin);
806 }
807 EXPORT_SYMBOL(inet_getname);
808
inet_send_prepare(struct sock * sk)809 int inet_send_prepare(struct sock *sk)
810 {
811 sock_rps_record_flow(sk);
812
813 /* We may need to bind the socket. */
814 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
815 inet_autobind(sk))
816 return -EAGAIN;
817
818 return 0;
819 }
820 EXPORT_SYMBOL_GPL(inet_send_prepare);
821
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)822 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
823 {
824 struct sock *sk = sock->sk;
825
826 if (unlikely(inet_send_prepare(sk)))
827 return -EAGAIN;
828
829 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
830 sk, msg, size);
831 }
832 EXPORT_SYMBOL(inet_sendmsg);
833
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)834 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
835 size_t size, int flags)
836 {
837 struct sock *sk = sock->sk;
838 const struct proto *prot;
839
840 if (unlikely(inet_send_prepare(sk)))
841 return -EAGAIN;
842
843 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
844 prot = READ_ONCE(sk->sk_prot);
845 if (prot->sendpage)
846 return prot->sendpage(sk, page, offset, size, flags);
847 return sock_no_sendpage(sock, page, offset, size, flags);
848 }
849 EXPORT_SYMBOL(inet_sendpage);
850
851 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
852 size_t, int, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)853 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
854 int flags)
855 {
856 struct sock *sk = sock->sk;
857 int addr_len = 0;
858 int err;
859
860 if (likely(!(flags & MSG_ERRQUEUE)))
861 sock_rps_record_flow(sk);
862
863 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
864 sk, msg, size, flags & MSG_DONTWAIT,
865 flags & ~MSG_DONTWAIT, &addr_len);
866 if (err >= 0)
867 msg->msg_namelen = addr_len;
868 return err;
869 }
870 EXPORT_SYMBOL(inet_recvmsg);
871
inet_shutdown(struct socket * sock,int how)872 int inet_shutdown(struct socket *sock, int how)
873 {
874 struct sock *sk = sock->sk;
875 int err = 0;
876
877 /* This should really check to make sure
878 * the socket is a TCP socket. (WHY AC...)
879 */
880 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
881 1->2 bit 2 snds.
882 2->3 */
883 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
884 return -EINVAL;
885
886 lock_sock(sk);
887 if (sock->state == SS_CONNECTING) {
888 if ((1 << sk->sk_state) &
889 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
890 sock->state = SS_DISCONNECTING;
891 else
892 sock->state = SS_CONNECTED;
893 }
894
895 switch (sk->sk_state) {
896 case TCP_CLOSE:
897 err = -ENOTCONN;
898 /* Hack to wake up other listeners, who can poll for
899 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
900 fallthrough;
901 default:
902 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
903 if (sk->sk_prot->shutdown)
904 sk->sk_prot->shutdown(sk, how);
905 break;
906
907 /* Remaining two branches are temporary solution for missing
908 * close() in multithreaded environment. It is _not_ a good idea,
909 * but we have no choice until close() is repaired at VFS level.
910 */
911 case TCP_LISTEN:
912 if (!(how & RCV_SHUTDOWN))
913 break;
914 fallthrough;
915 case TCP_SYN_SENT:
916 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
917 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
918 break;
919 }
920
921 /* Wake up anyone sleeping in poll. */
922 sk->sk_state_change(sk);
923 release_sock(sk);
924 return err;
925 }
926 EXPORT_SYMBOL(inet_shutdown);
927
928 /*
929 * ioctl() calls you can issue on an INET socket. Most of these are
930 * device configuration and stuff and very rarely used. Some ioctls
931 * pass on to the socket itself.
932 *
933 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
934 * loads the devconfigure module does its configuring and unloads it.
935 * There's a good 20K of config code hanging around the kernel.
936 */
937
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)938 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
939 {
940 struct sock *sk = sock->sk;
941 int err = 0;
942 struct net *net = sock_net(sk);
943 void __user *p = (void __user *)arg;
944 struct ifreq ifr;
945 struct rtentry rt;
946
947 switch (cmd) {
948 case SIOCADDRT:
949 case SIOCDELRT:
950 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
951 return -EFAULT;
952 err = ip_rt_ioctl(net, cmd, &rt);
953 break;
954 case SIOCRTMSG:
955 err = -EINVAL;
956 break;
957 case SIOCDARP:
958 case SIOCGARP:
959 case SIOCSARP:
960 err = arp_ioctl(net, cmd, (void __user *)arg);
961 break;
962 case SIOCGIFADDR:
963 case SIOCGIFBRDADDR:
964 case SIOCGIFNETMASK:
965 case SIOCGIFDSTADDR:
966 case SIOCGIFPFLAGS:
967 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
968 return -EFAULT;
969 err = devinet_ioctl(net, cmd, &ifr);
970 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
971 err = -EFAULT;
972 break;
973
974 case SIOCSIFADDR:
975 case SIOCSIFBRDADDR:
976 case SIOCSIFNETMASK:
977 case SIOCSIFDSTADDR:
978 case SIOCSIFPFLAGS:
979 case SIOCSIFFLAGS:
980 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
981 return -EFAULT;
982 err = devinet_ioctl(net, cmd, &ifr);
983 break;
984 default:
985 if (sk->sk_prot->ioctl)
986 err = sk->sk_prot->ioctl(sk, cmd, arg);
987 else
988 err = -ENOIOCTLCMD;
989 break;
990 }
991 return err;
992 }
993 EXPORT_SYMBOL(inet_ioctl);
994
995 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)996 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
997 struct compat_rtentry __user *ur)
998 {
999 compat_uptr_t rtdev;
1000 struct rtentry rt;
1001
1002 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1003 3 * sizeof(struct sockaddr)) ||
1004 get_user(rt.rt_flags, &ur->rt_flags) ||
1005 get_user(rt.rt_metric, &ur->rt_metric) ||
1006 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1007 get_user(rt.rt_window, &ur->rt_window) ||
1008 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1009 get_user(rtdev, &ur->rt_dev))
1010 return -EFAULT;
1011
1012 rt.rt_dev = compat_ptr(rtdev);
1013 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1014 }
1015
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1016 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1017 {
1018 void __user *argp = compat_ptr(arg);
1019 struct sock *sk = sock->sk;
1020
1021 switch (cmd) {
1022 case SIOCADDRT:
1023 case SIOCDELRT:
1024 return inet_compat_routing_ioctl(sk, cmd, argp);
1025 default:
1026 if (!sk->sk_prot->compat_ioctl)
1027 return -ENOIOCTLCMD;
1028 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1029 }
1030 }
1031 #endif /* CONFIG_COMPAT */
1032
1033 const struct proto_ops inet_stream_ops = {
1034 .family = PF_INET,
1035 .owner = THIS_MODULE,
1036 .release = inet_release,
1037 .bind = inet_bind,
1038 .connect = inet_stream_connect,
1039 .socketpair = sock_no_socketpair,
1040 .accept = inet_accept,
1041 .getname = inet_getname,
1042 .poll = tcp_poll,
1043 .ioctl = inet_ioctl,
1044 .gettstamp = sock_gettstamp,
1045 .listen = inet_listen,
1046 .shutdown = inet_shutdown,
1047 .setsockopt = sock_common_setsockopt,
1048 .getsockopt = sock_common_getsockopt,
1049 .sendmsg = inet_sendmsg,
1050 .recvmsg = inet_recvmsg,
1051 #ifdef CONFIG_MMU
1052 .mmap = tcp_mmap,
1053 #endif
1054 .sendpage = inet_sendpage,
1055 .splice_read = tcp_splice_read,
1056 .read_sock = tcp_read_sock,
1057 .sendmsg_locked = tcp_sendmsg_locked,
1058 .sendpage_locked = tcp_sendpage_locked,
1059 .peek_len = tcp_peek_len,
1060 #ifdef CONFIG_COMPAT
1061 .compat_ioctl = inet_compat_ioctl,
1062 #endif
1063 .set_rcvlowat = tcp_set_rcvlowat,
1064 };
1065 EXPORT_SYMBOL(inet_stream_ops);
1066
1067 const struct proto_ops inet_dgram_ops = {
1068 .family = PF_INET,
1069 .owner = THIS_MODULE,
1070 .release = inet_release,
1071 .bind = inet_bind,
1072 .connect = inet_dgram_connect,
1073 .socketpair = sock_no_socketpair,
1074 .accept = sock_no_accept,
1075 .getname = inet_getname,
1076 .poll = udp_poll,
1077 .ioctl = inet_ioctl,
1078 .gettstamp = sock_gettstamp,
1079 .listen = sock_no_listen,
1080 .shutdown = inet_shutdown,
1081 .setsockopt = sock_common_setsockopt,
1082 .getsockopt = sock_common_getsockopt,
1083 .sendmsg = inet_sendmsg,
1084 .recvmsg = inet_recvmsg,
1085 .mmap = sock_no_mmap,
1086 .sendpage = inet_sendpage,
1087 .set_peek_off = sk_set_peek_off,
1088 #ifdef CONFIG_COMPAT
1089 .compat_ioctl = inet_compat_ioctl,
1090 #endif
1091 };
1092 EXPORT_SYMBOL(inet_dgram_ops);
1093
1094 /*
1095 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1096 * udp_poll
1097 */
1098 static const struct proto_ops inet_sockraw_ops = {
1099 .family = PF_INET,
1100 .owner = THIS_MODULE,
1101 .release = inet_release,
1102 .bind = inet_bind,
1103 .connect = inet_dgram_connect,
1104 .socketpair = sock_no_socketpair,
1105 .accept = sock_no_accept,
1106 .getname = inet_getname,
1107 .poll = datagram_poll,
1108 .ioctl = inet_ioctl,
1109 .gettstamp = sock_gettstamp,
1110 .listen = sock_no_listen,
1111 .shutdown = inet_shutdown,
1112 .setsockopt = sock_common_setsockopt,
1113 .getsockopt = sock_common_getsockopt,
1114 .sendmsg = inet_sendmsg,
1115 .recvmsg = inet_recvmsg,
1116 .mmap = sock_no_mmap,
1117 .sendpage = inet_sendpage,
1118 #ifdef CONFIG_COMPAT
1119 .compat_ioctl = inet_compat_ioctl,
1120 #endif
1121 };
1122
1123 static const struct net_proto_family inet_family_ops = {
1124 .family = PF_INET,
1125 .create = inet_create,
1126 .owner = THIS_MODULE,
1127 };
1128
1129 /* Upon startup we insert all the elements in inetsw_array[] into
1130 * the linked list inetsw.
1131 */
1132 static struct inet_protosw inetsw_array[] =
1133 {
1134 {
1135 .type = SOCK_STREAM,
1136 .protocol = IPPROTO_TCP,
1137 .prot = &tcp_prot,
1138 .ops = &inet_stream_ops,
1139 .flags = INET_PROTOSW_PERMANENT |
1140 INET_PROTOSW_ICSK,
1141 },
1142
1143 {
1144 .type = SOCK_DGRAM,
1145 .protocol = IPPROTO_UDP,
1146 .prot = &udp_prot,
1147 .ops = &inet_dgram_ops,
1148 .flags = INET_PROTOSW_PERMANENT,
1149 },
1150
1151 {
1152 .type = SOCK_DGRAM,
1153 .protocol = IPPROTO_ICMP,
1154 .prot = &ping_prot,
1155 .ops = &inet_sockraw_ops,
1156 .flags = INET_PROTOSW_REUSE,
1157 },
1158
1159 {
1160 .type = SOCK_RAW,
1161 .protocol = IPPROTO_IP, /* wild card */
1162 .prot = &raw_prot,
1163 .ops = &inet_sockraw_ops,
1164 .flags = INET_PROTOSW_REUSE,
1165 }
1166 };
1167
1168 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1169
inet_register_protosw(struct inet_protosw * p)1170 void inet_register_protosw(struct inet_protosw *p)
1171 {
1172 struct list_head *lh;
1173 struct inet_protosw *answer;
1174 int protocol = p->protocol;
1175 struct list_head *last_perm;
1176
1177 spin_lock_bh(&inetsw_lock);
1178
1179 if (p->type >= SOCK_MAX)
1180 goto out_illegal;
1181
1182 /* If we are trying to override a permanent protocol, bail. */
1183 last_perm = &inetsw[p->type];
1184 list_for_each(lh, &inetsw[p->type]) {
1185 answer = list_entry(lh, struct inet_protosw, list);
1186 /* Check only the non-wild match. */
1187 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1188 break;
1189 if (protocol == answer->protocol)
1190 goto out_permanent;
1191 last_perm = lh;
1192 }
1193
1194 /* Add the new entry after the last permanent entry if any, so that
1195 * the new entry does not override a permanent entry when matched with
1196 * a wild-card protocol. But it is allowed to override any existing
1197 * non-permanent entry. This means that when we remove this entry, the
1198 * system automatically returns to the old behavior.
1199 */
1200 list_add_rcu(&p->list, last_perm);
1201 out:
1202 spin_unlock_bh(&inetsw_lock);
1203
1204 return;
1205
1206 out_permanent:
1207 pr_err("Attempt to override permanent protocol %d\n", protocol);
1208 goto out;
1209
1210 out_illegal:
1211 pr_err("Ignoring attempt to register invalid socket type %d\n",
1212 p->type);
1213 goto out;
1214 }
1215 EXPORT_SYMBOL(inet_register_protosw);
1216
inet_unregister_protosw(struct inet_protosw * p)1217 void inet_unregister_protosw(struct inet_protosw *p)
1218 {
1219 if (INET_PROTOSW_PERMANENT & p->flags) {
1220 pr_err("Attempt to unregister permanent protocol %d\n",
1221 p->protocol);
1222 } else {
1223 spin_lock_bh(&inetsw_lock);
1224 list_del_rcu(&p->list);
1225 spin_unlock_bh(&inetsw_lock);
1226
1227 synchronize_net();
1228 }
1229 }
1230 EXPORT_SYMBOL(inet_unregister_protosw);
1231
inet_sk_reselect_saddr(struct sock * sk)1232 static int inet_sk_reselect_saddr(struct sock *sk)
1233 {
1234 struct inet_sock *inet = inet_sk(sk);
1235 __be32 old_saddr = inet->inet_saddr;
1236 __be32 daddr = inet->inet_daddr;
1237 struct flowi4 *fl4;
1238 struct rtable *rt;
1239 __be32 new_saddr;
1240 struct ip_options_rcu *inet_opt;
1241
1242 inet_opt = rcu_dereference_protected(inet->inet_opt,
1243 lockdep_sock_is_held(sk));
1244 if (inet_opt && inet_opt->opt.srr)
1245 daddr = inet_opt->opt.faddr;
1246
1247 /* Query new route. */
1248 fl4 = &inet->cork.fl.u.ip4;
1249 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1250 sk->sk_bound_dev_if, sk->sk_protocol,
1251 inet->inet_sport, inet->inet_dport, sk);
1252 if (IS_ERR(rt))
1253 return PTR_ERR(rt);
1254
1255 sk_setup_caps(sk, &rt->dst);
1256
1257 new_saddr = fl4->saddr;
1258
1259 if (new_saddr == old_saddr)
1260 return 0;
1261
1262 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1263 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1264 __func__, &old_saddr, &new_saddr);
1265 }
1266
1267 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1268
1269 /*
1270 * XXX The only one ugly spot where we need to
1271 * XXX really change the sockets identity after
1272 * XXX it has entered the hashes. -DaveM
1273 *
1274 * Besides that, it does not check for connection
1275 * uniqueness. Wait for troubles.
1276 */
1277 return __sk_prot_rehash(sk);
1278 }
1279
inet_sk_rebuild_header(struct sock * sk)1280 int inet_sk_rebuild_header(struct sock *sk)
1281 {
1282 struct inet_sock *inet = inet_sk(sk);
1283 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1284 __be32 daddr;
1285 struct ip_options_rcu *inet_opt;
1286 struct flowi4 *fl4;
1287 int err;
1288
1289 /* Route is OK, nothing to do. */
1290 if (rt)
1291 return 0;
1292
1293 /* Reroute. */
1294 rcu_read_lock();
1295 inet_opt = rcu_dereference(inet->inet_opt);
1296 daddr = inet->inet_daddr;
1297 if (inet_opt && inet_opt->opt.srr)
1298 daddr = inet_opt->opt.faddr;
1299 rcu_read_unlock();
1300 fl4 = &inet->cork.fl.u.ip4;
1301 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1302 inet->inet_dport, inet->inet_sport,
1303 sk->sk_protocol, RT_CONN_FLAGS(sk),
1304 sk->sk_bound_dev_if);
1305 if (!IS_ERR(rt)) {
1306 err = 0;
1307 sk_setup_caps(sk, &rt->dst);
1308 } else {
1309 err = PTR_ERR(rt);
1310
1311 /* Routing failed... */
1312 sk->sk_route_caps = 0;
1313 /*
1314 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1315 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1316 */
1317 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1318 sk->sk_state != TCP_SYN_SENT ||
1319 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1320 (err = inet_sk_reselect_saddr(sk)) != 0)
1321 sk->sk_err_soft = -err;
1322 }
1323
1324 return err;
1325 }
1326 EXPORT_SYMBOL(inet_sk_rebuild_header);
1327
inet_sk_set_state(struct sock * sk,int state)1328 void inet_sk_set_state(struct sock *sk, int state)
1329 {
1330 trace_inet_sock_set_state(sk, sk->sk_state, state);
1331 sk->sk_state = state;
1332 }
1333 EXPORT_SYMBOL(inet_sk_set_state);
1334
inet_sk_state_store(struct sock * sk,int newstate)1335 void inet_sk_state_store(struct sock *sk, int newstate)
1336 {
1337 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1338 smp_store_release(&sk->sk_state, newstate);
1339 }
1340
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1341 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1342 netdev_features_t features)
1343 {
1344 bool udpfrag = false, fixedid = false, gso_partial, encap;
1345 struct sk_buff *segs = ERR_PTR(-EINVAL);
1346 const struct net_offload *ops;
1347 unsigned int offset = 0;
1348 struct iphdr *iph;
1349 int proto, tot_len;
1350 int nhoff;
1351 int ihl;
1352 int id;
1353
1354 skb_reset_network_header(skb);
1355 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1356 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1357 goto out;
1358
1359 iph = ip_hdr(skb);
1360 ihl = iph->ihl * 4;
1361 if (ihl < sizeof(*iph))
1362 goto out;
1363
1364 id = ntohs(iph->id);
1365 proto = iph->protocol;
1366
1367 /* Warning: after this point, iph might be no longer valid */
1368 if (unlikely(!pskb_may_pull(skb, ihl)))
1369 goto out;
1370 __skb_pull(skb, ihl);
1371
1372 encap = SKB_GSO_CB(skb)->encap_level > 0;
1373 if (encap)
1374 features &= skb->dev->hw_enc_features;
1375 SKB_GSO_CB(skb)->encap_level += ihl;
1376
1377 skb_reset_transport_header(skb);
1378
1379 segs = ERR_PTR(-EPROTONOSUPPORT);
1380
1381 if (!skb->encapsulation || encap) {
1382 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1383 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1384
1385 /* fixed ID is invalid if DF bit is not set */
1386 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1387 goto out;
1388 }
1389
1390 ops = rcu_dereference(inet_offloads[proto]);
1391 if (likely(ops && ops->callbacks.gso_segment)) {
1392 segs = ops->callbacks.gso_segment(skb, features);
1393 if (!segs)
1394 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1395 }
1396
1397 if (IS_ERR_OR_NULL(segs))
1398 goto out;
1399
1400 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1401
1402 skb = segs;
1403 do {
1404 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1405 if (udpfrag) {
1406 iph->frag_off = htons(offset >> 3);
1407 if (skb->next)
1408 iph->frag_off |= htons(IP_MF);
1409 offset += skb->len - nhoff - ihl;
1410 tot_len = skb->len - nhoff;
1411 } else if (skb_is_gso(skb)) {
1412 if (!fixedid) {
1413 iph->id = htons(id);
1414 id += skb_shinfo(skb)->gso_segs;
1415 }
1416
1417 if (gso_partial)
1418 tot_len = skb_shinfo(skb)->gso_size +
1419 SKB_GSO_CB(skb)->data_offset +
1420 skb->head - (unsigned char *)iph;
1421 else
1422 tot_len = skb->len - nhoff;
1423 } else {
1424 if (!fixedid)
1425 iph->id = htons(id++);
1426 tot_len = skb->len - nhoff;
1427 }
1428 iph->tot_len = htons(tot_len);
1429 ip_send_check(iph);
1430 if (encap)
1431 skb_reset_inner_headers(skb);
1432 skb->network_header = (u8 *)iph - skb->head;
1433 skb_reset_mac_len(skb);
1434 } while ((skb = skb->next));
1435
1436 out:
1437 return segs;
1438 }
1439 EXPORT_SYMBOL(inet_gso_segment);
1440
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1441 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1442 netdev_features_t features)
1443 {
1444 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1445 return ERR_PTR(-EINVAL);
1446
1447 return inet_gso_segment(skb, features);
1448 }
1449
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1450 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1451 {
1452 const struct net_offload *ops;
1453 struct sk_buff *pp = NULL;
1454 const struct iphdr *iph;
1455 struct sk_buff *p;
1456 unsigned int hlen;
1457 unsigned int off;
1458 unsigned int id;
1459 int flush = 1;
1460 int proto;
1461
1462 off = skb_gro_offset(skb);
1463 hlen = off + sizeof(*iph);
1464 iph = skb_gro_header_fast(skb, off);
1465 if (skb_gro_header_hard(skb, hlen)) {
1466 iph = skb_gro_header_slow(skb, hlen, off);
1467 if (unlikely(!iph))
1468 goto out;
1469 }
1470
1471 proto = iph->protocol;
1472
1473 rcu_read_lock();
1474 ops = rcu_dereference(inet_offloads[proto]);
1475 if (!ops || !ops->callbacks.gro_receive)
1476 goto out_unlock;
1477
1478 if (*(u8 *)iph != 0x45)
1479 goto out_unlock;
1480
1481 if (ip_is_fragment(iph))
1482 goto out_unlock;
1483
1484 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1485 goto out_unlock;
1486
1487 id = ntohl(*(__be32 *)&iph->id);
1488 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1489 id >>= 16;
1490
1491 list_for_each_entry(p, head, list) {
1492 struct iphdr *iph2;
1493 u16 flush_id;
1494
1495 if (!NAPI_GRO_CB(p)->same_flow)
1496 continue;
1497
1498 iph2 = (struct iphdr *)(p->data + off);
1499 /* The above works because, with the exception of the top
1500 * (inner most) layer, we only aggregate pkts with the same
1501 * hdr length so all the hdrs we'll need to verify will start
1502 * at the same offset.
1503 */
1504 if ((iph->protocol ^ iph2->protocol) |
1505 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1506 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1507 NAPI_GRO_CB(p)->same_flow = 0;
1508 continue;
1509 }
1510
1511 /* All fields must match except length and checksum. */
1512 NAPI_GRO_CB(p)->flush |=
1513 (iph->ttl ^ iph2->ttl) |
1514 (iph->tos ^ iph2->tos) |
1515 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1516
1517 NAPI_GRO_CB(p)->flush |= flush;
1518
1519 /* We need to store of the IP ID check to be included later
1520 * when we can verify that this packet does in fact belong
1521 * to a given flow.
1522 */
1523 flush_id = (u16)(id - ntohs(iph2->id));
1524
1525 /* This bit of code makes it much easier for us to identify
1526 * the cases where we are doing atomic vs non-atomic IP ID
1527 * checks. Specifically an atomic check can return IP ID
1528 * values 0 - 0xFFFF, while a non-atomic check can only
1529 * return 0 or 0xFFFF.
1530 */
1531 if (!NAPI_GRO_CB(p)->is_atomic ||
1532 !(iph->frag_off & htons(IP_DF))) {
1533 flush_id ^= NAPI_GRO_CB(p)->count;
1534 flush_id = flush_id ? 0xFFFF : 0;
1535 }
1536
1537 /* If the previous IP ID value was based on an atomic
1538 * datagram we can overwrite the value and ignore it.
1539 */
1540 if (NAPI_GRO_CB(skb)->is_atomic)
1541 NAPI_GRO_CB(p)->flush_id = flush_id;
1542 else
1543 NAPI_GRO_CB(p)->flush_id |= flush_id;
1544 }
1545
1546 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1547 NAPI_GRO_CB(skb)->flush |= flush;
1548 skb_set_network_header(skb, off);
1549 /* The above will be needed by the transport layer if there is one
1550 * immediately following this IP hdr.
1551 */
1552
1553 /* Note : No need to call skb_gro_postpull_rcsum() here,
1554 * as we already checked checksum over ipv4 header was 0
1555 */
1556 skb_gro_pull(skb, sizeof(*iph));
1557 skb_set_transport_header(skb, skb_gro_offset(skb));
1558
1559 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1560 ops->callbacks.gro_receive, head, skb);
1561
1562 out_unlock:
1563 rcu_read_unlock();
1564
1565 out:
1566 skb_gro_flush_final(skb, pp, flush);
1567
1568 return pp;
1569 }
1570 EXPORT_SYMBOL(inet_gro_receive);
1571
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1572 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1573 struct sk_buff *skb)
1574 {
1575 if (NAPI_GRO_CB(skb)->encap_mark) {
1576 NAPI_GRO_CB(skb)->flush = 1;
1577 return NULL;
1578 }
1579
1580 NAPI_GRO_CB(skb)->encap_mark = 1;
1581
1582 return inet_gro_receive(head, skb);
1583 }
1584
1585 #define SECONDS_PER_DAY 86400
1586
1587 /* inet_current_timestamp - Return IP network timestamp
1588 *
1589 * Return milliseconds since midnight in network byte order.
1590 */
inet_current_timestamp(void)1591 __be32 inet_current_timestamp(void)
1592 {
1593 u32 secs;
1594 u32 msecs;
1595 struct timespec64 ts;
1596
1597 ktime_get_real_ts64(&ts);
1598
1599 /* Get secs since midnight. */
1600 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1601 /* Convert to msecs. */
1602 msecs = secs * MSEC_PER_SEC;
1603 /* Convert nsec to msec. */
1604 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1605
1606 /* Convert to network byte order. */
1607 return htonl(msecs);
1608 }
1609 EXPORT_SYMBOL(inet_current_timestamp);
1610
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1611 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1612 {
1613 unsigned int family = READ_ONCE(sk->sk_family);
1614
1615 if (family == AF_INET)
1616 return ip_recv_error(sk, msg, len, addr_len);
1617 #if IS_ENABLED(CONFIG_IPV6)
1618 if (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