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/gro.h>
103 #include <net/gso.h>
104 #include <net/tcp.h>
105 #include <net/udp.h>
106 #include <net/udplite.h>
107 #include <net/ping.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
110 #include <net/raw.h>
111 #include <net/icmp.h>
112 #include <net/inet_common.h>
113 #include <net/ip_tunnels.h>
114 #include <net/xfrm.h>
115 #include <net/net_namespace.h>
116 #include <net/secure_seq.h>
117 #ifdef CONFIG_IP_MROUTE
118 #include <linux/mroute.h>
119 #endif
120 #include <net/l3mdev.h>
121 #include <net/compat.h>
122
123 #include <trace/events/sock.h>
124
125 /* The inetsw table contains everything that inet_create needs to
126 * build a new socket.
127 */
128 static struct list_head inetsw[SOCK_MAX];
129 static DEFINE_SPINLOCK(inetsw_lock);
130
131 /* New destruction routine */
132
inet_sock_destruct(struct sock * sk)133 void inet_sock_destruct(struct sock *sk)
134 {
135 struct inet_sock *inet = inet_sk(sk);
136
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
139
140 sk_mem_reclaim_final(sk);
141
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 pr_err("Attempt to release TCP socket in state %d %p\n",
144 sk->sk_state, sk);
145 return;
146 }
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 pr_err("Attempt to release alive inet socket %p\n", sk);
149 return;
150 }
151
152 WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc));
153 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
154 WARN_ON_ONCE(sk->sk_wmem_queued);
155 WARN_ON_ONCE(sk_forward_alloc_get(sk));
156
157 kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
159 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
160 }
161 EXPORT_SYMBOL(inet_sock_destruct);
162
163 /*
164 * The routines beyond this point handle the behaviour of an AF_INET
165 * socket object. Mostly it punts to the subprotocols of IP to do
166 * the work.
167 */
168
169 /*
170 * Automatically bind an unbound socket.
171 */
172
inet_autobind(struct sock * sk)173 static int inet_autobind(struct sock *sk)
174 {
175 struct inet_sock *inet;
176 /* We may need to bind the socket. */
177 lock_sock(sk);
178 inet = inet_sk(sk);
179 if (!inet->inet_num) {
180 if (sk->sk_prot->get_port(sk, 0)) {
181 release_sock(sk);
182 return -EAGAIN;
183 }
184 inet->inet_sport = htons(inet->inet_num);
185 }
186 release_sock(sk);
187 return 0;
188 }
189
__inet_listen_sk(struct sock * sk,int backlog)190 int __inet_listen_sk(struct sock *sk, int backlog)
191 {
192 unsigned char old_state = sk->sk_state;
193 int err, tcp_fastopen;
194
195 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
196 return -EINVAL;
197
198 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
199 /* Really, if the socket is already in listen state
200 * we can only allow the backlog to be adjusted.
201 */
202 if (old_state != TCP_LISTEN) {
203 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
204 * Note that only TCP sockets (SOCK_STREAM) will reach here.
205 * Also fastopen backlog may already been set via the option
206 * because the socket was in TCP_LISTEN state previously but
207 * was shutdown() rather than close().
208 */
209 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
210 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
211 (tcp_fastopen & TFO_SERVER_ENABLE) &&
212 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
213 fastopen_queue_tune(sk, backlog);
214 tcp_fastopen_init_key_once(sock_net(sk));
215 }
216
217 err = inet_csk_listen_start(sk);
218 if (err)
219 return err;
220
221 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
222 }
223 return 0;
224 }
225
226 /*
227 * Move a socket into listening state.
228 */
inet_listen(struct socket * sock,int backlog)229 int inet_listen(struct socket *sock, int backlog)
230 {
231 struct sock *sk = sock->sk;
232 int err = -EINVAL;
233
234 lock_sock(sk);
235
236 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
237 goto out;
238
239 err = __inet_listen_sk(sk, backlog);
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 = -ENOMEM;
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_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
338
339 inet_clear_bit(NODEFRAG, sk);
340
341 if (SOCK_RAW == sock->type) {
342 inet->inet_num = protocol;
343 if (IPPROTO_RAW == protocol)
344 inet_set_bit(HDRINCL, sk);
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 atomic_set(&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 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
360
361 inet->uc_ttl = -1;
362 inet_set_bit(MC_LOOP, sk);
363 inet->mc_ttl = 1;
364 inet_set_bit(MC_ALL, sk);
365 inet->mc_index = 0;
366 inet->mc_list = NULL;
367 inet->rcv_tos = 0;
368
369 if (inet->inet_num) {
370 /* It assumes that any protocol which allows
371 * the user to assign a number at socket
372 * creation time automatically
373 * shares.
374 */
375 inet->inet_sport = htons(inet->inet_num);
376 /* Add to protocol hash chains. */
377 err = sk->sk_prot->hash(sk);
378 if (err) {
379 sk_common_release(sk);
380 goto out;
381 }
382 }
383
384 if (sk->sk_prot->init) {
385 err = sk->sk_prot->init(sk);
386 if (err) {
387 sk_common_release(sk);
388 goto out;
389 }
390 }
391
392 if (!kern) {
393 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
394 if (err) {
395 sk_common_release(sk);
396 goto out;
397 }
398 }
399 out:
400 return err;
401 out_rcu_unlock:
402 rcu_read_unlock();
403 goto out;
404 }
405
406
407 /*
408 * The peer socket should always be NULL (or else). When we call this
409 * function we are destroying the object and from then on nobody
410 * should refer to it.
411 */
inet_release(struct socket * sock)412 int inet_release(struct socket *sock)
413 {
414 struct sock *sk = sock->sk;
415
416 if (sk) {
417 long timeout;
418
419 if (!sk->sk_kern_sock)
420 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
421
422 /* Applications forget to leave groups before exiting */
423 ip_mc_drop_socket(sk);
424
425 /* If linger is set, we don't return until the close
426 * is complete. Otherwise we return immediately. The
427 * actually closing is done the same either way.
428 *
429 * If the close is due to the process exiting, we never
430 * linger..
431 */
432 timeout = 0;
433 if (sock_flag(sk, SOCK_LINGER) &&
434 !(current->flags & PF_EXITING))
435 timeout = sk->sk_lingertime;
436 sk->sk_prot->close(sk, timeout);
437 sock->sk = NULL;
438 }
439 return 0;
440 }
441 EXPORT_SYMBOL(inet_release);
442
inet_bind_sk(struct sock * sk,struct sockaddr * uaddr,int addr_len)443 int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
444 {
445 u32 flags = BIND_WITH_LOCK;
446 int err;
447
448 /* If the socket has its own bind function then use it. (RAW) */
449 if (sk->sk_prot->bind) {
450 return sk->sk_prot->bind(sk, uaddr, addr_len);
451 }
452 if (addr_len < sizeof(struct sockaddr_in))
453 return -EINVAL;
454
455 /* BPF prog is run before any checks are done so that if the prog
456 * changes context in a wrong way it will be caught.
457 */
458 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
459 CGROUP_INET4_BIND, &flags);
460 if (err)
461 return err;
462
463 return __inet_bind(sk, uaddr, addr_len, flags);
464 }
465
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)466 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
467 {
468 return inet_bind_sk(sock->sk, uaddr, addr_len);
469 }
470 EXPORT_SYMBOL(inet_bind);
471
__inet_bind(struct sock * sk,struct sockaddr * uaddr,int addr_len,u32 flags)472 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
473 u32 flags)
474 {
475 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
476 struct inet_sock *inet = inet_sk(sk);
477 struct net *net = sock_net(sk);
478 unsigned short snum;
479 int chk_addr_ret;
480 u32 tb_id = RT_TABLE_LOCAL;
481 int err;
482
483 if (addr->sin_family != AF_INET) {
484 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
485 * only if s_addr is INADDR_ANY.
486 */
487 err = -EAFNOSUPPORT;
488 if (addr->sin_family != AF_UNSPEC ||
489 addr->sin_addr.s_addr != htonl(INADDR_ANY))
490 goto out;
491 }
492
493 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
494 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
495
496 /* Not specified by any standard per-se, however it breaks too
497 * many applications when removed. It is unfortunate since
498 * allowing applications to make a non-local bind solves
499 * several problems with systems using dynamic addressing.
500 * (ie. your servers still start up even if your ISDN link
501 * is temporarily down)
502 */
503 err = -EADDRNOTAVAIL;
504 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
505 chk_addr_ret))
506 goto out;
507
508 snum = ntohs(addr->sin_port);
509 err = -EACCES;
510 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
511 snum && inet_port_requires_bind_service(net, snum) &&
512 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
513 goto out;
514
515 /* We keep a pair of addresses. rcv_saddr is the one
516 * used by hash lookups, and saddr is used for transmit.
517 *
518 * In the BSD API these are the same except where it
519 * would be illegal to use them (multicast/broadcast) in
520 * which case the sending device address is used.
521 */
522 if (flags & BIND_WITH_LOCK)
523 lock_sock(sk);
524
525 /* Check these errors (active socket, double bind). */
526 err = -EINVAL;
527 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
528 goto out_release_sock;
529
530 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
531 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
532 inet->inet_saddr = 0; /* Use device */
533
534 /* Make sure we are allowed to bind here. */
535 if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
536 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
537 err = sk->sk_prot->get_port(sk, snum);
538 if (err) {
539 inet->inet_saddr = inet->inet_rcv_saddr = 0;
540 goto out_release_sock;
541 }
542 if (!(flags & BIND_FROM_BPF)) {
543 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
544 if (err) {
545 inet->inet_saddr = inet->inet_rcv_saddr = 0;
546 if (sk->sk_prot->put_port)
547 sk->sk_prot->put_port(sk);
548 goto out_release_sock;
549 }
550 }
551 }
552
553 if (inet->inet_rcv_saddr)
554 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
555 if (snum)
556 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
557 inet->inet_sport = htons(inet->inet_num);
558 inet->inet_daddr = 0;
559 inet->inet_dport = 0;
560 sk_dst_reset(sk);
561 err = 0;
562 out_release_sock:
563 if (flags & BIND_WITH_LOCK)
564 release_sock(sk);
565 out:
566 return err;
567 }
568
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)569 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
570 int addr_len, int flags)
571 {
572 struct sock *sk = sock->sk;
573 const struct proto *prot;
574 int err;
575
576 if (addr_len < sizeof(uaddr->sa_family))
577 return -EINVAL;
578
579 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
580 prot = READ_ONCE(sk->sk_prot);
581
582 if (uaddr->sa_family == AF_UNSPEC)
583 return prot->disconnect(sk, flags);
584
585 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
586 err = prot->pre_connect(sk, uaddr, addr_len);
587 if (err)
588 return err;
589 }
590
591 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
592 return -EAGAIN;
593 return prot->connect(sk, uaddr, addr_len);
594 }
595 EXPORT_SYMBOL(inet_dgram_connect);
596
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)597 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
598 {
599 DEFINE_WAIT_FUNC(wait, woken_wake_function);
600
601 add_wait_queue(sk_sleep(sk), &wait);
602 sk->sk_write_pending += writebias;
603
604 /* Basic assumption: if someone sets sk->sk_err, he _must_
605 * change state of the socket from TCP_SYN_*.
606 * Connect() does not allow to get error notifications
607 * without closing the socket.
608 */
609 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
610 release_sock(sk);
611 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
612 lock_sock(sk);
613 if (signal_pending(current) || !timeo)
614 break;
615 }
616 remove_wait_queue(sk_sleep(sk), &wait);
617 sk->sk_write_pending -= writebias;
618 return timeo;
619 }
620
621 /*
622 * Connect to a remote host. There is regrettably still a little
623 * TCP 'magic' in here.
624 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags,int is_sendmsg)625 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
626 int addr_len, int flags, int is_sendmsg)
627 {
628 struct sock *sk = sock->sk;
629 int err;
630 long timeo;
631
632 /*
633 * uaddr can be NULL and addr_len can be 0 if:
634 * sk is a TCP fastopen active socket and
635 * TCP_FASTOPEN_CONNECT sockopt is set and
636 * we already have a valid cookie for this socket.
637 * In this case, user can call write() after connect().
638 * write() will invoke tcp_sendmsg_fastopen() which calls
639 * __inet_stream_connect().
640 */
641 if (uaddr) {
642 if (addr_len < sizeof(uaddr->sa_family))
643 return -EINVAL;
644
645 if (uaddr->sa_family == AF_UNSPEC) {
646 sk->sk_disconnects++;
647 err = sk->sk_prot->disconnect(sk, flags);
648 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
649 goto out;
650 }
651 }
652
653 switch (sock->state) {
654 default:
655 err = -EINVAL;
656 goto out;
657 case SS_CONNECTED:
658 err = -EISCONN;
659 goto out;
660 case SS_CONNECTING:
661 if (inet_test_bit(DEFER_CONNECT, sk))
662 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
663 else
664 err = -EALREADY;
665 /* Fall out of switch with err, set for this state */
666 break;
667 case SS_UNCONNECTED:
668 err = -EISCONN;
669 if (sk->sk_state != TCP_CLOSE)
670 goto out;
671
672 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
673 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
674 if (err)
675 goto out;
676 }
677
678 err = sk->sk_prot->connect(sk, uaddr, addr_len);
679 if (err < 0)
680 goto out;
681
682 sock->state = SS_CONNECTING;
683
684 if (!err && inet_test_bit(DEFER_CONNECT, sk))
685 goto out;
686
687 /* Just entered SS_CONNECTING state; the only
688 * difference is that return value in non-blocking
689 * case is EINPROGRESS, rather than EALREADY.
690 */
691 err = -EINPROGRESS;
692 break;
693 }
694
695 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
696
697 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
698 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
699 tcp_sk(sk)->fastopen_req &&
700 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
701 int dis = sk->sk_disconnects;
702
703 /* Error code is set above */
704 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
705 goto out;
706
707 err = sock_intr_errno(timeo);
708 if (signal_pending(current))
709 goto out;
710
711 if (dis != sk->sk_disconnects) {
712 err = -EPIPE;
713 goto out;
714 }
715 }
716
717 /* Connection was closed by RST, timeout, ICMP error
718 * or another process disconnected us.
719 */
720 if (sk->sk_state == TCP_CLOSE)
721 goto sock_error;
722
723 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
724 * and error was received after socket entered established state.
725 * Hence, it is handled normally after connect() return successfully.
726 */
727
728 sock->state = SS_CONNECTED;
729 err = 0;
730 out:
731 return err;
732
733 sock_error:
734 err = sock_error(sk) ? : -ECONNABORTED;
735 sock->state = SS_UNCONNECTED;
736 sk->sk_disconnects++;
737 if (sk->sk_prot->disconnect(sk, flags))
738 sock->state = SS_DISCONNECTING;
739 goto out;
740 }
741 EXPORT_SYMBOL(__inet_stream_connect);
742
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)743 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
744 int addr_len, int flags)
745 {
746 int err;
747
748 lock_sock(sock->sk);
749 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
750 release_sock(sock->sk);
751 return err;
752 }
753 EXPORT_SYMBOL(inet_stream_connect);
754
__inet_accept(struct socket * sock,struct socket * newsock,struct sock * newsk)755 void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
756 {
757 sock_rps_record_flow(newsk);
758 WARN_ON(!((1 << newsk->sk_state) &
759 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
760 TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
761 TCPF_CLOSING | TCPF_CLOSE_WAIT |
762 TCPF_CLOSE)));
763
764 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
765 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
766 sock_graft(newsk, newsock);
767
768 newsock->state = SS_CONNECTED;
769 }
770
771 /*
772 * Accept a pending connection. The TCP layer now gives BSD semantics.
773 */
774
inet_accept(struct socket * sock,struct socket * newsock,int flags,bool kern)775 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
776 bool kern)
777 {
778 struct sock *sk1 = sock->sk, *sk2;
779 int err = -EINVAL;
780
781 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
782 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
783 if (!sk2)
784 return err;
785
786 lock_sock(sk2);
787 __inet_accept(sock, newsock, sk2);
788 release_sock(sk2);
789 return 0;
790 }
791 EXPORT_SYMBOL(inet_accept);
792
793 /*
794 * This does both peername and sockname.
795 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int peer)796 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
797 int peer)
798 {
799 struct sock *sk = sock->sk;
800 struct inet_sock *inet = inet_sk(sk);
801 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
802 int sin_addr_len = sizeof(*sin);
803
804 sin->sin_family = AF_INET;
805 lock_sock(sk);
806 if (peer) {
807 if (!inet->inet_dport ||
808 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
809 peer == 1)) {
810 release_sock(sk);
811 return -ENOTCONN;
812 }
813 sin->sin_port = inet->inet_dport;
814 sin->sin_addr.s_addr = inet->inet_daddr;
815 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
816 CGROUP_INET4_GETPEERNAME);
817 } else {
818 __be32 addr = inet->inet_rcv_saddr;
819 if (!addr)
820 addr = inet->inet_saddr;
821 sin->sin_port = inet->inet_sport;
822 sin->sin_addr.s_addr = addr;
823 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
824 CGROUP_INET4_GETSOCKNAME);
825 }
826 release_sock(sk);
827 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
828 return sin_addr_len;
829 }
830 EXPORT_SYMBOL(inet_getname);
831
inet_send_prepare(struct sock * sk)832 int inet_send_prepare(struct sock *sk)
833 {
834 sock_rps_record_flow(sk);
835
836 /* We may need to bind the socket. */
837 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
838 inet_autobind(sk))
839 return -EAGAIN;
840
841 return 0;
842 }
843 EXPORT_SYMBOL_GPL(inet_send_prepare);
844
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)845 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
846 {
847 struct sock *sk = sock->sk;
848
849 if (unlikely(inet_send_prepare(sk)))
850 return -EAGAIN;
851
852 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
853 sk, msg, size);
854 }
855 EXPORT_SYMBOL(inet_sendmsg);
856
inet_splice_eof(struct socket * sock)857 void inet_splice_eof(struct socket *sock)
858 {
859 const struct proto *prot;
860 struct sock *sk = sock->sk;
861
862 if (unlikely(inet_send_prepare(sk)))
863 return;
864
865 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
866 prot = READ_ONCE(sk->sk_prot);
867 if (prot->splice_eof)
868 prot->splice_eof(sock);
869 }
870 EXPORT_SYMBOL_GPL(inet_splice_eof);
871
872 INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
873 size_t, int, int *));
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)874 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
875 int flags)
876 {
877 struct sock *sk = sock->sk;
878 int addr_len = 0;
879 int err;
880
881 if (likely(!(flags & MSG_ERRQUEUE)))
882 sock_rps_record_flow(sk);
883
884 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
885 sk, msg, size, flags, &addr_len);
886 if (err >= 0)
887 msg->msg_namelen = addr_len;
888 return err;
889 }
890 EXPORT_SYMBOL(inet_recvmsg);
891
inet_shutdown(struct socket * sock,int how)892 int inet_shutdown(struct socket *sock, int how)
893 {
894 struct sock *sk = sock->sk;
895 int err = 0;
896
897 /* This should really check to make sure
898 * the socket is a TCP socket. (WHY AC...)
899 */
900 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
901 1->2 bit 2 snds.
902 2->3 */
903 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
904 return -EINVAL;
905
906 lock_sock(sk);
907 if (sock->state == SS_CONNECTING) {
908 if ((1 << sk->sk_state) &
909 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
910 sock->state = SS_DISCONNECTING;
911 else
912 sock->state = SS_CONNECTED;
913 }
914
915 switch (sk->sk_state) {
916 case TCP_CLOSE:
917 err = -ENOTCONN;
918 /* Hack to wake up other listeners, who can poll for
919 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
920 fallthrough;
921 default:
922 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
923 if (sk->sk_prot->shutdown)
924 sk->sk_prot->shutdown(sk, how);
925 break;
926
927 /* Remaining two branches are temporary solution for missing
928 * close() in multithreaded environment. It is _not_ a good idea,
929 * but we have no choice until close() is repaired at VFS level.
930 */
931 case TCP_LISTEN:
932 if (!(how & RCV_SHUTDOWN))
933 break;
934 fallthrough;
935 case TCP_SYN_SENT:
936 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
937 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
938 break;
939 }
940
941 /* Wake up anyone sleeping in poll. */
942 sk->sk_state_change(sk);
943 release_sock(sk);
944 return err;
945 }
946 EXPORT_SYMBOL(inet_shutdown);
947
948 /*
949 * ioctl() calls you can issue on an INET socket. Most of these are
950 * device configuration and stuff and very rarely used. Some ioctls
951 * pass on to the socket itself.
952 *
953 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
954 * loads the devconfigure module does its configuring and unloads it.
955 * There's a good 20K of config code hanging around the kernel.
956 */
957
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)958 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
959 {
960 struct sock *sk = sock->sk;
961 int err = 0;
962 struct net *net = sock_net(sk);
963 void __user *p = (void __user *)arg;
964 struct ifreq ifr;
965 struct rtentry rt;
966
967 switch (cmd) {
968 case SIOCADDRT:
969 case SIOCDELRT:
970 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
971 return -EFAULT;
972 err = ip_rt_ioctl(net, cmd, &rt);
973 break;
974 case SIOCRTMSG:
975 err = -EINVAL;
976 break;
977 case SIOCDARP:
978 case SIOCGARP:
979 case SIOCSARP:
980 err = arp_ioctl(net, cmd, (void __user *)arg);
981 break;
982 case SIOCGIFADDR:
983 case SIOCGIFBRDADDR:
984 case SIOCGIFNETMASK:
985 case SIOCGIFDSTADDR:
986 case SIOCGIFPFLAGS:
987 if (get_user_ifreq(&ifr, NULL, p))
988 return -EFAULT;
989 err = devinet_ioctl(net, cmd, &ifr);
990 if (!err && put_user_ifreq(&ifr, p))
991 err = -EFAULT;
992 break;
993
994 case SIOCSIFADDR:
995 case SIOCSIFBRDADDR:
996 case SIOCSIFNETMASK:
997 case SIOCSIFDSTADDR:
998 case SIOCSIFPFLAGS:
999 case SIOCSIFFLAGS:
1000 if (get_user_ifreq(&ifr, NULL, p))
1001 return -EFAULT;
1002 err = devinet_ioctl(net, cmd, &ifr);
1003 break;
1004 default:
1005 if (sk->sk_prot->ioctl)
1006 err = sk_ioctl(sk, cmd, (void __user *)arg);
1007 else
1008 err = -ENOIOCTLCMD;
1009 break;
1010 }
1011 return err;
1012 }
1013 EXPORT_SYMBOL(inet_ioctl);
1014
1015 #ifdef CONFIG_COMPAT
inet_compat_routing_ioctl(struct sock * sk,unsigned int cmd,struct compat_rtentry __user * ur)1016 static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1017 struct compat_rtentry __user *ur)
1018 {
1019 compat_uptr_t rtdev;
1020 struct rtentry rt;
1021
1022 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1023 3 * sizeof(struct sockaddr)) ||
1024 get_user(rt.rt_flags, &ur->rt_flags) ||
1025 get_user(rt.rt_metric, &ur->rt_metric) ||
1026 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1027 get_user(rt.rt_window, &ur->rt_window) ||
1028 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1029 get_user(rtdev, &ur->rt_dev))
1030 return -EFAULT;
1031
1032 rt.rt_dev = compat_ptr(rtdev);
1033 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1034 }
1035
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1036 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1037 {
1038 void __user *argp = compat_ptr(arg);
1039 struct sock *sk = sock->sk;
1040
1041 switch (cmd) {
1042 case SIOCADDRT:
1043 case SIOCDELRT:
1044 return inet_compat_routing_ioctl(sk, cmd, argp);
1045 default:
1046 if (!sk->sk_prot->compat_ioctl)
1047 return -ENOIOCTLCMD;
1048 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1049 }
1050 }
1051 #endif /* CONFIG_COMPAT */
1052
1053 const struct proto_ops inet_stream_ops = {
1054 .family = PF_INET,
1055 .owner = THIS_MODULE,
1056 .release = inet_release,
1057 .bind = inet_bind,
1058 .connect = inet_stream_connect,
1059 .socketpair = sock_no_socketpair,
1060 .accept = inet_accept,
1061 .getname = inet_getname,
1062 .poll = tcp_poll,
1063 .ioctl = inet_ioctl,
1064 .gettstamp = sock_gettstamp,
1065 .listen = inet_listen,
1066 .shutdown = inet_shutdown,
1067 .setsockopt = sock_common_setsockopt,
1068 .getsockopt = sock_common_getsockopt,
1069 .sendmsg = inet_sendmsg,
1070 .recvmsg = inet_recvmsg,
1071 #ifdef CONFIG_MMU
1072 .mmap = tcp_mmap,
1073 #endif
1074 .splice_eof = inet_splice_eof,
1075 .splice_read = tcp_splice_read,
1076 .read_sock = tcp_read_sock,
1077 .read_skb = tcp_read_skb,
1078 .sendmsg_locked = tcp_sendmsg_locked,
1079 .peek_len = tcp_peek_len,
1080 #ifdef CONFIG_COMPAT
1081 .compat_ioctl = inet_compat_ioctl,
1082 #endif
1083 .set_rcvlowat = tcp_set_rcvlowat,
1084 };
1085 EXPORT_SYMBOL(inet_stream_ops);
1086
1087 const struct proto_ops inet_dgram_ops = {
1088 .family = PF_INET,
1089 .owner = THIS_MODULE,
1090 .release = inet_release,
1091 .bind = inet_bind,
1092 .connect = inet_dgram_connect,
1093 .socketpair = sock_no_socketpair,
1094 .accept = sock_no_accept,
1095 .getname = inet_getname,
1096 .poll = udp_poll,
1097 .ioctl = inet_ioctl,
1098 .gettstamp = sock_gettstamp,
1099 .listen = sock_no_listen,
1100 .shutdown = inet_shutdown,
1101 .setsockopt = sock_common_setsockopt,
1102 .getsockopt = sock_common_getsockopt,
1103 .sendmsg = inet_sendmsg,
1104 .read_skb = udp_read_skb,
1105 .recvmsg = inet_recvmsg,
1106 .mmap = sock_no_mmap,
1107 .splice_eof = inet_splice_eof,
1108 .set_peek_off = sk_set_peek_off,
1109 #ifdef CONFIG_COMPAT
1110 .compat_ioctl = inet_compat_ioctl,
1111 #endif
1112 };
1113 EXPORT_SYMBOL(inet_dgram_ops);
1114
1115 /*
1116 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1117 * udp_poll
1118 */
1119 static const struct proto_ops inet_sockraw_ops = {
1120 .family = PF_INET,
1121 .owner = THIS_MODULE,
1122 .release = inet_release,
1123 .bind = inet_bind,
1124 .connect = inet_dgram_connect,
1125 .socketpair = sock_no_socketpair,
1126 .accept = sock_no_accept,
1127 .getname = inet_getname,
1128 .poll = datagram_poll,
1129 .ioctl = inet_ioctl,
1130 .gettstamp = sock_gettstamp,
1131 .listen = sock_no_listen,
1132 .shutdown = inet_shutdown,
1133 .setsockopt = sock_common_setsockopt,
1134 .getsockopt = sock_common_getsockopt,
1135 .sendmsg = inet_sendmsg,
1136 .recvmsg = inet_recvmsg,
1137 .mmap = sock_no_mmap,
1138 .splice_eof = inet_splice_eof,
1139 #ifdef CONFIG_COMPAT
1140 .compat_ioctl = inet_compat_ioctl,
1141 #endif
1142 };
1143
1144 static const struct net_proto_family inet_family_ops = {
1145 .family = PF_INET,
1146 .create = inet_create,
1147 .owner = THIS_MODULE,
1148 };
1149
1150 /* Upon startup we insert all the elements in inetsw_array[] into
1151 * the linked list inetsw.
1152 */
1153 static struct inet_protosw inetsw_array[] =
1154 {
1155 {
1156 .type = SOCK_STREAM,
1157 .protocol = IPPROTO_TCP,
1158 .prot = &tcp_prot,
1159 .ops = &inet_stream_ops,
1160 .flags = INET_PROTOSW_PERMANENT |
1161 INET_PROTOSW_ICSK,
1162 },
1163
1164 {
1165 .type = SOCK_DGRAM,
1166 .protocol = IPPROTO_UDP,
1167 .prot = &udp_prot,
1168 .ops = &inet_dgram_ops,
1169 .flags = INET_PROTOSW_PERMANENT,
1170 },
1171
1172 {
1173 .type = SOCK_DGRAM,
1174 .protocol = IPPROTO_ICMP,
1175 .prot = &ping_prot,
1176 .ops = &inet_sockraw_ops,
1177 .flags = INET_PROTOSW_REUSE,
1178 },
1179
1180 {
1181 .type = SOCK_RAW,
1182 .protocol = IPPROTO_IP, /* wild card */
1183 .prot = &raw_prot,
1184 .ops = &inet_sockraw_ops,
1185 .flags = INET_PROTOSW_REUSE,
1186 }
1187 };
1188
1189 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1190
inet_register_protosw(struct inet_protosw * p)1191 void inet_register_protosw(struct inet_protosw *p)
1192 {
1193 struct list_head *lh;
1194 struct inet_protosw *answer;
1195 int protocol = p->protocol;
1196 struct list_head *last_perm;
1197
1198 spin_lock_bh(&inetsw_lock);
1199
1200 if (p->type >= SOCK_MAX)
1201 goto out_illegal;
1202
1203 /* If we are trying to override a permanent protocol, bail. */
1204 last_perm = &inetsw[p->type];
1205 list_for_each(lh, &inetsw[p->type]) {
1206 answer = list_entry(lh, struct inet_protosw, list);
1207 /* Check only the non-wild match. */
1208 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1209 break;
1210 if (protocol == answer->protocol)
1211 goto out_permanent;
1212 last_perm = lh;
1213 }
1214
1215 /* Add the new entry after the last permanent entry if any, so that
1216 * the new entry does not override a permanent entry when matched with
1217 * a wild-card protocol. But it is allowed to override any existing
1218 * non-permanent entry. This means that when we remove this entry, the
1219 * system automatically returns to the old behavior.
1220 */
1221 list_add_rcu(&p->list, last_perm);
1222 out:
1223 spin_unlock_bh(&inetsw_lock);
1224
1225 return;
1226
1227 out_permanent:
1228 pr_err("Attempt to override permanent protocol %d\n", protocol);
1229 goto out;
1230
1231 out_illegal:
1232 pr_err("Ignoring attempt to register invalid socket type %d\n",
1233 p->type);
1234 goto out;
1235 }
1236 EXPORT_SYMBOL(inet_register_protosw);
1237
inet_unregister_protosw(struct inet_protosw * p)1238 void inet_unregister_protosw(struct inet_protosw *p)
1239 {
1240 if (INET_PROTOSW_PERMANENT & p->flags) {
1241 pr_err("Attempt to unregister permanent protocol %d\n",
1242 p->protocol);
1243 } else {
1244 spin_lock_bh(&inetsw_lock);
1245 list_del_rcu(&p->list);
1246 spin_unlock_bh(&inetsw_lock);
1247
1248 synchronize_net();
1249 }
1250 }
1251 EXPORT_SYMBOL(inet_unregister_protosw);
1252
inet_sk_reselect_saddr(struct sock * sk)1253 static int inet_sk_reselect_saddr(struct sock *sk)
1254 {
1255 struct inet_sock *inet = inet_sk(sk);
1256 __be32 old_saddr = inet->inet_saddr;
1257 __be32 daddr = inet->inet_daddr;
1258 struct flowi4 *fl4;
1259 struct rtable *rt;
1260 __be32 new_saddr;
1261 struct ip_options_rcu *inet_opt;
1262 int err;
1263
1264 inet_opt = rcu_dereference_protected(inet->inet_opt,
1265 lockdep_sock_is_held(sk));
1266 if (inet_opt && inet_opt->opt.srr)
1267 daddr = inet_opt->opt.faddr;
1268
1269 /* Query new route. */
1270 fl4 = &inet->cork.fl.u.ip4;
1271 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1272 sk->sk_protocol, inet->inet_sport,
1273 inet->inet_dport, sk);
1274 if (IS_ERR(rt))
1275 return PTR_ERR(rt);
1276
1277 new_saddr = fl4->saddr;
1278
1279 if (new_saddr == old_saddr) {
1280 sk_setup_caps(sk, &rt->dst);
1281 return 0;
1282 }
1283
1284 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1285 if (err) {
1286 ip_rt_put(rt);
1287 return err;
1288 }
1289
1290 sk_setup_caps(sk, &rt->dst);
1291
1292 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1293 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1294 __func__, &old_saddr, &new_saddr);
1295 }
1296
1297 /*
1298 * XXX The only one ugly spot where we need to
1299 * XXX really change the sockets identity after
1300 * XXX it has entered the hashes. -DaveM
1301 *
1302 * Besides that, it does not check for connection
1303 * uniqueness. Wait for troubles.
1304 */
1305 return __sk_prot_rehash(sk);
1306 }
1307
inet_sk_rebuild_header(struct sock * sk)1308 int inet_sk_rebuild_header(struct sock *sk)
1309 {
1310 struct inet_sock *inet = inet_sk(sk);
1311 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1312 __be32 daddr;
1313 struct ip_options_rcu *inet_opt;
1314 struct flowi4 *fl4;
1315 int err;
1316
1317 /* Route is OK, nothing to do. */
1318 if (rt)
1319 return 0;
1320
1321 /* Reroute. */
1322 rcu_read_lock();
1323 inet_opt = rcu_dereference(inet->inet_opt);
1324 daddr = inet->inet_daddr;
1325 if (inet_opt && inet_opt->opt.srr)
1326 daddr = inet_opt->opt.faddr;
1327 rcu_read_unlock();
1328 fl4 = &inet->cork.fl.u.ip4;
1329 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1330 inet->inet_dport, inet->inet_sport,
1331 sk->sk_protocol, RT_CONN_FLAGS(sk),
1332 sk->sk_bound_dev_if);
1333 if (!IS_ERR(rt)) {
1334 err = 0;
1335 sk_setup_caps(sk, &rt->dst);
1336 } else {
1337 err = PTR_ERR(rt);
1338
1339 /* Routing failed... */
1340 sk->sk_route_caps = 0;
1341 /*
1342 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1343 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1344 */
1345 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1346 sk->sk_state != TCP_SYN_SENT ||
1347 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1348 (err = inet_sk_reselect_saddr(sk)) != 0)
1349 WRITE_ONCE(sk->sk_err_soft, -err);
1350 }
1351
1352 return err;
1353 }
1354 EXPORT_SYMBOL(inet_sk_rebuild_header);
1355
inet_sk_set_state(struct sock * sk,int state)1356 void inet_sk_set_state(struct sock *sk, int state)
1357 {
1358 trace_inet_sock_set_state(sk, sk->sk_state, state);
1359 sk->sk_state = state;
1360 }
1361 EXPORT_SYMBOL(inet_sk_set_state);
1362
inet_sk_state_store(struct sock * sk,int newstate)1363 void inet_sk_state_store(struct sock *sk, int newstate)
1364 {
1365 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1366 smp_store_release(&sk->sk_state, newstate);
1367 }
1368
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1369 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1370 netdev_features_t features)
1371 {
1372 bool udpfrag = false, fixedid = false, gso_partial, encap;
1373 struct sk_buff *segs = ERR_PTR(-EINVAL);
1374 const struct net_offload *ops;
1375 unsigned int offset = 0;
1376 struct iphdr *iph;
1377 int proto, tot_len;
1378 int nhoff;
1379 int ihl;
1380 int id;
1381
1382 skb_reset_network_header(skb);
1383 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1384 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1385 goto out;
1386
1387 iph = ip_hdr(skb);
1388 ihl = iph->ihl * 4;
1389 if (ihl < sizeof(*iph))
1390 goto out;
1391
1392 id = ntohs(iph->id);
1393 proto = iph->protocol;
1394
1395 /* Warning: after this point, iph might be no longer valid */
1396 if (unlikely(!pskb_may_pull(skb, ihl)))
1397 goto out;
1398 __skb_pull(skb, ihl);
1399
1400 encap = SKB_GSO_CB(skb)->encap_level > 0;
1401 if (encap)
1402 features &= skb->dev->hw_enc_features;
1403 SKB_GSO_CB(skb)->encap_level += ihl;
1404
1405 skb_reset_transport_header(skb);
1406
1407 segs = ERR_PTR(-EPROTONOSUPPORT);
1408
1409 if (!skb->encapsulation || encap) {
1410 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1411 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1412
1413 /* fixed ID is invalid if DF bit is not set */
1414 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1415 goto out;
1416 }
1417
1418 ops = rcu_dereference(inet_offloads[proto]);
1419 if (likely(ops && ops->callbacks.gso_segment)) {
1420 segs = ops->callbacks.gso_segment(skb, features);
1421 if (!segs)
1422 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1423 }
1424
1425 if (IS_ERR_OR_NULL(segs))
1426 goto out;
1427
1428 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1429
1430 skb = segs;
1431 do {
1432 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1433 if (udpfrag) {
1434 iph->frag_off = htons(offset >> 3);
1435 if (skb->next)
1436 iph->frag_off |= htons(IP_MF);
1437 offset += skb->len - nhoff - ihl;
1438 tot_len = skb->len - nhoff;
1439 } else if (skb_is_gso(skb)) {
1440 if (!fixedid) {
1441 iph->id = htons(id);
1442 id += skb_shinfo(skb)->gso_segs;
1443 }
1444
1445 if (gso_partial)
1446 tot_len = skb_shinfo(skb)->gso_size +
1447 SKB_GSO_CB(skb)->data_offset +
1448 skb->head - (unsigned char *)iph;
1449 else
1450 tot_len = skb->len - nhoff;
1451 } else {
1452 if (!fixedid)
1453 iph->id = htons(id++);
1454 tot_len = skb->len - nhoff;
1455 }
1456 iph->tot_len = htons(tot_len);
1457 ip_send_check(iph);
1458 if (encap)
1459 skb_reset_inner_headers(skb);
1460 skb->network_header = (u8 *)iph - skb->head;
1461 skb_reset_mac_len(skb);
1462 } while ((skb = skb->next));
1463
1464 out:
1465 return segs;
1466 }
1467
ipip_gso_segment(struct sk_buff * skb,netdev_features_t features)1468 static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1469 netdev_features_t features)
1470 {
1471 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1472 return ERR_PTR(-EINVAL);
1473
1474 return inet_gso_segment(skb, features);
1475 }
1476
inet_gro_receive(struct list_head * head,struct sk_buff * skb)1477 struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1478 {
1479 const struct net_offload *ops;
1480 struct sk_buff *pp = NULL;
1481 const struct iphdr *iph;
1482 struct sk_buff *p;
1483 unsigned int hlen;
1484 unsigned int off;
1485 unsigned int id;
1486 int flush = 1;
1487 int proto;
1488
1489 off = skb_gro_offset(skb);
1490 hlen = off + sizeof(*iph);
1491 iph = skb_gro_header(skb, hlen, off);
1492 if (unlikely(!iph))
1493 goto out;
1494
1495 proto = iph->protocol;
1496
1497 ops = rcu_dereference(inet_offloads[proto]);
1498 if (!ops || !ops->callbacks.gro_receive)
1499 goto out;
1500
1501 if (*(u8 *)iph != 0x45)
1502 goto out;
1503
1504 if (ip_is_fragment(iph))
1505 goto out;
1506
1507 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1508 goto out;
1509
1510 NAPI_GRO_CB(skb)->proto = proto;
1511 id = ntohl(*(__be32 *)&iph->id);
1512 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1513 id >>= 16;
1514
1515 list_for_each_entry(p, head, list) {
1516 struct iphdr *iph2;
1517 u16 flush_id;
1518
1519 if (!NAPI_GRO_CB(p)->same_flow)
1520 continue;
1521
1522 iph2 = (struct iphdr *)(p->data + off);
1523 /* The above works because, with the exception of the top
1524 * (inner most) layer, we only aggregate pkts with the same
1525 * hdr length so all the hdrs we'll need to verify will start
1526 * at the same offset.
1527 */
1528 if ((iph->protocol ^ iph2->protocol) |
1529 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1530 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1531 NAPI_GRO_CB(p)->same_flow = 0;
1532 continue;
1533 }
1534
1535 /* All fields must match except length and checksum. */
1536 NAPI_GRO_CB(p)->flush |=
1537 (iph->ttl ^ iph2->ttl) |
1538 (iph->tos ^ iph2->tos) |
1539 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1540
1541 NAPI_GRO_CB(p)->flush |= flush;
1542
1543 /* We need to store of the IP ID check to be included later
1544 * when we can verify that this packet does in fact belong
1545 * to a given flow.
1546 */
1547 flush_id = (u16)(id - ntohs(iph2->id));
1548
1549 /* This bit of code makes it much easier for us to identify
1550 * the cases where we are doing atomic vs non-atomic IP ID
1551 * checks. Specifically an atomic check can return IP ID
1552 * values 0 - 0xFFFF, while a non-atomic check can only
1553 * return 0 or 0xFFFF.
1554 */
1555 if (!NAPI_GRO_CB(p)->is_atomic ||
1556 !(iph->frag_off & htons(IP_DF))) {
1557 flush_id ^= NAPI_GRO_CB(p)->count;
1558 flush_id = flush_id ? 0xFFFF : 0;
1559 }
1560
1561 /* If the previous IP ID value was based on an atomic
1562 * datagram we can overwrite the value and ignore it.
1563 */
1564 if (NAPI_GRO_CB(skb)->is_atomic)
1565 NAPI_GRO_CB(p)->flush_id = flush_id;
1566 else
1567 NAPI_GRO_CB(p)->flush_id |= flush_id;
1568 }
1569
1570 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1571 NAPI_GRO_CB(skb)->flush |= flush;
1572 skb_set_network_header(skb, off);
1573 /* The above will be needed by the transport layer if there is one
1574 * immediately following this IP hdr.
1575 */
1576 NAPI_GRO_CB(skb)->inner_network_offset = off;
1577
1578 /* Note : No need to call skb_gro_postpull_rcsum() here,
1579 * as we already checked checksum over ipv4 header was 0
1580 */
1581 skb_gro_pull(skb, sizeof(*iph));
1582 skb_set_transport_header(skb, skb_gro_offset(skb));
1583
1584 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1585 ops->callbacks.gro_receive, head, skb);
1586
1587 out:
1588 skb_gro_flush_final(skb, pp, flush);
1589
1590 return pp;
1591 }
1592
ipip_gro_receive(struct list_head * head,struct sk_buff * skb)1593 static struct sk_buff *ipip_gro_receive(struct list_head *head,
1594 struct sk_buff *skb)
1595 {
1596 if (NAPI_GRO_CB(skb)->encap_mark) {
1597 NAPI_GRO_CB(skb)->flush = 1;
1598 return NULL;
1599 }
1600
1601 NAPI_GRO_CB(skb)->encap_mark = 1;
1602
1603 return inet_gro_receive(head, skb);
1604 }
1605
1606 #define SECONDS_PER_DAY 86400
1607
1608 /* inet_current_timestamp - Return IP network timestamp
1609 *
1610 * Return milliseconds since midnight in network byte order.
1611 */
inet_current_timestamp(void)1612 __be32 inet_current_timestamp(void)
1613 {
1614 u32 secs;
1615 u32 msecs;
1616 struct timespec64 ts;
1617
1618 ktime_get_real_ts64(&ts);
1619
1620 /* Get secs since midnight. */
1621 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1622 /* Convert to msecs. */
1623 msecs = secs * MSEC_PER_SEC;
1624 /* Convert nsec to msec. */
1625 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1626
1627 /* Convert to network byte order. */
1628 return htonl(msecs);
1629 }
1630 EXPORT_SYMBOL(inet_current_timestamp);
1631
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1632 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1633 {
1634 unsigned int family = READ_ONCE(sk->sk_family);
1635
1636 if (family == AF_INET)
1637 return ip_recv_error(sk, msg, len, addr_len);
1638 #if IS_ENABLED(CONFIG_IPV6)
1639 if (family == AF_INET6)
1640 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1641 #endif
1642 return -EINVAL;
1643 }
1644 EXPORT_SYMBOL(inet_recv_error);
1645
inet_gro_complete(struct sk_buff * skb,int nhoff)1646 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1647 {
1648 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1649 const struct net_offload *ops;
1650 __be16 totlen = iph->tot_len;
1651 int proto = iph->protocol;
1652 int err = -ENOSYS;
1653
1654 if (skb->encapsulation) {
1655 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1656 skb_set_inner_network_header(skb, nhoff);
1657 }
1658
1659 iph_set_totlen(iph, skb->len - nhoff);
1660 csum_replace2(&iph->check, totlen, iph->tot_len);
1661
1662 ops = rcu_dereference(inet_offloads[proto]);
1663 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1664 goto out;
1665
1666 /* Only need to add sizeof(*iph) to get to the next hdr below
1667 * because any hdr with option will have been flushed in
1668 * inet_gro_receive().
1669 */
1670 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1671 tcp4_gro_complete, udp4_gro_complete,
1672 skb, nhoff + sizeof(*iph));
1673
1674 out:
1675 return err;
1676 }
1677
ipip_gro_complete(struct sk_buff * skb,int nhoff)1678 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1679 {
1680 skb->encapsulation = 1;
1681 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1682 return inet_gro_complete(skb, nhoff);
1683 }
1684
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1685 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1686 unsigned short type, unsigned char protocol,
1687 struct net *net)
1688 {
1689 struct socket *sock;
1690 int rc = sock_create_kern(net, family, type, protocol, &sock);
1691
1692 if (rc == 0) {
1693 *sk = sock->sk;
1694 (*sk)->sk_allocation = GFP_ATOMIC;
1695 (*sk)->sk_use_task_frag = false;
1696 /*
1697 * Unhash it so that IP input processing does not even see it,
1698 * we do not wish this socket to see incoming packets.
1699 */
1700 (*sk)->sk_prot->unhash(*sk);
1701 }
1702 return rc;
1703 }
1704 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1705
snmp_fold_field(void __percpu * mib,int offt)1706 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1707 {
1708 unsigned long res = 0;
1709 int i;
1710
1711 for_each_possible_cpu(i)
1712 res += snmp_get_cpu_field(mib, i, offt);
1713 return res;
1714 }
1715 EXPORT_SYMBOL_GPL(snmp_fold_field);
1716
1717 #if BITS_PER_LONG==32
1718
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1719 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1720 size_t syncp_offset)
1721 {
1722 void *bhptr;
1723 struct u64_stats_sync *syncp;
1724 u64 v;
1725 unsigned int start;
1726
1727 bhptr = per_cpu_ptr(mib, cpu);
1728 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1729 do {
1730 start = u64_stats_fetch_begin(syncp);
1731 v = *(((u64 *)bhptr) + offt);
1732 } while (u64_stats_fetch_retry(syncp, start));
1733
1734 return v;
1735 }
1736 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1737
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1738 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1739 {
1740 u64 res = 0;
1741 int cpu;
1742
1743 for_each_possible_cpu(cpu) {
1744 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1745 }
1746 return res;
1747 }
1748 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1749 #endif
1750
1751 #ifdef CONFIG_IP_MULTICAST
1752 static const struct net_protocol igmp_protocol = {
1753 .handler = igmp_rcv,
1754 };
1755 #endif
1756
1757 static const struct net_protocol tcp_protocol = {
1758 .handler = tcp_v4_rcv,
1759 .err_handler = tcp_v4_err,
1760 .no_policy = 1,
1761 .icmp_strict_tag_validation = 1,
1762 };
1763
1764 static const struct net_protocol udp_protocol = {
1765 .handler = udp_rcv,
1766 .err_handler = udp_err,
1767 .no_policy = 1,
1768 };
1769
1770 static const struct net_protocol icmp_protocol = {
1771 .handler = icmp_rcv,
1772 .err_handler = icmp_err,
1773 .no_policy = 1,
1774 };
1775
ipv4_mib_init_net(struct net * net)1776 static __net_init int ipv4_mib_init_net(struct net *net)
1777 {
1778 int i;
1779
1780 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1781 if (!net->mib.tcp_statistics)
1782 goto err_tcp_mib;
1783 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1784 if (!net->mib.ip_statistics)
1785 goto err_ip_mib;
1786
1787 for_each_possible_cpu(i) {
1788 struct ipstats_mib *af_inet_stats;
1789 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1790 u64_stats_init(&af_inet_stats->syncp);
1791 }
1792
1793 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1794 if (!net->mib.net_statistics)
1795 goto err_net_mib;
1796 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1797 if (!net->mib.udp_statistics)
1798 goto err_udp_mib;
1799 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1800 if (!net->mib.udplite_statistics)
1801 goto err_udplite_mib;
1802 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1803 if (!net->mib.icmp_statistics)
1804 goto err_icmp_mib;
1805 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1806 GFP_KERNEL);
1807 if (!net->mib.icmpmsg_statistics)
1808 goto err_icmpmsg_mib;
1809
1810 tcp_mib_init(net);
1811 return 0;
1812
1813 err_icmpmsg_mib:
1814 free_percpu(net->mib.icmp_statistics);
1815 err_icmp_mib:
1816 free_percpu(net->mib.udplite_statistics);
1817 err_udplite_mib:
1818 free_percpu(net->mib.udp_statistics);
1819 err_udp_mib:
1820 free_percpu(net->mib.net_statistics);
1821 err_net_mib:
1822 free_percpu(net->mib.ip_statistics);
1823 err_ip_mib:
1824 free_percpu(net->mib.tcp_statistics);
1825 err_tcp_mib:
1826 return -ENOMEM;
1827 }
1828
ipv4_mib_exit_net(struct net * net)1829 static __net_exit void ipv4_mib_exit_net(struct net *net)
1830 {
1831 kfree(net->mib.icmpmsg_statistics);
1832 free_percpu(net->mib.icmp_statistics);
1833 free_percpu(net->mib.udplite_statistics);
1834 free_percpu(net->mib.udp_statistics);
1835 free_percpu(net->mib.net_statistics);
1836 free_percpu(net->mib.ip_statistics);
1837 free_percpu(net->mib.tcp_statistics);
1838 #ifdef CONFIG_MPTCP
1839 /* allocated on demand, see mptcp_init_sock() */
1840 free_percpu(net->mib.mptcp_statistics);
1841 #endif
1842 }
1843
1844 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1845 .init = ipv4_mib_init_net,
1846 .exit = ipv4_mib_exit_net,
1847 };
1848
init_ipv4_mibs(void)1849 static int __init init_ipv4_mibs(void)
1850 {
1851 return register_pernet_subsys(&ipv4_mib_ops);
1852 }
1853
inet_init_net(struct net * net)1854 static __net_init int inet_init_net(struct net *net)
1855 {
1856 /*
1857 * Set defaults for local port range
1858 */
1859 seqlock_init(&net->ipv4.ip_local_ports.lock);
1860 net->ipv4.ip_local_ports.range[0] = 32768;
1861 net->ipv4.ip_local_ports.range[1] = 60999;
1862
1863 seqlock_init(&net->ipv4.ping_group_range.lock);
1864 /*
1865 * Sane defaults - nobody may create ping sockets.
1866 * Boot scripts should set this to distro-specific group.
1867 */
1868 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1869 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1870
1871 /* Default values for sysctl-controlled parameters.
1872 * We set them here, in case sysctl is not compiled.
1873 */
1874 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1875 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1876 net->ipv4.sysctl_ip_dynaddr = 0;
1877 net->ipv4.sysctl_ip_early_demux = 1;
1878 net->ipv4.sysctl_udp_early_demux = 1;
1879 net->ipv4.sysctl_tcp_early_demux = 1;
1880 net->ipv4.sysctl_nexthop_compat_mode = 1;
1881 #ifdef CONFIG_SYSCTL
1882 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1883 #endif
1884
1885 /* Some igmp sysctl, whose values are always used */
1886 net->ipv4.sysctl_igmp_max_memberships = 20;
1887 net->ipv4.sysctl_igmp_max_msf = 10;
1888 /* IGMP reports for link-local multicast groups are enabled by default */
1889 net->ipv4.sysctl_igmp_llm_reports = 1;
1890 net->ipv4.sysctl_igmp_qrv = 2;
1891
1892 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1893
1894 return 0;
1895 }
1896
1897 static __net_initdata struct pernet_operations af_inet_ops = {
1898 .init = inet_init_net,
1899 };
1900
init_inet_pernet_ops(void)1901 static int __init init_inet_pernet_ops(void)
1902 {
1903 return register_pernet_subsys(&af_inet_ops);
1904 }
1905
1906 static int ipv4_proc_init(void);
1907
1908 /*
1909 * IP protocol layer initialiser
1910 */
1911
1912 static struct packet_offload ip_packet_offload __read_mostly = {
1913 .type = cpu_to_be16(ETH_P_IP),
1914 .callbacks = {
1915 .gso_segment = inet_gso_segment,
1916 .gro_receive = inet_gro_receive,
1917 .gro_complete = inet_gro_complete,
1918 },
1919 };
1920
1921 static const struct net_offload ipip_offload = {
1922 .callbacks = {
1923 .gso_segment = ipip_gso_segment,
1924 .gro_receive = ipip_gro_receive,
1925 .gro_complete = ipip_gro_complete,
1926 },
1927 };
1928
ipip_offload_init(void)1929 static int __init ipip_offload_init(void)
1930 {
1931 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1932 }
1933
ipv4_offload_init(void)1934 static int __init ipv4_offload_init(void)
1935 {
1936 /*
1937 * Add offloads
1938 */
1939 if (udpv4_offload_init() < 0)
1940 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1941 if (tcpv4_offload_init() < 0)
1942 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1943 if (ipip_offload_init() < 0)
1944 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1945
1946 dev_add_offload(&ip_packet_offload);
1947 return 0;
1948 }
1949
1950 fs_initcall(ipv4_offload_init);
1951
1952 static struct packet_type ip_packet_type __read_mostly = {
1953 .type = cpu_to_be16(ETH_P_IP),
1954 .func = ip_rcv,
1955 .list_func = ip_list_rcv,
1956 };
1957
inet_init(void)1958 static int __init inet_init(void)
1959 {
1960 struct inet_protosw *q;
1961 struct list_head *r;
1962 int rc;
1963
1964 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1965
1966 raw_hashinfo_init(&raw_v4_hashinfo);
1967
1968 rc = proto_register(&tcp_prot, 1);
1969 if (rc)
1970 goto out;
1971
1972 rc = proto_register(&udp_prot, 1);
1973 if (rc)
1974 goto out_unregister_tcp_proto;
1975
1976 rc = proto_register(&raw_prot, 1);
1977 if (rc)
1978 goto out_unregister_udp_proto;
1979
1980 rc = proto_register(&ping_prot, 1);
1981 if (rc)
1982 goto out_unregister_raw_proto;
1983
1984 /*
1985 * Tell SOCKET that we are alive...
1986 */
1987
1988 (void)sock_register(&inet_family_ops);
1989
1990 #ifdef CONFIG_SYSCTL
1991 ip_static_sysctl_init();
1992 #endif
1993
1994 /*
1995 * Add all the base protocols.
1996 */
1997
1998 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1999 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
2000 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
2001 pr_crit("%s: Cannot add UDP protocol\n", __func__);
2002 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
2003 pr_crit("%s: Cannot add TCP protocol\n", __func__);
2004 #ifdef CONFIG_IP_MULTICAST
2005 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
2006 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
2007 #endif
2008
2009 /* Register the socket-side information for inet_create. */
2010 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2011 INIT_LIST_HEAD(r);
2012
2013 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2014 inet_register_protosw(q);
2015
2016 /*
2017 * Set the ARP module up
2018 */
2019
2020 arp_init();
2021
2022 /*
2023 * Set the IP module up
2024 */
2025
2026 ip_init();
2027
2028 /* Initialise per-cpu ipv4 mibs */
2029 if (init_ipv4_mibs())
2030 panic("%s: Cannot init ipv4 mibs\n", __func__);
2031
2032 /* Setup TCP slab cache for open requests. */
2033 tcp_init();
2034
2035 /* Setup UDP memory threshold */
2036 udp_init();
2037
2038 /* Add UDP-Lite (RFC 3828) */
2039 udplite4_register();
2040
2041 raw_init();
2042
2043 ping_init();
2044
2045 /*
2046 * Set the ICMP layer up
2047 */
2048
2049 if (icmp_init() < 0)
2050 panic("Failed to create the ICMP control socket.\n");
2051
2052 /*
2053 * Initialise the multicast router
2054 */
2055 #if defined(CONFIG_IP_MROUTE)
2056 if (ip_mr_init())
2057 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2058 #endif
2059
2060 if (init_inet_pernet_ops())
2061 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2062
2063 ipv4_proc_init();
2064
2065 ipfrag_init();
2066
2067 dev_add_pack(&ip_packet_type);
2068
2069 ip_tunnel_core_init();
2070
2071 rc = 0;
2072 out:
2073 return rc;
2074 out_unregister_raw_proto:
2075 proto_unregister(&raw_prot);
2076 out_unregister_udp_proto:
2077 proto_unregister(&udp_prot);
2078 out_unregister_tcp_proto:
2079 proto_unregister(&tcp_prot);
2080 goto out;
2081 }
2082
2083 fs_initcall(inet_init);
2084
2085 /* ------------------------------------------------------------------------ */
2086
2087 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)2088 static int __init ipv4_proc_init(void)
2089 {
2090 int rc = 0;
2091
2092 if (raw_proc_init())
2093 goto out_raw;
2094 if (tcp4_proc_init())
2095 goto out_tcp;
2096 if (udp4_proc_init())
2097 goto out_udp;
2098 if (ping_proc_init())
2099 goto out_ping;
2100 if (ip_misc_proc_init())
2101 goto out_misc;
2102 out:
2103 return rc;
2104 out_misc:
2105 ping_proc_exit();
2106 out_ping:
2107 udp4_proc_exit();
2108 out_udp:
2109 tcp4_proc_exit();
2110 out_tcp:
2111 raw_proc_exit();
2112 out_raw:
2113 rc = -ENOMEM;
2114 goto out;
2115 }
2116
2117 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)2118 static int __init ipv4_proc_init(void)
2119 {
2120 return 0;
2121 }
2122 #endif /* CONFIG_PROC_FS */
2123