1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PF_INET protocol family socket handler.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
12 *
13 * Changes (see also sock.c)
14 *
15 * piggy,
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
23 * structures
24 * when accept() ed
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
30 * Alan Cox,
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
40 * compactness.
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
46 * dumbly.
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
61 *
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
66 */
67
68 #define pr_fmt(fmt) "IPv4: " fmt
69
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
74 #include <linux/in.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
84 #include <linux/mm.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
92 #include <linux/netfilter/xt_qtaguid.h>
93
94 #include <asm/uaccess.h>
95
96 #include <linux/inet.h>
97 #include <linux/igmp.h>
98 #include <linux/inetdevice.h>
99 #include <linux/netdevice.h>
100 #include <net/checksum.h>
101 #include <net/ip.h>
102 #include <net/protocol.h>
103 #include <net/arp.h>
104 #include <net/route.h>
105 #include <net/ip_fib.h>
106 #include <net/inet_connection_sock.h>
107 #include <net/tcp.h>
108 #include <net/udp.h>
109 #include <net/udplite.h>
110 #include <net/ping.h>
111 #include <linux/skbuff.h>
112 #include <net/sock.h>
113 #include <net/raw.h>
114 #include <net/icmp.h>
115 #include <net/inet_common.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #include <net/secure_seq.h>
119 #ifdef CONFIG_IP_MROUTE
120 #include <linux/mroute.h>
121 #endif
122
123 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
124 #include <linux/android_aid.h>
125
current_has_network(void)126 static inline int current_has_network(void)
127 {
128 return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
129 }
130 #else
current_has_network(void)131 static inline int current_has_network(void)
132 {
133 return 1;
134 }
135 #endif
136
137 /* The inetsw table contains everything that inet_create needs to
138 * build a new socket.
139 */
140 static struct list_head inetsw[SOCK_MAX];
141 static DEFINE_SPINLOCK(inetsw_lock);
142
143 /* New destruction routine */
144
inet_sock_destruct(struct sock * sk)145 void inet_sock_destruct(struct sock *sk)
146 {
147 struct inet_sock *inet = inet_sk(sk);
148
149 __skb_queue_purge(&sk->sk_receive_queue);
150 __skb_queue_purge(&sk->sk_error_queue);
151
152 sk_mem_reclaim(sk);
153
154 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
155 pr_err("Attempt to release TCP socket in state %d %p\n",
156 sk->sk_state, sk);
157 return;
158 }
159 if (!sock_flag(sk, SOCK_DEAD)) {
160 pr_err("Attempt to release alive inet socket %p\n", sk);
161 return;
162 }
163
164 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
165 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
166 WARN_ON(sk->sk_wmem_queued);
167 WARN_ON(sk->sk_forward_alloc);
168
169 kfree(rcu_dereference_protected(inet->inet_opt, 1));
170 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
171 dst_release(sk->sk_rx_dst);
172 sk_refcnt_debug_dec(sk);
173 }
174 EXPORT_SYMBOL(inet_sock_destruct);
175
176 /*
177 * The routines beyond this point handle the behaviour of an AF_INET
178 * socket object. Mostly it punts to the subprotocols of IP to do
179 * the work.
180 */
181
182 /*
183 * Automatically bind an unbound socket.
184 */
185
inet_autobind(struct sock * sk)186 static int inet_autobind(struct sock *sk)
187 {
188 struct inet_sock *inet;
189 /* We may need to bind the socket. */
190 lock_sock(sk);
191 inet = inet_sk(sk);
192 if (!inet->inet_num) {
193 if (sk->sk_prot->get_port(sk, 0)) {
194 release_sock(sk);
195 return -EAGAIN;
196 }
197 inet->inet_sport = htons(inet->inet_num);
198 }
199 release_sock(sk);
200 return 0;
201 }
202
203 /*
204 * Move a socket into listening state.
205 */
inet_listen(struct socket * sock,int backlog)206 int inet_listen(struct socket *sock, int backlog)
207 {
208 struct sock *sk = sock->sk;
209 unsigned char old_state;
210 int err;
211
212 lock_sock(sk);
213
214 err = -EINVAL;
215 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
216 goto out;
217
218 old_state = sk->sk_state;
219 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
220 goto out;
221
222 /* Really, if the socket is already in listen state
223 * we can only allow the backlog to be adjusted.
224 */
225 if (old_state != TCP_LISTEN) {
226 /* Check special setups for testing purpose to enable TFO w/o
227 * requiring TCP_FASTOPEN sockopt.
228 * Note that only TCP sockets (SOCK_STREAM) will reach here.
229 * Also fastopenq may already been allocated because this
230 * socket was in TCP_LISTEN state previously but was
231 * shutdown() (rather than close()).
232 */
233 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
234 inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
235 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
236 err = fastopen_init_queue(sk, backlog);
237 else if ((sysctl_tcp_fastopen &
238 TFO_SERVER_WO_SOCKOPT2) != 0)
239 err = fastopen_init_queue(sk,
240 ((uint)sysctl_tcp_fastopen) >> 16);
241 else
242 err = 0;
243 if (err)
244 goto out;
245
246 tcp_fastopen_init_key_once(true);
247 }
248 err = inet_csk_listen_start(sk, backlog);
249 if (err)
250 goto out;
251 }
252 sk->sk_max_ack_backlog = backlog;
253 err = 0;
254
255 out:
256 release_sock(sk);
257 return err;
258 }
259 EXPORT_SYMBOL(inet_listen);
260
261 /*
262 * Create an inet socket.
263 */
264
inet_create(struct net * net,struct socket * sock,int protocol,int kern)265 static int inet_create(struct net *net, struct socket *sock, int protocol,
266 int kern)
267 {
268 struct sock *sk;
269 struct inet_protosw *answer;
270 struct inet_sock *inet;
271 struct proto *answer_prot;
272 unsigned char answer_flags;
273 int try_loading_module = 0;
274 int err;
275
276 if (!current_has_network())
277 return -EACCES;
278
279 if (protocol < 0 || protocol >= IPPROTO_MAX)
280 return -EINVAL;
281
282 sock->state = SS_UNCONNECTED;
283
284 /* Look for the requested type/protocol pair. */
285 lookup_protocol:
286 err = -ESOCKTNOSUPPORT;
287 rcu_read_lock();
288 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
289
290 err = 0;
291 /* Check the non-wild match. */
292 if (protocol == answer->protocol) {
293 if (protocol != IPPROTO_IP)
294 break;
295 } else {
296 /* Check for the two wild cases. */
297 if (IPPROTO_IP == protocol) {
298 protocol = answer->protocol;
299 break;
300 }
301 if (IPPROTO_IP == answer->protocol)
302 break;
303 }
304 err = -EPROTONOSUPPORT;
305 }
306
307 if (unlikely(err)) {
308 if (try_loading_module < 2) {
309 rcu_read_unlock();
310 /*
311 * Be more specific, e.g. net-pf-2-proto-132-type-1
312 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
313 */
314 if (++try_loading_module == 1)
315 request_module("net-pf-%d-proto-%d-type-%d",
316 PF_INET, protocol, sock->type);
317 /*
318 * Fall back to generic, e.g. net-pf-2-proto-132
319 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
320 */
321 else
322 request_module("net-pf-%d-proto-%d",
323 PF_INET, protocol);
324 goto lookup_protocol;
325 } else
326 goto out_rcu_unlock;
327 }
328
329 err = -EPERM;
330 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
331 goto out_rcu_unlock;
332
333 sock->ops = answer->ops;
334 answer_prot = answer->prot;
335 answer_flags = answer->flags;
336 rcu_read_unlock();
337
338 WARN_ON(answer_prot->slab == NULL);
339
340 err = -ENOBUFS;
341 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
342 if (sk == NULL)
343 goto out;
344
345 err = 0;
346 if (INET_PROTOSW_REUSE & answer_flags)
347 sk->sk_reuse = SK_CAN_REUSE;
348
349 inet = inet_sk(sk);
350 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
351
352 inet->nodefrag = 0;
353
354 if (SOCK_RAW == sock->type) {
355 inet->inet_num = protocol;
356 if (IPPROTO_RAW == protocol)
357 inet->hdrincl = 1;
358 }
359
360 if (net->ipv4.sysctl_ip_no_pmtu_disc)
361 inet->pmtudisc = IP_PMTUDISC_DONT;
362 else
363 inet->pmtudisc = IP_PMTUDISC_WANT;
364
365 inet->inet_id = 0;
366
367 sock_init_data(sock, sk);
368
369 sk->sk_destruct = inet_sock_destruct;
370 sk->sk_protocol = protocol;
371 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
372
373 inet->uc_ttl = -1;
374 inet->mc_loop = 1;
375 inet->mc_ttl = 1;
376 inet->mc_all = 1;
377 inet->mc_index = 0;
378 inet->mc_list = NULL;
379 inet->rcv_tos = 0;
380
381 sk_refcnt_debug_inc(sk);
382
383 if (inet->inet_num) {
384 /* It assumes that any protocol which allows
385 * the user to assign a number at socket
386 * creation time automatically
387 * shares.
388 */
389 inet->inet_sport = htons(inet->inet_num);
390 /* Add to protocol hash chains. */
391 sk->sk_prot->hash(sk);
392 }
393
394 if (sk->sk_prot->init) {
395 err = sk->sk_prot->init(sk);
396 if (err)
397 sk_common_release(sk);
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 #ifdef CONFIG_NETFILTER_XT_MATCH_QTAGUID
420 qtaguid_untag(sock, true);
421 #endif
422 sock_rps_reset_flow(sk);
423
424 /* Applications forget to leave groups before exiting */
425 ip_mc_drop_socket(sk);
426
427 /* If linger is set, we don't return until the close
428 * is complete. Otherwise we return immediately. The
429 * actually closing is done the same either way.
430 *
431 * If the close is due to the process exiting, we never
432 * linger..
433 */
434 timeout = 0;
435 if (sock_flag(sk, SOCK_LINGER) &&
436 !(current->flags & PF_EXITING))
437 timeout = sk->sk_lingertime;
438 sock->sk = NULL;
439 sk->sk_prot->close(sk, timeout);
440 }
441 return 0;
442 }
443 EXPORT_SYMBOL(inet_release);
444
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)445 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
446 {
447 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
448 struct sock *sk = sock->sk;
449 struct inet_sock *inet = inet_sk(sk);
450 struct net *net = sock_net(sk);
451 unsigned short snum;
452 int chk_addr_ret;
453 int err;
454
455 /* If the socket has its own bind function then use it. (RAW) */
456 if (sk->sk_prot->bind) {
457 err = sk->sk_prot->bind(sk, uaddr, addr_len);
458 goto out;
459 }
460 err = -EINVAL;
461 if (addr_len < sizeof(struct sockaddr_in))
462 goto out;
463
464 if (addr->sin_family != AF_INET) {
465 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
466 * only if s_addr is INADDR_ANY.
467 */
468 err = -EAFNOSUPPORT;
469 if (addr->sin_family != AF_UNSPEC ||
470 addr->sin_addr.s_addr != htonl(INADDR_ANY))
471 goto out;
472 }
473
474 chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
475
476 /* Not specified by any standard per-se, however it breaks too
477 * many applications when removed. It is unfortunate since
478 * allowing applications to make a non-local bind solves
479 * several problems with systems using dynamic addressing.
480 * (ie. your servers still start up even if your ISDN link
481 * is temporarily down)
482 */
483 err = -EADDRNOTAVAIL;
484 if (!net->ipv4.sysctl_ip_nonlocal_bind &&
485 !(inet->freebind || inet->transparent) &&
486 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
487 chk_addr_ret != RTN_LOCAL &&
488 chk_addr_ret != RTN_MULTICAST &&
489 chk_addr_ret != RTN_BROADCAST)
490 goto out;
491
492 snum = ntohs(addr->sin_port);
493 err = -EACCES;
494 if (snum && snum < PROT_SOCK &&
495 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
496 goto out;
497
498 /* We keep a pair of addresses. rcv_saddr is the one
499 * used by hash lookups, and saddr is used for transmit.
500 *
501 * In the BSD API these are the same except where it
502 * would be illegal to use them (multicast/broadcast) in
503 * which case the sending device address is used.
504 */
505 lock_sock(sk);
506
507 /* Check these errors (active socket, double bind). */
508 err = -EINVAL;
509 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
510 goto out_release_sock;
511
512 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
513 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
514 inet->inet_saddr = 0; /* Use device */
515
516 /* Make sure we are allowed to bind here. */
517 if (sk->sk_prot->get_port(sk, snum)) {
518 inet->inet_saddr = inet->inet_rcv_saddr = 0;
519 err = -EADDRINUSE;
520 goto out_release_sock;
521 }
522
523 if (inet->inet_rcv_saddr)
524 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
525 if (snum)
526 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
527 inet->inet_sport = htons(inet->inet_num);
528 inet->inet_daddr = 0;
529 inet->inet_dport = 0;
530 sk_dst_reset(sk);
531 err = 0;
532 out_release_sock:
533 release_sock(sk);
534 out:
535 return err;
536 }
537 EXPORT_SYMBOL(inet_bind);
538
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)539 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
540 int addr_len, int flags)
541 {
542 struct sock *sk = sock->sk;
543
544 if (addr_len < sizeof(uaddr->sa_family))
545 return -EINVAL;
546 if (uaddr->sa_family == AF_UNSPEC)
547 return sk->sk_prot->disconnect(sk, flags);
548
549 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
550 return -EAGAIN;
551 return sk->sk_prot->connect(sk, uaddr, addr_len);
552 }
553 EXPORT_SYMBOL(inet_dgram_connect);
554
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)555 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
556 {
557 DEFINE_WAIT(wait);
558
559 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
560 sk->sk_write_pending += writebias;
561
562 /* Basic assumption: if someone sets sk->sk_err, he _must_
563 * change state of the socket from TCP_SYN_*.
564 * Connect() does not allow to get error notifications
565 * without closing the socket.
566 */
567 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
568 release_sock(sk);
569 timeo = schedule_timeout(timeo);
570 lock_sock(sk);
571 if (signal_pending(current) || !timeo)
572 break;
573 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
574 }
575 finish_wait(sk_sleep(sk), &wait);
576 sk->sk_write_pending -= writebias;
577 return timeo;
578 }
579
580 /*
581 * Connect to a remote host. There is regrettably still a little
582 * TCP 'magic' in here.
583 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)584 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
585 int addr_len, int flags)
586 {
587 struct sock *sk = sock->sk;
588 int err;
589 long timeo;
590
591 if (addr_len < sizeof(uaddr->sa_family))
592 return -EINVAL;
593
594 if (uaddr->sa_family == AF_UNSPEC) {
595 err = sk->sk_prot->disconnect(sk, flags);
596 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
597 goto out;
598 }
599
600 switch (sock->state) {
601 default:
602 err = -EINVAL;
603 goto out;
604 case SS_CONNECTED:
605 err = -EISCONN;
606 goto out;
607 case SS_CONNECTING:
608 err = -EALREADY;
609 /* Fall out of switch with err, set for this state */
610 break;
611 case SS_UNCONNECTED:
612 err = -EISCONN;
613 if (sk->sk_state != TCP_CLOSE)
614 goto out;
615
616 err = sk->sk_prot->connect(sk, uaddr, addr_len);
617 if (err < 0)
618 goto out;
619
620 sock->state = SS_CONNECTING;
621
622 /* Just entered SS_CONNECTING state; the only
623 * difference is that return value in non-blocking
624 * case is EINPROGRESS, rather than EALREADY.
625 */
626 err = -EINPROGRESS;
627 break;
628 }
629
630 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
631
632 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
633 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
634 tcp_sk(sk)->fastopen_req &&
635 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
636
637 /* Error code is set above */
638 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
639 goto out;
640
641 err = sock_intr_errno(timeo);
642 if (signal_pending(current))
643 goto out;
644 }
645
646 /* Connection was closed by RST, timeout, ICMP error
647 * or another process disconnected us.
648 */
649 if (sk->sk_state == TCP_CLOSE)
650 goto sock_error;
651
652 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
653 * and error was received after socket entered established state.
654 * Hence, it is handled normally after connect() return successfully.
655 */
656
657 sock->state = SS_CONNECTED;
658 err = 0;
659 out:
660 return err;
661
662 sock_error:
663 err = sock_error(sk) ? : -ECONNABORTED;
664 sock->state = SS_UNCONNECTED;
665 if (sk->sk_prot->disconnect(sk, flags))
666 sock->state = SS_DISCONNECTING;
667 goto out;
668 }
669 EXPORT_SYMBOL(__inet_stream_connect);
670
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)671 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
672 int addr_len, int flags)
673 {
674 int err;
675
676 lock_sock(sock->sk);
677 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
678 release_sock(sock->sk);
679 return err;
680 }
681 EXPORT_SYMBOL(inet_stream_connect);
682
683 /*
684 * Accept a pending connection. The TCP layer now gives BSD semantics.
685 */
686
inet_accept(struct socket * sock,struct socket * newsock,int flags)687 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
688 {
689 struct sock *sk1 = sock->sk;
690 int err = -EINVAL;
691 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
692
693 if (!sk2)
694 goto do_err;
695
696 lock_sock(sk2);
697
698 sock_rps_record_flow(sk2);
699 WARN_ON(!((1 << sk2->sk_state) &
700 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
701 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
702
703 sock_graft(sk2, newsock);
704
705 newsock->state = SS_CONNECTED;
706 err = 0;
707 release_sock(sk2);
708 do_err:
709 return err;
710 }
711 EXPORT_SYMBOL(inet_accept);
712
713
714 /*
715 * This does both peername and sockname.
716 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int * uaddr_len,int peer)717 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
718 int *uaddr_len, int peer)
719 {
720 struct sock *sk = sock->sk;
721 struct inet_sock *inet = inet_sk(sk);
722 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
723
724 sin->sin_family = AF_INET;
725 if (peer) {
726 if (!inet->inet_dport ||
727 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
728 peer == 1))
729 return -ENOTCONN;
730 sin->sin_port = inet->inet_dport;
731 sin->sin_addr.s_addr = inet->inet_daddr;
732 } else {
733 __be32 addr = inet->inet_rcv_saddr;
734 if (!addr)
735 addr = inet->inet_saddr;
736 sin->sin_port = inet->inet_sport;
737 sin->sin_addr.s_addr = addr;
738 }
739 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
740 *uaddr_len = sizeof(*sin);
741 return 0;
742 }
743 EXPORT_SYMBOL(inet_getname);
744
inet_sendmsg(struct kiocb * iocb,struct socket * sock,struct msghdr * msg,size_t size)745 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
746 size_t size)
747 {
748 struct sock *sk = sock->sk;
749
750 sock_rps_record_flow(sk);
751
752 /* We may need to bind the socket. */
753 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
754 inet_autobind(sk))
755 return -EAGAIN;
756
757 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
758 }
759 EXPORT_SYMBOL(inet_sendmsg);
760
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)761 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
762 size_t size, int flags)
763 {
764 struct sock *sk = sock->sk;
765
766 sock_rps_record_flow(sk);
767
768 /* We may need to bind the socket. */
769 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
770 inet_autobind(sk))
771 return -EAGAIN;
772
773 if (sk->sk_prot->sendpage)
774 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
775 return sock_no_sendpage(sock, page, offset, size, flags);
776 }
777 EXPORT_SYMBOL(inet_sendpage);
778
inet_recvmsg(struct kiocb * iocb,struct socket * sock,struct msghdr * msg,size_t size,int flags)779 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
780 size_t size, int flags)
781 {
782 struct sock *sk = sock->sk;
783 int addr_len = 0;
784 int err;
785
786 sock_rps_record_flow(sk);
787
788 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
789 flags & ~MSG_DONTWAIT, &addr_len);
790 if (err >= 0)
791 msg->msg_namelen = addr_len;
792 return err;
793 }
794 EXPORT_SYMBOL(inet_recvmsg);
795
inet_shutdown(struct socket * sock,int how)796 int inet_shutdown(struct socket *sock, int how)
797 {
798 struct sock *sk = sock->sk;
799 int err = 0;
800
801 /* This should really check to make sure
802 * the socket is a TCP socket. (WHY AC...)
803 */
804 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
805 1->2 bit 2 snds.
806 2->3 */
807 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
808 return -EINVAL;
809
810 lock_sock(sk);
811 if (sock->state == SS_CONNECTING) {
812 if ((1 << sk->sk_state) &
813 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
814 sock->state = SS_DISCONNECTING;
815 else
816 sock->state = SS_CONNECTED;
817 }
818
819 switch (sk->sk_state) {
820 case TCP_CLOSE:
821 err = -ENOTCONN;
822 /* Hack to wake up other listeners, who can poll for
823 POLLHUP, even on eg. unconnected UDP sockets -- RR */
824 default:
825 sk->sk_shutdown |= how;
826 if (sk->sk_prot->shutdown)
827 sk->sk_prot->shutdown(sk, how);
828 break;
829
830 /* Remaining two branches are temporary solution for missing
831 * close() in multithreaded environment. It is _not_ a good idea,
832 * but we have no choice until close() is repaired at VFS level.
833 */
834 case TCP_LISTEN:
835 if (!(how & RCV_SHUTDOWN))
836 break;
837 /* Fall through */
838 case TCP_SYN_SENT:
839 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
840 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
841 break;
842 }
843
844 /* Wake up anyone sleeping in poll. */
845 sk->sk_state_change(sk);
846 release_sock(sk);
847 return err;
848 }
849 EXPORT_SYMBOL(inet_shutdown);
850
851 /*
852 * ioctl() calls you can issue on an INET socket. Most of these are
853 * device configuration and stuff and very rarely used. Some ioctls
854 * pass on to the socket itself.
855 *
856 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
857 * loads the devconfigure module does its configuring and unloads it.
858 * There's a good 20K of config code hanging around the kernel.
859 */
860
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)861 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
862 {
863 struct sock *sk = sock->sk;
864 int err = 0;
865 struct net *net = sock_net(sk);
866
867 switch (cmd) {
868 case SIOCGSTAMP:
869 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
870 break;
871 case SIOCGSTAMPNS:
872 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
873 break;
874 case SIOCADDRT:
875 case SIOCDELRT:
876 case SIOCRTMSG:
877 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
878 break;
879 case SIOCDARP:
880 case SIOCGARP:
881 case SIOCSARP:
882 err = arp_ioctl(net, cmd, (void __user *)arg);
883 break;
884 case SIOCGIFADDR:
885 case SIOCSIFADDR:
886 case SIOCGIFBRDADDR:
887 case SIOCSIFBRDADDR:
888 case SIOCGIFNETMASK:
889 case SIOCSIFNETMASK:
890 case SIOCGIFDSTADDR:
891 case SIOCSIFDSTADDR:
892 case SIOCSIFPFLAGS:
893 case SIOCGIFPFLAGS:
894 case SIOCSIFFLAGS:
895 err = devinet_ioctl(net, cmd, (void __user *)arg);
896 break;
897 default:
898 if (sk->sk_prot->ioctl)
899 err = sk->sk_prot->ioctl(sk, cmd, arg);
900 else
901 err = -ENOIOCTLCMD;
902 break;
903 }
904 return err;
905 }
906 EXPORT_SYMBOL(inet_ioctl);
907
908 #ifdef CONFIG_COMPAT
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)909 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
910 {
911 struct sock *sk = sock->sk;
912 int err = -ENOIOCTLCMD;
913
914 if (sk->sk_prot->compat_ioctl)
915 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
916
917 return err;
918 }
919 #endif
920
921 const struct proto_ops inet_stream_ops = {
922 .family = PF_INET,
923 .owner = THIS_MODULE,
924 .release = inet_release,
925 .bind = inet_bind,
926 .connect = inet_stream_connect,
927 .socketpair = sock_no_socketpair,
928 .accept = inet_accept,
929 .getname = inet_getname,
930 .poll = tcp_poll,
931 .ioctl = inet_ioctl,
932 .listen = inet_listen,
933 .shutdown = inet_shutdown,
934 .setsockopt = sock_common_setsockopt,
935 .getsockopt = sock_common_getsockopt,
936 .sendmsg = inet_sendmsg,
937 .recvmsg = inet_recvmsg,
938 .mmap = sock_no_mmap,
939 .sendpage = inet_sendpage,
940 .splice_read = tcp_splice_read,
941 #ifdef CONFIG_COMPAT
942 .compat_setsockopt = compat_sock_common_setsockopt,
943 .compat_getsockopt = compat_sock_common_getsockopt,
944 .compat_ioctl = inet_compat_ioctl,
945 #endif
946 };
947 EXPORT_SYMBOL(inet_stream_ops);
948
949 const struct proto_ops inet_dgram_ops = {
950 .family = PF_INET,
951 .owner = THIS_MODULE,
952 .release = inet_release,
953 .bind = inet_bind,
954 .connect = inet_dgram_connect,
955 .socketpair = sock_no_socketpair,
956 .accept = sock_no_accept,
957 .getname = inet_getname,
958 .poll = udp_poll,
959 .ioctl = inet_ioctl,
960 .listen = sock_no_listen,
961 .shutdown = inet_shutdown,
962 .setsockopt = sock_common_setsockopt,
963 .getsockopt = sock_common_getsockopt,
964 .sendmsg = inet_sendmsg,
965 .recvmsg = inet_recvmsg,
966 .mmap = sock_no_mmap,
967 .sendpage = inet_sendpage,
968 #ifdef CONFIG_COMPAT
969 .compat_setsockopt = compat_sock_common_setsockopt,
970 .compat_getsockopt = compat_sock_common_getsockopt,
971 .compat_ioctl = inet_compat_ioctl,
972 #endif
973 };
974 EXPORT_SYMBOL(inet_dgram_ops);
975
976 /*
977 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
978 * udp_poll
979 */
980 static const struct proto_ops inet_sockraw_ops = {
981 .family = PF_INET,
982 .owner = THIS_MODULE,
983 .release = inet_release,
984 .bind = inet_bind,
985 .connect = inet_dgram_connect,
986 .socketpair = sock_no_socketpair,
987 .accept = sock_no_accept,
988 .getname = inet_getname,
989 .poll = datagram_poll,
990 .ioctl = inet_ioctl,
991 .listen = sock_no_listen,
992 .shutdown = inet_shutdown,
993 .setsockopt = sock_common_setsockopt,
994 .getsockopt = sock_common_getsockopt,
995 .sendmsg = inet_sendmsg,
996 .recvmsg = inet_recvmsg,
997 .mmap = sock_no_mmap,
998 .sendpage = inet_sendpage,
999 #ifdef CONFIG_COMPAT
1000 .compat_setsockopt = compat_sock_common_setsockopt,
1001 .compat_getsockopt = compat_sock_common_getsockopt,
1002 .compat_ioctl = inet_compat_ioctl,
1003 #endif
1004 };
1005
1006 static const struct net_proto_family inet_family_ops = {
1007 .family = PF_INET,
1008 .create = inet_create,
1009 .owner = THIS_MODULE,
1010 };
1011
1012 /* Upon startup we insert all the elements in inetsw_array[] into
1013 * the linked list inetsw.
1014 */
1015 static struct inet_protosw inetsw_array[] =
1016 {
1017 {
1018 .type = SOCK_STREAM,
1019 .protocol = IPPROTO_TCP,
1020 .prot = &tcp_prot,
1021 .ops = &inet_stream_ops,
1022 .flags = INET_PROTOSW_PERMANENT |
1023 INET_PROTOSW_ICSK,
1024 },
1025
1026 {
1027 .type = SOCK_DGRAM,
1028 .protocol = IPPROTO_UDP,
1029 .prot = &udp_prot,
1030 .ops = &inet_dgram_ops,
1031 .flags = INET_PROTOSW_PERMANENT,
1032 },
1033
1034 {
1035 .type = SOCK_DGRAM,
1036 .protocol = IPPROTO_ICMP,
1037 .prot = &ping_prot,
1038 .ops = &inet_sockraw_ops,
1039 .flags = INET_PROTOSW_REUSE,
1040 },
1041
1042 {
1043 .type = SOCK_RAW,
1044 .protocol = IPPROTO_IP, /* wild card */
1045 .prot = &raw_prot,
1046 .ops = &inet_sockraw_ops,
1047 .flags = INET_PROTOSW_REUSE,
1048 }
1049 };
1050
1051 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1052
inet_register_protosw(struct inet_protosw * p)1053 void inet_register_protosw(struct inet_protosw *p)
1054 {
1055 struct list_head *lh;
1056 struct inet_protosw *answer;
1057 int protocol = p->protocol;
1058 struct list_head *last_perm;
1059
1060 spin_lock_bh(&inetsw_lock);
1061
1062 if (p->type >= SOCK_MAX)
1063 goto out_illegal;
1064
1065 /* If we are trying to override a permanent protocol, bail. */
1066 answer = NULL;
1067 last_perm = &inetsw[p->type];
1068 list_for_each(lh, &inetsw[p->type]) {
1069 answer = list_entry(lh, struct inet_protosw, list);
1070
1071 /* Check only the non-wild match. */
1072 if (INET_PROTOSW_PERMANENT & answer->flags) {
1073 if (protocol == answer->protocol)
1074 break;
1075 last_perm = lh;
1076 }
1077
1078 answer = NULL;
1079 }
1080 if (answer)
1081 goto out_permanent;
1082
1083 /* Add the new entry after the last permanent entry if any, so that
1084 * the new entry does not override a permanent entry when matched with
1085 * a wild-card protocol. But it is allowed to override any existing
1086 * non-permanent entry. This means that when we remove this entry, the
1087 * system automatically returns to the old behavior.
1088 */
1089 list_add_rcu(&p->list, last_perm);
1090 out:
1091 spin_unlock_bh(&inetsw_lock);
1092
1093 return;
1094
1095 out_permanent:
1096 pr_err("Attempt to override permanent protocol %d\n", protocol);
1097 goto out;
1098
1099 out_illegal:
1100 pr_err("Ignoring attempt to register invalid socket type %d\n",
1101 p->type);
1102 goto out;
1103 }
1104 EXPORT_SYMBOL(inet_register_protosw);
1105
inet_unregister_protosw(struct inet_protosw * p)1106 void inet_unregister_protosw(struct inet_protosw *p)
1107 {
1108 if (INET_PROTOSW_PERMANENT & p->flags) {
1109 pr_err("Attempt to unregister permanent protocol %d\n",
1110 p->protocol);
1111 } else {
1112 spin_lock_bh(&inetsw_lock);
1113 list_del_rcu(&p->list);
1114 spin_unlock_bh(&inetsw_lock);
1115
1116 synchronize_net();
1117 }
1118 }
1119 EXPORT_SYMBOL(inet_unregister_protosw);
1120
1121 /*
1122 * Shall we try to damage output packets if routing dev changes?
1123 */
1124
1125 int sysctl_ip_dynaddr __read_mostly;
1126
inet_sk_reselect_saddr(struct sock * sk)1127 static int inet_sk_reselect_saddr(struct sock *sk)
1128 {
1129 struct inet_sock *inet = inet_sk(sk);
1130 __be32 old_saddr = inet->inet_saddr;
1131 __be32 daddr = inet->inet_daddr;
1132 struct flowi4 *fl4;
1133 struct rtable *rt;
1134 __be32 new_saddr;
1135 struct ip_options_rcu *inet_opt;
1136
1137 inet_opt = rcu_dereference_protected(inet->inet_opt,
1138 sock_owned_by_user(sk));
1139 if (inet_opt && inet_opt->opt.srr)
1140 daddr = inet_opt->opt.faddr;
1141
1142 /* Query new route. */
1143 fl4 = &inet->cork.fl.u.ip4;
1144 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1145 sk->sk_bound_dev_if, sk->sk_protocol,
1146 inet->inet_sport, inet->inet_dport, sk);
1147 if (IS_ERR(rt))
1148 return PTR_ERR(rt);
1149
1150 sk_setup_caps(sk, &rt->dst);
1151
1152 new_saddr = fl4->saddr;
1153
1154 if (new_saddr == old_saddr)
1155 return 0;
1156
1157 if (sysctl_ip_dynaddr > 1) {
1158 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1159 __func__, &old_saddr, &new_saddr);
1160 }
1161
1162 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1163
1164 /*
1165 * XXX The only one ugly spot where we need to
1166 * XXX really change the sockets identity after
1167 * XXX it has entered the hashes. -DaveM
1168 *
1169 * Besides that, it does not check for connection
1170 * uniqueness. Wait for troubles.
1171 */
1172 __sk_prot_rehash(sk);
1173 return 0;
1174 }
1175
inet_sk_rebuild_header(struct sock * sk)1176 int inet_sk_rebuild_header(struct sock *sk)
1177 {
1178 struct inet_sock *inet = inet_sk(sk);
1179 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1180 __be32 daddr;
1181 struct ip_options_rcu *inet_opt;
1182 struct flowi4 *fl4;
1183 int err;
1184
1185 /* Route is OK, nothing to do. */
1186 if (rt)
1187 return 0;
1188
1189 /* Reroute. */
1190 rcu_read_lock();
1191 inet_opt = rcu_dereference(inet->inet_opt);
1192 daddr = inet->inet_daddr;
1193 if (inet_opt && inet_opt->opt.srr)
1194 daddr = inet_opt->opt.faddr;
1195 rcu_read_unlock();
1196 fl4 = &inet->cork.fl.u.ip4;
1197 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1198 inet->inet_dport, inet->inet_sport,
1199 sk->sk_protocol, RT_CONN_FLAGS(sk),
1200 sk->sk_bound_dev_if);
1201 if (!IS_ERR(rt)) {
1202 err = 0;
1203 sk_setup_caps(sk, &rt->dst);
1204 } else {
1205 err = PTR_ERR(rt);
1206
1207 /* Routing failed... */
1208 sk->sk_route_caps = 0;
1209 /*
1210 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1211 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1212 */
1213 if (!sysctl_ip_dynaddr ||
1214 sk->sk_state != TCP_SYN_SENT ||
1215 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1216 (err = inet_sk_reselect_saddr(sk)) != 0)
1217 sk->sk_err_soft = -err;
1218 }
1219
1220 return err;
1221 }
1222 EXPORT_SYMBOL(inet_sk_rebuild_header);
1223
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1224 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1225 netdev_features_t features)
1226 {
1227 struct sk_buff *segs = ERR_PTR(-EINVAL);
1228 const struct net_offload *ops;
1229 unsigned int offset = 0;
1230 bool udpfrag, encap;
1231 struct iphdr *iph;
1232 int proto;
1233 int nhoff;
1234 int ihl;
1235 int id;
1236
1237 if (unlikely(skb_shinfo(skb)->gso_type &
1238 ~(SKB_GSO_TCPV4 |
1239 SKB_GSO_UDP |
1240 SKB_GSO_DODGY |
1241 SKB_GSO_TCP_ECN |
1242 SKB_GSO_GRE |
1243 SKB_GSO_GRE_CSUM |
1244 SKB_GSO_IPIP |
1245 SKB_GSO_SIT |
1246 SKB_GSO_TCPV6 |
1247 SKB_GSO_UDP_TUNNEL |
1248 SKB_GSO_UDP_TUNNEL_CSUM |
1249 SKB_GSO_MPLS |
1250 0)))
1251 goto out;
1252
1253 skb_reset_network_header(skb);
1254 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1255 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1256 goto out;
1257
1258 iph = ip_hdr(skb);
1259 ihl = iph->ihl * 4;
1260 if (ihl < sizeof(*iph))
1261 goto out;
1262
1263 id = ntohs(iph->id);
1264 proto = iph->protocol;
1265
1266 /* Warning: after this point, iph might be no longer valid */
1267 if (unlikely(!pskb_may_pull(skb, ihl)))
1268 goto out;
1269 __skb_pull(skb, ihl);
1270
1271 encap = SKB_GSO_CB(skb)->encap_level > 0;
1272 if (encap)
1273 features &= skb->dev->hw_enc_features;
1274 SKB_GSO_CB(skb)->encap_level += ihl;
1275
1276 skb_reset_transport_header(skb);
1277
1278 segs = ERR_PTR(-EPROTONOSUPPORT);
1279
1280 if (skb->encapsulation &&
1281 skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1282 udpfrag = proto == IPPROTO_UDP && encap;
1283 else
1284 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1285
1286 ops = rcu_dereference(inet_offloads[proto]);
1287 if (likely(ops && ops->callbacks.gso_segment))
1288 segs = ops->callbacks.gso_segment(skb, features);
1289
1290 if (IS_ERR_OR_NULL(segs))
1291 goto out;
1292
1293 skb = segs;
1294 do {
1295 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1296 if (udpfrag) {
1297 iph->id = htons(id);
1298 iph->frag_off = htons(offset >> 3);
1299 if (skb->next != NULL)
1300 iph->frag_off |= htons(IP_MF);
1301 offset += skb->len - nhoff - ihl;
1302 } else {
1303 iph->id = htons(id++);
1304 }
1305 iph->tot_len = htons(skb->len - nhoff);
1306 ip_send_check(iph);
1307 if (encap)
1308 skb_reset_inner_headers(skb);
1309 skb->network_header = (u8 *)iph - skb->head;
1310 } while ((skb = skb->next));
1311
1312 out:
1313 return segs;
1314 }
1315
inet_gro_receive(struct sk_buff ** head,struct sk_buff * skb)1316 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1317 struct sk_buff *skb)
1318 {
1319 const struct net_offload *ops;
1320 struct sk_buff **pp = NULL;
1321 struct sk_buff *p;
1322 const struct iphdr *iph;
1323 unsigned int hlen;
1324 unsigned int off;
1325 unsigned int id;
1326 int flush = 1;
1327 int proto;
1328
1329 off = skb_gro_offset(skb);
1330 hlen = off + sizeof(*iph);
1331 iph = skb_gro_header_fast(skb, off);
1332 if (skb_gro_header_hard(skb, hlen)) {
1333 iph = skb_gro_header_slow(skb, hlen, off);
1334 if (unlikely(!iph))
1335 goto out;
1336 }
1337
1338 proto = iph->protocol;
1339
1340 rcu_read_lock();
1341 ops = rcu_dereference(inet_offloads[proto]);
1342 if (!ops || !ops->callbacks.gro_receive)
1343 goto out_unlock;
1344
1345 if (*(u8 *)iph != 0x45)
1346 goto out_unlock;
1347
1348 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1349 goto out_unlock;
1350
1351 id = ntohl(*(__be32 *)&iph->id);
1352 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1353 id >>= 16;
1354
1355 for (p = *head; p; p = p->next) {
1356 struct iphdr *iph2;
1357
1358 if (!NAPI_GRO_CB(p)->same_flow)
1359 continue;
1360
1361 iph2 = (struct iphdr *)(p->data + off);
1362 /* The above works because, with the exception of the top
1363 * (inner most) layer, we only aggregate pkts with the same
1364 * hdr length so all the hdrs we'll need to verify will start
1365 * at the same offset.
1366 */
1367 if ((iph->protocol ^ iph2->protocol) |
1368 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1369 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1370 NAPI_GRO_CB(p)->same_flow = 0;
1371 continue;
1372 }
1373
1374 /* All fields must match except length and checksum. */
1375 NAPI_GRO_CB(p)->flush |=
1376 (iph->ttl ^ iph2->ttl) |
1377 (iph->tos ^ iph2->tos) |
1378 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1379
1380 /* Save the IP ID check to be included later when we get to
1381 * the transport layer so only the inner most IP ID is checked.
1382 * This is because some GSO/TSO implementations do not
1383 * correctly increment the IP ID for the outer hdrs.
1384 */
1385 NAPI_GRO_CB(p)->flush_id =
1386 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1387 NAPI_GRO_CB(p)->flush |= flush;
1388 }
1389
1390 NAPI_GRO_CB(skb)->flush |= flush;
1391 skb_set_network_header(skb, off);
1392 /* The above will be needed by the transport layer if there is one
1393 * immediately following this IP hdr.
1394 */
1395
1396 /* Note : No need to call skb_gro_postpull_rcsum() here,
1397 * as we already checked checksum over ipv4 header was 0
1398 */
1399 skb_gro_pull(skb, sizeof(*iph));
1400 skb_set_transport_header(skb, skb_gro_offset(skb));
1401
1402 pp = ops->callbacks.gro_receive(head, skb);
1403
1404 out_unlock:
1405 rcu_read_unlock();
1406
1407 out:
1408 NAPI_GRO_CB(skb)->flush |= flush;
1409
1410 return pp;
1411 }
1412
ipip_gro_receive(struct sk_buff ** head,struct sk_buff * skb)1413 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1414 struct sk_buff *skb)
1415 {
1416 if (NAPI_GRO_CB(skb)->encap_mark) {
1417 NAPI_GRO_CB(skb)->flush = 1;
1418 return NULL;
1419 }
1420
1421 NAPI_GRO_CB(skb)->encap_mark = 1;
1422
1423 return inet_gro_receive(head, skb);
1424 }
1425
1426 #define SECONDS_PER_DAY 86400
1427
1428 /* inet_current_timestamp - Return IP network timestamp
1429 *
1430 * Return milliseconds since midnight in network byte order.
1431 */
inet_current_timestamp(void)1432 __be32 inet_current_timestamp(void)
1433 {
1434 u32 secs;
1435 u32 msecs;
1436 struct timespec64 ts;
1437
1438 ktime_get_real_ts64(&ts);
1439
1440 /* Get secs since midnight. */
1441 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1442 /* Convert to msecs. */
1443 msecs = secs * MSEC_PER_SEC;
1444 /* Convert nsec to msec. */
1445 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1446
1447 /* Convert to network byte order. */
1448 return htons(msecs);
1449 }
1450 EXPORT_SYMBOL(inet_current_timestamp);
1451
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1452 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1453 {
1454 if (sk->sk_family == AF_INET)
1455 return ip_recv_error(sk, msg, len, addr_len);
1456 #if IS_ENABLED(CONFIG_IPV6)
1457 if (sk->sk_family == AF_INET6)
1458 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1459 #endif
1460 return -EINVAL;
1461 }
1462
inet_gro_complete(struct sk_buff * skb,int nhoff)1463 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1464 {
1465 __be16 newlen = htons(skb->len - nhoff);
1466 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1467 const struct net_offload *ops;
1468 int proto = iph->protocol;
1469 int err = -ENOSYS;
1470
1471 if (skb->encapsulation)
1472 skb_set_inner_network_header(skb, nhoff);
1473
1474 csum_replace2(&iph->check, iph->tot_len, newlen);
1475 iph->tot_len = newlen;
1476
1477 rcu_read_lock();
1478 ops = rcu_dereference(inet_offloads[proto]);
1479 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1480 goto out_unlock;
1481
1482 /* Only need to add sizeof(*iph) to get to the next hdr below
1483 * because any hdr with option will have been flushed in
1484 * inet_gro_receive().
1485 */
1486 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1487
1488 out_unlock:
1489 rcu_read_unlock();
1490
1491 return err;
1492 }
1493
ipip_gro_complete(struct sk_buff * skb,int nhoff)1494 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1495 {
1496 skb->encapsulation = 1;
1497 skb_shinfo(skb)->gso_type |= SKB_GSO_IPIP;
1498 return inet_gro_complete(skb, nhoff);
1499 }
1500
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1501 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1502 unsigned short type, unsigned char protocol,
1503 struct net *net)
1504 {
1505 struct socket *sock;
1506 int rc = sock_create_kern(family, type, protocol, &sock);
1507
1508 if (rc == 0) {
1509 *sk = sock->sk;
1510 (*sk)->sk_allocation = GFP_ATOMIC;
1511 /*
1512 * Unhash it so that IP input processing does not even see it,
1513 * we do not wish this socket to see incoming packets.
1514 */
1515 (*sk)->sk_prot->unhash(*sk);
1516
1517 sk_change_net(*sk, net);
1518 }
1519 return rc;
1520 }
1521 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1522
snmp_fold_field(void __percpu * mib,int offt)1523 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1524 {
1525 unsigned long res = 0;
1526 int i;
1527
1528 for_each_possible_cpu(i)
1529 res += *(((unsigned long *) per_cpu_ptr(mib, i)) + offt);
1530 return res;
1531 }
1532 EXPORT_SYMBOL_GPL(snmp_fold_field);
1533
1534 #if BITS_PER_LONG==32
1535
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1536 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1537 {
1538 u64 res = 0;
1539 int cpu;
1540
1541 for_each_possible_cpu(cpu) {
1542 void *bhptr;
1543 struct u64_stats_sync *syncp;
1544 u64 v;
1545 unsigned int start;
1546
1547 bhptr = per_cpu_ptr(mib, cpu);
1548 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1549 do {
1550 start = u64_stats_fetch_begin_irq(syncp);
1551 v = *(((u64 *) bhptr) + offt);
1552 } while (u64_stats_fetch_retry_irq(syncp, start));
1553
1554 res += v;
1555 }
1556 return res;
1557 }
1558 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1559 #endif
1560
1561 #ifdef CONFIG_IP_MULTICAST
1562 static const struct net_protocol igmp_protocol = {
1563 .handler = igmp_rcv,
1564 .netns_ok = 1,
1565 };
1566 #endif
1567
1568 static const struct net_protocol tcp_protocol = {
1569 .early_demux = tcp_v4_early_demux,
1570 .handler = tcp_v4_rcv,
1571 .err_handler = tcp_v4_err,
1572 .no_policy = 1,
1573 .netns_ok = 1,
1574 .icmp_strict_tag_validation = 1,
1575 };
1576
1577 static const struct net_protocol udp_protocol = {
1578 .early_demux = udp_v4_early_demux,
1579 .handler = udp_rcv,
1580 .err_handler = udp_err,
1581 .no_policy = 1,
1582 .netns_ok = 1,
1583 };
1584
1585 static const struct net_protocol icmp_protocol = {
1586 .handler = icmp_rcv,
1587 .err_handler = icmp_err,
1588 .no_policy = 1,
1589 .netns_ok = 1,
1590 };
1591
ipv4_mib_init_net(struct net * net)1592 static __net_init int ipv4_mib_init_net(struct net *net)
1593 {
1594 int i;
1595
1596 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1597 if (!net->mib.tcp_statistics)
1598 goto err_tcp_mib;
1599 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1600 if (!net->mib.ip_statistics)
1601 goto err_ip_mib;
1602
1603 for_each_possible_cpu(i) {
1604 struct ipstats_mib *af_inet_stats;
1605 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1606 u64_stats_init(&af_inet_stats->syncp);
1607 }
1608
1609 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1610 if (!net->mib.net_statistics)
1611 goto err_net_mib;
1612 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1613 if (!net->mib.udp_statistics)
1614 goto err_udp_mib;
1615 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1616 if (!net->mib.udplite_statistics)
1617 goto err_udplite_mib;
1618 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1619 if (!net->mib.icmp_statistics)
1620 goto err_icmp_mib;
1621 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1622 GFP_KERNEL);
1623 if (!net->mib.icmpmsg_statistics)
1624 goto err_icmpmsg_mib;
1625
1626 tcp_mib_init(net);
1627 return 0;
1628
1629 err_icmpmsg_mib:
1630 free_percpu(net->mib.icmp_statistics);
1631 err_icmp_mib:
1632 free_percpu(net->mib.udplite_statistics);
1633 err_udplite_mib:
1634 free_percpu(net->mib.udp_statistics);
1635 err_udp_mib:
1636 free_percpu(net->mib.net_statistics);
1637 err_net_mib:
1638 free_percpu(net->mib.ip_statistics);
1639 err_ip_mib:
1640 free_percpu(net->mib.tcp_statistics);
1641 err_tcp_mib:
1642 return -ENOMEM;
1643 }
1644
ipv4_mib_exit_net(struct net * net)1645 static __net_exit void ipv4_mib_exit_net(struct net *net)
1646 {
1647 kfree(net->mib.icmpmsg_statistics);
1648 free_percpu(net->mib.icmp_statistics);
1649 free_percpu(net->mib.udplite_statistics);
1650 free_percpu(net->mib.udp_statistics);
1651 free_percpu(net->mib.net_statistics);
1652 free_percpu(net->mib.ip_statistics);
1653 free_percpu(net->mib.tcp_statistics);
1654 }
1655
1656 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1657 .init = ipv4_mib_init_net,
1658 .exit = ipv4_mib_exit_net,
1659 };
1660
init_ipv4_mibs(void)1661 static int __init init_ipv4_mibs(void)
1662 {
1663 return register_pernet_subsys(&ipv4_mib_ops);
1664 }
1665
inet_init_net(struct net * net)1666 static __net_init int inet_init_net(struct net *net)
1667 {
1668 /*
1669 * Set defaults for local port range
1670 */
1671 seqlock_init(&net->ipv4.ip_local_ports.lock);
1672 net->ipv4.ip_local_ports.range[0] = 32768;
1673 net->ipv4.ip_local_ports.range[1] = 61000;
1674
1675 seqlock_init(&net->ipv4.ping_group_range.lock);
1676 /*
1677 * Sane defaults - nobody may create ping sockets.
1678 * Boot scripts should set this to distro-specific group.
1679 */
1680 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1681 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1682 return 0;
1683 }
1684
inet_exit_net(struct net * net)1685 static __net_exit void inet_exit_net(struct net *net)
1686 {
1687 }
1688
1689 static __net_initdata struct pernet_operations af_inet_ops = {
1690 .init = inet_init_net,
1691 .exit = inet_exit_net,
1692 };
1693
init_inet_pernet_ops(void)1694 static int __init init_inet_pernet_ops(void)
1695 {
1696 return register_pernet_subsys(&af_inet_ops);
1697 }
1698
1699 static int ipv4_proc_init(void);
1700
1701 /*
1702 * IP protocol layer initialiser
1703 */
1704
1705 static struct packet_offload ip_packet_offload __read_mostly = {
1706 .type = cpu_to_be16(ETH_P_IP),
1707 .callbacks = {
1708 .gso_segment = inet_gso_segment,
1709 .gro_receive = inet_gro_receive,
1710 .gro_complete = inet_gro_complete,
1711 },
1712 };
1713
1714 static const struct net_offload ipip_offload = {
1715 .callbacks = {
1716 .gso_segment = inet_gso_segment,
1717 .gro_receive = ipip_gro_receive,
1718 .gro_complete = ipip_gro_complete,
1719 },
1720 };
1721
ipv4_offload_init(void)1722 static int __init ipv4_offload_init(void)
1723 {
1724 /*
1725 * Add offloads
1726 */
1727 if (udpv4_offload_init() < 0)
1728 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1729 if (tcpv4_offload_init() < 0)
1730 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1731
1732 dev_add_offload(&ip_packet_offload);
1733 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1734 return 0;
1735 }
1736
1737 fs_initcall(ipv4_offload_init);
1738
1739 static struct packet_type ip_packet_type __read_mostly = {
1740 .type = cpu_to_be16(ETH_P_IP),
1741 .func = ip_rcv,
1742 };
1743
inet_init(void)1744 static int __init inet_init(void)
1745 {
1746 struct inet_protosw *q;
1747 struct list_head *r;
1748 int rc = -EINVAL;
1749
1750 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1751
1752 rc = proto_register(&tcp_prot, 1);
1753 if (rc)
1754 goto out;
1755
1756 rc = proto_register(&udp_prot, 1);
1757 if (rc)
1758 goto out_unregister_tcp_proto;
1759
1760 rc = proto_register(&raw_prot, 1);
1761 if (rc)
1762 goto out_unregister_udp_proto;
1763
1764 rc = proto_register(&ping_prot, 1);
1765 if (rc)
1766 goto out_unregister_raw_proto;
1767
1768 /*
1769 * Tell SOCKET that we are alive...
1770 */
1771
1772 (void)sock_register(&inet_family_ops);
1773
1774 #ifdef CONFIG_SYSCTL
1775 ip_static_sysctl_init();
1776 #endif
1777
1778 /*
1779 * Add all the base protocols.
1780 */
1781
1782 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1783 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1784 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1785 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1786 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1787 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1788 #ifdef CONFIG_IP_MULTICAST
1789 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1790 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1791 #endif
1792
1793 /* Register the socket-side information for inet_create. */
1794 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1795 INIT_LIST_HEAD(r);
1796
1797 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1798 inet_register_protosw(q);
1799
1800 /*
1801 * Set the ARP module up
1802 */
1803
1804 arp_init();
1805
1806 /*
1807 * Set the IP module up
1808 */
1809
1810 ip_init();
1811
1812 tcp_v4_init();
1813
1814 /* Setup TCP slab cache for open requests. */
1815 tcp_init();
1816
1817 /* Setup UDP memory threshold */
1818 udp_init();
1819
1820 /* Add UDP-Lite (RFC 3828) */
1821 udplite4_register();
1822
1823 ping_init();
1824
1825 /*
1826 * Set the ICMP layer up
1827 */
1828
1829 if (icmp_init() < 0)
1830 panic("Failed to create the ICMP control socket.\n");
1831
1832 /*
1833 * Initialise the multicast router
1834 */
1835 #if defined(CONFIG_IP_MROUTE)
1836 if (ip_mr_init())
1837 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1838 #endif
1839
1840 if (init_inet_pernet_ops())
1841 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1842 /*
1843 * Initialise per-cpu ipv4 mibs
1844 */
1845
1846 if (init_ipv4_mibs())
1847 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1848
1849 ipv4_proc_init();
1850
1851 ipfrag_init();
1852
1853 dev_add_pack(&ip_packet_type);
1854
1855 rc = 0;
1856 out:
1857 return rc;
1858 out_unregister_raw_proto:
1859 proto_unregister(&raw_prot);
1860 out_unregister_udp_proto:
1861 proto_unregister(&udp_prot);
1862 out_unregister_tcp_proto:
1863 proto_unregister(&tcp_prot);
1864 goto out;
1865 }
1866
1867 fs_initcall(inet_init);
1868
1869 /* ------------------------------------------------------------------------ */
1870
1871 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)1872 static int __init ipv4_proc_init(void)
1873 {
1874 int rc = 0;
1875
1876 if (raw_proc_init())
1877 goto out_raw;
1878 if (tcp4_proc_init())
1879 goto out_tcp;
1880 if (udp4_proc_init())
1881 goto out_udp;
1882 if (ping_proc_init())
1883 goto out_ping;
1884 if (ip_misc_proc_init())
1885 goto out_misc;
1886 out:
1887 return rc;
1888 out_misc:
1889 ping_proc_exit();
1890 out_ping:
1891 udp4_proc_exit();
1892 out_udp:
1893 tcp4_proc_exit();
1894 out_tcp:
1895 raw_proc_exit();
1896 out_raw:
1897 rc = -ENOMEM;
1898 goto out;
1899 }
1900
1901 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)1902 static int __init ipv4_proc_init(void)
1903 {
1904 return 0;
1905 }
1906 #endif /* CONFIG_PROC_FS */
1907
1908 MODULE_ALIAS_NETPROTO(PF_INET);
1909
1910