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/kmod.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/ip_tunnels.h>
117 #include <net/xfrm.h>
118 #include <net/net_namespace.h>
119 #include <net/secure_seq.h>
120 #ifdef CONFIG_IP_MROUTE
121 #include <linux/mroute.h>
122 #endif
123 #include <net/l3mdev.h>
124
125 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
126 #include <linux/android_aid.h>
127
current_has_network(void)128 static inline int current_has_network(void)
129 {
130 return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
131 }
132 #else
current_has_network(void)133 static inline int current_has_network(void)
134 {
135 return 1;
136 }
137 #endif
138
139 /* The inetsw table contains everything that inet_create needs to
140 * build a new socket.
141 */
142 static struct list_head inetsw[SOCK_MAX];
143 static DEFINE_SPINLOCK(inetsw_lock);
144
145 /* New destruction routine */
146
inet_sock_destruct(struct sock * sk)147 void inet_sock_destruct(struct sock *sk)
148 {
149 struct inet_sock *inet = inet_sk(sk);
150
151 __skb_queue_purge(&sk->sk_receive_queue);
152 __skb_queue_purge(&sk->sk_error_queue);
153
154 sk_mem_reclaim(sk);
155
156 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
157 pr_err("Attempt to release TCP socket in state %d %p\n",
158 sk->sk_state, sk);
159 return;
160 }
161 if (!sock_flag(sk, SOCK_DEAD)) {
162 pr_err("Attempt to release alive inet socket %p\n", sk);
163 return;
164 }
165
166 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
167 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
168 WARN_ON(sk->sk_wmem_queued);
169 WARN_ON(sk->sk_forward_alloc);
170
171 kfree(rcu_dereference_protected(inet->inet_opt, 1));
172 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
173 dst_release(sk->sk_rx_dst);
174 sk_refcnt_debug_dec(sk);
175 }
176 EXPORT_SYMBOL(inet_sock_destruct);
177
178 /*
179 * The routines beyond this point handle the behaviour of an AF_INET
180 * socket object. Mostly it punts to the subprotocols of IP to do
181 * the work.
182 */
183
184 /*
185 * Automatically bind an unbound socket.
186 */
187
inet_autobind(struct sock * sk)188 static int inet_autobind(struct sock *sk)
189 {
190 struct inet_sock *inet;
191 /* We may need to bind the socket. */
192 lock_sock(sk);
193 inet = inet_sk(sk);
194 if (!inet->inet_num) {
195 if (sk->sk_prot->get_port(sk, 0)) {
196 release_sock(sk);
197 return -EAGAIN;
198 }
199 inet->inet_sport = htons(inet->inet_num);
200 }
201 release_sock(sk);
202 return 0;
203 }
204
205 /*
206 * Move a socket into listening state.
207 */
inet_listen(struct socket * sock,int backlog)208 int inet_listen(struct socket *sock, int backlog)
209 {
210 struct sock *sk = sock->sk;
211 unsigned char old_state;
212 int err;
213
214 lock_sock(sk);
215
216 err = -EINVAL;
217 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
218 goto out;
219
220 old_state = sk->sk_state;
221 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
222 goto out;
223
224 /* Really, if the socket is already in listen state
225 * we can only allow the backlog to be adjusted.
226 */
227 if (old_state != TCP_LISTEN) {
228 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
229 * Note that only TCP sockets (SOCK_STREAM) will reach here.
230 * Also fastopen backlog may already been set via the option
231 * because the socket was in TCP_LISTEN state previously but
232 * was shutdown() rather than close().
233 */
234 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
235 (sysctl_tcp_fastopen & TFO_SERVER_ENABLE) &&
236 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
237 fastopen_queue_tune(sk, backlog);
238 tcp_fastopen_init_key_once(true);
239 }
240
241 err = inet_csk_listen_start(sk, backlog);
242 if (err)
243 goto out;
244 }
245 sk->sk_max_ack_backlog = backlog;
246 err = 0;
247
248 out:
249 release_sock(sk);
250 return err;
251 }
252 EXPORT_SYMBOL(inet_listen);
253
254 /*
255 * Create an inet socket.
256 */
257
inet_create(struct net * net,struct socket * sock,int protocol,int kern)258 static int inet_create(struct net *net, struct socket *sock, int protocol,
259 int kern)
260 {
261 struct sock *sk;
262 struct inet_protosw *answer;
263 struct inet_sock *inet;
264 struct proto *answer_prot;
265 unsigned char answer_flags;
266 int try_loading_module = 0;
267 int err;
268
269 if (protocol < 0 || protocol >= IPPROTO_MAX)
270 return -EINVAL;
271
272 if (!current_has_network())
273 return -EACCES;
274
275 sock->state = SS_UNCONNECTED;
276
277 /* Look for the requested type/protocol pair. */
278 lookup_protocol:
279 err = -ESOCKTNOSUPPORT;
280 rcu_read_lock();
281 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
282
283 err = 0;
284 /* Check the non-wild match. */
285 if (protocol == answer->protocol) {
286 if (protocol != IPPROTO_IP)
287 break;
288 } else {
289 /* Check for the two wild cases. */
290 if (IPPROTO_IP == protocol) {
291 protocol = answer->protocol;
292 break;
293 }
294 if (IPPROTO_IP == answer->protocol)
295 break;
296 }
297 err = -EPROTONOSUPPORT;
298 }
299
300 if (unlikely(err)) {
301 if (try_loading_module < 2) {
302 rcu_read_unlock();
303 /*
304 * Be more specific, e.g. net-pf-2-proto-132-type-1
305 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
306 */
307 if (++try_loading_module == 1)
308 request_module("net-pf-%d-proto-%d-type-%d",
309 PF_INET, protocol, sock->type);
310 /*
311 * Fall back to generic, e.g. net-pf-2-proto-132
312 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
313 */
314 else
315 request_module("net-pf-%d-proto-%d",
316 PF_INET, protocol);
317 goto lookup_protocol;
318 } else
319 goto out_rcu_unlock;
320 }
321
322 err = -EPERM;
323 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
324 goto out_rcu_unlock;
325
326 sock->ops = answer->ops;
327 answer_prot = answer->prot;
328 answer_flags = answer->flags;
329 rcu_read_unlock();
330
331 WARN_ON(!answer_prot->slab);
332
333 err = -ENOBUFS;
334 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
335 if (!sk)
336 goto out;
337
338 err = 0;
339 if (INET_PROTOSW_REUSE & answer_flags)
340 sk->sk_reuse = SK_CAN_REUSE;
341
342 inet = inet_sk(sk);
343 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
344
345 inet->nodefrag = 0;
346
347 if (SOCK_RAW == sock->type) {
348 inet->inet_num = protocol;
349 if (IPPROTO_RAW == protocol)
350 inet->hdrincl = 1;
351 }
352
353 if (net->ipv4.sysctl_ip_no_pmtu_disc)
354 inet->pmtudisc = IP_PMTUDISC_DONT;
355 else
356 inet->pmtudisc = IP_PMTUDISC_WANT;
357
358 inet->inet_id = 0;
359
360 sock_init_data(sock, sk);
361
362 sk->sk_destruct = inet_sock_destruct;
363 sk->sk_protocol = protocol;
364 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
365
366 inet->uc_ttl = -1;
367 inet->mc_loop = 1;
368 inet->mc_ttl = 1;
369 inet->mc_all = 1;
370 inet->mc_index = 0;
371 inet->mc_list = NULL;
372 inet->rcv_tos = 0;
373
374 sk_refcnt_debug_inc(sk);
375
376 if (inet->inet_num) {
377 /* It assumes that any protocol which allows
378 * the user to assign a number at socket
379 * creation time automatically
380 * shares.
381 */
382 inet->inet_sport = htons(inet->inet_num);
383 /* Add to protocol hash chains. */
384 err = sk->sk_prot->hash(sk);
385 if (err) {
386 sk_common_release(sk);
387 goto out;
388 }
389 }
390
391 if (sk->sk_prot->init) {
392 err = sk->sk_prot->init(sk);
393 if (err)
394 sk_common_release(sk);
395 }
396 out:
397 return err;
398 out_rcu_unlock:
399 rcu_read_unlock();
400 goto out;
401 }
402
403
404 /*
405 * The peer socket should always be NULL (or else). When we call this
406 * function we are destroying the object and from then on nobody
407 * should refer to it.
408 */
inet_release(struct socket * sock)409 int inet_release(struct socket *sock)
410 {
411 struct sock *sk = sock->sk;
412
413 if (sk) {
414 long timeout;
415
416 #ifdef CONFIG_NETFILTER_XT_MATCH_QTAGUID
417 qtaguid_untag(sock, true);
418 #endif
419 /* Applications forget to leave groups before exiting */
420 ip_mc_drop_socket(sk);
421
422 /* If linger is set, we don't return until the close
423 * is complete. Otherwise we return immediately. The
424 * actually closing is done the same either way.
425 *
426 * If the close is due to the process exiting, we never
427 * linger..
428 */
429 timeout = 0;
430 if (sock_flag(sk, SOCK_LINGER) &&
431 !(current->flags & PF_EXITING))
432 timeout = sk->sk_lingertime;
433 sock->sk = NULL;
434 sk->sk_prot->close(sk, timeout);
435 }
436 return 0;
437 }
438 EXPORT_SYMBOL(inet_release);
439
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)440 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
441 {
442 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
443 struct sock *sk = sock->sk;
444 struct inet_sock *inet = inet_sk(sk);
445 struct net *net = sock_net(sk);
446 unsigned short snum;
447 int chk_addr_ret;
448 u32 tb_id = RT_TABLE_LOCAL;
449 int err;
450
451 /* If the socket has its own bind function then use it. (RAW) */
452 if (sk->sk_prot->bind) {
453 err = sk->sk_prot->bind(sk, uaddr, addr_len);
454 goto out;
455 }
456 err = -EINVAL;
457 if (addr_len < sizeof(struct sockaddr_in))
458 goto out;
459
460 if (addr->sin_family != AF_INET) {
461 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
462 * only if s_addr is INADDR_ANY.
463 */
464 err = -EAFNOSUPPORT;
465 if (addr->sin_family != AF_UNSPEC ||
466 addr->sin_addr.s_addr != htonl(INADDR_ANY))
467 goto out;
468 }
469
470 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
471 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
472
473 /* Not specified by any standard per-se, however it breaks too
474 * many applications when removed. It is unfortunate since
475 * allowing applications to make a non-local bind solves
476 * several problems with systems using dynamic addressing.
477 * (ie. your servers still start up even if your ISDN link
478 * is temporarily down)
479 */
480 err = -EADDRNOTAVAIL;
481 if (!net->ipv4.sysctl_ip_nonlocal_bind &&
482 !(inet->freebind || inet->transparent) &&
483 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
484 chk_addr_ret != RTN_LOCAL &&
485 chk_addr_ret != RTN_MULTICAST &&
486 chk_addr_ret != RTN_BROADCAST)
487 goto out;
488
489 snum = ntohs(addr->sin_port);
490 err = -EACCES;
491 if (snum && snum < PROT_SOCK &&
492 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
493 goto out;
494
495 /* We keep a pair of addresses. rcv_saddr is the one
496 * used by hash lookups, and saddr is used for transmit.
497 *
498 * In the BSD API these are the same except where it
499 * would be illegal to use them (multicast/broadcast) in
500 * which case the sending device address is used.
501 */
502 lock_sock(sk);
503
504 /* Check these errors (active socket, double bind). */
505 err = -EINVAL;
506 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
507 goto out_release_sock;
508
509 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
510 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
511 inet->inet_saddr = 0; /* Use device */
512
513 /* Make sure we are allowed to bind here. */
514 if ((snum || !inet->bind_address_no_port) &&
515 sk->sk_prot->get_port(sk, snum)) {
516 inet->inet_saddr = inet->inet_rcv_saddr = 0;
517 err = -EADDRINUSE;
518 goto out_release_sock;
519 }
520
521 if (inet->inet_rcv_saddr)
522 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
523 if (snum)
524 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
525 inet->inet_sport = htons(inet->inet_num);
526 inet->inet_daddr = 0;
527 inet->inet_dport = 0;
528 sk_dst_reset(sk);
529 err = 0;
530 out_release_sock:
531 release_sock(sk);
532 out:
533 return err;
534 }
535 EXPORT_SYMBOL(inet_bind);
536
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)537 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
538 int addr_len, int flags)
539 {
540 struct sock *sk = sock->sk;
541
542 if (addr_len < sizeof(uaddr->sa_family))
543 return -EINVAL;
544 if (uaddr->sa_family == AF_UNSPEC)
545 return sk->sk_prot->disconnect(sk, flags);
546
547 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
548 return -EAGAIN;
549 return sk->sk_prot->connect(sk, uaddr, addr_len);
550 }
551 EXPORT_SYMBOL(inet_dgram_connect);
552
inet_wait_for_connect(struct sock * sk,long timeo,int writebias)553 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
554 {
555 DEFINE_WAIT_FUNC(wait, woken_wake_function);
556
557 add_wait_queue(sk_sleep(sk), &wait);
558 sk->sk_write_pending += writebias;
559
560 /* Basic assumption: if someone sets sk->sk_err, he _must_
561 * change state of the socket from TCP_SYN_*.
562 * Connect() does not allow to get error notifications
563 * without closing the socket.
564 */
565 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
566 release_sock(sk);
567 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
568 lock_sock(sk);
569 if (signal_pending(current) || !timeo)
570 break;
571 }
572 remove_wait_queue(sk_sleep(sk), &wait);
573 sk->sk_write_pending -= writebias;
574 return timeo;
575 }
576
577 /*
578 * Connect to a remote host. There is regrettably still a little
579 * TCP 'magic' in here.
580 */
__inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)581 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
582 int addr_len, int flags)
583 {
584 struct sock *sk = sock->sk;
585 int err;
586 long timeo;
587
588 /*
589 * uaddr can be NULL and addr_len can be 0 if:
590 * sk is a TCP fastopen active socket and
591 * TCP_FASTOPEN_CONNECT sockopt is set and
592 * we already have a valid cookie for this socket.
593 * In this case, user can call write() after connect().
594 * write() will invoke tcp_sendmsg_fastopen() which calls
595 * __inet_stream_connect().
596 */
597 if (uaddr) {
598 if (addr_len < sizeof(uaddr->sa_family))
599 return -EINVAL;
600
601 if (uaddr->sa_family == AF_UNSPEC) {
602 err = sk->sk_prot->disconnect(sk, flags);
603 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
604 goto out;
605 }
606 }
607
608 switch (sock->state) {
609 default:
610 err = -EINVAL;
611 goto out;
612 case SS_CONNECTED:
613 err = -EISCONN;
614 goto out;
615 case SS_CONNECTING:
616 if (inet_sk(sk)->defer_connect)
617 err = -EINPROGRESS;
618 else
619 err = -EALREADY;
620 /* Fall out of switch with err, set for this state */
621 break;
622 case SS_UNCONNECTED:
623 err = -EISCONN;
624 if (sk->sk_state != TCP_CLOSE)
625 goto out;
626
627 err = sk->sk_prot->connect(sk, uaddr, addr_len);
628 if (err < 0)
629 goto out;
630
631 sock->state = SS_CONNECTING;
632
633 if (!err && inet_sk(sk)->defer_connect)
634 goto out;
635
636 /* Just entered SS_CONNECTING state; the only
637 * difference is that return value in non-blocking
638 * case is EINPROGRESS, rather than EALREADY.
639 */
640 err = -EINPROGRESS;
641 break;
642 }
643
644 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
645
646 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
647 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
648 tcp_sk(sk)->fastopen_req &&
649 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
650
651 /* Error code is set above */
652 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
653 goto out;
654
655 err = sock_intr_errno(timeo);
656 if (signal_pending(current))
657 goto out;
658 }
659
660 /* Connection was closed by RST, timeout, ICMP error
661 * or another process disconnected us.
662 */
663 if (sk->sk_state == TCP_CLOSE)
664 goto sock_error;
665
666 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
667 * and error was received after socket entered established state.
668 * Hence, it is handled normally after connect() return successfully.
669 */
670
671 sock->state = SS_CONNECTED;
672 err = 0;
673 out:
674 return err;
675
676 sock_error:
677 err = sock_error(sk) ? : -ECONNABORTED;
678 sock->state = SS_UNCONNECTED;
679 if (sk->sk_prot->disconnect(sk, flags))
680 sock->state = SS_DISCONNECTING;
681 goto out;
682 }
683 EXPORT_SYMBOL(__inet_stream_connect);
684
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)685 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
686 int addr_len, int flags)
687 {
688 int err;
689
690 lock_sock(sock->sk);
691 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
692 release_sock(sock->sk);
693 return err;
694 }
695 EXPORT_SYMBOL(inet_stream_connect);
696
697 /*
698 * Accept a pending connection. The TCP layer now gives BSD semantics.
699 */
700
inet_accept(struct socket * sock,struct socket * newsock,int flags)701 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
702 {
703 struct sock *sk1 = sock->sk;
704 int err = -EINVAL;
705 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
706
707 if (!sk2)
708 goto do_err;
709
710 lock_sock(sk2);
711
712 sock_rps_record_flow(sk2);
713 WARN_ON(!((1 << sk2->sk_state) &
714 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
715 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
716
717 sock_graft(sk2, newsock);
718
719 newsock->state = SS_CONNECTED;
720 err = 0;
721 release_sock(sk2);
722 do_err:
723 return err;
724 }
725 EXPORT_SYMBOL(inet_accept);
726
727
728 /*
729 * This does both peername and sockname.
730 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int * uaddr_len,int peer)731 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
732 int *uaddr_len, int peer)
733 {
734 struct sock *sk = sock->sk;
735 struct inet_sock *inet = inet_sk(sk);
736 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
737
738 sin->sin_family = AF_INET;
739 if (peer) {
740 if (!inet->inet_dport ||
741 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
742 peer == 1))
743 return -ENOTCONN;
744 sin->sin_port = inet->inet_dport;
745 sin->sin_addr.s_addr = inet->inet_daddr;
746 } else {
747 __be32 addr = inet->inet_rcv_saddr;
748 if (!addr)
749 addr = inet->inet_saddr;
750 sin->sin_port = inet->inet_sport;
751 sin->sin_addr.s_addr = addr;
752 }
753 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
754 *uaddr_len = sizeof(*sin);
755 return 0;
756 }
757 EXPORT_SYMBOL(inet_getname);
758
inet_sendmsg(struct socket * sock,struct msghdr * msg,size_t size)759 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
760 {
761 struct sock *sk = sock->sk;
762
763 sock_rps_record_flow(sk);
764
765 /* We may need to bind the socket. */
766 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
767 inet_autobind(sk))
768 return -EAGAIN;
769
770 return sk->sk_prot->sendmsg(sk, msg, size);
771 }
772 EXPORT_SYMBOL(inet_sendmsg);
773
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)774 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
775 size_t size, int flags)
776 {
777 struct sock *sk = sock->sk;
778
779 sock_rps_record_flow(sk);
780
781 /* We may need to bind the socket. */
782 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
783 inet_autobind(sk))
784 return -EAGAIN;
785
786 if (sk->sk_prot->sendpage)
787 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
788 return sock_no_sendpage(sock, page, offset, size, flags);
789 }
790 EXPORT_SYMBOL(inet_sendpage);
791
inet_recvmsg(struct socket * sock,struct msghdr * msg,size_t size,int flags)792 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
793 int flags)
794 {
795 struct sock *sk = sock->sk;
796 int addr_len = 0;
797 int err;
798
799 sock_rps_record_flow(sk);
800
801 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
802 flags & ~MSG_DONTWAIT, &addr_len);
803 if (err >= 0)
804 msg->msg_namelen = addr_len;
805 return err;
806 }
807 EXPORT_SYMBOL(inet_recvmsg);
808
inet_shutdown(struct socket * sock,int how)809 int inet_shutdown(struct socket *sock, int how)
810 {
811 struct sock *sk = sock->sk;
812 int err = 0;
813
814 /* This should really check to make sure
815 * the socket is a TCP socket. (WHY AC...)
816 */
817 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
818 1->2 bit 2 snds.
819 2->3 */
820 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
821 return -EINVAL;
822
823 lock_sock(sk);
824 if (sock->state == SS_CONNECTING) {
825 if ((1 << sk->sk_state) &
826 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
827 sock->state = SS_DISCONNECTING;
828 else
829 sock->state = SS_CONNECTED;
830 }
831
832 switch (sk->sk_state) {
833 case TCP_CLOSE:
834 err = -ENOTCONN;
835 /* Hack to wake up other listeners, who can poll for
836 POLLHUP, even on eg. unconnected UDP sockets -- RR */
837 default:
838 sk->sk_shutdown |= how;
839 if (sk->sk_prot->shutdown)
840 sk->sk_prot->shutdown(sk, how);
841 break;
842
843 /* Remaining two branches are temporary solution for missing
844 * close() in multithreaded environment. It is _not_ a good idea,
845 * but we have no choice until close() is repaired at VFS level.
846 */
847 case TCP_LISTEN:
848 if (!(how & RCV_SHUTDOWN))
849 break;
850 /* Fall through */
851 case TCP_SYN_SENT:
852 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
853 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
854 break;
855 }
856
857 /* Wake up anyone sleeping in poll. */
858 sk->sk_state_change(sk);
859 release_sock(sk);
860 return err;
861 }
862 EXPORT_SYMBOL(inet_shutdown);
863
864 /*
865 * ioctl() calls you can issue on an INET socket. Most of these are
866 * device configuration and stuff and very rarely used. Some ioctls
867 * pass on to the socket itself.
868 *
869 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
870 * loads the devconfigure module does its configuring and unloads it.
871 * There's a good 20K of config code hanging around the kernel.
872 */
873
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)874 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
875 {
876 struct sock *sk = sock->sk;
877 int err = 0;
878 struct net *net = sock_net(sk);
879
880 switch (cmd) {
881 case SIOCGSTAMP:
882 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
883 break;
884 case SIOCGSTAMPNS:
885 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
886 break;
887 case SIOCADDRT:
888 case SIOCDELRT:
889 case SIOCRTMSG:
890 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
891 break;
892 case SIOCDARP:
893 case SIOCGARP:
894 case SIOCSARP:
895 err = arp_ioctl(net, cmd, (void __user *)arg);
896 break;
897 case SIOCGIFADDR:
898 case SIOCSIFADDR:
899 case SIOCGIFBRDADDR:
900 case SIOCSIFBRDADDR:
901 case SIOCGIFNETMASK:
902 case SIOCSIFNETMASK:
903 case SIOCGIFDSTADDR:
904 case SIOCSIFDSTADDR:
905 case SIOCSIFPFLAGS:
906 case SIOCGIFPFLAGS:
907 case SIOCSIFFLAGS:
908 err = devinet_ioctl(net, cmd, (void __user *)arg);
909 break;
910 default:
911 if (sk->sk_prot->ioctl)
912 err = sk->sk_prot->ioctl(sk, cmd, arg);
913 else
914 err = -ENOIOCTLCMD;
915 break;
916 }
917 return err;
918 }
919 EXPORT_SYMBOL(inet_ioctl);
920
921 #ifdef CONFIG_COMPAT
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)922 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
923 {
924 struct sock *sk = sock->sk;
925 int err = -ENOIOCTLCMD;
926
927 if (sk->sk_prot->compat_ioctl)
928 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
929
930 return err;
931 }
932 #endif
933
934 const struct proto_ops inet_stream_ops = {
935 .family = PF_INET,
936 .owner = THIS_MODULE,
937 .release = inet_release,
938 .bind = inet_bind,
939 .connect = inet_stream_connect,
940 .socketpair = sock_no_socketpair,
941 .accept = inet_accept,
942 .getname = inet_getname,
943 .poll = tcp_poll,
944 .ioctl = inet_ioctl,
945 .listen = inet_listen,
946 .shutdown = inet_shutdown,
947 .setsockopt = sock_common_setsockopt,
948 .getsockopt = sock_common_getsockopt,
949 .sendmsg = inet_sendmsg,
950 .recvmsg = inet_recvmsg,
951 .mmap = sock_no_mmap,
952 .sendpage = inet_sendpage,
953 .splice_read = tcp_splice_read,
954 .read_sock = tcp_read_sock,
955 .peek_len = tcp_peek_len,
956 #ifdef CONFIG_COMPAT
957 .compat_setsockopt = compat_sock_common_setsockopt,
958 .compat_getsockopt = compat_sock_common_getsockopt,
959 .compat_ioctl = inet_compat_ioctl,
960 #endif
961 };
962 EXPORT_SYMBOL(inet_stream_ops);
963
964 const struct proto_ops inet_dgram_ops = {
965 .family = PF_INET,
966 .owner = THIS_MODULE,
967 .release = inet_release,
968 .bind = inet_bind,
969 .connect = inet_dgram_connect,
970 .socketpair = sock_no_socketpair,
971 .accept = sock_no_accept,
972 .getname = inet_getname,
973 .poll = udp_poll,
974 .ioctl = inet_ioctl,
975 .listen = sock_no_listen,
976 .shutdown = inet_shutdown,
977 .setsockopt = sock_common_setsockopt,
978 .getsockopt = sock_common_getsockopt,
979 .sendmsg = inet_sendmsg,
980 .recvmsg = inet_recvmsg,
981 .mmap = sock_no_mmap,
982 .sendpage = inet_sendpage,
983 .set_peek_off = sk_set_peek_off,
984 #ifdef CONFIG_COMPAT
985 .compat_setsockopt = compat_sock_common_setsockopt,
986 .compat_getsockopt = compat_sock_common_getsockopt,
987 .compat_ioctl = inet_compat_ioctl,
988 #endif
989 };
990 EXPORT_SYMBOL(inet_dgram_ops);
991
992 /*
993 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
994 * udp_poll
995 */
996 static const struct proto_ops inet_sockraw_ops = {
997 .family = PF_INET,
998 .owner = THIS_MODULE,
999 .release = inet_release,
1000 .bind = inet_bind,
1001 .connect = inet_dgram_connect,
1002 .socketpair = sock_no_socketpair,
1003 .accept = sock_no_accept,
1004 .getname = inet_getname,
1005 .poll = datagram_poll,
1006 .ioctl = inet_ioctl,
1007 .listen = sock_no_listen,
1008 .shutdown = inet_shutdown,
1009 .setsockopt = sock_common_setsockopt,
1010 .getsockopt = sock_common_getsockopt,
1011 .sendmsg = inet_sendmsg,
1012 .recvmsg = inet_recvmsg,
1013 .mmap = sock_no_mmap,
1014 .sendpage = inet_sendpage,
1015 #ifdef CONFIG_COMPAT
1016 .compat_setsockopt = compat_sock_common_setsockopt,
1017 .compat_getsockopt = compat_sock_common_getsockopt,
1018 .compat_ioctl = inet_compat_ioctl,
1019 #endif
1020 };
1021
1022 static const struct net_proto_family inet_family_ops = {
1023 .family = PF_INET,
1024 .create = inet_create,
1025 .owner = THIS_MODULE,
1026 };
1027
1028 /* Upon startup we insert all the elements in inetsw_array[] into
1029 * the linked list inetsw.
1030 */
1031 static struct inet_protosw inetsw_array[] =
1032 {
1033 {
1034 .type = SOCK_STREAM,
1035 .protocol = IPPROTO_TCP,
1036 .prot = &tcp_prot,
1037 .ops = &inet_stream_ops,
1038 .flags = INET_PROTOSW_PERMANENT |
1039 INET_PROTOSW_ICSK,
1040 },
1041
1042 {
1043 .type = SOCK_DGRAM,
1044 .protocol = IPPROTO_UDP,
1045 .prot = &udp_prot,
1046 .ops = &inet_dgram_ops,
1047 .flags = INET_PROTOSW_PERMANENT,
1048 },
1049
1050 {
1051 .type = SOCK_DGRAM,
1052 .protocol = IPPROTO_ICMP,
1053 .prot = &ping_prot,
1054 .ops = &inet_sockraw_ops,
1055 .flags = INET_PROTOSW_REUSE,
1056 },
1057
1058 {
1059 .type = SOCK_RAW,
1060 .protocol = IPPROTO_IP, /* wild card */
1061 .prot = &raw_prot,
1062 .ops = &inet_sockraw_ops,
1063 .flags = INET_PROTOSW_REUSE,
1064 }
1065 };
1066
1067 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1068
inet_register_protosw(struct inet_protosw * p)1069 void inet_register_protosw(struct inet_protosw *p)
1070 {
1071 struct list_head *lh;
1072 struct inet_protosw *answer;
1073 int protocol = p->protocol;
1074 struct list_head *last_perm;
1075
1076 spin_lock_bh(&inetsw_lock);
1077
1078 if (p->type >= SOCK_MAX)
1079 goto out_illegal;
1080
1081 /* If we are trying to override a permanent protocol, bail. */
1082 last_perm = &inetsw[p->type];
1083 list_for_each(lh, &inetsw[p->type]) {
1084 answer = list_entry(lh, struct inet_protosw, list);
1085 /* Check only the non-wild match. */
1086 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1087 break;
1088 if (protocol == answer->protocol)
1089 goto out_permanent;
1090 last_perm = lh;
1091 }
1092
1093 /* Add the new entry after the last permanent entry if any, so that
1094 * the new entry does not override a permanent entry when matched with
1095 * a wild-card protocol. But it is allowed to override any existing
1096 * non-permanent entry. This means that when we remove this entry, the
1097 * system automatically returns to the old behavior.
1098 */
1099 list_add_rcu(&p->list, last_perm);
1100 out:
1101 spin_unlock_bh(&inetsw_lock);
1102
1103 return;
1104
1105 out_permanent:
1106 pr_err("Attempt to override permanent protocol %d\n", protocol);
1107 goto out;
1108
1109 out_illegal:
1110 pr_err("Ignoring attempt to register invalid socket type %d\n",
1111 p->type);
1112 goto out;
1113 }
1114 EXPORT_SYMBOL(inet_register_protosw);
1115
inet_unregister_protosw(struct inet_protosw * p)1116 void inet_unregister_protosw(struct inet_protosw *p)
1117 {
1118 if (INET_PROTOSW_PERMANENT & p->flags) {
1119 pr_err("Attempt to unregister permanent protocol %d\n",
1120 p->protocol);
1121 } else {
1122 spin_lock_bh(&inetsw_lock);
1123 list_del_rcu(&p->list);
1124 spin_unlock_bh(&inetsw_lock);
1125
1126 synchronize_net();
1127 }
1128 }
1129 EXPORT_SYMBOL(inet_unregister_protosw);
1130
inet_sk_reselect_saddr(struct sock * sk)1131 static int inet_sk_reselect_saddr(struct sock *sk)
1132 {
1133 struct inet_sock *inet = inet_sk(sk);
1134 __be32 old_saddr = inet->inet_saddr;
1135 __be32 daddr = inet->inet_daddr;
1136 struct flowi4 *fl4;
1137 struct rtable *rt;
1138 __be32 new_saddr;
1139 struct ip_options_rcu *inet_opt;
1140
1141 inet_opt = rcu_dereference_protected(inet->inet_opt,
1142 lockdep_sock_is_held(sk));
1143 if (inet_opt && inet_opt->opt.srr)
1144 daddr = inet_opt->opt.faddr;
1145
1146 /* Query new route. */
1147 fl4 = &inet->cork.fl.u.ip4;
1148 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1149 sk->sk_bound_dev_if, sk->sk_protocol,
1150 inet->inet_sport, inet->inet_dport, sk);
1151 if (IS_ERR(rt))
1152 return PTR_ERR(rt);
1153
1154 sk_setup_caps(sk, &rt->dst);
1155
1156 new_saddr = fl4->saddr;
1157
1158 if (new_saddr == old_saddr)
1159 return 0;
1160
1161 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1162 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1163 __func__, &old_saddr, &new_saddr);
1164 }
1165
1166 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1167
1168 /*
1169 * XXX The only one ugly spot where we need to
1170 * XXX really change the sockets identity after
1171 * XXX it has entered the hashes. -DaveM
1172 *
1173 * Besides that, it does not check for connection
1174 * uniqueness. Wait for troubles.
1175 */
1176 return __sk_prot_rehash(sk);
1177 }
1178
inet_sk_rebuild_header(struct sock * sk)1179 int inet_sk_rebuild_header(struct sock *sk)
1180 {
1181 struct inet_sock *inet = inet_sk(sk);
1182 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1183 __be32 daddr;
1184 struct ip_options_rcu *inet_opt;
1185 struct flowi4 *fl4;
1186 int err;
1187
1188 /* Route is OK, nothing to do. */
1189 if (rt)
1190 return 0;
1191
1192 /* Reroute. */
1193 rcu_read_lock();
1194 inet_opt = rcu_dereference(inet->inet_opt);
1195 daddr = inet->inet_daddr;
1196 if (inet_opt && inet_opt->opt.srr)
1197 daddr = inet_opt->opt.faddr;
1198 rcu_read_unlock();
1199 fl4 = &inet->cork.fl.u.ip4;
1200 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1201 inet->inet_dport, inet->inet_sport,
1202 sk->sk_protocol, RT_CONN_FLAGS(sk),
1203 sk->sk_bound_dev_if);
1204 if (!IS_ERR(rt)) {
1205 err = 0;
1206 sk_setup_caps(sk, &rt->dst);
1207 } else {
1208 err = PTR_ERR(rt);
1209
1210 /* Routing failed... */
1211 sk->sk_route_caps = 0;
1212 /*
1213 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1214 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1215 */
1216 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1217 sk->sk_state != TCP_SYN_SENT ||
1218 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1219 (err = inet_sk_reselect_saddr(sk)) != 0)
1220 sk->sk_err_soft = -err;
1221 }
1222
1223 return err;
1224 }
1225 EXPORT_SYMBOL(inet_sk_rebuild_header);
1226
inet_gso_segment(struct sk_buff * skb,netdev_features_t features)1227 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1228 netdev_features_t features)
1229 {
1230 bool udpfrag = false, fixedid = false, gso_partial, encap;
1231 struct sk_buff *segs = ERR_PTR(-EINVAL);
1232 const struct net_offload *ops;
1233 unsigned int offset = 0;
1234 struct iphdr *iph;
1235 int proto, tot_len;
1236 int nhoff;
1237 int ihl;
1238 int id;
1239
1240 skb_reset_network_header(skb);
1241 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1242 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1243 goto out;
1244
1245 iph = ip_hdr(skb);
1246 ihl = iph->ihl * 4;
1247 if (ihl < sizeof(*iph))
1248 goto out;
1249
1250 id = ntohs(iph->id);
1251 proto = iph->protocol;
1252
1253 /* Warning: after this point, iph might be no longer valid */
1254 if (unlikely(!pskb_may_pull(skb, ihl)))
1255 goto out;
1256 __skb_pull(skb, ihl);
1257
1258 encap = SKB_GSO_CB(skb)->encap_level > 0;
1259 if (encap)
1260 features &= skb->dev->hw_enc_features;
1261 SKB_GSO_CB(skb)->encap_level += ihl;
1262
1263 skb_reset_transport_header(skb);
1264
1265 segs = ERR_PTR(-EPROTONOSUPPORT);
1266
1267 if (!skb->encapsulation || encap) {
1268 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1269 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1270
1271 /* fixed ID is invalid if DF bit is not set */
1272 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1273 goto out;
1274 }
1275
1276 ops = rcu_dereference(inet_offloads[proto]);
1277 if (likely(ops && ops->callbacks.gso_segment))
1278 segs = ops->callbacks.gso_segment(skb, features);
1279
1280 if (IS_ERR_OR_NULL(segs))
1281 goto out;
1282
1283 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1284
1285 skb = segs;
1286 do {
1287 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1288 if (udpfrag) {
1289 iph->frag_off = htons(offset >> 3);
1290 if (skb->next)
1291 iph->frag_off |= htons(IP_MF);
1292 offset += skb->len - nhoff - ihl;
1293 tot_len = skb->len - nhoff;
1294 } else if (skb_is_gso(skb)) {
1295 if (!fixedid) {
1296 iph->id = htons(id);
1297 id += skb_shinfo(skb)->gso_segs;
1298 }
1299
1300 if (gso_partial)
1301 tot_len = skb_shinfo(skb)->gso_size +
1302 SKB_GSO_CB(skb)->data_offset +
1303 skb->head - (unsigned char *)iph;
1304 else
1305 tot_len = skb->len - nhoff;
1306 } else {
1307 if (!fixedid)
1308 iph->id = htons(id++);
1309 tot_len = skb->len - nhoff;
1310 }
1311 iph->tot_len = htons(tot_len);
1312 ip_send_check(iph);
1313 if (encap)
1314 skb_reset_inner_headers(skb);
1315 skb->network_header = (u8 *)iph - skb->head;
1316 } while ((skb = skb->next));
1317
1318 out:
1319 return segs;
1320 }
1321 EXPORT_SYMBOL(inet_gso_segment);
1322
inet_gro_receive(struct sk_buff ** head,struct sk_buff * skb)1323 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
1324 {
1325 const struct net_offload *ops;
1326 struct sk_buff **pp = NULL;
1327 struct sk_buff *p;
1328 const struct iphdr *iph;
1329 unsigned int hlen;
1330 unsigned int off;
1331 unsigned int id;
1332 int flush = 1;
1333 int proto;
1334
1335 off = skb_gro_offset(skb);
1336 hlen = off + sizeof(*iph);
1337 iph = skb_gro_header_fast(skb, off);
1338 if (skb_gro_header_hard(skb, hlen)) {
1339 iph = skb_gro_header_slow(skb, hlen, off);
1340 if (unlikely(!iph))
1341 goto out;
1342 }
1343
1344 proto = iph->protocol;
1345
1346 rcu_read_lock();
1347 ops = rcu_dereference(inet_offloads[proto]);
1348 if (!ops || !ops->callbacks.gro_receive)
1349 goto out_unlock;
1350
1351 if (*(u8 *)iph != 0x45)
1352 goto out_unlock;
1353
1354 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1355 goto out_unlock;
1356
1357 id = ntohl(*(__be32 *)&iph->id);
1358 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1359 id >>= 16;
1360
1361 for (p = *head; p; p = p->next) {
1362 struct iphdr *iph2;
1363 u16 flush_id;
1364
1365 if (!NAPI_GRO_CB(p)->same_flow)
1366 continue;
1367
1368 iph2 = (struct iphdr *)(p->data + off);
1369 /* The above works because, with the exception of the top
1370 * (inner most) layer, we only aggregate pkts with the same
1371 * hdr length so all the hdrs we'll need to verify will start
1372 * at the same offset.
1373 */
1374 if ((iph->protocol ^ iph2->protocol) |
1375 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1376 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1377 NAPI_GRO_CB(p)->same_flow = 0;
1378 continue;
1379 }
1380
1381 /* All fields must match except length and checksum. */
1382 NAPI_GRO_CB(p)->flush |=
1383 (iph->ttl ^ iph2->ttl) |
1384 (iph->tos ^ iph2->tos) |
1385 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1386
1387 NAPI_GRO_CB(p)->flush |= flush;
1388
1389 /* We need to store of the IP ID check to be included later
1390 * when we can verify that this packet does in fact belong
1391 * to a given flow.
1392 */
1393 flush_id = (u16)(id - ntohs(iph2->id));
1394
1395 /* This bit of code makes it much easier for us to identify
1396 * the cases where we are doing atomic vs non-atomic IP ID
1397 * checks. Specifically an atomic check can return IP ID
1398 * values 0 - 0xFFFF, while a non-atomic check can only
1399 * return 0 or 0xFFFF.
1400 */
1401 if (!NAPI_GRO_CB(p)->is_atomic ||
1402 !(iph->frag_off & htons(IP_DF))) {
1403 flush_id ^= NAPI_GRO_CB(p)->count;
1404 flush_id = flush_id ? 0xFFFF : 0;
1405 }
1406
1407 /* If the previous IP ID value was based on an atomic
1408 * datagram we can overwrite the value and ignore it.
1409 */
1410 if (NAPI_GRO_CB(skb)->is_atomic)
1411 NAPI_GRO_CB(p)->flush_id = flush_id;
1412 else
1413 NAPI_GRO_CB(p)->flush_id |= flush_id;
1414 }
1415
1416 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1417 NAPI_GRO_CB(skb)->flush |= flush;
1418 skb_set_network_header(skb, off);
1419 /* The above will be needed by the transport layer if there is one
1420 * immediately following this IP hdr.
1421 */
1422
1423 /* Note : No need to call skb_gro_postpull_rcsum() here,
1424 * as we already checked checksum over ipv4 header was 0
1425 */
1426 skb_gro_pull(skb, sizeof(*iph));
1427 skb_set_transport_header(skb, skb_gro_offset(skb));
1428
1429 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1430
1431 out_unlock:
1432 rcu_read_unlock();
1433
1434 out:
1435 NAPI_GRO_CB(skb)->flush |= flush;
1436
1437 return pp;
1438 }
1439 EXPORT_SYMBOL(inet_gro_receive);
1440
ipip_gro_receive(struct sk_buff ** head,struct sk_buff * skb)1441 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1442 struct sk_buff *skb)
1443 {
1444 if (NAPI_GRO_CB(skb)->encap_mark) {
1445 NAPI_GRO_CB(skb)->flush = 1;
1446 return NULL;
1447 }
1448
1449 NAPI_GRO_CB(skb)->encap_mark = 1;
1450
1451 return inet_gro_receive(head, skb);
1452 }
1453
1454 #define SECONDS_PER_DAY 86400
1455
1456 /* inet_current_timestamp - Return IP network timestamp
1457 *
1458 * Return milliseconds since midnight in network byte order.
1459 */
inet_current_timestamp(void)1460 __be32 inet_current_timestamp(void)
1461 {
1462 u32 secs;
1463 u32 msecs;
1464 struct timespec64 ts;
1465
1466 ktime_get_real_ts64(&ts);
1467
1468 /* Get secs since midnight. */
1469 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1470 /* Convert to msecs. */
1471 msecs = secs * MSEC_PER_SEC;
1472 /* Convert nsec to msec. */
1473 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1474
1475 /* Convert to network byte order. */
1476 return htonl(msecs);
1477 }
1478 EXPORT_SYMBOL(inet_current_timestamp);
1479
inet_recv_error(struct sock * sk,struct msghdr * msg,int len,int * addr_len)1480 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1481 {
1482 if (sk->sk_family == AF_INET)
1483 return ip_recv_error(sk, msg, len, addr_len);
1484 #if IS_ENABLED(CONFIG_IPV6)
1485 if (sk->sk_family == AF_INET6)
1486 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1487 #endif
1488 return -EINVAL;
1489 }
1490
inet_gro_complete(struct sk_buff * skb,int nhoff)1491 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1492 {
1493 __be16 newlen = htons(skb->len - nhoff);
1494 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1495 const struct net_offload *ops;
1496 int proto = iph->protocol;
1497 int err = -ENOSYS;
1498
1499 if (skb->encapsulation) {
1500 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1501 skb_set_inner_network_header(skb, nhoff);
1502 }
1503
1504 csum_replace2(&iph->check, iph->tot_len, newlen);
1505 iph->tot_len = newlen;
1506
1507 rcu_read_lock();
1508 ops = rcu_dereference(inet_offloads[proto]);
1509 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1510 goto out_unlock;
1511
1512 /* Only need to add sizeof(*iph) to get to the next hdr below
1513 * because any hdr with option will have been flushed in
1514 * inet_gro_receive().
1515 */
1516 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1517
1518 out_unlock:
1519 rcu_read_unlock();
1520
1521 return err;
1522 }
1523 EXPORT_SYMBOL(inet_gro_complete);
1524
ipip_gro_complete(struct sk_buff * skb,int nhoff)1525 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1526 {
1527 skb->encapsulation = 1;
1528 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1529 return inet_gro_complete(skb, nhoff);
1530 }
1531
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1532 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1533 unsigned short type, unsigned char protocol,
1534 struct net *net)
1535 {
1536 struct socket *sock;
1537 int rc = sock_create_kern(net, family, type, protocol, &sock);
1538
1539 if (rc == 0) {
1540 *sk = sock->sk;
1541 (*sk)->sk_allocation = GFP_ATOMIC;
1542 /*
1543 * Unhash it so that IP input processing does not even see it,
1544 * we do not wish this socket to see incoming packets.
1545 */
1546 (*sk)->sk_prot->unhash(*sk);
1547 }
1548 return rc;
1549 }
1550 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1551
snmp_get_cpu_field(void __percpu * mib,int cpu,int offt)1552 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1553 {
1554 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1555 }
1556 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1557
snmp_fold_field(void __percpu * mib,int offt)1558 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1559 {
1560 unsigned long res = 0;
1561 int i;
1562
1563 for_each_possible_cpu(i)
1564 res += snmp_get_cpu_field(mib, i, offt);
1565 return res;
1566 }
1567 EXPORT_SYMBOL_GPL(snmp_fold_field);
1568
1569 #if BITS_PER_LONG==32
1570
snmp_get_cpu_field64(void __percpu * mib,int cpu,int offt,size_t syncp_offset)1571 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1572 size_t syncp_offset)
1573 {
1574 void *bhptr;
1575 struct u64_stats_sync *syncp;
1576 u64 v;
1577 unsigned int start;
1578
1579 bhptr = per_cpu_ptr(mib, cpu);
1580 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1581 do {
1582 start = u64_stats_fetch_begin_irq(syncp);
1583 v = *(((u64 *)bhptr) + offt);
1584 } while (u64_stats_fetch_retry_irq(syncp, start));
1585
1586 return v;
1587 }
1588 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1589
snmp_fold_field64(void __percpu * mib,int offt,size_t syncp_offset)1590 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1591 {
1592 u64 res = 0;
1593 int cpu;
1594
1595 for_each_possible_cpu(cpu) {
1596 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1597 }
1598 return res;
1599 }
1600 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1601 #endif
1602
1603 #ifdef CONFIG_IP_MULTICAST
1604 static const struct net_protocol igmp_protocol = {
1605 .handler = igmp_rcv,
1606 .netns_ok = 1,
1607 };
1608 #endif
1609
1610 static const struct net_protocol tcp_protocol = {
1611 .early_demux = tcp_v4_early_demux,
1612 .handler = tcp_v4_rcv,
1613 .err_handler = tcp_v4_err,
1614 .no_policy = 1,
1615 .netns_ok = 1,
1616 .icmp_strict_tag_validation = 1,
1617 };
1618
1619 static const struct net_protocol udp_protocol = {
1620 .early_demux = udp_v4_early_demux,
1621 .handler = udp_rcv,
1622 .err_handler = udp_err,
1623 .no_policy = 1,
1624 .netns_ok = 1,
1625 };
1626
1627 static const struct net_protocol icmp_protocol = {
1628 .handler = icmp_rcv,
1629 .err_handler = icmp_err,
1630 .no_policy = 1,
1631 .netns_ok = 1,
1632 };
1633
ipv4_mib_init_net(struct net * net)1634 static __net_init int ipv4_mib_init_net(struct net *net)
1635 {
1636 int i;
1637
1638 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1639 if (!net->mib.tcp_statistics)
1640 goto err_tcp_mib;
1641 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1642 if (!net->mib.ip_statistics)
1643 goto err_ip_mib;
1644
1645 for_each_possible_cpu(i) {
1646 struct ipstats_mib *af_inet_stats;
1647 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1648 u64_stats_init(&af_inet_stats->syncp);
1649 }
1650
1651 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1652 if (!net->mib.net_statistics)
1653 goto err_net_mib;
1654 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1655 if (!net->mib.udp_statistics)
1656 goto err_udp_mib;
1657 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1658 if (!net->mib.udplite_statistics)
1659 goto err_udplite_mib;
1660 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1661 if (!net->mib.icmp_statistics)
1662 goto err_icmp_mib;
1663 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1664 GFP_KERNEL);
1665 if (!net->mib.icmpmsg_statistics)
1666 goto err_icmpmsg_mib;
1667
1668 tcp_mib_init(net);
1669 return 0;
1670
1671 err_icmpmsg_mib:
1672 free_percpu(net->mib.icmp_statistics);
1673 err_icmp_mib:
1674 free_percpu(net->mib.udplite_statistics);
1675 err_udplite_mib:
1676 free_percpu(net->mib.udp_statistics);
1677 err_udp_mib:
1678 free_percpu(net->mib.net_statistics);
1679 err_net_mib:
1680 free_percpu(net->mib.ip_statistics);
1681 err_ip_mib:
1682 free_percpu(net->mib.tcp_statistics);
1683 err_tcp_mib:
1684 return -ENOMEM;
1685 }
1686
ipv4_mib_exit_net(struct net * net)1687 static __net_exit void ipv4_mib_exit_net(struct net *net)
1688 {
1689 kfree(net->mib.icmpmsg_statistics);
1690 free_percpu(net->mib.icmp_statistics);
1691 free_percpu(net->mib.udplite_statistics);
1692 free_percpu(net->mib.udp_statistics);
1693 free_percpu(net->mib.net_statistics);
1694 free_percpu(net->mib.ip_statistics);
1695 free_percpu(net->mib.tcp_statistics);
1696 }
1697
1698 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1699 .init = ipv4_mib_init_net,
1700 .exit = ipv4_mib_exit_net,
1701 };
1702
init_ipv4_mibs(void)1703 static int __init init_ipv4_mibs(void)
1704 {
1705 return register_pernet_subsys(&ipv4_mib_ops);
1706 }
1707
inet_init_net(struct net * net)1708 static __net_init int inet_init_net(struct net *net)
1709 {
1710 /*
1711 * Set defaults for local port range
1712 */
1713 seqlock_init(&net->ipv4.ip_local_ports.lock);
1714 net->ipv4.ip_local_ports.range[0] = 32768;
1715 net->ipv4.ip_local_ports.range[1] = 60999;
1716
1717 seqlock_init(&net->ipv4.ping_group_range.lock);
1718 /*
1719 * Sane defaults - nobody may create ping sockets.
1720 * Boot scripts should set this to distro-specific group.
1721 */
1722 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1723 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1724
1725 /* Default values for sysctl-controlled parameters.
1726 * We set them here, in case sysctl is not compiled.
1727 */
1728 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1729 net->ipv4.sysctl_ip_dynaddr = 0;
1730 net->ipv4.sysctl_ip_early_demux = 1;
1731
1732 /* Some igmp sysctl, whose values are always used */
1733 net->ipv4.sysctl_igmp_max_memberships = 20;
1734 net->ipv4.sysctl_igmp_max_msf = 10;
1735 /* IGMP reports for link-local multicast groups are enabled by default */
1736 net->ipv4.sysctl_igmp_llm_reports = 1;
1737 net->ipv4.sysctl_igmp_qrv = 2;
1738
1739 return 0;
1740 }
1741
inet_exit_net(struct net * net)1742 static __net_exit void inet_exit_net(struct net *net)
1743 {
1744 }
1745
1746 static __net_initdata struct pernet_operations af_inet_ops = {
1747 .init = inet_init_net,
1748 .exit = inet_exit_net,
1749 };
1750
init_inet_pernet_ops(void)1751 static int __init init_inet_pernet_ops(void)
1752 {
1753 return register_pernet_subsys(&af_inet_ops);
1754 }
1755
1756 static int ipv4_proc_init(void);
1757
1758 /*
1759 * IP protocol layer initialiser
1760 */
1761
1762 static struct packet_offload ip_packet_offload __read_mostly = {
1763 .type = cpu_to_be16(ETH_P_IP),
1764 .callbacks = {
1765 .gso_segment = inet_gso_segment,
1766 .gro_receive = inet_gro_receive,
1767 .gro_complete = inet_gro_complete,
1768 },
1769 };
1770
1771 static const struct net_offload ipip_offload = {
1772 .callbacks = {
1773 .gso_segment = inet_gso_segment,
1774 .gro_receive = ipip_gro_receive,
1775 .gro_complete = ipip_gro_complete,
1776 },
1777 };
1778
ipv4_offload_init(void)1779 static int __init ipv4_offload_init(void)
1780 {
1781 /*
1782 * Add offloads
1783 */
1784 if (udpv4_offload_init() < 0)
1785 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1786 if (tcpv4_offload_init() < 0)
1787 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1788
1789 dev_add_offload(&ip_packet_offload);
1790 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1791 return 0;
1792 }
1793
1794 fs_initcall(ipv4_offload_init);
1795
1796 static struct packet_type ip_packet_type __read_mostly = {
1797 .type = cpu_to_be16(ETH_P_IP),
1798 .func = ip_rcv,
1799 };
1800
inet_init(void)1801 static int __init inet_init(void)
1802 {
1803 struct inet_protosw *q;
1804 struct list_head *r;
1805 int rc = -EINVAL;
1806
1807 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1808
1809 rc = proto_register(&tcp_prot, 1);
1810 if (rc)
1811 goto out;
1812
1813 rc = proto_register(&udp_prot, 1);
1814 if (rc)
1815 goto out_unregister_tcp_proto;
1816
1817 rc = proto_register(&raw_prot, 1);
1818 if (rc)
1819 goto out_unregister_udp_proto;
1820
1821 rc = proto_register(&ping_prot, 1);
1822 if (rc)
1823 goto out_unregister_raw_proto;
1824
1825 /*
1826 * Tell SOCKET that we are alive...
1827 */
1828
1829 (void)sock_register(&inet_family_ops);
1830
1831 #ifdef CONFIG_SYSCTL
1832 ip_static_sysctl_init();
1833 #endif
1834
1835 /*
1836 * Add all the base protocols.
1837 */
1838
1839 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1840 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1841 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1842 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1843 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1844 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1845 #ifdef CONFIG_IP_MULTICAST
1846 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1847 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1848 #endif
1849
1850 /* Register the socket-side information for inet_create. */
1851 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1852 INIT_LIST_HEAD(r);
1853
1854 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1855 inet_register_protosw(q);
1856
1857 /*
1858 * Set the ARP module up
1859 */
1860
1861 arp_init();
1862
1863 /*
1864 * Set the IP module up
1865 */
1866
1867 ip_init();
1868
1869 tcp_v4_init();
1870
1871 /* Setup TCP slab cache for open requests. */
1872 tcp_init();
1873
1874 /* Setup UDP memory threshold */
1875 udp_init();
1876
1877 /* Add UDP-Lite (RFC 3828) */
1878 udplite4_register();
1879
1880 ping_init();
1881
1882 /*
1883 * Set the ICMP layer up
1884 */
1885
1886 if (icmp_init() < 0)
1887 panic("Failed to create the ICMP control socket.\n");
1888
1889 /*
1890 * Initialise the multicast router
1891 */
1892 #if defined(CONFIG_IP_MROUTE)
1893 if (ip_mr_init())
1894 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1895 #endif
1896
1897 if (init_inet_pernet_ops())
1898 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1899 /*
1900 * Initialise per-cpu ipv4 mibs
1901 */
1902
1903 if (init_ipv4_mibs())
1904 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1905
1906 ipv4_proc_init();
1907
1908 ipfrag_init();
1909
1910 dev_add_pack(&ip_packet_type);
1911
1912 ip_tunnel_core_init();
1913
1914 rc = 0;
1915 out:
1916 return rc;
1917 out_unregister_raw_proto:
1918 proto_unregister(&raw_prot);
1919 out_unregister_udp_proto:
1920 proto_unregister(&udp_prot);
1921 out_unregister_tcp_proto:
1922 proto_unregister(&tcp_prot);
1923 goto out;
1924 }
1925
1926 fs_initcall(inet_init);
1927
1928 /* ------------------------------------------------------------------------ */
1929
1930 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)1931 static int __init ipv4_proc_init(void)
1932 {
1933 int rc = 0;
1934
1935 if (raw_proc_init())
1936 goto out_raw;
1937 if (tcp4_proc_init())
1938 goto out_tcp;
1939 if (udp4_proc_init())
1940 goto out_udp;
1941 if (ping_proc_init())
1942 goto out_ping;
1943 if (ip_misc_proc_init())
1944 goto out_misc;
1945 out:
1946 return rc;
1947 out_misc:
1948 ping_proc_exit();
1949 out_ping:
1950 udp4_proc_exit();
1951 out_udp:
1952 tcp4_proc_exit();
1953 out_tcp:
1954 raw_proc_exit();
1955 out_raw:
1956 rc = -ENOMEM;
1957 goto out;
1958 }
1959
1960 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)1961 static int __init ipv4_proc_init(void)
1962 {
1963 return 0;
1964 }
1965 #endif /* CONFIG_PROC_FS */
1966