projects / GitHub / moto-9609 / android_kernel_motorola_exynos9610.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge 4.14.72 into android-4.14-p
[GitHub/moto-9609/android_kernel_motorola_exynos9610.git] / net / ipv4 / af_inet.c
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 <linux/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
128 static inline int current_has_network(void)
129 {
130 return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
131 }
132 #else
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
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(refcount_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
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 */
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
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 goto out;
396 }
397 }
398
399 if (!kern) {
400 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
401 if (err) {
402 sk_common_release(sk);
403 goto out;
404 }
405 }
406 out:
407 return err;
408 out_rcu_unlock:
409 rcu_read_unlock();
410 goto out;
411 }
412
413
414 /*
415 * The peer socket should always be NULL (or else). When we call this
416 * function we are destroying the object and from then on nobody
417 * should refer to it.
418 */
419 int inet_release(struct socket *sock)
420 {
421 struct sock *sk = sock->sk;
422
423 if (sk) {
424 long timeout;
425
426 #ifdef CONFIG_NETFILTER_XT_MATCH_QTAGUID
427 qtaguid_untag(sock, true);
428 #endif
429 /* Applications forget to leave groups before exiting */
430 ip_mc_drop_socket(sk);
431
432 /* If linger is set, we don't return until the close
433 * is complete. Otherwise we return immediately. The
434 * actually closing is done the same either way.
435 *
436 * If the close is due to the process exiting, we never
437 * linger..
438 */
439 timeout = 0;
440 if (sock_flag(sk, SOCK_LINGER) &&
441 !(current->flags & PF_EXITING))
442 timeout = sk->sk_lingertime;
443 sock->sk = NULL;
444 sk->sk_prot->close(sk, timeout);
445 }
446 return 0;
447 }
448 EXPORT_SYMBOL(inet_release);
449
450 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
451 {
452 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
453 struct sock *sk = sock->sk;
454 struct inet_sock *inet = inet_sk(sk);
455 struct net *net = sock_net(sk);
456 unsigned short snum;
457 int chk_addr_ret;
458 u32 tb_id = RT_TABLE_LOCAL;
459 int err;
460
461 /* If the socket has its own bind function then use it. (RAW) */
462 if (sk->sk_prot->bind) {
463 err = sk->sk_prot->bind(sk, uaddr, addr_len);
464 goto out;
465 }
466 err = -EINVAL;
467 if (addr_len < sizeof(struct sockaddr_in))
468 goto out;
469
470 if (addr->sin_family != AF_INET) {
471 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
472 * only if s_addr is INADDR_ANY.
473 */
474 err = -EAFNOSUPPORT;
475 if (addr->sin_family != AF_UNSPEC ||
476 addr->sin_addr.s_addr != htonl(INADDR_ANY))
477 goto out;
478 }
479
480 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
481 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
482
483 /* Not specified by any standard per-se, however it breaks too
484 * many applications when removed. It is unfortunate since
485 * allowing applications to make a non-local bind solves
486 * several problems with systems using dynamic addressing.
487 * (ie. your servers still start up even if your ISDN link
488 * is temporarily down)
489 */
490 err = -EADDRNOTAVAIL;
491 if (!net->ipv4.sysctl_ip_nonlocal_bind &&
492 !(inet->freebind || inet->transparent) &&
493 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
494 chk_addr_ret != RTN_LOCAL &&
495 chk_addr_ret != RTN_MULTICAST &&
496 chk_addr_ret != RTN_BROADCAST)
497 goto out;
498
499 snum = ntohs(addr->sin_port);
500 err = -EACCES;
501 if (snum && snum < inet_prot_sock(net) &&
502 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
503 goto out;
504
505 /* We keep a pair of addresses. rcv_saddr is the one
506 * used by hash lookups, and saddr is used for transmit.
507 *
508 * In the BSD API these are the same except where it
509 * would be illegal to use them (multicast/broadcast) in
510 * which case the sending device address is used.
511 */
512 lock_sock(sk);
513
514 /* Check these errors (active socket, double bind). */
515 err = -EINVAL;
516 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
517 goto out_release_sock;
518
519 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
520 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
521 inet->inet_saddr = 0; /* Use device */
522
523 /* Make sure we are allowed to bind here. */
524 if ((snum || !inet->bind_address_no_port) &&
525 sk->sk_prot->get_port(sk, snum)) {
526 inet->inet_saddr = inet->inet_rcv_saddr = 0;
527 err = -EADDRINUSE;
528 goto out_release_sock;
529 }
530
531 if (inet->inet_rcv_saddr)
532 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
533 if (snum)
534 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
535 inet->inet_sport = htons(inet->inet_num);
536 inet->inet_daddr = 0;
537 inet->inet_dport = 0;
538 sk_dst_reset(sk);
539 err = 0;
540 out_release_sock:
541 release_sock(sk);
542 out:
543 return err;
544 }
545 EXPORT_SYMBOL(inet_bind);
546
547 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
548 int addr_len, int flags)
549 {
550 struct sock *sk = sock->sk;
551
552 if (addr_len < sizeof(uaddr->sa_family))
553 return -EINVAL;
554 if (uaddr->sa_family == AF_UNSPEC)
555 return sk->sk_prot->disconnect(sk, flags);
556
557 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
558 return -EAGAIN;
559 return sk->sk_prot->connect(sk, uaddr, addr_len);
560 }
561 EXPORT_SYMBOL(inet_dgram_connect);
562
563 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
564 {
565 DEFINE_WAIT_FUNC(wait, woken_wake_function);
566
567 add_wait_queue(sk_sleep(sk), &wait);
568 sk->sk_write_pending += writebias;
569
570 /* Basic assumption: if someone sets sk->sk_err, he _must_
571 * change state of the socket from TCP_SYN_*.
572 * Connect() does not allow to get error notifications
573 * without closing the socket.
574 */
575 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
576 release_sock(sk);
577 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
578 lock_sock(sk);
579 if (signal_pending(current) || !timeo)
580 break;
581 }
582 remove_wait_queue(sk_sleep(sk), &wait);
583 sk->sk_write_pending -= writebias;
584 return timeo;
585 }
586
587 /*
588 * Connect to a remote host. There is regrettably still a little
589 * TCP 'magic' in here.
590 */
591 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
592 int addr_len, int flags, int is_sendmsg)
593 {
594 struct sock *sk = sock->sk;
595 int err;
596 long timeo;
597
598 /*
599 * uaddr can be NULL and addr_len can be 0 if:
600 * sk is a TCP fastopen active socket and
601 * TCP_FASTOPEN_CONNECT sockopt is set and
602 * we already have a valid cookie for this socket.
603 * In this case, user can call write() after connect().
604 * write() will invoke tcp_sendmsg_fastopen() which calls
605 * __inet_stream_connect().
606 */
607 if (uaddr) {
608 if (addr_len < sizeof(uaddr->sa_family))
609 return -EINVAL;
610
611 if (uaddr->sa_family == AF_UNSPEC) {
612 err = sk->sk_prot->disconnect(sk, flags);
613 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
614 goto out;
615 }
616 }
617
618 switch (sock->state) {
619 default:
620 err = -EINVAL;
621 goto out;
622 case SS_CONNECTED:
623 err = -EISCONN;
624 goto out;
625 case SS_CONNECTING:
626 if (inet_sk(sk)->defer_connect)
627 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
628 else
629 err = -EALREADY;
630 /* Fall out of switch with err, set for this state */
631 break;
632 case SS_UNCONNECTED:
633 err = -EISCONN;
634 if (sk->sk_state != TCP_CLOSE)
635 goto out;
636
637 err = sk->sk_prot->connect(sk, uaddr, addr_len);
638 if (err < 0)
639 goto out;
640
641 sock->state = SS_CONNECTING;
642
643 if (!err && inet_sk(sk)->defer_connect)
644 goto out;
645
646 /* Just entered SS_CONNECTING state; the only
647 * difference is that return value in non-blocking
648 * case is EINPROGRESS, rather than EALREADY.
649 */
650 err = -EINPROGRESS;
651 break;
652 }
653
654 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
655
656 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
657 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
658 tcp_sk(sk)->fastopen_req &&
659 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
660
661 /* Error code is set above */
662 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
663 goto out;
664
665 err = sock_intr_errno(timeo);
666 if (signal_pending(current))
667 goto out;
668 }
669
670 /* Connection was closed by RST, timeout, ICMP error
671 * or another process disconnected us.
672 */
673 if (sk->sk_state == TCP_CLOSE)
674 goto sock_error;
675
676 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
677 * and error was received after socket entered established state.
678 * Hence, it is handled normally after connect() return successfully.
679 */
680
681 sock->state = SS_CONNECTED;
682 err = 0;
683 out:
684 return err;
685
686 sock_error:
687 err = sock_error(sk) ? : -ECONNABORTED;
688 sock->state = SS_UNCONNECTED;
689 if (sk->sk_prot->disconnect(sk, flags))
690 sock->state = SS_DISCONNECTING;
691 goto out;
692 }
693 EXPORT_SYMBOL(__inet_stream_connect);
694
695 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
696 int addr_len, int flags)
697 {
698 int err;
699
700 lock_sock(sock->sk);
701 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
702 release_sock(sock->sk);
703 return err;
704 }
705 EXPORT_SYMBOL(inet_stream_connect);
706
707 /*
708 * Accept a pending connection. The TCP layer now gives BSD semantics.
709 */
710
711 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
712 bool kern)
713 {
714 struct sock *sk1 = sock->sk;
715 int err = -EINVAL;
716 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
717
718 if (!sk2)
719 goto do_err;
720
721 lock_sock(sk2);
722
723 sock_rps_record_flow(sk2);
724 WARN_ON(!((1 << sk2->sk_state) &
725 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
726 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
727
728 sock_graft(sk2, newsock);
729
730 newsock->state = SS_CONNECTED;
731 err = 0;
732 release_sock(sk2);
733 do_err:
734 return err;
735 }
736 EXPORT_SYMBOL(inet_accept);
737
738
739 /*
740 * This does both peername and sockname.
741 */
742 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
743 int *uaddr_len, int peer)
744 {
745 struct sock *sk = sock->sk;
746 struct inet_sock *inet = inet_sk(sk);
747 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
748
749 sin->sin_family = AF_INET;
750 if (peer) {
751 if (!inet->inet_dport ||
752 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
753 peer == 1))
754 return -ENOTCONN;
755 sin->sin_port = inet->inet_dport;
756 sin->sin_addr.s_addr = inet->inet_daddr;
757 } else {
758 __be32 addr = inet->inet_rcv_saddr;
759 if (!addr)
760 addr = inet->inet_saddr;
761 sin->sin_port = inet->inet_sport;
762 sin->sin_addr.s_addr = addr;
763 }
764 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
765 *uaddr_len = sizeof(*sin);
766 return 0;
767 }
768 EXPORT_SYMBOL(inet_getname);
769
770 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
771 {
772 struct sock *sk = sock->sk;
773
774 sock_rps_record_flow(sk);
775
776 /* We may need to bind the socket. */
777 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
778 inet_autobind(sk))
779 return -EAGAIN;
780
781 return sk->sk_prot->sendmsg(sk, msg, size);
782 }
783 EXPORT_SYMBOL(inet_sendmsg);
784
785 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
786 size_t size, int flags)
787 {
788 struct sock *sk = sock->sk;
789
790 sock_rps_record_flow(sk);
791
792 /* We may need to bind the socket. */
793 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
794 inet_autobind(sk))
795 return -EAGAIN;
796
797 if (sk->sk_prot->sendpage)
798 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
799 return sock_no_sendpage(sock, page, offset, size, flags);
800 }
801 EXPORT_SYMBOL(inet_sendpage);
802
803 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
804 int flags)
805 {
806 struct sock *sk = sock->sk;
807 int addr_len = 0;
808 int err;
809
810 sock_rps_record_flow(sk);
811
812 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
813 flags & ~MSG_DONTWAIT, &addr_len);
814 if (err >= 0)
815 msg->msg_namelen = addr_len;
816 return err;
817 }
818 EXPORT_SYMBOL(inet_recvmsg);
819
820 int inet_shutdown(struct socket *sock, int how)
821 {
822 struct sock *sk = sock->sk;
823 int err = 0;
824
825 /* This should really check to make sure
826 * the socket is a TCP socket. (WHY AC...)
827 */
828 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
829 1->2 bit 2 snds.
830 2->3 */
831 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
832 return -EINVAL;
833
834 lock_sock(sk);
835 if (sock->state == SS_CONNECTING) {
836 if ((1 << sk->sk_state) &
837 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
838 sock->state = SS_DISCONNECTING;
839 else
840 sock->state = SS_CONNECTED;
841 }
842
843 switch (sk->sk_state) {
844 case TCP_CLOSE:
845 err = -ENOTCONN;
846 /* Hack to wake up other listeners, who can poll for
847 POLLHUP, even on eg. unconnected UDP sockets -- RR */
848 default:
849 sk->sk_shutdown |= how;
850 if (sk->sk_prot->shutdown)
851 sk->sk_prot->shutdown(sk, how);
852 break;
853
854 /* Remaining two branches are temporary solution for missing
855 * close() in multithreaded environment. It is _not_ a good idea,
856 * but we have no choice until close() is repaired at VFS level.
857 */
858 case TCP_LISTEN:
859 if (!(how & RCV_SHUTDOWN))
860 break;
861 /* Fall through */
862 case TCP_SYN_SENT:
863 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
864 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
865 break;
866 }
867
868 /* Wake up anyone sleeping in poll. */
869 sk->sk_state_change(sk);
870 release_sock(sk);
871 return err;
872 }
873 EXPORT_SYMBOL(inet_shutdown);
874
875 /*
876 * ioctl() calls you can issue on an INET socket. Most of these are
877 * device configuration and stuff and very rarely used. Some ioctls
878 * pass on to the socket itself.
879 *
880 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
881 * loads the devconfigure module does its configuring and unloads it.
882 * There's a good 20K of config code hanging around the kernel.
883 */
884
885 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
886 {
887 struct sock *sk = sock->sk;
888 int err = 0;
889 struct net *net = sock_net(sk);
890
891 switch (cmd) {
892 case SIOCGSTAMP:
893 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
894 break;
895 case SIOCGSTAMPNS:
896 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
897 break;
898 case SIOCADDRT:
899 case SIOCDELRT:
900 case SIOCRTMSG:
901 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
902 break;
903 case SIOCDARP:
904 case SIOCGARP:
905 case SIOCSARP:
906 err = arp_ioctl(net, cmd, (void __user *)arg);
907 break;
908 case SIOCGIFADDR:
909 case SIOCSIFADDR:
910 case SIOCGIFBRDADDR:
911 case SIOCSIFBRDADDR:
912 case SIOCGIFNETMASK:
913 case SIOCSIFNETMASK:
914 case SIOCGIFDSTADDR:
915 case SIOCSIFDSTADDR:
916 case SIOCSIFPFLAGS:
917 case SIOCGIFPFLAGS:
918 case SIOCSIFFLAGS:
919 err = devinet_ioctl(net, cmd, (void __user *)arg);
920 break;
921 default:
922 if (sk->sk_prot->ioctl)
923 err = sk->sk_prot->ioctl(sk, cmd, arg);
924 else
925 err = -ENOIOCTLCMD;
926 break;
927 }
928 return err;
929 }
930 EXPORT_SYMBOL(inet_ioctl);
931
932 #ifdef CONFIG_COMPAT
933 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
934 {
935 struct sock *sk = sock->sk;
936 int err = -ENOIOCTLCMD;
937
938 if (sk->sk_prot->compat_ioctl)
939 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
940
941 return err;
942 }
943 #endif
944
945 const struct proto_ops inet_stream_ops = {
946 .family = PF_INET,
947 .owner = THIS_MODULE,
948 .release = inet_release,
949 .bind = inet_bind,
950 .connect = inet_stream_connect,
951 .socketpair = sock_no_socketpair,
952 .accept = inet_accept,
953 .getname = inet_getname,
954 .poll = tcp_poll,
955 .ioctl = inet_ioctl,
956 .listen = inet_listen,
957 .shutdown = inet_shutdown,
958 .setsockopt = sock_common_setsockopt,
959 .getsockopt = sock_common_getsockopt,
960 .sendmsg = inet_sendmsg,
961 .recvmsg = inet_recvmsg,
962 .mmap = sock_no_mmap,
963 .sendpage = inet_sendpage,
964 .splice_read = tcp_splice_read,
965 .read_sock = tcp_read_sock,
966 .sendmsg_locked = tcp_sendmsg_locked,
967 .sendpage_locked = tcp_sendpage_locked,
968 .peek_len = tcp_peek_len,
969 #ifdef CONFIG_COMPAT
970 .compat_setsockopt = compat_sock_common_setsockopt,
971 .compat_getsockopt = compat_sock_common_getsockopt,
972 .compat_ioctl = inet_compat_ioctl,
973 #endif
974 };
975 EXPORT_SYMBOL(inet_stream_ops);
976
977 const struct proto_ops inet_dgram_ops = {
978 .family = PF_INET,
979 .owner = THIS_MODULE,
980 .release = inet_release,
981 .bind = inet_bind,
982 .connect = inet_dgram_connect,
983 .socketpair = sock_no_socketpair,
984 .accept = sock_no_accept,
985 .getname = inet_getname,
986 .poll = udp_poll,
987 .ioctl = inet_ioctl,
988 .listen = sock_no_listen,
989 .shutdown = inet_shutdown,
990 .setsockopt = sock_common_setsockopt,
991 .getsockopt = sock_common_getsockopt,
992 .sendmsg = inet_sendmsg,
993 .recvmsg = inet_recvmsg,
994 .mmap = sock_no_mmap,
995 .sendpage = inet_sendpage,
996 .set_peek_off = sk_set_peek_off,
997 #ifdef CONFIG_COMPAT
998 .compat_setsockopt = compat_sock_common_setsockopt,
999 .compat_getsockopt = compat_sock_common_getsockopt,
1000 .compat_ioctl = inet_compat_ioctl,
1001 #endif
1002 };
1003 EXPORT_SYMBOL(inet_dgram_ops);
1004
1005 /*
1006 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1007 * udp_poll
1008 */
1009 static const struct proto_ops inet_sockraw_ops = {
1010 .family = PF_INET,
1011 .owner = THIS_MODULE,
1012 .release = inet_release,
1013 .bind = inet_bind,
1014 .connect = inet_dgram_connect,
1015 .socketpair = sock_no_socketpair,
1016 .accept = sock_no_accept,
1017 .getname = inet_getname,
1018 .poll = datagram_poll,
1019 .ioctl = inet_ioctl,
1020 .listen = sock_no_listen,
1021 .shutdown = inet_shutdown,
1022 .setsockopt = sock_common_setsockopt,
1023 .getsockopt = sock_common_getsockopt,
1024 .sendmsg = inet_sendmsg,
1025 .recvmsg = inet_recvmsg,
1026 .mmap = sock_no_mmap,
1027 .sendpage = inet_sendpage,
1028 #ifdef CONFIG_COMPAT
1029 .compat_setsockopt = compat_sock_common_setsockopt,
1030 .compat_getsockopt = compat_sock_common_getsockopt,
1031 .compat_ioctl = inet_compat_ioctl,
1032 #endif
1033 };
1034
1035 static const struct net_proto_family inet_family_ops = {
1036 .family = PF_INET,
1037 .create = inet_create,
1038 .owner = THIS_MODULE,
1039 };
1040
1041 /* Upon startup we insert all the elements in inetsw_array[] into
1042 * the linked list inetsw.
1043 */
1044 static struct inet_protosw inetsw_array[] =
1045 {
1046 {
1047 .type = SOCK_STREAM,
1048 .protocol = IPPROTO_TCP,
1049 .prot = &tcp_prot,
1050 .ops = &inet_stream_ops,
1051 .flags = INET_PROTOSW_PERMANENT |
1052 INET_PROTOSW_ICSK,
1053 },
1054
1055 {
1056 .type = SOCK_DGRAM,
1057 .protocol = IPPROTO_UDP,
1058 .prot = &udp_prot,
1059 .ops = &inet_dgram_ops,
1060 .flags = INET_PROTOSW_PERMANENT,
1061 },
1062
1063 {
1064 .type = SOCK_DGRAM,
1065 .protocol = IPPROTO_ICMP,
1066 .prot = &ping_prot,
1067 .ops = &inet_sockraw_ops,
1068 .flags = INET_PROTOSW_REUSE,
1069 },
1070
1071 {
1072 .type = SOCK_RAW,
1073 .protocol = IPPROTO_IP, /* wild card */
1074 .prot = &raw_prot,
1075 .ops = &inet_sockraw_ops,
1076 .flags = INET_PROTOSW_REUSE,
1077 }
1078 };
1079
1080 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1081
1082 void inet_register_protosw(struct inet_protosw *p)
1083 {
1084 struct list_head *lh;
1085 struct inet_protosw *answer;
1086 int protocol = p->protocol;
1087 struct list_head *last_perm;
1088
1089 spin_lock_bh(&inetsw_lock);
1090
1091 if (p->type >= SOCK_MAX)
1092 goto out_illegal;
1093
1094 /* If we are trying to override a permanent protocol, bail. */
1095 last_perm = &inetsw[p->type];
1096 list_for_each(lh, &inetsw[p->type]) {
1097 answer = list_entry(lh, struct inet_protosw, list);
1098 /* Check only the non-wild match. */
1099 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1100 break;
1101 if (protocol == answer->protocol)
1102 goto out_permanent;
1103 last_perm = lh;
1104 }
1105
1106 /* Add the new entry after the last permanent entry if any, so that
1107 * the new entry does not override a permanent entry when matched with
1108 * a wild-card protocol. But it is allowed to override any existing
1109 * non-permanent entry. This means that when we remove this entry, the
1110 * system automatically returns to the old behavior.
1111 */
1112 list_add_rcu(&p->list, last_perm);
1113 out:
1114 spin_unlock_bh(&inetsw_lock);
1115
1116 return;
1117
1118 out_permanent:
1119 pr_err("Attempt to override permanent protocol %d\n", protocol);
1120 goto out;
1121
1122 out_illegal:
1123 pr_err("Ignoring attempt to register invalid socket type %d\n",
1124 p->type);
1125 goto out;
1126 }
1127 EXPORT_SYMBOL(inet_register_protosw);
1128
1129 void inet_unregister_protosw(struct inet_protosw *p)
1130 {
1131 if (INET_PROTOSW_PERMANENT & p->flags) {
1132 pr_err("Attempt to unregister permanent protocol %d\n",
1133 p->protocol);
1134 } else {
1135 spin_lock_bh(&inetsw_lock);
1136 list_del_rcu(&p->list);
1137 spin_unlock_bh(&inetsw_lock);
1138
1139 synchronize_net();
1140 }
1141 }
1142 EXPORT_SYMBOL(inet_unregister_protosw);
1143
1144 static int inet_sk_reselect_saddr(struct sock *sk)
1145 {
1146 struct inet_sock *inet = inet_sk(sk);
1147 __be32 old_saddr = inet->inet_saddr;
1148 __be32 daddr = inet->inet_daddr;
1149 struct flowi4 *fl4;
1150 struct rtable *rt;
1151 __be32 new_saddr;
1152 struct ip_options_rcu *inet_opt;
1153
1154 inet_opt = rcu_dereference_protected(inet->inet_opt,
1155 lockdep_sock_is_held(sk));
1156 if (inet_opt && inet_opt->opt.srr)
1157 daddr = inet_opt->opt.faddr;
1158
1159 /* Query new route. */
1160 fl4 = &inet->cork.fl.u.ip4;
1161 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1162 sk->sk_bound_dev_if, sk->sk_protocol,
1163 inet->inet_sport, inet->inet_dport, sk);
1164 if (IS_ERR(rt))
1165 return PTR_ERR(rt);
1166
1167 sk_setup_caps(sk, &rt->dst);
1168
1169 new_saddr = fl4->saddr;
1170
1171 if (new_saddr == old_saddr)
1172 return 0;
1173
1174 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1175 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1176 __func__, &old_saddr, &new_saddr);
1177 }
1178
1179 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1180
1181 /*
1182 * XXX The only one ugly spot where we need to
1183 * XXX really change the sockets identity after
1184 * XXX it has entered the hashes. -DaveM
1185 *
1186 * Besides that, it does not check for connection
1187 * uniqueness. Wait for troubles.
1188 */
1189 return __sk_prot_rehash(sk);
1190 }
1191
1192 int inet_sk_rebuild_header(struct sock *sk)
1193 {
1194 struct inet_sock *inet = inet_sk(sk);
1195 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1196 __be32 daddr;
1197 struct ip_options_rcu *inet_opt;
1198 struct flowi4 *fl4;
1199 int err;
1200
1201 /* Route is OK, nothing to do. */
1202 if (rt)
1203 return 0;
1204
1205 /* Reroute. */
1206 rcu_read_lock();
1207 inet_opt = rcu_dereference(inet->inet_opt);
1208 daddr = inet->inet_daddr;
1209 if (inet_opt && inet_opt->opt.srr)
1210 daddr = inet_opt->opt.faddr;
1211 rcu_read_unlock();
1212 fl4 = &inet->cork.fl.u.ip4;
1213 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1214 inet->inet_dport, inet->inet_sport,
1215 sk->sk_protocol, RT_CONN_FLAGS(sk),
1216 sk->sk_bound_dev_if);
1217 if (!IS_ERR(rt)) {
1218 err = 0;
1219 sk_setup_caps(sk, &rt->dst);
1220 } else {
1221 err = PTR_ERR(rt);
1222
1223 /* Routing failed... */
1224 sk->sk_route_caps = 0;
1225 /*
1226 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1227 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1228 */
1229 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1230 sk->sk_state != TCP_SYN_SENT ||
1231 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1232 (err = inet_sk_reselect_saddr(sk)) != 0)
1233 sk->sk_err_soft = -err;
1234 }
1235
1236 return err;
1237 }
1238 EXPORT_SYMBOL(inet_sk_rebuild_header);
1239
1240 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1241 netdev_features_t features)
1242 {
1243 bool udpfrag = false, fixedid = false, gso_partial, encap;
1244 struct sk_buff *segs = ERR_PTR(-EINVAL);
1245 const struct net_offload *ops;
1246 unsigned int offset = 0;
1247 struct iphdr *iph;
1248 int proto, tot_len;
1249 int nhoff;
1250 int ihl;
1251 int id;
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 || encap) {
1281 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1282 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1283
1284 /* fixed ID is invalid if DF bit is not set */
1285 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1286 goto out;
1287 }
1288
1289 ops = rcu_dereference(inet_offloads[proto]);
1290 if (likely(ops && ops->callbacks.gso_segment))
1291 segs = ops->callbacks.gso_segment(skb, features);
1292
1293 if (IS_ERR_OR_NULL(segs))
1294 goto out;
1295
1296 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1297
1298 skb = segs;
1299 do {
1300 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1301 if (udpfrag) {
1302 iph->frag_off = htons(offset >> 3);
1303 if (skb->next)
1304 iph->frag_off |= htons(IP_MF);
1305 offset += skb->len - nhoff - ihl;
1306 tot_len = skb->len - nhoff;
1307 } else if (skb_is_gso(skb)) {
1308 if (!fixedid) {
1309 iph->id = htons(id);
1310 id += skb_shinfo(skb)->gso_segs;
1311 }
1312
1313 if (gso_partial)
1314 tot_len = skb_shinfo(skb)->gso_size +
1315 SKB_GSO_CB(skb)->data_offset +
1316 skb->head - (unsigned char *)iph;
1317 else
1318 tot_len = skb->len - nhoff;
1319 } else {
1320 if (!fixedid)
1321 iph->id = htons(id++);
1322 tot_len = skb->len - nhoff;
1323 }
1324 iph->tot_len = htons(tot_len);
1325 ip_send_check(iph);
1326 if (encap)
1327 skb_reset_inner_headers(skb);
1328 skb->network_header = (u8 *)iph - skb->head;
1329 } while ((skb = skb->next));
1330
1331 out:
1332 return segs;
1333 }
1334 EXPORT_SYMBOL(inet_gso_segment);
1335
1336 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
1337 {
1338 const struct net_offload *ops;
1339 struct sk_buff **pp = NULL;
1340 struct sk_buff *p;
1341 const struct iphdr *iph;
1342 unsigned int hlen;
1343 unsigned int off;
1344 unsigned int id;
1345 int flush = 1;
1346 int proto;
1347
1348 off = skb_gro_offset(skb);
1349 hlen = off + sizeof(*iph);
1350 iph = skb_gro_header_fast(skb, off);
1351 if (skb_gro_header_hard(skb, hlen)) {
1352 iph = skb_gro_header_slow(skb, hlen, off);
1353 if (unlikely(!iph))
1354 goto out;
1355 }
1356
1357 proto = iph->protocol;
1358
1359 rcu_read_lock();
1360 ops = rcu_dereference(inet_offloads[proto]);
1361 if (!ops || !ops->callbacks.gro_receive)
1362 goto out_unlock;
1363
1364 if (*(u8 *)iph != 0x45)
1365 goto out_unlock;
1366
1367 if (ip_is_fragment(iph))
1368 goto out_unlock;
1369
1370 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1371 goto out_unlock;
1372
1373 id = ntohl(*(__be32 *)&iph->id);
1374 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1375 id >>= 16;
1376
1377 for (p = *head; p; p = p->next) {
1378 struct iphdr *iph2;
1379 u16 flush_id;
1380
1381 if (!NAPI_GRO_CB(p)->same_flow)
1382 continue;
1383
1384 iph2 = (struct iphdr *)(p->data + off);
1385 /* The above works because, with the exception of the top
1386 * (inner most) layer, we only aggregate pkts with the same
1387 * hdr length so all the hdrs we'll need to verify will start
1388 * at the same offset.
1389 */
1390 if ((iph->protocol ^ iph2->protocol) |
1391 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1392 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1393 NAPI_GRO_CB(p)->same_flow = 0;
1394 continue;
1395 }
1396
1397 /* All fields must match except length and checksum. */
1398 NAPI_GRO_CB(p)->flush |=
1399 (iph->ttl ^ iph2->ttl) |
1400 (iph->tos ^ iph2->tos) |
1401 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1402
1403 NAPI_GRO_CB(p)->flush |= flush;
1404
1405 /* We need to store of the IP ID check to be included later
1406 * when we can verify that this packet does in fact belong
1407 * to a given flow.
1408 */
1409 flush_id = (u16)(id - ntohs(iph2->id));
1410
1411 /* This bit of code makes it much easier for us to identify
1412 * the cases where we are doing atomic vs non-atomic IP ID
1413 * checks. Specifically an atomic check can return IP ID
1414 * values 0 - 0xFFFF, while a non-atomic check can only
1415 * return 0 or 0xFFFF.
1416 */
1417 if (!NAPI_GRO_CB(p)->is_atomic ||
1418 !(iph->frag_off & htons(IP_DF))) {
1419 flush_id ^= NAPI_GRO_CB(p)->count;
1420 flush_id = flush_id ? 0xFFFF : 0;
1421 }
1422
1423 /* If the previous IP ID value was based on an atomic
1424 * datagram we can overwrite the value and ignore it.
1425 */
1426 if (NAPI_GRO_CB(skb)->is_atomic)
1427 NAPI_GRO_CB(p)->flush_id = flush_id;
1428 else
1429 NAPI_GRO_CB(p)->flush_id |= flush_id;
1430 }
1431
1432 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1433 NAPI_GRO_CB(skb)->flush |= flush;
1434 skb_set_network_header(skb, off);
1435 /* The above will be needed by the transport layer if there is one
1436 * immediately following this IP hdr.
1437 */
1438
1439 /* Note : No need to call skb_gro_postpull_rcsum() here,
1440 * as we already checked checksum over ipv4 header was 0
1441 */
1442 skb_gro_pull(skb, sizeof(*iph));
1443 skb_set_transport_header(skb, skb_gro_offset(skb));
1444
1445 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1446
1447 out_unlock:
1448 rcu_read_unlock();
1449
1450 out:
1451 skb_gro_flush_final(skb, pp, flush);
1452
1453 return pp;
1454 }
1455 EXPORT_SYMBOL(inet_gro_receive);
1456
1457 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1458 struct sk_buff *skb)
1459 {
1460 if (NAPI_GRO_CB(skb)->encap_mark) {
1461 NAPI_GRO_CB(skb)->flush = 1;
1462 return NULL;
1463 }
1464
1465 NAPI_GRO_CB(skb)->encap_mark = 1;
1466
1467 return inet_gro_receive(head, skb);
1468 }
1469
1470 #define SECONDS_PER_DAY 86400
1471
1472 /* inet_current_timestamp - Return IP network timestamp
1473 *
1474 * Return milliseconds since midnight in network byte order.
1475 */
1476 __be32 inet_current_timestamp(void)
1477 {
1478 u32 secs;
1479 u32 msecs;
1480 struct timespec64 ts;
1481
1482 ktime_get_real_ts64(&ts);
1483
1484 /* Get secs since midnight. */
1485 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1486 /* Convert to msecs. */
1487 msecs = secs * MSEC_PER_SEC;
1488 /* Convert nsec to msec. */
1489 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1490
1491 /* Convert to network byte order. */
1492 return htonl(msecs);
1493 }
1494 EXPORT_SYMBOL(inet_current_timestamp);
1495
1496 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1497 {
1498 if (sk->sk_family == AF_INET)
1499 return ip_recv_error(sk, msg, len, addr_len);
1500 #if IS_ENABLED(CONFIG_IPV6)
1501 if (sk->sk_family == AF_INET6)
1502 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1503 #endif
1504 return -EINVAL;
1505 }
1506
1507 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1508 {
1509 __be16 newlen = htons(skb->len - nhoff);
1510 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1511 const struct net_offload *ops;
1512 int proto = iph->protocol;
1513 int err = -ENOSYS;
1514
1515 if (skb->encapsulation) {
1516 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1517 skb_set_inner_network_header(skb, nhoff);
1518 }
1519
1520 csum_replace2(&iph->check, iph->tot_len, newlen);
1521 iph->tot_len = newlen;
1522
1523 rcu_read_lock();
1524 ops = rcu_dereference(inet_offloads[proto]);
1525 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1526 goto out_unlock;
1527
1528 /* Only need to add sizeof(*iph) to get to the next hdr below
1529 * because any hdr with option will have been flushed in
1530 * inet_gro_receive().
1531 */
1532 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1533
1534 out_unlock:
1535 rcu_read_unlock();
1536
1537 return err;
1538 }
1539 EXPORT_SYMBOL(inet_gro_complete);
1540
1541 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1542 {
1543 skb->encapsulation = 1;
1544 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1545 return inet_gro_complete(skb, nhoff);
1546 }
1547
1548 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1549 unsigned short type, unsigned char protocol,
1550 struct net *net)
1551 {
1552 struct socket *sock;
1553 int rc = sock_create_kern(net, family, type, protocol, &sock);
1554
1555 if (rc == 0) {
1556 *sk = sock->sk;
1557 (*sk)->sk_allocation = GFP_ATOMIC;
1558 /*
1559 * Unhash it so that IP input processing does not even see it,
1560 * we do not wish this socket to see incoming packets.
1561 */
1562 (*sk)->sk_prot->unhash(*sk);
1563 }
1564 return rc;
1565 }
1566 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1567
1568 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1569 {
1570 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1571 }
1572 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1573
1574 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1575 {
1576 unsigned long res = 0;
1577 int i;
1578
1579 for_each_possible_cpu(i)
1580 res += snmp_get_cpu_field(mib, i, offt);
1581 return res;
1582 }
1583 EXPORT_SYMBOL_GPL(snmp_fold_field);
1584
1585 #if BITS_PER_LONG==32
1586
1587 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1588 size_t syncp_offset)
1589 {
1590 void *bhptr;
1591 struct u64_stats_sync *syncp;
1592 u64 v;
1593 unsigned int start;
1594
1595 bhptr = per_cpu_ptr(mib, cpu);
1596 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1597 do {
1598 start = u64_stats_fetch_begin_irq(syncp);
1599 v = *(((u64 *)bhptr) + offt);
1600 } while (u64_stats_fetch_retry_irq(syncp, start));
1601
1602 return v;
1603 }
1604 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1605
1606 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1607 {
1608 u64 res = 0;
1609 int cpu;
1610
1611 for_each_possible_cpu(cpu) {
1612 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1613 }
1614 return res;
1615 }
1616 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1617 #endif
1618
1619 #ifdef CONFIG_IP_MULTICAST
1620 static const struct net_protocol igmp_protocol = {
1621 .handler = igmp_rcv,
1622 .netns_ok = 1,
1623 };
1624 #endif
1625
1626 /* thinking of making this const? Don't.
1627 * early_demux can change based on sysctl.
1628 */
1629 static struct net_protocol tcp_protocol = {
1630 .early_demux = tcp_v4_early_demux,
1631 .early_demux_handler = tcp_v4_early_demux,
1632 .handler = tcp_v4_rcv,
1633 .err_handler = tcp_v4_err,
1634 .no_policy = 1,
1635 .netns_ok = 1,
1636 .icmp_strict_tag_validation = 1,
1637 };
1638
1639 /* thinking of making this const? Don't.
1640 * early_demux can change based on sysctl.
1641 */
1642 static struct net_protocol udp_protocol = {
1643 .early_demux = udp_v4_early_demux,
1644 .early_demux_handler = udp_v4_early_demux,
1645 .handler = udp_rcv,
1646 .err_handler = udp_err,
1647 .no_policy = 1,
1648 .netns_ok = 1,
1649 };
1650
1651 static const struct net_protocol icmp_protocol = {
1652 .handler = icmp_rcv,
1653 .err_handler = icmp_err,
1654 .no_policy = 1,
1655 .netns_ok = 1,
1656 };
1657
1658 static __net_init int ipv4_mib_init_net(struct net *net)
1659 {
1660 int i;
1661
1662 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1663 if (!net->mib.tcp_statistics)
1664 goto err_tcp_mib;
1665 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1666 if (!net->mib.ip_statistics)
1667 goto err_ip_mib;
1668
1669 for_each_possible_cpu(i) {
1670 struct ipstats_mib *af_inet_stats;
1671 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1672 u64_stats_init(&af_inet_stats->syncp);
1673 }
1674
1675 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1676 if (!net->mib.net_statistics)
1677 goto err_net_mib;
1678 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1679 if (!net->mib.udp_statistics)
1680 goto err_udp_mib;
1681 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1682 if (!net->mib.udplite_statistics)
1683 goto err_udplite_mib;
1684 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1685 if (!net->mib.icmp_statistics)
1686 goto err_icmp_mib;
1687 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1688 GFP_KERNEL);
1689 if (!net->mib.icmpmsg_statistics)
1690 goto err_icmpmsg_mib;
1691
1692 tcp_mib_init(net);
1693 return 0;
1694
1695 err_icmpmsg_mib:
1696 free_percpu(net->mib.icmp_statistics);
1697 err_icmp_mib:
1698 free_percpu(net->mib.udplite_statistics);
1699 err_udplite_mib:
1700 free_percpu(net->mib.udp_statistics);
1701 err_udp_mib:
1702 free_percpu(net->mib.net_statistics);
1703 err_net_mib:
1704 free_percpu(net->mib.ip_statistics);
1705 err_ip_mib:
1706 free_percpu(net->mib.tcp_statistics);
1707 err_tcp_mib:
1708 return -ENOMEM;
1709 }
1710
1711 static __net_exit void ipv4_mib_exit_net(struct net *net)
1712 {
1713 kfree(net->mib.icmpmsg_statistics);
1714 free_percpu(net->mib.icmp_statistics);
1715 free_percpu(net->mib.udplite_statistics);
1716 free_percpu(net->mib.udp_statistics);
1717 free_percpu(net->mib.net_statistics);
1718 free_percpu(net->mib.ip_statistics);
1719 free_percpu(net->mib.tcp_statistics);
1720 }
1721
1722 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1723 .init = ipv4_mib_init_net,
1724 .exit = ipv4_mib_exit_net,
1725 };
1726
1727 static int __init init_ipv4_mibs(void)
1728 {
1729 return register_pernet_subsys(&ipv4_mib_ops);
1730 }
1731
1732 static __net_init int inet_init_net(struct net *net)
1733 {
1734 /*
1735 * Set defaults for local port range
1736 */
1737 seqlock_init(&net->ipv4.ip_local_ports.lock);
1738 net->ipv4.ip_local_ports.range[0] = 32768;
1739 net->ipv4.ip_local_ports.range[1] = 60999;
1740
1741 seqlock_init(&net->ipv4.ping_group_range.lock);
1742 /*
1743 * Sane defaults - nobody may create ping sockets.
1744 * Boot scripts should set this to distro-specific group.
1745 */
1746 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1747 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1748
1749 /* Default values for sysctl-controlled parameters.
1750 * We set them here, in case sysctl is not compiled.
1751 */
1752 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1753 net->ipv4.sysctl_ip_dynaddr = 0;
1754 net->ipv4.sysctl_ip_early_demux = 1;
1755 net->ipv4.sysctl_udp_early_demux = 1;
1756 net->ipv4.sysctl_tcp_early_demux = 1;
1757 #ifdef CONFIG_SYSCTL
1758 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1759 #endif
1760
1761 /* Some igmp sysctl, whose values are always used */
1762 net->ipv4.sysctl_igmp_max_memberships = 20;
1763 net->ipv4.sysctl_igmp_max_msf = 10;
1764 /* IGMP reports for link-local multicast groups are enabled by default */
1765 net->ipv4.sysctl_igmp_llm_reports = 1;
1766 net->ipv4.sysctl_igmp_qrv = 2;
1767
1768 return 0;
1769 }
1770
1771 static __net_exit void inet_exit_net(struct net *net)
1772 {
1773 }
1774
1775 static __net_initdata struct pernet_operations af_inet_ops = {
1776 .init = inet_init_net,
1777 .exit = inet_exit_net,
1778 };
1779
1780 static int __init init_inet_pernet_ops(void)
1781 {
1782 return register_pernet_subsys(&af_inet_ops);
1783 }
1784
1785 static int ipv4_proc_init(void);
1786
1787 /*
1788 * IP protocol layer initialiser
1789 */
1790
1791 static struct packet_offload ip_packet_offload __read_mostly = {
1792 .type = cpu_to_be16(ETH_P_IP),
1793 .callbacks = {
1794 .gso_segment = inet_gso_segment,
1795 .gro_receive = inet_gro_receive,
1796 .gro_complete = inet_gro_complete,
1797 },
1798 };
1799
1800 static const struct net_offload ipip_offload = {
1801 .callbacks = {
1802 .gso_segment = inet_gso_segment,
1803 .gro_receive = ipip_gro_receive,
1804 .gro_complete = ipip_gro_complete,
1805 },
1806 };
1807
1808 static int __init ipip_offload_init(void)
1809 {
1810 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1811 }
1812
1813 static int __init ipv4_offload_init(void)
1814 {
1815 /*
1816 * Add offloads
1817 */
1818 if (udpv4_offload_init() < 0)
1819 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1820 if (tcpv4_offload_init() < 0)
1821 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1822 if (ipip_offload_init() < 0)
1823 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1824
1825 dev_add_offload(&ip_packet_offload);
1826 return 0;
1827 }
1828
1829 fs_initcall(ipv4_offload_init);
1830
1831 static struct packet_type ip_packet_type __read_mostly = {
1832 .type = cpu_to_be16(ETH_P_IP),
1833 .func = ip_rcv,
1834 };
1835
1836 static int __init inet_init(void)
1837 {
1838 struct inet_protosw *q;
1839 struct list_head *r;
1840 int rc = -EINVAL;
1841
1842 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1843
1844 rc = proto_register(&tcp_prot, 1);
1845 if (rc)
1846 goto out;
1847
1848 rc = proto_register(&udp_prot, 1);
1849 if (rc)
1850 goto out_unregister_tcp_proto;
1851
1852 rc = proto_register(&raw_prot, 1);
1853 if (rc)
1854 goto out_unregister_udp_proto;
1855
1856 rc = proto_register(&ping_prot, 1);
1857 if (rc)
1858 goto out_unregister_raw_proto;
1859
1860 /*
1861 * Tell SOCKET that we are alive...
1862 */
1863
1864 (void)sock_register(&inet_family_ops);
1865
1866 #ifdef CONFIG_SYSCTL
1867 ip_static_sysctl_init();
1868 #endif
1869
1870 /*
1871 * Add all the base protocols.
1872 */
1873
1874 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1875 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1876 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1877 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1878 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1879 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1880 #ifdef CONFIG_IP_MULTICAST
1881 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1882 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1883 #endif
1884
1885 /* Register the socket-side information for inet_create. */
1886 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1887 INIT_LIST_HEAD(r);
1888
1889 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1890 inet_register_protosw(q);
1891
1892 /*
1893 * Set the ARP module up
1894 */
1895
1896 arp_init();
1897
1898 /*
1899 * Set the IP module up
1900 */
1901
1902 ip_init();
1903
1904 /* Setup TCP slab cache for open requests. */
1905 tcp_init();
1906
1907 /* Setup UDP memory threshold */
1908 udp_init();
1909
1910 /* Add UDP-Lite (RFC 3828) */
1911 udplite4_register();
1912
1913 ping_init();
1914
1915 /*
1916 * Set the ICMP layer up
1917 */
1918
1919 if (icmp_init() < 0)
1920 panic("Failed to create the ICMP control socket.\n");
1921
1922 /*
1923 * Initialise the multicast router
1924 */
1925 #if defined(CONFIG_IP_MROUTE)
1926 if (ip_mr_init())
1927 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1928 #endif
1929
1930 if (init_inet_pernet_ops())
1931 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1932 /*
1933 * Initialise per-cpu ipv4 mibs
1934 */
1935
1936 if (init_ipv4_mibs())
1937 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1938
1939 ipv4_proc_init();
1940
1941 ipfrag_init();
1942
1943 dev_add_pack(&ip_packet_type);
1944
1945 ip_tunnel_core_init();
1946
1947 rc = 0;
1948 out:
1949 return rc;
1950 out_unregister_raw_proto:
1951 proto_unregister(&raw_prot);
1952 out_unregister_udp_proto:
1953 proto_unregister(&udp_prot);
1954 out_unregister_tcp_proto:
1955 proto_unregister(&tcp_prot);
1956 goto out;
1957 }
1958
1959 fs_initcall(inet_init);
1960
1961 /* ------------------------------------------------------------------------ */
1962
1963 #ifdef CONFIG_PROC_FS
1964 static int __init ipv4_proc_init(void)
1965 {
1966 int rc = 0;
1967
1968 if (raw_proc_init())
1969 goto out_raw;
1970 if (tcp4_proc_init())
1971 goto out_tcp;
1972 if (udp4_proc_init())
1973 goto out_udp;
1974 if (ping_proc_init())
1975 goto out_ping;
1976 if (ip_misc_proc_init())
1977 goto out_misc;
1978 out:
1979 return rc;
1980 out_misc:
1981 ping_proc_exit();
1982 out_ping:
1983 udp4_proc_exit();
1984 out_udp:
1985 tcp4_proc_exit();
1986 out_tcp:
1987 raw_proc_exit();
1988 out_raw:
1989 rc = -ENOMEM;
1990 goto out;
1991 }
1992
1993 #else /* CONFIG_PROC_FS */
1994 static int __init ipv4_proc_init(void)
1995 {
1996 return 0;
1997 }
1998 #endif /* CONFIG_PROC_FS */