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
93#include <linux/uaccess.h>
94
95#include <linux/inet.h>
96#include <linux/igmp.h>
97#include <linux/inetdevice.h>
98#include <linux/netdevice.h>
99#include <net/checksum.h>
100#include <net/ip.h>
101#include <net/protocol.h>
102#include <net/arp.h>
103#include <net/route.h>
104#include <net/ip_fib.h>
105#include <net/inet_connection_sock.h>
106#include <net/tcp.h>
107#include <net/udp.h>
108#include <net/udplite.h>
109#include <net/ping.h>
110#include <linux/skbuff.h>
111#include <net/sock.h>
112#include <net/raw.h>
113#include <net/icmp.h>
114#include <net/inet_common.h>
115#include <net/ip_tunnels.h>
116#include <net/xfrm.h>
117#include <net/net_namespace.h>
118#include <net/secure_seq.h>
119#ifdef CONFIG_IP_MROUTE
120#include <linux/mroute.h>
121#endif
122#include <net/l3mdev.h>
123
124#include <trace/events/sock.h>
125
126/* The inetsw table contains everything that inet_create needs to
127 * build a new socket.
128 */
129static struct list_head inetsw[SOCK_MAX];
130static DEFINE_SPINLOCK(inetsw_lock);
131
132/* New destruction routine */
133
134void inet_sock_destruct(struct sock *sk)
135{
136 struct inet_sock *inet = inet_sk(sk);
137
138 __skb_queue_purge(&sk->sk_receive_queue);
139 __skb_queue_purge(&sk->sk_error_queue);
140
141 sk_mem_reclaim(sk);
142
143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144 pr_err("Attempt to release TCP socket in state %d %p\n",
145 sk->sk_state, sk);
146 return;
147 }
148 if (!sock_flag(sk, SOCK_DEAD)) {
149 pr_err("Attempt to release alive inet socket %p\n", sk);
150 return;
151 }
152
153 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
155 WARN_ON(sk->sk_wmem_queued);
156 WARN_ON(sk->sk_forward_alloc);
157
158 kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
160 dst_release(sk->sk_rx_dst);
161 sk_refcnt_debug_dec(sk);
162}
163EXPORT_SYMBOL(inet_sock_destruct);
164
165/*
166 * The routines beyond this point handle the behaviour of an AF_INET
167 * socket object. Mostly it punts to the subprotocols of IP to do
168 * the work.
169 */
170
171/*
172 * Automatically bind an unbound socket.
173 */
174
175static int inet_autobind(struct sock *sk)
176{
177 struct inet_sock *inet;
178 /* We may need to bind the socket. */
179 lock_sock(sk);
180 inet = inet_sk(sk);
181 if (!inet->inet_num) {
182 if (sk->sk_prot->get_port(sk, 0)) {
183 release_sock(sk);
184 return -EAGAIN;
185 }
186 inet->inet_sport = htons(inet->inet_num);
187 }
188 release_sock(sk);
189 return 0;
190}
191
192/*
193 * Move a socket into listening state.
194 */
195int inet_listen(struct socket *sock, int backlog)
196{
197 struct sock *sk = sock->sk;
198 unsigned char old_state;
199 int err, tcp_fastopen;
200
201 lock_sock(sk);
202
203 err = -EINVAL;
204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205 goto out;
206
207 old_state = sk->sk_state;
208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209 goto out;
210
211 /* Really, if the socket is already in listen state
212 * we can only allow the backlog to be adjusted.
213 */
214 if (old_state != TCP_LISTEN) {
215 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
216 * Note that only TCP sockets (SOCK_STREAM) will reach here.
217 * Also fastopen backlog may already been set via the option
218 * because the socket was in TCP_LISTEN state previously but
219 * was shutdown() rather than close().
220 */
221 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
222 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
223 (tcp_fastopen & TFO_SERVER_ENABLE) &&
224 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
225 fastopen_queue_tune(sk, backlog);
226 tcp_fastopen_init_key_once(sock_net(sk));
227 }
228
229 err = inet_csk_listen_start(sk, backlog);
230 if (err)
231 goto out;
232 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
233 }
234 sk->sk_max_ack_backlog = backlog;
235 err = 0;
236
237out:
238 release_sock(sk);
239 return err;
240}
241EXPORT_SYMBOL(inet_listen);
242
243/*
244 * Create an inet socket.
245 */
246
247static int inet_create(struct net *net, struct socket *sock, int protocol,
248 int kern)
249{
250 struct sock *sk;
251 struct inet_protosw *answer;
252 struct inet_sock *inet;
253 struct proto *answer_prot;
254 unsigned char answer_flags;
255 int try_loading_module = 0;
256 int err;
257
258 if (protocol < 0 || protocol >= IPPROTO_MAX)
259 return -EINVAL;
260
261 sock->state = SS_UNCONNECTED;
262
263 /* Look for the requested type/protocol pair. */
264lookup_protocol:
265 err = -ESOCKTNOSUPPORT;
266 rcu_read_lock();
267 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
268
269 err = 0;
270 /* Check the non-wild match. */
271 if (protocol == answer->protocol) {
272 if (protocol != IPPROTO_IP)
273 break;
274 } else {
275 /* Check for the two wild cases. */
276 if (IPPROTO_IP == protocol) {
277 protocol = answer->protocol;
278 break;
279 }
280 if (IPPROTO_IP == answer->protocol)
281 break;
282 }
283 err = -EPROTONOSUPPORT;
284 }
285
286 if (unlikely(err)) {
287 if (try_loading_module < 2) {
288 rcu_read_unlock();
289 /*
290 * Be more specific, e.g. net-pf-2-proto-132-type-1
291 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
292 */
293 if (++try_loading_module == 1)
294 request_module("net-pf-%d-proto-%d-type-%d",
295 PF_INET, protocol, sock->type);
296 /*
297 * Fall back to generic, e.g. net-pf-2-proto-132
298 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
299 */
300 else
301 request_module("net-pf-%d-proto-%d",
302 PF_INET, protocol);
303 goto lookup_protocol;
304 } else
305 goto out_rcu_unlock;
306 }
307
308 err = -EPERM;
309 if (sock->type == SOCK_RAW && !kern &&
310 !ns_capable(net->user_ns, CAP_NET_RAW))
311 goto out_rcu_unlock;
312
313 sock->ops = answer->ops;
314 answer_prot = answer->prot;
315 answer_flags = answer->flags;
316 rcu_read_unlock();
317
318 WARN_ON(!answer_prot->slab);
319
320 err = -ENOBUFS;
321 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
322 if (!sk)
323 goto out;
324
325 err = 0;
326 if (INET_PROTOSW_REUSE & answer_flags)
327 sk->sk_reuse = SK_CAN_REUSE;
328
329 inet = inet_sk(sk);
330 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
331
332 inet->nodefrag = 0;
333
334 if (SOCK_RAW == sock->type) {
335 inet->inet_num = protocol;
336 if (IPPROTO_RAW == protocol)
337 inet->hdrincl = 1;
338 }
339
340 if (net->ipv4.sysctl_ip_no_pmtu_disc)
341 inet->pmtudisc = IP_PMTUDISC_DONT;
342 else
343 inet->pmtudisc = IP_PMTUDISC_WANT;
344
345 inet->inet_id = 0;
346
347 sock_init_data(sock, sk);
348
349 sk->sk_destruct = inet_sock_destruct;
350 sk->sk_protocol = protocol;
351 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
352
353 inet->uc_ttl = -1;
354 inet->mc_loop = 1;
355 inet->mc_ttl = 1;
356 inet->mc_all = 1;
357 inet->mc_index = 0;
358 inet->mc_list = NULL;
359 inet->rcv_tos = 0;
360
361 sk_refcnt_debug_inc(sk);
362
363 if (inet->inet_num) {
364 /* It assumes that any protocol which allows
365 * the user to assign a number at socket
366 * creation time automatically
367 * shares.
368 */
369 inet->inet_sport = htons(inet->inet_num);
370 /* Add to protocol hash chains. */
371 err = sk->sk_prot->hash(sk);
372 if (err) {
373 sk_common_release(sk);
374 goto out;
375 }
376 }
377
378 if (sk->sk_prot->init) {
379 err = sk->sk_prot->init(sk);
380 if (err) {
381 sk_common_release(sk);
382 goto out;
383 }
384 }
385
386 if (!kern) {
387 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
388 if (err) {
389 sk_common_release(sk);
390 goto out;
391 }
392 }
393out:
394 return err;
395out_rcu_unlock:
396 rcu_read_unlock();
397 goto out;
398}
399
400
401/*
402 * The peer socket should always be NULL (or else). When we call this
403 * function we are destroying the object and from then on nobody
404 * should refer to it.
405 */
406int inet_release(struct socket *sock)
407{
408 struct sock *sk = sock->sk;
409
410 if (sk) {
411 long timeout;
412
413 /* Applications forget to leave groups before exiting */
414 ip_mc_drop_socket(sk);
415
416 /* If linger is set, we don't return until the close
417 * is complete. Otherwise we return immediately. The
418 * actually closing is done the same either way.
419 *
420 * If the close is due to the process exiting, we never
421 * linger..
422 */
423 timeout = 0;
424 if (sock_flag(sk, SOCK_LINGER) &&
425 !(current->flags & PF_EXITING))
426 timeout = sk->sk_lingertime;
427 sock->sk = NULL;
428 sk->sk_prot->close(sk, timeout);
429 }
430 return 0;
431}
432EXPORT_SYMBOL(inet_release);
433
434int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
435{
436 struct sock *sk = sock->sk;
437 int err;
438
439 /* If the socket has its own bind function then use it. (RAW) */
440 if (sk->sk_prot->bind) {
441 return sk->sk_prot->bind(sk, uaddr, addr_len);
442 }
443 if (addr_len < sizeof(struct sockaddr_in))
444 return -EINVAL;
445
446 /* BPF prog is run before any checks are done so that if the prog
447 * changes context in a wrong way it will be caught.
448 */
449 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
450 if (err)
451 return err;
452
453 return __inet_bind(sk, uaddr, addr_len, false, true);
454}
455EXPORT_SYMBOL(inet_bind);
456
457int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
458 bool force_bind_address_no_port, bool with_lock)
459{
460 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
461 struct inet_sock *inet = inet_sk(sk);
462 struct net *net = sock_net(sk);
463 unsigned short snum;
464 int chk_addr_ret;
465 u32 tb_id = RT_TABLE_LOCAL;
466 int err;
467
468 if (addr->sin_family != AF_INET) {
469 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
470 * only if s_addr is INADDR_ANY.
471 */
472 err = -EAFNOSUPPORT;
473 if (addr->sin_family != AF_UNSPEC ||
474 addr->sin_addr.s_addr != htonl(INADDR_ANY))
475 goto out;
476 }
477
478 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
479 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
480
481 /* Not specified by any standard per-se, however it breaks too
482 * many applications when removed. It is unfortunate since
483 * allowing applications to make a non-local bind solves
484 * several problems with systems using dynamic addressing.
485 * (ie. your servers still start up even if your ISDN link
486 * is temporarily down)
487 */
488 err = -EADDRNOTAVAIL;
489 if (!inet_can_nonlocal_bind(net, inet) &&
490 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
491 chk_addr_ret != RTN_LOCAL &&
492 chk_addr_ret != RTN_MULTICAST &&
493 chk_addr_ret != RTN_BROADCAST)
494 goto out;
495
496 snum = ntohs(addr->sin_port);
497 err = -EACCES;
498 if (snum && snum < inet_prot_sock(net) &&
499 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
500 goto out;
501
502 /* We keep a pair of addresses. rcv_saddr is the one
503 * used by hash lookups, and saddr is used for transmit.
504 *
505 * In the BSD API these are the same except where it
506 * would be illegal to use them (multicast/broadcast) in
507 * which case the sending device address is used.
508 */
509 if (with_lock)
510 lock_sock(sk);
511
512 /* Check these errors (active socket, double bind). */
513 err = -EINVAL;
514 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
515 goto out_release_sock;
516
517 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
518 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
519 inet->inet_saddr = 0; /* Use device */
520
521 /* Make sure we are allowed to bind here. */
522 if (snum || !(inet->bind_address_no_port ||
523 force_bind_address_no_port)) {
524 if (sk->sk_prot->get_port(sk, snum)) {
525 inet->inet_saddr = inet->inet_rcv_saddr = 0;
526 err = -EADDRINUSE;
527 goto out_release_sock;
528 }
529 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
530 if (err) {
531 inet->inet_saddr = inet->inet_rcv_saddr = 0;
532 goto out_release_sock;
533 }
534 }
535
536 if (inet->inet_rcv_saddr)
537 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
538 if (snum)
539 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
540 inet->inet_sport = htons(inet->inet_num);
541 inet->inet_daddr = 0;
542 inet->inet_dport = 0;
543 sk_dst_reset(sk);
544 err = 0;
545out_release_sock:
546 if (with_lock)
547 release_sock(sk);
548out:
549 return err;
550}
551
552int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
553 int addr_len, int flags)
554{
555 struct sock *sk = sock->sk;
556 int err;
557
558 if (addr_len < sizeof(uaddr->sa_family))
559 return -EINVAL;
560 if (uaddr->sa_family == AF_UNSPEC)
561 return sk->sk_prot->disconnect(sk, flags);
562
563 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
564 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
565 if (err)
566 return err;
567 }
568
569 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
570 return -EAGAIN;
571 return sk->sk_prot->connect(sk, uaddr, addr_len);
572}
573EXPORT_SYMBOL(inet_dgram_connect);
574
575static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
576{
577 DEFINE_WAIT_FUNC(wait, woken_wake_function);
578
579 add_wait_queue(sk_sleep(sk), &wait);
580 sk->sk_write_pending += writebias;
581
582 /* Basic assumption: if someone sets sk->sk_err, he _must_
583 * change state of the socket from TCP_SYN_*.
584 * Connect() does not allow to get error notifications
585 * without closing the socket.
586 */
587 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
588 release_sock(sk);
589 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
590 lock_sock(sk);
591 if (signal_pending(current) || !timeo)
592 break;
593 }
594 remove_wait_queue(sk_sleep(sk), &wait);
595 sk->sk_write_pending -= writebias;
596 return timeo;
597}
598
599/*
600 * Connect to a remote host. There is regrettably still a little
601 * TCP 'magic' in here.
602 */
603int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
604 int addr_len, int flags, int is_sendmsg)
605{
606 struct sock *sk = sock->sk;
607 int err;
608 long timeo;
609
610 /*
611 * uaddr can be NULL and addr_len can be 0 if:
612 * sk is a TCP fastopen active socket and
613 * TCP_FASTOPEN_CONNECT sockopt is set and
614 * we already have a valid cookie for this socket.
615 * In this case, user can call write() after connect().
616 * write() will invoke tcp_sendmsg_fastopen() which calls
617 * __inet_stream_connect().
618 */
619 if (uaddr) {
620 if (addr_len < sizeof(uaddr->sa_family))
621 return -EINVAL;
622
623 if (uaddr->sa_family == AF_UNSPEC) {
624 err = sk->sk_prot->disconnect(sk, flags);
625 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
626 goto out;
627 }
628 }
629
630 switch (sock->state) {
631 default:
632 err = -EINVAL;
633 goto out;
634 case SS_CONNECTED:
635 err = -EISCONN;
636 goto out;
637 case SS_CONNECTING:
638 if (inet_sk(sk)->defer_connect)
639 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
640 else
641 err = -EALREADY;
642 /* Fall out of switch with err, set for this state */
643 break;
644 case SS_UNCONNECTED:
645 err = -EISCONN;
646 if (sk->sk_state != TCP_CLOSE)
647 goto out;
648
649 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
650 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
651 if (err)
652 goto out;
653 }
654
655 err = sk->sk_prot->connect(sk, uaddr, addr_len);
656 if (err < 0)
657 goto out;
658
659 sock->state = SS_CONNECTING;
660
661 if (!err && inet_sk(sk)->defer_connect)
662 goto out;
663
664 /* Just entered SS_CONNECTING state; the only
665 * difference is that return value in non-blocking
666 * case is EINPROGRESS, rather than EALREADY.
667 */
668 err = -EINPROGRESS;
669 break;
670 }
671
672 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
673
674 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
675 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
676 tcp_sk(sk)->fastopen_req &&
677 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
678
679 /* Error code is set above */
680 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
681 goto out;
682
683 err = sock_intr_errno(timeo);
684 if (signal_pending(current))
685 goto out;
686 }
687
688 /* Connection was closed by RST, timeout, ICMP error
689 * or another process disconnected us.
690 */
691 if (sk->sk_state == TCP_CLOSE)
692 goto sock_error;
693
694 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
695 * and error was received after socket entered established state.
696 * Hence, it is handled normally after connect() return successfully.
697 */
698
699 sock->state = SS_CONNECTED;
700 err = 0;
701out:
702 return err;
703
704sock_error:
705 err = sock_error(sk) ? : -ECONNABORTED;
706 sock->state = SS_UNCONNECTED;
707 if (sk->sk_prot->disconnect(sk, flags))
708 sock->state = SS_DISCONNECTING;
709 goto out;
710}
711EXPORT_SYMBOL(__inet_stream_connect);
712
713int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
714 int addr_len, int flags)
715{
716 int err;
717
718 lock_sock(sock->sk);
719 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
720 release_sock(sock->sk);
721 return err;
722}
723EXPORT_SYMBOL(inet_stream_connect);
724
725/*
726 * Accept a pending connection. The TCP layer now gives BSD semantics.
727 */
728
729int inet_accept(struct socket *sock, struct socket *newsock, int flags,
730 bool kern)
731{
732 struct sock *sk1 = sock->sk;
733 int err = -EINVAL;
734 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
735
736 if (!sk2)
737 goto do_err;
738
739 lock_sock(sk2);
740
741 sock_rps_record_flow(sk2);
742 WARN_ON(!((1 << sk2->sk_state) &
743 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
744 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
745
746 sock_graft(sk2, newsock);
747
748 newsock->state = SS_CONNECTED;
749 err = 0;
750 release_sock(sk2);
751do_err:
752 return err;
753}
754EXPORT_SYMBOL(inet_accept);
755
756
757/*
758 * This does both peername and sockname.
759 */
760int inet_getname(struct socket *sock, struct sockaddr *uaddr,
761 int peer)
762{
763 struct sock *sk = sock->sk;
764 struct inet_sock *inet = inet_sk(sk);
765 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
766
767 sin->sin_family = AF_INET;
768 if (peer) {
769 if (!inet->inet_dport ||
770 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
771 peer == 1))
772 return -ENOTCONN;
773 sin->sin_port = inet->inet_dport;
774 sin->sin_addr.s_addr = inet->inet_daddr;
775 } else {
776 __be32 addr = inet->inet_rcv_saddr;
777 if (!addr)
778 addr = inet->inet_saddr;
779 sin->sin_port = inet->inet_sport;
780 sin->sin_addr.s_addr = addr;
781 }
782 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
783 return sizeof(*sin);
784}
785EXPORT_SYMBOL(inet_getname);
786
787int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
788{
789 struct sock *sk = sock->sk;
790
791 sock_rps_record_flow(sk);
792
793 /* We may need to bind the socket. */
794 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
795 inet_autobind(sk))
796 return -EAGAIN;
797
798 return sk->sk_prot->sendmsg(sk, msg, size);
799}
800EXPORT_SYMBOL(inet_sendmsg);
801
802ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
803 size_t size, int flags)
804{
805 struct sock *sk = sock->sk;
806
807 sock_rps_record_flow(sk);
808
809 /* We may need to bind the socket. */
810 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
811 inet_autobind(sk))
812 return -EAGAIN;
813
814 if (sk->sk_prot->sendpage)
815 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
816 return sock_no_sendpage(sock, page, offset, size, flags);
817}
818EXPORT_SYMBOL(inet_sendpage);
819
820int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
821 int flags)
822{
823 struct sock *sk = sock->sk;
824 int addr_len = 0;
825 int err;
826
827 if (likely(!(flags & MSG_ERRQUEUE)))
828 sock_rps_record_flow(sk);
829
830 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
831 flags & ~MSG_DONTWAIT, &addr_len);
832 if (err >= 0)
833 msg->msg_namelen = addr_len;
834 return err;
835}
836EXPORT_SYMBOL(inet_recvmsg);
837
838int inet_shutdown(struct socket *sock, int how)
839{
840 struct sock *sk = sock->sk;
841 int err = 0;
842
843 /* This should really check to make sure
844 * the socket is a TCP socket. (WHY AC...)
845 */
846 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
847 1->2 bit 2 snds.
848 2->3 */
849 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
850 return -EINVAL;
851
852 lock_sock(sk);
853 if (sock->state == SS_CONNECTING) {
854 if ((1 << sk->sk_state) &
855 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
856 sock->state = SS_DISCONNECTING;
857 else
858 sock->state = SS_CONNECTED;
859 }
860
861 switch (sk->sk_state) {
862 case TCP_CLOSE:
863 err = -ENOTCONN;
864 /* Hack to wake up other listeners, who can poll for
865 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
866 /* fall through */
867 default:
868 sk->sk_shutdown |= how;
869 if (sk->sk_prot->shutdown)
870 sk->sk_prot->shutdown(sk, how);
871 break;
872
873 /* Remaining two branches are temporary solution for missing
874 * close() in multithreaded environment. It is _not_ a good idea,
875 * but we have no choice until close() is repaired at VFS level.
876 */
877 case TCP_LISTEN:
878 if (!(how & RCV_SHUTDOWN))
879 break;
880 /* fall through */
881 case TCP_SYN_SENT:
882 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
883 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
884 break;
885 }
886
887 /* Wake up anyone sleeping in poll. */
888 sk->sk_state_change(sk);
889 release_sock(sk);
890 return err;
891}
892EXPORT_SYMBOL(inet_shutdown);
893
894/*
895 * ioctl() calls you can issue on an INET socket. Most of these are
896 * device configuration and stuff and very rarely used. Some ioctls
897 * pass on to the socket itself.
898 *
899 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
900 * loads the devconfigure module does its configuring and unloads it.
901 * There's a good 20K of config code hanging around the kernel.
902 */
903
904int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
905{
906 struct sock *sk = sock->sk;
907 int err = 0;
908 struct net *net = sock_net(sk);
909 void __user *p = (void __user *)arg;
910 struct ifreq ifr;
911 struct rtentry rt;
912
913 switch (cmd) {
914 case SIOCGSTAMP:
915 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
916 break;
917 case SIOCGSTAMPNS:
918 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
919 break;
920 case SIOCADDRT:
921 case SIOCDELRT:
922 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
923 return -EFAULT;
924 err = ip_rt_ioctl(net, cmd, &rt);
925 break;
926 case SIOCRTMSG:
927 err = -EINVAL;
928 break;
929 case SIOCDARP:
930 case SIOCGARP:
931 case SIOCSARP:
932 err = arp_ioctl(net, cmd, (void __user *)arg);
933 break;
934 case SIOCGIFADDR:
935 case SIOCGIFBRDADDR:
936 case SIOCGIFNETMASK:
937 case SIOCGIFDSTADDR:
938 case SIOCGIFPFLAGS:
939 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
940 return -EFAULT;
941 err = devinet_ioctl(net, cmd, &ifr);
942 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
943 err = -EFAULT;
944 break;
945
946 case SIOCSIFADDR:
947 case SIOCSIFBRDADDR:
948 case SIOCSIFNETMASK:
949 case SIOCSIFDSTADDR:
950 case SIOCSIFPFLAGS:
951 case SIOCSIFFLAGS:
952 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
953 return -EFAULT;
954 err = devinet_ioctl(net, cmd, &ifr);
955 break;
956 default:
957 if (sk->sk_prot->ioctl)
958 err = sk->sk_prot->ioctl(sk, cmd, arg);
959 else
960 err = -ENOIOCTLCMD;
961 break;
962 }
963 return err;
964}
965EXPORT_SYMBOL(inet_ioctl);
966
967#ifdef CONFIG_COMPAT
968static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
969{
970 struct sock *sk = sock->sk;
971 int err = -ENOIOCTLCMD;
972
973 if (sk->sk_prot->compat_ioctl)
974 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
975
976 return err;
977}
978#endif
979
980const struct proto_ops inet_stream_ops = {
981 .family = PF_INET,
982 .owner = THIS_MODULE,
983 .release = inet_release,
984 .bind = inet_bind,
985 .connect = inet_stream_connect,
986 .socketpair = sock_no_socketpair,
987 .accept = inet_accept,
988 .getname = inet_getname,
989 .poll = tcp_poll,
990 .ioctl = inet_ioctl,
991 .listen = inet_listen,
992 .shutdown = inet_shutdown,
993 .setsockopt = sock_common_setsockopt,
994 .getsockopt = sock_common_getsockopt,
995 .sendmsg = inet_sendmsg,
996 .recvmsg = inet_recvmsg,
997#ifdef CONFIG_MMU
998 .mmap = tcp_mmap,
999#endif
1000 .sendpage = inet_sendpage,
1001 .splice_read = tcp_splice_read,
1002 .read_sock = tcp_read_sock,
1003 .sendmsg_locked = tcp_sendmsg_locked,
1004 .sendpage_locked = tcp_sendpage_locked,
1005 .peek_len = tcp_peek_len,
1006#ifdef CONFIG_COMPAT
1007 .compat_setsockopt = compat_sock_common_setsockopt,
1008 .compat_getsockopt = compat_sock_common_getsockopt,
1009 .compat_ioctl = inet_compat_ioctl,
1010#endif
1011 .set_rcvlowat = tcp_set_rcvlowat,
1012};
1013EXPORT_SYMBOL(inet_stream_ops);
1014
1015const struct proto_ops inet_dgram_ops = {
1016 .family = PF_INET,
1017 .owner = THIS_MODULE,
1018 .release = inet_release,
1019 .bind = inet_bind,
1020 .connect = inet_dgram_connect,
1021 .socketpair = sock_no_socketpair,
1022 .accept = sock_no_accept,
1023 .getname = inet_getname,
1024 .poll = udp_poll,
1025 .ioctl = inet_ioctl,
1026 .listen = sock_no_listen,
1027 .shutdown = inet_shutdown,
1028 .setsockopt = sock_common_setsockopt,
1029 .getsockopt = sock_common_getsockopt,
1030 .sendmsg = inet_sendmsg,
1031 .recvmsg = inet_recvmsg,
1032 .mmap = sock_no_mmap,
1033 .sendpage = inet_sendpage,
1034 .set_peek_off = sk_set_peek_off,
1035#ifdef CONFIG_COMPAT
1036 .compat_setsockopt = compat_sock_common_setsockopt,
1037 .compat_getsockopt = compat_sock_common_getsockopt,
1038 .compat_ioctl = inet_compat_ioctl,
1039#endif
1040};
1041EXPORT_SYMBOL(inet_dgram_ops);
1042
1043/*
1044 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1045 * udp_poll
1046 */
1047static const struct proto_ops inet_sockraw_ops = {
1048 .family = PF_INET,
1049 .owner = THIS_MODULE,
1050 .release = inet_release,
1051 .bind = inet_bind,
1052 .connect = inet_dgram_connect,
1053 .socketpair = sock_no_socketpair,
1054 .accept = sock_no_accept,
1055 .getname = inet_getname,
1056 .poll = datagram_poll,
1057 .ioctl = inet_ioctl,
1058 .listen = sock_no_listen,
1059 .shutdown = inet_shutdown,
1060 .setsockopt = sock_common_setsockopt,
1061 .getsockopt = sock_common_getsockopt,
1062 .sendmsg = inet_sendmsg,
1063 .recvmsg = inet_recvmsg,
1064 .mmap = sock_no_mmap,
1065 .sendpage = inet_sendpage,
1066#ifdef CONFIG_COMPAT
1067 .compat_setsockopt = compat_sock_common_setsockopt,
1068 .compat_getsockopt = compat_sock_common_getsockopt,
1069 .compat_ioctl = inet_compat_ioctl,
1070#endif
1071};
1072
1073static const struct net_proto_family inet_family_ops = {
1074 .family = PF_INET,
1075 .create = inet_create,
1076 .owner = THIS_MODULE,
1077};
1078
1079/* Upon startup we insert all the elements in inetsw_array[] into
1080 * the linked list inetsw.
1081 */
1082static struct inet_protosw inetsw_array[] =
1083{
1084 {
1085 .type = SOCK_STREAM,
1086 .protocol = IPPROTO_TCP,
1087 .prot = &tcp_prot,
1088 .ops = &inet_stream_ops,
1089 .flags = INET_PROTOSW_PERMANENT |
1090 INET_PROTOSW_ICSK,
1091 },
1092
1093 {
1094 .type = SOCK_DGRAM,
1095 .protocol = IPPROTO_UDP,
1096 .prot = &udp_prot,
1097 .ops = &inet_dgram_ops,
1098 .flags = INET_PROTOSW_PERMANENT,
1099 },
1100
1101 {
1102 .type = SOCK_DGRAM,
1103 .protocol = IPPROTO_ICMP,
1104 .prot = &ping_prot,
1105 .ops = &inet_sockraw_ops,
1106 .flags = INET_PROTOSW_REUSE,
1107 },
1108
1109 {
1110 .type = SOCK_RAW,
1111 .protocol = IPPROTO_IP, /* wild card */
1112 .prot = &raw_prot,
1113 .ops = &inet_sockraw_ops,
1114 .flags = INET_PROTOSW_REUSE,
1115 }
1116};
1117
1118#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1119
1120void inet_register_protosw(struct inet_protosw *p)
1121{
1122 struct list_head *lh;
1123 struct inet_protosw *answer;
1124 int protocol = p->protocol;
1125 struct list_head *last_perm;
1126
1127 spin_lock_bh(&inetsw_lock);
1128
1129 if (p->type >= SOCK_MAX)
1130 goto out_illegal;
1131
1132 /* If we are trying to override a permanent protocol, bail. */
1133 last_perm = &inetsw[p->type];
1134 list_for_each(lh, &inetsw[p->type]) {
1135 answer = list_entry(lh, struct inet_protosw, list);
1136 /* Check only the non-wild match. */
1137 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1138 break;
1139 if (protocol == answer->protocol)
1140 goto out_permanent;
1141 last_perm = lh;
1142 }
1143
1144 /* Add the new entry after the last permanent entry if any, so that
1145 * the new entry does not override a permanent entry when matched with
1146 * a wild-card protocol. But it is allowed to override any existing
1147 * non-permanent entry. This means that when we remove this entry, the
1148 * system automatically returns to the old behavior.
1149 */
1150 list_add_rcu(&p->list, last_perm);
1151out:
1152 spin_unlock_bh(&inetsw_lock);
1153
1154 return;
1155
1156out_permanent:
1157 pr_err("Attempt to override permanent protocol %d\n", protocol);
1158 goto out;
1159
1160out_illegal:
1161 pr_err("Ignoring attempt to register invalid socket type %d\n",
1162 p->type);
1163 goto out;
1164}
1165EXPORT_SYMBOL(inet_register_protosw);
1166
1167void inet_unregister_protosw(struct inet_protosw *p)
1168{
1169 if (INET_PROTOSW_PERMANENT & p->flags) {
1170 pr_err("Attempt to unregister permanent protocol %d\n",
1171 p->protocol);
1172 } else {
1173 spin_lock_bh(&inetsw_lock);
1174 list_del_rcu(&p->list);
1175 spin_unlock_bh(&inetsw_lock);
1176
1177 synchronize_net();
1178 }
1179}
1180EXPORT_SYMBOL(inet_unregister_protosw);
1181
1182static int inet_sk_reselect_saddr(struct sock *sk)
1183{
1184 struct inet_sock *inet = inet_sk(sk);
1185 __be32 old_saddr = inet->inet_saddr;
1186 __be32 daddr = inet->inet_daddr;
1187 struct flowi4 *fl4;
1188 struct rtable *rt;
1189 __be32 new_saddr;
1190 struct ip_options_rcu *inet_opt;
1191
1192 inet_opt = rcu_dereference_protected(inet->inet_opt,
1193 lockdep_sock_is_held(sk));
1194 if (inet_opt && inet_opt->opt.srr)
1195 daddr = inet_opt->opt.faddr;
1196
1197 /* Query new route. */
1198 fl4 = &inet->cork.fl.u.ip4;
1199 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1200 sk->sk_bound_dev_if, sk->sk_protocol,
1201 inet->inet_sport, inet->inet_dport, sk);
1202 if (IS_ERR(rt))
1203 return PTR_ERR(rt);
1204
1205 sk_setup_caps(sk, &rt->dst);
1206
1207 new_saddr = fl4->saddr;
1208
1209 if (new_saddr == old_saddr)
1210 return 0;
1211
1212 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1213 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1214 __func__, &old_saddr, &new_saddr);
1215 }
1216
1217 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1218
1219 /*
1220 * XXX The only one ugly spot where we need to
1221 * XXX really change the sockets identity after
1222 * XXX it has entered the hashes. -DaveM
1223 *
1224 * Besides that, it does not check for connection
1225 * uniqueness. Wait for troubles.
1226 */
1227 return __sk_prot_rehash(sk);
1228}
1229
1230int inet_sk_rebuild_header(struct sock *sk)
1231{
1232 struct inet_sock *inet = inet_sk(sk);
1233 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1234 __be32 daddr;
1235 struct ip_options_rcu *inet_opt;
1236 struct flowi4 *fl4;
1237 int err;
1238
1239 /* Route is OK, nothing to do. */
1240 if (rt)
1241 return 0;
1242
1243 /* Reroute. */
1244 rcu_read_lock();
1245 inet_opt = rcu_dereference(inet->inet_opt);
1246 daddr = inet->inet_daddr;
1247 if (inet_opt && inet_opt->opt.srr)
1248 daddr = inet_opt->opt.faddr;
1249 rcu_read_unlock();
1250 fl4 = &inet->cork.fl.u.ip4;
1251 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1252 inet->inet_dport, inet->inet_sport,
1253 sk->sk_protocol, RT_CONN_FLAGS(sk),
1254 sk->sk_bound_dev_if);
1255 if (!IS_ERR(rt)) {
1256 err = 0;
1257 sk_setup_caps(sk, &rt->dst);
1258 } else {
1259 err = PTR_ERR(rt);
1260
1261 /* Routing failed... */
1262 sk->sk_route_caps = 0;
1263 /*
1264 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1265 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1266 */
1267 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1268 sk->sk_state != TCP_SYN_SENT ||
1269 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1270 (err = inet_sk_reselect_saddr(sk)) != 0)
1271 sk->sk_err_soft = -err;
1272 }
1273
1274 return err;
1275}
1276EXPORT_SYMBOL(inet_sk_rebuild_header);
1277
1278void inet_sk_set_state(struct sock *sk, int state)
1279{
1280 trace_inet_sock_set_state(sk, sk->sk_state, state);
1281 sk->sk_state = state;
1282}
1283EXPORT_SYMBOL(inet_sk_set_state);
1284
1285void inet_sk_state_store(struct sock *sk, int newstate)
1286{
1287 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1288 smp_store_release(&sk->sk_state, newstate);
1289}
1290
1291struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1292 netdev_features_t features)
1293{
1294 bool udpfrag = false, fixedid = false, gso_partial, encap;
1295 struct sk_buff *segs = ERR_PTR(-EINVAL);
1296 const struct net_offload *ops;
1297 unsigned int offset = 0;
1298 struct iphdr *iph;
1299 int proto, tot_len;
1300 int nhoff;
1301 int ihl;
1302 int id;
1303
1304 skb_reset_network_header(skb);
1305 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1306 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1307 goto out;
1308
1309 iph = ip_hdr(skb);
1310 ihl = iph->ihl * 4;
1311 if (ihl < sizeof(*iph))
1312 goto out;
1313
1314 id = ntohs(iph->id);
1315 proto = iph->protocol;
1316
1317 /* Warning: after this point, iph might be no longer valid */
1318 if (unlikely(!pskb_may_pull(skb, ihl)))
1319 goto out;
1320 __skb_pull(skb, ihl);
1321
1322 encap = SKB_GSO_CB(skb)->encap_level > 0;
1323 if (encap)
1324 features &= skb->dev->hw_enc_features;
1325 SKB_GSO_CB(skb)->encap_level += ihl;
1326
1327 skb_reset_transport_header(skb);
1328
1329 segs = ERR_PTR(-EPROTONOSUPPORT);
1330
1331 if (!skb->encapsulation || encap) {
1332 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1333 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1334
1335 /* fixed ID is invalid if DF bit is not set */
1336 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1337 goto out;
1338 }
1339
1340 ops = rcu_dereference(inet_offloads[proto]);
1341 if (likely(ops && ops->callbacks.gso_segment))
1342 segs = ops->callbacks.gso_segment(skb, features);
1343
1344 if (IS_ERR_OR_NULL(segs))
1345 goto out;
1346
1347 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1348
1349 skb = segs;
1350 do {
1351 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1352 if (udpfrag) {
1353 iph->frag_off = htons(offset >> 3);
1354 if (skb->next)
1355 iph->frag_off |= htons(IP_MF);
1356 offset += skb->len - nhoff - ihl;
1357 tot_len = skb->len - nhoff;
1358 } else if (skb_is_gso(skb)) {
1359 if (!fixedid) {
1360 iph->id = htons(id);
1361 id += skb_shinfo(skb)->gso_segs;
1362 }
1363
1364 if (gso_partial)
1365 tot_len = skb_shinfo(skb)->gso_size +
1366 SKB_GSO_CB(skb)->data_offset +
1367 skb->head - (unsigned char *)iph;
1368 else
1369 tot_len = skb->len - nhoff;
1370 } else {
1371 if (!fixedid)
1372 iph->id = htons(id++);
1373 tot_len = skb->len - nhoff;
1374 }
1375 iph->tot_len = htons(tot_len);
1376 ip_send_check(iph);
1377 if (encap)
1378 skb_reset_inner_headers(skb);
1379 skb->network_header = (u8 *)iph - skb->head;
1380 } while ((skb = skb->next));
1381
1382out:
1383 return segs;
1384}
1385EXPORT_SYMBOL(inet_gso_segment);
1386
1387struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1388{
1389 const struct net_offload *ops;
1390 struct sk_buff *pp = NULL;
1391 const struct iphdr *iph;
1392 struct sk_buff *p;
1393 unsigned int hlen;
1394 unsigned int off;
1395 unsigned int id;
1396 int flush = 1;
1397 int proto;
1398
1399 off = skb_gro_offset(skb);
1400 hlen = off + sizeof(*iph);
1401 iph = skb_gro_header_fast(skb, off);
1402 if (skb_gro_header_hard(skb, hlen)) {
1403 iph = skb_gro_header_slow(skb, hlen, off);
1404 if (unlikely(!iph))
1405 goto out;
1406 }
1407
1408 proto = iph->protocol;
1409
1410 rcu_read_lock();
1411 ops = rcu_dereference(inet_offloads[proto]);
1412 if (!ops || !ops->callbacks.gro_receive)
1413 goto out_unlock;
1414
1415 if (*(u8 *)iph != 0x45)
1416 goto out_unlock;
1417
1418 if (ip_is_fragment(iph))
1419 goto out_unlock;
1420
1421 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1422 goto out_unlock;
1423
1424 id = ntohl(*(__be32 *)&iph->id);
1425 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1426 id >>= 16;
1427
1428 list_for_each_entry(p, head, list) {
1429 struct iphdr *iph2;
1430 u16 flush_id;
1431
1432 if (!NAPI_GRO_CB(p)->same_flow)
1433 continue;
1434
1435 iph2 = (struct iphdr *)(p->data + off);
1436 /* The above works because, with the exception of the top
1437 * (inner most) layer, we only aggregate pkts with the same
1438 * hdr length so all the hdrs we'll need to verify will start
1439 * at the same offset.
1440 */
1441 if ((iph->protocol ^ iph2->protocol) |
1442 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1443 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1444 NAPI_GRO_CB(p)->same_flow = 0;
1445 continue;
1446 }
1447
1448 /* All fields must match except length and checksum. */
1449 NAPI_GRO_CB(p)->flush |=
1450 (iph->ttl ^ iph2->ttl) |
1451 (iph->tos ^ iph2->tos) |
1452 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1453
1454 NAPI_GRO_CB(p)->flush |= flush;
1455
1456 /* We need to store of the IP ID check to be included later
1457 * when we can verify that this packet does in fact belong
1458 * to a given flow.
1459 */
1460 flush_id = (u16)(id - ntohs(iph2->id));
1461
1462 /* This bit of code makes it much easier for us to identify
1463 * the cases where we are doing atomic vs non-atomic IP ID
1464 * checks. Specifically an atomic check can return IP ID
1465 * values 0 - 0xFFFF, while a non-atomic check can only
1466 * return 0 or 0xFFFF.
1467 */
1468 if (!NAPI_GRO_CB(p)->is_atomic ||
1469 !(iph->frag_off & htons(IP_DF))) {
1470 flush_id ^= NAPI_GRO_CB(p)->count;
1471 flush_id = flush_id ? 0xFFFF : 0;
1472 }
1473
1474 /* If the previous IP ID value was based on an atomic
1475 * datagram we can overwrite the value and ignore it.
1476 */
1477 if (NAPI_GRO_CB(skb)->is_atomic)
1478 NAPI_GRO_CB(p)->flush_id = flush_id;
1479 else
1480 NAPI_GRO_CB(p)->flush_id |= flush_id;
1481 }
1482
1483 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1484 NAPI_GRO_CB(skb)->flush |= flush;
1485 skb_set_network_header(skb, off);
1486 /* The above will be needed by the transport layer if there is one
1487 * immediately following this IP hdr.
1488 */
1489
1490 /* Note : No need to call skb_gro_postpull_rcsum() here,
1491 * as we already checked checksum over ipv4 header was 0
1492 */
1493 skb_gro_pull(skb, sizeof(*iph));
1494 skb_set_transport_header(skb, skb_gro_offset(skb));
1495
1496 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1497
1498out_unlock:
1499 rcu_read_unlock();
1500
1501out:
1502 skb_gro_flush_final(skb, pp, flush);
1503
1504 return pp;
1505}
1506EXPORT_SYMBOL(inet_gro_receive);
1507
1508static struct sk_buff *ipip_gro_receive(struct list_head *head,
1509 struct sk_buff *skb)
1510{
1511 if (NAPI_GRO_CB(skb)->encap_mark) {
1512 NAPI_GRO_CB(skb)->flush = 1;
1513 return NULL;
1514 }
1515
1516 NAPI_GRO_CB(skb)->encap_mark = 1;
1517
1518 return inet_gro_receive(head, skb);
1519}
1520
1521#define SECONDS_PER_DAY 86400
1522
1523/* inet_current_timestamp - Return IP network timestamp
1524 *
1525 * Return milliseconds since midnight in network byte order.
1526 */
1527__be32 inet_current_timestamp(void)
1528{
1529 u32 secs;
1530 u32 msecs;
1531 struct timespec64 ts;
1532
1533 ktime_get_real_ts64(&ts);
1534
1535 /* Get secs since midnight. */
1536 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1537 /* Convert to msecs. */
1538 msecs = secs * MSEC_PER_SEC;
1539 /* Convert nsec to msec. */
1540 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1541
1542 /* Convert to network byte order. */
1543 return htonl(msecs);
1544}
1545EXPORT_SYMBOL(inet_current_timestamp);
1546
1547int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1548{
1549 if (sk->sk_family == AF_INET)
1550 return ip_recv_error(sk, msg, len, addr_len);
1551#if IS_ENABLED(CONFIG_IPV6)
1552 if (sk->sk_family == AF_INET6)
1553 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1554#endif
1555 return -EINVAL;
1556}
1557
1558int inet_gro_complete(struct sk_buff *skb, int nhoff)
1559{
1560 __be16 newlen = htons(skb->len - nhoff);
1561 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1562 const struct net_offload *ops;
1563 int proto = iph->protocol;
1564 int err = -ENOSYS;
1565
1566 if (skb->encapsulation) {
1567 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1568 skb_set_inner_network_header(skb, nhoff);
1569 }
1570
1571 csum_replace2(&iph->check, iph->tot_len, newlen);
1572 iph->tot_len = newlen;
1573
1574 rcu_read_lock();
1575 ops = rcu_dereference(inet_offloads[proto]);
1576 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1577 goto out_unlock;
1578
1579 /* Only need to add sizeof(*iph) to get to the next hdr below
1580 * because any hdr with option will have been flushed in
1581 * inet_gro_receive().
1582 */
1583 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1584
1585out_unlock:
1586 rcu_read_unlock();
1587
1588 return err;
1589}
1590EXPORT_SYMBOL(inet_gro_complete);
1591
1592static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1593{
1594 skb->encapsulation = 1;
1595 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1596 return inet_gro_complete(skb, nhoff);
1597}
1598
1599int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1600 unsigned short type, unsigned char protocol,
1601 struct net *net)
1602{
1603 struct socket *sock;
1604 int rc = sock_create_kern(net, family, type, protocol, &sock);
1605
1606 if (rc == 0) {
1607 *sk = sock->sk;
1608 (*sk)->sk_allocation = GFP_ATOMIC;
1609 /*
1610 * Unhash it so that IP input processing does not even see it,
1611 * we do not wish this socket to see incoming packets.
1612 */
1613 (*sk)->sk_prot->unhash(*sk);
1614 }
1615 return rc;
1616}
1617EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1618
1619u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1620{
1621 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1622}
1623EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1624
1625unsigned long snmp_fold_field(void __percpu *mib, int offt)
1626{
1627 unsigned long res = 0;
1628 int i;
1629
1630 for_each_possible_cpu(i)
1631 res += snmp_get_cpu_field(mib, i, offt);
1632 return res;
1633}
1634EXPORT_SYMBOL_GPL(snmp_fold_field);
1635
1636#if BITS_PER_LONG==32
1637
1638u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1639 size_t syncp_offset)
1640{
1641 void *bhptr;
1642 struct u64_stats_sync *syncp;
1643 u64 v;
1644 unsigned int start;
1645
1646 bhptr = per_cpu_ptr(mib, cpu);
1647 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1648 do {
1649 start = u64_stats_fetch_begin_irq(syncp);
1650 v = *(((u64 *)bhptr) + offt);
1651 } while (u64_stats_fetch_retry_irq(syncp, start));
1652
1653 return v;
1654}
1655EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1656
1657u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1658{
1659 u64 res = 0;
1660 int cpu;
1661
1662 for_each_possible_cpu(cpu) {
1663 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1664 }
1665 return res;
1666}
1667EXPORT_SYMBOL_GPL(snmp_fold_field64);
1668#endif
1669
1670#ifdef CONFIG_IP_MULTICAST
1671static const struct net_protocol igmp_protocol = {
1672 .handler = igmp_rcv,
1673 .netns_ok = 1,
1674};
1675#endif
1676
1677/* thinking of making this const? Don't.
1678 * early_demux can change based on sysctl.
1679 */
1680static struct net_protocol tcp_protocol = {
1681 .early_demux = tcp_v4_early_demux,
1682 .early_demux_handler = tcp_v4_early_demux,
1683 .handler = tcp_v4_rcv,
1684 .err_handler = tcp_v4_err,
1685 .no_policy = 1,
1686 .netns_ok = 1,
1687 .icmp_strict_tag_validation = 1,
1688};
1689
1690/* thinking of making this const? Don't.
1691 * early_demux can change based on sysctl.
1692 */
1693static struct net_protocol udp_protocol = {
1694 .early_demux = udp_v4_early_demux,
1695 .early_demux_handler = udp_v4_early_demux,
1696 .handler = udp_rcv,
1697 .err_handler = udp_err,
1698 .no_policy = 1,
1699 .netns_ok = 1,
1700};
1701
1702static const struct net_protocol icmp_protocol = {
1703 .handler = icmp_rcv,
1704 .err_handler = icmp_err,
1705 .no_policy = 1,
1706 .netns_ok = 1,
1707};
1708
1709static __net_init int ipv4_mib_init_net(struct net *net)
1710{
1711 int i;
1712
1713 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1714 if (!net->mib.tcp_statistics)
1715 goto err_tcp_mib;
1716 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1717 if (!net->mib.ip_statistics)
1718 goto err_ip_mib;
1719
1720 for_each_possible_cpu(i) {
1721 struct ipstats_mib *af_inet_stats;
1722 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1723 u64_stats_init(&af_inet_stats->syncp);
1724 }
1725
1726 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1727 if (!net->mib.net_statistics)
1728 goto err_net_mib;
1729 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1730 if (!net->mib.udp_statistics)
1731 goto err_udp_mib;
1732 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1733 if (!net->mib.udplite_statistics)
1734 goto err_udplite_mib;
1735 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1736 if (!net->mib.icmp_statistics)
1737 goto err_icmp_mib;
1738 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1739 GFP_KERNEL);
1740 if (!net->mib.icmpmsg_statistics)
1741 goto err_icmpmsg_mib;
1742
1743 tcp_mib_init(net);
1744 return 0;
1745
1746err_icmpmsg_mib:
1747 free_percpu(net->mib.icmp_statistics);
1748err_icmp_mib:
1749 free_percpu(net->mib.udplite_statistics);
1750err_udplite_mib:
1751 free_percpu(net->mib.udp_statistics);
1752err_udp_mib:
1753 free_percpu(net->mib.net_statistics);
1754err_net_mib:
1755 free_percpu(net->mib.ip_statistics);
1756err_ip_mib:
1757 free_percpu(net->mib.tcp_statistics);
1758err_tcp_mib:
1759 return -ENOMEM;
1760}
1761
1762static __net_exit void ipv4_mib_exit_net(struct net *net)
1763{
1764 kfree(net->mib.icmpmsg_statistics);
1765 free_percpu(net->mib.icmp_statistics);
1766 free_percpu(net->mib.udplite_statistics);
1767 free_percpu(net->mib.udp_statistics);
1768 free_percpu(net->mib.net_statistics);
1769 free_percpu(net->mib.ip_statistics);
1770 free_percpu(net->mib.tcp_statistics);
1771}
1772
1773static __net_initdata struct pernet_operations ipv4_mib_ops = {
1774 .init = ipv4_mib_init_net,
1775 .exit = ipv4_mib_exit_net,
1776};
1777
1778static int __init init_ipv4_mibs(void)
1779{
1780 return register_pernet_subsys(&ipv4_mib_ops);
1781}
1782
1783static __net_init int inet_init_net(struct net *net)
1784{
1785 /*
1786 * Set defaults for local port range
1787 */
1788 seqlock_init(&net->ipv4.ip_local_ports.lock);
1789 net->ipv4.ip_local_ports.range[0] = 32768;
1790 net->ipv4.ip_local_ports.range[1] = 60999;
1791
1792 seqlock_init(&net->ipv4.ping_group_range.lock);
1793 /*
1794 * Sane defaults - nobody may create ping sockets.
1795 * Boot scripts should set this to distro-specific group.
1796 */
1797 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1798 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1799
1800 /* Default values for sysctl-controlled parameters.
1801 * We set them here, in case sysctl is not compiled.
1802 */
1803 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1804 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1805 net->ipv4.sysctl_ip_dynaddr = 0;
1806 net->ipv4.sysctl_ip_early_demux = 1;
1807 net->ipv4.sysctl_udp_early_demux = 1;
1808 net->ipv4.sysctl_tcp_early_demux = 1;
1809#ifdef CONFIG_SYSCTL
1810 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1811#endif
1812
1813 /* Some igmp sysctl, whose values are always used */
1814 net->ipv4.sysctl_igmp_max_memberships = 20;
1815 net->ipv4.sysctl_igmp_max_msf = 10;
1816 /* IGMP reports for link-local multicast groups are enabled by default */
1817 net->ipv4.sysctl_igmp_llm_reports = 1;
1818 net->ipv4.sysctl_igmp_qrv = 2;
1819
1820 return 0;
1821}
1822
1823static __net_exit void inet_exit_net(struct net *net)
1824{
1825}
1826
1827static __net_initdata struct pernet_operations af_inet_ops = {
1828 .init = inet_init_net,
1829 .exit = inet_exit_net,
1830};
1831
1832static int __init init_inet_pernet_ops(void)
1833{
1834 return register_pernet_subsys(&af_inet_ops);
1835}
1836
1837static int ipv4_proc_init(void);
1838
1839/*
1840 * IP protocol layer initialiser
1841 */
1842
1843static struct packet_offload ip_packet_offload __read_mostly = {
1844 .type = cpu_to_be16(ETH_P_IP),
1845 .callbacks = {
1846 .gso_segment = inet_gso_segment,
1847 .gro_receive = inet_gro_receive,
1848 .gro_complete = inet_gro_complete,
1849 },
1850};
1851
1852static const struct net_offload ipip_offload = {
1853 .callbacks = {
1854 .gso_segment = inet_gso_segment,
1855 .gro_receive = ipip_gro_receive,
1856 .gro_complete = ipip_gro_complete,
1857 },
1858};
1859
1860static int __init ipip_offload_init(void)
1861{
1862 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1863}
1864
1865static int __init ipv4_offload_init(void)
1866{
1867 /*
1868 * Add offloads
1869 */
1870 if (udpv4_offload_init() < 0)
1871 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1872 if (tcpv4_offload_init() < 0)
1873 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1874 if (ipip_offload_init() < 0)
1875 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1876
1877 dev_add_offload(&ip_packet_offload);
1878 return 0;
1879}
1880
1881fs_initcall(ipv4_offload_init);
1882
1883static struct packet_type ip_packet_type __read_mostly = {
1884 .type = cpu_to_be16(ETH_P_IP),
1885 .func = ip_rcv,
1886 .list_func = ip_list_rcv,
1887};
1888
1889static int __init inet_init(void)
1890{
1891 struct inet_protosw *q;
1892 struct list_head *r;
1893 int rc = -EINVAL;
1894
1895 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1896
1897 rc = proto_register(&tcp_prot, 1);
1898 if (rc)
1899 goto out;
1900
1901 rc = proto_register(&udp_prot, 1);
1902 if (rc)
1903 goto out_unregister_tcp_proto;
1904
1905 rc = proto_register(&raw_prot, 1);
1906 if (rc)
1907 goto out_unregister_udp_proto;
1908
1909 rc = proto_register(&ping_prot, 1);
1910 if (rc)
1911 goto out_unregister_raw_proto;
1912
1913 /*
1914 * Tell SOCKET that we are alive...
1915 */
1916
1917 (void)sock_register(&inet_family_ops);
1918
1919#ifdef CONFIG_SYSCTL
1920 ip_static_sysctl_init();
1921#endif
1922
1923 /*
1924 * Add all the base protocols.
1925 */
1926
1927 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1928 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1929 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1930 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1931 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1932 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1933#ifdef CONFIG_IP_MULTICAST
1934 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1935 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1936#endif
1937
1938 /* Register the socket-side information for inet_create. */
1939 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1940 INIT_LIST_HEAD(r);
1941
1942 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1943 inet_register_protosw(q);
1944
1945 /*
1946 * Set the ARP module up
1947 */
1948
1949 arp_init();
1950
1951 /*
1952 * Set the IP module up
1953 */
1954
1955 ip_init();
1956
1957 /* Setup TCP slab cache for open requests. */
1958 tcp_init();
1959
1960 /* Setup UDP memory threshold */
1961 udp_init();
1962
1963 /* Add UDP-Lite (RFC 3828) */
1964 udplite4_register();
1965
1966 ping_init();
1967
1968 /*
1969 * Set the ICMP layer up
1970 */
1971
1972 if (icmp_init() < 0)
1973 panic("Failed to create the ICMP control socket.\n");
1974
1975 /*
1976 * Initialise the multicast router
1977 */
1978#if defined(CONFIG_IP_MROUTE)
1979 if (ip_mr_init())
1980 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1981#endif
1982
1983 if (init_inet_pernet_ops())
1984 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1985 /*
1986 * Initialise per-cpu ipv4 mibs
1987 */
1988
1989 if (init_ipv4_mibs())
1990 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1991
1992 ipv4_proc_init();
1993
1994 ipfrag_init();
1995
1996 dev_add_pack(&ip_packet_type);
1997
1998 ip_tunnel_core_init();
1999
2000 rc = 0;
2001out:
2002 return rc;
2003out_unregister_raw_proto:
2004 proto_unregister(&raw_prot);
2005out_unregister_udp_proto:
2006 proto_unregister(&udp_prot);
2007out_unregister_tcp_proto:
2008 proto_unregister(&tcp_prot);
2009 goto out;
2010}
2011
2012fs_initcall(inet_init);
2013
2014/* ------------------------------------------------------------------------ */
2015
2016#ifdef CONFIG_PROC_FS
2017static int __init ipv4_proc_init(void)
2018{
2019 int rc = 0;
2020
2021 if (raw_proc_init())
2022 goto out_raw;
2023 if (tcp4_proc_init())
2024 goto out_tcp;
2025 if (udp4_proc_init())
2026 goto out_udp;
2027 if (ping_proc_init())
2028 goto out_ping;
2029 if (ip_misc_proc_init())
2030 goto out_misc;
2031out:
2032 return rc;
2033out_misc:
2034 ping_proc_exit();
2035out_ping:
2036 udp4_proc_exit();
2037out_udp:
2038 tcp4_proc_exit();
2039out_tcp:
2040 raw_proc_exit();
2041out_raw:
2042 rc = -ENOMEM;
2043 goto out;
2044}
2045
2046#else /* CONFIG_PROC_FS */
2047static int __init ipv4_proc_init(void)
2048{
2049 return 0;
2050}
2051#endif /* CONFIG_PROC_FS */
2052