Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * PF_INET protocol family socket handler.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Florian La Roche, <flla@stud.uni-sb.de>
12 * Alan Cox, <A.Cox@swansea.ac.uk>
13 *
14 * Changes (see also sock.c)
15 *
16 * piggy,
17 * Karl Knutson : Socket protocol table
18 * A.N.Kuznetsov : Socket death error in accept().
19 * John Richardson : Fix non blocking error in connect()
20 * so sockets that fail to connect
21 * don't return -EINPROGRESS.
22 * Alan Cox : Asynchronous I/O support
23 * Alan Cox : Keep correct socket pointer on sock
24 * structures
25 * when accept() ed
26 * Alan Cox : Semantics of SO_LINGER aren't state
27 * moved to close when you look carefully.
28 * With this fixed and the accept bug fixed
29 * some RPC stuff seems happier.
30 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Alan Cox,
32 * Tony Gale : Fixed reuse semantics.
33 * Alan Cox : bind() shouldn't abort existing but dead
34 * sockets. Stops FTP netin:.. I hope.
35 * Alan Cox : bind() works correctly for RAW sockets.
36 * Note that FreeBSD at least was broken
37 * in this respect so be careful with
38 * compatibility tests...
39 * Alan Cox : routing cache support
40 * Alan Cox : memzero the socket structure for
41 * compactness.
42 * Matt Day : nonblock connect error handler
43 * Alan Cox : Allow large numbers of pending sockets
44 * (eg for big web sites), but only if
45 * specifically application requested.
46 * Alan Cox : New buffering throughout IP. Used
47 * dumbly.
48 * Alan Cox : New buffering now used smartly.
49 * Alan Cox : BSD rather than common sense
50 * interpretation of listen.
51 * Germano Caronni : Assorted small races.
52 * Alan Cox : sendmsg/recvmsg basic support.
53 * Alan Cox : Only sendmsg/recvmsg now supported.
54 * Alan Cox : Locked down bind (see security list).
55 * Alan Cox : Loosened bind a little.
56 * Mike McLagan : ADD/DEL DLCI Ioctls
57 * Willy Konynenberg : Transparent proxying support.
58 * David S. Miller : New socket lookup architecture.
59 * Some other random speedups.
60 * Cyrus Durgin : Cleaned up file for kmod hacks.
61 * Andi Kleen : Fix inet_stream_connect TCP race.
62 */
63
64#define pr_fmt(fmt) "IPv4: " fmt
65
66#include <linux/err.h>
67#include <linux/errno.h>
68#include <linux/types.h>
69#include <linux/socket.h>
70#include <linux/in.h>
71#include <linux/kernel.h>
72#include <linux/kmod.h>
73#include <linux/sched.h>
74#include <linux/timer.h>
75#include <linux/string.h>
76#include <linux/sockios.h>
77#include <linux/net.h>
78#include <linux/capability.h>
79#include <linux/fcntl.h>
80#include <linux/mm.h>
81#include <linux/interrupt.h>
82#include <linux/stat.h>
83#include <linux/init.h>
84#include <linux/poll.h>
85#include <linux/netfilter_ipv4.h>
86#include <linux/random.h>
87#include <linux/slab.h>
88
89#include <linux/uaccess.h>
90
91#include <linux/inet.h>
92#include <linux/igmp.h>
93#include <linux/inetdevice.h>
94#include <linux/netdevice.h>
95#include <net/checksum.h>
96#include <net/ip.h>
97#include <net/protocol.h>
98#include <net/arp.h>
99#include <net/route.h>
100#include <net/ip_fib.h>
101#include <net/inet_connection_sock.h>
102#include <net/tcp.h>
103#include <net/udp.h>
104#include <net/udplite.h>
105#include <net/ping.h>
106#include <linux/skbuff.h>
107#include <net/sock.h>
108#include <net/raw.h>
109#include <net/icmp.h>
110#include <net/inet_common.h>
111#include <net/ip_tunnels.h>
112#include <net/xfrm.h>
113#include <net/net_namespace.h>
114#include <net/secure_seq.h>
115#ifdef CONFIG_IP_MROUTE
116#include <linux/mroute.h>
117#endif
118#include <net/l3mdev.h>
119
120#include <trace/events/sock.h>
121
122/* The inetsw table contains everything that inet_create needs to
123 * build a new socket.
124 */
125static struct list_head inetsw[SOCK_MAX];
126static DEFINE_SPINLOCK(inetsw_lock);
127
128/* New destruction routine */
129
130void inet_sock_destruct(struct sock *sk)
131{
132 struct inet_sock *inet = inet_sk(sk);
133
134 __skb_queue_purge(&sk->sk_receive_queue);
135 if (sk->sk_rx_skb_cache) {
136 __kfree_skb(sk->sk_rx_skb_cache);
137 sk->sk_rx_skb_cache = NULL;
138 }
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_protected(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 sk->sk_max_ack_backlog = backlog;
212 /* Really, if the socket is already in listen state
213 * we can only allow the backlog to be adjusted.
214 */
215 if (old_state != TCP_LISTEN) {
216 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
217 * Note that only TCP sockets (SOCK_STREAM) will reach here.
218 * Also fastopen backlog may already been set via the option
219 * because the socket was in TCP_LISTEN state previously but
220 * was shutdown() rather than close().
221 */
222 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
223 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
224 (tcp_fastopen & TFO_SERVER_ENABLE) &&
225 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
226 fastopen_queue_tune(sk, backlog);
227 tcp_fastopen_init_key_once(sock_net(sk));
228 }
229
230 err = inet_csk_listen_start(sk, backlog);
231 if (err)
232 goto out;
233 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
234 }
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 sk->sk_prot->close(sk, timeout);
428 sock->sk = NULL;
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_send_prepare(struct sock *sk)
788{
789 sock_rps_record_flow(sk);
790
791 /* We may need to bind the socket. */
792 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
793 inet_autobind(sk))
794 return -EAGAIN;
795
796 return 0;
797}
798EXPORT_SYMBOL_GPL(inet_send_prepare);
799
800int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
801{
802 struct sock *sk = sock->sk;
803
804 if (unlikely(inet_send_prepare(sk)))
805 return -EAGAIN;
806
807 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
808 sk, msg, size);
809}
810EXPORT_SYMBOL(inet_sendmsg);
811
812ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
813 size_t size, int flags)
814{
815 struct sock *sk = sock->sk;
816
817 if (unlikely(inet_send_prepare(sk)))
818 return -EAGAIN;
819
820 if (sk->sk_prot->sendpage)
821 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
822 return sock_no_sendpage(sock, page, offset, size, flags);
823}
824EXPORT_SYMBOL(inet_sendpage);
825
826INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
827 size_t, int, int, int *));
828int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
829 int flags)
830{
831 struct sock *sk = sock->sk;
832 int addr_len = 0;
833 int err;
834
835 if (likely(!(flags & MSG_ERRQUEUE)))
836 sock_rps_record_flow(sk);
837
838 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
839 sk, msg, size, flags & MSG_DONTWAIT,
840 flags & ~MSG_DONTWAIT, &addr_len);
841 if (err >= 0)
842 msg->msg_namelen = addr_len;
843 return err;
844}
845EXPORT_SYMBOL(inet_recvmsg);
846
847int inet_shutdown(struct socket *sock, int how)
848{
849 struct sock *sk = sock->sk;
850 int err = 0;
851
852 /* This should really check to make sure
853 * the socket is a TCP socket. (WHY AC...)
854 */
855 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
856 1->2 bit 2 snds.
857 2->3 */
858 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
859 return -EINVAL;
860
861 lock_sock(sk);
862 if (sock->state == SS_CONNECTING) {
863 if ((1 << sk->sk_state) &
864 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
865 sock->state = SS_DISCONNECTING;
866 else
867 sock->state = SS_CONNECTED;
868 }
869
870 switch (sk->sk_state) {
871 case TCP_CLOSE:
872 err = -ENOTCONN;
873 /* Hack to wake up other listeners, who can poll for
874 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
875 /* fall through */
876 default:
877 sk->sk_shutdown |= how;
878 if (sk->sk_prot->shutdown)
879 sk->sk_prot->shutdown(sk, how);
880 break;
881
882 /* Remaining two branches are temporary solution for missing
883 * close() in multithreaded environment. It is _not_ a good idea,
884 * but we have no choice until close() is repaired at VFS level.
885 */
886 case TCP_LISTEN:
887 if (!(how & RCV_SHUTDOWN))
888 break;
889 /* fall through */
890 case TCP_SYN_SENT:
891 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
892 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
893 break;
894 }
895
896 /* Wake up anyone sleeping in poll. */
897 sk->sk_state_change(sk);
898 release_sock(sk);
899 return err;
900}
901EXPORT_SYMBOL(inet_shutdown);
902
903/*
904 * ioctl() calls you can issue on an INET socket. Most of these are
905 * device configuration and stuff and very rarely used. Some ioctls
906 * pass on to the socket itself.
907 *
908 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
909 * loads the devconfigure module does its configuring and unloads it.
910 * There's a good 20K of config code hanging around the kernel.
911 */
912
913int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
914{
915 struct sock *sk = sock->sk;
916 int err = 0;
917 struct net *net = sock_net(sk);
918 void __user *p = (void __user *)arg;
919 struct ifreq ifr;
920 struct rtentry rt;
921
922 switch (cmd) {
923 case SIOCADDRT:
924 case SIOCDELRT:
925 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
926 return -EFAULT;
927 err = ip_rt_ioctl(net, cmd, &rt);
928 break;
929 case SIOCRTMSG:
930 err = -EINVAL;
931 break;
932 case SIOCDARP:
933 case SIOCGARP:
934 case SIOCSARP:
935 err = arp_ioctl(net, cmd, (void __user *)arg);
936 break;
937 case SIOCGIFADDR:
938 case SIOCGIFBRDADDR:
939 case SIOCGIFNETMASK:
940 case SIOCGIFDSTADDR:
941 case SIOCGIFPFLAGS:
942 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
943 return -EFAULT;
944 err = devinet_ioctl(net, cmd, &ifr);
945 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
946 err = -EFAULT;
947 break;
948
949 case SIOCSIFADDR:
950 case SIOCSIFBRDADDR:
951 case SIOCSIFNETMASK:
952 case SIOCSIFDSTADDR:
953 case SIOCSIFPFLAGS:
954 case SIOCSIFFLAGS:
955 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
956 return -EFAULT;
957 err = devinet_ioctl(net, cmd, &ifr);
958 break;
959 default:
960 if (sk->sk_prot->ioctl)
961 err = sk->sk_prot->ioctl(sk, cmd, arg);
962 else
963 err = -ENOIOCTLCMD;
964 break;
965 }
966 return err;
967}
968EXPORT_SYMBOL(inet_ioctl);
969
970#ifdef CONFIG_COMPAT
971static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
972{
973 struct sock *sk = sock->sk;
974 int err = -ENOIOCTLCMD;
975
976 if (sk->sk_prot->compat_ioctl)
977 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
978
979 return err;
980}
981#endif
982
983const struct proto_ops inet_stream_ops = {
984 .family = PF_INET,
985 .owner = THIS_MODULE,
986 .release = inet_release,
987 .bind = inet_bind,
988 .connect = inet_stream_connect,
989 .socketpair = sock_no_socketpair,
990 .accept = inet_accept,
991 .getname = inet_getname,
992 .poll = tcp_poll,
993 .ioctl = inet_ioctl,
994 .gettstamp = sock_gettstamp,
995 .listen = inet_listen,
996 .shutdown = inet_shutdown,
997 .setsockopt = sock_common_setsockopt,
998 .getsockopt = sock_common_getsockopt,
999 .sendmsg = inet_sendmsg,
1000 .recvmsg = inet_recvmsg,
1001#ifdef CONFIG_MMU
1002 .mmap = tcp_mmap,
1003#endif
1004 .sendpage = inet_sendpage,
1005 .splice_read = tcp_splice_read,
1006 .read_sock = tcp_read_sock,
1007 .sendmsg_locked = tcp_sendmsg_locked,
1008 .sendpage_locked = tcp_sendpage_locked,
1009 .peek_len = tcp_peek_len,
1010#ifdef CONFIG_COMPAT
1011 .compat_setsockopt = compat_sock_common_setsockopt,
1012 .compat_getsockopt = compat_sock_common_getsockopt,
1013 .compat_ioctl = inet_compat_ioctl,
1014#endif
1015 .set_rcvlowat = tcp_set_rcvlowat,
1016};
1017EXPORT_SYMBOL(inet_stream_ops);
1018
1019const struct proto_ops inet_dgram_ops = {
1020 .family = PF_INET,
1021 .owner = THIS_MODULE,
1022 .release = inet_release,
1023 .bind = inet_bind,
1024 .connect = inet_dgram_connect,
1025 .socketpair = sock_no_socketpair,
1026 .accept = sock_no_accept,
1027 .getname = inet_getname,
1028 .poll = udp_poll,
1029 .ioctl = inet_ioctl,
1030 .gettstamp = sock_gettstamp,
1031 .listen = sock_no_listen,
1032 .shutdown = inet_shutdown,
1033 .setsockopt = sock_common_setsockopt,
1034 .getsockopt = sock_common_getsockopt,
1035 .sendmsg = inet_sendmsg,
1036 .recvmsg = inet_recvmsg,
1037 .mmap = sock_no_mmap,
1038 .sendpage = inet_sendpage,
1039 .set_peek_off = sk_set_peek_off,
1040#ifdef CONFIG_COMPAT
1041 .compat_setsockopt = compat_sock_common_setsockopt,
1042 .compat_getsockopt = compat_sock_common_getsockopt,
1043 .compat_ioctl = inet_compat_ioctl,
1044#endif
1045};
1046EXPORT_SYMBOL(inet_dgram_ops);
1047
1048/*
1049 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1050 * udp_poll
1051 */
1052static const struct proto_ops inet_sockraw_ops = {
1053 .family = PF_INET,
1054 .owner = THIS_MODULE,
1055 .release = inet_release,
1056 .bind = inet_bind,
1057 .connect = inet_dgram_connect,
1058 .socketpair = sock_no_socketpair,
1059 .accept = sock_no_accept,
1060 .getname = inet_getname,
1061 .poll = datagram_poll,
1062 .ioctl = inet_ioctl,
1063 .gettstamp = sock_gettstamp,
1064 .listen = sock_no_listen,
1065 .shutdown = inet_shutdown,
1066 .setsockopt = sock_common_setsockopt,
1067 .getsockopt = sock_common_getsockopt,
1068 .sendmsg = inet_sendmsg,
1069 .recvmsg = inet_recvmsg,
1070 .mmap = sock_no_mmap,
1071 .sendpage = inet_sendpage,
1072#ifdef CONFIG_COMPAT
1073 .compat_setsockopt = compat_sock_common_setsockopt,
1074 .compat_getsockopt = compat_sock_common_getsockopt,
1075 .compat_ioctl = inet_compat_ioctl,
1076#endif
1077};
1078
1079static const struct net_proto_family inet_family_ops = {
1080 .family = PF_INET,
1081 .create = inet_create,
1082 .owner = THIS_MODULE,
1083};
1084
1085/* Upon startup we insert all the elements in inetsw_array[] into
1086 * the linked list inetsw.
1087 */
1088static struct inet_protosw inetsw_array[] =
1089{
1090 {
1091 .type = SOCK_STREAM,
1092 .protocol = IPPROTO_TCP,
1093 .prot = &tcp_prot,
1094 .ops = &inet_stream_ops,
1095 .flags = INET_PROTOSW_PERMANENT |
1096 INET_PROTOSW_ICSK,
1097 },
1098
1099 {
1100 .type = SOCK_DGRAM,
1101 .protocol = IPPROTO_UDP,
1102 .prot = &udp_prot,
1103 .ops = &inet_dgram_ops,
1104 .flags = INET_PROTOSW_PERMANENT,
1105 },
1106
1107 {
1108 .type = SOCK_DGRAM,
1109 .protocol = IPPROTO_ICMP,
1110 .prot = &ping_prot,
1111 .ops = &inet_sockraw_ops,
1112 .flags = INET_PROTOSW_REUSE,
1113 },
1114
1115 {
1116 .type = SOCK_RAW,
1117 .protocol = IPPROTO_IP, /* wild card */
1118 .prot = &raw_prot,
1119 .ops = &inet_sockraw_ops,
1120 .flags = INET_PROTOSW_REUSE,
1121 }
1122};
1123
1124#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1125
1126void inet_register_protosw(struct inet_protosw *p)
1127{
1128 struct list_head *lh;
1129 struct inet_protosw *answer;
1130 int protocol = p->protocol;
1131 struct list_head *last_perm;
1132
1133 spin_lock_bh(&inetsw_lock);
1134
1135 if (p->type >= SOCK_MAX)
1136 goto out_illegal;
1137
1138 /* If we are trying to override a permanent protocol, bail. */
1139 last_perm = &inetsw[p->type];
1140 list_for_each(lh, &inetsw[p->type]) {
1141 answer = list_entry(lh, struct inet_protosw, list);
1142 /* Check only the non-wild match. */
1143 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1144 break;
1145 if (protocol == answer->protocol)
1146 goto out_permanent;
1147 last_perm = lh;
1148 }
1149
1150 /* Add the new entry after the last permanent entry if any, so that
1151 * the new entry does not override a permanent entry when matched with
1152 * a wild-card protocol. But it is allowed to override any existing
1153 * non-permanent entry. This means that when we remove this entry, the
1154 * system automatically returns to the old behavior.
1155 */
1156 list_add_rcu(&p->list, last_perm);
1157out:
1158 spin_unlock_bh(&inetsw_lock);
1159
1160 return;
1161
1162out_permanent:
1163 pr_err("Attempt to override permanent protocol %d\n", protocol);
1164 goto out;
1165
1166out_illegal:
1167 pr_err("Ignoring attempt to register invalid socket type %d\n",
1168 p->type);
1169 goto out;
1170}
1171EXPORT_SYMBOL(inet_register_protosw);
1172
1173void inet_unregister_protosw(struct inet_protosw *p)
1174{
1175 if (INET_PROTOSW_PERMANENT & p->flags) {
1176 pr_err("Attempt to unregister permanent protocol %d\n",
1177 p->protocol);
1178 } else {
1179 spin_lock_bh(&inetsw_lock);
1180 list_del_rcu(&p->list);
1181 spin_unlock_bh(&inetsw_lock);
1182
1183 synchronize_net();
1184 }
1185}
1186EXPORT_SYMBOL(inet_unregister_protosw);
1187
1188static int inet_sk_reselect_saddr(struct sock *sk)
1189{
1190 struct inet_sock *inet = inet_sk(sk);
1191 __be32 old_saddr = inet->inet_saddr;
1192 __be32 daddr = inet->inet_daddr;
1193 struct flowi4 *fl4;
1194 struct rtable *rt;
1195 __be32 new_saddr;
1196 struct ip_options_rcu *inet_opt;
1197
1198 inet_opt = rcu_dereference_protected(inet->inet_opt,
1199 lockdep_sock_is_held(sk));
1200 if (inet_opt && inet_opt->opt.srr)
1201 daddr = inet_opt->opt.faddr;
1202
1203 /* Query new route. */
1204 fl4 = &inet->cork.fl.u.ip4;
1205 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1206 sk->sk_bound_dev_if, sk->sk_protocol,
1207 inet->inet_sport, inet->inet_dport, sk);
1208 if (IS_ERR(rt))
1209 return PTR_ERR(rt);
1210
1211 sk_setup_caps(sk, &rt->dst);
1212
1213 new_saddr = fl4->saddr;
1214
1215 if (new_saddr == old_saddr)
1216 return 0;
1217
1218 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1219 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1220 __func__, &old_saddr, &new_saddr);
1221 }
1222
1223 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1224
1225 /*
1226 * XXX The only one ugly spot where we need to
1227 * XXX really change the sockets identity after
1228 * XXX it has entered the hashes. -DaveM
1229 *
1230 * Besides that, it does not check for connection
1231 * uniqueness. Wait for troubles.
1232 */
1233 return __sk_prot_rehash(sk);
1234}
1235
1236int inet_sk_rebuild_header(struct sock *sk)
1237{
1238 struct inet_sock *inet = inet_sk(sk);
1239 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1240 __be32 daddr;
1241 struct ip_options_rcu *inet_opt;
1242 struct flowi4 *fl4;
1243 int err;
1244
1245 /* Route is OK, nothing to do. */
1246 if (rt)
1247 return 0;
1248
1249 /* Reroute. */
1250 rcu_read_lock();
1251 inet_opt = rcu_dereference(inet->inet_opt);
1252 daddr = inet->inet_daddr;
1253 if (inet_opt && inet_opt->opt.srr)
1254 daddr = inet_opt->opt.faddr;
1255 rcu_read_unlock();
1256 fl4 = &inet->cork.fl.u.ip4;
1257 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1258 inet->inet_dport, inet->inet_sport,
1259 sk->sk_protocol, RT_CONN_FLAGS(sk),
1260 sk->sk_bound_dev_if);
1261 if (!IS_ERR(rt)) {
1262 err = 0;
1263 sk_setup_caps(sk, &rt->dst);
1264 } else {
1265 err = PTR_ERR(rt);
1266
1267 /* Routing failed... */
1268 sk->sk_route_caps = 0;
1269 /*
1270 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1271 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1272 */
1273 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1274 sk->sk_state != TCP_SYN_SENT ||
1275 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1276 (err = inet_sk_reselect_saddr(sk)) != 0)
1277 sk->sk_err_soft = -err;
1278 }
1279
1280 return err;
1281}
1282EXPORT_SYMBOL(inet_sk_rebuild_header);
1283
1284void inet_sk_set_state(struct sock *sk, int state)
1285{
1286 trace_inet_sock_set_state(sk, sk->sk_state, state);
1287 sk->sk_state = state;
1288}
1289EXPORT_SYMBOL(inet_sk_set_state);
1290
1291void inet_sk_state_store(struct sock *sk, int newstate)
1292{
1293 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1294 smp_store_release(&sk->sk_state, newstate);
1295}
1296
1297struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1298 netdev_features_t features)
1299{
1300 bool udpfrag = false, fixedid = false, gso_partial, encap;
1301 struct sk_buff *segs = ERR_PTR(-EINVAL);
1302 const struct net_offload *ops;
1303 unsigned int offset = 0;
1304 struct iphdr *iph;
1305 int proto, tot_len;
1306 int nhoff;
1307 int ihl;
1308 int id;
1309
1310 skb_reset_network_header(skb);
1311 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1312 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1313 goto out;
1314
1315 iph = ip_hdr(skb);
1316 ihl = iph->ihl * 4;
1317 if (ihl < sizeof(*iph))
1318 goto out;
1319
1320 id = ntohs(iph->id);
1321 proto = iph->protocol;
1322
1323 /* Warning: after this point, iph might be no longer valid */
1324 if (unlikely(!pskb_may_pull(skb, ihl)))
1325 goto out;
1326 __skb_pull(skb, ihl);
1327
1328 encap = SKB_GSO_CB(skb)->encap_level > 0;
1329 if (encap)
1330 features &= skb->dev->hw_enc_features;
1331 SKB_GSO_CB(skb)->encap_level += ihl;
1332
1333 skb_reset_transport_header(skb);
1334
1335 segs = ERR_PTR(-EPROTONOSUPPORT);
1336
1337 if (!skb->encapsulation || encap) {
1338 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1339 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1340
1341 /* fixed ID is invalid if DF bit is not set */
1342 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1343 goto out;
1344 }
1345
1346 ops = rcu_dereference(inet_offloads[proto]);
1347 if (likely(ops && ops->callbacks.gso_segment))
1348 segs = ops->callbacks.gso_segment(skb, features);
1349
1350 if (IS_ERR_OR_NULL(segs))
1351 goto out;
1352
1353 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1354
1355 skb = segs;
1356 do {
1357 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1358 if (udpfrag) {
1359 iph->frag_off = htons(offset >> 3);
1360 if (skb->next)
1361 iph->frag_off |= htons(IP_MF);
1362 offset += skb->len - nhoff - ihl;
1363 tot_len = skb->len - nhoff;
1364 } else if (skb_is_gso(skb)) {
1365 if (!fixedid) {
1366 iph->id = htons(id);
1367 id += skb_shinfo(skb)->gso_segs;
1368 }
1369
1370 if (gso_partial)
1371 tot_len = skb_shinfo(skb)->gso_size +
1372 SKB_GSO_CB(skb)->data_offset +
1373 skb->head - (unsigned char *)iph;
1374 else
1375 tot_len = skb->len - nhoff;
1376 } else {
1377 if (!fixedid)
1378 iph->id = htons(id++);
1379 tot_len = skb->len - nhoff;
1380 }
1381 iph->tot_len = htons(tot_len);
1382 ip_send_check(iph);
1383 if (encap)
1384 skb_reset_inner_headers(skb);
1385 skb->network_header = (u8 *)iph - skb->head;
1386 skb_reset_mac_len(skb);
1387 } while ((skb = skb->next));
1388
1389out:
1390 return segs;
1391}
1392EXPORT_SYMBOL(inet_gso_segment);
1393
1394static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1395 netdev_features_t features)
1396{
1397 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1398 return ERR_PTR(-EINVAL);
1399
1400 return inet_gso_segment(skb, features);
1401}
1402
1403INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *,
1404 struct sk_buff *));
1405INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *,
1406 struct sk_buff *));
1407struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1408{
1409 const struct net_offload *ops;
1410 struct sk_buff *pp = NULL;
1411 const struct iphdr *iph;
1412 struct sk_buff *p;
1413 unsigned int hlen;
1414 unsigned int off;
1415 unsigned int id;
1416 int flush = 1;
1417 int proto;
1418
1419 off = skb_gro_offset(skb);
1420 hlen = off + sizeof(*iph);
1421 iph = skb_gro_header_fast(skb, off);
1422 if (skb_gro_header_hard(skb, hlen)) {
1423 iph = skb_gro_header_slow(skb, hlen, off);
1424 if (unlikely(!iph))
1425 goto out;
1426 }
1427
1428 proto = iph->protocol;
1429
1430 rcu_read_lock();
1431 ops = rcu_dereference(inet_offloads[proto]);
1432 if (!ops || !ops->callbacks.gro_receive)
1433 goto out_unlock;
1434
1435 if (*(u8 *)iph != 0x45)
1436 goto out_unlock;
1437
1438 if (ip_is_fragment(iph))
1439 goto out_unlock;
1440
1441 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1442 goto out_unlock;
1443
1444 id = ntohl(*(__be32 *)&iph->id);
1445 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1446 id >>= 16;
1447
1448 list_for_each_entry(p, head, list) {
1449 struct iphdr *iph2;
1450 u16 flush_id;
1451
1452 if (!NAPI_GRO_CB(p)->same_flow)
1453 continue;
1454
1455 iph2 = (struct iphdr *)(p->data + off);
1456 /* The above works because, with the exception of the top
1457 * (inner most) layer, we only aggregate pkts with the same
1458 * hdr length so all the hdrs we'll need to verify will start
1459 * at the same offset.
1460 */
1461 if ((iph->protocol ^ iph2->protocol) |
1462 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1463 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1464 NAPI_GRO_CB(p)->same_flow = 0;
1465 continue;
1466 }
1467
1468 /* All fields must match except length and checksum. */
1469 NAPI_GRO_CB(p)->flush |=
1470 (iph->ttl ^ iph2->ttl) |
1471 (iph->tos ^ iph2->tos) |
1472 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1473
1474 NAPI_GRO_CB(p)->flush |= flush;
1475
1476 /* We need to store of the IP ID check to be included later
1477 * when we can verify that this packet does in fact belong
1478 * to a given flow.
1479 */
1480 flush_id = (u16)(id - ntohs(iph2->id));
1481
1482 /* This bit of code makes it much easier for us to identify
1483 * the cases where we are doing atomic vs non-atomic IP ID
1484 * checks. Specifically an atomic check can return IP ID
1485 * values 0 - 0xFFFF, while a non-atomic check can only
1486 * return 0 or 0xFFFF.
1487 */
1488 if (!NAPI_GRO_CB(p)->is_atomic ||
1489 !(iph->frag_off & htons(IP_DF))) {
1490 flush_id ^= NAPI_GRO_CB(p)->count;
1491 flush_id = flush_id ? 0xFFFF : 0;
1492 }
1493
1494 /* If the previous IP ID value was based on an atomic
1495 * datagram we can overwrite the value and ignore it.
1496 */
1497 if (NAPI_GRO_CB(skb)->is_atomic)
1498 NAPI_GRO_CB(p)->flush_id = flush_id;
1499 else
1500 NAPI_GRO_CB(p)->flush_id |= flush_id;
1501 }
1502
1503 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1504 NAPI_GRO_CB(skb)->flush |= flush;
1505 skb_set_network_header(skb, off);
1506 /* The above will be needed by the transport layer if there is one
1507 * immediately following this IP hdr.
1508 */
1509
1510 /* Note : No need to call skb_gro_postpull_rcsum() here,
1511 * as we already checked checksum over ipv4 header was 0
1512 */
1513 skb_gro_pull(skb, sizeof(*iph));
1514 skb_set_transport_header(skb, skb_gro_offset(skb));
1515
1516 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1517 ops->callbacks.gro_receive, head, skb);
1518
1519out_unlock:
1520 rcu_read_unlock();
1521
1522out:
1523 skb_gro_flush_final(skb, pp, flush);
1524
1525 return pp;
1526}
1527EXPORT_SYMBOL(inet_gro_receive);
1528
1529static struct sk_buff *ipip_gro_receive(struct list_head *head,
1530 struct sk_buff *skb)
1531{
1532 if (NAPI_GRO_CB(skb)->encap_mark) {
1533 NAPI_GRO_CB(skb)->flush = 1;
1534 return NULL;
1535 }
1536
1537 NAPI_GRO_CB(skb)->encap_mark = 1;
1538
1539 return inet_gro_receive(head, skb);
1540}
1541
1542#define SECONDS_PER_DAY 86400
1543
1544/* inet_current_timestamp - Return IP network timestamp
1545 *
1546 * Return milliseconds since midnight in network byte order.
1547 */
1548__be32 inet_current_timestamp(void)
1549{
1550 u32 secs;
1551 u32 msecs;
1552 struct timespec64 ts;
1553
1554 ktime_get_real_ts64(&ts);
1555
1556 /* Get secs since midnight. */
1557 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1558 /* Convert to msecs. */
1559 msecs = secs * MSEC_PER_SEC;
1560 /* Convert nsec to msec. */
1561 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1562
1563 /* Convert to network byte order. */
1564 return htonl(msecs);
1565}
1566EXPORT_SYMBOL(inet_current_timestamp);
1567
1568int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1569{
1570 if (sk->sk_family == AF_INET)
1571 return ip_recv_error(sk, msg, len, addr_len);
1572#if IS_ENABLED(CONFIG_IPV6)
1573 if (sk->sk_family == AF_INET6)
1574 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1575#endif
1576 return -EINVAL;
1577}
1578
1579INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *, int));
1580INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int));
1581int inet_gro_complete(struct sk_buff *skb, int nhoff)
1582{
1583 __be16 newlen = htons(skb->len - nhoff);
1584 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1585 const struct net_offload *ops;
1586 int proto = iph->protocol;
1587 int err = -ENOSYS;
1588
1589 if (skb->encapsulation) {
1590 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1591 skb_set_inner_network_header(skb, nhoff);
1592 }
1593
1594 csum_replace2(&iph->check, iph->tot_len, newlen);
1595 iph->tot_len = newlen;
1596
1597 rcu_read_lock();
1598 ops = rcu_dereference(inet_offloads[proto]);
1599 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1600 goto out_unlock;
1601
1602 /* Only need to add sizeof(*iph) to get to the next hdr below
1603 * because any hdr with option will have been flushed in
1604 * inet_gro_receive().
1605 */
1606 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1607 tcp4_gro_complete, udp4_gro_complete,
1608 skb, nhoff + sizeof(*iph));
1609
1610out_unlock:
1611 rcu_read_unlock();
1612
1613 return err;
1614}
1615EXPORT_SYMBOL(inet_gro_complete);
1616
1617static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1618{
1619 skb->encapsulation = 1;
1620 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1621 return inet_gro_complete(skb, nhoff);
1622}
1623
1624int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1625 unsigned short type, unsigned char protocol,
1626 struct net *net)
1627{
1628 struct socket *sock;
1629 int rc = sock_create_kern(net, family, type, protocol, &sock);
1630
1631 if (rc == 0) {
1632 *sk = sock->sk;
1633 (*sk)->sk_allocation = GFP_ATOMIC;
1634 /*
1635 * Unhash it so that IP input processing does not even see it,
1636 * we do not wish this socket to see incoming packets.
1637 */
1638 (*sk)->sk_prot->unhash(*sk);
1639 }
1640 return rc;
1641}
1642EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1643
1644u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1645{
1646 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1647}
1648EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1649
1650unsigned long snmp_fold_field(void __percpu *mib, int offt)
1651{
1652 unsigned long res = 0;
1653 int i;
1654
1655 for_each_possible_cpu(i)
1656 res += snmp_get_cpu_field(mib, i, offt);
1657 return res;
1658}
1659EXPORT_SYMBOL_GPL(snmp_fold_field);
1660
1661#if BITS_PER_LONG==32
1662
1663u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1664 size_t syncp_offset)
1665{
1666 void *bhptr;
1667 struct u64_stats_sync *syncp;
1668 u64 v;
1669 unsigned int start;
1670
1671 bhptr = per_cpu_ptr(mib, cpu);
1672 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1673 do {
1674 start = u64_stats_fetch_begin_irq(syncp);
1675 v = *(((u64 *)bhptr) + offt);
1676 } while (u64_stats_fetch_retry_irq(syncp, start));
1677
1678 return v;
1679}
1680EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1681
1682u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1683{
1684 u64 res = 0;
1685 int cpu;
1686
1687 for_each_possible_cpu(cpu) {
1688 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1689 }
1690 return res;
1691}
1692EXPORT_SYMBOL_GPL(snmp_fold_field64);
1693#endif
1694
1695#ifdef CONFIG_IP_MULTICAST
1696static const struct net_protocol igmp_protocol = {
1697 .handler = igmp_rcv,
1698 .netns_ok = 1,
1699};
1700#endif
1701
1702/* thinking of making this const? Don't.
1703 * early_demux can change based on sysctl.
1704 */
1705static struct net_protocol tcp_protocol = {
1706 .early_demux = tcp_v4_early_demux,
1707 .early_demux_handler = tcp_v4_early_demux,
1708 .handler = tcp_v4_rcv,
1709 .err_handler = tcp_v4_err,
1710 .no_policy = 1,
1711 .netns_ok = 1,
1712 .icmp_strict_tag_validation = 1,
1713};
1714
1715/* thinking of making this const? Don't.
1716 * early_demux can change based on sysctl.
1717 */
1718static struct net_protocol udp_protocol = {
1719 .early_demux = udp_v4_early_demux,
1720 .early_demux_handler = udp_v4_early_demux,
1721 .handler = udp_rcv,
1722 .err_handler = udp_err,
1723 .no_policy = 1,
1724 .netns_ok = 1,
1725};
1726
1727static const struct net_protocol icmp_protocol = {
1728 .handler = icmp_rcv,
1729 .err_handler = icmp_err,
1730 .no_policy = 1,
1731 .netns_ok = 1,
1732};
1733
1734static __net_init int ipv4_mib_init_net(struct net *net)
1735{
1736 int i;
1737
1738 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1739 if (!net->mib.tcp_statistics)
1740 goto err_tcp_mib;
1741 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1742 if (!net->mib.ip_statistics)
1743 goto err_ip_mib;
1744
1745 for_each_possible_cpu(i) {
1746 struct ipstats_mib *af_inet_stats;
1747 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1748 u64_stats_init(&af_inet_stats->syncp);
1749 }
1750
1751 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1752 if (!net->mib.net_statistics)
1753 goto err_net_mib;
1754 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1755 if (!net->mib.udp_statistics)
1756 goto err_udp_mib;
1757 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1758 if (!net->mib.udplite_statistics)
1759 goto err_udplite_mib;
1760 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1761 if (!net->mib.icmp_statistics)
1762 goto err_icmp_mib;
1763 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1764 GFP_KERNEL);
1765 if (!net->mib.icmpmsg_statistics)
1766 goto err_icmpmsg_mib;
1767
1768 tcp_mib_init(net);
1769 return 0;
1770
1771err_icmpmsg_mib:
1772 free_percpu(net->mib.icmp_statistics);
1773err_icmp_mib:
1774 free_percpu(net->mib.udplite_statistics);
1775err_udplite_mib:
1776 free_percpu(net->mib.udp_statistics);
1777err_udp_mib:
1778 free_percpu(net->mib.net_statistics);
1779err_net_mib:
1780 free_percpu(net->mib.ip_statistics);
1781err_ip_mib:
1782 free_percpu(net->mib.tcp_statistics);
1783err_tcp_mib:
1784 return -ENOMEM;
1785}
1786
1787static __net_exit void ipv4_mib_exit_net(struct net *net)
1788{
1789 kfree(net->mib.icmpmsg_statistics);
1790 free_percpu(net->mib.icmp_statistics);
1791 free_percpu(net->mib.udplite_statistics);
1792 free_percpu(net->mib.udp_statistics);
1793 free_percpu(net->mib.net_statistics);
1794 free_percpu(net->mib.ip_statistics);
1795 free_percpu(net->mib.tcp_statistics);
1796}
1797
1798static __net_initdata struct pernet_operations ipv4_mib_ops = {
1799 .init = ipv4_mib_init_net,
1800 .exit = ipv4_mib_exit_net,
1801};
1802
1803static int __init init_ipv4_mibs(void)
1804{
1805 return register_pernet_subsys(&ipv4_mib_ops);
1806}
1807
1808static __net_init int inet_init_net(struct net *net)
1809{
1810 /*
1811 * Set defaults for local port range
1812 */
1813 seqlock_init(&net->ipv4.ip_local_ports.lock);
1814 net->ipv4.ip_local_ports.range[0] = 32768;
1815 net->ipv4.ip_local_ports.range[1] = 60999;
1816
1817 seqlock_init(&net->ipv4.ping_group_range.lock);
1818 /*
1819 * Sane defaults - nobody may create ping sockets.
1820 * Boot scripts should set this to distro-specific group.
1821 */
1822 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1823 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1824
1825 /* Default values for sysctl-controlled parameters.
1826 * We set them here, in case sysctl is not compiled.
1827 */
1828 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1829 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1830 net->ipv4.sysctl_ip_dynaddr = 0;
1831 net->ipv4.sysctl_ip_early_demux = 1;
1832 net->ipv4.sysctl_udp_early_demux = 1;
1833 net->ipv4.sysctl_tcp_early_demux = 1;
1834#ifdef CONFIG_SYSCTL
1835 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1836#endif
1837
1838 /* Some igmp sysctl, whose values are always used */
1839 net->ipv4.sysctl_igmp_max_memberships = 20;
1840 net->ipv4.sysctl_igmp_max_msf = 10;
1841 /* IGMP reports for link-local multicast groups are enabled by default */
1842 net->ipv4.sysctl_igmp_llm_reports = 1;
1843 net->ipv4.sysctl_igmp_qrv = 2;
1844
1845 return 0;
1846}
1847
1848static __net_initdata struct pernet_operations af_inet_ops = {
1849 .init = inet_init_net,
1850};
1851
1852static int __init init_inet_pernet_ops(void)
1853{
1854 return register_pernet_subsys(&af_inet_ops);
1855}
1856
1857static int ipv4_proc_init(void);
1858
1859/*
1860 * IP protocol layer initialiser
1861 */
1862
1863static struct packet_offload ip_packet_offload __read_mostly = {
1864 .type = cpu_to_be16(ETH_P_IP),
1865 .callbacks = {
1866 .gso_segment = inet_gso_segment,
1867 .gro_receive = inet_gro_receive,
1868 .gro_complete = inet_gro_complete,
1869 },
1870};
1871
1872static const struct net_offload ipip_offload = {
1873 .callbacks = {
1874 .gso_segment = ipip_gso_segment,
1875 .gro_receive = ipip_gro_receive,
1876 .gro_complete = ipip_gro_complete,
1877 },
1878};
1879
1880static int __init ipip_offload_init(void)
1881{
1882 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1883}
1884
1885static int __init ipv4_offload_init(void)
1886{
1887 /*
1888 * Add offloads
1889 */
1890 if (udpv4_offload_init() < 0)
1891 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1892 if (tcpv4_offload_init() < 0)
1893 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1894 if (ipip_offload_init() < 0)
1895 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1896
1897 dev_add_offload(&ip_packet_offload);
1898 return 0;
1899}
1900
1901fs_initcall(ipv4_offload_init);
1902
1903static struct packet_type ip_packet_type __read_mostly = {
1904 .type = cpu_to_be16(ETH_P_IP),
1905 .func = ip_rcv,
1906 .list_func = ip_list_rcv,
1907};
1908
1909static int __init inet_init(void)
1910{
1911 struct inet_protosw *q;
1912 struct list_head *r;
1913 int rc = -EINVAL;
1914
1915 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1916
1917 rc = proto_register(&tcp_prot, 1);
1918 if (rc)
1919 goto out;
1920
1921 rc = proto_register(&udp_prot, 1);
1922 if (rc)
1923 goto out_unregister_tcp_proto;
1924
1925 rc = proto_register(&raw_prot, 1);
1926 if (rc)
1927 goto out_unregister_udp_proto;
1928
1929 rc = proto_register(&ping_prot, 1);
1930 if (rc)
1931 goto out_unregister_raw_proto;
1932
1933 /*
1934 * Tell SOCKET that we are alive...
1935 */
1936
1937 (void)sock_register(&inet_family_ops);
1938
1939#ifdef CONFIG_SYSCTL
1940 ip_static_sysctl_init();
1941#endif
1942
1943 /*
1944 * Add all the base protocols.
1945 */
1946
1947 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1948 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1949 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1950 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1951 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1952 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1953#ifdef CONFIG_IP_MULTICAST
1954 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1955 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1956#endif
1957
1958 /* Register the socket-side information for inet_create. */
1959 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1960 INIT_LIST_HEAD(r);
1961
1962 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1963 inet_register_protosw(q);
1964
1965 /*
1966 * Set the ARP module up
1967 */
1968
1969 arp_init();
1970
1971 /*
1972 * Set the IP module up
1973 */
1974
1975 ip_init();
1976
1977 /* Setup TCP slab cache for open requests. */
1978 tcp_init();
1979
1980 /* Setup UDP memory threshold */
1981 udp_init();
1982
1983 /* Add UDP-Lite (RFC 3828) */
1984 udplite4_register();
1985
1986 raw_init();
1987
1988 ping_init();
1989
1990 /*
1991 * Set the ICMP layer up
1992 */
1993
1994 if (icmp_init() < 0)
1995 panic("Failed to create the ICMP control socket.\n");
1996
1997 /*
1998 * Initialise the multicast router
1999 */
2000#if defined(CONFIG_IP_MROUTE)
2001 if (ip_mr_init())
2002 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2003#endif
2004
2005 if (init_inet_pernet_ops())
2006 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2007 /*
2008 * Initialise per-cpu ipv4 mibs
2009 */
2010
2011 if (init_ipv4_mibs())
2012 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
2013
2014 ipv4_proc_init();
2015
2016 ipfrag_init();
2017
2018 dev_add_pack(&ip_packet_type);
2019
2020 ip_tunnel_core_init();
2021
2022 rc = 0;
2023out:
2024 return rc;
2025out_unregister_raw_proto:
2026 proto_unregister(&raw_prot);
2027out_unregister_udp_proto:
2028 proto_unregister(&udp_prot);
2029out_unregister_tcp_proto:
2030 proto_unregister(&tcp_prot);
2031 goto out;
2032}
2033
2034fs_initcall(inet_init);
2035
2036/* ------------------------------------------------------------------------ */
2037
2038#ifdef CONFIG_PROC_FS
2039static int __init ipv4_proc_init(void)
2040{
2041 int rc = 0;
2042
2043 if (raw_proc_init())
2044 goto out_raw;
2045 if (tcp4_proc_init())
2046 goto out_tcp;
2047 if (udp4_proc_init())
2048 goto out_udp;
2049 if (ping_proc_init())
2050 goto out_ping;
2051 if (ip_misc_proc_init())
2052 goto out_misc;
2053out:
2054 return rc;
2055out_misc:
2056 ping_proc_exit();
2057out_ping:
2058 udp4_proc_exit();
2059out_udp:
2060 tcp4_proc_exit();
2061out_tcp:
2062 raw_proc_exit();
2063out_raw:
2064 rc = -ENOMEM;
2065 goto out;
2066}
2067
2068#else /* CONFIG_PROC_FS */
2069static int __init ipv4_proc_init(void)
2070{
2071 return 0;
2072}
2073#endif /* CONFIG_PROC_FS */
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
125/* The inetsw table contains everything that inet_create needs to
126 * build a new socket.
127 */
128static struct list_head inetsw[SOCK_MAX];
129static DEFINE_SPINLOCK(inetsw_lock);
130
131/* New destruction routine */
132
133void inet_sock_destruct(struct sock *sk)
134{
135 struct inet_sock *inet = inet_sk(sk);
136
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
139
140 sk_mem_reclaim(sk);
141
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 pr_err("Attempt to release TCP socket in state %d %p\n",
144 sk->sk_state, sk);
145 return;
146 }
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 pr_err("Attempt to release alive inet socket %p\n", sk);
149 return;
150 }
151
152 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 WARN_ON(sk->sk_wmem_queued);
155 WARN_ON(sk->sk_forward_alloc);
156
157 kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159 dst_release(sk->sk_rx_dst);
160 sk_refcnt_debug_dec(sk);
161}
162EXPORT_SYMBOL(inet_sock_destruct);
163
164/*
165 * The routines beyond this point handle the behaviour of an AF_INET
166 * socket object. Mostly it punts to the subprotocols of IP to do
167 * the work.
168 */
169
170/*
171 * Automatically bind an unbound socket.
172 */
173
174static int inet_autobind(struct sock *sk)
175{
176 struct inet_sock *inet;
177 /* We may need to bind the socket. */
178 lock_sock(sk);
179 inet = inet_sk(sk);
180 if (!inet->inet_num) {
181 if (sk->sk_prot->get_port(sk, 0)) {
182 release_sock(sk);
183 return -EAGAIN;
184 }
185 inet->inet_sport = htons(inet->inet_num);
186 }
187 release_sock(sk);
188 return 0;
189}
190
191/*
192 * Move a socket into listening state.
193 */
194int inet_listen(struct socket *sock, int backlog)
195{
196 struct sock *sk = sock->sk;
197 unsigned char old_state;
198 int err;
199
200 lock_sock(sk);
201
202 err = -EINVAL;
203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 goto out;
205
206 old_state = sk->sk_state;
207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 goto out;
209
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
212 */
213 if (old_state != TCP_LISTEN) {
214 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
215 * Note that only TCP sockets (SOCK_STREAM) will reach here.
216 * Also fastopen backlog may already been set via the option
217 * because the socket was in TCP_LISTEN state previously but
218 * was shutdown() rather than close().
219 */
220 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
221 (sysctl_tcp_fastopen & TFO_SERVER_ENABLE) &&
222 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
223 fastopen_queue_tune(sk, backlog);
224 tcp_fastopen_init_key_once(true);
225 }
226
227 err = inet_csk_listen_start(sk, backlog);
228 if (err)
229 goto out;
230 }
231 sk->sk_max_ack_backlog = backlog;
232 err = 0;
233
234out:
235 release_sock(sk);
236 return err;
237}
238EXPORT_SYMBOL(inet_listen);
239
240/*
241 * Create an inet socket.
242 */
243
244static int inet_create(struct net *net, struct socket *sock, int protocol,
245 int kern)
246{
247 struct sock *sk;
248 struct inet_protosw *answer;
249 struct inet_sock *inet;
250 struct proto *answer_prot;
251 unsigned char answer_flags;
252 int try_loading_module = 0;
253 int err;
254
255 if (protocol < 0 || protocol >= IPPROTO_MAX)
256 return -EINVAL;
257
258 sock->state = SS_UNCONNECTED;
259
260 /* Look for the requested type/protocol pair. */
261lookup_protocol:
262 err = -ESOCKTNOSUPPORT;
263 rcu_read_lock();
264 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
265
266 err = 0;
267 /* Check the non-wild match. */
268 if (protocol == answer->protocol) {
269 if (protocol != IPPROTO_IP)
270 break;
271 } else {
272 /* Check for the two wild cases. */
273 if (IPPROTO_IP == protocol) {
274 protocol = answer->protocol;
275 break;
276 }
277 if (IPPROTO_IP == answer->protocol)
278 break;
279 }
280 err = -EPROTONOSUPPORT;
281 }
282
283 if (unlikely(err)) {
284 if (try_loading_module < 2) {
285 rcu_read_unlock();
286 /*
287 * Be more specific, e.g. net-pf-2-proto-132-type-1
288 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
289 */
290 if (++try_loading_module == 1)
291 request_module("net-pf-%d-proto-%d-type-%d",
292 PF_INET, protocol, sock->type);
293 /*
294 * Fall back to generic, e.g. net-pf-2-proto-132
295 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
296 */
297 else
298 request_module("net-pf-%d-proto-%d",
299 PF_INET, protocol);
300 goto lookup_protocol;
301 } else
302 goto out_rcu_unlock;
303 }
304
305 err = -EPERM;
306 if (sock->type == SOCK_RAW && !kern &&
307 !ns_capable(net->user_ns, CAP_NET_RAW))
308 goto out_rcu_unlock;
309
310 sock->ops = answer->ops;
311 answer_prot = answer->prot;
312 answer_flags = answer->flags;
313 rcu_read_unlock();
314
315 WARN_ON(!answer_prot->slab);
316
317 err = -ENOBUFS;
318 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
319 if (!sk)
320 goto out;
321
322 err = 0;
323 if (INET_PROTOSW_REUSE & answer_flags)
324 sk->sk_reuse = SK_CAN_REUSE;
325
326 inet = inet_sk(sk);
327 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
328
329 inet->nodefrag = 0;
330
331 if (SOCK_RAW == sock->type) {
332 inet->inet_num = protocol;
333 if (IPPROTO_RAW == protocol)
334 inet->hdrincl = 1;
335 }
336
337 if (net->ipv4.sysctl_ip_no_pmtu_disc)
338 inet->pmtudisc = IP_PMTUDISC_DONT;
339 else
340 inet->pmtudisc = IP_PMTUDISC_WANT;
341
342 inet->inet_id = 0;
343
344 sock_init_data(sock, sk);
345
346 sk->sk_destruct = inet_sock_destruct;
347 sk->sk_protocol = protocol;
348 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
349
350 inet->uc_ttl = -1;
351 inet->mc_loop = 1;
352 inet->mc_ttl = 1;
353 inet->mc_all = 1;
354 inet->mc_index = 0;
355 inet->mc_list = NULL;
356 inet->rcv_tos = 0;
357
358 sk_refcnt_debug_inc(sk);
359
360 if (inet->inet_num) {
361 /* It assumes that any protocol which allows
362 * the user to assign a number at socket
363 * creation time automatically
364 * shares.
365 */
366 inet->inet_sport = htons(inet->inet_num);
367 /* Add to protocol hash chains. */
368 err = sk->sk_prot->hash(sk);
369 if (err) {
370 sk_common_release(sk);
371 goto out;
372 }
373 }
374
375 if (sk->sk_prot->init) {
376 err = sk->sk_prot->init(sk);
377 if (err) {
378 sk_common_release(sk);
379 goto out;
380 }
381 }
382
383 if (!kern) {
384 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
385 if (err) {
386 sk_common_release(sk);
387 goto out;
388 }
389 }
390out:
391 return err;
392out_rcu_unlock:
393 rcu_read_unlock();
394 goto out;
395}
396
397
398/*
399 * The peer socket should always be NULL (or else). When we call this
400 * function we are destroying the object and from then on nobody
401 * should refer to it.
402 */
403int inet_release(struct socket *sock)
404{
405 struct sock *sk = sock->sk;
406
407 if (sk) {
408 long timeout;
409
410 /* Applications forget to leave groups before exiting */
411 ip_mc_drop_socket(sk);
412
413 /* If linger is set, we don't return until the close
414 * is complete. Otherwise we return immediately. The
415 * actually closing is done the same either way.
416 *
417 * If the close is due to the process exiting, we never
418 * linger..
419 */
420 timeout = 0;
421 if (sock_flag(sk, SOCK_LINGER) &&
422 !(current->flags & PF_EXITING))
423 timeout = sk->sk_lingertime;
424 sock->sk = NULL;
425 sk->sk_prot->close(sk, timeout);
426 }
427 return 0;
428}
429EXPORT_SYMBOL(inet_release);
430
431int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
432{
433 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
434 struct sock *sk = sock->sk;
435 struct inet_sock *inet = inet_sk(sk);
436 struct net *net = sock_net(sk);
437 unsigned short snum;
438 int chk_addr_ret;
439 u32 tb_id = RT_TABLE_LOCAL;
440 int err;
441
442 /* If the socket has its own bind function then use it. (RAW) */
443 if (sk->sk_prot->bind) {
444 err = sk->sk_prot->bind(sk, uaddr, addr_len);
445 goto out;
446 }
447 err = -EINVAL;
448 if (addr_len < sizeof(struct sockaddr_in))
449 goto out;
450
451 if (addr->sin_family != AF_INET) {
452 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
453 * only if s_addr is INADDR_ANY.
454 */
455 err = -EAFNOSUPPORT;
456 if (addr->sin_family != AF_UNSPEC ||
457 addr->sin_addr.s_addr != htonl(INADDR_ANY))
458 goto out;
459 }
460
461 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
462 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
463
464 /* Not specified by any standard per-se, however it breaks too
465 * many applications when removed. It is unfortunate since
466 * allowing applications to make a non-local bind solves
467 * several problems with systems using dynamic addressing.
468 * (ie. your servers still start up even if your ISDN link
469 * is temporarily down)
470 */
471 err = -EADDRNOTAVAIL;
472 if (!net->ipv4.sysctl_ip_nonlocal_bind &&
473 !(inet->freebind || inet->transparent) &&
474 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
475 chk_addr_ret != RTN_LOCAL &&
476 chk_addr_ret != RTN_MULTICAST &&
477 chk_addr_ret != RTN_BROADCAST)
478 goto out;
479
480 snum = ntohs(addr->sin_port);
481 err = -EACCES;
482 if (snum && snum < PROT_SOCK &&
483 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
484 goto out;
485
486 /* We keep a pair of addresses. rcv_saddr is the one
487 * used by hash lookups, and saddr is used for transmit.
488 *
489 * In the BSD API these are the same except where it
490 * would be illegal to use them (multicast/broadcast) in
491 * which case the sending device address is used.
492 */
493 lock_sock(sk);
494
495 /* Check these errors (active socket, double bind). */
496 err = -EINVAL;
497 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
498 goto out_release_sock;
499
500 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
501 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
502 inet->inet_saddr = 0; /* Use device */
503
504 /* Make sure we are allowed to bind here. */
505 if ((snum || !inet->bind_address_no_port) &&
506 sk->sk_prot->get_port(sk, snum)) {
507 inet->inet_saddr = inet->inet_rcv_saddr = 0;
508 err = -EADDRINUSE;
509 goto out_release_sock;
510 }
511
512 if (inet->inet_rcv_saddr)
513 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
514 if (snum)
515 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
516 inet->inet_sport = htons(inet->inet_num);
517 inet->inet_daddr = 0;
518 inet->inet_dport = 0;
519 sk_dst_reset(sk);
520 err = 0;
521out_release_sock:
522 release_sock(sk);
523out:
524 return err;
525}
526EXPORT_SYMBOL(inet_bind);
527
528int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
529 int addr_len, int flags)
530{
531 struct sock *sk = sock->sk;
532
533 if (addr_len < sizeof(uaddr->sa_family))
534 return -EINVAL;
535 if (uaddr->sa_family == AF_UNSPEC)
536 return sk->sk_prot->disconnect(sk, flags);
537
538 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
539 return -EAGAIN;
540 return sk->sk_prot->connect(sk, uaddr, addr_len);
541}
542EXPORT_SYMBOL(inet_dgram_connect);
543
544static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
545{
546 DEFINE_WAIT_FUNC(wait, woken_wake_function);
547
548 add_wait_queue(sk_sleep(sk), &wait);
549 sk->sk_write_pending += writebias;
550
551 /* Basic assumption: if someone sets sk->sk_err, he _must_
552 * change state of the socket from TCP_SYN_*.
553 * Connect() does not allow to get error notifications
554 * without closing the socket.
555 */
556 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
557 release_sock(sk);
558 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
559 lock_sock(sk);
560 if (signal_pending(current) || !timeo)
561 break;
562 }
563 remove_wait_queue(sk_sleep(sk), &wait);
564 sk->sk_write_pending -= writebias;
565 return timeo;
566}
567
568/*
569 * Connect to a remote host. There is regrettably still a little
570 * TCP 'magic' in here.
571 */
572int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
573 int addr_len, int flags)
574{
575 struct sock *sk = sock->sk;
576 int err;
577 long timeo;
578
579 if (addr_len < sizeof(uaddr->sa_family))
580 return -EINVAL;
581
582 if (uaddr->sa_family == AF_UNSPEC) {
583 err = sk->sk_prot->disconnect(sk, flags);
584 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
585 goto out;
586 }
587
588 switch (sock->state) {
589 default:
590 err = -EINVAL;
591 goto out;
592 case SS_CONNECTED:
593 err = -EISCONN;
594 goto out;
595 case SS_CONNECTING:
596 err = -EALREADY;
597 /* Fall out of switch with err, set for this state */
598 break;
599 case SS_UNCONNECTED:
600 err = -EISCONN;
601 if (sk->sk_state != TCP_CLOSE)
602 goto out;
603
604 err = sk->sk_prot->connect(sk, uaddr, addr_len);
605 if (err < 0)
606 goto out;
607
608 sock->state = SS_CONNECTING;
609
610 /* Just entered SS_CONNECTING state; the only
611 * difference is that return value in non-blocking
612 * case is EINPROGRESS, rather than EALREADY.
613 */
614 err = -EINPROGRESS;
615 break;
616 }
617
618 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
619
620 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
621 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
622 tcp_sk(sk)->fastopen_req &&
623 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
624
625 /* Error code is set above */
626 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
627 goto out;
628
629 err = sock_intr_errno(timeo);
630 if (signal_pending(current))
631 goto out;
632 }
633
634 /* Connection was closed by RST, timeout, ICMP error
635 * or another process disconnected us.
636 */
637 if (sk->sk_state == TCP_CLOSE)
638 goto sock_error;
639
640 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
641 * and error was received after socket entered established state.
642 * Hence, it is handled normally after connect() return successfully.
643 */
644
645 sock->state = SS_CONNECTED;
646 err = 0;
647out:
648 return err;
649
650sock_error:
651 err = sock_error(sk) ? : -ECONNABORTED;
652 sock->state = SS_UNCONNECTED;
653 if (sk->sk_prot->disconnect(sk, flags))
654 sock->state = SS_DISCONNECTING;
655 goto out;
656}
657EXPORT_SYMBOL(__inet_stream_connect);
658
659int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
660 int addr_len, int flags)
661{
662 int err;
663
664 lock_sock(sock->sk);
665 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
666 release_sock(sock->sk);
667 return err;
668}
669EXPORT_SYMBOL(inet_stream_connect);
670
671/*
672 * Accept a pending connection. The TCP layer now gives BSD semantics.
673 */
674
675int inet_accept(struct socket *sock, struct socket *newsock, int flags)
676{
677 struct sock *sk1 = sock->sk;
678 int err = -EINVAL;
679 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
680
681 if (!sk2)
682 goto do_err;
683
684 lock_sock(sk2);
685
686 sock_rps_record_flow(sk2);
687 WARN_ON(!((1 << sk2->sk_state) &
688 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
689 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
690
691 sock_graft(sk2, newsock);
692
693 newsock->state = SS_CONNECTED;
694 err = 0;
695 release_sock(sk2);
696do_err:
697 return err;
698}
699EXPORT_SYMBOL(inet_accept);
700
701
702/*
703 * This does both peername and sockname.
704 */
705int inet_getname(struct socket *sock, struct sockaddr *uaddr,
706 int *uaddr_len, int peer)
707{
708 struct sock *sk = sock->sk;
709 struct inet_sock *inet = inet_sk(sk);
710 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
711
712 sin->sin_family = AF_INET;
713 if (peer) {
714 if (!inet->inet_dport ||
715 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
716 peer == 1))
717 return -ENOTCONN;
718 sin->sin_port = inet->inet_dport;
719 sin->sin_addr.s_addr = inet->inet_daddr;
720 } else {
721 __be32 addr = inet->inet_rcv_saddr;
722 if (!addr)
723 addr = inet->inet_saddr;
724 sin->sin_port = inet->inet_sport;
725 sin->sin_addr.s_addr = addr;
726 }
727 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
728 *uaddr_len = sizeof(*sin);
729 return 0;
730}
731EXPORT_SYMBOL(inet_getname);
732
733int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
734{
735 struct sock *sk = sock->sk;
736
737 sock_rps_record_flow(sk);
738
739 /* We may need to bind the socket. */
740 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
741 inet_autobind(sk))
742 return -EAGAIN;
743
744 return sk->sk_prot->sendmsg(sk, msg, size);
745}
746EXPORT_SYMBOL(inet_sendmsg);
747
748ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
749 size_t size, int flags)
750{
751 struct sock *sk = sock->sk;
752
753 sock_rps_record_flow(sk);
754
755 /* We may need to bind the socket. */
756 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
757 inet_autobind(sk))
758 return -EAGAIN;
759
760 if (sk->sk_prot->sendpage)
761 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
762 return sock_no_sendpage(sock, page, offset, size, flags);
763}
764EXPORT_SYMBOL(inet_sendpage);
765
766int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
767 int flags)
768{
769 struct sock *sk = sock->sk;
770 int addr_len = 0;
771 int err;
772
773 sock_rps_record_flow(sk);
774
775 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
776 flags & ~MSG_DONTWAIT, &addr_len);
777 if (err >= 0)
778 msg->msg_namelen = addr_len;
779 return err;
780}
781EXPORT_SYMBOL(inet_recvmsg);
782
783int inet_shutdown(struct socket *sock, int how)
784{
785 struct sock *sk = sock->sk;
786 int err = 0;
787
788 /* This should really check to make sure
789 * the socket is a TCP socket. (WHY AC...)
790 */
791 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
792 1->2 bit 2 snds.
793 2->3 */
794 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
795 return -EINVAL;
796
797 lock_sock(sk);
798 if (sock->state == SS_CONNECTING) {
799 if ((1 << sk->sk_state) &
800 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
801 sock->state = SS_DISCONNECTING;
802 else
803 sock->state = SS_CONNECTED;
804 }
805
806 switch (sk->sk_state) {
807 case TCP_CLOSE:
808 err = -ENOTCONN;
809 /* Hack to wake up other listeners, who can poll for
810 POLLHUP, even on eg. unconnected UDP sockets -- RR */
811 default:
812 sk->sk_shutdown |= how;
813 if (sk->sk_prot->shutdown)
814 sk->sk_prot->shutdown(sk, how);
815 break;
816
817 /* Remaining two branches are temporary solution for missing
818 * close() in multithreaded environment. It is _not_ a good idea,
819 * but we have no choice until close() is repaired at VFS level.
820 */
821 case TCP_LISTEN:
822 if (!(how & RCV_SHUTDOWN))
823 break;
824 /* Fall through */
825 case TCP_SYN_SENT:
826 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
827 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
828 break;
829 }
830
831 /* Wake up anyone sleeping in poll. */
832 sk->sk_state_change(sk);
833 release_sock(sk);
834 return err;
835}
836EXPORT_SYMBOL(inet_shutdown);
837
838/*
839 * ioctl() calls you can issue on an INET socket. Most of these are
840 * device configuration and stuff and very rarely used. Some ioctls
841 * pass on to the socket itself.
842 *
843 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
844 * loads the devconfigure module does its configuring and unloads it.
845 * There's a good 20K of config code hanging around the kernel.
846 */
847
848int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
849{
850 struct sock *sk = sock->sk;
851 int err = 0;
852 struct net *net = sock_net(sk);
853
854 switch (cmd) {
855 case SIOCGSTAMP:
856 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
857 break;
858 case SIOCGSTAMPNS:
859 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
860 break;
861 case SIOCADDRT:
862 case SIOCDELRT:
863 case SIOCRTMSG:
864 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
865 break;
866 case SIOCDARP:
867 case SIOCGARP:
868 case SIOCSARP:
869 err = arp_ioctl(net, cmd, (void __user *)arg);
870 break;
871 case SIOCGIFADDR:
872 case SIOCSIFADDR:
873 case SIOCGIFBRDADDR:
874 case SIOCSIFBRDADDR:
875 case SIOCGIFNETMASK:
876 case SIOCSIFNETMASK:
877 case SIOCGIFDSTADDR:
878 case SIOCSIFDSTADDR:
879 case SIOCSIFPFLAGS:
880 case SIOCGIFPFLAGS:
881 case SIOCSIFFLAGS:
882 err = devinet_ioctl(net, cmd, (void __user *)arg);
883 break;
884 default:
885 if (sk->sk_prot->ioctl)
886 err = sk->sk_prot->ioctl(sk, cmd, arg);
887 else
888 err = -ENOIOCTLCMD;
889 break;
890 }
891 return err;
892}
893EXPORT_SYMBOL(inet_ioctl);
894
895#ifdef CONFIG_COMPAT
896static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
897{
898 struct sock *sk = sock->sk;
899 int err = -ENOIOCTLCMD;
900
901 if (sk->sk_prot->compat_ioctl)
902 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
903
904 return err;
905}
906#endif
907
908const struct proto_ops inet_stream_ops = {
909 .family = PF_INET,
910 .owner = THIS_MODULE,
911 .release = inet_release,
912 .bind = inet_bind,
913 .connect = inet_stream_connect,
914 .socketpair = sock_no_socketpair,
915 .accept = inet_accept,
916 .getname = inet_getname,
917 .poll = tcp_poll,
918 .ioctl = inet_ioctl,
919 .listen = inet_listen,
920 .shutdown = inet_shutdown,
921 .setsockopt = sock_common_setsockopt,
922 .getsockopt = sock_common_getsockopt,
923 .sendmsg = inet_sendmsg,
924 .recvmsg = inet_recvmsg,
925 .mmap = sock_no_mmap,
926 .sendpage = inet_sendpage,
927 .splice_read = tcp_splice_read,
928 .read_sock = tcp_read_sock,
929 .peek_len = tcp_peek_len,
930#ifdef CONFIG_COMPAT
931 .compat_setsockopt = compat_sock_common_setsockopt,
932 .compat_getsockopt = compat_sock_common_getsockopt,
933 .compat_ioctl = inet_compat_ioctl,
934#endif
935};
936EXPORT_SYMBOL(inet_stream_ops);
937
938const struct proto_ops inet_dgram_ops = {
939 .family = PF_INET,
940 .owner = THIS_MODULE,
941 .release = inet_release,
942 .bind = inet_bind,
943 .connect = inet_dgram_connect,
944 .socketpair = sock_no_socketpair,
945 .accept = sock_no_accept,
946 .getname = inet_getname,
947 .poll = udp_poll,
948 .ioctl = inet_ioctl,
949 .listen = sock_no_listen,
950 .shutdown = inet_shutdown,
951 .setsockopt = sock_common_setsockopt,
952 .getsockopt = sock_common_getsockopt,
953 .sendmsg = inet_sendmsg,
954 .recvmsg = inet_recvmsg,
955 .mmap = sock_no_mmap,
956 .sendpage = inet_sendpage,
957 .set_peek_off = sk_set_peek_off,
958#ifdef CONFIG_COMPAT
959 .compat_setsockopt = compat_sock_common_setsockopt,
960 .compat_getsockopt = compat_sock_common_getsockopt,
961 .compat_ioctl = inet_compat_ioctl,
962#endif
963};
964EXPORT_SYMBOL(inet_dgram_ops);
965
966/*
967 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
968 * udp_poll
969 */
970static const struct proto_ops inet_sockraw_ops = {
971 .family = PF_INET,
972 .owner = THIS_MODULE,
973 .release = inet_release,
974 .bind = inet_bind,
975 .connect = inet_dgram_connect,
976 .socketpair = sock_no_socketpair,
977 .accept = sock_no_accept,
978 .getname = inet_getname,
979 .poll = datagram_poll,
980 .ioctl = inet_ioctl,
981 .listen = sock_no_listen,
982 .shutdown = inet_shutdown,
983 .setsockopt = sock_common_setsockopt,
984 .getsockopt = sock_common_getsockopt,
985 .sendmsg = inet_sendmsg,
986 .recvmsg = inet_recvmsg,
987 .mmap = sock_no_mmap,
988 .sendpage = inet_sendpage,
989#ifdef CONFIG_COMPAT
990 .compat_setsockopt = compat_sock_common_setsockopt,
991 .compat_getsockopt = compat_sock_common_getsockopt,
992 .compat_ioctl = inet_compat_ioctl,
993#endif
994};
995
996static const struct net_proto_family inet_family_ops = {
997 .family = PF_INET,
998 .create = inet_create,
999 .owner = THIS_MODULE,
1000};
1001
1002/* Upon startup we insert all the elements in inetsw_array[] into
1003 * the linked list inetsw.
1004 */
1005static struct inet_protosw inetsw_array[] =
1006{
1007 {
1008 .type = SOCK_STREAM,
1009 .protocol = IPPROTO_TCP,
1010 .prot = &tcp_prot,
1011 .ops = &inet_stream_ops,
1012 .flags = INET_PROTOSW_PERMANENT |
1013 INET_PROTOSW_ICSK,
1014 },
1015
1016 {
1017 .type = SOCK_DGRAM,
1018 .protocol = IPPROTO_UDP,
1019 .prot = &udp_prot,
1020 .ops = &inet_dgram_ops,
1021 .flags = INET_PROTOSW_PERMANENT,
1022 },
1023
1024 {
1025 .type = SOCK_DGRAM,
1026 .protocol = IPPROTO_ICMP,
1027 .prot = &ping_prot,
1028 .ops = &inet_dgram_ops,
1029 .flags = INET_PROTOSW_REUSE,
1030 },
1031
1032 {
1033 .type = SOCK_RAW,
1034 .protocol = IPPROTO_IP, /* wild card */
1035 .prot = &raw_prot,
1036 .ops = &inet_sockraw_ops,
1037 .flags = INET_PROTOSW_REUSE,
1038 }
1039};
1040
1041#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1042
1043void inet_register_protosw(struct inet_protosw *p)
1044{
1045 struct list_head *lh;
1046 struct inet_protosw *answer;
1047 int protocol = p->protocol;
1048 struct list_head *last_perm;
1049
1050 spin_lock_bh(&inetsw_lock);
1051
1052 if (p->type >= SOCK_MAX)
1053 goto out_illegal;
1054
1055 /* If we are trying to override a permanent protocol, bail. */
1056 last_perm = &inetsw[p->type];
1057 list_for_each(lh, &inetsw[p->type]) {
1058 answer = list_entry(lh, struct inet_protosw, list);
1059 /* Check only the non-wild match. */
1060 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1061 break;
1062 if (protocol == answer->protocol)
1063 goto out_permanent;
1064 last_perm = lh;
1065 }
1066
1067 /* Add the new entry after the last permanent entry if any, so that
1068 * the new entry does not override a permanent entry when matched with
1069 * a wild-card protocol. But it is allowed to override any existing
1070 * non-permanent entry. This means that when we remove this entry, the
1071 * system automatically returns to the old behavior.
1072 */
1073 list_add_rcu(&p->list, last_perm);
1074out:
1075 spin_unlock_bh(&inetsw_lock);
1076
1077 return;
1078
1079out_permanent:
1080 pr_err("Attempt to override permanent protocol %d\n", protocol);
1081 goto out;
1082
1083out_illegal:
1084 pr_err("Ignoring attempt to register invalid socket type %d\n",
1085 p->type);
1086 goto out;
1087}
1088EXPORT_SYMBOL(inet_register_protosw);
1089
1090void inet_unregister_protosw(struct inet_protosw *p)
1091{
1092 if (INET_PROTOSW_PERMANENT & p->flags) {
1093 pr_err("Attempt to unregister permanent protocol %d\n",
1094 p->protocol);
1095 } else {
1096 spin_lock_bh(&inetsw_lock);
1097 list_del_rcu(&p->list);
1098 spin_unlock_bh(&inetsw_lock);
1099
1100 synchronize_net();
1101 }
1102}
1103EXPORT_SYMBOL(inet_unregister_protosw);
1104
1105static int inet_sk_reselect_saddr(struct sock *sk)
1106{
1107 struct inet_sock *inet = inet_sk(sk);
1108 __be32 old_saddr = inet->inet_saddr;
1109 __be32 daddr = inet->inet_daddr;
1110 struct flowi4 *fl4;
1111 struct rtable *rt;
1112 __be32 new_saddr;
1113 struct ip_options_rcu *inet_opt;
1114
1115 inet_opt = rcu_dereference_protected(inet->inet_opt,
1116 lockdep_sock_is_held(sk));
1117 if (inet_opt && inet_opt->opt.srr)
1118 daddr = inet_opt->opt.faddr;
1119
1120 /* Query new route. */
1121 fl4 = &inet->cork.fl.u.ip4;
1122 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1123 sk->sk_bound_dev_if, sk->sk_protocol,
1124 inet->inet_sport, inet->inet_dport, sk);
1125 if (IS_ERR(rt))
1126 return PTR_ERR(rt);
1127
1128 sk_setup_caps(sk, &rt->dst);
1129
1130 new_saddr = fl4->saddr;
1131
1132 if (new_saddr == old_saddr)
1133 return 0;
1134
1135 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1136 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1137 __func__, &old_saddr, &new_saddr);
1138 }
1139
1140 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1141
1142 /*
1143 * XXX The only one ugly spot where we need to
1144 * XXX really change the sockets identity after
1145 * XXX it has entered the hashes. -DaveM
1146 *
1147 * Besides that, it does not check for connection
1148 * uniqueness. Wait for troubles.
1149 */
1150 return __sk_prot_rehash(sk);
1151}
1152
1153int inet_sk_rebuild_header(struct sock *sk)
1154{
1155 struct inet_sock *inet = inet_sk(sk);
1156 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1157 __be32 daddr;
1158 struct ip_options_rcu *inet_opt;
1159 struct flowi4 *fl4;
1160 int err;
1161
1162 /* Route is OK, nothing to do. */
1163 if (rt)
1164 return 0;
1165
1166 /* Reroute. */
1167 rcu_read_lock();
1168 inet_opt = rcu_dereference(inet->inet_opt);
1169 daddr = inet->inet_daddr;
1170 if (inet_opt && inet_opt->opt.srr)
1171 daddr = inet_opt->opt.faddr;
1172 rcu_read_unlock();
1173 fl4 = &inet->cork.fl.u.ip4;
1174 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1175 inet->inet_dport, inet->inet_sport,
1176 sk->sk_protocol, RT_CONN_FLAGS(sk),
1177 sk->sk_bound_dev_if);
1178 if (!IS_ERR(rt)) {
1179 err = 0;
1180 sk_setup_caps(sk, &rt->dst);
1181 } else {
1182 err = PTR_ERR(rt);
1183
1184 /* Routing failed... */
1185 sk->sk_route_caps = 0;
1186 /*
1187 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1188 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1189 */
1190 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1191 sk->sk_state != TCP_SYN_SENT ||
1192 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1193 (err = inet_sk_reselect_saddr(sk)) != 0)
1194 sk->sk_err_soft = -err;
1195 }
1196
1197 return err;
1198}
1199EXPORT_SYMBOL(inet_sk_rebuild_header);
1200
1201struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1202 netdev_features_t features)
1203{
1204 bool udpfrag = false, fixedid = false, gso_partial, encap;
1205 struct sk_buff *segs = ERR_PTR(-EINVAL);
1206 const struct net_offload *ops;
1207 unsigned int offset = 0;
1208 struct iphdr *iph;
1209 int proto, tot_len;
1210 int nhoff;
1211 int ihl;
1212 int id;
1213
1214 skb_reset_network_header(skb);
1215 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1216 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1217 goto out;
1218
1219 iph = ip_hdr(skb);
1220 ihl = iph->ihl * 4;
1221 if (ihl < sizeof(*iph))
1222 goto out;
1223
1224 id = ntohs(iph->id);
1225 proto = iph->protocol;
1226
1227 /* Warning: after this point, iph might be no longer valid */
1228 if (unlikely(!pskb_may_pull(skb, ihl)))
1229 goto out;
1230 __skb_pull(skb, ihl);
1231
1232 encap = SKB_GSO_CB(skb)->encap_level > 0;
1233 if (encap)
1234 features &= skb->dev->hw_enc_features;
1235 SKB_GSO_CB(skb)->encap_level += ihl;
1236
1237 skb_reset_transport_header(skb);
1238
1239 segs = ERR_PTR(-EPROTONOSUPPORT);
1240
1241 if (!skb->encapsulation || encap) {
1242 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1243 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1244
1245 /* fixed ID is invalid if DF bit is not set */
1246 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1247 goto out;
1248 }
1249
1250 ops = rcu_dereference(inet_offloads[proto]);
1251 if (likely(ops && ops->callbacks.gso_segment))
1252 segs = ops->callbacks.gso_segment(skb, features);
1253
1254 if (IS_ERR_OR_NULL(segs))
1255 goto out;
1256
1257 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1258
1259 skb = segs;
1260 do {
1261 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1262 if (udpfrag) {
1263 iph->frag_off = htons(offset >> 3);
1264 if (skb->next)
1265 iph->frag_off |= htons(IP_MF);
1266 offset += skb->len - nhoff - ihl;
1267 tot_len = skb->len - nhoff;
1268 } else if (skb_is_gso(skb)) {
1269 if (!fixedid) {
1270 iph->id = htons(id);
1271 id += skb_shinfo(skb)->gso_segs;
1272 }
1273
1274 if (gso_partial)
1275 tot_len = skb_shinfo(skb)->gso_size +
1276 SKB_GSO_CB(skb)->data_offset +
1277 skb->head - (unsigned char *)iph;
1278 else
1279 tot_len = skb->len - nhoff;
1280 } else {
1281 if (!fixedid)
1282 iph->id = htons(id++);
1283 tot_len = skb->len - nhoff;
1284 }
1285 iph->tot_len = htons(tot_len);
1286 ip_send_check(iph);
1287 if (encap)
1288 skb_reset_inner_headers(skb);
1289 skb->network_header = (u8 *)iph - skb->head;
1290 } while ((skb = skb->next));
1291
1292out:
1293 return segs;
1294}
1295EXPORT_SYMBOL(inet_gso_segment);
1296
1297struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
1298{
1299 const struct net_offload *ops;
1300 struct sk_buff **pp = NULL;
1301 struct sk_buff *p;
1302 const struct iphdr *iph;
1303 unsigned int hlen;
1304 unsigned int off;
1305 unsigned int id;
1306 int flush = 1;
1307 int proto;
1308
1309 off = skb_gro_offset(skb);
1310 hlen = off + sizeof(*iph);
1311 iph = skb_gro_header_fast(skb, off);
1312 if (skb_gro_header_hard(skb, hlen)) {
1313 iph = skb_gro_header_slow(skb, hlen, off);
1314 if (unlikely(!iph))
1315 goto out;
1316 }
1317
1318 proto = iph->protocol;
1319
1320 rcu_read_lock();
1321 ops = rcu_dereference(inet_offloads[proto]);
1322 if (!ops || !ops->callbacks.gro_receive)
1323 goto out_unlock;
1324
1325 if (*(u8 *)iph != 0x45)
1326 goto out_unlock;
1327
1328 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1329 goto out_unlock;
1330
1331 id = ntohl(*(__be32 *)&iph->id);
1332 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1333 id >>= 16;
1334
1335 for (p = *head; p; p = p->next) {
1336 struct iphdr *iph2;
1337 u16 flush_id;
1338
1339 if (!NAPI_GRO_CB(p)->same_flow)
1340 continue;
1341
1342 iph2 = (struct iphdr *)(p->data + off);
1343 /* The above works because, with the exception of the top
1344 * (inner most) layer, we only aggregate pkts with the same
1345 * hdr length so all the hdrs we'll need to verify will start
1346 * at the same offset.
1347 */
1348 if ((iph->protocol ^ iph2->protocol) |
1349 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1350 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1351 NAPI_GRO_CB(p)->same_flow = 0;
1352 continue;
1353 }
1354
1355 /* All fields must match except length and checksum. */
1356 NAPI_GRO_CB(p)->flush |=
1357 (iph->ttl ^ iph2->ttl) |
1358 (iph->tos ^ iph2->tos) |
1359 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1360
1361 NAPI_GRO_CB(p)->flush |= flush;
1362
1363 /* We need to store of the IP ID check to be included later
1364 * when we can verify that this packet does in fact belong
1365 * to a given flow.
1366 */
1367 flush_id = (u16)(id - ntohs(iph2->id));
1368
1369 /* This bit of code makes it much easier for us to identify
1370 * the cases where we are doing atomic vs non-atomic IP ID
1371 * checks. Specifically an atomic check can return IP ID
1372 * values 0 - 0xFFFF, while a non-atomic check can only
1373 * return 0 or 0xFFFF.
1374 */
1375 if (!NAPI_GRO_CB(p)->is_atomic ||
1376 !(iph->frag_off & htons(IP_DF))) {
1377 flush_id ^= NAPI_GRO_CB(p)->count;
1378 flush_id = flush_id ? 0xFFFF : 0;
1379 }
1380
1381 /* If the previous IP ID value was based on an atomic
1382 * datagram we can overwrite the value and ignore it.
1383 */
1384 if (NAPI_GRO_CB(skb)->is_atomic)
1385 NAPI_GRO_CB(p)->flush_id = flush_id;
1386 else
1387 NAPI_GRO_CB(p)->flush_id |= flush_id;
1388 }
1389
1390 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1391 NAPI_GRO_CB(skb)->flush |= flush;
1392 skb_set_network_header(skb, off);
1393 /* The above will be needed by the transport layer if there is one
1394 * immediately following this IP hdr.
1395 */
1396
1397 /* Note : No need to call skb_gro_postpull_rcsum() here,
1398 * as we already checked checksum over ipv4 header was 0
1399 */
1400 skb_gro_pull(skb, sizeof(*iph));
1401 skb_set_transport_header(skb, skb_gro_offset(skb));
1402
1403 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1404
1405out_unlock:
1406 rcu_read_unlock();
1407
1408out:
1409 NAPI_GRO_CB(skb)->flush |= flush;
1410
1411 return pp;
1412}
1413EXPORT_SYMBOL(inet_gro_receive);
1414
1415static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1416 struct sk_buff *skb)
1417{
1418 if (NAPI_GRO_CB(skb)->encap_mark) {
1419 NAPI_GRO_CB(skb)->flush = 1;
1420 return NULL;
1421 }
1422
1423 NAPI_GRO_CB(skb)->encap_mark = 1;
1424
1425 return inet_gro_receive(head, skb);
1426}
1427
1428#define SECONDS_PER_DAY 86400
1429
1430/* inet_current_timestamp - Return IP network timestamp
1431 *
1432 * Return milliseconds since midnight in network byte order.
1433 */
1434__be32 inet_current_timestamp(void)
1435{
1436 u32 secs;
1437 u32 msecs;
1438 struct timespec64 ts;
1439
1440 ktime_get_real_ts64(&ts);
1441
1442 /* Get secs since midnight. */
1443 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1444 /* Convert to msecs. */
1445 msecs = secs * MSEC_PER_SEC;
1446 /* Convert nsec to msec. */
1447 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1448
1449 /* Convert to network byte order. */
1450 return htonl(msecs);
1451}
1452EXPORT_SYMBOL(inet_current_timestamp);
1453
1454int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1455{
1456 if (sk->sk_family == AF_INET)
1457 return ip_recv_error(sk, msg, len, addr_len);
1458#if IS_ENABLED(CONFIG_IPV6)
1459 if (sk->sk_family == AF_INET6)
1460 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1461#endif
1462 return -EINVAL;
1463}
1464
1465int inet_gro_complete(struct sk_buff *skb, int nhoff)
1466{
1467 __be16 newlen = htons(skb->len - nhoff);
1468 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1469 const struct net_offload *ops;
1470 int proto = iph->protocol;
1471 int err = -ENOSYS;
1472
1473 if (skb->encapsulation) {
1474 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1475 skb_set_inner_network_header(skb, nhoff);
1476 }
1477
1478 csum_replace2(&iph->check, iph->tot_len, newlen);
1479 iph->tot_len = newlen;
1480
1481 rcu_read_lock();
1482 ops = rcu_dereference(inet_offloads[proto]);
1483 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1484 goto out_unlock;
1485
1486 /* Only need to add sizeof(*iph) to get to the next hdr below
1487 * because any hdr with option will have been flushed in
1488 * inet_gro_receive().
1489 */
1490 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1491
1492out_unlock:
1493 rcu_read_unlock();
1494
1495 return err;
1496}
1497EXPORT_SYMBOL(inet_gro_complete);
1498
1499static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1500{
1501 skb->encapsulation = 1;
1502 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1503 return inet_gro_complete(skb, nhoff);
1504}
1505
1506int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1507 unsigned short type, unsigned char protocol,
1508 struct net *net)
1509{
1510 struct socket *sock;
1511 int rc = sock_create_kern(net, family, type, protocol, &sock);
1512
1513 if (rc == 0) {
1514 *sk = sock->sk;
1515 (*sk)->sk_allocation = GFP_ATOMIC;
1516 /*
1517 * Unhash it so that IP input processing does not even see it,
1518 * we do not wish this socket to see incoming packets.
1519 */
1520 (*sk)->sk_prot->unhash(*sk);
1521 }
1522 return rc;
1523}
1524EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1525
1526u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1527{
1528 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1529}
1530EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1531
1532unsigned long snmp_fold_field(void __percpu *mib, int offt)
1533{
1534 unsigned long res = 0;
1535 int i;
1536
1537 for_each_possible_cpu(i)
1538 res += snmp_get_cpu_field(mib, i, offt);
1539 return res;
1540}
1541EXPORT_SYMBOL_GPL(snmp_fold_field);
1542
1543#if BITS_PER_LONG==32
1544
1545u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1546 size_t syncp_offset)
1547{
1548 void *bhptr;
1549 struct u64_stats_sync *syncp;
1550 u64 v;
1551 unsigned int start;
1552
1553 bhptr = per_cpu_ptr(mib, cpu);
1554 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1555 do {
1556 start = u64_stats_fetch_begin_irq(syncp);
1557 v = *(((u64 *)bhptr) + offt);
1558 } while (u64_stats_fetch_retry_irq(syncp, start));
1559
1560 return v;
1561}
1562EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1563
1564u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1565{
1566 u64 res = 0;
1567 int cpu;
1568
1569 for_each_possible_cpu(cpu) {
1570 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1571 }
1572 return res;
1573}
1574EXPORT_SYMBOL_GPL(snmp_fold_field64);
1575#endif
1576
1577#ifdef CONFIG_IP_MULTICAST
1578static const struct net_protocol igmp_protocol = {
1579 .handler = igmp_rcv,
1580 .netns_ok = 1,
1581};
1582#endif
1583
1584static const struct net_protocol tcp_protocol = {
1585 .early_demux = tcp_v4_early_demux,
1586 .handler = tcp_v4_rcv,
1587 .err_handler = tcp_v4_err,
1588 .no_policy = 1,
1589 .netns_ok = 1,
1590 .icmp_strict_tag_validation = 1,
1591};
1592
1593static const struct net_protocol udp_protocol = {
1594 .early_demux = udp_v4_early_demux,
1595 .handler = udp_rcv,
1596 .err_handler = udp_err,
1597 .no_policy = 1,
1598 .netns_ok = 1,
1599};
1600
1601static const struct net_protocol icmp_protocol = {
1602 .handler = icmp_rcv,
1603 .err_handler = icmp_err,
1604 .no_policy = 1,
1605 .netns_ok = 1,
1606};
1607
1608static __net_init int ipv4_mib_init_net(struct net *net)
1609{
1610 int i;
1611
1612 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1613 if (!net->mib.tcp_statistics)
1614 goto err_tcp_mib;
1615 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1616 if (!net->mib.ip_statistics)
1617 goto err_ip_mib;
1618
1619 for_each_possible_cpu(i) {
1620 struct ipstats_mib *af_inet_stats;
1621 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1622 u64_stats_init(&af_inet_stats->syncp);
1623 }
1624
1625 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1626 if (!net->mib.net_statistics)
1627 goto err_net_mib;
1628 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1629 if (!net->mib.udp_statistics)
1630 goto err_udp_mib;
1631 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1632 if (!net->mib.udplite_statistics)
1633 goto err_udplite_mib;
1634 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1635 if (!net->mib.icmp_statistics)
1636 goto err_icmp_mib;
1637 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1638 GFP_KERNEL);
1639 if (!net->mib.icmpmsg_statistics)
1640 goto err_icmpmsg_mib;
1641
1642 tcp_mib_init(net);
1643 return 0;
1644
1645err_icmpmsg_mib:
1646 free_percpu(net->mib.icmp_statistics);
1647err_icmp_mib:
1648 free_percpu(net->mib.udplite_statistics);
1649err_udplite_mib:
1650 free_percpu(net->mib.udp_statistics);
1651err_udp_mib:
1652 free_percpu(net->mib.net_statistics);
1653err_net_mib:
1654 free_percpu(net->mib.ip_statistics);
1655err_ip_mib:
1656 free_percpu(net->mib.tcp_statistics);
1657err_tcp_mib:
1658 return -ENOMEM;
1659}
1660
1661static __net_exit void ipv4_mib_exit_net(struct net *net)
1662{
1663 kfree(net->mib.icmpmsg_statistics);
1664 free_percpu(net->mib.icmp_statistics);
1665 free_percpu(net->mib.udplite_statistics);
1666 free_percpu(net->mib.udp_statistics);
1667 free_percpu(net->mib.net_statistics);
1668 free_percpu(net->mib.ip_statistics);
1669 free_percpu(net->mib.tcp_statistics);
1670}
1671
1672static __net_initdata struct pernet_operations ipv4_mib_ops = {
1673 .init = ipv4_mib_init_net,
1674 .exit = ipv4_mib_exit_net,
1675};
1676
1677static int __init init_ipv4_mibs(void)
1678{
1679 return register_pernet_subsys(&ipv4_mib_ops);
1680}
1681
1682static __net_init int inet_init_net(struct net *net)
1683{
1684 /*
1685 * Set defaults for local port range
1686 */
1687 seqlock_init(&net->ipv4.ip_local_ports.lock);
1688 net->ipv4.ip_local_ports.range[0] = 32768;
1689 net->ipv4.ip_local_ports.range[1] = 60999;
1690
1691 seqlock_init(&net->ipv4.ping_group_range.lock);
1692 /*
1693 * Sane defaults - nobody may create ping sockets.
1694 * Boot scripts should set this to distro-specific group.
1695 */
1696 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1697 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1698
1699 /* Default values for sysctl-controlled parameters.
1700 * We set them here, in case sysctl is not compiled.
1701 */
1702 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1703 net->ipv4.sysctl_ip_dynaddr = 0;
1704 net->ipv4.sysctl_ip_early_demux = 1;
1705
1706 return 0;
1707}
1708
1709static __net_exit void inet_exit_net(struct net *net)
1710{
1711}
1712
1713static __net_initdata struct pernet_operations af_inet_ops = {
1714 .init = inet_init_net,
1715 .exit = inet_exit_net,
1716};
1717
1718static int __init init_inet_pernet_ops(void)
1719{
1720 return register_pernet_subsys(&af_inet_ops);
1721}
1722
1723static int ipv4_proc_init(void);
1724
1725/*
1726 * IP protocol layer initialiser
1727 */
1728
1729static struct packet_offload ip_packet_offload __read_mostly = {
1730 .type = cpu_to_be16(ETH_P_IP),
1731 .callbacks = {
1732 .gso_segment = inet_gso_segment,
1733 .gro_receive = inet_gro_receive,
1734 .gro_complete = inet_gro_complete,
1735 },
1736};
1737
1738static const struct net_offload ipip_offload = {
1739 .callbacks = {
1740 .gso_segment = inet_gso_segment,
1741 .gro_receive = ipip_gro_receive,
1742 .gro_complete = ipip_gro_complete,
1743 },
1744};
1745
1746static int __init ipv4_offload_init(void)
1747{
1748 /*
1749 * Add offloads
1750 */
1751 if (udpv4_offload_init() < 0)
1752 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1753 if (tcpv4_offload_init() < 0)
1754 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1755
1756 dev_add_offload(&ip_packet_offload);
1757 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1758 return 0;
1759}
1760
1761fs_initcall(ipv4_offload_init);
1762
1763static struct packet_type ip_packet_type __read_mostly = {
1764 .type = cpu_to_be16(ETH_P_IP),
1765 .func = ip_rcv,
1766};
1767
1768static int __init inet_init(void)
1769{
1770 struct inet_protosw *q;
1771 struct list_head *r;
1772 int rc = -EINVAL;
1773
1774 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1775
1776 rc = proto_register(&tcp_prot, 1);
1777 if (rc)
1778 goto out;
1779
1780 rc = proto_register(&udp_prot, 1);
1781 if (rc)
1782 goto out_unregister_tcp_proto;
1783
1784 rc = proto_register(&raw_prot, 1);
1785 if (rc)
1786 goto out_unregister_udp_proto;
1787
1788 rc = proto_register(&ping_prot, 1);
1789 if (rc)
1790 goto out_unregister_raw_proto;
1791
1792 /*
1793 * Tell SOCKET that we are alive...
1794 */
1795
1796 (void)sock_register(&inet_family_ops);
1797
1798#ifdef CONFIG_SYSCTL
1799 ip_static_sysctl_init();
1800#endif
1801
1802 /*
1803 * Add all the base protocols.
1804 */
1805
1806 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1807 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1808 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1809 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1810 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1811 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1812#ifdef CONFIG_IP_MULTICAST
1813 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1814 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1815#endif
1816
1817 /* Register the socket-side information for inet_create. */
1818 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1819 INIT_LIST_HEAD(r);
1820
1821 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1822 inet_register_protosw(q);
1823
1824 /*
1825 * Set the ARP module up
1826 */
1827
1828 arp_init();
1829
1830 /*
1831 * Set the IP module up
1832 */
1833
1834 ip_init();
1835
1836 tcp_v4_init();
1837
1838 /* Setup TCP slab cache for open requests. */
1839 tcp_init();
1840
1841 /* Setup UDP memory threshold */
1842 udp_init();
1843
1844 /* Add UDP-Lite (RFC 3828) */
1845 udplite4_register();
1846
1847 ping_init();
1848
1849 /*
1850 * Set the ICMP layer up
1851 */
1852
1853 if (icmp_init() < 0)
1854 panic("Failed to create the ICMP control socket.\n");
1855
1856 /*
1857 * Initialise the multicast router
1858 */
1859#if defined(CONFIG_IP_MROUTE)
1860 if (ip_mr_init())
1861 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1862#endif
1863
1864 if (init_inet_pernet_ops())
1865 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1866 /*
1867 * Initialise per-cpu ipv4 mibs
1868 */
1869
1870 if (init_ipv4_mibs())
1871 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1872
1873 ipv4_proc_init();
1874
1875 ipfrag_init();
1876
1877 dev_add_pack(&ip_packet_type);
1878
1879 ip_tunnel_core_init();
1880
1881 rc = 0;
1882out:
1883 return rc;
1884out_unregister_raw_proto:
1885 proto_unregister(&raw_prot);
1886out_unregister_udp_proto:
1887 proto_unregister(&udp_prot);
1888out_unregister_tcp_proto:
1889 proto_unregister(&tcp_prot);
1890 goto out;
1891}
1892
1893fs_initcall(inet_init);
1894
1895/* ------------------------------------------------------------------------ */
1896
1897#ifdef CONFIG_PROC_FS
1898static int __init ipv4_proc_init(void)
1899{
1900 int rc = 0;
1901
1902 if (raw_proc_init())
1903 goto out_raw;
1904 if (tcp4_proc_init())
1905 goto out_tcp;
1906 if (udp4_proc_init())
1907 goto out_udp;
1908 if (ping_proc_init())
1909 goto out_ping;
1910 if (ip_misc_proc_init())
1911 goto out_misc;
1912out:
1913 return rc;
1914out_misc:
1915 ping_proc_exit();
1916out_ping:
1917 udp4_proc_exit();
1918out_udp:
1919 tcp4_proc_exit();
1920out_tcp:
1921 raw_proc_exit();
1922out_raw:
1923 rc = -ENOMEM;
1924 goto out;
1925}
1926
1927#else /* CONFIG_PROC_FS */
1928static int __init ipv4_proc_init(void)
1929{
1930 return 0;
1931}
1932#endif /* CONFIG_PROC_FS */