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/gro.h>
103#include <net/gso.h>
104#include <net/tcp.h>
105#include <net/udp.h>
106#include <net/udplite.h>
107#include <net/ping.h>
108#include <linux/skbuff.h>
109#include <net/sock.h>
110#include <net/raw.h>
111#include <net/icmp.h>
112#include <net/inet_common.h>
113#include <net/ip_tunnels.h>
114#include <net/xfrm.h>
115#include <net/net_namespace.h>
116#include <net/secure_seq.h>
117#ifdef CONFIG_IP_MROUTE
118#include <linux/mroute.h>
119#endif
120#include <net/l3mdev.h>
121#include <net/compat.h>
122#include <net/rps.h>
123
124#include <trace/events/sock.h>
125
126/* The inetsw table contains everything that inet_create needs to
127 * build a new socket.
128 */
129static struct list_head inetsw[SOCK_MAX];
130static DEFINE_SPINLOCK(inetsw_lock);
131
132/* New destruction routine */
133
134void inet_sock_destruct(struct sock *sk)
135{
136 struct inet_sock *inet = inet_sk(sk);
137
138 __skb_queue_purge(&sk->sk_receive_queue);
139 __skb_queue_purge(&sk->sk_error_queue);
140
141 sk_mem_reclaim_final(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_ONCE(atomic_read(&sk->sk_rmem_alloc));
154 WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc));
155 WARN_ON_ONCE(sk->sk_wmem_queued);
156 WARN_ON_ONCE(sk_forward_alloc_get(sk));
157
158 kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1));
160 dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1));
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
191int __inet_listen_sk(struct sock *sk, int backlog)
192{
193 unsigned char old_state = sk->sk_state;
194 int err, tcp_fastopen;
195
196 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
197 return -EINVAL;
198
199 WRITE_ONCE(sk->sk_max_ack_backlog, backlog);
200 /* Really, if the socket is already in listen state
201 * we can only allow the backlog to be adjusted.
202 */
203 if (old_state != TCP_LISTEN) {
204 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
205 * Note that only TCP sockets (SOCK_STREAM) will reach here.
206 * Also fastopen backlog may already been set via the option
207 * because the socket was in TCP_LISTEN state previously but
208 * was shutdown() rather than close().
209 */
210 tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen);
211 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
212 (tcp_fastopen & TFO_SERVER_ENABLE) &&
213 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
214 fastopen_queue_tune(sk, backlog);
215 tcp_fastopen_init_key_once(sock_net(sk));
216 }
217
218 err = inet_csk_listen_start(sk);
219 if (err)
220 return err;
221
222 tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_LISTEN_CB, 0, NULL);
223 }
224 return 0;
225}
226
227/*
228 * Move a socket into listening state.
229 */
230int inet_listen(struct socket *sock, int backlog)
231{
232 struct sock *sk = sock->sk;
233 int err = -EINVAL;
234
235 lock_sock(sk);
236
237 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
238 goto out;
239
240 err = __inet_listen_sk(sk, backlog);
241
242out:
243 release_sock(sk);
244 return err;
245}
246EXPORT_SYMBOL(inet_listen);
247
248/*
249 * Create an inet socket.
250 */
251
252static int inet_create(struct net *net, struct socket *sock, int protocol,
253 int kern)
254{
255 struct sock *sk;
256 struct inet_protosw *answer;
257 struct inet_sock *inet;
258 struct proto *answer_prot;
259 unsigned char answer_flags;
260 int try_loading_module = 0;
261 int err;
262
263 if (protocol < 0 || protocol >= IPPROTO_MAX)
264 return -EINVAL;
265
266 sock->state = SS_UNCONNECTED;
267
268 /* Look for the requested type/protocol pair. */
269lookup_protocol:
270 err = -ESOCKTNOSUPPORT;
271 rcu_read_lock();
272 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
273
274 err = 0;
275 /* Check the non-wild match. */
276 if (protocol == answer->protocol) {
277 if (protocol != IPPROTO_IP)
278 break;
279 } else {
280 /* Check for the two wild cases. */
281 if (IPPROTO_IP == protocol) {
282 protocol = answer->protocol;
283 break;
284 }
285 if (IPPROTO_IP == answer->protocol)
286 break;
287 }
288 err = -EPROTONOSUPPORT;
289 }
290
291 if (unlikely(err)) {
292 if (try_loading_module < 2) {
293 rcu_read_unlock();
294 /*
295 * Be more specific, e.g. net-pf-2-proto-132-type-1
296 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
297 */
298 if (++try_loading_module == 1)
299 request_module("net-pf-%d-proto-%d-type-%d",
300 PF_INET, protocol, sock->type);
301 /*
302 * Fall back to generic, e.g. net-pf-2-proto-132
303 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
304 */
305 else
306 request_module("net-pf-%d-proto-%d",
307 PF_INET, protocol);
308 goto lookup_protocol;
309 } else
310 goto out_rcu_unlock;
311 }
312
313 err = -EPERM;
314 if (sock->type == SOCK_RAW && !kern &&
315 !ns_capable(net->user_ns, CAP_NET_RAW))
316 goto out_rcu_unlock;
317
318 sock->ops = answer->ops;
319 answer_prot = answer->prot;
320 answer_flags = answer->flags;
321 rcu_read_unlock();
322
323 WARN_ON(!answer_prot->slab);
324
325 err = -ENOMEM;
326 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
327 if (!sk)
328 goto out;
329
330 err = 0;
331 if (INET_PROTOSW_REUSE & answer_flags)
332 sk->sk_reuse = SK_CAN_REUSE;
333
334 if (INET_PROTOSW_ICSK & answer_flags)
335 inet_init_csk_locks(sk);
336
337 inet = inet_sk(sk);
338 inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags);
339
340 inet_clear_bit(NODEFRAG, sk);
341
342 if (SOCK_RAW == sock->type) {
343 inet->inet_num = protocol;
344 if (IPPROTO_RAW == protocol)
345 inet_set_bit(HDRINCL, sk);
346 }
347
348 if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc))
349 inet->pmtudisc = IP_PMTUDISC_DONT;
350 else
351 inet->pmtudisc = IP_PMTUDISC_WANT;
352
353 atomic_set(&inet->inet_id, 0);
354
355 sock_init_data(sock, sk);
356
357 sk->sk_destruct = inet_sock_destruct;
358 sk->sk_protocol = protocol;
359 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
360 sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash);
361
362 inet->uc_ttl = -1;
363 inet_set_bit(MC_LOOP, sk);
364 inet->mc_ttl = 1;
365 inet_set_bit(MC_ALL, sk);
366 inet->mc_index = 0;
367 inet->mc_list = NULL;
368 inet->rcv_tos = 0;
369
370 if (inet->inet_num) {
371 /* It assumes that any protocol which allows
372 * the user to assign a number at socket
373 * creation time automatically
374 * shares.
375 */
376 inet->inet_sport = htons(inet->inet_num);
377 /* Add to protocol hash chains. */
378 err = sk->sk_prot->hash(sk);
379 if (err) {
380 sk_common_release(sk);
381 goto out;
382 }
383 }
384
385 if (sk->sk_prot->init) {
386 err = sk->sk_prot->init(sk);
387 if (err) {
388 sk_common_release(sk);
389 goto out;
390 }
391 }
392
393 if (!kern) {
394 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
395 if (err) {
396 sk_common_release(sk);
397 goto out;
398 }
399 }
400out:
401 return err;
402out_rcu_unlock:
403 rcu_read_unlock();
404 goto out;
405}
406
407
408/*
409 * The peer socket should always be NULL (or else). When we call this
410 * function we are destroying the object and from then on nobody
411 * should refer to it.
412 */
413int inet_release(struct socket *sock)
414{
415 struct sock *sk = sock->sk;
416
417 if (sk) {
418 long timeout;
419
420 if (!sk->sk_kern_sock)
421 BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
422
423 /* Applications forget to leave groups before exiting */
424 ip_mc_drop_socket(sk);
425
426 /* If linger is set, we don't return until the close
427 * is complete. Otherwise we return immediately. The
428 * actually closing is done the same either way.
429 *
430 * If the close is due to the process exiting, we never
431 * linger..
432 */
433 timeout = 0;
434 if (sock_flag(sk, SOCK_LINGER) &&
435 !(current->flags & PF_EXITING))
436 timeout = sk->sk_lingertime;
437 sk->sk_prot->close(sk, timeout);
438 sock->sk = NULL;
439 }
440 return 0;
441}
442EXPORT_SYMBOL(inet_release);
443
444int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len)
445{
446 u32 flags = BIND_WITH_LOCK;
447 int err;
448
449 /* If the socket has its own bind function then use it. (RAW) */
450 if (sk->sk_prot->bind) {
451 return sk->sk_prot->bind(sk, uaddr, addr_len);
452 }
453 if (addr_len < sizeof(struct sockaddr_in))
454 return -EINVAL;
455
456 /* BPF prog is run before any checks are done so that if the prog
457 * changes context in a wrong way it will be caught.
458 */
459 err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len,
460 CGROUP_INET4_BIND, &flags);
461 if (err)
462 return err;
463
464 return __inet_bind(sk, uaddr, addr_len, flags);
465}
466
467int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
468{
469 return inet_bind_sk(sock->sk, uaddr, addr_len);
470}
471EXPORT_SYMBOL(inet_bind);
472
473int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
474 u32 flags)
475{
476 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
477 struct inet_sock *inet = inet_sk(sk);
478 struct net *net = sock_net(sk);
479 unsigned short snum;
480 int chk_addr_ret;
481 u32 tb_id = RT_TABLE_LOCAL;
482 int err;
483
484 if (addr->sin_family != AF_INET) {
485 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
486 * only if s_addr is INADDR_ANY.
487 */
488 err = -EAFNOSUPPORT;
489 if (addr->sin_family != AF_UNSPEC ||
490 addr->sin_addr.s_addr != htonl(INADDR_ANY))
491 goto out;
492 }
493
494 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
495 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
496
497 /* Not specified by any standard per-se, however it breaks too
498 * many applications when removed. It is unfortunate since
499 * allowing applications to make a non-local bind solves
500 * several problems with systems using dynamic addressing.
501 * (ie. your servers still start up even if your ISDN link
502 * is temporarily down)
503 */
504 err = -EADDRNOTAVAIL;
505 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
506 chk_addr_ret))
507 goto out;
508
509 snum = ntohs(addr->sin_port);
510 err = -EACCES;
511 if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) &&
512 snum && inet_port_requires_bind_service(net, snum) &&
513 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
514 goto out;
515
516 /* We keep a pair of addresses. rcv_saddr is the one
517 * used by hash lookups, and saddr is used for transmit.
518 *
519 * In the BSD API these are the same except where it
520 * would be illegal to use them (multicast/broadcast) in
521 * which case the sending device address is used.
522 */
523 if (flags & BIND_WITH_LOCK)
524 lock_sock(sk);
525
526 /* Check these errors (active socket, double bind). */
527 err = -EINVAL;
528 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
529 goto out_release_sock;
530
531 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
532 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
533 inet->inet_saddr = 0; /* Use device */
534
535 /* Make sure we are allowed to bind here. */
536 if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) ||
537 (flags & BIND_FORCE_ADDRESS_NO_PORT))) {
538 err = sk->sk_prot->get_port(sk, snum);
539 if (err) {
540 inet->inet_saddr = inet->inet_rcv_saddr = 0;
541 goto out_release_sock;
542 }
543 if (!(flags & BIND_FROM_BPF)) {
544 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
545 if (err) {
546 inet->inet_saddr = inet->inet_rcv_saddr = 0;
547 if (sk->sk_prot->put_port)
548 sk->sk_prot->put_port(sk);
549 goto out_release_sock;
550 }
551 }
552 }
553
554 if (inet->inet_rcv_saddr)
555 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
556 if (snum)
557 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
558 inet->inet_sport = htons(inet->inet_num);
559 inet->inet_daddr = 0;
560 inet->inet_dport = 0;
561 sk_dst_reset(sk);
562 err = 0;
563out_release_sock:
564 if (flags & BIND_WITH_LOCK)
565 release_sock(sk);
566out:
567 return err;
568}
569
570int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
571 int addr_len, int flags)
572{
573 struct sock *sk = sock->sk;
574 const struct proto *prot;
575 int err;
576
577 if (addr_len < sizeof(uaddr->sa_family))
578 return -EINVAL;
579
580 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
581 prot = READ_ONCE(sk->sk_prot);
582
583 if (uaddr->sa_family == AF_UNSPEC)
584 return prot->disconnect(sk, flags);
585
586 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
587 err = prot->pre_connect(sk, uaddr, addr_len);
588 if (err)
589 return err;
590 }
591
592 if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk))
593 return -EAGAIN;
594 return prot->connect(sk, uaddr, addr_len);
595}
596EXPORT_SYMBOL(inet_dgram_connect);
597
598static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
599{
600 DEFINE_WAIT_FUNC(wait, woken_wake_function);
601
602 add_wait_queue(sk_sleep(sk), &wait);
603 sk->sk_write_pending += writebias;
604
605 /* Basic assumption: if someone sets sk->sk_err, he _must_
606 * change state of the socket from TCP_SYN_*.
607 * Connect() does not allow to get error notifications
608 * without closing the socket.
609 */
610 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
611 release_sock(sk);
612 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
613 lock_sock(sk);
614 if (signal_pending(current) || !timeo)
615 break;
616 }
617 remove_wait_queue(sk_sleep(sk), &wait);
618 sk->sk_write_pending -= writebias;
619 return timeo;
620}
621
622/*
623 * Connect to a remote host. There is regrettably still a little
624 * TCP 'magic' in here.
625 */
626int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
627 int addr_len, int flags, int is_sendmsg)
628{
629 struct sock *sk = sock->sk;
630 int err;
631 long timeo;
632
633 /*
634 * uaddr can be NULL and addr_len can be 0 if:
635 * sk is a TCP fastopen active socket and
636 * TCP_FASTOPEN_CONNECT sockopt is set and
637 * we already have a valid cookie for this socket.
638 * In this case, user can call write() after connect().
639 * write() will invoke tcp_sendmsg_fastopen() which calls
640 * __inet_stream_connect().
641 */
642 if (uaddr) {
643 if (addr_len < sizeof(uaddr->sa_family))
644 return -EINVAL;
645
646 if (uaddr->sa_family == AF_UNSPEC) {
647 sk->sk_disconnects++;
648 err = sk->sk_prot->disconnect(sk, flags);
649 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
650 goto out;
651 }
652 }
653
654 switch (sock->state) {
655 default:
656 err = -EINVAL;
657 goto out;
658 case SS_CONNECTED:
659 err = -EISCONN;
660 goto out;
661 case SS_CONNECTING:
662 if (inet_test_bit(DEFER_CONNECT, sk))
663 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
664 else
665 err = -EALREADY;
666 /* Fall out of switch with err, set for this state */
667 break;
668 case SS_UNCONNECTED:
669 err = -EISCONN;
670 if (sk->sk_state != TCP_CLOSE)
671 goto out;
672
673 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
674 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
675 if (err)
676 goto out;
677 }
678
679 err = sk->sk_prot->connect(sk, uaddr, addr_len);
680 if (err < 0)
681 goto out;
682
683 sock->state = SS_CONNECTING;
684
685 if (!err && inet_test_bit(DEFER_CONNECT, sk))
686 goto out;
687
688 /* Just entered SS_CONNECTING state; the only
689 * difference is that return value in non-blocking
690 * case is EINPROGRESS, rather than EALREADY.
691 */
692 err = -EINPROGRESS;
693 break;
694 }
695
696 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
697
698 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
699 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
700 tcp_sk(sk)->fastopen_req &&
701 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
702 int dis = sk->sk_disconnects;
703
704 /* Error code is set above */
705 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
706 goto out;
707
708 err = sock_intr_errno(timeo);
709 if (signal_pending(current))
710 goto out;
711
712 if (dis != sk->sk_disconnects) {
713 err = -EPIPE;
714 goto out;
715 }
716 }
717
718 /* Connection was closed by RST, timeout, ICMP error
719 * or another process disconnected us.
720 */
721 if (sk->sk_state == TCP_CLOSE)
722 goto sock_error;
723
724 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
725 * and error was received after socket entered established state.
726 * Hence, it is handled normally after connect() return successfully.
727 */
728
729 sock->state = SS_CONNECTED;
730 err = 0;
731out:
732 return err;
733
734sock_error:
735 err = sock_error(sk) ? : -ECONNABORTED;
736 sock->state = SS_UNCONNECTED;
737 sk->sk_disconnects++;
738 if (sk->sk_prot->disconnect(sk, flags))
739 sock->state = SS_DISCONNECTING;
740 goto out;
741}
742EXPORT_SYMBOL(__inet_stream_connect);
743
744int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
745 int addr_len, int flags)
746{
747 int err;
748
749 lock_sock(sock->sk);
750 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
751 release_sock(sock->sk);
752 return err;
753}
754EXPORT_SYMBOL(inet_stream_connect);
755
756void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk)
757{
758 sock_rps_record_flow(newsk);
759 WARN_ON(!((1 << newsk->sk_state) &
760 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
761 TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 |
762 TCPF_CLOSING | TCPF_CLOSE_WAIT |
763 TCPF_CLOSE)));
764
765 if (test_bit(SOCK_SUPPORT_ZC, &sock->flags))
766 set_bit(SOCK_SUPPORT_ZC, &newsock->flags);
767 sock_graft(newsk, newsock);
768
769 newsock->state = SS_CONNECTED;
770}
771
772/*
773 * Accept a pending connection. The TCP layer now gives BSD semantics.
774 */
775
776int inet_accept(struct socket *sock, struct socket *newsock, int flags,
777 bool kern)
778{
779 struct sock *sk1 = sock->sk, *sk2;
780 int err = -EINVAL;
781
782 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
783 sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern);
784 if (!sk2)
785 return err;
786
787 lock_sock(sk2);
788 __inet_accept(sock, newsock, sk2);
789 release_sock(sk2);
790 return 0;
791}
792EXPORT_SYMBOL(inet_accept);
793
794/*
795 * This does both peername and sockname.
796 */
797int inet_getname(struct socket *sock, struct sockaddr *uaddr,
798 int peer)
799{
800 struct sock *sk = sock->sk;
801 struct inet_sock *inet = inet_sk(sk);
802 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
803 int sin_addr_len = sizeof(*sin);
804
805 sin->sin_family = AF_INET;
806 lock_sock(sk);
807 if (peer) {
808 if (!inet->inet_dport ||
809 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
810 peer == 1)) {
811 release_sock(sk);
812 return -ENOTCONN;
813 }
814 sin->sin_port = inet->inet_dport;
815 sin->sin_addr.s_addr = inet->inet_daddr;
816 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
817 CGROUP_INET4_GETPEERNAME);
818 } else {
819 __be32 addr = inet->inet_rcv_saddr;
820 if (!addr)
821 addr = inet->inet_saddr;
822 sin->sin_port = inet->inet_sport;
823 sin->sin_addr.s_addr = addr;
824 BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len,
825 CGROUP_INET4_GETSOCKNAME);
826 }
827 release_sock(sk);
828 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
829 return sin_addr_len;
830}
831EXPORT_SYMBOL(inet_getname);
832
833int inet_send_prepare(struct sock *sk)
834{
835 sock_rps_record_flow(sk);
836
837 /* We may need to bind the socket. */
838 if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind &&
839 inet_autobind(sk))
840 return -EAGAIN;
841
842 return 0;
843}
844EXPORT_SYMBOL_GPL(inet_send_prepare);
845
846int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
847{
848 struct sock *sk = sock->sk;
849
850 if (unlikely(inet_send_prepare(sk)))
851 return -EAGAIN;
852
853 return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg,
854 sk, msg, size);
855}
856EXPORT_SYMBOL(inet_sendmsg);
857
858void inet_splice_eof(struct socket *sock)
859{
860 const struct proto *prot;
861 struct sock *sk = sock->sk;
862
863 if (unlikely(inet_send_prepare(sk)))
864 return;
865
866 /* IPV6_ADDRFORM can change sk->sk_prot under us. */
867 prot = READ_ONCE(sk->sk_prot);
868 if (prot->splice_eof)
869 prot->splice_eof(sock);
870}
871EXPORT_SYMBOL_GPL(inet_splice_eof);
872
873INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *,
874 size_t, int, int *));
875int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
876 int flags)
877{
878 struct sock *sk = sock->sk;
879 int addr_len = 0;
880 int err;
881
882 if (likely(!(flags & MSG_ERRQUEUE)))
883 sock_rps_record_flow(sk);
884
885 err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
886 sk, msg, size, flags, &addr_len);
887 if (err >= 0)
888 msg->msg_namelen = addr_len;
889 return err;
890}
891EXPORT_SYMBOL(inet_recvmsg);
892
893int inet_shutdown(struct socket *sock, int how)
894{
895 struct sock *sk = sock->sk;
896 int err = 0;
897
898 /* This should really check to make sure
899 * the socket is a TCP socket. (WHY AC...)
900 */
901 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
902 1->2 bit 2 snds.
903 2->3 */
904 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
905 return -EINVAL;
906
907 lock_sock(sk);
908 if (sock->state == SS_CONNECTING) {
909 if ((1 << sk->sk_state) &
910 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
911 sock->state = SS_DISCONNECTING;
912 else
913 sock->state = SS_CONNECTED;
914 }
915
916 switch (sk->sk_state) {
917 case TCP_CLOSE:
918 err = -ENOTCONN;
919 /* Hack to wake up other listeners, who can poll for
920 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
921 fallthrough;
922 default:
923 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how);
924 if (sk->sk_prot->shutdown)
925 sk->sk_prot->shutdown(sk, how);
926 break;
927
928 /* Remaining two branches are temporary solution for missing
929 * close() in multithreaded environment. It is _not_ a good idea,
930 * but we have no choice until close() is repaired at VFS level.
931 */
932 case TCP_LISTEN:
933 if (!(how & RCV_SHUTDOWN))
934 break;
935 fallthrough;
936 case TCP_SYN_SENT:
937 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
938 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
939 break;
940 }
941
942 /* Wake up anyone sleeping in poll. */
943 sk->sk_state_change(sk);
944 release_sock(sk);
945 return err;
946}
947EXPORT_SYMBOL(inet_shutdown);
948
949/*
950 * ioctl() calls you can issue on an INET socket. Most of these are
951 * device configuration and stuff and very rarely used. Some ioctls
952 * pass on to the socket itself.
953 *
954 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
955 * loads the devconfigure module does its configuring and unloads it.
956 * There's a good 20K of config code hanging around the kernel.
957 */
958
959int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
960{
961 struct sock *sk = sock->sk;
962 int err = 0;
963 struct net *net = sock_net(sk);
964 void __user *p = (void __user *)arg;
965 struct ifreq ifr;
966 struct rtentry rt;
967
968 switch (cmd) {
969 case SIOCADDRT:
970 case SIOCDELRT:
971 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
972 return -EFAULT;
973 err = ip_rt_ioctl(net, cmd, &rt);
974 break;
975 case SIOCRTMSG:
976 err = -EINVAL;
977 break;
978 case SIOCDARP:
979 case SIOCGARP:
980 case SIOCSARP:
981 err = arp_ioctl(net, cmd, (void __user *)arg);
982 break;
983 case SIOCGIFADDR:
984 case SIOCGIFBRDADDR:
985 case SIOCGIFNETMASK:
986 case SIOCGIFDSTADDR:
987 case SIOCGIFPFLAGS:
988 if (get_user_ifreq(&ifr, NULL, p))
989 return -EFAULT;
990 err = devinet_ioctl(net, cmd, &ifr);
991 if (!err && put_user_ifreq(&ifr, p))
992 err = -EFAULT;
993 break;
994
995 case SIOCSIFADDR:
996 case SIOCSIFBRDADDR:
997 case SIOCSIFNETMASK:
998 case SIOCSIFDSTADDR:
999 case SIOCSIFPFLAGS:
1000 case SIOCSIFFLAGS:
1001 if (get_user_ifreq(&ifr, NULL, p))
1002 return -EFAULT;
1003 err = devinet_ioctl(net, cmd, &ifr);
1004 break;
1005 default:
1006 if (sk->sk_prot->ioctl)
1007 err = sk_ioctl(sk, cmd, (void __user *)arg);
1008 else
1009 err = -ENOIOCTLCMD;
1010 break;
1011 }
1012 return err;
1013}
1014EXPORT_SYMBOL(inet_ioctl);
1015
1016#ifdef CONFIG_COMPAT
1017static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd,
1018 struct compat_rtentry __user *ur)
1019{
1020 compat_uptr_t rtdev;
1021 struct rtentry rt;
1022
1023 if (copy_from_user(&rt.rt_dst, &ur->rt_dst,
1024 3 * sizeof(struct sockaddr)) ||
1025 get_user(rt.rt_flags, &ur->rt_flags) ||
1026 get_user(rt.rt_metric, &ur->rt_metric) ||
1027 get_user(rt.rt_mtu, &ur->rt_mtu) ||
1028 get_user(rt.rt_window, &ur->rt_window) ||
1029 get_user(rt.rt_irtt, &ur->rt_irtt) ||
1030 get_user(rtdev, &ur->rt_dev))
1031 return -EFAULT;
1032
1033 rt.rt_dev = compat_ptr(rtdev);
1034 return ip_rt_ioctl(sock_net(sk), cmd, &rt);
1035}
1036
1037static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1038{
1039 void __user *argp = compat_ptr(arg);
1040 struct sock *sk = sock->sk;
1041
1042 switch (cmd) {
1043 case SIOCADDRT:
1044 case SIOCDELRT:
1045 return inet_compat_routing_ioctl(sk, cmd, argp);
1046 default:
1047 if (!sk->sk_prot->compat_ioctl)
1048 return -ENOIOCTLCMD;
1049 return sk->sk_prot->compat_ioctl(sk, cmd, arg);
1050 }
1051}
1052#endif /* CONFIG_COMPAT */
1053
1054const struct proto_ops inet_stream_ops = {
1055 .family = PF_INET,
1056 .owner = THIS_MODULE,
1057 .release = inet_release,
1058 .bind = inet_bind,
1059 .connect = inet_stream_connect,
1060 .socketpair = sock_no_socketpair,
1061 .accept = inet_accept,
1062 .getname = inet_getname,
1063 .poll = tcp_poll,
1064 .ioctl = inet_ioctl,
1065 .gettstamp = sock_gettstamp,
1066 .listen = inet_listen,
1067 .shutdown = inet_shutdown,
1068 .setsockopt = sock_common_setsockopt,
1069 .getsockopt = sock_common_getsockopt,
1070 .sendmsg = inet_sendmsg,
1071 .recvmsg = inet_recvmsg,
1072#ifdef CONFIG_MMU
1073 .mmap = tcp_mmap,
1074#endif
1075 .splice_eof = inet_splice_eof,
1076 .splice_read = tcp_splice_read,
1077 .read_sock = tcp_read_sock,
1078 .read_skb = tcp_read_skb,
1079 .sendmsg_locked = tcp_sendmsg_locked,
1080 .peek_len = tcp_peek_len,
1081#ifdef CONFIG_COMPAT
1082 .compat_ioctl = inet_compat_ioctl,
1083#endif
1084 .set_rcvlowat = tcp_set_rcvlowat,
1085};
1086EXPORT_SYMBOL(inet_stream_ops);
1087
1088const struct proto_ops inet_dgram_ops = {
1089 .family = PF_INET,
1090 .owner = THIS_MODULE,
1091 .release = inet_release,
1092 .bind = inet_bind,
1093 .connect = inet_dgram_connect,
1094 .socketpair = sock_no_socketpair,
1095 .accept = sock_no_accept,
1096 .getname = inet_getname,
1097 .poll = udp_poll,
1098 .ioctl = inet_ioctl,
1099 .gettstamp = sock_gettstamp,
1100 .listen = sock_no_listen,
1101 .shutdown = inet_shutdown,
1102 .setsockopt = sock_common_setsockopt,
1103 .getsockopt = sock_common_getsockopt,
1104 .sendmsg = inet_sendmsg,
1105 .read_skb = udp_read_skb,
1106 .recvmsg = inet_recvmsg,
1107 .mmap = sock_no_mmap,
1108 .splice_eof = inet_splice_eof,
1109 .set_peek_off = udp_set_peek_off,
1110#ifdef CONFIG_COMPAT
1111 .compat_ioctl = inet_compat_ioctl,
1112#endif
1113};
1114EXPORT_SYMBOL(inet_dgram_ops);
1115
1116/*
1117 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1118 * udp_poll
1119 */
1120static const struct proto_ops inet_sockraw_ops = {
1121 .family = PF_INET,
1122 .owner = THIS_MODULE,
1123 .release = inet_release,
1124 .bind = inet_bind,
1125 .connect = inet_dgram_connect,
1126 .socketpair = sock_no_socketpair,
1127 .accept = sock_no_accept,
1128 .getname = inet_getname,
1129 .poll = datagram_poll,
1130 .ioctl = inet_ioctl,
1131 .gettstamp = sock_gettstamp,
1132 .listen = sock_no_listen,
1133 .shutdown = inet_shutdown,
1134 .setsockopt = sock_common_setsockopt,
1135 .getsockopt = sock_common_getsockopt,
1136 .sendmsg = inet_sendmsg,
1137 .recvmsg = inet_recvmsg,
1138 .mmap = sock_no_mmap,
1139 .splice_eof = inet_splice_eof,
1140#ifdef CONFIG_COMPAT
1141 .compat_ioctl = inet_compat_ioctl,
1142#endif
1143};
1144
1145static const struct net_proto_family inet_family_ops = {
1146 .family = PF_INET,
1147 .create = inet_create,
1148 .owner = THIS_MODULE,
1149};
1150
1151/* Upon startup we insert all the elements in inetsw_array[] into
1152 * the linked list inetsw.
1153 */
1154static struct inet_protosw inetsw_array[] =
1155{
1156 {
1157 .type = SOCK_STREAM,
1158 .protocol = IPPROTO_TCP,
1159 .prot = &tcp_prot,
1160 .ops = &inet_stream_ops,
1161 .flags = INET_PROTOSW_PERMANENT |
1162 INET_PROTOSW_ICSK,
1163 },
1164
1165 {
1166 .type = SOCK_DGRAM,
1167 .protocol = IPPROTO_UDP,
1168 .prot = &udp_prot,
1169 .ops = &inet_dgram_ops,
1170 .flags = INET_PROTOSW_PERMANENT,
1171 },
1172
1173 {
1174 .type = SOCK_DGRAM,
1175 .protocol = IPPROTO_ICMP,
1176 .prot = &ping_prot,
1177 .ops = &inet_sockraw_ops,
1178 .flags = INET_PROTOSW_REUSE,
1179 },
1180
1181 {
1182 .type = SOCK_RAW,
1183 .protocol = IPPROTO_IP, /* wild card */
1184 .prot = &raw_prot,
1185 .ops = &inet_sockraw_ops,
1186 .flags = INET_PROTOSW_REUSE,
1187 }
1188};
1189
1190#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1191
1192void inet_register_protosw(struct inet_protosw *p)
1193{
1194 struct list_head *lh;
1195 struct inet_protosw *answer;
1196 int protocol = p->protocol;
1197 struct list_head *last_perm;
1198
1199 spin_lock_bh(&inetsw_lock);
1200
1201 if (p->type >= SOCK_MAX)
1202 goto out_illegal;
1203
1204 /* If we are trying to override a permanent protocol, bail. */
1205 last_perm = &inetsw[p->type];
1206 list_for_each(lh, &inetsw[p->type]) {
1207 answer = list_entry(lh, struct inet_protosw, list);
1208 /* Check only the non-wild match. */
1209 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1210 break;
1211 if (protocol == answer->protocol)
1212 goto out_permanent;
1213 last_perm = lh;
1214 }
1215
1216 /* Add the new entry after the last permanent entry if any, so that
1217 * the new entry does not override a permanent entry when matched with
1218 * a wild-card protocol. But it is allowed to override any existing
1219 * non-permanent entry. This means that when we remove this entry, the
1220 * system automatically returns to the old behavior.
1221 */
1222 list_add_rcu(&p->list, last_perm);
1223out:
1224 spin_unlock_bh(&inetsw_lock);
1225
1226 return;
1227
1228out_permanent:
1229 pr_err("Attempt to override permanent protocol %d\n", protocol);
1230 goto out;
1231
1232out_illegal:
1233 pr_err("Ignoring attempt to register invalid socket type %d\n",
1234 p->type);
1235 goto out;
1236}
1237EXPORT_SYMBOL(inet_register_protosw);
1238
1239void inet_unregister_protosw(struct inet_protosw *p)
1240{
1241 if (INET_PROTOSW_PERMANENT & p->flags) {
1242 pr_err("Attempt to unregister permanent protocol %d\n",
1243 p->protocol);
1244 } else {
1245 spin_lock_bh(&inetsw_lock);
1246 list_del_rcu(&p->list);
1247 spin_unlock_bh(&inetsw_lock);
1248
1249 synchronize_net();
1250 }
1251}
1252EXPORT_SYMBOL(inet_unregister_protosw);
1253
1254static int inet_sk_reselect_saddr(struct sock *sk)
1255{
1256 struct inet_sock *inet = inet_sk(sk);
1257 __be32 old_saddr = inet->inet_saddr;
1258 __be32 daddr = inet->inet_daddr;
1259 struct flowi4 *fl4;
1260 struct rtable *rt;
1261 __be32 new_saddr;
1262 struct ip_options_rcu *inet_opt;
1263 int err;
1264
1265 inet_opt = rcu_dereference_protected(inet->inet_opt,
1266 lockdep_sock_is_held(sk));
1267 if (inet_opt && inet_opt->opt.srr)
1268 daddr = inet_opt->opt.faddr;
1269
1270 /* Query new route. */
1271 fl4 = &inet->cork.fl.u.ip4;
1272 rt = ip_route_connect(fl4, daddr, 0, sk->sk_bound_dev_if,
1273 sk->sk_protocol, inet->inet_sport,
1274 inet->inet_dport, sk);
1275 if (IS_ERR(rt))
1276 return PTR_ERR(rt);
1277
1278 new_saddr = fl4->saddr;
1279
1280 if (new_saddr == old_saddr) {
1281 sk_setup_caps(sk, &rt->dst);
1282 return 0;
1283 }
1284
1285 err = inet_bhash2_update_saddr(sk, &new_saddr, AF_INET);
1286 if (err) {
1287 ip_rt_put(rt);
1288 return err;
1289 }
1290
1291 sk_setup_caps(sk, &rt->dst);
1292
1293 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) {
1294 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1295 __func__, &old_saddr, &new_saddr);
1296 }
1297
1298 /*
1299 * XXX The only one ugly spot where we need to
1300 * XXX really change the sockets identity after
1301 * XXX it has entered the hashes. -DaveM
1302 *
1303 * Besides that, it does not check for connection
1304 * uniqueness. Wait for troubles.
1305 */
1306 return __sk_prot_rehash(sk);
1307}
1308
1309int inet_sk_rebuild_header(struct sock *sk)
1310{
1311 struct rtable *rt = dst_rtable(__sk_dst_check(sk, 0));
1312 struct inet_sock *inet = inet_sk(sk);
1313 __be32 daddr;
1314 struct ip_options_rcu *inet_opt;
1315 struct flowi4 *fl4;
1316 int err;
1317
1318 /* Route is OK, nothing to do. */
1319 if (rt)
1320 return 0;
1321
1322 /* Reroute. */
1323 rcu_read_lock();
1324 inet_opt = rcu_dereference(inet->inet_opt);
1325 daddr = inet->inet_daddr;
1326 if (inet_opt && inet_opt->opt.srr)
1327 daddr = inet_opt->opt.faddr;
1328 rcu_read_unlock();
1329 fl4 = &inet->cork.fl.u.ip4;
1330 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1331 inet->inet_dport, inet->inet_sport,
1332 sk->sk_protocol, ip_sock_rt_tos(sk),
1333 sk->sk_bound_dev_if);
1334 if (!IS_ERR(rt)) {
1335 err = 0;
1336 sk_setup_caps(sk, &rt->dst);
1337 } else {
1338 err = PTR_ERR(rt);
1339
1340 /* Routing failed... */
1341 sk->sk_route_caps = 0;
1342 /*
1343 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1344 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1345 */
1346 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) ||
1347 sk->sk_state != TCP_SYN_SENT ||
1348 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1349 (err = inet_sk_reselect_saddr(sk)) != 0)
1350 WRITE_ONCE(sk->sk_err_soft, -err);
1351 }
1352
1353 return err;
1354}
1355EXPORT_SYMBOL(inet_sk_rebuild_header);
1356
1357void inet_sk_set_state(struct sock *sk, int state)
1358{
1359 trace_inet_sock_set_state(sk, sk->sk_state, state);
1360 sk->sk_state = state;
1361}
1362EXPORT_SYMBOL(inet_sk_set_state);
1363
1364void inet_sk_state_store(struct sock *sk, int newstate)
1365{
1366 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1367 smp_store_release(&sk->sk_state, newstate);
1368}
1369
1370struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1371 netdev_features_t features)
1372{
1373 bool udpfrag = false, fixedid = false, gso_partial, encap;
1374 struct sk_buff *segs = ERR_PTR(-EINVAL);
1375 const struct net_offload *ops;
1376 unsigned int offset = 0;
1377 struct iphdr *iph;
1378 int proto, tot_len;
1379 int nhoff;
1380 int ihl;
1381 int id;
1382
1383 skb_reset_network_header(skb);
1384 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1385 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1386 goto out;
1387
1388 iph = ip_hdr(skb);
1389 ihl = iph->ihl * 4;
1390 if (ihl < sizeof(*iph))
1391 goto out;
1392
1393 id = ntohs(iph->id);
1394 proto = iph->protocol;
1395
1396 /* Warning: after this point, iph might be no longer valid */
1397 if (unlikely(!pskb_may_pull(skb, ihl)))
1398 goto out;
1399 __skb_pull(skb, ihl);
1400
1401 encap = SKB_GSO_CB(skb)->encap_level > 0;
1402 if (encap)
1403 features &= skb->dev->hw_enc_features;
1404 SKB_GSO_CB(skb)->encap_level += ihl;
1405
1406 skb_reset_transport_header(skb);
1407
1408 segs = ERR_PTR(-EPROTONOSUPPORT);
1409
1410 if (!skb->encapsulation || encap) {
1411 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1412 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1413
1414 /* fixed ID is invalid if DF bit is not set */
1415 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1416 goto out;
1417 }
1418
1419 ops = rcu_dereference(inet_offloads[proto]);
1420 if (likely(ops && ops->callbacks.gso_segment)) {
1421 segs = ops->callbacks.gso_segment(skb, features);
1422 if (!segs)
1423 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
1424 }
1425
1426 if (IS_ERR_OR_NULL(segs))
1427 goto out;
1428
1429 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1430
1431 skb = segs;
1432 do {
1433 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1434 if (udpfrag) {
1435 iph->frag_off = htons(offset >> 3);
1436 if (skb->next)
1437 iph->frag_off |= htons(IP_MF);
1438 offset += skb->len - nhoff - ihl;
1439 tot_len = skb->len - nhoff;
1440 } else if (skb_is_gso(skb)) {
1441 if (!fixedid) {
1442 iph->id = htons(id);
1443 id += skb_shinfo(skb)->gso_segs;
1444 }
1445
1446 if (gso_partial)
1447 tot_len = skb_shinfo(skb)->gso_size +
1448 SKB_GSO_CB(skb)->data_offset +
1449 skb->head - (unsigned char *)iph;
1450 else
1451 tot_len = skb->len - nhoff;
1452 } else {
1453 if (!fixedid)
1454 iph->id = htons(id++);
1455 tot_len = skb->len - nhoff;
1456 }
1457 iph->tot_len = htons(tot_len);
1458 ip_send_check(iph);
1459 if (encap)
1460 skb_reset_inner_headers(skb);
1461 skb->network_header = (u8 *)iph - skb->head;
1462 skb_reset_mac_len(skb);
1463 } while ((skb = skb->next));
1464
1465out:
1466 return segs;
1467}
1468
1469static struct sk_buff *ipip_gso_segment(struct sk_buff *skb,
1470 netdev_features_t features)
1471{
1472 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
1473 return ERR_PTR(-EINVAL);
1474
1475 return inet_gso_segment(skb, features);
1476}
1477
1478struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb)
1479{
1480 const struct net_offload *ops;
1481 struct sk_buff *pp = NULL;
1482 const struct iphdr *iph;
1483 struct sk_buff *p;
1484 unsigned int hlen;
1485 unsigned int off;
1486 unsigned int id;
1487 int flush = 1;
1488 int proto;
1489
1490 off = skb_gro_offset(skb);
1491 hlen = off + sizeof(*iph);
1492 iph = skb_gro_header(skb, hlen, off);
1493 if (unlikely(!iph))
1494 goto out;
1495
1496 proto = iph->protocol;
1497
1498 ops = rcu_dereference(inet_offloads[proto]);
1499 if (!ops || !ops->callbacks.gro_receive)
1500 goto out;
1501
1502 if (*(u8 *)iph != 0x45)
1503 goto out;
1504
1505 if (ip_is_fragment(iph))
1506 goto out;
1507
1508 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1509 goto out;
1510
1511 NAPI_GRO_CB(skb)->proto = proto;
1512 id = ntohl(*(__be32 *)&iph->id);
1513 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1514 id >>= 16;
1515
1516 list_for_each_entry(p, head, list) {
1517 struct iphdr *iph2;
1518 u16 flush_id;
1519
1520 if (!NAPI_GRO_CB(p)->same_flow)
1521 continue;
1522
1523 iph2 = (struct iphdr *)(p->data + off);
1524 /* The above works because, with the exception of the top
1525 * (inner most) layer, we only aggregate pkts with the same
1526 * hdr length so all the hdrs we'll need to verify will start
1527 * at the same offset.
1528 */
1529 if ((iph->protocol ^ iph2->protocol) |
1530 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1531 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1532 NAPI_GRO_CB(p)->same_flow = 0;
1533 continue;
1534 }
1535
1536 /* All fields must match except length and checksum. */
1537 NAPI_GRO_CB(p)->flush |=
1538 (iph->ttl ^ iph2->ttl) |
1539 (iph->tos ^ iph2->tos) |
1540 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1541
1542 NAPI_GRO_CB(p)->flush |= flush;
1543
1544 /* We need to store of the IP ID check to be included later
1545 * when we can verify that this packet does in fact belong
1546 * to a given flow.
1547 */
1548 flush_id = (u16)(id - ntohs(iph2->id));
1549
1550 /* This bit of code makes it much easier for us to identify
1551 * the cases where we are doing atomic vs non-atomic IP ID
1552 * checks. Specifically an atomic check can return IP ID
1553 * values 0 - 0xFFFF, while a non-atomic check can only
1554 * return 0 or 0xFFFF.
1555 */
1556 if (!NAPI_GRO_CB(p)->is_atomic ||
1557 !(iph->frag_off & htons(IP_DF))) {
1558 flush_id ^= NAPI_GRO_CB(p)->count;
1559 flush_id = flush_id ? 0xFFFF : 0;
1560 }
1561
1562 /* If the previous IP ID value was based on an atomic
1563 * datagram we can overwrite the value and ignore it.
1564 */
1565 if (NAPI_GRO_CB(skb)->is_atomic)
1566 NAPI_GRO_CB(p)->flush_id = flush_id;
1567 else
1568 NAPI_GRO_CB(p)->flush_id |= flush_id;
1569 }
1570
1571 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1572 NAPI_GRO_CB(skb)->flush |= flush;
1573 skb_set_network_header(skb, off);
1574 /* The above will be needed by the transport layer if there is one
1575 * immediately following this IP hdr.
1576 */
1577 NAPI_GRO_CB(skb)->inner_network_offset = off;
1578
1579 /* Note : No need to call skb_gro_postpull_rcsum() here,
1580 * as we already checked checksum over ipv4 header was 0
1581 */
1582 skb_gro_pull(skb, sizeof(*iph));
1583 skb_set_transport_header(skb, skb_gro_offset(skb));
1584
1585 pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive,
1586 ops->callbacks.gro_receive, head, skb);
1587
1588out:
1589 skb_gro_flush_final(skb, pp, flush);
1590
1591 return pp;
1592}
1593
1594static struct sk_buff *ipip_gro_receive(struct list_head *head,
1595 struct sk_buff *skb)
1596{
1597 if (NAPI_GRO_CB(skb)->encap_mark) {
1598 NAPI_GRO_CB(skb)->flush = 1;
1599 return NULL;
1600 }
1601
1602 NAPI_GRO_CB(skb)->encap_mark = 1;
1603
1604 return inet_gro_receive(head, skb);
1605}
1606
1607#define SECONDS_PER_DAY 86400
1608
1609/* inet_current_timestamp - Return IP network timestamp
1610 *
1611 * Return milliseconds since midnight in network byte order.
1612 */
1613__be32 inet_current_timestamp(void)
1614{
1615 u32 secs;
1616 u32 msecs;
1617 struct timespec64 ts;
1618
1619 ktime_get_real_ts64(&ts);
1620
1621 /* Get secs since midnight. */
1622 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1623 /* Convert to msecs. */
1624 msecs = secs * MSEC_PER_SEC;
1625 /* Convert nsec to msec. */
1626 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1627
1628 /* Convert to network byte order. */
1629 return htonl(msecs);
1630}
1631EXPORT_SYMBOL(inet_current_timestamp);
1632
1633int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1634{
1635 unsigned int family = READ_ONCE(sk->sk_family);
1636
1637 if (family == AF_INET)
1638 return ip_recv_error(sk, msg, len, addr_len);
1639#if IS_ENABLED(CONFIG_IPV6)
1640 if (family == AF_INET6)
1641 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1642#endif
1643 return -EINVAL;
1644}
1645EXPORT_SYMBOL(inet_recv_error);
1646
1647int inet_gro_complete(struct sk_buff *skb, int nhoff)
1648{
1649 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1650 const struct net_offload *ops;
1651 __be16 totlen = iph->tot_len;
1652 int proto = iph->protocol;
1653 int err = -ENOSYS;
1654
1655 if (skb->encapsulation) {
1656 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1657 skb_set_inner_network_header(skb, nhoff);
1658 }
1659
1660 iph_set_totlen(iph, skb->len - nhoff);
1661 csum_replace2(&iph->check, totlen, iph->tot_len);
1662
1663 ops = rcu_dereference(inet_offloads[proto]);
1664 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1665 goto out;
1666
1667 /* Only need to add sizeof(*iph) to get to the next hdr below
1668 * because any hdr with option will have been flushed in
1669 * inet_gro_receive().
1670 */
1671 err = INDIRECT_CALL_2(ops->callbacks.gro_complete,
1672 tcp4_gro_complete, udp4_gro_complete,
1673 skb, nhoff + sizeof(*iph));
1674
1675out:
1676 return err;
1677}
1678
1679static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1680{
1681 skb->encapsulation = 1;
1682 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1683 return inet_gro_complete(skb, nhoff);
1684}
1685
1686int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1687 unsigned short type, unsigned char protocol,
1688 struct net *net)
1689{
1690 struct socket *sock;
1691 int rc = sock_create_kern(net, family, type, protocol, &sock);
1692
1693 if (rc == 0) {
1694 *sk = sock->sk;
1695 (*sk)->sk_allocation = GFP_ATOMIC;
1696 (*sk)->sk_use_task_frag = false;
1697 /*
1698 * Unhash it so that IP input processing does not even see it,
1699 * we do not wish this socket to see incoming packets.
1700 */
1701 (*sk)->sk_prot->unhash(*sk);
1702 }
1703 return rc;
1704}
1705EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1706
1707unsigned long snmp_fold_field(void __percpu *mib, int offt)
1708{
1709 unsigned long res = 0;
1710 int i;
1711
1712 for_each_possible_cpu(i)
1713 res += snmp_get_cpu_field(mib, i, offt);
1714 return res;
1715}
1716EXPORT_SYMBOL_GPL(snmp_fold_field);
1717
1718#if BITS_PER_LONG==32
1719
1720u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1721 size_t syncp_offset)
1722{
1723 void *bhptr;
1724 struct u64_stats_sync *syncp;
1725 u64 v;
1726 unsigned int start;
1727
1728 bhptr = per_cpu_ptr(mib, cpu);
1729 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1730 do {
1731 start = u64_stats_fetch_begin(syncp);
1732 v = *(((u64 *)bhptr) + offt);
1733 } while (u64_stats_fetch_retry(syncp, start));
1734
1735 return v;
1736}
1737EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1738
1739u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1740{
1741 u64 res = 0;
1742 int cpu;
1743
1744 for_each_possible_cpu(cpu) {
1745 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1746 }
1747 return res;
1748}
1749EXPORT_SYMBOL_GPL(snmp_fold_field64);
1750#endif
1751
1752#ifdef CONFIG_IP_MULTICAST
1753static const struct net_protocol igmp_protocol = {
1754 .handler = igmp_rcv,
1755};
1756#endif
1757
1758static const struct net_protocol icmp_protocol = {
1759 .handler = icmp_rcv,
1760 .err_handler = icmp_err,
1761 .no_policy = 1,
1762};
1763
1764static __net_init int ipv4_mib_init_net(struct net *net)
1765{
1766 int i;
1767
1768 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1769 if (!net->mib.tcp_statistics)
1770 goto err_tcp_mib;
1771 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1772 if (!net->mib.ip_statistics)
1773 goto err_ip_mib;
1774
1775 for_each_possible_cpu(i) {
1776 struct ipstats_mib *af_inet_stats;
1777 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1778 u64_stats_init(&af_inet_stats->syncp);
1779 }
1780
1781 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1782 if (!net->mib.net_statistics)
1783 goto err_net_mib;
1784 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1785 if (!net->mib.udp_statistics)
1786 goto err_udp_mib;
1787 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1788 if (!net->mib.udplite_statistics)
1789 goto err_udplite_mib;
1790 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1791 if (!net->mib.icmp_statistics)
1792 goto err_icmp_mib;
1793 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1794 GFP_KERNEL);
1795 if (!net->mib.icmpmsg_statistics)
1796 goto err_icmpmsg_mib;
1797
1798 tcp_mib_init(net);
1799 return 0;
1800
1801err_icmpmsg_mib:
1802 free_percpu(net->mib.icmp_statistics);
1803err_icmp_mib:
1804 free_percpu(net->mib.udplite_statistics);
1805err_udplite_mib:
1806 free_percpu(net->mib.udp_statistics);
1807err_udp_mib:
1808 free_percpu(net->mib.net_statistics);
1809err_net_mib:
1810 free_percpu(net->mib.ip_statistics);
1811err_ip_mib:
1812 free_percpu(net->mib.tcp_statistics);
1813err_tcp_mib:
1814 return -ENOMEM;
1815}
1816
1817static __net_exit void ipv4_mib_exit_net(struct net *net)
1818{
1819 kfree(net->mib.icmpmsg_statistics);
1820 free_percpu(net->mib.icmp_statistics);
1821 free_percpu(net->mib.udplite_statistics);
1822 free_percpu(net->mib.udp_statistics);
1823 free_percpu(net->mib.net_statistics);
1824 free_percpu(net->mib.ip_statistics);
1825 free_percpu(net->mib.tcp_statistics);
1826#ifdef CONFIG_MPTCP
1827 /* allocated on demand, see mptcp_init_sock() */
1828 free_percpu(net->mib.mptcp_statistics);
1829#endif
1830}
1831
1832static __net_initdata struct pernet_operations ipv4_mib_ops = {
1833 .init = ipv4_mib_init_net,
1834 .exit = ipv4_mib_exit_net,
1835};
1836
1837static int __init init_ipv4_mibs(void)
1838{
1839 return register_pernet_subsys(&ipv4_mib_ops);
1840}
1841
1842static __net_init int inet_init_net(struct net *net)
1843{
1844 /*
1845 * Set defaults for local port range
1846 */
1847 net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u;
1848
1849 seqlock_init(&net->ipv4.ping_group_range.lock);
1850 /*
1851 * Sane defaults - nobody may create ping sockets.
1852 * Boot scripts should set this to distro-specific group.
1853 */
1854 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1855 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1856
1857 /* Default values for sysctl-controlled parameters.
1858 * We set them here, in case sysctl is not compiled.
1859 */
1860 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1861 net->ipv4.sysctl_ip_fwd_update_priority = 1;
1862 net->ipv4.sysctl_ip_dynaddr = 0;
1863 net->ipv4.sysctl_ip_early_demux = 1;
1864 net->ipv4.sysctl_udp_early_demux = 1;
1865 net->ipv4.sysctl_tcp_early_demux = 1;
1866 net->ipv4.sysctl_nexthop_compat_mode = 1;
1867#ifdef CONFIG_SYSCTL
1868 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1869#endif
1870
1871 /* Some igmp sysctl, whose values are always used */
1872 net->ipv4.sysctl_igmp_max_memberships = 20;
1873 net->ipv4.sysctl_igmp_max_msf = 10;
1874 /* IGMP reports for link-local multicast groups are enabled by default */
1875 net->ipv4.sysctl_igmp_llm_reports = 1;
1876 net->ipv4.sysctl_igmp_qrv = 2;
1877
1878 net->ipv4.sysctl_fib_notify_on_flag_change = 0;
1879
1880 return 0;
1881}
1882
1883static __net_initdata struct pernet_operations af_inet_ops = {
1884 .init = inet_init_net,
1885};
1886
1887static int __init init_inet_pernet_ops(void)
1888{
1889 return register_pernet_subsys(&af_inet_ops);
1890}
1891
1892static int ipv4_proc_init(void);
1893
1894/*
1895 * IP protocol layer initialiser
1896 */
1897
1898
1899static const struct net_offload ipip_offload = {
1900 .callbacks = {
1901 .gso_segment = ipip_gso_segment,
1902 .gro_receive = ipip_gro_receive,
1903 .gro_complete = ipip_gro_complete,
1904 },
1905};
1906
1907static int __init ipip_offload_init(void)
1908{
1909 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1910}
1911
1912static int __init ipv4_offload_init(void)
1913{
1914 /*
1915 * Add offloads
1916 */
1917 if (udpv4_offload_init() < 0)
1918 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1919 if (tcpv4_offload_init() < 0)
1920 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1921 if (ipip_offload_init() < 0)
1922 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1923
1924 net_hotdata.ip_packet_offload = (struct packet_offload) {
1925 .type = cpu_to_be16(ETH_P_IP),
1926 .callbacks = {
1927 .gso_segment = inet_gso_segment,
1928 .gro_receive = inet_gro_receive,
1929 .gro_complete = inet_gro_complete,
1930 },
1931 };
1932 dev_add_offload(&net_hotdata.ip_packet_offload);
1933 return 0;
1934}
1935
1936fs_initcall(ipv4_offload_init);
1937
1938static struct packet_type ip_packet_type __read_mostly = {
1939 .type = cpu_to_be16(ETH_P_IP),
1940 .func = ip_rcv,
1941 .list_func = ip_list_rcv,
1942};
1943
1944static int __init inet_init(void)
1945{
1946 struct inet_protosw *q;
1947 struct list_head *r;
1948 int rc;
1949
1950 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1951
1952 raw_hashinfo_init(&raw_v4_hashinfo);
1953
1954 rc = proto_register(&tcp_prot, 1);
1955 if (rc)
1956 goto out;
1957
1958 rc = proto_register(&udp_prot, 1);
1959 if (rc)
1960 goto out_unregister_tcp_proto;
1961
1962 rc = proto_register(&raw_prot, 1);
1963 if (rc)
1964 goto out_unregister_udp_proto;
1965
1966 rc = proto_register(&ping_prot, 1);
1967 if (rc)
1968 goto out_unregister_raw_proto;
1969
1970 /*
1971 * Tell SOCKET that we are alive...
1972 */
1973
1974 (void)sock_register(&inet_family_ops);
1975
1976#ifdef CONFIG_SYSCTL
1977 ip_static_sysctl_init();
1978#endif
1979
1980 /*
1981 * Add all the base protocols.
1982 */
1983
1984 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1985 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1986
1987 net_hotdata.udp_protocol = (struct net_protocol) {
1988 .handler = udp_rcv,
1989 .err_handler = udp_err,
1990 .no_policy = 1,
1991 };
1992 if (inet_add_protocol(&net_hotdata.udp_protocol, IPPROTO_UDP) < 0)
1993 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1994
1995 net_hotdata.tcp_protocol = (struct net_protocol) {
1996 .handler = tcp_v4_rcv,
1997 .err_handler = tcp_v4_err,
1998 .no_policy = 1,
1999 .icmp_strict_tag_validation = 1,
2000 };
2001 if (inet_add_protocol(&net_hotdata.tcp_protocol, IPPROTO_TCP) < 0)
2002 pr_crit("%s: Cannot add TCP protocol\n", __func__);
2003#ifdef CONFIG_IP_MULTICAST
2004 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
2005 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
2006#endif
2007
2008 /* Register the socket-side information for inet_create. */
2009 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
2010 INIT_LIST_HEAD(r);
2011
2012 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
2013 inet_register_protosw(q);
2014
2015 /*
2016 * Set the ARP module up
2017 */
2018
2019 arp_init();
2020
2021 /*
2022 * Set the IP module up
2023 */
2024
2025 ip_init();
2026
2027 /* Initialise per-cpu ipv4 mibs */
2028 if (init_ipv4_mibs())
2029 panic("%s: Cannot init ipv4 mibs\n", __func__);
2030
2031 /* Setup TCP slab cache for open requests. */
2032 tcp_init();
2033
2034 /* Setup UDP memory threshold */
2035 udp_init();
2036
2037 /* Add UDP-Lite (RFC 3828) */
2038 udplite4_register();
2039
2040 raw_init();
2041
2042 ping_init();
2043
2044 /*
2045 * Set the ICMP layer up
2046 */
2047
2048 if (icmp_init() < 0)
2049 panic("Failed to create the ICMP control socket.\n");
2050
2051 /*
2052 * Initialise the multicast router
2053 */
2054#if defined(CONFIG_IP_MROUTE)
2055 if (ip_mr_init())
2056 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
2057#endif
2058
2059 if (init_inet_pernet_ops())
2060 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
2061
2062 ipv4_proc_init();
2063
2064 ipfrag_init();
2065
2066 dev_add_pack(&ip_packet_type);
2067
2068 ip_tunnel_core_init();
2069
2070 rc = 0;
2071out:
2072 return rc;
2073out_unregister_raw_proto:
2074 proto_unregister(&raw_prot);
2075out_unregister_udp_proto:
2076 proto_unregister(&udp_prot);
2077out_unregister_tcp_proto:
2078 proto_unregister(&tcp_prot);
2079 goto out;
2080}
2081
2082fs_initcall(inet_init);
2083
2084/* ------------------------------------------------------------------------ */
2085
2086#ifdef CONFIG_PROC_FS
2087static int __init ipv4_proc_init(void)
2088{
2089 int rc = 0;
2090
2091 if (raw_proc_init())
2092 goto out_raw;
2093 if (tcp4_proc_init())
2094 goto out_tcp;
2095 if (udp4_proc_init())
2096 goto out_udp;
2097 if (ping_proc_init())
2098 goto out_ping;
2099 if (ip_misc_proc_init())
2100 goto out_misc;
2101out:
2102 return rc;
2103out_misc:
2104 ping_proc_exit();
2105out_ping:
2106 udp4_proc_exit();
2107out_udp:
2108 tcp4_proc_exit();
2109out_tcp:
2110 raw_proc_exit();
2111out_raw:
2112 rc = -ENOMEM;
2113 goto out;
2114}
2115
2116#else /* CONFIG_PROC_FS */
2117static int __init ipv4_proc_init(void)
2118{
2119 return 0;
2120}
2121#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#include <linux/err.h>
69#include <linux/errno.h>
70#include <linux/types.h>
71#include <linux/socket.h>
72#include <linux/in.h>
73#include <linux/kernel.h>
74#include <linux/module.h>
75#include <linux/sched.h>
76#include <linux/timer.h>
77#include <linux/string.h>
78#include <linux/sockios.h>
79#include <linux/net.h>
80#include <linux/capability.h>
81#include <linux/fcntl.h>
82#include <linux/mm.h>
83#include <linux/interrupt.h>
84#include <linux/stat.h>
85#include <linux/init.h>
86#include <linux/poll.h>
87#include <linux/netfilter_ipv4.h>
88#include <linux/random.h>
89#include <linux/slab.h>
90
91#include <asm/uaccess.h>
92#include <asm/system.h>
93
94#include <linux/inet.h>
95#include <linux/igmp.h>
96#include <linux/inetdevice.h>
97#include <linux/netdevice.h>
98#include <net/checksum.h>
99#include <net/ip.h>
100#include <net/protocol.h>
101#include <net/arp.h>
102#include <net/route.h>
103#include <net/ip_fib.h>
104#include <net/inet_connection_sock.h>
105#include <net/tcp.h>
106#include <net/udp.h>
107#include <net/udplite.h>
108#include <net/ping.h>
109#include <linux/skbuff.h>
110#include <net/sock.h>
111#include <net/raw.h>
112#include <net/icmp.h>
113#include <net/ipip.h>
114#include <net/inet_common.h>
115#include <net/xfrm.h>
116#include <net/net_namespace.h>
117#ifdef CONFIG_IP_MROUTE
118#include <linux/mroute.h>
119#endif
120
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
128struct ipv4_config ipv4_config;
129EXPORT_SYMBOL(ipv4_config);
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 sk_refcnt_debug_dec(sk);
160}
161EXPORT_SYMBOL(inet_sock_destruct);
162
163/*
164 * The routines beyond this point handle the behaviour of an AF_INET
165 * socket object. Mostly it punts to the subprotocols of IP to do
166 * the work.
167 */
168
169/*
170 * Automatically bind an unbound socket.
171 */
172
173static int inet_autobind(struct sock *sk)
174{
175 struct inet_sock *inet;
176 /* We may need to bind the socket. */
177 lock_sock(sk);
178 inet = inet_sk(sk);
179 if (!inet->inet_num) {
180 if (sk->sk_prot->get_port(sk, 0)) {
181 release_sock(sk);
182 return -EAGAIN;
183 }
184 inet->inet_sport = htons(inet->inet_num);
185 }
186 release_sock(sk);
187 return 0;
188}
189
190/*
191 * Move a socket into listening state.
192 */
193int inet_listen(struct socket *sock, int backlog)
194{
195 struct sock *sk = sock->sk;
196 unsigned char old_state;
197 int err;
198
199 lock_sock(sk);
200
201 err = -EINVAL;
202 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
203 goto out;
204
205 old_state = sk->sk_state;
206 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
207 goto out;
208
209 /* Really, if the socket is already in listen state
210 * we can only allow the backlog to be adjusted.
211 */
212 if (old_state != TCP_LISTEN) {
213 err = inet_csk_listen_start(sk, backlog);
214 if (err)
215 goto out;
216 }
217 sk->sk_max_ack_backlog = backlog;
218 err = 0;
219
220out:
221 release_sock(sk);
222 return err;
223}
224EXPORT_SYMBOL(inet_listen);
225
226u32 inet_ehash_secret __read_mostly;
227EXPORT_SYMBOL(inet_ehash_secret);
228
229/*
230 * inet_ehash_secret must be set exactly once
231 */
232void build_ehash_secret(void)
233{
234 u32 rnd;
235
236 do {
237 get_random_bytes(&rnd, sizeof(rnd));
238 } while (rnd == 0);
239
240 cmpxchg(&inet_ehash_secret, 0, rnd);
241}
242EXPORT_SYMBOL(build_ehash_secret);
243
244static inline int inet_netns_ok(struct net *net, int protocol)
245{
246 int hash;
247 const struct net_protocol *ipprot;
248
249 if (net_eq(net, &init_net))
250 return 1;
251
252 hash = protocol & (MAX_INET_PROTOS - 1);
253 ipprot = rcu_dereference(inet_protos[hash]);
254
255 if (ipprot == NULL)
256 /* raw IP is OK */
257 return 1;
258 return ipprot->netns_ok;
259}
260
261/*
262 * Create an inet socket.
263 */
264
265static int inet_create(struct net *net, struct socket *sock, int protocol,
266 int kern)
267{
268 struct sock *sk;
269 struct inet_protosw *answer;
270 struct inet_sock *inet;
271 struct proto *answer_prot;
272 unsigned char answer_flags;
273 char answer_no_check;
274 int try_loading_module = 0;
275 int err;
276
277 if (unlikely(!inet_ehash_secret))
278 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
279 build_ehash_secret();
280
281 sock->state = SS_UNCONNECTED;
282
283 /* Look for the requested type/protocol pair. */
284lookup_protocol:
285 err = -ESOCKTNOSUPPORT;
286 rcu_read_lock();
287 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
288
289 err = 0;
290 /* Check the non-wild match. */
291 if (protocol == answer->protocol) {
292 if (protocol != IPPROTO_IP)
293 break;
294 } else {
295 /* Check for the two wild cases. */
296 if (IPPROTO_IP == protocol) {
297 protocol = answer->protocol;
298 break;
299 }
300 if (IPPROTO_IP == answer->protocol)
301 break;
302 }
303 err = -EPROTONOSUPPORT;
304 }
305
306 if (unlikely(err)) {
307 if (try_loading_module < 2) {
308 rcu_read_unlock();
309 /*
310 * Be more specific, e.g. net-pf-2-proto-132-type-1
311 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
312 */
313 if (++try_loading_module == 1)
314 request_module("net-pf-%d-proto-%d-type-%d",
315 PF_INET, protocol, sock->type);
316 /*
317 * Fall back to generic, e.g. net-pf-2-proto-132
318 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
319 */
320 else
321 request_module("net-pf-%d-proto-%d",
322 PF_INET, protocol);
323 goto lookup_protocol;
324 } else
325 goto out_rcu_unlock;
326 }
327
328 err = -EPERM;
329 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
330 goto out_rcu_unlock;
331
332 err = -EAFNOSUPPORT;
333 if (!inet_netns_ok(net, protocol))
334 goto out_rcu_unlock;
335
336 sock->ops = answer->ops;
337 answer_prot = answer->prot;
338 answer_no_check = answer->no_check;
339 answer_flags = answer->flags;
340 rcu_read_unlock();
341
342 WARN_ON(answer_prot->slab == NULL);
343
344 err = -ENOBUFS;
345 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
346 if (sk == NULL)
347 goto out;
348
349 err = 0;
350 sk->sk_no_check = answer_no_check;
351 if (INET_PROTOSW_REUSE & answer_flags)
352 sk->sk_reuse = 1;
353
354 inet = inet_sk(sk);
355 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
356
357 inet->nodefrag = 0;
358
359 if (SOCK_RAW == sock->type) {
360 inet->inet_num = protocol;
361 if (IPPROTO_RAW == protocol)
362 inet->hdrincl = 1;
363 }
364
365 if (ipv4_config.no_pmtu_disc)
366 inet->pmtudisc = IP_PMTUDISC_DONT;
367 else
368 inet->pmtudisc = IP_PMTUDISC_WANT;
369
370 inet->inet_id = 0;
371
372 sock_init_data(sock, sk);
373
374 sk->sk_destruct = inet_sock_destruct;
375 sk->sk_protocol = protocol;
376 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
377
378 inet->uc_ttl = -1;
379 inet->mc_loop = 1;
380 inet->mc_ttl = 1;
381 inet->mc_all = 1;
382 inet->mc_index = 0;
383 inet->mc_list = NULL;
384
385 sk_refcnt_debug_inc(sk);
386
387 if (inet->inet_num) {
388 /* It assumes that any protocol which allows
389 * the user to assign a number at socket
390 * creation time automatically
391 * shares.
392 */
393 inet->inet_sport = htons(inet->inet_num);
394 /* Add to protocol hash chains. */
395 sk->sk_prot->hash(sk);
396 }
397
398 if (sk->sk_prot->init) {
399 err = sk->sk_prot->init(sk);
400 if (err)
401 sk_common_release(sk);
402 }
403out:
404 return err;
405out_rcu_unlock:
406 rcu_read_unlock();
407 goto out;
408}
409
410
411/*
412 * The peer socket should always be NULL (or else). When we call this
413 * function we are destroying the object and from then on nobody
414 * should refer to it.
415 */
416int inet_release(struct socket *sock)
417{
418 struct sock *sk = sock->sk;
419
420 if (sk) {
421 long timeout;
422
423 sock_rps_reset_flow(sk);
424
425 /* Applications forget to leave groups before exiting */
426 ip_mc_drop_socket(sk);
427
428 /* If linger is set, we don't return until the close
429 * is complete. Otherwise we return immediately. The
430 * actually closing is done the same either way.
431 *
432 * If the close is due to the process exiting, we never
433 * linger..
434 */
435 timeout = 0;
436 if (sock_flag(sk, SOCK_LINGER) &&
437 !(current->flags & PF_EXITING))
438 timeout = sk->sk_lingertime;
439 sock->sk = NULL;
440 sk->sk_prot->close(sk, timeout);
441 }
442 return 0;
443}
444EXPORT_SYMBOL(inet_release);
445
446/* It is off by default, see below. */
447int sysctl_ip_nonlocal_bind __read_mostly;
448EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
449
450int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
451{
452 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
453 struct sock *sk = sock->sk;
454 struct inet_sock *inet = inet_sk(sk);
455 unsigned short snum;
456 int chk_addr_ret;
457 int err;
458
459 /* If the socket has its own bind function then use it. (RAW) */
460 if (sk->sk_prot->bind) {
461 err = sk->sk_prot->bind(sk, uaddr, addr_len);
462 goto out;
463 }
464 err = -EINVAL;
465 if (addr_len < sizeof(struct sockaddr_in))
466 goto out;
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 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
479
480 /* Not specified by any standard per-se, however it breaks too
481 * many applications when removed. It is unfortunate since
482 * allowing applications to make a non-local bind solves
483 * several problems with systems using dynamic addressing.
484 * (ie. your servers still start up even if your ISDN link
485 * is temporarily down)
486 */
487 err = -EADDRNOTAVAIL;
488 if (!sysctl_ip_nonlocal_bind &&
489 !(inet->freebind || inet->transparent) &&
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 < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
499 goto out;
500
501 /* We keep a pair of addresses. rcv_saddr is the one
502 * used by hash lookups, and saddr is used for transmit.
503 *
504 * In the BSD API these are the same except where it
505 * would be illegal to use them (multicast/broadcast) in
506 * which case the sending device address is used.
507 */
508 lock_sock(sk);
509
510 /* Check these errors (active socket, double bind). */
511 err = -EINVAL;
512 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
513 goto out_release_sock;
514
515 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
516 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
517 inet->inet_saddr = 0; /* Use device */
518
519 /* Make sure we are allowed to bind here. */
520 if (sk->sk_prot->get_port(sk, snum)) {
521 inet->inet_saddr = inet->inet_rcv_saddr = 0;
522 err = -EADDRINUSE;
523 goto out_release_sock;
524 }
525
526 if (inet->inet_rcv_saddr)
527 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
528 if (snum)
529 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
530 inet->inet_sport = htons(inet->inet_num);
531 inet->inet_daddr = 0;
532 inet->inet_dport = 0;
533 sk_dst_reset(sk);
534 err = 0;
535out_release_sock:
536 release_sock(sk);
537out:
538 return err;
539}
540EXPORT_SYMBOL(inet_bind);
541
542int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
543 int addr_len, int flags)
544{
545 struct sock *sk = sock->sk;
546
547 if (addr_len < sizeof(uaddr->sa_family))
548 return -EINVAL;
549 if (uaddr->sa_family == AF_UNSPEC)
550 return sk->sk_prot->disconnect(sk, flags);
551
552 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
553 return -EAGAIN;
554 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
555}
556EXPORT_SYMBOL(inet_dgram_connect);
557
558static long inet_wait_for_connect(struct sock *sk, long timeo)
559{
560 DEFINE_WAIT(wait);
561
562 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
563
564 /* Basic assumption: if someone sets sk->sk_err, he _must_
565 * change state of the socket from TCP_SYN_*.
566 * Connect() does not allow to get error notifications
567 * without closing the socket.
568 */
569 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
570 release_sock(sk);
571 timeo = schedule_timeout(timeo);
572 lock_sock(sk);
573 if (signal_pending(current) || !timeo)
574 break;
575 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
576 }
577 finish_wait(sk_sleep(sk), &wait);
578 return timeo;
579}
580
581/*
582 * Connect to a remote host. There is regrettably still a little
583 * TCP 'magic' in here.
584 */
585int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
586 int addr_len, int flags)
587{
588 struct sock *sk = sock->sk;
589 int err;
590 long timeo;
591
592 if (addr_len < sizeof(uaddr->sa_family))
593 return -EINVAL;
594
595 lock_sock(sk);
596
597 if (uaddr->sa_family == AF_UNSPEC) {
598 err = sk->sk_prot->disconnect(sk, flags);
599 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
600 goto out;
601 }
602
603 switch (sock->state) {
604 default:
605 err = -EINVAL;
606 goto out;
607 case SS_CONNECTED:
608 err = -EISCONN;
609 goto out;
610 case SS_CONNECTING:
611 err = -EALREADY;
612 /* Fall out of switch with err, set for this state */
613 break;
614 case SS_UNCONNECTED:
615 err = -EISCONN;
616 if (sk->sk_state != TCP_CLOSE)
617 goto out;
618
619 err = sk->sk_prot->connect(sk, uaddr, addr_len);
620 if (err < 0)
621 goto out;
622
623 sock->state = SS_CONNECTING;
624
625 /* Just entered SS_CONNECTING state; the only
626 * difference is that return value in non-blocking
627 * case is EINPROGRESS, rather than EALREADY.
628 */
629 err = -EINPROGRESS;
630 break;
631 }
632
633 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
634
635 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
636 /* Error code is set above */
637 if (!timeo || !inet_wait_for_connect(sk, timeo))
638 goto out;
639
640 err = sock_intr_errno(timeo);
641 if (signal_pending(current))
642 goto out;
643 }
644
645 /* Connection was closed by RST, timeout, ICMP error
646 * or another process disconnected us.
647 */
648 if (sk->sk_state == TCP_CLOSE)
649 goto sock_error;
650
651 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
652 * and error was received after socket entered established state.
653 * Hence, it is handled normally after connect() return successfully.
654 */
655
656 sock->state = SS_CONNECTED;
657 err = 0;
658out:
659 release_sock(sk);
660 return err;
661
662sock_error:
663 err = sock_error(sk) ? : -ECONNABORTED;
664 sock->state = SS_UNCONNECTED;
665 if (sk->sk_prot->disconnect(sk, flags))
666 sock->state = SS_DISCONNECTING;
667 goto out;
668}
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_CLOSE_WAIT | TCPF_CLOSE)));
689
690 sock_graft(sk2, newsock);
691
692 newsock->state = SS_CONNECTED;
693 err = 0;
694 release_sock(sk2);
695do_err:
696 return err;
697}
698EXPORT_SYMBOL(inet_accept);
699
700
701/*
702 * This does both peername and sockname.
703 */
704int inet_getname(struct socket *sock, struct sockaddr *uaddr,
705 int *uaddr_len, int peer)
706{
707 struct sock *sk = sock->sk;
708 struct inet_sock *inet = inet_sk(sk);
709 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
710
711 sin->sin_family = AF_INET;
712 if (peer) {
713 if (!inet->inet_dport ||
714 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
715 peer == 1))
716 return -ENOTCONN;
717 sin->sin_port = inet->inet_dport;
718 sin->sin_addr.s_addr = inet->inet_daddr;
719 } else {
720 __be32 addr = inet->inet_rcv_saddr;
721 if (!addr)
722 addr = inet->inet_saddr;
723 sin->sin_port = inet->inet_sport;
724 sin->sin_addr.s_addr = addr;
725 }
726 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
727 *uaddr_len = sizeof(*sin);
728 return 0;
729}
730EXPORT_SYMBOL(inet_getname);
731
732int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
733 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(iocb, 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 kiocb *iocb, struct socket *sock, struct msghdr *msg,
767 size_t size, 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(iocb, 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
896int 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#ifdef CONFIG_COMPAT
929 .compat_setsockopt = compat_sock_common_setsockopt,
930 .compat_getsockopt = compat_sock_common_getsockopt,
931 .compat_ioctl = inet_compat_ioctl,
932#endif
933};
934EXPORT_SYMBOL(inet_stream_ops);
935
936const struct proto_ops inet_dgram_ops = {
937 .family = PF_INET,
938 .owner = THIS_MODULE,
939 .release = inet_release,
940 .bind = inet_bind,
941 .connect = inet_dgram_connect,
942 .socketpair = sock_no_socketpair,
943 .accept = sock_no_accept,
944 .getname = inet_getname,
945 .poll = udp_poll,
946 .ioctl = inet_ioctl,
947 .listen = sock_no_listen,
948 .shutdown = inet_shutdown,
949 .setsockopt = sock_common_setsockopt,
950 .getsockopt = sock_common_getsockopt,
951 .sendmsg = inet_sendmsg,
952 .recvmsg = inet_recvmsg,
953 .mmap = sock_no_mmap,
954 .sendpage = inet_sendpage,
955#ifdef CONFIG_COMPAT
956 .compat_setsockopt = compat_sock_common_setsockopt,
957 .compat_getsockopt = compat_sock_common_getsockopt,
958 .compat_ioctl = inet_compat_ioctl,
959#endif
960};
961EXPORT_SYMBOL(inet_dgram_ops);
962
963/*
964 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
965 * udp_poll
966 */
967static const struct proto_ops inet_sockraw_ops = {
968 .family = PF_INET,
969 .owner = THIS_MODULE,
970 .release = inet_release,
971 .bind = inet_bind,
972 .connect = inet_dgram_connect,
973 .socketpair = sock_no_socketpair,
974 .accept = sock_no_accept,
975 .getname = inet_getname,
976 .poll = datagram_poll,
977 .ioctl = inet_ioctl,
978 .listen = sock_no_listen,
979 .shutdown = inet_shutdown,
980 .setsockopt = sock_common_setsockopt,
981 .getsockopt = sock_common_getsockopt,
982 .sendmsg = inet_sendmsg,
983 .recvmsg = inet_recvmsg,
984 .mmap = sock_no_mmap,
985 .sendpage = inet_sendpage,
986#ifdef CONFIG_COMPAT
987 .compat_setsockopt = compat_sock_common_setsockopt,
988 .compat_getsockopt = compat_sock_common_getsockopt,
989 .compat_ioctl = inet_compat_ioctl,
990#endif
991};
992
993static const struct net_proto_family inet_family_ops = {
994 .family = PF_INET,
995 .create = inet_create,
996 .owner = THIS_MODULE,
997};
998
999/* Upon startup we insert all the elements in inetsw_array[] into
1000 * the linked list inetsw.
1001 */
1002static struct inet_protosw inetsw_array[] =
1003{
1004 {
1005 .type = SOCK_STREAM,
1006 .protocol = IPPROTO_TCP,
1007 .prot = &tcp_prot,
1008 .ops = &inet_stream_ops,
1009 .no_check = 0,
1010 .flags = INET_PROTOSW_PERMANENT |
1011 INET_PROTOSW_ICSK,
1012 },
1013
1014 {
1015 .type = SOCK_DGRAM,
1016 .protocol = IPPROTO_UDP,
1017 .prot = &udp_prot,
1018 .ops = &inet_dgram_ops,
1019 .no_check = UDP_CSUM_DEFAULT,
1020 .flags = INET_PROTOSW_PERMANENT,
1021 },
1022
1023 {
1024 .type = SOCK_DGRAM,
1025 .protocol = IPPROTO_ICMP,
1026 .prot = &ping_prot,
1027 .ops = &inet_dgram_ops,
1028 .no_check = UDP_CSUM_DEFAULT,
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 .no_check = UDP_CSUM_DEFAULT,
1038 .flags = INET_PROTOSW_REUSE,
1039 }
1040};
1041
1042#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1043
1044void inet_register_protosw(struct inet_protosw *p)
1045{
1046 struct list_head *lh;
1047 struct inet_protosw *answer;
1048 int protocol = p->protocol;
1049 struct list_head *last_perm;
1050
1051 spin_lock_bh(&inetsw_lock);
1052
1053 if (p->type >= SOCK_MAX)
1054 goto out_illegal;
1055
1056 /* If we are trying to override a permanent protocol, bail. */
1057 answer = NULL;
1058 last_perm = &inetsw[p->type];
1059 list_for_each(lh, &inetsw[p->type]) {
1060 answer = list_entry(lh, struct inet_protosw, list);
1061
1062 /* Check only the non-wild match. */
1063 if (INET_PROTOSW_PERMANENT & answer->flags) {
1064 if (protocol == answer->protocol)
1065 break;
1066 last_perm = lh;
1067 }
1068
1069 answer = NULL;
1070 }
1071 if (answer)
1072 goto out_permanent;
1073
1074 /* Add the new entry after the last permanent entry if any, so that
1075 * the new entry does not override a permanent entry when matched with
1076 * a wild-card protocol. But it is allowed to override any existing
1077 * non-permanent entry. This means that when we remove this entry, the
1078 * system automatically returns to the old behavior.
1079 */
1080 list_add_rcu(&p->list, last_perm);
1081out:
1082 spin_unlock_bh(&inetsw_lock);
1083
1084 return;
1085
1086out_permanent:
1087 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1088 protocol);
1089 goto out;
1090
1091out_illegal:
1092 printk(KERN_ERR
1093 "Ignoring attempt to register invalid socket type %d.\n",
1094 p->type);
1095 goto out;
1096}
1097EXPORT_SYMBOL(inet_register_protosw);
1098
1099void inet_unregister_protosw(struct inet_protosw *p)
1100{
1101 if (INET_PROTOSW_PERMANENT & p->flags) {
1102 printk(KERN_ERR
1103 "Attempt to unregister permanent protocol %d.\n",
1104 p->protocol);
1105 } else {
1106 spin_lock_bh(&inetsw_lock);
1107 list_del_rcu(&p->list);
1108 spin_unlock_bh(&inetsw_lock);
1109
1110 synchronize_net();
1111 }
1112}
1113EXPORT_SYMBOL(inet_unregister_protosw);
1114
1115/*
1116 * Shall we try to damage output packets if routing dev changes?
1117 */
1118
1119int sysctl_ip_dynaddr __read_mostly;
1120
1121static int inet_sk_reselect_saddr(struct sock *sk)
1122{
1123 struct inet_sock *inet = inet_sk(sk);
1124 __be32 old_saddr = inet->inet_saddr;
1125 __be32 daddr = inet->inet_daddr;
1126 struct flowi4 *fl4;
1127 struct rtable *rt;
1128 __be32 new_saddr;
1129 struct ip_options_rcu *inet_opt;
1130
1131 inet_opt = rcu_dereference_protected(inet->inet_opt,
1132 sock_owned_by_user(sk));
1133 if (inet_opt && inet_opt->opt.srr)
1134 daddr = inet_opt->opt.faddr;
1135
1136 /* Query new route. */
1137 fl4 = &inet->cork.fl.u.ip4;
1138 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1139 sk->sk_bound_dev_if, sk->sk_protocol,
1140 inet->inet_sport, inet->inet_dport, sk, false);
1141 if (IS_ERR(rt))
1142 return PTR_ERR(rt);
1143
1144 sk_setup_caps(sk, &rt->dst);
1145
1146 new_saddr = fl4->saddr;
1147
1148 if (new_saddr == old_saddr)
1149 return 0;
1150
1151 if (sysctl_ip_dynaddr > 1) {
1152 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1153 __func__, &old_saddr, &new_saddr);
1154 }
1155
1156 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1157
1158 /*
1159 * XXX The only one ugly spot where we need to
1160 * XXX really change the sockets identity after
1161 * XXX it has entered the hashes. -DaveM
1162 *
1163 * Besides that, it does not check for connection
1164 * uniqueness. Wait for troubles.
1165 */
1166 __sk_prot_rehash(sk);
1167 return 0;
1168}
1169
1170int inet_sk_rebuild_header(struct sock *sk)
1171{
1172 struct inet_sock *inet = inet_sk(sk);
1173 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1174 __be32 daddr;
1175 struct ip_options_rcu *inet_opt;
1176 struct flowi4 *fl4;
1177 int err;
1178
1179 /* Route is OK, nothing to do. */
1180 if (rt)
1181 return 0;
1182
1183 /* Reroute. */
1184 rcu_read_lock();
1185 inet_opt = rcu_dereference(inet->inet_opt);
1186 daddr = inet->inet_daddr;
1187 if (inet_opt && inet_opt->opt.srr)
1188 daddr = inet_opt->opt.faddr;
1189 rcu_read_unlock();
1190 fl4 = &inet->cork.fl.u.ip4;
1191 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1192 inet->inet_dport, inet->inet_sport,
1193 sk->sk_protocol, RT_CONN_FLAGS(sk),
1194 sk->sk_bound_dev_if);
1195 if (!IS_ERR(rt)) {
1196 err = 0;
1197 sk_setup_caps(sk, &rt->dst);
1198 } else {
1199 err = PTR_ERR(rt);
1200
1201 /* Routing failed... */
1202 sk->sk_route_caps = 0;
1203 /*
1204 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1205 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1206 */
1207 if (!sysctl_ip_dynaddr ||
1208 sk->sk_state != TCP_SYN_SENT ||
1209 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1210 (err = inet_sk_reselect_saddr(sk)) != 0)
1211 sk->sk_err_soft = -err;
1212 }
1213
1214 return err;
1215}
1216EXPORT_SYMBOL(inet_sk_rebuild_header);
1217
1218static int inet_gso_send_check(struct sk_buff *skb)
1219{
1220 const struct iphdr *iph;
1221 const struct net_protocol *ops;
1222 int proto;
1223 int ihl;
1224 int err = -EINVAL;
1225
1226 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1227 goto out;
1228
1229 iph = ip_hdr(skb);
1230 ihl = iph->ihl * 4;
1231 if (ihl < sizeof(*iph))
1232 goto out;
1233
1234 if (unlikely(!pskb_may_pull(skb, ihl)))
1235 goto out;
1236
1237 __skb_pull(skb, ihl);
1238 skb_reset_transport_header(skb);
1239 iph = ip_hdr(skb);
1240 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1241 err = -EPROTONOSUPPORT;
1242
1243 rcu_read_lock();
1244 ops = rcu_dereference(inet_protos[proto]);
1245 if (likely(ops && ops->gso_send_check))
1246 err = ops->gso_send_check(skb);
1247 rcu_read_unlock();
1248
1249out:
1250 return err;
1251}
1252
1253static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1254{
1255 struct sk_buff *segs = ERR_PTR(-EINVAL);
1256 struct iphdr *iph;
1257 const struct net_protocol *ops;
1258 int proto;
1259 int ihl;
1260 int id;
1261 unsigned int offset = 0;
1262
1263 if (!(features & NETIF_F_V4_CSUM))
1264 features &= ~NETIF_F_SG;
1265
1266 if (unlikely(skb_shinfo(skb)->gso_type &
1267 ~(SKB_GSO_TCPV4 |
1268 SKB_GSO_UDP |
1269 SKB_GSO_DODGY |
1270 SKB_GSO_TCP_ECN |
1271 0)))
1272 goto out;
1273
1274 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1275 goto out;
1276
1277 iph = ip_hdr(skb);
1278 ihl = iph->ihl * 4;
1279 if (ihl < sizeof(*iph))
1280 goto out;
1281
1282 if (unlikely(!pskb_may_pull(skb, ihl)))
1283 goto out;
1284
1285 __skb_pull(skb, ihl);
1286 skb_reset_transport_header(skb);
1287 iph = ip_hdr(skb);
1288 id = ntohs(iph->id);
1289 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1290 segs = ERR_PTR(-EPROTONOSUPPORT);
1291
1292 rcu_read_lock();
1293 ops = rcu_dereference(inet_protos[proto]);
1294 if (likely(ops && ops->gso_segment))
1295 segs = ops->gso_segment(skb, features);
1296 rcu_read_unlock();
1297
1298 if (!segs || IS_ERR(segs))
1299 goto out;
1300
1301 skb = segs;
1302 do {
1303 iph = ip_hdr(skb);
1304 if (proto == IPPROTO_UDP) {
1305 iph->id = htons(id);
1306 iph->frag_off = htons(offset >> 3);
1307 if (skb->next != NULL)
1308 iph->frag_off |= htons(IP_MF);
1309 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1310 } else
1311 iph->id = htons(id++);
1312 iph->tot_len = htons(skb->len - skb->mac_len);
1313 iph->check = 0;
1314 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1315 } while ((skb = skb->next));
1316
1317out:
1318 return segs;
1319}
1320
1321static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1322 struct sk_buff *skb)
1323{
1324 const struct net_protocol *ops;
1325 struct sk_buff **pp = NULL;
1326 struct sk_buff *p;
1327 const struct iphdr *iph;
1328 unsigned int hlen;
1329 unsigned int off;
1330 unsigned int id;
1331 int flush = 1;
1332 int proto;
1333
1334 off = skb_gro_offset(skb);
1335 hlen = off + sizeof(*iph);
1336 iph = skb_gro_header_fast(skb, off);
1337 if (skb_gro_header_hard(skb, hlen)) {
1338 iph = skb_gro_header_slow(skb, hlen, off);
1339 if (unlikely(!iph))
1340 goto out;
1341 }
1342
1343 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1344
1345 rcu_read_lock();
1346 ops = rcu_dereference(inet_protos[proto]);
1347 if (!ops || !ops->gro_receive)
1348 goto out_unlock;
1349
1350 if (*(u8 *)iph != 0x45)
1351 goto out_unlock;
1352
1353 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1354 goto out_unlock;
1355
1356 id = ntohl(*(__be32 *)&iph->id);
1357 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1358 id >>= 16;
1359
1360 for (p = *head; p; p = p->next) {
1361 struct iphdr *iph2;
1362
1363 if (!NAPI_GRO_CB(p)->same_flow)
1364 continue;
1365
1366 iph2 = ip_hdr(p);
1367
1368 if ((iph->protocol ^ iph2->protocol) |
1369 (iph->tos ^ iph2->tos) |
1370 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1371 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1372 NAPI_GRO_CB(p)->same_flow = 0;
1373 continue;
1374 }
1375
1376 /* All fields must match except length and checksum. */
1377 NAPI_GRO_CB(p)->flush |=
1378 (iph->ttl ^ iph2->ttl) |
1379 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1380
1381 NAPI_GRO_CB(p)->flush |= flush;
1382 }
1383
1384 NAPI_GRO_CB(skb)->flush |= flush;
1385 skb_gro_pull(skb, sizeof(*iph));
1386 skb_set_transport_header(skb, skb_gro_offset(skb));
1387
1388 pp = ops->gro_receive(head, skb);
1389
1390out_unlock:
1391 rcu_read_unlock();
1392
1393out:
1394 NAPI_GRO_CB(skb)->flush |= flush;
1395
1396 return pp;
1397}
1398
1399static int inet_gro_complete(struct sk_buff *skb)
1400{
1401 const struct net_protocol *ops;
1402 struct iphdr *iph = ip_hdr(skb);
1403 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1404 int err = -ENOSYS;
1405 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1406
1407 csum_replace2(&iph->check, iph->tot_len, newlen);
1408 iph->tot_len = newlen;
1409
1410 rcu_read_lock();
1411 ops = rcu_dereference(inet_protos[proto]);
1412 if (WARN_ON(!ops || !ops->gro_complete))
1413 goto out_unlock;
1414
1415 err = ops->gro_complete(skb);
1416
1417out_unlock:
1418 rcu_read_unlock();
1419
1420 return err;
1421}
1422
1423int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1424 unsigned short type, unsigned char protocol,
1425 struct net *net)
1426{
1427 struct socket *sock;
1428 int rc = sock_create_kern(family, type, protocol, &sock);
1429
1430 if (rc == 0) {
1431 *sk = sock->sk;
1432 (*sk)->sk_allocation = GFP_ATOMIC;
1433 /*
1434 * Unhash it so that IP input processing does not even see it,
1435 * we do not wish this socket to see incoming packets.
1436 */
1437 (*sk)->sk_prot->unhash(*sk);
1438
1439 sk_change_net(*sk, net);
1440 }
1441 return rc;
1442}
1443EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1444
1445unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1446{
1447 unsigned long res = 0;
1448 int i, j;
1449
1450 for_each_possible_cpu(i) {
1451 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1452 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1453 }
1454 return res;
1455}
1456EXPORT_SYMBOL_GPL(snmp_fold_field);
1457
1458#if BITS_PER_LONG==32
1459
1460u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1461{
1462 u64 res = 0;
1463 int cpu;
1464
1465 for_each_possible_cpu(cpu) {
1466 void *bhptr;
1467 struct u64_stats_sync *syncp;
1468 u64 v;
1469 unsigned int start;
1470
1471 bhptr = per_cpu_ptr(mib[0], cpu);
1472 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1473 do {
1474 start = u64_stats_fetch_begin_bh(syncp);
1475 v = *(((u64 *) bhptr) + offt);
1476 } while (u64_stats_fetch_retry_bh(syncp, start));
1477
1478 res += v;
1479 }
1480 return res;
1481}
1482EXPORT_SYMBOL_GPL(snmp_fold_field64);
1483#endif
1484
1485int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1486{
1487 BUG_ON(ptr == NULL);
1488 ptr[0] = __alloc_percpu(mibsize, align);
1489 if (!ptr[0])
1490 return -ENOMEM;
1491#if SNMP_ARRAY_SZ == 2
1492 ptr[1] = __alloc_percpu(mibsize, align);
1493 if (!ptr[1]) {
1494 free_percpu(ptr[0]);
1495 ptr[0] = NULL;
1496 return -ENOMEM;
1497 }
1498#endif
1499 return 0;
1500}
1501EXPORT_SYMBOL_GPL(snmp_mib_init);
1502
1503void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1504{
1505 int i;
1506
1507 BUG_ON(ptr == NULL);
1508 for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1509 free_percpu(ptr[i]);
1510 ptr[i] = NULL;
1511 }
1512}
1513EXPORT_SYMBOL_GPL(snmp_mib_free);
1514
1515#ifdef CONFIG_IP_MULTICAST
1516static const struct net_protocol igmp_protocol = {
1517 .handler = igmp_rcv,
1518 .netns_ok = 1,
1519};
1520#endif
1521
1522static const struct net_protocol tcp_protocol = {
1523 .handler = tcp_v4_rcv,
1524 .err_handler = tcp_v4_err,
1525 .gso_send_check = tcp_v4_gso_send_check,
1526 .gso_segment = tcp_tso_segment,
1527 .gro_receive = tcp4_gro_receive,
1528 .gro_complete = tcp4_gro_complete,
1529 .no_policy = 1,
1530 .netns_ok = 1,
1531};
1532
1533static const struct net_protocol udp_protocol = {
1534 .handler = udp_rcv,
1535 .err_handler = udp_err,
1536 .gso_send_check = udp4_ufo_send_check,
1537 .gso_segment = udp4_ufo_fragment,
1538 .no_policy = 1,
1539 .netns_ok = 1,
1540};
1541
1542static const struct net_protocol icmp_protocol = {
1543 .handler = icmp_rcv,
1544 .err_handler = ping_err,
1545 .no_policy = 1,
1546 .netns_ok = 1,
1547};
1548
1549static __net_init int ipv4_mib_init_net(struct net *net)
1550{
1551 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1552 sizeof(struct tcp_mib),
1553 __alignof__(struct tcp_mib)) < 0)
1554 goto err_tcp_mib;
1555 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1556 sizeof(struct ipstats_mib),
1557 __alignof__(struct ipstats_mib)) < 0)
1558 goto err_ip_mib;
1559 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1560 sizeof(struct linux_mib),
1561 __alignof__(struct linux_mib)) < 0)
1562 goto err_net_mib;
1563 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1564 sizeof(struct udp_mib),
1565 __alignof__(struct udp_mib)) < 0)
1566 goto err_udp_mib;
1567 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1568 sizeof(struct udp_mib),
1569 __alignof__(struct udp_mib)) < 0)
1570 goto err_udplite_mib;
1571 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1572 sizeof(struct icmp_mib),
1573 __alignof__(struct icmp_mib)) < 0)
1574 goto err_icmp_mib;
1575 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1576 sizeof(struct icmpmsg_mib),
1577 __alignof__(struct icmpmsg_mib)) < 0)
1578 goto err_icmpmsg_mib;
1579
1580 tcp_mib_init(net);
1581 return 0;
1582
1583err_icmpmsg_mib:
1584 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1585err_icmp_mib:
1586 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1587err_udplite_mib:
1588 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1589err_udp_mib:
1590 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1591err_net_mib:
1592 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1593err_ip_mib:
1594 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1595err_tcp_mib:
1596 return -ENOMEM;
1597}
1598
1599static __net_exit void ipv4_mib_exit_net(struct net *net)
1600{
1601 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1602 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1603 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1604 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1605 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1606 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1607 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1608}
1609
1610static __net_initdata struct pernet_operations ipv4_mib_ops = {
1611 .init = ipv4_mib_init_net,
1612 .exit = ipv4_mib_exit_net,
1613};
1614
1615static int __init init_ipv4_mibs(void)
1616{
1617 return register_pernet_subsys(&ipv4_mib_ops);
1618}
1619
1620static int ipv4_proc_init(void);
1621
1622/*
1623 * IP protocol layer initialiser
1624 */
1625
1626static struct packet_type ip_packet_type __read_mostly = {
1627 .type = cpu_to_be16(ETH_P_IP),
1628 .func = ip_rcv,
1629 .gso_send_check = inet_gso_send_check,
1630 .gso_segment = inet_gso_segment,
1631 .gro_receive = inet_gro_receive,
1632 .gro_complete = inet_gro_complete,
1633};
1634
1635static int __init inet_init(void)
1636{
1637 struct sk_buff *dummy_skb;
1638 struct inet_protosw *q;
1639 struct list_head *r;
1640 int rc = -EINVAL;
1641
1642 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1643
1644 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1645 if (!sysctl_local_reserved_ports)
1646 goto out;
1647
1648 rc = proto_register(&tcp_prot, 1);
1649 if (rc)
1650 goto out_free_reserved_ports;
1651
1652 rc = proto_register(&udp_prot, 1);
1653 if (rc)
1654 goto out_unregister_tcp_proto;
1655
1656 rc = proto_register(&raw_prot, 1);
1657 if (rc)
1658 goto out_unregister_udp_proto;
1659
1660 rc = proto_register(&ping_prot, 1);
1661 if (rc)
1662 goto out_unregister_raw_proto;
1663
1664 /*
1665 * Tell SOCKET that we are alive...
1666 */
1667
1668 (void)sock_register(&inet_family_ops);
1669
1670#ifdef CONFIG_SYSCTL
1671 ip_static_sysctl_init();
1672#endif
1673
1674 /*
1675 * Add all the base protocols.
1676 */
1677
1678 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1679 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1680 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1681 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1682 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1683 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1684#ifdef CONFIG_IP_MULTICAST
1685 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1686 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1687#endif
1688
1689 /* Register the socket-side information for inet_create. */
1690 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1691 INIT_LIST_HEAD(r);
1692
1693 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1694 inet_register_protosw(q);
1695
1696 /*
1697 * Set the ARP module up
1698 */
1699
1700 arp_init();
1701
1702 /*
1703 * Set the IP module up
1704 */
1705
1706 ip_init();
1707
1708 tcp_v4_init();
1709
1710 /* Setup TCP slab cache for open requests. */
1711 tcp_init();
1712
1713 /* Setup UDP memory threshold */
1714 udp_init();
1715
1716 /* Add UDP-Lite (RFC 3828) */
1717 udplite4_register();
1718
1719 ping_init();
1720
1721 /*
1722 * Set the ICMP layer up
1723 */
1724
1725 if (icmp_init() < 0)
1726 panic("Failed to create the ICMP control socket.\n");
1727
1728 /*
1729 * Initialise the multicast router
1730 */
1731#if defined(CONFIG_IP_MROUTE)
1732 if (ip_mr_init())
1733 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1734#endif
1735 /*
1736 * Initialise per-cpu ipv4 mibs
1737 */
1738
1739 if (init_ipv4_mibs())
1740 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1741
1742 ipv4_proc_init();
1743
1744 ipfrag_init();
1745
1746 dev_add_pack(&ip_packet_type);
1747
1748 rc = 0;
1749out:
1750 return rc;
1751out_unregister_raw_proto:
1752 proto_unregister(&raw_prot);
1753out_unregister_udp_proto:
1754 proto_unregister(&udp_prot);
1755out_unregister_tcp_proto:
1756 proto_unregister(&tcp_prot);
1757out_free_reserved_ports:
1758 kfree(sysctl_local_reserved_ports);
1759 goto out;
1760}
1761
1762fs_initcall(inet_init);
1763
1764/* ------------------------------------------------------------------------ */
1765
1766#ifdef CONFIG_PROC_FS
1767static int __init ipv4_proc_init(void)
1768{
1769 int rc = 0;
1770
1771 if (raw_proc_init())
1772 goto out_raw;
1773 if (tcp4_proc_init())
1774 goto out_tcp;
1775 if (udp4_proc_init())
1776 goto out_udp;
1777 if (ping_proc_init())
1778 goto out_ping;
1779 if (ip_misc_proc_init())
1780 goto out_misc;
1781out:
1782 return rc;
1783out_misc:
1784 ping_proc_exit();
1785out_ping:
1786 udp4_proc_exit();
1787out_udp:
1788 tcp4_proc_exit();
1789out_tcp:
1790 raw_proc_exit();
1791out_raw:
1792 rc = -ENOMEM;
1793 goto out;
1794}
1795
1796#else /* CONFIG_PROC_FS */
1797static int __init ipv4_proc_init(void)
1798{
1799 return 0;
1800}
1801#endif /* CONFIG_PROC_FS */
1802
1803MODULE_ALIAS_NETPROTO(PF_INET);
1804