~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ipv4/af_inet.c

Version: ~ [ linux-5.10-rc1 ] ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
  3  *              operating system.  INET is implemented using the  BSD Socket
  4  *              interface as the means of communication with the user level.
  5  *
  6  *              PF_INET protocol family socket handler.
  7  *
  8  * Authors:     Ross Biro
  9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 10  *              Florian La Roche, <flla@stud.uni-sb.de>
 11  *              Alan Cox, <A.Cox@swansea.ac.uk>
 12  *
 13  * Changes (see also sock.c)
 14  *
 15  *              piggy,
 16  *              Karl Knutson    :       Socket protocol table
 17  *              A.N.Kuznetsov   :       Socket death error in accept().
 18  *              John Richardson :       Fix non blocking error in connect()
 19  *                                      so sockets that fail to connect
 20  *                                      don't return -EINPROGRESS.
 21  *              Alan Cox        :       Asynchronous I/O support
 22  *              Alan Cox        :       Keep correct socket pointer on sock
 23  *                                      structures
 24  *                                      when accept() ed
 25  *              Alan Cox        :       Semantics of SO_LINGER aren't state
 26  *                                      moved to close when you look carefully.
 27  *                                      With this fixed and the accept bug fixed
 28  *                                      some RPC stuff seems happier.
 29  *              Niibe Yutaka    :       4.4BSD style write async I/O
 30  *              Alan Cox,
 31  *              Tony Gale       :       Fixed reuse semantics.
 32  *              Alan Cox        :       bind() shouldn't abort existing but dead
 33  *                                      sockets. Stops FTP netin:.. I hope.
 34  *              Alan Cox        :       bind() works correctly for RAW sockets.
 35  *                                      Note that FreeBSD at least was broken
 36  *                                      in this respect so be careful with
 37  *                                      compatibility tests...
 38  *              Alan Cox        :       routing cache support
 39  *              Alan Cox        :       memzero the socket structure for
 40  *                                      compactness.
 41  *              Matt Day        :       nonblock connect error handler
 42  *              Alan Cox        :       Allow large numbers of pending sockets
 43  *                                      (eg for big web sites), but only if
 44  *                                      specifically application requested.
 45  *              Alan Cox        :       New buffering throughout IP. Used
 46  *                                      dumbly.
 47  *              Alan Cox        :       New buffering now used smartly.
 48  *              Alan Cox        :       BSD rather than common sense
 49  *                                      interpretation of listen.
 50  *              Germano Caronni :       Assorted small races.
 51  *              Alan Cox        :       sendmsg/recvmsg basic support.
 52  *              Alan Cox        :       Only sendmsg/recvmsg now supported.
 53  *              Alan Cox        :       Locked down bind (see security list).
 54  *              Alan Cox        :       Loosened bind a little.
 55  *              Mike McLagan    :       ADD/DEL DLCI Ioctls
 56  *      Willy Konynenberg       :       Transparent proxying support.
 57  *              David S. Miller :       New socket lookup architecture.
 58  *                                      Some other random speedups.
 59  *              Cyrus Durgin    :       Cleaned up file for kmod hacks.
 60  *              Andi Kleen      :       Fix inet_stream_connect TCP race.
 61  *
 62  *              This program is free software; you can redistribute it and/or
 63  *              modify it under the terms of the GNU General Public License
 64  *              as published by the Free Software Foundation; either version
 65  *              2 of the License, or (at your option) any later version.
 66  */
 67 
 68 #define pr_fmt(fmt) "IPv4: " fmt
 69 
 70 #include <linux/err.h>
 71 #include <linux/errno.h>
 72 #include <linux/types.h>
 73 #include <linux/socket.h>
 74 #include <linux/in.h>
 75 #include <linux/kernel.h>
 76 #include <linux/module.h>
 77 #include <linux/sched.h>
 78 #include <linux/timer.h>
 79 #include <linux/string.h>
 80 #include <linux/sockios.h>
 81 #include <linux/net.h>
 82 #include <linux/capability.h>
 83 #include <linux/fcntl.h>
 84 #include <linux/mm.h>
 85 #include <linux/interrupt.h>
 86 #include <linux/stat.h>
 87 #include <linux/init.h>
 88 #include <linux/poll.h>
 89 #include <linux/netfilter_ipv4.h>
 90 #include <linux/random.h>
 91 #include <linux/slab.h>
 92 
 93 #include <asm/uaccess.h>
 94 
 95 #include <linux/inet.h>
 96 #include <linux/igmp.h>
 97 #include <linux/inetdevice.h>
 98 #include <linux/netdevice.h>
 99 #include <net/checksum.h>
100 #include <net/ip.h>
101 #include <net/protocol.h>
102 #include <net/arp.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
106 #include <net/tcp.h>
107 #include <net/udp.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
112 #include <net/raw.h>
113 #include <net/icmp.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #include <net/secure_seq.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
120 #endif
121 
122 
123 /* The inetsw table contains everything that inet_create needs to
124  * build a new socket.
125  */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128 
129 struct ipv4_config ipv4_config;
130 EXPORT_SYMBOL(ipv4_config);
131 
132 /* New destruction routine */
133 
134 void inet_sock_destruct(struct sock *sk)
135 {
136         struct inet_sock *inet = inet_sk(sk);
137 
138         __skb_queue_purge(&sk->sk_receive_queue);
139         __skb_queue_purge(&sk->sk_error_queue);
140 
141         sk_mem_reclaim(sk);
142 
143         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144                 pr_err("Attempt to release TCP socket in state %d %p\n",
145                        sk->sk_state, sk);
146                 return;
147         }
148         if (!sock_flag(sk, SOCK_DEAD)) {
149                 pr_err("Attempt to release alive inet socket %p\n", sk);
150                 return;
151         }
152 
153         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
155         WARN_ON(sk->sk_wmem_queued);
156         WARN_ON(sk->sk_forward_alloc);
157 
158         kfree(rcu_dereference_protected(inet->inet_opt, 1));
159         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
160         dst_release(sk->sk_rx_dst);
161         sk_refcnt_debug_dec(sk);
162 }
163 EXPORT_SYMBOL(inet_sock_destruct);
164 
165 /*
166  *      The routines beyond this point handle the behaviour of an AF_INET
167  *      socket object. Mostly it punts to the subprotocols of IP to do
168  *      the work.
169  */
170 
171 /*
172  *      Automatically bind an unbound socket.
173  */
174 
175 static int inet_autobind(struct sock *sk)
176 {
177         struct inet_sock *inet;
178         /* We may need to bind the socket. */
179         lock_sock(sk);
180         inet = inet_sk(sk);
181         if (!inet->inet_num) {
182                 if (sk->sk_prot->get_port(sk, 0)) {
183                         release_sock(sk);
184                         return -EAGAIN;
185                 }
186                 inet->inet_sport = htons(inet->inet_num);
187         }
188         release_sock(sk);
189         return 0;
190 }
191 
192 /*
193  *      Move a socket into listening state.
194  */
195 int inet_listen(struct socket *sock, int backlog)
196 {
197         struct sock *sk = sock->sk;
198         unsigned char old_state;
199         int err;
200 
201         lock_sock(sk);
202 
203         err = -EINVAL;
204         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205                 goto out;
206 
207         old_state = sk->sk_state;
208         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209                 goto out;
210 
211         /* Really, if the socket is already in listen state
212          * we can only allow the backlog to be adjusted.
213          */
214         if (old_state != TCP_LISTEN) {
215                 /* Check special setups for testing purpose to enable TFO w/o
216                  * requiring TCP_FASTOPEN sockopt.
217                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
218                  * Also fastopenq may already been allocated because this
219                  * socket was in TCP_LISTEN state previously but was
220                  * shutdown() (rather than close()).
221                  */
222                 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
223                     inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
224                         if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
225                                 err = fastopen_init_queue(sk, backlog);
226                         else if ((sysctl_tcp_fastopen &
227                                   TFO_SERVER_WO_SOCKOPT2) != 0)
228                                 err = fastopen_init_queue(sk,
229                                     ((uint)sysctl_tcp_fastopen) >> 16);
230                         else
231                                 err = 0;
232                         if (err)
233                                 goto out;
234                 }
235                 err = inet_csk_listen_start(sk, backlog);
236                 if (err)
237                         goto out;
238         }
239         sk->sk_max_ack_backlog = backlog;
240         err = 0;
241 
242 out:
243         release_sock(sk);
244         return err;
245 }
246 EXPORT_SYMBOL(inet_listen);
247 
248 u32 inet_ehash_secret __read_mostly;
249 EXPORT_SYMBOL(inet_ehash_secret);
250 
251 u32 ipv6_hash_secret __read_mostly;
252 EXPORT_SYMBOL(ipv6_hash_secret);
253 
254 /*
255  * inet_ehash_secret must be set exactly once, and to a non nul value
256  * ipv6_hash_secret must be set exactly once.
257  */
258 void build_ehash_secret(void)
259 {
260         u32 rnd;
261 
262         do {
263                 get_random_bytes(&rnd, sizeof(rnd));
264         } while (rnd == 0);
265 
266         if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
267                 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
268 }
269 EXPORT_SYMBOL(build_ehash_secret);
270 
271 /*
272  *      Create an inet socket.
273  */
274 
275 static int inet_create(struct net *net, struct socket *sock, int protocol,
276                        int kern)
277 {
278         struct sock *sk;
279         struct inet_protosw *answer;
280         struct inet_sock *inet;
281         struct proto *answer_prot;
282         unsigned char answer_flags;
283         char answer_no_check;
284         int try_loading_module = 0;
285         int err;
286 
287         if (unlikely(!inet_ehash_secret))
288                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
289                         build_ehash_secret();
290 
291         if (protocol < 0 || protocol >= IPPROTO_MAX)
292                 return -EINVAL;
293 
294         sock->state = SS_UNCONNECTED;
295 
296         /* Look for the requested type/protocol pair. */
297 lookup_protocol:
298         err = -ESOCKTNOSUPPORT;
299         rcu_read_lock();
300         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
301 
302                 err = 0;
303                 /* Check the non-wild match. */
304                 if (protocol == answer->protocol) {
305                         if (protocol != IPPROTO_IP)
306                                 break;
307                 } else {
308                         /* Check for the two wild cases. */
309                         if (IPPROTO_IP == protocol) {
310                                 protocol = answer->protocol;
311                                 break;
312                         }
313                         if (IPPROTO_IP == answer->protocol)
314                                 break;
315                 }
316                 err = -EPROTONOSUPPORT;
317         }
318 
319         if (unlikely(err)) {
320                 if (try_loading_module < 2) {
321                         rcu_read_unlock();
322                         /*
323                          * Be more specific, e.g. net-pf-2-proto-132-type-1
324                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
325                          */
326                         if (++try_loading_module == 1)
327                                 request_module("net-pf-%d-proto-%d-type-%d",
328                                                PF_INET, protocol, sock->type);
329                         /*
330                          * Fall back to generic, e.g. net-pf-2-proto-132
331                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
332                          */
333                         else
334                                 request_module("net-pf-%d-proto-%d",
335                                                PF_INET, protocol);
336                         goto lookup_protocol;
337                 } else
338                         goto out_rcu_unlock;
339         }
340 
341         err = -EPERM;
342         if (sock->type == SOCK_RAW && !kern &&
343             !ns_capable(net->user_ns, CAP_NET_RAW))
344                 goto out_rcu_unlock;
345 
346         sock->ops = answer->ops;
347         answer_prot = answer->prot;
348         answer_no_check = answer->no_check;
349         answer_flags = answer->flags;
350         rcu_read_unlock();
351 
352         WARN_ON(answer_prot->slab == NULL);
353 
354         err = -ENOBUFS;
355         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
356         if (sk == NULL)
357                 goto out;
358 
359         err = 0;
360         sk->sk_no_check = answer_no_check;
361         if (INET_PROTOSW_REUSE & answer_flags)
362                 sk->sk_reuse = SK_CAN_REUSE;
363 
364         inet = inet_sk(sk);
365         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
366 
367         inet->nodefrag = 0;
368 
369         if (SOCK_RAW == sock->type) {
370                 inet->inet_num = protocol;
371                 if (IPPROTO_RAW == protocol)
372                         inet->hdrincl = 1;
373         }
374 
375         if (ipv4_config.no_pmtu_disc)
376                 inet->pmtudisc = IP_PMTUDISC_DONT;
377         else
378                 inet->pmtudisc = IP_PMTUDISC_WANT;
379 
380         inet->inet_id = 0;
381 
382         sock_init_data(sock, sk);
383 
384         sk->sk_destruct    = inet_sock_destruct;
385         sk->sk_protocol    = protocol;
386         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
387 
388         inet->uc_ttl    = -1;
389         inet->mc_loop   = 1;
390         inet->mc_ttl    = 1;
391         inet->mc_all    = 1;
392         inet->mc_index  = 0;
393         inet->mc_list   = NULL;
394         inet->rcv_tos   = 0;
395 
396         sk_refcnt_debug_inc(sk);
397 
398         if (inet->inet_num) {
399                 /* It assumes that any protocol which allows
400                  * the user to assign a number at socket
401                  * creation time automatically
402                  * shares.
403                  */
404                 inet->inet_sport = htons(inet->inet_num);
405                 /* Add to protocol hash chains. */
406                 sk->sk_prot->hash(sk);
407         }
408 
409         if (sk->sk_prot->init) {
410                 err = sk->sk_prot->init(sk);
411                 if (err)
412                         sk_common_release(sk);
413         }
414 out:
415         return err;
416 out_rcu_unlock:
417         rcu_read_unlock();
418         goto out;
419 }
420 
421 
422 /*
423  *      The peer socket should always be NULL (or else). When we call this
424  *      function we are destroying the object and from then on nobody
425  *      should refer to it.
426  */
427 int inet_release(struct socket *sock)
428 {
429         struct sock *sk = sock->sk;
430 
431         if (sk) {
432                 long timeout;
433 
434                 sock_rps_reset_flow(sk);
435 
436                 /* Applications forget to leave groups before exiting */
437                 ip_mc_drop_socket(sk);
438 
439                 /* If linger is set, we don't return until the close
440                  * is complete.  Otherwise we return immediately. The
441                  * actually closing is done the same either way.
442                  *
443                  * If the close is due to the process exiting, we never
444                  * linger..
445                  */
446                 timeout = 0;
447                 if (sock_flag(sk, SOCK_LINGER) &&
448                     !(current->flags & PF_EXITING))
449                         timeout = sk->sk_lingertime;
450                 sock->sk = NULL;
451                 sk->sk_prot->close(sk, timeout);
452         }
453         return 0;
454 }
455 EXPORT_SYMBOL(inet_release);
456 
457 /* It is off by default, see below. */
458 int sysctl_ip_nonlocal_bind __read_mostly;
459 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
460 
461 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
462 {
463         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
464         struct sock *sk = sock->sk;
465         struct inet_sock *inet = inet_sk(sk);
466         struct net *net = sock_net(sk);
467         unsigned short snum;
468         int chk_addr_ret;
469         int err;
470 
471         /* If the socket has its own bind function then use it. (RAW) */
472         if (sk->sk_prot->bind) {
473                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
474                 goto out;
475         }
476         err = -EINVAL;
477         if (addr_len < sizeof(struct sockaddr_in))
478                 goto out;
479 
480         if (addr->sin_family != AF_INET) {
481                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
482                  * only if s_addr is INADDR_ANY.
483                  */
484                 err = -EAFNOSUPPORT;
485                 if (addr->sin_family != AF_UNSPEC ||
486                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
487                         goto out;
488         }
489 
490         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
491 
492         /* Not specified by any standard per-se, however it breaks too
493          * many applications when removed.  It is unfortunate since
494          * allowing applications to make a non-local bind solves
495          * several problems with systems using dynamic addressing.
496          * (ie. your servers still start up even if your ISDN link
497          *  is temporarily down)
498          */
499         err = -EADDRNOTAVAIL;
500         if (!sysctl_ip_nonlocal_bind &&
501             !(inet->freebind || inet->transparent) &&
502             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
503             chk_addr_ret != RTN_LOCAL &&
504             chk_addr_ret != RTN_MULTICAST &&
505             chk_addr_ret != RTN_BROADCAST)
506                 goto out;
507 
508         snum = ntohs(addr->sin_port);
509         err = -EACCES;
510         if (snum && snum < PROT_SOCK &&
511             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
512                 goto out;
513 
514         /*      We keep a pair of addresses. rcv_saddr is the one
515          *      used by hash lookups, and saddr is used for transmit.
516          *
517          *      In the BSD API these are the same except where it
518          *      would be illegal to use them (multicast/broadcast) in
519          *      which case the sending device address is used.
520          */
521         lock_sock(sk);
522 
523         /* Check these errors (active socket, double bind). */
524         err = -EINVAL;
525         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
526                 goto out_release_sock;
527 
528         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
529         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
530                 inet->inet_saddr = 0;  /* Use device */
531 
532         /* Make sure we are allowed to bind here. */
533         if (sk->sk_prot->get_port(sk, snum)) {
534                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
535                 err = -EADDRINUSE;
536                 goto out_release_sock;
537         }
538 
539         if (inet->inet_rcv_saddr)
540                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
541         if (snum)
542                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
543         inet->inet_sport = htons(inet->inet_num);
544         inet->inet_daddr = 0;
545         inet->inet_dport = 0;
546         sk_dst_reset(sk);
547         err = 0;
548 out_release_sock:
549         release_sock(sk);
550 out:
551         return err;
552 }
553 EXPORT_SYMBOL(inet_bind);
554 
555 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
556                        int addr_len, int flags)
557 {
558         struct sock *sk = sock->sk;
559 
560         if (addr_len < sizeof(uaddr->sa_family))
561                 return -EINVAL;
562         if (uaddr->sa_family == AF_UNSPEC)
563                 return sk->sk_prot->disconnect(sk, flags);
564 
565         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
566                 return -EAGAIN;
567         return sk->sk_prot->connect(sk, uaddr, addr_len);
568 }
569 EXPORT_SYMBOL(inet_dgram_connect);
570 
571 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
572 {
573         DEFINE_WAIT(wait);
574 
575         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
576         sk->sk_write_pending += writebias;
577 
578         /* Basic assumption: if someone sets sk->sk_err, he _must_
579          * change state of the socket from TCP_SYN_*.
580          * Connect() does not allow to get error notifications
581          * without closing the socket.
582          */
583         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
584                 release_sock(sk);
585                 timeo = schedule_timeout(timeo);
586                 lock_sock(sk);
587                 if (signal_pending(current) || !timeo)
588                         break;
589                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
590         }
591         finish_wait(sk_sleep(sk), &wait);
592         sk->sk_write_pending -= writebias;
593         return timeo;
594 }
595 
596 /*
597  *      Connect to a remote host. There is regrettably still a little
598  *      TCP 'magic' in here.
599  */
600 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
601                           int addr_len, int flags)
602 {
603         struct sock *sk = sock->sk;
604         int err;
605         long timeo;
606 
607         if (addr_len < sizeof(uaddr->sa_family))
608                 return -EINVAL;
609 
610         if (uaddr->sa_family == AF_UNSPEC) {
611                 err = sk->sk_prot->disconnect(sk, flags);
612                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
613                 goto out;
614         }
615 
616         switch (sock->state) {
617         default:
618                 err = -EINVAL;
619                 goto out;
620         case SS_CONNECTED:
621                 err = -EISCONN;
622                 goto out;
623         case SS_CONNECTING:
624                 err = -EALREADY;
625                 /* Fall out of switch with err, set for this state */
626                 break;
627         case SS_UNCONNECTED:
628                 err = -EISCONN;
629                 if (sk->sk_state != TCP_CLOSE)
630                         goto out;
631 
632                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
633                 if (err < 0)
634                         goto out;
635 
636                 sock->state = SS_CONNECTING;
637 
638                 /* Just entered SS_CONNECTING state; the only
639                  * difference is that return value in non-blocking
640                  * case is EINPROGRESS, rather than EALREADY.
641                  */
642                 err = -EINPROGRESS;
643                 break;
644         }
645 
646         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
647 
648         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
649                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
650                                 tcp_sk(sk)->fastopen_req &&
651                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
652 
653                 /* Error code is set above */
654                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
655                         goto out;
656 
657                 err = sock_intr_errno(timeo);
658                 if (signal_pending(current))
659                         goto out;
660         }
661 
662         /* Connection was closed by RST, timeout, ICMP error
663          * or another process disconnected us.
664          */
665         if (sk->sk_state == TCP_CLOSE)
666                 goto sock_error;
667 
668         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
669          * and error was received after socket entered established state.
670          * Hence, it is handled normally after connect() return successfully.
671          */
672 
673         sock->state = SS_CONNECTED;
674         err = 0;
675 out:
676         return err;
677 
678 sock_error:
679         err = sock_error(sk) ? : -ECONNABORTED;
680         sock->state = SS_UNCONNECTED;
681         if (sk->sk_prot->disconnect(sk, flags))
682                 sock->state = SS_DISCONNECTING;
683         goto out;
684 }
685 EXPORT_SYMBOL(__inet_stream_connect);
686 
687 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
688                         int addr_len, int flags)
689 {
690         int err;
691 
692         lock_sock(sock->sk);
693         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
694         release_sock(sock->sk);
695         return err;
696 }
697 EXPORT_SYMBOL(inet_stream_connect);
698 
699 /*
700  *      Accept a pending connection. The TCP layer now gives BSD semantics.
701  */
702 
703 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
704 {
705         struct sock *sk1 = sock->sk;
706         int err = -EINVAL;
707         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
708 
709         if (!sk2)
710                 goto do_err;
711 
712         lock_sock(sk2);
713 
714         sock_rps_record_flow(sk2);
715         WARN_ON(!((1 << sk2->sk_state) &
716                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
717                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
718 
719         sock_graft(sk2, newsock);
720 
721         newsock->state = SS_CONNECTED;
722         err = 0;
723         release_sock(sk2);
724 do_err:
725         return err;
726 }
727 EXPORT_SYMBOL(inet_accept);
728 
729 
730 /*
731  *      This does both peername and sockname.
732  */
733 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
734                         int *uaddr_len, int peer)
735 {
736         struct sock *sk         = sock->sk;
737         struct inet_sock *inet  = inet_sk(sk);
738         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
739 
740         sin->sin_family = AF_INET;
741         if (peer) {
742                 if (!inet->inet_dport ||
743                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
744                      peer == 1))
745                         return -ENOTCONN;
746                 sin->sin_port = inet->inet_dport;
747                 sin->sin_addr.s_addr = inet->inet_daddr;
748         } else {
749                 __be32 addr = inet->inet_rcv_saddr;
750                 if (!addr)
751                         addr = inet->inet_saddr;
752                 sin->sin_port = inet->inet_sport;
753                 sin->sin_addr.s_addr = addr;
754         }
755         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
756         *uaddr_len = sizeof(*sin);
757         return 0;
758 }
759 EXPORT_SYMBOL(inet_getname);
760 
761 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
762                  size_t size)
763 {
764         struct sock *sk = sock->sk;
765 
766         sock_rps_record_flow(sk);
767 
768         /* We may need to bind the socket. */
769         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
770             inet_autobind(sk))
771                 return -EAGAIN;
772 
773         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
774 }
775 EXPORT_SYMBOL(inet_sendmsg);
776 
777 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
778                       size_t size, int flags)
779 {
780         struct sock *sk = sock->sk;
781 
782         sock_rps_record_flow(sk);
783 
784         /* We may need to bind the socket. */
785         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
786             inet_autobind(sk))
787                 return -EAGAIN;
788 
789         if (sk->sk_prot->sendpage)
790                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
791         return sock_no_sendpage(sock, page, offset, size, flags);
792 }
793 EXPORT_SYMBOL(inet_sendpage);
794 
795 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
796                  size_t size, int flags)
797 {
798         struct sock *sk = sock->sk;
799         int addr_len = 0;
800         int err;
801 
802         sock_rps_record_flow(sk);
803 
804         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
805                                    flags & ~MSG_DONTWAIT, &addr_len);
806         if (err >= 0)
807                 msg->msg_namelen = addr_len;
808         return err;
809 }
810 EXPORT_SYMBOL(inet_recvmsg);
811 
812 int inet_shutdown(struct socket *sock, int how)
813 {
814         struct sock *sk = sock->sk;
815         int err = 0;
816 
817         /* This should really check to make sure
818          * the socket is a TCP socket. (WHY AC...)
819          */
820         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
821                        1->2 bit 2 snds.
822                        2->3 */
823         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
824                 return -EINVAL;
825 
826         lock_sock(sk);
827         if (sock->state == SS_CONNECTING) {
828                 if ((1 << sk->sk_state) &
829                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
830                         sock->state = SS_DISCONNECTING;
831                 else
832                         sock->state = SS_CONNECTED;
833         }
834 
835         switch (sk->sk_state) {
836         case TCP_CLOSE:
837                 err = -ENOTCONN;
838                 /* Hack to wake up other listeners, who can poll for
839                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
840         default:
841                 sk->sk_shutdown |= how;
842                 if (sk->sk_prot->shutdown)
843                         sk->sk_prot->shutdown(sk, how);
844                 break;
845 
846         /* Remaining two branches are temporary solution for missing
847          * close() in multithreaded environment. It is _not_ a good idea,
848          * but we have no choice until close() is repaired at VFS level.
849          */
850         case TCP_LISTEN:
851                 if (!(how & RCV_SHUTDOWN))
852                         break;
853                 /* Fall through */
854         case TCP_SYN_SENT:
855                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
856                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
857                 break;
858         }
859 
860         /* Wake up anyone sleeping in poll. */
861         sk->sk_state_change(sk);
862         release_sock(sk);
863         return err;
864 }
865 EXPORT_SYMBOL(inet_shutdown);
866 
867 /*
868  *      ioctl() calls you can issue on an INET socket. Most of these are
869  *      device configuration and stuff and very rarely used. Some ioctls
870  *      pass on to the socket itself.
871  *
872  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
873  *      loads the devconfigure module does its configuring and unloads it.
874  *      There's a good 20K of config code hanging around the kernel.
875  */
876 
877 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
878 {
879         struct sock *sk = sock->sk;
880         int err = 0;
881         struct net *net = sock_net(sk);
882 
883         switch (cmd) {
884         case SIOCGSTAMP:
885                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
886                 break;
887         case SIOCGSTAMPNS:
888                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
889                 break;
890         case SIOCADDRT:
891         case SIOCDELRT:
892         case SIOCRTMSG:
893                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
894                 break;
895         case SIOCDARP:
896         case SIOCGARP:
897         case SIOCSARP:
898                 err = arp_ioctl(net, cmd, (void __user *)arg);
899                 break;
900         case SIOCGIFADDR:
901         case SIOCSIFADDR:
902         case SIOCGIFBRDADDR:
903         case SIOCSIFBRDADDR:
904         case SIOCGIFNETMASK:
905         case SIOCSIFNETMASK:
906         case SIOCGIFDSTADDR:
907         case SIOCSIFDSTADDR:
908         case SIOCSIFPFLAGS:
909         case SIOCGIFPFLAGS:
910         case SIOCSIFFLAGS:
911                 err = devinet_ioctl(net, cmd, (void __user *)arg);
912                 break;
913         default:
914                 if (sk->sk_prot->ioctl)
915                         err = sk->sk_prot->ioctl(sk, cmd, arg);
916                 else
917                         err = -ENOIOCTLCMD;
918                 break;
919         }
920         return err;
921 }
922 EXPORT_SYMBOL(inet_ioctl);
923 
924 #ifdef CONFIG_COMPAT
925 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
926 {
927         struct sock *sk = sock->sk;
928         int err = -ENOIOCTLCMD;
929 
930         if (sk->sk_prot->compat_ioctl)
931                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
932 
933         return err;
934 }
935 #endif
936 
937 const struct proto_ops inet_stream_ops = {
938         .family            = PF_INET,
939         .owner             = THIS_MODULE,
940         .release           = inet_release,
941         .bind              = inet_bind,
942         .connect           = inet_stream_connect,
943         .socketpair        = sock_no_socketpair,
944         .accept            = inet_accept,
945         .getname           = inet_getname,
946         .poll              = tcp_poll,
947         .ioctl             = inet_ioctl,
948         .listen            = inet_listen,
949         .shutdown          = inet_shutdown,
950         .setsockopt        = sock_common_setsockopt,
951         .getsockopt        = sock_common_getsockopt,
952         .sendmsg           = inet_sendmsg,
953         .recvmsg           = inet_recvmsg,
954         .mmap              = sock_no_mmap,
955         .sendpage          = inet_sendpage,
956         .splice_read       = tcp_splice_read,
957 #ifdef CONFIG_COMPAT
958         .compat_setsockopt = compat_sock_common_setsockopt,
959         .compat_getsockopt = compat_sock_common_getsockopt,
960         .compat_ioctl      = inet_compat_ioctl,
961 #endif
962 };
963 EXPORT_SYMBOL(inet_stream_ops);
964 
965 const struct proto_ops inet_dgram_ops = {
966         .family            = PF_INET,
967         .owner             = THIS_MODULE,
968         .release           = inet_release,
969         .bind              = inet_bind,
970         .connect           = inet_dgram_connect,
971         .socketpair        = sock_no_socketpair,
972         .accept            = sock_no_accept,
973         .getname           = inet_getname,
974         .poll              = udp_poll,
975         .ioctl             = inet_ioctl,
976         .listen            = sock_no_listen,
977         .shutdown          = inet_shutdown,
978         .setsockopt        = sock_common_setsockopt,
979         .getsockopt        = sock_common_getsockopt,
980         .sendmsg           = inet_sendmsg,
981         .recvmsg           = inet_recvmsg,
982         .mmap              = sock_no_mmap,
983         .sendpage          = inet_sendpage,
984 #ifdef CONFIG_COMPAT
985         .compat_setsockopt = compat_sock_common_setsockopt,
986         .compat_getsockopt = compat_sock_common_getsockopt,
987         .compat_ioctl      = inet_compat_ioctl,
988 #endif
989 };
990 EXPORT_SYMBOL(inet_dgram_ops);
991 
992 /*
993  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
994  * udp_poll
995  */
996 static const struct proto_ops inet_sockraw_ops = {
997         .family            = PF_INET,
998         .owner             = THIS_MODULE,
999         .release           = inet_release,
1000         .bind              = inet_bind,
1001         .connect           = inet_dgram_connect,
1002         .socketpair        = sock_no_socketpair,
1003         .accept            = sock_no_accept,
1004         .getname           = inet_getname,
1005         .poll              = datagram_poll,
1006         .ioctl             = inet_ioctl,
1007         .listen            = sock_no_listen,
1008         .shutdown          = inet_shutdown,
1009         .setsockopt        = sock_common_setsockopt,
1010         .getsockopt        = sock_common_getsockopt,
1011         .sendmsg           = inet_sendmsg,
1012         .recvmsg           = inet_recvmsg,
1013         .mmap              = sock_no_mmap,
1014         .sendpage          = inet_sendpage,
1015 #ifdef CONFIG_COMPAT
1016         .compat_setsockopt = compat_sock_common_setsockopt,
1017         .compat_getsockopt = compat_sock_common_getsockopt,
1018         .compat_ioctl      = inet_compat_ioctl,
1019 #endif
1020 };
1021 
1022 static const struct net_proto_family inet_family_ops = {
1023         .family = PF_INET,
1024         .create = inet_create,
1025         .owner  = THIS_MODULE,
1026 };
1027 
1028 /* Upon startup we insert all the elements in inetsw_array[] into
1029  * the linked list inetsw.
1030  */
1031 static struct inet_protosw inetsw_array[] =
1032 {
1033         {
1034                 .type =       SOCK_STREAM,
1035                 .protocol =   IPPROTO_TCP,
1036                 .prot =       &tcp_prot,
1037                 .ops =        &inet_stream_ops,
1038                 .no_check =   0,
1039                 .flags =      INET_PROTOSW_PERMANENT |
1040                               INET_PROTOSW_ICSK,
1041         },
1042 
1043         {
1044                 .type =       SOCK_DGRAM,
1045                 .protocol =   IPPROTO_UDP,
1046                 .prot =       &udp_prot,
1047                 .ops =        &inet_dgram_ops,
1048                 .no_check =   UDP_CSUM_DEFAULT,
1049                 .flags =      INET_PROTOSW_PERMANENT,
1050        },
1051 
1052        {
1053                 .type =       SOCK_DGRAM,
1054                 .protocol =   IPPROTO_ICMP,
1055                 .prot =       &ping_prot,
1056                 .ops =        &inet_sockraw_ops,
1057                 .no_check =   UDP_CSUM_DEFAULT,
1058                 .flags =      INET_PROTOSW_REUSE,
1059        },
1060 
1061        {
1062                .type =       SOCK_RAW,
1063                .protocol =   IPPROTO_IP,        /* wild card */
1064                .prot =       &raw_prot,
1065                .ops =        &inet_sockraw_ops,
1066                .no_check =   UDP_CSUM_DEFAULT,
1067                .flags =      INET_PROTOSW_REUSE,
1068        }
1069 };
1070 
1071 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1072 
1073 void inet_register_protosw(struct inet_protosw *p)
1074 {
1075         struct list_head *lh;
1076         struct inet_protosw *answer;
1077         int protocol = p->protocol;
1078         struct list_head *last_perm;
1079 
1080         spin_lock_bh(&inetsw_lock);
1081 
1082         if (p->type >= SOCK_MAX)
1083                 goto out_illegal;
1084 
1085         /* If we are trying to override a permanent protocol, bail. */
1086         answer = NULL;
1087         last_perm = &inetsw[p->type];
1088         list_for_each(lh, &inetsw[p->type]) {
1089                 answer = list_entry(lh, struct inet_protosw, list);
1090 
1091                 /* Check only the non-wild match. */
1092                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1093                         if (protocol == answer->protocol)
1094                                 break;
1095                         last_perm = lh;
1096                 }
1097 
1098                 answer = NULL;
1099         }
1100         if (answer)
1101                 goto out_permanent;
1102 
1103         /* Add the new entry after the last permanent entry if any, so that
1104          * the new entry does not override a permanent entry when matched with
1105          * a wild-card protocol. But it is allowed to override any existing
1106          * non-permanent entry.  This means that when we remove this entry, the
1107          * system automatically returns to the old behavior.
1108          */
1109         list_add_rcu(&p->list, last_perm);
1110 out:
1111         spin_unlock_bh(&inetsw_lock);
1112 
1113         return;
1114 
1115 out_permanent:
1116         pr_err("Attempt to override permanent protocol %d\n", protocol);
1117         goto out;
1118 
1119 out_illegal:
1120         pr_err("Ignoring attempt to register invalid socket type %d\n",
1121                p->type);
1122         goto out;
1123 }
1124 EXPORT_SYMBOL(inet_register_protosw);
1125 
1126 void inet_unregister_protosw(struct inet_protosw *p)
1127 {
1128         if (INET_PROTOSW_PERMANENT & p->flags) {
1129                 pr_err("Attempt to unregister permanent protocol %d\n",
1130                        p->protocol);
1131         } else {
1132                 spin_lock_bh(&inetsw_lock);
1133                 list_del_rcu(&p->list);
1134                 spin_unlock_bh(&inetsw_lock);
1135 
1136                 synchronize_net();
1137         }
1138 }
1139 EXPORT_SYMBOL(inet_unregister_protosw);
1140 
1141 /*
1142  *      Shall we try to damage output packets if routing dev changes?
1143  */
1144 
1145 int sysctl_ip_dynaddr __read_mostly;
1146 
1147 static int inet_sk_reselect_saddr(struct sock *sk)
1148 {
1149         struct inet_sock *inet = inet_sk(sk);
1150         __be32 old_saddr = inet->inet_saddr;
1151         __be32 daddr = inet->inet_daddr;
1152         struct flowi4 *fl4;
1153         struct rtable *rt;
1154         __be32 new_saddr;
1155         struct ip_options_rcu *inet_opt;
1156 
1157         inet_opt = rcu_dereference_protected(inet->inet_opt,
1158                                              sock_owned_by_user(sk));
1159         if (inet_opt && inet_opt->opt.srr)
1160                 daddr = inet_opt->opt.faddr;
1161 
1162         /* Query new route. */
1163         fl4 = &inet->cork.fl.u.ip4;
1164         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1165                               sk->sk_bound_dev_if, sk->sk_protocol,
1166                               inet->inet_sport, inet->inet_dport, sk, false);
1167         if (IS_ERR(rt))
1168                 return PTR_ERR(rt);
1169 
1170         sk_setup_caps(sk, &rt->dst);
1171 
1172         new_saddr = fl4->saddr;
1173 
1174         if (new_saddr == old_saddr)
1175                 return 0;
1176 
1177         if (sysctl_ip_dynaddr > 1) {
1178                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1179                         __func__, &old_saddr, &new_saddr);
1180         }
1181 
1182         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1183 
1184         /*
1185          * XXX The only one ugly spot where we need to
1186          * XXX really change the sockets identity after
1187          * XXX it has entered the hashes. -DaveM
1188          *
1189          * Besides that, it does not check for connection
1190          * uniqueness. Wait for troubles.
1191          */
1192         __sk_prot_rehash(sk);
1193         return 0;
1194 }
1195 
1196 int inet_sk_rebuild_header(struct sock *sk)
1197 {
1198         struct inet_sock *inet = inet_sk(sk);
1199         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1200         __be32 daddr;
1201         struct ip_options_rcu *inet_opt;
1202         struct flowi4 *fl4;
1203         int err;
1204 
1205         /* Route is OK, nothing to do. */
1206         if (rt)
1207                 return 0;
1208 
1209         /* Reroute. */
1210         rcu_read_lock();
1211         inet_opt = rcu_dereference(inet->inet_opt);
1212         daddr = inet->inet_daddr;
1213         if (inet_opt && inet_opt->opt.srr)
1214                 daddr = inet_opt->opt.faddr;
1215         rcu_read_unlock();
1216         fl4 = &inet->cork.fl.u.ip4;
1217         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1218                                    inet->inet_dport, inet->inet_sport,
1219                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1220                                    sk->sk_bound_dev_if);
1221         if (!IS_ERR(rt)) {
1222                 err = 0;
1223                 sk_setup_caps(sk, &rt->dst);
1224         } else {
1225                 err = PTR_ERR(rt);
1226 
1227                 /* Routing failed... */
1228                 sk->sk_route_caps = 0;
1229                 /*
1230                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1231                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1232                  */
1233                 if (!sysctl_ip_dynaddr ||
1234                     sk->sk_state != TCP_SYN_SENT ||
1235                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1236                     (err = inet_sk_reselect_saddr(sk)) != 0)
1237                         sk->sk_err_soft = -err;
1238         }
1239 
1240         return err;
1241 }
1242 EXPORT_SYMBOL(inet_sk_rebuild_header);
1243 
1244 static int inet_gso_send_check(struct sk_buff *skb)
1245 {
1246         const struct net_offload *ops;
1247         const struct iphdr *iph;
1248         int proto;
1249         int ihl;
1250         int err = -EINVAL;
1251 
1252         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1253                 goto out;
1254 
1255         iph = ip_hdr(skb);
1256         ihl = iph->ihl * 4;
1257         if (ihl < sizeof(*iph))
1258                 goto out;
1259 
1260         if (unlikely(!pskb_may_pull(skb, ihl)))
1261                 goto out;
1262 
1263         __skb_pull(skb, ihl);
1264         skb_reset_transport_header(skb);
1265         iph = ip_hdr(skb);
1266         proto = iph->protocol;
1267         err = -EPROTONOSUPPORT;
1268 
1269         rcu_read_lock();
1270         ops = rcu_dereference(inet_offloads[proto]);
1271         if (likely(ops && ops->callbacks.gso_send_check))
1272                 err = ops->callbacks.gso_send_check(skb);
1273         rcu_read_unlock();
1274 
1275 out:
1276         return err;
1277 }
1278 
1279 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1280         netdev_features_t features)
1281 {
1282         struct sk_buff *segs = ERR_PTR(-EINVAL);
1283         const struct net_offload *ops;
1284         struct iphdr *iph;
1285         int proto;
1286         int ihl;
1287         int id;
1288         unsigned int offset = 0;
1289         bool tunnel;
1290 
1291         if (unlikely(skb_shinfo(skb)->gso_type &
1292                      ~(SKB_GSO_TCPV4 |
1293                        SKB_GSO_UDP |
1294                        SKB_GSO_DODGY |
1295                        SKB_GSO_TCP_ECN |
1296                        SKB_GSO_GRE |
1297                        SKB_GSO_TCPV6 |
1298                        SKB_GSO_UDP_TUNNEL |
1299                        0)))
1300                 goto out;
1301 
1302         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1303                 goto out;
1304 
1305         iph = ip_hdr(skb);
1306         ihl = iph->ihl * 4;
1307         if (ihl < sizeof(*iph))
1308                 goto out;
1309 
1310         if (unlikely(!pskb_may_pull(skb, ihl)))
1311                 goto out;
1312 
1313         tunnel = !!skb->encapsulation;
1314 
1315         __skb_pull(skb, ihl);
1316         skb_reset_transport_header(skb);
1317         iph = ip_hdr(skb);
1318         id = ntohs(iph->id);
1319         proto = iph->protocol;
1320         segs = ERR_PTR(-EPROTONOSUPPORT);
1321 
1322         rcu_read_lock();
1323         ops = rcu_dereference(inet_offloads[proto]);
1324         if (likely(ops && ops->callbacks.gso_segment))
1325                 segs = ops->callbacks.gso_segment(skb, features);
1326         rcu_read_unlock();
1327 
1328         if (IS_ERR_OR_NULL(segs))
1329                 goto out;
1330 
1331         skb = segs;
1332         do {
1333                 iph = ip_hdr(skb);
1334                 if (!tunnel && proto == IPPROTO_UDP) {
1335                         iph->id = htons(id);
1336                         iph->frag_off = htons(offset >> 3);
1337                         if (skb->next != NULL)
1338                                 iph->frag_off |= htons(IP_MF);
1339                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1340                 } else  {
1341                         iph->id = htons(id++);
1342                 }
1343                 iph->tot_len = htons(skb->len - skb->mac_len);
1344                 iph->check = 0;
1345                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1346         } while ((skb = skb->next));
1347 
1348 out:
1349         return segs;
1350 }
1351 
1352 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1353                                          struct sk_buff *skb)
1354 {
1355         const struct net_offload *ops;
1356         struct sk_buff **pp = NULL;
1357         struct sk_buff *p;
1358         const struct iphdr *iph;
1359         unsigned int hlen;
1360         unsigned int off;
1361         unsigned int id;
1362         int flush = 1;
1363         int proto;
1364 
1365         off = skb_gro_offset(skb);
1366         hlen = off + sizeof(*iph);
1367         iph = skb_gro_header_fast(skb, off);
1368         if (skb_gro_header_hard(skb, hlen)) {
1369                 iph = skb_gro_header_slow(skb, hlen, off);
1370                 if (unlikely(!iph))
1371                         goto out;
1372         }
1373 
1374         proto = iph->protocol;
1375 
1376         rcu_read_lock();
1377         ops = rcu_dereference(inet_offloads[proto]);
1378         if (!ops || !ops->callbacks.gro_receive)
1379                 goto out_unlock;
1380 
1381         if (*(u8 *)iph != 0x45)
1382                 goto out_unlock;
1383 
1384         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1385                 goto out_unlock;
1386 
1387         id = ntohl(*(__be32 *)&iph->id);
1388         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1389         id >>= 16;
1390 
1391         for (p = *head; p; p = p->next) {
1392                 struct iphdr *iph2;
1393 
1394                 if (!NAPI_GRO_CB(p)->same_flow)
1395                         continue;
1396 
1397                 iph2 = ip_hdr(p);
1398 
1399                 if ((iph->protocol ^ iph2->protocol) |
1400                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1401                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1402                         NAPI_GRO_CB(p)->same_flow = 0;
1403                         continue;
1404                 }
1405 
1406                 /* All fields must match except length and checksum. */
1407                 NAPI_GRO_CB(p)->flush |=
1408                         (iph->ttl ^ iph2->ttl) |
1409                         (iph->tos ^ iph2->tos) |
1410                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1411 
1412                 NAPI_GRO_CB(p)->flush |= flush;
1413         }
1414 
1415         NAPI_GRO_CB(skb)->flush |= flush;
1416         skb_gro_pull(skb, sizeof(*iph));
1417         skb_set_transport_header(skb, skb_gro_offset(skb));
1418 
1419         pp = ops->callbacks.gro_receive(head, skb);
1420 
1421 out_unlock:
1422         rcu_read_unlock();
1423 
1424 out:
1425         NAPI_GRO_CB(skb)->flush |= flush;
1426 
1427         return pp;
1428 }
1429 
1430 static int inet_gro_complete(struct sk_buff *skb)
1431 {
1432         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1433         struct iphdr *iph = ip_hdr(skb);
1434         const struct net_offload *ops;
1435         int proto = iph->protocol;
1436         int err = -ENOSYS;
1437 
1438         csum_replace2(&iph->check, iph->tot_len, newlen);
1439         iph->tot_len = newlen;
1440 
1441         rcu_read_lock();
1442         ops = rcu_dereference(inet_offloads[proto]);
1443         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1444                 goto out_unlock;
1445 
1446         err = ops->callbacks.gro_complete(skb);
1447 
1448 out_unlock:
1449         rcu_read_unlock();
1450 
1451         return err;
1452 }
1453 
1454 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1455                          unsigned short type, unsigned char protocol,
1456                          struct net *net)
1457 {
1458         struct socket *sock;
1459         int rc = sock_create_kern(family, type, protocol, &sock);
1460 
1461         if (rc == 0) {
1462                 *sk = sock->sk;
1463                 (*sk)->sk_allocation = GFP_ATOMIC;
1464                 /*
1465                  * Unhash it so that IP input processing does not even see it,
1466                  * we do not wish this socket to see incoming packets.
1467                  */
1468                 (*sk)->sk_prot->unhash(*sk);
1469 
1470                 sk_change_net(*sk, net);
1471         }
1472         return rc;
1473 }
1474 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1475 
1476 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1477 {
1478         unsigned long res = 0;
1479         int i, j;
1480 
1481         for_each_possible_cpu(i) {
1482                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1483                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1484         }
1485         return res;
1486 }
1487 EXPORT_SYMBOL_GPL(snmp_fold_field);
1488 
1489 #if BITS_PER_LONG==32
1490 
1491 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1492 {
1493         u64 res = 0;
1494         int cpu;
1495 
1496         for_each_possible_cpu(cpu) {
1497                 void *bhptr;
1498                 struct u64_stats_sync *syncp;
1499                 u64 v;
1500                 unsigned int start;
1501 
1502                 bhptr = per_cpu_ptr(mib[0], cpu);
1503                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1504                 do {
1505                         start = u64_stats_fetch_begin_bh(syncp);
1506                         v = *(((u64 *) bhptr) + offt);
1507                 } while (u64_stats_fetch_retry_bh(syncp, start));
1508 
1509                 res += v;
1510         }
1511         return res;
1512 }
1513 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1514 #endif
1515 
1516 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1517 {
1518         BUG_ON(ptr == NULL);
1519         ptr[0] = __alloc_percpu(mibsize, align);
1520         if (!ptr[0])
1521                 return -ENOMEM;
1522 #if SNMP_ARRAY_SZ == 2
1523         ptr[1] = __alloc_percpu(mibsize, align);
1524         if (!ptr[1]) {
1525                 free_percpu(ptr[0]);
1526                 ptr[0] = NULL;
1527                 return -ENOMEM;
1528         }
1529 #endif
1530         return 0;
1531 }
1532 EXPORT_SYMBOL_GPL(snmp_mib_init);
1533 
1534 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1535 {
1536         int i;
1537 
1538         BUG_ON(ptr == NULL);
1539         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1540                 free_percpu(ptr[i]);
1541                 ptr[i] = NULL;
1542         }
1543 }
1544 EXPORT_SYMBOL_GPL(snmp_mib_free);
1545 
1546 #ifdef CONFIG_IP_MULTICAST
1547 static const struct net_protocol igmp_protocol = {
1548         .handler =      igmp_rcv,
1549         .netns_ok =     1,
1550 };
1551 #endif
1552 
1553 static const struct net_protocol tcp_protocol = {
1554         .early_demux    =       tcp_v4_early_demux,
1555         .handler        =       tcp_v4_rcv,
1556         .err_handler    =       tcp_v4_err,
1557         .no_policy      =       1,
1558         .netns_ok       =       1,
1559 };
1560 
1561 static const struct net_offload tcp_offload = {
1562         .callbacks = {
1563                 .gso_send_check =       tcp_v4_gso_send_check,
1564                 .gso_segment    =       tcp_tso_segment,
1565                 .gro_receive    =       tcp4_gro_receive,
1566                 .gro_complete   =       tcp4_gro_complete,
1567         },
1568 };
1569 
1570 static const struct net_protocol udp_protocol = {
1571         .handler =      udp_rcv,
1572         .err_handler =  udp_err,
1573         .no_policy =    1,
1574         .netns_ok =     1,
1575 };
1576 
1577 static const struct net_offload udp_offload = {
1578         .callbacks = {
1579                 .gso_send_check = udp4_ufo_send_check,
1580                 .gso_segment = udp4_ufo_fragment,
1581         },
1582 };
1583 
1584 static const struct net_protocol icmp_protocol = {
1585         .handler =      icmp_rcv,
1586         .err_handler =  icmp_err,
1587         .no_policy =    1,
1588         .netns_ok =     1,
1589 };
1590 
1591 static __net_init int ipv4_mib_init_net(struct net *net)
1592 {
1593         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1594                           sizeof(struct tcp_mib),
1595                           __alignof__(struct tcp_mib)) < 0)
1596                 goto err_tcp_mib;
1597         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1598                           sizeof(struct ipstats_mib),
1599                           __alignof__(struct ipstats_mib)) < 0)
1600                 goto err_ip_mib;
1601         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1602                           sizeof(struct linux_mib),
1603                           __alignof__(struct linux_mib)) < 0)
1604                 goto err_net_mib;
1605         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1606                           sizeof(struct udp_mib),
1607                           __alignof__(struct udp_mib)) < 0)
1608                 goto err_udp_mib;
1609         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1610                           sizeof(struct udp_mib),
1611                           __alignof__(struct udp_mib)) < 0)
1612                 goto err_udplite_mib;
1613         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1614                           sizeof(struct icmp_mib),
1615                           __alignof__(struct icmp_mib)) < 0)
1616                 goto err_icmp_mib;
1617         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1618                                               GFP_KERNEL);
1619         if (!net->mib.icmpmsg_statistics)
1620                 goto err_icmpmsg_mib;
1621 
1622         tcp_mib_init(net);
1623         return 0;
1624 
1625 err_icmpmsg_mib:
1626         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1627 err_icmp_mib:
1628         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1629 err_udplite_mib:
1630         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1631 err_udp_mib:
1632         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1633 err_net_mib:
1634         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1635 err_ip_mib:
1636         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1637 err_tcp_mib:
1638         return -ENOMEM;
1639 }
1640 
1641 static __net_exit void ipv4_mib_exit_net(struct net *net)
1642 {
1643         kfree(net->mib.icmpmsg_statistics);
1644         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1645         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1646         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1647         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1648         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1649         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1650 }
1651 
1652 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1653         .init = ipv4_mib_init_net,
1654         .exit = ipv4_mib_exit_net,
1655 };
1656 
1657 static int __init init_ipv4_mibs(void)
1658 {
1659         return register_pernet_subsys(&ipv4_mib_ops);
1660 }
1661 
1662 static int ipv4_proc_init(void);
1663 
1664 /*
1665  *      IP protocol layer initialiser
1666  */
1667 
1668 static struct packet_offload ip_packet_offload __read_mostly = {
1669         .type = cpu_to_be16(ETH_P_IP),
1670         .callbacks = {
1671                 .gso_send_check = inet_gso_send_check,
1672                 .gso_segment = inet_gso_segment,
1673                 .gro_receive = inet_gro_receive,
1674                 .gro_complete = inet_gro_complete,
1675         },
1676 };
1677 
1678 static int __init ipv4_offload_init(void)
1679 {
1680         /*
1681          * Add offloads
1682          */
1683         if (inet_add_offload(&udp_offload, IPPROTO_UDP) < 0)
1684                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1685         if (inet_add_offload(&tcp_offload, IPPROTO_TCP) < 0)
1686                 pr_crit("%s: Cannot add TCP protocol offlaod\n", __func__);
1687 
1688         dev_add_offload(&ip_packet_offload);
1689         return 0;
1690 }
1691 
1692 fs_initcall(ipv4_offload_init);
1693 
1694 static struct packet_type ip_packet_type __read_mostly = {
1695         .type = cpu_to_be16(ETH_P_IP),
1696         .func = ip_rcv,
1697 };
1698 
1699 static int __init inet_init(void)
1700 {
1701         struct inet_protosw *q;
1702         struct list_head *r;
1703         int rc = -EINVAL;
1704 
1705         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1706 
1707         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1708         if (!sysctl_local_reserved_ports)
1709                 goto out;
1710 
1711         rc = proto_register(&tcp_prot, 1);
1712         if (rc)
1713                 goto out_free_reserved_ports;
1714 
1715         rc = proto_register(&udp_prot, 1);
1716         if (rc)
1717                 goto out_unregister_tcp_proto;
1718 
1719         rc = proto_register(&raw_prot, 1);
1720         if (rc)
1721                 goto out_unregister_udp_proto;
1722 
1723         rc = proto_register(&ping_prot, 1);
1724         if (rc)
1725                 goto out_unregister_raw_proto;
1726 
1727         /*
1728          *      Tell SOCKET that we are alive...
1729          */
1730 
1731         (void)sock_register(&inet_family_ops);
1732 
1733 #ifdef CONFIG_SYSCTL
1734         ip_static_sysctl_init();
1735 #endif
1736 
1737         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1738 
1739         /*
1740          *      Add all the base protocols.
1741          */
1742 
1743         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1744                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1745         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1746                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1747         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1748                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1749 #ifdef CONFIG_IP_MULTICAST
1750         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1751                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1752 #endif
1753 
1754         /* Register the socket-side information for inet_create. */
1755         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1756                 INIT_LIST_HEAD(r);
1757 
1758         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1759                 inet_register_protosw(q);
1760 
1761         /*
1762          *      Set the ARP module up
1763          */
1764 
1765         arp_init();
1766 
1767         /*
1768          *      Set the IP module up
1769          */
1770 
1771         ip_init();
1772 
1773         tcp_v4_init();
1774 
1775         /* Setup TCP slab cache for open requests. */
1776         tcp_init();
1777 
1778         /* Setup UDP memory threshold */
1779         udp_init();
1780 
1781         /* Add UDP-Lite (RFC 3828) */
1782         udplite4_register();
1783 
1784         ping_init();
1785 
1786         /*
1787          *      Set the ICMP layer up
1788          */
1789 
1790         if (icmp_init() < 0)
1791                 panic("Failed to create the ICMP control socket.\n");
1792 
1793         /*
1794          *      Initialise the multicast router
1795          */
1796 #if defined(CONFIG_IP_MROUTE)
1797         if (ip_mr_init())
1798                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1799 #endif
1800         /*
1801          *      Initialise per-cpu ipv4 mibs
1802          */
1803 
1804         if (init_ipv4_mibs())
1805                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1806 
1807         ipv4_proc_init();
1808 
1809         ipfrag_init();
1810 
1811         dev_add_pack(&ip_packet_type);
1812 
1813         rc = 0;
1814 out:
1815         return rc;
1816 out_unregister_raw_proto:
1817         proto_unregister(&raw_prot);
1818 out_unregister_udp_proto:
1819         proto_unregister(&udp_prot);
1820 out_unregister_tcp_proto:
1821         proto_unregister(&tcp_prot);
1822 out_free_reserved_ports:
1823         kfree(sysctl_local_reserved_ports);
1824         goto out;
1825 }
1826 
1827 fs_initcall(inet_init);
1828 
1829 /* ------------------------------------------------------------------------ */
1830 
1831 #ifdef CONFIG_PROC_FS
1832 static int __init ipv4_proc_init(void)
1833 {
1834         int rc = 0;
1835 
1836         if (raw_proc_init())
1837                 goto out_raw;
1838         if (tcp4_proc_init())
1839                 goto out_tcp;
1840         if (udp4_proc_init())
1841                 goto out_udp;
1842         if (ping_proc_init())
1843                 goto out_ping;
1844         if (ip_misc_proc_init())
1845                 goto out_misc;
1846 out:
1847         return rc;
1848 out_misc:
1849         ping_proc_exit();
1850 out_ping:
1851         udp4_proc_exit();
1852 out_udp:
1853         tcp4_proc_exit();
1854 out_tcp:
1855         raw_proc_exit();
1856 out_raw:
1857         rc = -ENOMEM;
1858         goto out;
1859 }
1860 
1861 #else /* CONFIG_PROC_FS */
1862 static int __init ipv4_proc_init(void)
1863 {
1864         return 0;
1865 }
1866 #endif /* CONFIG_PROC_FS */
1867 
1868 MODULE_ALIAS_NETPROTO(PF_INET);
1869 
1870 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp