Merge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[pandora-kernel.git] / net / ipv4 / tcp_ipv4.c
1 /*
2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
3  *              operating system.  INET is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              Implementation of the Transmission Control Protocol(TCP).
7  *
8  *              IPv4 specific functions
9  *
10  *
11  *              code split from:
12  *              linux/ipv4/tcp.c
13  *              linux/ipv4/tcp_input.c
14  *              linux/ipv4/tcp_output.c
15  *
16  *              See tcp.c for author information
17  *
18  *      This program is free software; you can redistribute it and/or
19  *      modify it under the terms of the GNU General Public License
20  *      as published by the Free Software Foundation; either version
21  *      2 of the License, or (at your option) any later version.
22  */
23
24 /*
25  * Changes:
26  *              David S. Miller :       New socket lookup architecture.
27  *                                      This code is dedicated to John Dyson.
28  *              David S. Miller :       Change semantics of established hash,
29  *                                      half is devoted to TIME_WAIT sockets
30  *                                      and the rest go in the other half.
31  *              Andi Kleen :            Add support for syncookies and fixed
32  *                                      some bugs: ip options weren't passed to
33  *                                      the TCP layer, missed a check for an
34  *                                      ACK bit.
35  *              Andi Kleen :            Implemented fast path mtu discovery.
36  *                                      Fixed many serious bugs in the
37  *                                      request_sock handling and moved
38  *                                      most of it into the af independent code.
39  *                                      Added tail drop and some other bugfixes.
40  *                                      Added new listen semantics.
41  *              Mike McLagan    :       Routing by source
42  *      Juan Jose Ciarlante:            ip_dynaddr bits
43  *              Andi Kleen:             various fixes.
44  *      Vitaly E. Lavrov        :       Transparent proxy revived after year
45  *                                      coma.
46  *      Andi Kleen              :       Fix new listen.
47  *      Andi Kleen              :       Fix accept error reporting.
48  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
49  *      Alexey Kuznetsov                allow both IPv4 and IPv6 sockets to bind
50  *                                      a single port at the same time.
51  */
52
53
54 #include <linux/bottom_half.h>
55 #include <linux/types.h>
56 #include <linux/fcntl.h>
57 #include <linux/module.h>
58 #include <linux/random.h>
59 #include <linux/cache.h>
60 #include <linux/jhash.h>
61 #include <linux/init.h>
62 #include <linux/times.h>
63
64 #include <net/net_namespace.h>
65 #include <net/icmp.h>
66 #include <net/inet_hashtables.h>
67 #include <net/tcp.h>
68 #include <net/transp_v6.h>
69 #include <net/ipv6.h>
70 #include <net/inet_common.h>
71 #include <net/timewait_sock.h>
72 #include <net/xfrm.h>
73 #include <net/netdma.h>
74
75 #include <linux/inet.h>
76 #include <linux/ipv6.h>
77 #include <linux/stddef.h>
78 #include <linux/proc_fs.h>
79 #include <linux/seq_file.h>
80
81 #include <linux/crypto.h>
82 #include <linux/scatterlist.h>
83
84 int sysctl_tcp_tw_reuse __read_mostly;
85 int sysctl_tcp_low_latency __read_mostly;
86
87
88 #ifdef CONFIG_TCP_MD5SIG
89 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
90                                                    __be32 addr);
91 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
92                                __be32 daddr, __be32 saddr, struct tcphdr *th);
93 #else
94 static inline
95 struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
96 {
97         return NULL;
98 }
99 #endif
100
101 struct inet_hashinfo tcp_hashinfo;
102
103 static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
104 {
105         return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
106                                           ip_hdr(skb)->saddr,
107                                           tcp_hdr(skb)->dest,
108                                           tcp_hdr(skb)->source);
109 }
110
111 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
112 {
113         const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
114         struct tcp_sock *tp = tcp_sk(sk);
115
116         /* With PAWS, it is safe from the viewpoint
117            of data integrity. Even without PAWS it is safe provided sequence
118            spaces do not overlap i.e. at data rates <= 80Mbit/sec.
119
120            Actually, the idea is close to VJ's one, only timestamp cache is
121            held not per host, but per port pair and TW bucket is used as state
122            holder.
123
124            If TW bucket has been already destroyed we fall back to VJ's scheme
125            and use initial timestamp retrieved from peer table.
126          */
127         if (tcptw->tw_ts_recent_stamp &&
128             (twp == NULL || (sysctl_tcp_tw_reuse &&
129                              get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
130                 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
131                 if (tp->write_seq == 0)
132                         tp->write_seq = 1;
133                 tp->rx_opt.ts_recent       = tcptw->tw_ts_recent;
134                 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
135                 sock_hold(sktw);
136                 return 1;
137         }
138
139         return 0;
140 }
141
142 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
143
144 /* This will initiate an outgoing connection. */
145 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
146 {
147         struct inet_sock *inet = inet_sk(sk);
148         struct tcp_sock *tp = tcp_sk(sk);
149         struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
150         struct rtable *rt;
151         __be32 daddr, nexthop;
152         int tmp;
153         int err;
154
155         if (addr_len < sizeof(struct sockaddr_in))
156                 return -EINVAL;
157
158         if (usin->sin_family != AF_INET)
159                 return -EAFNOSUPPORT;
160
161         nexthop = daddr = usin->sin_addr.s_addr;
162         if (inet->opt && inet->opt->srr) {
163                 if (!daddr)
164                         return -EINVAL;
165                 nexthop = inet->opt->faddr;
166         }
167
168         tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
169                                RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
170                                IPPROTO_TCP,
171                                inet->inet_sport, usin->sin_port, sk, 1);
172         if (tmp < 0) {
173                 if (tmp == -ENETUNREACH)
174                         IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
175                 return tmp;
176         }
177
178         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
179                 ip_rt_put(rt);
180                 return -ENETUNREACH;
181         }
182
183         if (!inet->opt || !inet->opt->srr)
184                 daddr = rt->rt_dst;
185
186         if (!inet->inet_saddr)
187                 inet->inet_saddr = rt->rt_src;
188         inet->inet_rcv_saddr = inet->inet_saddr;
189
190         if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
191                 /* Reset inherited state */
192                 tp->rx_opt.ts_recent       = 0;
193                 tp->rx_opt.ts_recent_stamp = 0;
194                 tp->write_seq              = 0;
195         }
196
197         if (tcp_death_row.sysctl_tw_recycle &&
198             !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
199                 struct inet_peer *peer = rt_get_peer(rt);
200                 /*
201                  * VJ's idea. We save last timestamp seen from
202                  * the destination in peer table, when entering state
203                  * TIME-WAIT * and initialize rx_opt.ts_recent from it,
204                  * when trying new connection.
205                  */
206                 if (peer != NULL &&
207                     (u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
208                         tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
209                         tp->rx_opt.ts_recent = peer->tcp_ts;
210                 }
211         }
212
213         inet->inet_dport = usin->sin_port;
214         inet->inet_daddr = daddr;
215
216         inet_csk(sk)->icsk_ext_hdr_len = 0;
217         if (inet->opt)
218                 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
219
220         tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
221
222         /* Socket identity is still unknown (sport may be zero).
223          * However we set state to SYN-SENT and not releasing socket
224          * lock select source port, enter ourselves into the hash tables and
225          * complete initialization after this.
226          */
227         tcp_set_state(sk, TCP_SYN_SENT);
228         err = inet_hash_connect(&tcp_death_row, sk);
229         if (err)
230                 goto failure;
231
232         err = ip_route_newports(&rt, IPPROTO_TCP,
233                                 inet->inet_sport, inet->inet_dport, sk);
234         if (err)
235                 goto failure;
236
237         /* OK, now commit destination to socket.  */
238         sk->sk_gso_type = SKB_GSO_TCPV4;
239         sk_setup_caps(sk, &rt->u.dst);
240
241         if (!tp->write_seq)
242                 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
243                                                            inet->inet_daddr,
244                                                            inet->inet_sport,
245                                                            usin->sin_port);
246
247         inet->inet_id = tp->write_seq ^ jiffies;
248
249         err = tcp_connect(sk);
250         rt = NULL;
251         if (err)
252                 goto failure;
253
254         return 0;
255
256 failure:
257         /*
258          * This unhashes the socket and releases the local port,
259          * if necessary.
260          */
261         tcp_set_state(sk, TCP_CLOSE);
262         ip_rt_put(rt);
263         sk->sk_route_caps = 0;
264         inet->inet_dport = 0;
265         return err;
266 }
267
268 /*
269  * This routine does path mtu discovery as defined in RFC1191.
270  */
271 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
272 {
273         struct dst_entry *dst;
274         struct inet_sock *inet = inet_sk(sk);
275
276         /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
277          * send out by Linux are always <576bytes so they should go through
278          * unfragmented).
279          */
280         if (sk->sk_state == TCP_LISTEN)
281                 return;
282
283         /* We don't check in the destentry if pmtu discovery is forbidden
284          * on this route. We just assume that no packet_to_big packets
285          * are send back when pmtu discovery is not active.
286          * There is a small race when the user changes this flag in the
287          * route, but I think that's acceptable.
288          */
289         if ((dst = __sk_dst_check(sk, 0)) == NULL)
290                 return;
291
292         dst->ops->update_pmtu(dst, mtu);
293
294         /* Something is about to be wrong... Remember soft error
295          * for the case, if this connection will not able to recover.
296          */
297         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
298                 sk->sk_err_soft = EMSGSIZE;
299
300         mtu = dst_mtu(dst);
301
302         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
303             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
304                 tcp_sync_mss(sk, mtu);
305
306                 /* Resend the TCP packet because it's
307                  * clear that the old packet has been
308                  * dropped. This is the new "fast" path mtu
309                  * discovery.
310                  */
311                 tcp_simple_retransmit(sk);
312         } /* else let the usual retransmit timer handle it */
313 }
314
315 /*
316  * This routine is called by the ICMP module when it gets some
317  * sort of error condition.  If err < 0 then the socket should
318  * be closed and the error returned to the user.  If err > 0
319  * it's just the icmp type << 8 | icmp code.  After adjustment
320  * header points to the first 8 bytes of the tcp header.  We need
321  * to find the appropriate port.
322  *
323  * The locking strategy used here is very "optimistic". When
324  * someone else accesses the socket the ICMP is just dropped
325  * and for some paths there is no check at all.
326  * A more general error queue to queue errors for later handling
327  * is probably better.
328  *
329  */
330
331 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
332 {
333         struct iphdr *iph = (struct iphdr *)icmp_skb->data;
334         struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
335         struct inet_connection_sock *icsk;
336         struct tcp_sock *tp;
337         struct inet_sock *inet;
338         const int type = icmp_hdr(icmp_skb)->type;
339         const int code = icmp_hdr(icmp_skb)->code;
340         struct sock *sk;
341         struct sk_buff *skb;
342         __u32 seq;
343         __u32 remaining;
344         int err;
345         struct net *net = dev_net(icmp_skb->dev);
346
347         if (icmp_skb->len < (iph->ihl << 2) + 8) {
348                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
349                 return;
350         }
351
352         sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
353                         iph->saddr, th->source, inet_iif(icmp_skb));
354         if (!sk) {
355                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
356                 return;
357         }
358         if (sk->sk_state == TCP_TIME_WAIT) {
359                 inet_twsk_put(inet_twsk(sk));
360                 return;
361         }
362
363         bh_lock_sock(sk);
364         /* If too many ICMPs get dropped on busy
365          * servers this needs to be solved differently.
366          */
367         if (sock_owned_by_user(sk))
368                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
369
370         if (sk->sk_state == TCP_CLOSE)
371                 goto out;
372
373         icsk = inet_csk(sk);
374         tp = tcp_sk(sk);
375         seq = ntohl(th->seq);
376         if (sk->sk_state != TCP_LISTEN &&
377             !between(seq, tp->snd_una, tp->snd_nxt)) {
378                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
379                 goto out;
380         }
381
382         switch (type) {
383         case ICMP_SOURCE_QUENCH:
384                 /* Just silently ignore these. */
385                 goto out;
386         case ICMP_PARAMETERPROB:
387                 err = EPROTO;
388                 break;
389         case ICMP_DEST_UNREACH:
390                 if (code > NR_ICMP_UNREACH)
391                         goto out;
392
393                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
394                         if (!sock_owned_by_user(sk))
395                                 do_pmtu_discovery(sk, iph, info);
396                         goto out;
397                 }
398
399                 err = icmp_err_convert[code].errno;
400                 /* check if icmp_skb allows revert of backoff
401                  * (see draft-zimmermann-tcp-lcd) */
402                 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
403                         break;
404                 if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
405                     !icsk->icsk_backoff)
406                         break;
407
408                 icsk->icsk_backoff--;
409                 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
410                                          icsk->icsk_backoff;
411                 tcp_bound_rto(sk);
412
413                 skb = tcp_write_queue_head(sk);
414                 BUG_ON(!skb);
415
416                 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
417                                 tcp_time_stamp - TCP_SKB_CB(skb)->when);
418
419                 if (remaining) {
420                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
421                                                   remaining, TCP_RTO_MAX);
422                 } else if (sock_owned_by_user(sk)) {
423                         /* RTO revert clocked out retransmission,
424                          * but socket is locked. Will defer. */
425                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
426                                                   HZ/20, TCP_RTO_MAX);
427                 } else {
428                         /* RTO revert clocked out retransmission.
429                          * Will retransmit now */
430                         tcp_retransmit_timer(sk);
431                 }
432
433                 break;
434         case ICMP_TIME_EXCEEDED:
435                 err = EHOSTUNREACH;
436                 break;
437         default:
438                 goto out;
439         }
440
441         switch (sk->sk_state) {
442                 struct request_sock *req, **prev;
443         case TCP_LISTEN:
444                 if (sock_owned_by_user(sk))
445                         goto out;
446
447                 req = inet_csk_search_req(sk, &prev, th->dest,
448                                           iph->daddr, iph->saddr);
449                 if (!req)
450                         goto out;
451
452                 /* ICMPs are not backlogged, hence we cannot get
453                    an established socket here.
454                  */
455                 WARN_ON(req->sk);
456
457                 if (seq != tcp_rsk(req)->snt_isn) {
458                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
459                         goto out;
460                 }
461
462                 /*
463                  * Still in SYN_RECV, just remove it silently.
464                  * There is no good way to pass the error to the newly
465                  * created socket, and POSIX does not want network
466                  * errors returned from accept().
467                  */
468                 inet_csk_reqsk_queue_drop(sk, req, prev);
469                 goto out;
470
471         case TCP_SYN_SENT:
472         case TCP_SYN_RECV:  /* Cannot happen.
473                                It can f.e. if SYNs crossed.
474                              */
475                 if (!sock_owned_by_user(sk)) {
476                         sk->sk_err = err;
477
478                         sk->sk_error_report(sk);
479
480                         tcp_done(sk);
481                 } else {
482                         sk->sk_err_soft = err;
483                 }
484                 goto out;
485         }
486
487         /* If we've already connected we will keep trying
488          * until we time out, or the user gives up.
489          *
490          * rfc1122 4.2.3.9 allows to consider as hard errors
491          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
492          * but it is obsoleted by pmtu discovery).
493          *
494          * Note, that in modern internet, where routing is unreliable
495          * and in each dark corner broken firewalls sit, sending random
496          * errors ordered by their masters even this two messages finally lose
497          * their original sense (even Linux sends invalid PORT_UNREACHs)
498          *
499          * Now we are in compliance with RFCs.
500          *                                                      --ANK (980905)
501          */
502
503         inet = inet_sk(sk);
504         if (!sock_owned_by_user(sk) && inet->recverr) {
505                 sk->sk_err = err;
506                 sk->sk_error_report(sk);
507         } else  { /* Only an error on timeout */
508                 sk->sk_err_soft = err;
509         }
510
511 out:
512         bh_unlock_sock(sk);
513         sock_put(sk);
514 }
515
516 /* This routine computes an IPv4 TCP checksum. */
517 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
518 {
519         struct inet_sock *inet = inet_sk(sk);
520         struct tcphdr *th = tcp_hdr(skb);
521
522         if (skb->ip_summed == CHECKSUM_PARTIAL) {
523                 th->check = ~tcp_v4_check(len, inet->inet_saddr,
524                                           inet->inet_daddr, 0);
525                 skb->csum_start = skb_transport_header(skb) - skb->head;
526                 skb->csum_offset = offsetof(struct tcphdr, check);
527         } else {
528                 th->check = tcp_v4_check(len, inet->inet_saddr,
529                                          inet->inet_daddr,
530                                          csum_partial(th,
531                                                       th->doff << 2,
532                                                       skb->csum));
533         }
534 }
535
536 int tcp_v4_gso_send_check(struct sk_buff *skb)
537 {
538         const struct iphdr *iph;
539         struct tcphdr *th;
540
541         if (!pskb_may_pull(skb, sizeof(*th)))
542                 return -EINVAL;
543
544         iph = ip_hdr(skb);
545         th = tcp_hdr(skb);
546
547         th->check = 0;
548         th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0);
549         skb->csum_start = skb_transport_header(skb) - skb->head;
550         skb->csum_offset = offsetof(struct tcphdr, check);
551         skb->ip_summed = CHECKSUM_PARTIAL;
552         return 0;
553 }
554
555 /*
556  *      This routine will send an RST to the other tcp.
557  *
558  *      Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
559  *                    for reset.
560  *      Answer: if a packet caused RST, it is not for a socket
561  *              existing in our system, if it is matched to a socket,
562  *              it is just duplicate segment or bug in other side's TCP.
563  *              So that we build reply only basing on parameters
564  *              arrived with segment.
565  *      Exception: precedence violation. We do not implement it in any case.
566  */
567
568 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
569 {
570         struct tcphdr *th = tcp_hdr(skb);
571         struct {
572                 struct tcphdr th;
573 #ifdef CONFIG_TCP_MD5SIG
574                 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
575 #endif
576         } rep;
577         struct ip_reply_arg arg;
578 #ifdef CONFIG_TCP_MD5SIG
579         struct tcp_md5sig_key *key;
580 #endif
581         struct net *net;
582
583         /* Never send a reset in response to a reset. */
584         if (th->rst)
585                 return;
586
587         if (skb_rtable(skb)->rt_type != RTN_LOCAL)
588                 return;
589
590         /* Swap the send and the receive. */
591         memset(&rep, 0, sizeof(rep));
592         rep.th.dest   = th->source;
593         rep.th.source = th->dest;
594         rep.th.doff   = sizeof(struct tcphdr) / 4;
595         rep.th.rst    = 1;
596
597         if (th->ack) {
598                 rep.th.seq = th->ack_seq;
599         } else {
600                 rep.th.ack = 1;
601                 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
602                                        skb->len - (th->doff << 2));
603         }
604
605         memset(&arg, 0, sizeof(arg));
606         arg.iov[0].iov_base = (unsigned char *)&rep;
607         arg.iov[0].iov_len  = sizeof(rep.th);
608
609 #ifdef CONFIG_TCP_MD5SIG
610         key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
611         if (key) {
612                 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
613                                    (TCPOPT_NOP << 16) |
614                                    (TCPOPT_MD5SIG << 8) |
615                                    TCPOLEN_MD5SIG);
616                 /* Update length and the length the header thinks exists */
617                 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
618                 rep.th.doff = arg.iov[0].iov_len / 4;
619
620                 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
621                                      key, ip_hdr(skb)->saddr,
622                                      ip_hdr(skb)->daddr, &rep.th);
623         }
624 #endif
625         arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
626                                       ip_hdr(skb)->saddr, /* XXX */
627                                       arg.iov[0].iov_len, IPPROTO_TCP, 0);
628         arg.csumoffset = offsetof(struct tcphdr, check) / 2;
629         arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
630
631         net = dev_net(skb_dst(skb)->dev);
632         ip_send_reply(net->ipv4.tcp_sock, skb,
633                       &arg, arg.iov[0].iov_len);
634
635         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
636         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
637 }
638
639 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
640    outside socket context is ugly, certainly. What can I do?
641  */
642
643 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
644                             u32 win, u32 ts, int oif,
645                             struct tcp_md5sig_key *key,
646                             int reply_flags)
647 {
648         struct tcphdr *th = tcp_hdr(skb);
649         struct {
650                 struct tcphdr th;
651                 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
652 #ifdef CONFIG_TCP_MD5SIG
653                            + (TCPOLEN_MD5SIG_ALIGNED >> 2)
654 #endif
655                         ];
656         } rep;
657         struct ip_reply_arg arg;
658         struct net *net = dev_net(skb_dst(skb)->dev);
659
660         memset(&rep.th, 0, sizeof(struct tcphdr));
661         memset(&arg, 0, sizeof(arg));
662
663         arg.iov[0].iov_base = (unsigned char *)&rep;
664         arg.iov[0].iov_len  = sizeof(rep.th);
665         if (ts) {
666                 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
667                                    (TCPOPT_TIMESTAMP << 8) |
668                                    TCPOLEN_TIMESTAMP);
669                 rep.opt[1] = htonl(tcp_time_stamp);
670                 rep.opt[2] = htonl(ts);
671                 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
672         }
673
674         /* Swap the send and the receive. */
675         rep.th.dest    = th->source;
676         rep.th.source  = th->dest;
677         rep.th.doff    = arg.iov[0].iov_len / 4;
678         rep.th.seq     = htonl(seq);
679         rep.th.ack_seq = htonl(ack);
680         rep.th.ack     = 1;
681         rep.th.window  = htons(win);
682
683 #ifdef CONFIG_TCP_MD5SIG
684         if (key) {
685                 int offset = (ts) ? 3 : 0;
686
687                 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
688                                           (TCPOPT_NOP << 16) |
689                                           (TCPOPT_MD5SIG << 8) |
690                                           TCPOLEN_MD5SIG);
691                 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
692                 rep.th.doff = arg.iov[0].iov_len/4;
693
694                 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
695                                     key, ip_hdr(skb)->saddr,
696                                     ip_hdr(skb)->daddr, &rep.th);
697         }
698 #endif
699         arg.flags = reply_flags;
700         arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
701                                       ip_hdr(skb)->saddr, /* XXX */
702                                       arg.iov[0].iov_len, IPPROTO_TCP, 0);
703         arg.csumoffset = offsetof(struct tcphdr, check) / 2;
704         if (oif)
705                 arg.bound_dev_if = oif;
706
707         ip_send_reply(net->ipv4.tcp_sock, skb,
708                       &arg, arg.iov[0].iov_len);
709
710         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
711 }
712
713 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
714 {
715         struct inet_timewait_sock *tw = inet_twsk(sk);
716         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
717
718         tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
719                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
720                         tcptw->tw_ts_recent,
721                         tw->tw_bound_dev_if,
722                         tcp_twsk_md5_key(tcptw),
723                         tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
724                         );
725
726         inet_twsk_put(tw);
727 }
728
729 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
730                                   struct request_sock *req)
731 {
732         tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
733                         tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
734                         req->ts_recent,
735                         0,
736                         tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
737                         inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
738 }
739
740 /*
741  *      Send a SYN-ACK after having received a SYN.
742  *      This still operates on a request_sock only, not on a big
743  *      socket.
744  */
745 static int __tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
746                                 struct request_sock *req,
747                                 struct request_values *rvp)
748 {
749         const struct inet_request_sock *ireq = inet_rsk(req);
750         int err = -1;
751         struct sk_buff * skb;
752
753         /* First, grab a route. */
754         if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
755                 return -1;
756
757         skb = tcp_make_synack(sk, dst, req, rvp);
758
759         if (skb) {
760                 struct tcphdr *th = tcp_hdr(skb);
761
762                 th->check = tcp_v4_check(skb->len,
763                                          ireq->loc_addr,
764                                          ireq->rmt_addr,
765                                          csum_partial(th, skb->len,
766                                                       skb->csum));
767
768                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
769                                             ireq->rmt_addr,
770                                             ireq->opt);
771                 err = net_xmit_eval(err);
772         }
773
774         dst_release(dst);
775         return err;
776 }
777
778 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
779                               struct request_values *rvp)
780 {
781         return __tcp_v4_send_synack(sk, NULL, req, rvp);
782 }
783
784 /*
785  *      IPv4 request_sock destructor.
786  */
787 static void tcp_v4_reqsk_destructor(struct request_sock *req)
788 {
789         kfree(inet_rsk(req)->opt);
790 }
791
792 #ifdef CONFIG_SYN_COOKIES
793 static void syn_flood_warning(struct sk_buff *skb)
794 {
795         static unsigned long warntime;
796
797         if (time_after(jiffies, (warntime + HZ * 60))) {
798                 warntime = jiffies;
799                 printk(KERN_INFO
800                        "possible SYN flooding on port %d. Sending cookies.\n",
801                        ntohs(tcp_hdr(skb)->dest));
802         }
803 }
804 #endif
805
806 /*
807  * Save and compile IPv4 options into the request_sock if needed.
808  */
809 static struct ip_options *tcp_v4_save_options(struct sock *sk,
810                                               struct sk_buff *skb)
811 {
812         struct ip_options *opt = &(IPCB(skb)->opt);
813         struct ip_options *dopt = NULL;
814
815         if (opt && opt->optlen) {
816                 int opt_size = optlength(opt);
817                 dopt = kmalloc(opt_size, GFP_ATOMIC);
818                 if (dopt) {
819                         if (ip_options_echo(dopt, skb)) {
820                                 kfree(dopt);
821                                 dopt = NULL;
822                         }
823                 }
824         }
825         return dopt;
826 }
827
828 #ifdef CONFIG_TCP_MD5SIG
829 /*
830  * RFC2385 MD5 checksumming requires a mapping of
831  * IP address->MD5 Key.
832  * We need to maintain these in the sk structure.
833  */
834
835 /* Find the Key structure for an address.  */
836 static struct tcp_md5sig_key *
837                         tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
838 {
839         struct tcp_sock *tp = tcp_sk(sk);
840         int i;
841
842         if (!tp->md5sig_info || !tp->md5sig_info->entries4)
843                 return NULL;
844         for (i = 0; i < tp->md5sig_info->entries4; i++) {
845                 if (tp->md5sig_info->keys4[i].addr == addr)
846                         return &tp->md5sig_info->keys4[i].base;
847         }
848         return NULL;
849 }
850
851 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
852                                          struct sock *addr_sk)
853 {
854         return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
855 }
856
857 EXPORT_SYMBOL(tcp_v4_md5_lookup);
858
859 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
860                                                       struct request_sock *req)
861 {
862         return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
863 }
864
865 /* This can be called on a newly created socket, from other files */
866 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
867                       u8 *newkey, u8 newkeylen)
868 {
869         /* Add Key to the list */
870         struct tcp_md5sig_key *key;
871         struct tcp_sock *tp = tcp_sk(sk);
872         struct tcp4_md5sig_key *keys;
873
874         key = tcp_v4_md5_do_lookup(sk, addr);
875         if (key) {
876                 /* Pre-existing entry - just update that one. */
877                 kfree(key->key);
878                 key->key = newkey;
879                 key->keylen = newkeylen;
880         } else {
881                 struct tcp_md5sig_info *md5sig;
882
883                 if (!tp->md5sig_info) {
884                         tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
885                                                   GFP_ATOMIC);
886                         if (!tp->md5sig_info) {
887                                 kfree(newkey);
888                                 return -ENOMEM;
889                         }
890                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
891                 }
892                 if (tcp_alloc_md5sig_pool(sk) == NULL) {
893                         kfree(newkey);
894                         return -ENOMEM;
895                 }
896                 md5sig = tp->md5sig_info;
897
898                 if (md5sig->alloced4 == md5sig->entries4) {
899                         keys = kmalloc((sizeof(*keys) *
900                                         (md5sig->entries4 + 1)), GFP_ATOMIC);
901                         if (!keys) {
902                                 kfree(newkey);
903                                 tcp_free_md5sig_pool();
904                                 return -ENOMEM;
905                         }
906
907                         if (md5sig->entries4)
908                                 memcpy(keys, md5sig->keys4,
909                                        sizeof(*keys) * md5sig->entries4);
910
911                         /* Free old key list, and reference new one */
912                         kfree(md5sig->keys4);
913                         md5sig->keys4 = keys;
914                         md5sig->alloced4++;
915                 }
916                 md5sig->entries4++;
917                 md5sig->keys4[md5sig->entries4 - 1].addr        = addr;
918                 md5sig->keys4[md5sig->entries4 - 1].base.key    = newkey;
919                 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
920         }
921         return 0;
922 }
923
924 EXPORT_SYMBOL(tcp_v4_md5_do_add);
925
926 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
927                                u8 *newkey, u8 newkeylen)
928 {
929         return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
930                                  newkey, newkeylen);
931 }
932
933 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
934 {
935         struct tcp_sock *tp = tcp_sk(sk);
936         int i;
937
938         for (i = 0; i < tp->md5sig_info->entries4; i++) {
939                 if (tp->md5sig_info->keys4[i].addr == addr) {
940                         /* Free the key */
941                         kfree(tp->md5sig_info->keys4[i].base.key);
942                         tp->md5sig_info->entries4--;
943
944                         if (tp->md5sig_info->entries4 == 0) {
945                                 kfree(tp->md5sig_info->keys4);
946                                 tp->md5sig_info->keys4 = NULL;
947                                 tp->md5sig_info->alloced4 = 0;
948                         } else if (tp->md5sig_info->entries4 != i) {
949                                 /* Need to do some manipulation */
950                                 memmove(&tp->md5sig_info->keys4[i],
951                                         &tp->md5sig_info->keys4[i+1],
952                                         (tp->md5sig_info->entries4 - i) *
953                                          sizeof(struct tcp4_md5sig_key));
954                         }
955                         tcp_free_md5sig_pool();
956                         return 0;
957                 }
958         }
959         return -ENOENT;
960 }
961
962 EXPORT_SYMBOL(tcp_v4_md5_do_del);
963
964 static void tcp_v4_clear_md5_list(struct sock *sk)
965 {
966         struct tcp_sock *tp = tcp_sk(sk);
967
968         /* Free each key, then the set of key keys,
969          * the crypto element, and then decrement our
970          * hold on the last resort crypto.
971          */
972         if (tp->md5sig_info->entries4) {
973                 int i;
974                 for (i = 0; i < tp->md5sig_info->entries4; i++)
975                         kfree(tp->md5sig_info->keys4[i].base.key);
976                 tp->md5sig_info->entries4 = 0;
977                 tcp_free_md5sig_pool();
978         }
979         if (tp->md5sig_info->keys4) {
980                 kfree(tp->md5sig_info->keys4);
981                 tp->md5sig_info->keys4 = NULL;
982                 tp->md5sig_info->alloced4  = 0;
983         }
984 }
985
986 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
987                                  int optlen)
988 {
989         struct tcp_md5sig cmd;
990         struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
991         u8 *newkey;
992
993         if (optlen < sizeof(cmd))
994                 return -EINVAL;
995
996         if (copy_from_user(&cmd, optval, sizeof(cmd)))
997                 return -EFAULT;
998
999         if (sin->sin_family != AF_INET)
1000                 return -EINVAL;
1001
1002         if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1003                 if (!tcp_sk(sk)->md5sig_info)
1004                         return -ENOENT;
1005                 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1006         }
1007
1008         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1009                 return -EINVAL;
1010
1011         if (!tcp_sk(sk)->md5sig_info) {
1012                 struct tcp_sock *tp = tcp_sk(sk);
1013                 struct tcp_md5sig_info *p;
1014
1015                 p = kzalloc(sizeof(*p), sk->sk_allocation);
1016                 if (!p)
1017                         return -EINVAL;
1018
1019                 tp->md5sig_info = p;
1020                 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1021         }
1022
1023         newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1024         if (!newkey)
1025                 return -ENOMEM;
1026         return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1027                                  newkey, cmd.tcpm_keylen);
1028 }
1029
1030 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1031                                         __be32 daddr, __be32 saddr, int nbytes)
1032 {
1033         struct tcp4_pseudohdr *bp;
1034         struct scatterlist sg;
1035
1036         bp = &hp->md5_blk.ip4;
1037
1038         /*
1039          * 1. the TCP pseudo-header (in the order: source IP address,
1040          * destination IP address, zero-padded protocol number, and
1041          * segment length)
1042          */
1043         bp->saddr = saddr;
1044         bp->daddr = daddr;
1045         bp->pad = 0;
1046         bp->protocol = IPPROTO_TCP;
1047         bp->len = cpu_to_be16(nbytes);
1048
1049         sg_init_one(&sg, bp, sizeof(*bp));
1050         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1051 }
1052
1053 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1054                                __be32 daddr, __be32 saddr, struct tcphdr *th)
1055 {
1056         struct tcp_md5sig_pool *hp;
1057         struct hash_desc *desc;
1058
1059         hp = tcp_get_md5sig_pool();
1060         if (!hp)
1061                 goto clear_hash_noput;
1062         desc = &hp->md5_desc;
1063
1064         if (crypto_hash_init(desc))
1065                 goto clear_hash;
1066         if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1067                 goto clear_hash;
1068         if (tcp_md5_hash_header(hp, th))
1069                 goto clear_hash;
1070         if (tcp_md5_hash_key(hp, key))
1071                 goto clear_hash;
1072         if (crypto_hash_final(desc, md5_hash))
1073                 goto clear_hash;
1074
1075         tcp_put_md5sig_pool();
1076         return 0;
1077
1078 clear_hash:
1079         tcp_put_md5sig_pool();
1080 clear_hash_noput:
1081         memset(md5_hash, 0, 16);
1082         return 1;
1083 }
1084
1085 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1086                         struct sock *sk, struct request_sock *req,
1087                         struct sk_buff *skb)
1088 {
1089         struct tcp_md5sig_pool *hp;
1090         struct hash_desc *desc;
1091         struct tcphdr *th = tcp_hdr(skb);
1092         __be32 saddr, daddr;
1093
1094         if (sk) {
1095                 saddr = inet_sk(sk)->inet_saddr;
1096                 daddr = inet_sk(sk)->inet_daddr;
1097         } else if (req) {
1098                 saddr = inet_rsk(req)->loc_addr;
1099                 daddr = inet_rsk(req)->rmt_addr;
1100         } else {
1101                 const struct iphdr *iph = ip_hdr(skb);
1102                 saddr = iph->saddr;
1103                 daddr = iph->daddr;
1104         }
1105
1106         hp = tcp_get_md5sig_pool();
1107         if (!hp)
1108                 goto clear_hash_noput;
1109         desc = &hp->md5_desc;
1110
1111         if (crypto_hash_init(desc))
1112                 goto clear_hash;
1113
1114         if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1115                 goto clear_hash;
1116         if (tcp_md5_hash_header(hp, th))
1117                 goto clear_hash;
1118         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1119                 goto clear_hash;
1120         if (tcp_md5_hash_key(hp, key))
1121                 goto clear_hash;
1122         if (crypto_hash_final(desc, md5_hash))
1123                 goto clear_hash;
1124
1125         tcp_put_md5sig_pool();
1126         return 0;
1127
1128 clear_hash:
1129         tcp_put_md5sig_pool();
1130 clear_hash_noput:
1131         memset(md5_hash, 0, 16);
1132         return 1;
1133 }
1134
1135 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1136
1137 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1138 {
1139         /*
1140          * This gets called for each TCP segment that arrives
1141          * so we want to be efficient.
1142          * We have 3 drop cases:
1143          * o No MD5 hash and one expected.
1144          * o MD5 hash and we're not expecting one.
1145          * o MD5 hash and its wrong.
1146          */
1147         __u8 *hash_location = NULL;
1148         struct tcp_md5sig_key *hash_expected;
1149         const struct iphdr *iph = ip_hdr(skb);
1150         struct tcphdr *th = tcp_hdr(skb);
1151         int genhash;
1152         unsigned char newhash[16];
1153
1154         hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1155         hash_location = tcp_parse_md5sig_option(th);
1156
1157         /* We've parsed the options - do we have a hash? */
1158         if (!hash_expected && !hash_location)
1159                 return 0;
1160
1161         if (hash_expected && !hash_location) {
1162                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1163                 return 1;
1164         }
1165
1166         if (!hash_expected && hash_location) {
1167                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1168                 return 1;
1169         }
1170
1171         /* Okay, so this is hash_expected and hash_location -
1172          * so we need to calculate the checksum.
1173          */
1174         genhash = tcp_v4_md5_hash_skb(newhash,
1175                                       hash_expected,
1176                                       NULL, NULL, skb);
1177
1178         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1179                 if (net_ratelimit()) {
1180                         printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1181                                &iph->saddr, ntohs(th->source),
1182                                &iph->daddr, ntohs(th->dest),
1183                                genhash ? " tcp_v4_calc_md5_hash failed" : "");
1184                 }
1185                 return 1;
1186         }
1187         return 0;
1188 }
1189
1190 #endif
1191
1192 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1193         .family         =       PF_INET,
1194         .obj_size       =       sizeof(struct tcp_request_sock),
1195         .rtx_syn_ack    =       tcp_v4_send_synack,
1196         .send_ack       =       tcp_v4_reqsk_send_ack,
1197         .destructor     =       tcp_v4_reqsk_destructor,
1198         .send_reset     =       tcp_v4_send_reset,
1199 };
1200
1201 #ifdef CONFIG_TCP_MD5SIG
1202 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1203         .md5_lookup     =       tcp_v4_reqsk_md5_lookup,
1204         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1205 };
1206 #endif
1207
1208 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1209         .twsk_obj_size  = sizeof(struct tcp_timewait_sock),
1210         .twsk_unique    = tcp_twsk_unique,
1211         .twsk_destructor= tcp_twsk_destructor,
1212 };
1213
1214 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1215 {
1216         struct tcp_extend_values tmp_ext;
1217         struct tcp_options_received tmp_opt;
1218         u8 *hash_location;
1219         struct request_sock *req;
1220         struct inet_request_sock *ireq;
1221         struct tcp_sock *tp = tcp_sk(sk);
1222         struct dst_entry *dst = NULL;
1223         __be32 saddr = ip_hdr(skb)->saddr;
1224         __be32 daddr = ip_hdr(skb)->daddr;
1225         __u32 isn = TCP_SKB_CB(skb)->when;
1226 #ifdef CONFIG_SYN_COOKIES
1227         int want_cookie = 0;
1228 #else
1229 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1230 #endif
1231
1232         /* Never answer to SYNs send to broadcast or multicast */
1233         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1234                 goto drop;
1235
1236         /* TW buckets are converted to open requests without
1237          * limitations, they conserve resources and peer is
1238          * evidently real one.
1239          */
1240         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1241 #ifdef CONFIG_SYN_COOKIES
1242                 if (sysctl_tcp_syncookies) {
1243                         want_cookie = 1;
1244                 } else
1245 #endif
1246                 goto drop;
1247         }
1248
1249         /* Accept backlog is full. If we have already queued enough
1250          * of warm entries in syn queue, drop request. It is better than
1251          * clogging syn queue with openreqs with exponentially increasing
1252          * timeout.
1253          */
1254         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1255                 goto drop;
1256
1257         req = inet_reqsk_alloc(&tcp_request_sock_ops);
1258         if (!req)
1259                 goto drop;
1260
1261 #ifdef CONFIG_TCP_MD5SIG
1262         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1263 #endif
1264
1265         tcp_clear_options(&tmp_opt);
1266         tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1267         tmp_opt.user_mss  = tp->rx_opt.user_mss;
1268         tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1269
1270         if (tmp_opt.cookie_plus > 0 &&
1271             tmp_opt.saw_tstamp &&
1272             !tp->rx_opt.cookie_out_never &&
1273             (sysctl_tcp_cookie_size > 0 ||
1274              (tp->cookie_values != NULL &&
1275               tp->cookie_values->cookie_desired > 0))) {
1276                 u8 *c;
1277                 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1278                 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1279
1280                 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1281                         goto drop_and_release;
1282
1283                 /* Secret recipe starts with IP addresses */
1284                 *mess++ ^= daddr;
1285                 *mess++ ^= saddr;
1286
1287                 /* plus variable length Initiator Cookie */
1288                 c = (u8 *)mess;
1289                 while (l-- > 0)
1290                         *c++ ^= *hash_location++;
1291
1292 #ifdef CONFIG_SYN_COOKIES
1293                 want_cookie = 0;        /* not our kind of cookie */
1294 #endif
1295                 tmp_ext.cookie_out_never = 0; /* false */
1296                 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1297         } else if (!tp->rx_opt.cookie_in_always) {
1298                 /* redundant indications, but ensure initialization. */
1299                 tmp_ext.cookie_out_never = 1; /* true */
1300                 tmp_ext.cookie_plus = 0;
1301         } else {
1302                 goto drop_and_release;
1303         }
1304         tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1305
1306         if (want_cookie && !tmp_opt.saw_tstamp)
1307                 tcp_clear_options(&tmp_opt);
1308
1309         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1310         tcp_openreq_init(req, &tmp_opt, skb);
1311
1312         ireq = inet_rsk(req);
1313         ireq->loc_addr = daddr;
1314         ireq->rmt_addr = saddr;
1315         ireq->no_srccheck = inet_sk(sk)->transparent;
1316         ireq->opt = tcp_v4_save_options(sk, skb);
1317
1318         if (security_inet_conn_request(sk, skb, req))
1319                 goto drop_and_free;
1320
1321         if (!want_cookie)
1322                 TCP_ECN_create_request(req, tcp_hdr(skb));
1323
1324         if (want_cookie) {
1325 #ifdef CONFIG_SYN_COOKIES
1326                 syn_flood_warning(skb);
1327                 req->cookie_ts = tmp_opt.tstamp_ok;
1328 #endif
1329                 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1330         } else if (!isn) {
1331                 struct inet_peer *peer = NULL;
1332
1333                 /* VJ's idea. We save last timestamp seen
1334                  * from the destination in peer table, when entering
1335                  * state TIME-WAIT, and check against it before
1336                  * accepting new connection request.
1337                  *
1338                  * If "isn" is not zero, this request hit alive
1339                  * timewait bucket, so that all the necessary checks
1340                  * are made in the function processing timewait state.
1341                  */
1342                 if (tmp_opt.saw_tstamp &&
1343                     tcp_death_row.sysctl_tw_recycle &&
1344                     (dst = inet_csk_route_req(sk, req)) != NULL &&
1345                     (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1346                     peer->v4daddr == saddr) {
1347                         if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1348                             (s32)(peer->tcp_ts - req->ts_recent) >
1349                                                         TCP_PAWS_WINDOW) {
1350                                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1351                                 goto drop_and_release;
1352                         }
1353                 }
1354                 /* Kill the following clause, if you dislike this way. */
1355                 else if (!sysctl_tcp_syncookies &&
1356                          (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1357                           (sysctl_max_syn_backlog >> 2)) &&
1358                          (!peer || !peer->tcp_ts_stamp) &&
1359                          (!dst || !dst_metric(dst, RTAX_RTT))) {
1360                         /* Without syncookies last quarter of
1361                          * backlog is filled with destinations,
1362                          * proven to be alive.
1363                          * It means that we continue to communicate
1364                          * to destinations, already remembered
1365                          * to the moment of synflood.
1366                          */
1367                         LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1368                                        &saddr, ntohs(tcp_hdr(skb)->source));
1369                         goto drop_and_release;
1370                 }
1371
1372                 isn = tcp_v4_init_sequence(skb);
1373         }
1374         tcp_rsk(req)->snt_isn = isn;
1375
1376         if (__tcp_v4_send_synack(sk, dst, req,
1377                                  (struct request_values *)&tmp_ext) ||
1378             want_cookie)
1379                 goto drop_and_free;
1380
1381         inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1382         return 0;
1383
1384 drop_and_release:
1385         dst_release(dst);
1386 drop_and_free:
1387         reqsk_free(req);
1388 drop:
1389         return 0;
1390 }
1391
1392
1393 /*
1394  * The three way handshake has completed - we got a valid synack -
1395  * now create the new socket.
1396  */
1397 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1398                                   struct request_sock *req,
1399                                   struct dst_entry *dst)
1400 {
1401         struct inet_request_sock *ireq;
1402         struct inet_sock *newinet;
1403         struct tcp_sock *newtp;
1404         struct sock *newsk;
1405 #ifdef CONFIG_TCP_MD5SIG
1406         struct tcp_md5sig_key *key;
1407 #endif
1408
1409         if (sk_acceptq_is_full(sk))
1410                 goto exit_overflow;
1411
1412         if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1413                 goto exit;
1414
1415         newsk = tcp_create_openreq_child(sk, req, skb);
1416         if (!newsk)
1417                 goto exit;
1418
1419         newsk->sk_gso_type = SKB_GSO_TCPV4;
1420         sk_setup_caps(newsk, dst);
1421
1422         newtp                 = tcp_sk(newsk);
1423         newinet               = inet_sk(newsk);
1424         ireq                  = inet_rsk(req);
1425         newinet->inet_daddr   = ireq->rmt_addr;
1426         newinet->inet_rcv_saddr = ireq->loc_addr;
1427         newinet->inet_saddr           = ireq->loc_addr;
1428         newinet->opt          = ireq->opt;
1429         ireq->opt             = NULL;
1430         newinet->mc_index     = inet_iif(skb);
1431         newinet->mc_ttl       = ip_hdr(skb)->ttl;
1432         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1433         if (newinet->opt)
1434                 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
1435         newinet->inet_id = newtp->write_seq ^ jiffies;
1436
1437         tcp_mtup_init(newsk);
1438         tcp_sync_mss(newsk, dst_mtu(dst));
1439         newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1440         if (tcp_sk(sk)->rx_opt.user_mss &&
1441             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1442                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1443
1444         tcp_initialize_rcv_mss(newsk);
1445
1446 #ifdef CONFIG_TCP_MD5SIG
1447         /* Copy over the MD5 key from the original socket */
1448         key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1449         if (key != NULL) {
1450                 /*
1451                  * We're using one, so create a matching key
1452                  * on the newsk structure. If we fail to get
1453                  * memory, then we end up not copying the key
1454                  * across. Shucks.
1455                  */
1456                 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1457                 if (newkey != NULL)
1458                         tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1459                                           newkey, key->keylen);
1460                 newsk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1461         }
1462 #endif
1463
1464         __inet_hash_nolisten(newsk, NULL);
1465         __inet_inherit_port(sk, newsk);
1466
1467         return newsk;
1468
1469 exit_overflow:
1470         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1471 exit:
1472         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1473         dst_release(dst);
1474         return NULL;
1475 }
1476
1477 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1478 {
1479         struct tcphdr *th = tcp_hdr(skb);
1480         const struct iphdr *iph = ip_hdr(skb);
1481         struct sock *nsk;
1482         struct request_sock **prev;
1483         /* Find possible connection requests. */
1484         struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1485                                                        iph->saddr, iph->daddr);
1486         if (req)
1487                 return tcp_check_req(sk, skb, req, prev);
1488
1489         nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1490                         th->source, iph->daddr, th->dest, inet_iif(skb));
1491
1492         if (nsk) {
1493                 if (nsk->sk_state != TCP_TIME_WAIT) {
1494                         bh_lock_sock(nsk);
1495                         return nsk;
1496                 }
1497                 inet_twsk_put(inet_twsk(nsk));
1498                 return NULL;
1499         }
1500
1501 #ifdef CONFIG_SYN_COOKIES
1502         if (!th->rst && !th->syn && th->ack)
1503                 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1504 #endif
1505         return sk;
1506 }
1507
1508 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1509 {
1510         const struct iphdr *iph = ip_hdr(skb);
1511
1512         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1513                 if (!tcp_v4_check(skb->len, iph->saddr,
1514                                   iph->daddr, skb->csum)) {
1515                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1516                         return 0;
1517                 }
1518         }
1519
1520         skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1521                                        skb->len, IPPROTO_TCP, 0);
1522
1523         if (skb->len <= 76) {
1524                 return __skb_checksum_complete(skb);
1525         }
1526         return 0;
1527 }
1528
1529
1530 /* The socket must have it's spinlock held when we get
1531  * here.
1532  *
1533  * We have a potential double-lock case here, so even when
1534  * doing backlog processing we use the BH locking scheme.
1535  * This is because we cannot sleep with the original spinlock
1536  * held.
1537  */
1538 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1539 {
1540         struct sock *rsk;
1541 #ifdef CONFIG_TCP_MD5SIG
1542         /*
1543          * We really want to reject the packet as early as possible
1544          * if:
1545          *  o We're expecting an MD5'd packet and this is no MD5 tcp option
1546          *  o There is an MD5 option and we're not expecting one
1547          */
1548         if (tcp_v4_inbound_md5_hash(sk, skb))
1549                 goto discard;
1550 #endif
1551
1552         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1553                 TCP_CHECK_TIMER(sk);
1554                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1555                         rsk = sk;
1556                         goto reset;
1557                 }
1558                 TCP_CHECK_TIMER(sk);
1559                 return 0;
1560         }
1561
1562         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1563                 goto csum_err;
1564
1565         if (sk->sk_state == TCP_LISTEN) {
1566                 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1567                 if (!nsk)
1568                         goto discard;
1569
1570                 if (nsk != sk) {
1571                         if (tcp_child_process(sk, nsk, skb)) {
1572                                 rsk = nsk;
1573                                 goto reset;
1574                         }
1575                         return 0;
1576                 }
1577         }
1578
1579         TCP_CHECK_TIMER(sk);
1580         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1581                 rsk = sk;
1582                 goto reset;
1583         }
1584         TCP_CHECK_TIMER(sk);
1585         return 0;
1586
1587 reset:
1588         tcp_v4_send_reset(rsk, skb);
1589 discard:
1590         kfree_skb(skb);
1591         /* Be careful here. If this function gets more complicated and
1592          * gcc suffers from register pressure on the x86, sk (in %ebx)
1593          * might be destroyed here. This current version compiles correctly,
1594          * but you have been warned.
1595          */
1596         return 0;
1597
1598 csum_err:
1599         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1600         goto discard;
1601 }
1602
1603 /*
1604  *      From tcp_input.c
1605  */
1606
1607 int tcp_v4_rcv(struct sk_buff *skb)
1608 {
1609         const struct iphdr *iph;
1610         struct tcphdr *th;
1611         struct sock *sk;
1612         int ret;
1613         struct net *net = dev_net(skb->dev);
1614
1615         if (skb->pkt_type != PACKET_HOST)
1616                 goto discard_it;
1617
1618         /* Count it even if it's bad */
1619         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1620
1621         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1622                 goto discard_it;
1623
1624         th = tcp_hdr(skb);
1625
1626         if (th->doff < sizeof(struct tcphdr) / 4)
1627                 goto bad_packet;
1628         if (!pskb_may_pull(skb, th->doff * 4))
1629                 goto discard_it;
1630
1631         /* An explanation is required here, I think.
1632          * Packet length and doff are validated by header prediction,
1633          * provided case of th->doff==0 is eliminated.
1634          * So, we defer the checks. */
1635         if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1636                 goto bad_packet;
1637
1638         th = tcp_hdr(skb);
1639         iph = ip_hdr(skb);
1640         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1641         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1642                                     skb->len - th->doff * 4);
1643         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1644         TCP_SKB_CB(skb)->when    = 0;
1645         TCP_SKB_CB(skb)->flags   = iph->tos;
1646         TCP_SKB_CB(skb)->sacked  = 0;
1647
1648         sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1649         if (!sk)
1650                 goto no_tcp_socket;
1651
1652 process:
1653         if (sk->sk_state == TCP_TIME_WAIT)
1654                 goto do_time_wait;
1655
1656         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1657                 goto discard_and_relse;
1658         nf_reset(skb);
1659
1660         if (sk_filter(sk, skb))
1661                 goto discard_and_relse;
1662
1663         skb->dev = NULL;
1664
1665         bh_lock_sock_nested(sk);
1666         ret = 0;
1667         if (!sock_owned_by_user(sk)) {
1668 #ifdef CONFIG_NET_DMA
1669                 struct tcp_sock *tp = tcp_sk(sk);
1670                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1671                         tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1672                 if (tp->ucopy.dma_chan)
1673                         ret = tcp_v4_do_rcv(sk, skb);
1674                 else
1675 #endif
1676                 {
1677                         if (!tcp_prequeue(sk, skb))
1678                                 ret = tcp_v4_do_rcv(sk, skb);
1679                 }
1680         } else
1681                 sk_add_backlog(sk, skb);
1682         bh_unlock_sock(sk);
1683
1684         sock_put(sk);
1685
1686         return ret;
1687
1688 no_tcp_socket:
1689         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1690                 goto discard_it;
1691
1692         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1693 bad_packet:
1694                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1695         } else {
1696                 tcp_v4_send_reset(NULL, skb);
1697         }
1698
1699 discard_it:
1700         /* Discard frame. */
1701         kfree_skb(skb);
1702         return 0;
1703
1704 discard_and_relse:
1705         sock_put(sk);
1706         goto discard_it;
1707
1708 do_time_wait:
1709         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1710                 inet_twsk_put(inet_twsk(sk));
1711                 goto discard_it;
1712         }
1713
1714         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1715                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1716                 inet_twsk_put(inet_twsk(sk));
1717                 goto discard_it;
1718         }
1719         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1720         case TCP_TW_SYN: {
1721                 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1722                                                         &tcp_hashinfo,
1723                                                         iph->daddr, th->dest,
1724                                                         inet_iif(skb));
1725                 if (sk2) {
1726                         inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1727                         inet_twsk_put(inet_twsk(sk));
1728                         sk = sk2;
1729                         goto process;
1730                 }
1731                 /* Fall through to ACK */
1732         }
1733         case TCP_TW_ACK:
1734                 tcp_v4_timewait_ack(sk, skb);
1735                 break;
1736         case TCP_TW_RST:
1737                 goto no_tcp_socket;
1738         case TCP_TW_SUCCESS:;
1739         }
1740         goto discard_it;
1741 }
1742
1743 /* VJ's idea. Save last timestamp seen from this destination
1744  * and hold it at least for normal timewait interval to use for duplicate
1745  * segment detection in subsequent connections, before they enter synchronized
1746  * state.
1747  */
1748
1749 int tcp_v4_remember_stamp(struct sock *sk)
1750 {
1751         struct inet_sock *inet = inet_sk(sk);
1752         struct tcp_sock *tp = tcp_sk(sk);
1753         struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1754         struct inet_peer *peer = NULL;
1755         int release_it = 0;
1756
1757         if (!rt || rt->rt_dst != inet->inet_daddr) {
1758                 peer = inet_getpeer(inet->inet_daddr, 1);
1759                 release_it = 1;
1760         } else {
1761                 if (!rt->peer)
1762                         rt_bind_peer(rt, 1);
1763                 peer = rt->peer;
1764         }
1765
1766         if (peer) {
1767                 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1768                     ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1769                      peer->tcp_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
1770                         peer->tcp_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
1771                         peer->tcp_ts = tp->rx_opt.ts_recent;
1772                 }
1773                 if (release_it)
1774                         inet_putpeer(peer);
1775                 return 1;
1776         }
1777
1778         return 0;
1779 }
1780
1781 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1782 {
1783         struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1784
1785         if (peer) {
1786                 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1787
1788                 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1789                     ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1790                      peer->tcp_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
1791                         peer->tcp_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
1792                         peer->tcp_ts       = tcptw->tw_ts_recent;
1793                 }
1794                 inet_putpeer(peer);
1795                 return 1;
1796         }
1797
1798         return 0;
1799 }
1800
1801 const struct inet_connection_sock_af_ops ipv4_specific = {
1802         .queue_xmit        = ip_queue_xmit,
1803         .send_check        = tcp_v4_send_check,
1804         .rebuild_header    = inet_sk_rebuild_header,
1805         .conn_request      = tcp_v4_conn_request,
1806         .syn_recv_sock     = tcp_v4_syn_recv_sock,
1807         .remember_stamp    = tcp_v4_remember_stamp,
1808         .net_header_len    = sizeof(struct iphdr),
1809         .setsockopt        = ip_setsockopt,
1810         .getsockopt        = ip_getsockopt,
1811         .addr2sockaddr     = inet_csk_addr2sockaddr,
1812         .sockaddr_len      = sizeof(struct sockaddr_in),
1813         .bind_conflict     = inet_csk_bind_conflict,
1814 #ifdef CONFIG_COMPAT
1815         .compat_setsockopt = compat_ip_setsockopt,
1816         .compat_getsockopt = compat_ip_getsockopt,
1817 #endif
1818 };
1819
1820 #ifdef CONFIG_TCP_MD5SIG
1821 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1822         .md5_lookup             = tcp_v4_md5_lookup,
1823         .calc_md5_hash          = tcp_v4_md5_hash_skb,
1824         .md5_add                = tcp_v4_md5_add_func,
1825         .md5_parse              = tcp_v4_parse_md5_keys,
1826 };
1827 #endif
1828
1829 /* NOTE: A lot of things set to zero explicitly by call to
1830  *       sk_alloc() so need not be done here.
1831  */
1832 static int tcp_v4_init_sock(struct sock *sk)
1833 {
1834         struct inet_connection_sock *icsk = inet_csk(sk);
1835         struct tcp_sock *tp = tcp_sk(sk);
1836
1837         skb_queue_head_init(&tp->out_of_order_queue);
1838         tcp_init_xmit_timers(sk);
1839         tcp_prequeue_init(tp);
1840
1841         icsk->icsk_rto = TCP_TIMEOUT_INIT;
1842         tp->mdev = TCP_TIMEOUT_INIT;
1843
1844         /* So many TCP implementations out there (incorrectly) count the
1845          * initial SYN frame in their delayed-ACK and congestion control
1846          * algorithms that we must have the following bandaid to talk
1847          * efficiently to them.  -DaveM
1848          */
1849         tp->snd_cwnd = 2;
1850
1851         /* See draft-stevens-tcpca-spec-01 for discussion of the
1852          * initialization of these values.
1853          */
1854         tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1855         tp->snd_cwnd_clamp = ~0;
1856         tp->mss_cache = TCP_MSS_DEFAULT;
1857
1858         tp->reordering = sysctl_tcp_reordering;
1859         icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1860
1861         sk->sk_state = TCP_CLOSE;
1862
1863         sk->sk_write_space = sk_stream_write_space;
1864         sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1865
1866         icsk->icsk_af_ops = &ipv4_specific;
1867         icsk->icsk_sync_mss = tcp_sync_mss;
1868 #ifdef CONFIG_TCP_MD5SIG
1869         tp->af_specific = &tcp_sock_ipv4_specific;
1870 #endif
1871
1872         /* TCP Cookie Transactions */
1873         if (sysctl_tcp_cookie_size > 0) {
1874                 /* Default, cookies without s_data_payload. */
1875                 tp->cookie_values =
1876                         kzalloc(sizeof(*tp->cookie_values),
1877                                 sk->sk_allocation);
1878                 if (tp->cookie_values != NULL)
1879                         kref_init(&tp->cookie_values->kref);
1880         }
1881         /* Presumed zeroed, in order of appearance:
1882          *      cookie_in_always, cookie_out_never,
1883          *      s_data_constant, s_data_in, s_data_out
1884          */
1885         sk->sk_sndbuf = sysctl_tcp_wmem[1];
1886         sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1887
1888         local_bh_disable();
1889         percpu_counter_inc(&tcp_sockets_allocated);
1890         local_bh_enable();
1891
1892         return 0;
1893 }
1894
1895 void tcp_v4_destroy_sock(struct sock *sk)
1896 {
1897         struct tcp_sock *tp = tcp_sk(sk);
1898
1899         tcp_clear_xmit_timers(sk);
1900
1901         tcp_cleanup_congestion_control(sk);
1902
1903         /* Cleanup up the write buffer. */
1904         tcp_write_queue_purge(sk);
1905
1906         /* Cleans up our, hopefully empty, out_of_order_queue. */
1907         __skb_queue_purge(&tp->out_of_order_queue);
1908
1909 #ifdef CONFIG_TCP_MD5SIG
1910         /* Clean up the MD5 key list, if any */
1911         if (tp->md5sig_info) {
1912                 tcp_v4_clear_md5_list(sk);
1913                 kfree(tp->md5sig_info);
1914                 tp->md5sig_info = NULL;
1915         }
1916 #endif
1917
1918 #ifdef CONFIG_NET_DMA
1919         /* Cleans up our sk_async_wait_queue */
1920         __skb_queue_purge(&sk->sk_async_wait_queue);
1921 #endif
1922
1923         /* Clean prequeue, it must be empty really */
1924         __skb_queue_purge(&tp->ucopy.prequeue);
1925
1926         /* Clean up a referenced TCP bind bucket. */
1927         if (inet_csk(sk)->icsk_bind_hash)
1928                 inet_put_port(sk);
1929
1930         /*
1931          * If sendmsg cached page exists, toss it.
1932          */
1933         if (sk->sk_sndmsg_page) {
1934                 __free_page(sk->sk_sndmsg_page);
1935                 sk->sk_sndmsg_page = NULL;
1936         }
1937
1938         /* TCP Cookie Transactions */
1939         if (tp->cookie_values != NULL) {
1940                 kref_put(&tp->cookie_values->kref,
1941                          tcp_cookie_values_release);
1942                 tp->cookie_values = NULL;
1943         }
1944
1945         percpu_counter_dec(&tcp_sockets_allocated);
1946 }
1947
1948 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1949
1950 #ifdef CONFIG_PROC_FS
1951 /* Proc filesystem TCP sock list dumping. */
1952
1953 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1954 {
1955         return hlist_nulls_empty(head) ? NULL :
1956                 list_entry(head->first, struct inet_timewait_sock, tw_node);
1957 }
1958
1959 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1960 {
1961         return !is_a_nulls(tw->tw_node.next) ?
1962                 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1963 }
1964
1965 static void *listening_get_next(struct seq_file *seq, void *cur)
1966 {
1967         struct inet_connection_sock *icsk;
1968         struct hlist_nulls_node *node;
1969         struct sock *sk = cur;
1970         struct inet_listen_hashbucket *ilb;
1971         struct tcp_iter_state *st = seq->private;
1972         struct net *net = seq_file_net(seq);
1973
1974         if (!sk) {
1975                 st->bucket = 0;
1976                 ilb = &tcp_hashinfo.listening_hash[0];
1977                 spin_lock_bh(&ilb->lock);
1978                 sk = sk_nulls_head(&ilb->head);
1979                 goto get_sk;
1980         }
1981         ilb = &tcp_hashinfo.listening_hash[st->bucket];
1982         ++st->num;
1983
1984         if (st->state == TCP_SEQ_STATE_OPENREQ) {
1985                 struct request_sock *req = cur;
1986
1987                 icsk = inet_csk(st->syn_wait_sk);
1988                 req = req->dl_next;
1989                 while (1) {
1990                         while (req) {
1991                                 if (req->rsk_ops->family == st->family) {
1992                                         cur = req;
1993                                         goto out;
1994                                 }
1995                                 req = req->dl_next;
1996                         }
1997                         if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1998                                 break;
1999 get_req:
2000                         req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2001                 }
2002                 sk        = sk_next(st->syn_wait_sk);
2003                 st->state = TCP_SEQ_STATE_LISTENING;
2004                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2005         } else {
2006                 icsk = inet_csk(sk);
2007                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2008                 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2009                         goto start_req;
2010                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2011                 sk = sk_next(sk);
2012         }
2013 get_sk:
2014         sk_nulls_for_each_from(sk, node) {
2015                 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) {
2016                         cur = sk;
2017                         goto out;
2018                 }
2019                 icsk = inet_csk(sk);
2020                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2021                 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2022 start_req:
2023                         st->uid         = sock_i_uid(sk);
2024                         st->syn_wait_sk = sk;
2025                         st->state       = TCP_SEQ_STATE_OPENREQ;
2026                         st->sbucket     = 0;
2027                         goto get_req;
2028                 }
2029                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2030         }
2031         spin_unlock_bh(&ilb->lock);
2032         if (++st->bucket < INET_LHTABLE_SIZE) {
2033                 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2034                 spin_lock_bh(&ilb->lock);
2035                 sk = sk_nulls_head(&ilb->head);
2036                 goto get_sk;
2037         }
2038         cur = NULL;
2039 out:
2040         return cur;
2041 }
2042
2043 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2044 {
2045         void *rc = listening_get_next(seq, NULL);
2046
2047         while (rc && *pos) {
2048                 rc = listening_get_next(seq, rc);
2049                 --*pos;
2050         }
2051         return rc;
2052 }
2053
2054 static inline int empty_bucket(struct tcp_iter_state *st)
2055 {
2056         return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2057                 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2058 }
2059
2060 static void *established_get_first(struct seq_file *seq)
2061 {
2062         struct tcp_iter_state *st = seq->private;
2063         struct net *net = seq_file_net(seq);
2064         void *rc = NULL;
2065
2066         for (st->bucket = 0; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2067                 struct sock *sk;
2068                 struct hlist_nulls_node *node;
2069                 struct inet_timewait_sock *tw;
2070                 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2071
2072                 /* Lockless fast path for the common case of empty buckets */
2073                 if (empty_bucket(st))
2074                         continue;
2075
2076                 spin_lock_bh(lock);
2077                 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2078                         if (sk->sk_family != st->family ||
2079                             !net_eq(sock_net(sk), net)) {
2080                                 continue;
2081                         }
2082                         rc = sk;
2083                         goto out;
2084                 }
2085                 st->state = TCP_SEQ_STATE_TIME_WAIT;
2086                 inet_twsk_for_each(tw, node,
2087                                    &tcp_hashinfo.ehash[st->bucket].twchain) {
2088                         if (tw->tw_family != st->family ||
2089                             !net_eq(twsk_net(tw), net)) {
2090                                 continue;
2091                         }
2092                         rc = tw;
2093                         goto out;
2094                 }
2095                 spin_unlock_bh(lock);
2096                 st->state = TCP_SEQ_STATE_ESTABLISHED;
2097         }
2098 out:
2099         return rc;
2100 }
2101
2102 static void *established_get_next(struct seq_file *seq, void *cur)
2103 {
2104         struct sock *sk = cur;
2105         struct inet_timewait_sock *tw;
2106         struct hlist_nulls_node *node;
2107         struct tcp_iter_state *st = seq->private;
2108         struct net *net = seq_file_net(seq);
2109
2110         ++st->num;
2111
2112         if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2113                 tw = cur;
2114                 tw = tw_next(tw);
2115 get_tw:
2116                 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2117                         tw = tw_next(tw);
2118                 }
2119                 if (tw) {
2120                         cur = tw;
2121                         goto out;
2122                 }
2123                 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2124                 st->state = TCP_SEQ_STATE_ESTABLISHED;
2125
2126                 /* Look for next non empty bucket */
2127                 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2128                                 empty_bucket(st))
2129                         ;
2130                 if (st->bucket > tcp_hashinfo.ehash_mask)
2131                         return NULL;
2132
2133                 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2134                 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2135         } else
2136                 sk = sk_nulls_next(sk);
2137
2138         sk_nulls_for_each_from(sk, node) {
2139                 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2140                         goto found;
2141         }
2142
2143         st->state = TCP_SEQ_STATE_TIME_WAIT;
2144         tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2145         goto get_tw;
2146 found:
2147         cur = sk;
2148 out:
2149         return cur;
2150 }
2151
2152 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2153 {
2154         void *rc = established_get_first(seq);
2155
2156         while (rc && pos) {
2157                 rc = established_get_next(seq, rc);
2158                 --pos;
2159         }
2160         return rc;
2161 }
2162
2163 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2164 {
2165         void *rc;
2166         struct tcp_iter_state *st = seq->private;
2167
2168         st->state = TCP_SEQ_STATE_LISTENING;
2169         rc        = listening_get_idx(seq, &pos);
2170
2171         if (!rc) {
2172                 st->state = TCP_SEQ_STATE_ESTABLISHED;
2173                 rc        = established_get_idx(seq, pos);
2174         }
2175
2176         return rc;
2177 }
2178
2179 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2180 {
2181         struct tcp_iter_state *st = seq->private;
2182         st->state = TCP_SEQ_STATE_LISTENING;
2183         st->num = 0;
2184         return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2185 }
2186
2187 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2188 {
2189         void *rc = NULL;
2190         struct tcp_iter_state *st;
2191
2192         if (v == SEQ_START_TOKEN) {
2193                 rc = tcp_get_idx(seq, 0);
2194                 goto out;
2195         }
2196         st = seq->private;
2197
2198         switch (st->state) {
2199         case TCP_SEQ_STATE_OPENREQ:
2200         case TCP_SEQ_STATE_LISTENING:
2201                 rc = listening_get_next(seq, v);
2202                 if (!rc) {
2203                         st->state = TCP_SEQ_STATE_ESTABLISHED;
2204                         rc        = established_get_first(seq);
2205                 }
2206                 break;
2207         case TCP_SEQ_STATE_ESTABLISHED:
2208         case TCP_SEQ_STATE_TIME_WAIT:
2209                 rc = established_get_next(seq, v);
2210                 break;
2211         }
2212 out:
2213         ++*pos;
2214         return rc;
2215 }
2216
2217 static void tcp_seq_stop(struct seq_file *seq, void *v)
2218 {
2219         struct tcp_iter_state *st = seq->private;
2220
2221         switch (st->state) {
2222         case TCP_SEQ_STATE_OPENREQ:
2223                 if (v) {
2224                         struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2225                         read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2226                 }
2227         case TCP_SEQ_STATE_LISTENING:
2228                 if (v != SEQ_START_TOKEN)
2229                         spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2230                 break;
2231         case TCP_SEQ_STATE_TIME_WAIT:
2232         case TCP_SEQ_STATE_ESTABLISHED:
2233                 if (v)
2234                         spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2235                 break;
2236         }
2237 }
2238
2239 static int tcp_seq_open(struct inode *inode, struct file *file)
2240 {
2241         struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2242         struct tcp_iter_state *s;
2243         int err;
2244
2245         err = seq_open_net(inode, file, &afinfo->seq_ops,
2246                           sizeof(struct tcp_iter_state));
2247         if (err < 0)
2248                 return err;
2249
2250         s = ((struct seq_file *)file->private_data)->private;
2251         s->family               = afinfo->family;
2252         return 0;
2253 }
2254
2255 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2256 {
2257         int rc = 0;
2258         struct proc_dir_entry *p;
2259
2260         afinfo->seq_fops.open           = tcp_seq_open;
2261         afinfo->seq_fops.read           = seq_read;
2262         afinfo->seq_fops.llseek         = seq_lseek;
2263         afinfo->seq_fops.release        = seq_release_net;
2264
2265         afinfo->seq_ops.start           = tcp_seq_start;
2266         afinfo->seq_ops.next            = tcp_seq_next;
2267         afinfo->seq_ops.stop            = tcp_seq_stop;
2268
2269         p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2270                              &afinfo->seq_fops, afinfo);
2271         if (!p)
2272                 rc = -ENOMEM;
2273         return rc;
2274 }
2275
2276 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2277 {
2278         proc_net_remove(net, afinfo->name);
2279 }
2280
2281 static void get_openreq4(struct sock *sk, struct request_sock *req,
2282                          struct seq_file *f, int i, int uid, int *len)
2283 {
2284         const struct inet_request_sock *ireq = inet_rsk(req);
2285         int ttd = req->expires - jiffies;
2286
2287         seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2288                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
2289                 i,
2290                 ireq->loc_addr,
2291                 ntohs(inet_sk(sk)->inet_sport),
2292                 ireq->rmt_addr,
2293                 ntohs(ireq->rmt_port),
2294                 TCP_SYN_RECV,
2295                 0, 0, /* could print option size, but that is af dependent. */
2296                 1,    /* timers active (only the expire timer) */
2297                 jiffies_to_clock_t(ttd),
2298                 req->retrans,
2299                 uid,
2300                 0,  /* non standard timer */
2301                 0, /* open_requests have no inode */
2302                 atomic_read(&sk->sk_refcnt),
2303                 req,
2304                 len);
2305 }
2306
2307 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2308 {
2309         int timer_active;
2310         unsigned long timer_expires;
2311         struct tcp_sock *tp = tcp_sk(sk);
2312         const struct inet_connection_sock *icsk = inet_csk(sk);
2313         struct inet_sock *inet = inet_sk(sk);
2314         __be32 dest = inet->inet_daddr;
2315         __be32 src = inet->inet_rcv_saddr;
2316         __u16 destp = ntohs(inet->inet_dport);
2317         __u16 srcp = ntohs(inet->inet_sport);
2318         int rx_queue;
2319
2320         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2321                 timer_active    = 1;
2322                 timer_expires   = icsk->icsk_timeout;
2323         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2324                 timer_active    = 4;
2325                 timer_expires   = icsk->icsk_timeout;
2326         } else if (timer_pending(&sk->sk_timer)) {
2327                 timer_active    = 2;
2328                 timer_expires   = sk->sk_timer.expires;
2329         } else {
2330                 timer_active    = 0;
2331                 timer_expires = jiffies;
2332         }
2333
2334         if (sk->sk_state == TCP_LISTEN)
2335                 rx_queue = sk->sk_ack_backlog;
2336         else
2337                 /*
2338                  * because we dont lock socket, we might find a transient negative value
2339                  */
2340                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2341
2342         seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2343                         "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
2344                 i, src, srcp, dest, destp, sk->sk_state,
2345                 tp->write_seq - tp->snd_una,
2346                 rx_queue,
2347                 timer_active,
2348                 jiffies_to_clock_t(timer_expires - jiffies),
2349                 icsk->icsk_retransmits,
2350                 sock_i_uid(sk),
2351                 icsk->icsk_probes_out,
2352                 sock_i_ino(sk),
2353                 atomic_read(&sk->sk_refcnt), sk,
2354                 jiffies_to_clock_t(icsk->icsk_rto),
2355                 jiffies_to_clock_t(icsk->icsk_ack.ato),
2356                 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2357                 tp->snd_cwnd,
2358                 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2359                 len);
2360 }
2361
2362 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2363                                struct seq_file *f, int i, int *len)
2364 {
2365         __be32 dest, src;
2366         __u16 destp, srcp;
2367         int ttd = tw->tw_ttd - jiffies;
2368
2369         if (ttd < 0)
2370                 ttd = 0;
2371
2372         dest  = tw->tw_daddr;
2373         src   = tw->tw_rcv_saddr;
2374         destp = ntohs(tw->tw_dport);
2375         srcp  = ntohs(tw->tw_sport);
2376
2377         seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2378                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
2379                 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2380                 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2381                 atomic_read(&tw->tw_refcnt), tw, len);
2382 }
2383
2384 #define TMPSZ 150
2385
2386 static int tcp4_seq_show(struct seq_file *seq, void *v)
2387 {
2388         struct tcp_iter_state *st;
2389         int len;
2390
2391         if (v == SEQ_START_TOKEN) {
2392                 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2393                            "  sl  local_address rem_address   st tx_queue "
2394                            "rx_queue tr tm->when retrnsmt   uid  timeout "
2395                            "inode");
2396                 goto out;
2397         }
2398         st = seq->private;
2399
2400         switch (st->state) {
2401         case TCP_SEQ_STATE_LISTENING:
2402         case TCP_SEQ_STATE_ESTABLISHED:
2403                 get_tcp4_sock(v, seq, st->num, &len);
2404                 break;
2405         case TCP_SEQ_STATE_OPENREQ:
2406                 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2407                 break;
2408         case TCP_SEQ_STATE_TIME_WAIT:
2409                 get_timewait4_sock(v, seq, st->num, &len);
2410                 break;
2411         }
2412         seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2413 out:
2414         return 0;
2415 }
2416
2417 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2418         .name           = "tcp",
2419         .family         = AF_INET,
2420         .seq_fops       = {
2421                 .owner          = THIS_MODULE,
2422         },
2423         .seq_ops        = {
2424                 .show           = tcp4_seq_show,
2425         },
2426 };
2427
2428 static int tcp4_proc_init_net(struct net *net)
2429 {
2430         return tcp_proc_register(net, &tcp4_seq_afinfo);
2431 }
2432
2433 static void tcp4_proc_exit_net(struct net *net)
2434 {
2435         tcp_proc_unregister(net, &tcp4_seq_afinfo);
2436 }
2437
2438 static struct pernet_operations tcp4_net_ops = {
2439         .init = tcp4_proc_init_net,
2440         .exit = tcp4_proc_exit_net,
2441 };
2442
2443 int __init tcp4_proc_init(void)
2444 {
2445         return register_pernet_subsys(&tcp4_net_ops);
2446 }
2447
2448 void tcp4_proc_exit(void)
2449 {
2450         unregister_pernet_subsys(&tcp4_net_ops);
2451 }
2452 #endif /* CONFIG_PROC_FS */
2453
2454 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2455 {
2456         struct iphdr *iph = skb_gro_network_header(skb);
2457
2458         switch (skb->ip_summed) {
2459         case CHECKSUM_COMPLETE:
2460                 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2461                                   skb->csum)) {
2462                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2463                         break;
2464                 }
2465
2466                 /* fall through */
2467         case CHECKSUM_NONE:
2468                 NAPI_GRO_CB(skb)->flush = 1;
2469                 return NULL;
2470         }
2471
2472         return tcp_gro_receive(head, skb);
2473 }
2474 EXPORT_SYMBOL(tcp4_gro_receive);
2475
2476 int tcp4_gro_complete(struct sk_buff *skb)
2477 {
2478         struct iphdr *iph = ip_hdr(skb);
2479         struct tcphdr *th = tcp_hdr(skb);
2480
2481         th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2482                                   iph->saddr, iph->daddr, 0);
2483         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2484
2485         return tcp_gro_complete(skb);
2486 }
2487 EXPORT_SYMBOL(tcp4_gro_complete);
2488
2489 struct proto tcp_prot = {
2490         .name                   = "TCP",
2491         .owner                  = THIS_MODULE,
2492         .close                  = tcp_close,
2493         .connect                = tcp_v4_connect,
2494         .disconnect             = tcp_disconnect,
2495         .accept                 = inet_csk_accept,
2496         .ioctl                  = tcp_ioctl,
2497         .init                   = tcp_v4_init_sock,
2498         .destroy                = tcp_v4_destroy_sock,
2499         .shutdown               = tcp_shutdown,
2500         .setsockopt             = tcp_setsockopt,
2501         .getsockopt             = tcp_getsockopt,
2502         .recvmsg                = tcp_recvmsg,
2503         .backlog_rcv            = tcp_v4_do_rcv,
2504         .hash                   = inet_hash,
2505         .unhash                 = inet_unhash,
2506         .get_port               = inet_csk_get_port,
2507         .enter_memory_pressure  = tcp_enter_memory_pressure,
2508         .sockets_allocated      = &tcp_sockets_allocated,
2509         .orphan_count           = &tcp_orphan_count,
2510         .memory_allocated       = &tcp_memory_allocated,
2511         .memory_pressure        = &tcp_memory_pressure,
2512         .sysctl_mem             = sysctl_tcp_mem,
2513         .sysctl_wmem            = sysctl_tcp_wmem,
2514         .sysctl_rmem            = sysctl_tcp_rmem,
2515         .max_header             = MAX_TCP_HEADER,
2516         .obj_size               = sizeof(struct tcp_sock),
2517         .slab_flags             = SLAB_DESTROY_BY_RCU,
2518         .twsk_prot              = &tcp_timewait_sock_ops,
2519         .rsk_prot               = &tcp_request_sock_ops,
2520         .h.hashinfo             = &tcp_hashinfo,
2521 #ifdef CONFIG_COMPAT
2522         .compat_setsockopt      = compat_tcp_setsockopt,
2523         .compat_getsockopt      = compat_tcp_getsockopt,
2524 #endif
2525 };
2526
2527
2528 static int __net_init tcp_sk_init(struct net *net)
2529 {
2530         return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2531                                     PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2532 }
2533
2534 static void __net_exit tcp_sk_exit(struct net *net)
2535 {
2536         inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2537 }
2538
2539 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2540 {
2541         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2542 }
2543
2544 static struct pernet_operations __net_initdata tcp_sk_ops = {
2545        .init       = tcp_sk_init,
2546        .exit       = tcp_sk_exit,
2547        .exit_batch = tcp_sk_exit_batch,
2548 };
2549
2550 void __init tcp_v4_init(void)
2551 {
2552         inet_hashinfo_init(&tcp_hashinfo);
2553         if (register_pernet_subsys(&tcp_sk_ops))
2554                 panic("Failed to create the TCP control socket.\n");
2555 }
2556
2557 EXPORT_SYMBOL(ipv4_specific);
2558 EXPORT_SYMBOL(tcp_hashinfo);
2559 EXPORT_SYMBOL(tcp_prot);
2560 EXPORT_SYMBOL(tcp_v4_conn_request);
2561 EXPORT_SYMBOL(tcp_v4_connect);
2562 EXPORT_SYMBOL(tcp_v4_do_rcv);
2563 EXPORT_SYMBOL(tcp_v4_remember_stamp);
2564 EXPORT_SYMBOL(tcp_v4_send_check);
2565 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2566
2567 #ifdef CONFIG_PROC_FS
2568 EXPORT_SYMBOL(tcp_proc_register);
2569 EXPORT_SYMBOL(tcp_proc_unregister);
2570 #endif
2571 EXPORT_SYMBOL(sysctl_tcp_low_latency);
2572