tcp: fix possible socket refcount problem
[linux-3.10.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 #define pr_fmt(fmt) "TCP: " fmt
54
55 #include <linux/bottom_half.h>
56 #include <linux/types.h>
57 #include <linux/fcntl.h>
58 #include <linux/module.h>
59 #include <linux/random.h>
60 #include <linux/cache.h>
61 #include <linux/jhash.h>
62 #include <linux/init.h>
63 #include <linux/times.h>
64 #include <linux/slab.h>
65
66 #include <net/net_namespace.h>
67 #include <net/icmp.h>
68 #include <net/inet_hashtables.h>
69 #include <net/tcp.h>
70 #include <net/transp_v6.h>
71 #include <net/ipv6.h>
72 #include <net/inet_common.h>
73 #include <net/timewait_sock.h>
74 #include <net/xfrm.h>
75 #include <net/netdma.h>
76 #include <net/secure_seq.h>
77 #include <net/tcp_memcontrol.h>
78
79 #include <linux/inet.h>
80 #include <linux/ipv6.h>
81 #include <linux/stddef.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84
85 #include <linux/crypto.h>
86 #include <linux/scatterlist.h>
87
88 int sysctl_tcp_tw_reuse __read_mostly;
89 int sysctl_tcp_low_latency __read_mostly;
90 EXPORT_SYMBOL(sysctl_tcp_low_latency);
91
92
93 #ifdef CONFIG_TCP_MD5SIG
94 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
95                                __be32 daddr, __be32 saddr, const struct tcphdr *th);
96 #endif
97
98 struct inet_hashinfo tcp_hashinfo;
99 EXPORT_SYMBOL(tcp_hashinfo);
100
101 static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
102 {
103         return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
104                                           ip_hdr(skb)->saddr,
105                                           tcp_hdr(skb)->dest,
106                                           tcp_hdr(skb)->source);
107 }
108
109 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
110 {
111         const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
112         struct tcp_sock *tp = tcp_sk(sk);
113
114         /* With PAWS, it is safe from the viewpoint
115            of data integrity. Even without PAWS it is safe provided sequence
116            spaces do not overlap i.e. at data rates <= 80Mbit/sec.
117
118            Actually, the idea is close to VJ's one, only timestamp cache is
119            held not per host, but per port pair and TW bucket is used as state
120            holder.
121
122            If TW bucket has been already destroyed we fall back to VJ's scheme
123            and use initial timestamp retrieved from peer table.
124          */
125         if (tcptw->tw_ts_recent_stamp &&
126             (twp == NULL || (sysctl_tcp_tw_reuse &&
127                              get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
128                 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
129                 if (tp->write_seq == 0)
130                         tp->write_seq = 1;
131                 tp->rx_opt.ts_recent       = tcptw->tw_ts_recent;
132                 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
133                 sock_hold(sktw);
134                 return 1;
135         }
136
137         return 0;
138 }
139 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
140
141 static int tcp_repair_connect(struct sock *sk)
142 {
143         tcp_connect_init(sk);
144         tcp_finish_connect(sk, NULL);
145
146         return 0;
147 }
148
149 /* This will initiate an outgoing connection. */
150 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
151 {
152         struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
153         struct inet_sock *inet = inet_sk(sk);
154         struct tcp_sock *tp = tcp_sk(sk);
155         __be16 orig_sport, orig_dport;
156         __be32 daddr, nexthop;
157         struct flowi4 *fl4;
158         struct rtable *rt;
159         int err;
160         struct ip_options_rcu *inet_opt;
161
162         if (addr_len < sizeof(struct sockaddr_in))
163                 return -EINVAL;
164
165         if (usin->sin_family != AF_INET)
166                 return -EAFNOSUPPORT;
167
168         nexthop = daddr = usin->sin_addr.s_addr;
169         inet_opt = rcu_dereference_protected(inet->inet_opt,
170                                              sock_owned_by_user(sk));
171         if (inet_opt && inet_opt->opt.srr) {
172                 if (!daddr)
173                         return -EINVAL;
174                 nexthop = inet_opt->opt.faddr;
175         }
176
177         orig_sport = inet->inet_sport;
178         orig_dport = usin->sin_port;
179         fl4 = &inet->cork.fl.u.ip4;
180         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
181                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
182                               IPPROTO_TCP,
183                               orig_sport, orig_dport, sk, true);
184         if (IS_ERR(rt)) {
185                 err = PTR_ERR(rt);
186                 if (err == -ENETUNREACH)
187                         IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
188                 return err;
189         }
190
191         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
192                 ip_rt_put(rt);
193                 return -ENETUNREACH;
194         }
195
196         if (!inet_opt || !inet_opt->opt.srr)
197                 daddr = fl4->daddr;
198
199         if (!inet->inet_saddr)
200                 inet->inet_saddr = fl4->saddr;
201         inet->inet_rcv_saddr = inet->inet_saddr;
202
203         if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
204                 /* Reset inherited state */
205                 tp->rx_opt.ts_recent       = 0;
206                 tp->rx_opt.ts_recent_stamp = 0;
207                 if (likely(!tp->repair))
208                         tp->write_seq      = 0;
209         }
210
211         if (tcp_death_row.sysctl_tw_recycle &&
212             !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
213                 tcp_fetch_timewait_stamp(sk, &rt->dst);
214
215         inet->inet_dport = usin->sin_port;
216         inet->inet_daddr = daddr;
217
218         inet_csk(sk)->icsk_ext_hdr_len = 0;
219         if (inet_opt)
220                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
221
222         tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
223
224         /* Socket identity is still unknown (sport may be zero).
225          * However we set state to SYN-SENT and not releasing socket
226          * lock select source port, enter ourselves into the hash tables and
227          * complete initialization after this.
228          */
229         tcp_set_state(sk, TCP_SYN_SENT);
230         err = inet_hash_connect(&tcp_death_row, sk);
231         if (err)
232                 goto failure;
233
234         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
235                                inet->inet_sport, inet->inet_dport, sk);
236         if (IS_ERR(rt)) {
237                 err = PTR_ERR(rt);
238                 rt = NULL;
239                 goto failure;
240         }
241         /* OK, now commit destination to socket.  */
242         sk->sk_gso_type = SKB_GSO_TCPV4;
243         sk_setup_caps(sk, &rt->dst);
244
245         if (!tp->write_seq && likely(!tp->repair))
246                 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
247                                                            inet->inet_daddr,
248                                                            inet->inet_sport,
249                                                            usin->sin_port);
250
251         inet->inet_id = tp->write_seq ^ jiffies;
252
253         if (likely(!tp->repair))
254                 err = tcp_connect(sk);
255         else
256                 err = tcp_repair_connect(sk);
257
258         rt = NULL;
259         if (err)
260                 goto failure;
261
262         return 0;
263
264 failure:
265         /*
266          * This unhashes the socket and releases the local port,
267          * if necessary.
268          */
269         tcp_set_state(sk, TCP_CLOSE);
270         ip_rt_put(rt);
271         sk->sk_route_caps = 0;
272         inet->inet_dport = 0;
273         return err;
274 }
275 EXPORT_SYMBOL(tcp_v4_connect);
276
277 /*
278  * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
279  * It can be called through tcp_release_cb() if socket was owned by user
280  * at the time tcp_v4_err() was called to handle ICMP message.
281  */
282 static void tcp_v4_mtu_reduced(struct sock *sk)
283 {
284         struct dst_entry *dst;
285         struct inet_sock *inet = inet_sk(sk);
286         u32 mtu = tcp_sk(sk)->mtu_info;
287
288         /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
289          * send out by Linux are always <576bytes so they should go through
290          * unfragmented).
291          */
292         if (sk->sk_state == TCP_LISTEN)
293                 return;
294
295         dst = inet_csk_update_pmtu(sk, mtu);
296         if (!dst)
297                 return;
298
299         /* Something is about to be wrong... Remember soft error
300          * for the case, if this connection will not able to recover.
301          */
302         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
303                 sk->sk_err_soft = EMSGSIZE;
304
305         mtu = dst_mtu(dst);
306
307         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
308             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
309                 tcp_sync_mss(sk, mtu);
310
311                 /* Resend the TCP packet because it's
312                  * clear that the old packet has been
313                  * dropped. This is the new "fast" path mtu
314                  * discovery.
315                  */
316                 tcp_simple_retransmit(sk);
317         } /* else let the usual retransmit timer handle it */
318 }
319
320 static void do_redirect(struct sk_buff *skb, struct sock *sk)
321 {
322         struct dst_entry *dst = __sk_dst_check(sk, 0);
323
324         if (dst)
325                 dst->ops->redirect(dst, sk, skb);
326 }
327
328 /*
329  * This routine is called by the ICMP module when it gets some
330  * sort of error condition.  If err < 0 then the socket should
331  * be closed and the error returned to the user.  If err > 0
332  * it's just the icmp type << 8 | icmp code.  After adjustment
333  * header points to the first 8 bytes of the tcp header.  We need
334  * to find the appropriate port.
335  *
336  * The locking strategy used here is very "optimistic". When
337  * someone else accesses the socket the ICMP is just dropped
338  * and for some paths there is no check at all.
339  * A more general error queue to queue errors for later handling
340  * is probably better.
341  *
342  */
343
344 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
345 {
346         const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
347         struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
348         struct inet_connection_sock *icsk;
349         struct tcp_sock *tp;
350         struct inet_sock *inet;
351         const int type = icmp_hdr(icmp_skb)->type;
352         const int code = icmp_hdr(icmp_skb)->code;
353         struct sock *sk;
354         struct sk_buff *skb;
355         __u32 seq;
356         __u32 remaining;
357         int err;
358         struct net *net = dev_net(icmp_skb->dev);
359
360         if (icmp_skb->len < (iph->ihl << 2) + 8) {
361                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
362                 return;
363         }
364
365         sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
366                         iph->saddr, th->source, inet_iif(icmp_skb));
367         if (!sk) {
368                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
369                 return;
370         }
371         if (sk->sk_state == TCP_TIME_WAIT) {
372                 inet_twsk_put(inet_twsk(sk));
373                 return;
374         }
375
376         bh_lock_sock(sk);
377         /* If too many ICMPs get dropped on busy
378          * servers this needs to be solved differently.
379          * We do take care of PMTU discovery (RFC1191) special case :
380          * we can receive locally generated ICMP messages while socket is held.
381          */
382         if (sock_owned_by_user(sk) &&
383             type != ICMP_DEST_UNREACH &&
384             code != ICMP_FRAG_NEEDED)
385                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
386
387         if (sk->sk_state == TCP_CLOSE)
388                 goto out;
389
390         if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
391                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
392                 goto out;
393         }
394
395         icsk = inet_csk(sk);
396         tp = tcp_sk(sk);
397         seq = ntohl(th->seq);
398         if (sk->sk_state != TCP_LISTEN &&
399             !between(seq, tp->snd_una, tp->snd_nxt)) {
400                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
401                 goto out;
402         }
403
404         switch (type) {
405         case ICMP_REDIRECT:
406                 do_redirect(icmp_skb, sk);
407                 goto out;
408         case ICMP_SOURCE_QUENCH:
409                 /* Just silently ignore these. */
410                 goto out;
411         case ICMP_PARAMETERPROB:
412                 err = EPROTO;
413                 break;
414         case ICMP_DEST_UNREACH:
415                 if (code > NR_ICMP_UNREACH)
416                         goto out;
417
418                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
419                         tp->mtu_info = info;
420                         if (!sock_owned_by_user(sk)) {
421                                 tcp_v4_mtu_reduced(sk);
422                         } else {
423                                 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
424                                         sock_hold(sk);
425                         }
426                         goto out;
427                 }
428
429                 err = icmp_err_convert[code].errno;
430                 /* check if icmp_skb allows revert of backoff
431                  * (see draft-zimmermann-tcp-lcd) */
432                 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
433                         break;
434                 if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
435                     !icsk->icsk_backoff)
436                         break;
437
438                 if (sock_owned_by_user(sk))
439                         break;
440
441                 icsk->icsk_backoff--;
442                 inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
443                         TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
444                 tcp_bound_rto(sk);
445
446                 skb = tcp_write_queue_head(sk);
447                 BUG_ON(!skb);
448
449                 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
450                                 tcp_time_stamp - TCP_SKB_CB(skb)->when);
451
452                 if (remaining) {
453                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
454                                                   remaining, TCP_RTO_MAX);
455                 } else {
456                         /* RTO revert clocked out retransmission.
457                          * Will retransmit now */
458                         tcp_retransmit_timer(sk);
459                 }
460
461                 break;
462         case ICMP_TIME_EXCEEDED:
463                 err = EHOSTUNREACH;
464                 break;
465         default:
466                 goto out;
467         }
468
469         switch (sk->sk_state) {
470                 struct request_sock *req, **prev;
471         case TCP_LISTEN:
472                 if (sock_owned_by_user(sk))
473                         goto out;
474
475                 req = inet_csk_search_req(sk, &prev, th->dest,
476                                           iph->daddr, iph->saddr);
477                 if (!req)
478                         goto out;
479
480                 /* ICMPs are not backlogged, hence we cannot get
481                    an established socket here.
482                  */
483                 WARN_ON(req->sk);
484
485                 if (seq != tcp_rsk(req)->snt_isn) {
486                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
487                         goto out;
488                 }
489
490                 /*
491                  * Still in SYN_RECV, just remove it silently.
492                  * There is no good way to pass the error to the newly
493                  * created socket, and POSIX does not want network
494                  * errors returned from accept().
495                  */
496                 inet_csk_reqsk_queue_drop(sk, req, prev);
497                 goto out;
498
499         case TCP_SYN_SENT:
500         case TCP_SYN_RECV:  /* Cannot happen.
501                                It can f.e. if SYNs crossed.
502                              */
503                 if (!sock_owned_by_user(sk)) {
504                         sk->sk_err = err;
505
506                         sk->sk_error_report(sk);
507
508                         tcp_done(sk);
509                 } else {
510                         sk->sk_err_soft = err;
511                 }
512                 goto out;
513         }
514
515         /* If we've already connected we will keep trying
516          * until we time out, or the user gives up.
517          *
518          * rfc1122 4.2.3.9 allows to consider as hard errors
519          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
520          * but it is obsoleted by pmtu discovery).
521          *
522          * Note, that in modern internet, where routing is unreliable
523          * and in each dark corner broken firewalls sit, sending random
524          * errors ordered by their masters even this two messages finally lose
525          * their original sense (even Linux sends invalid PORT_UNREACHs)
526          *
527          * Now we are in compliance with RFCs.
528          *                                                      --ANK (980905)
529          */
530
531         inet = inet_sk(sk);
532         if (!sock_owned_by_user(sk) && inet->recverr) {
533                 sk->sk_err = err;
534                 sk->sk_error_report(sk);
535         } else  { /* Only an error on timeout */
536                 sk->sk_err_soft = err;
537         }
538
539 out:
540         bh_unlock_sock(sk);
541         sock_put(sk);
542 }
543
544 static void __tcp_v4_send_check(struct sk_buff *skb,
545                                 __be32 saddr, __be32 daddr)
546 {
547         struct tcphdr *th = tcp_hdr(skb);
548
549         if (skb->ip_summed == CHECKSUM_PARTIAL) {
550                 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
551                 skb->csum_start = skb_transport_header(skb) - skb->head;
552                 skb->csum_offset = offsetof(struct tcphdr, check);
553         } else {
554                 th->check = tcp_v4_check(skb->len, saddr, daddr,
555                                          csum_partial(th,
556                                                       th->doff << 2,
557                                                       skb->csum));
558         }
559 }
560
561 /* This routine computes an IPv4 TCP checksum. */
562 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
563 {
564         const struct inet_sock *inet = inet_sk(sk);
565
566         __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
567 }
568 EXPORT_SYMBOL(tcp_v4_send_check);
569
570 int tcp_v4_gso_send_check(struct sk_buff *skb)
571 {
572         const struct iphdr *iph;
573         struct tcphdr *th;
574
575         if (!pskb_may_pull(skb, sizeof(*th)))
576                 return -EINVAL;
577
578         iph = ip_hdr(skb);
579         th = tcp_hdr(skb);
580
581         th->check = 0;
582         skb->ip_summed = CHECKSUM_PARTIAL;
583         __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
584         return 0;
585 }
586
587 /*
588  *      This routine will send an RST to the other tcp.
589  *
590  *      Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
591  *                    for reset.
592  *      Answer: if a packet caused RST, it is not for a socket
593  *              existing in our system, if it is matched to a socket,
594  *              it is just duplicate segment or bug in other side's TCP.
595  *              So that we build reply only basing on parameters
596  *              arrived with segment.
597  *      Exception: precedence violation. We do not implement it in any case.
598  */
599
600 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
601 {
602         const struct tcphdr *th = tcp_hdr(skb);
603         struct {
604                 struct tcphdr th;
605 #ifdef CONFIG_TCP_MD5SIG
606                 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
607 #endif
608         } rep;
609         struct ip_reply_arg arg;
610 #ifdef CONFIG_TCP_MD5SIG
611         struct tcp_md5sig_key *key;
612         const __u8 *hash_location = NULL;
613         unsigned char newhash[16];
614         int genhash;
615         struct sock *sk1 = NULL;
616 #endif
617         struct net *net;
618
619         /* Never send a reset in response to a reset. */
620         if (th->rst)
621                 return;
622
623         if (skb_rtable(skb)->rt_type != RTN_LOCAL)
624                 return;
625
626         /* Swap the send and the receive. */
627         memset(&rep, 0, sizeof(rep));
628         rep.th.dest   = th->source;
629         rep.th.source = th->dest;
630         rep.th.doff   = sizeof(struct tcphdr) / 4;
631         rep.th.rst    = 1;
632
633         if (th->ack) {
634                 rep.th.seq = th->ack_seq;
635         } else {
636                 rep.th.ack = 1;
637                 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
638                                        skb->len - (th->doff << 2));
639         }
640
641         memset(&arg, 0, sizeof(arg));
642         arg.iov[0].iov_base = (unsigned char *)&rep;
643         arg.iov[0].iov_len  = sizeof(rep.th);
644
645 #ifdef CONFIG_TCP_MD5SIG
646         hash_location = tcp_parse_md5sig_option(th);
647         if (!sk && hash_location) {
648                 /*
649                  * active side is lost. Try to find listening socket through
650                  * source port, and then find md5 key through listening socket.
651                  * we are not loose security here:
652                  * Incoming packet is checked with md5 hash with finding key,
653                  * no RST generated if md5 hash doesn't match.
654                  */
655                 sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev),
656                                              &tcp_hashinfo, ip_hdr(skb)->daddr,
657                                              ntohs(th->source), inet_iif(skb));
658                 /* don't send rst if it can't find key */
659                 if (!sk1)
660                         return;
661                 rcu_read_lock();
662                 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
663                                         &ip_hdr(skb)->saddr, AF_INET);
664                 if (!key)
665                         goto release_sk1;
666
667                 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb);
668                 if (genhash || memcmp(hash_location, newhash, 16) != 0)
669                         goto release_sk1;
670         } else {
671                 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
672                                              &ip_hdr(skb)->saddr,
673                                              AF_INET) : NULL;
674         }
675
676         if (key) {
677                 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
678                                    (TCPOPT_NOP << 16) |
679                                    (TCPOPT_MD5SIG << 8) |
680                                    TCPOLEN_MD5SIG);
681                 /* Update length and the length the header thinks exists */
682                 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
683                 rep.th.doff = arg.iov[0].iov_len / 4;
684
685                 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
686                                      key, ip_hdr(skb)->saddr,
687                                      ip_hdr(skb)->daddr, &rep.th);
688         }
689 #endif
690         arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
691                                       ip_hdr(skb)->saddr, /* XXX */
692                                       arg.iov[0].iov_len, IPPROTO_TCP, 0);
693         arg.csumoffset = offsetof(struct tcphdr, check) / 2;
694         arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
695         /* When socket is gone, all binding information is lost.
696          * routing might fail in this case. using iif for oif to
697          * make sure we can deliver it
698          */
699         arg.bound_dev_if = sk ? sk->sk_bound_dev_if : inet_iif(skb);
700
701         net = dev_net(skb_dst(skb)->dev);
702         arg.tos = ip_hdr(skb)->tos;
703         ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
704                               ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
705
706         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
707         TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
708
709 #ifdef CONFIG_TCP_MD5SIG
710 release_sk1:
711         if (sk1) {
712                 rcu_read_unlock();
713                 sock_put(sk1);
714         }
715 #endif
716 }
717
718 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
719    outside socket context is ugly, certainly. What can I do?
720  */
721
722 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
723                             u32 win, u32 ts, int oif,
724                             struct tcp_md5sig_key *key,
725                             int reply_flags, u8 tos)
726 {
727         const struct tcphdr *th = tcp_hdr(skb);
728         struct {
729                 struct tcphdr th;
730                 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
731 #ifdef CONFIG_TCP_MD5SIG
732                            + (TCPOLEN_MD5SIG_ALIGNED >> 2)
733 #endif
734                         ];
735         } rep;
736         struct ip_reply_arg arg;
737         struct net *net = dev_net(skb_dst(skb)->dev);
738
739         memset(&rep.th, 0, sizeof(struct tcphdr));
740         memset(&arg, 0, sizeof(arg));
741
742         arg.iov[0].iov_base = (unsigned char *)&rep;
743         arg.iov[0].iov_len  = sizeof(rep.th);
744         if (ts) {
745                 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
746                                    (TCPOPT_TIMESTAMP << 8) |
747                                    TCPOLEN_TIMESTAMP);
748                 rep.opt[1] = htonl(tcp_time_stamp);
749                 rep.opt[2] = htonl(ts);
750                 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
751         }
752
753         /* Swap the send and the receive. */
754         rep.th.dest    = th->source;
755         rep.th.source  = th->dest;
756         rep.th.doff    = arg.iov[0].iov_len / 4;
757         rep.th.seq     = htonl(seq);
758         rep.th.ack_seq = htonl(ack);
759         rep.th.ack     = 1;
760         rep.th.window  = htons(win);
761
762 #ifdef CONFIG_TCP_MD5SIG
763         if (key) {
764                 int offset = (ts) ? 3 : 0;
765
766                 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
767                                           (TCPOPT_NOP << 16) |
768                                           (TCPOPT_MD5SIG << 8) |
769                                           TCPOLEN_MD5SIG);
770                 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
771                 rep.th.doff = arg.iov[0].iov_len/4;
772
773                 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
774                                     key, ip_hdr(skb)->saddr,
775                                     ip_hdr(skb)->daddr, &rep.th);
776         }
777 #endif
778         arg.flags = reply_flags;
779         arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
780                                       ip_hdr(skb)->saddr, /* XXX */
781                                       arg.iov[0].iov_len, IPPROTO_TCP, 0);
782         arg.csumoffset = offsetof(struct tcphdr, check) / 2;
783         if (oif)
784                 arg.bound_dev_if = oif;
785         arg.tos = tos;
786         ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
787                               ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
788
789         TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
790 }
791
792 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
793 {
794         struct inet_timewait_sock *tw = inet_twsk(sk);
795         struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
796
797         tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
798                         tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
799                         tcptw->tw_ts_recent,
800                         tw->tw_bound_dev_if,
801                         tcp_twsk_md5_key(tcptw),
802                         tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
803                         tw->tw_tos
804                         );
805
806         inet_twsk_put(tw);
807 }
808
809 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
810                                   struct request_sock *req)
811 {
812         tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
813                         tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
814                         req->ts_recent,
815                         0,
816                         tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
817                                           AF_INET),
818                         inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
819                         ip_hdr(skb)->tos);
820 }
821
822 /*
823  *      Send a SYN-ACK after having received a SYN.
824  *      This still operates on a request_sock only, not on a big
825  *      socket.
826  */
827 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
828                               struct request_sock *req,
829                               struct request_values *rvp,
830                               u16 queue_mapping,
831                               bool nocache)
832 {
833         const struct inet_request_sock *ireq = inet_rsk(req);
834         struct flowi4 fl4;
835         int err = -1;
836         struct sk_buff * skb;
837
838         /* First, grab a route. */
839         if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
840                 return -1;
841
842         skb = tcp_make_synack(sk, dst, req, rvp);
843
844         if (skb) {
845                 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
846
847                 skb_set_queue_mapping(skb, queue_mapping);
848                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
849                                             ireq->rmt_addr,
850                                             ireq->opt);
851                 err = net_xmit_eval(err);
852         }
853
854         return err;
855 }
856
857 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
858                               struct request_values *rvp)
859 {
860         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
861         return tcp_v4_send_synack(sk, NULL, req, rvp, 0, false);
862 }
863
864 /*
865  *      IPv4 request_sock destructor.
866  */
867 static void tcp_v4_reqsk_destructor(struct request_sock *req)
868 {
869         kfree(inet_rsk(req)->opt);
870 }
871
872 /*
873  * Return true if a syncookie should be sent
874  */
875 bool tcp_syn_flood_action(struct sock *sk,
876                          const struct sk_buff *skb,
877                          const char *proto)
878 {
879         const char *msg = "Dropping request";
880         bool want_cookie = false;
881         struct listen_sock *lopt;
882
883
884
885 #ifdef CONFIG_SYN_COOKIES
886         if (sysctl_tcp_syncookies) {
887                 msg = "Sending cookies";
888                 want_cookie = true;
889                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
890         } else
891 #endif
892                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
893
894         lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
895         if (!lopt->synflood_warned) {
896                 lopt->synflood_warned = 1;
897                 pr_info("%s: Possible SYN flooding on port %d. %s.  Check SNMP counters.\n",
898                         proto, ntohs(tcp_hdr(skb)->dest), msg);
899         }
900         return want_cookie;
901 }
902 EXPORT_SYMBOL(tcp_syn_flood_action);
903
904 /*
905  * Save and compile IPv4 options into the request_sock if needed.
906  */
907 static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk,
908                                                   struct sk_buff *skb)
909 {
910         const struct ip_options *opt = &(IPCB(skb)->opt);
911         struct ip_options_rcu *dopt = NULL;
912
913         if (opt && opt->optlen) {
914                 int opt_size = sizeof(*dopt) + opt->optlen;
915
916                 dopt = kmalloc(opt_size, GFP_ATOMIC);
917                 if (dopt) {
918                         if (ip_options_echo(&dopt->opt, skb)) {
919                                 kfree(dopt);
920                                 dopt = NULL;
921                         }
922                 }
923         }
924         return dopt;
925 }
926
927 #ifdef CONFIG_TCP_MD5SIG
928 /*
929  * RFC2385 MD5 checksumming requires a mapping of
930  * IP address->MD5 Key.
931  * We need to maintain these in the sk structure.
932  */
933
934 /* Find the Key structure for an address.  */
935 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
936                                          const union tcp_md5_addr *addr,
937                                          int family)
938 {
939         struct tcp_sock *tp = tcp_sk(sk);
940         struct tcp_md5sig_key *key;
941         struct hlist_node *pos;
942         unsigned int size = sizeof(struct in_addr);
943         struct tcp_md5sig_info *md5sig;
944
945         /* caller either holds rcu_read_lock() or socket lock */
946         md5sig = rcu_dereference_check(tp->md5sig_info,
947                                        sock_owned_by_user(sk) ||
948                                        lockdep_is_held(&sk->sk_lock.slock));
949         if (!md5sig)
950                 return NULL;
951 #if IS_ENABLED(CONFIG_IPV6)
952         if (family == AF_INET6)
953                 size = sizeof(struct in6_addr);
954 #endif
955         hlist_for_each_entry_rcu(key, pos, &md5sig->head, node) {
956                 if (key->family != family)
957                         continue;
958                 if (!memcmp(&key->addr, addr, size))
959                         return key;
960         }
961         return NULL;
962 }
963 EXPORT_SYMBOL(tcp_md5_do_lookup);
964
965 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
966                                          struct sock *addr_sk)
967 {
968         union tcp_md5_addr *addr;
969
970         addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
971         return tcp_md5_do_lookup(sk, addr, AF_INET);
972 }
973 EXPORT_SYMBOL(tcp_v4_md5_lookup);
974
975 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
976                                                       struct request_sock *req)
977 {
978         union tcp_md5_addr *addr;
979
980         addr = (union tcp_md5_addr *)&inet_rsk(req)->rmt_addr;
981         return tcp_md5_do_lookup(sk, addr, AF_INET);
982 }
983
984 /* This can be called on a newly created socket, from other files */
985 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
986                    int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
987 {
988         /* Add Key to the list */
989         struct tcp_md5sig_key *key;
990         struct tcp_sock *tp = tcp_sk(sk);
991         struct tcp_md5sig_info *md5sig;
992
993         key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
994         if (key) {
995                 /* Pre-existing entry - just update that one. */
996                 memcpy(key->key, newkey, newkeylen);
997                 key->keylen = newkeylen;
998                 return 0;
999         }
1000
1001         md5sig = rcu_dereference_protected(tp->md5sig_info,
1002                                            sock_owned_by_user(sk));
1003         if (!md5sig) {
1004                 md5sig = kmalloc(sizeof(*md5sig), gfp);
1005                 if (!md5sig)
1006                         return -ENOMEM;
1007
1008                 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1009                 INIT_HLIST_HEAD(&md5sig->head);
1010                 rcu_assign_pointer(tp->md5sig_info, md5sig);
1011         }
1012
1013         key = sock_kmalloc(sk, sizeof(*key), gfp);
1014         if (!key)
1015                 return -ENOMEM;
1016         if (hlist_empty(&md5sig->head) && !tcp_alloc_md5sig_pool(sk)) {
1017                 sock_kfree_s(sk, key, sizeof(*key));
1018                 return -ENOMEM;
1019         }
1020
1021         memcpy(key->key, newkey, newkeylen);
1022         key->keylen = newkeylen;
1023         key->family = family;
1024         memcpy(&key->addr, addr,
1025                (family == AF_INET6) ? sizeof(struct in6_addr) :
1026                                       sizeof(struct in_addr));
1027         hlist_add_head_rcu(&key->node, &md5sig->head);
1028         return 0;
1029 }
1030 EXPORT_SYMBOL(tcp_md5_do_add);
1031
1032 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
1033 {
1034         struct tcp_sock *tp = tcp_sk(sk);
1035         struct tcp_md5sig_key *key;
1036         struct tcp_md5sig_info *md5sig;
1037
1038         key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
1039         if (!key)
1040                 return -ENOENT;
1041         hlist_del_rcu(&key->node);
1042         atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1043         kfree_rcu(key, rcu);
1044         md5sig = rcu_dereference_protected(tp->md5sig_info,
1045                                            sock_owned_by_user(sk));
1046         if (hlist_empty(&md5sig->head))
1047                 tcp_free_md5sig_pool();
1048         return 0;
1049 }
1050 EXPORT_SYMBOL(tcp_md5_do_del);
1051
1052 void tcp_clear_md5_list(struct sock *sk)
1053 {
1054         struct tcp_sock *tp = tcp_sk(sk);
1055         struct tcp_md5sig_key *key;
1056         struct hlist_node *pos, *n;
1057         struct tcp_md5sig_info *md5sig;
1058
1059         md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1060
1061         if (!hlist_empty(&md5sig->head))
1062                 tcp_free_md5sig_pool();
1063         hlist_for_each_entry_safe(key, pos, n, &md5sig->head, node) {
1064                 hlist_del_rcu(&key->node);
1065                 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1066                 kfree_rcu(key, rcu);
1067         }
1068 }
1069
1070 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1071                                  int optlen)
1072 {
1073         struct tcp_md5sig cmd;
1074         struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1075
1076         if (optlen < sizeof(cmd))
1077                 return -EINVAL;
1078
1079         if (copy_from_user(&cmd, optval, sizeof(cmd)))
1080                 return -EFAULT;
1081
1082         if (sin->sin_family != AF_INET)
1083                 return -EINVAL;
1084
1085         if (!cmd.tcpm_key || !cmd.tcpm_keylen)
1086                 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1087                                       AF_INET);
1088
1089         if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1090                 return -EINVAL;
1091
1092         return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1093                               AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1094                               GFP_KERNEL);
1095 }
1096
1097 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1098                                         __be32 daddr, __be32 saddr, int nbytes)
1099 {
1100         struct tcp4_pseudohdr *bp;
1101         struct scatterlist sg;
1102
1103         bp = &hp->md5_blk.ip4;
1104
1105         /*
1106          * 1. the TCP pseudo-header (in the order: source IP address,
1107          * destination IP address, zero-padded protocol number, and
1108          * segment length)
1109          */
1110         bp->saddr = saddr;
1111         bp->daddr = daddr;
1112         bp->pad = 0;
1113         bp->protocol = IPPROTO_TCP;
1114         bp->len = cpu_to_be16(nbytes);
1115
1116         sg_init_one(&sg, bp, sizeof(*bp));
1117         return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1118 }
1119
1120 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1121                                __be32 daddr, __be32 saddr, const struct tcphdr *th)
1122 {
1123         struct tcp_md5sig_pool *hp;
1124         struct hash_desc *desc;
1125
1126         hp = tcp_get_md5sig_pool();
1127         if (!hp)
1128                 goto clear_hash_noput;
1129         desc = &hp->md5_desc;
1130
1131         if (crypto_hash_init(desc))
1132                 goto clear_hash;
1133         if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1134                 goto clear_hash;
1135         if (tcp_md5_hash_header(hp, th))
1136                 goto clear_hash;
1137         if (tcp_md5_hash_key(hp, key))
1138                 goto clear_hash;
1139         if (crypto_hash_final(desc, md5_hash))
1140                 goto clear_hash;
1141
1142         tcp_put_md5sig_pool();
1143         return 0;
1144
1145 clear_hash:
1146         tcp_put_md5sig_pool();
1147 clear_hash_noput:
1148         memset(md5_hash, 0, 16);
1149         return 1;
1150 }
1151
1152 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1153                         const struct sock *sk, const struct request_sock *req,
1154                         const struct sk_buff *skb)
1155 {
1156         struct tcp_md5sig_pool *hp;
1157         struct hash_desc *desc;
1158         const struct tcphdr *th = tcp_hdr(skb);
1159         __be32 saddr, daddr;
1160
1161         if (sk) {
1162                 saddr = inet_sk(sk)->inet_saddr;
1163                 daddr = inet_sk(sk)->inet_daddr;
1164         } else if (req) {
1165                 saddr = inet_rsk(req)->loc_addr;
1166                 daddr = inet_rsk(req)->rmt_addr;
1167         } else {
1168                 const struct iphdr *iph = ip_hdr(skb);
1169                 saddr = iph->saddr;
1170                 daddr = iph->daddr;
1171         }
1172
1173         hp = tcp_get_md5sig_pool();
1174         if (!hp)
1175                 goto clear_hash_noput;
1176         desc = &hp->md5_desc;
1177
1178         if (crypto_hash_init(desc))
1179                 goto clear_hash;
1180
1181         if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1182                 goto clear_hash;
1183         if (tcp_md5_hash_header(hp, th))
1184                 goto clear_hash;
1185         if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1186                 goto clear_hash;
1187         if (tcp_md5_hash_key(hp, key))
1188                 goto clear_hash;
1189         if (crypto_hash_final(desc, md5_hash))
1190                 goto clear_hash;
1191
1192         tcp_put_md5sig_pool();
1193         return 0;
1194
1195 clear_hash:
1196         tcp_put_md5sig_pool();
1197 clear_hash_noput:
1198         memset(md5_hash, 0, 16);
1199         return 1;
1200 }
1201 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1202
1203 static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1204 {
1205         /*
1206          * This gets called for each TCP segment that arrives
1207          * so we want to be efficient.
1208          * We have 3 drop cases:
1209          * o No MD5 hash and one expected.
1210          * o MD5 hash and we're not expecting one.
1211          * o MD5 hash and its wrong.
1212          */
1213         const __u8 *hash_location = NULL;
1214         struct tcp_md5sig_key *hash_expected;
1215         const struct iphdr *iph = ip_hdr(skb);
1216         const struct tcphdr *th = tcp_hdr(skb);
1217         int genhash;
1218         unsigned char newhash[16];
1219
1220         hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1221                                           AF_INET);
1222         hash_location = tcp_parse_md5sig_option(th);
1223
1224         /* We've parsed the options - do we have a hash? */
1225         if (!hash_expected && !hash_location)
1226                 return false;
1227
1228         if (hash_expected && !hash_location) {
1229                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1230                 return true;
1231         }
1232
1233         if (!hash_expected && hash_location) {
1234                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1235                 return true;
1236         }
1237
1238         /* Okay, so this is hash_expected and hash_location -
1239          * so we need to calculate the checksum.
1240          */
1241         genhash = tcp_v4_md5_hash_skb(newhash,
1242                                       hash_expected,
1243                                       NULL, NULL, skb);
1244
1245         if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1246                 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1247                                      &iph->saddr, ntohs(th->source),
1248                                      &iph->daddr, ntohs(th->dest),
1249                                      genhash ? " tcp_v4_calc_md5_hash failed"
1250                                      : "");
1251                 return true;
1252         }
1253         return false;
1254 }
1255
1256 #endif
1257
1258 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1259         .family         =       PF_INET,
1260         .obj_size       =       sizeof(struct tcp_request_sock),
1261         .rtx_syn_ack    =       tcp_v4_rtx_synack,
1262         .send_ack       =       tcp_v4_reqsk_send_ack,
1263         .destructor     =       tcp_v4_reqsk_destructor,
1264         .send_reset     =       tcp_v4_send_reset,
1265         .syn_ack_timeout =      tcp_syn_ack_timeout,
1266 };
1267
1268 #ifdef CONFIG_TCP_MD5SIG
1269 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1270         .md5_lookup     =       tcp_v4_reqsk_md5_lookup,
1271         .calc_md5_hash  =       tcp_v4_md5_hash_skb,
1272 };
1273 #endif
1274
1275 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1276 {
1277         struct tcp_extend_values tmp_ext;
1278         struct tcp_options_received tmp_opt;
1279         const u8 *hash_location;
1280         struct request_sock *req;
1281         struct inet_request_sock *ireq;
1282         struct tcp_sock *tp = tcp_sk(sk);
1283         struct dst_entry *dst = NULL;
1284         __be32 saddr = ip_hdr(skb)->saddr;
1285         __be32 daddr = ip_hdr(skb)->daddr;
1286         __u32 isn = TCP_SKB_CB(skb)->when;
1287         bool want_cookie = false;
1288
1289         /* Never answer to SYNs send to broadcast or multicast */
1290         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1291                 goto drop;
1292
1293         /* TW buckets are converted to open requests without
1294          * limitations, they conserve resources and peer is
1295          * evidently real one.
1296          */
1297         if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1298                 want_cookie = tcp_syn_flood_action(sk, skb, "TCP");
1299                 if (!want_cookie)
1300                         goto drop;
1301         }
1302
1303         /* Accept backlog is full. If we have already queued enough
1304          * of warm entries in syn queue, drop request. It is better than
1305          * clogging syn queue with openreqs with exponentially increasing
1306          * timeout.
1307          */
1308         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1309                 goto drop;
1310
1311         req = inet_reqsk_alloc(&tcp_request_sock_ops);
1312         if (!req)
1313                 goto drop;
1314
1315 #ifdef CONFIG_TCP_MD5SIG
1316         tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1317 #endif
1318
1319         tcp_clear_options(&tmp_opt);
1320         tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1321         tmp_opt.user_mss  = tp->rx_opt.user_mss;
1322         tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
1323
1324         if (tmp_opt.cookie_plus > 0 &&
1325             tmp_opt.saw_tstamp &&
1326             !tp->rx_opt.cookie_out_never &&
1327             (sysctl_tcp_cookie_size > 0 ||
1328              (tp->cookie_values != NULL &&
1329               tp->cookie_values->cookie_desired > 0))) {
1330                 u8 *c;
1331                 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1332                 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1333
1334                 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1335                         goto drop_and_release;
1336
1337                 /* Secret recipe starts with IP addresses */
1338                 *mess++ ^= (__force u32)daddr;
1339                 *mess++ ^= (__force u32)saddr;
1340
1341                 /* plus variable length Initiator Cookie */
1342                 c = (u8 *)mess;
1343                 while (l-- > 0)
1344                         *c++ ^= *hash_location++;
1345
1346                 want_cookie = false;    /* not our kind of cookie */
1347                 tmp_ext.cookie_out_never = 0; /* false */
1348                 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1349         } else if (!tp->rx_opt.cookie_in_always) {
1350                 /* redundant indications, but ensure initialization. */
1351                 tmp_ext.cookie_out_never = 1; /* true */
1352                 tmp_ext.cookie_plus = 0;
1353         } else {
1354                 goto drop_and_release;
1355         }
1356         tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1357
1358         if (want_cookie && !tmp_opt.saw_tstamp)
1359                 tcp_clear_options(&tmp_opt);
1360
1361         tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1362         tcp_openreq_init(req, &tmp_opt, skb);
1363
1364         ireq = inet_rsk(req);
1365         ireq->loc_addr = daddr;
1366         ireq->rmt_addr = saddr;
1367         ireq->no_srccheck = inet_sk(sk)->transparent;
1368         ireq->opt = tcp_v4_save_options(sk, skb);
1369
1370         if (security_inet_conn_request(sk, skb, req))
1371                 goto drop_and_free;
1372
1373         if (!want_cookie || tmp_opt.tstamp_ok)
1374                 TCP_ECN_create_request(req, skb);
1375
1376         if (want_cookie) {
1377                 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1378                 req->cookie_ts = tmp_opt.tstamp_ok;
1379         } else if (!isn) {
1380                 struct flowi4 fl4;
1381
1382                 /* VJ's idea. We save last timestamp seen
1383                  * from the destination in peer table, when entering
1384                  * state TIME-WAIT, and check against it before
1385                  * accepting new connection request.
1386                  *
1387                  * If "isn" is not zero, this request hit alive
1388                  * timewait bucket, so that all the necessary checks
1389                  * are made in the function processing timewait state.
1390                  */
1391                 if (tmp_opt.saw_tstamp &&
1392                     tcp_death_row.sysctl_tw_recycle &&
1393                     (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1394                     fl4.daddr == saddr) {
1395                         if (!tcp_peer_is_proven(req, dst, true)) {
1396                                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1397                                 goto drop_and_release;
1398                         }
1399                 }
1400                 /* Kill the following clause, if you dislike this way. */
1401                 else if (!sysctl_tcp_syncookies &&
1402                          (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1403                           (sysctl_max_syn_backlog >> 2)) &&
1404                          !tcp_peer_is_proven(req, dst, false)) {
1405                         /* Without syncookies last quarter of
1406                          * backlog is filled with destinations,
1407                          * proven to be alive.
1408                          * It means that we continue to communicate
1409                          * to destinations, already remembered
1410                          * to the moment of synflood.
1411                          */
1412                         LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"),
1413                                        &saddr, ntohs(tcp_hdr(skb)->source));
1414                         goto drop_and_release;
1415                 }
1416
1417                 isn = tcp_v4_init_sequence(skb);
1418         }
1419         tcp_rsk(req)->snt_isn = isn;
1420         tcp_rsk(req)->snt_synack = tcp_time_stamp;
1421
1422         if (tcp_v4_send_synack(sk, dst, req,
1423                                (struct request_values *)&tmp_ext,
1424                                skb_get_queue_mapping(skb),
1425                                want_cookie) ||
1426             want_cookie)
1427                 goto drop_and_free;
1428
1429         inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1430         return 0;
1431
1432 drop_and_release:
1433         dst_release(dst);
1434 drop_and_free:
1435         reqsk_free(req);
1436 drop:
1437         return 0;
1438 }
1439 EXPORT_SYMBOL(tcp_v4_conn_request);
1440
1441
1442 /*
1443  * The three way handshake has completed - we got a valid synack -
1444  * now create the new socket.
1445  */
1446 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1447                                   struct request_sock *req,
1448                                   struct dst_entry *dst)
1449 {
1450         struct inet_request_sock *ireq;
1451         struct inet_sock *newinet;
1452         struct tcp_sock *newtp;
1453         struct sock *newsk;
1454 #ifdef CONFIG_TCP_MD5SIG
1455         struct tcp_md5sig_key *key;
1456 #endif
1457         struct ip_options_rcu *inet_opt;
1458
1459         if (sk_acceptq_is_full(sk))
1460                 goto exit_overflow;
1461
1462         newsk = tcp_create_openreq_child(sk, req, skb);
1463         if (!newsk)
1464                 goto exit_nonewsk;
1465
1466         newsk->sk_gso_type = SKB_GSO_TCPV4;
1467         inet_sk_rx_dst_set(newsk, skb);
1468
1469         newtp                 = tcp_sk(newsk);
1470         newinet               = inet_sk(newsk);
1471         ireq                  = inet_rsk(req);
1472         newinet->inet_daddr   = ireq->rmt_addr;
1473         newinet->inet_rcv_saddr = ireq->loc_addr;
1474         newinet->inet_saddr           = ireq->loc_addr;
1475         inet_opt              = ireq->opt;
1476         rcu_assign_pointer(newinet->inet_opt, inet_opt);
1477         ireq->opt             = NULL;
1478         newinet->mc_index     = inet_iif(skb);
1479         newinet->mc_ttl       = ip_hdr(skb)->ttl;
1480         newinet->rcv_tos      = ip_hdr(skb)->tos;
1481         inet_csk(newsk)->icsk_ext_hdr_len = 0;
1482         if (inet_opt)
1483                 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1484         newinet->inet_id = newtp->write_seq ^ jiffies;
1485
1486         if (!dst) {
1487                 dst = inet_csk_route_child_sock(sk, newsk, req);
1488                 if (!dst)
1489                         goto put_and_exit;
1490         } else {
1491                 /* syncookie case : see end of cookie_v4_check() */
1492         }
1493         sk_setup_caps(newsk, dst);
1494
1495         tcp_mtup_init(newsk);
1496         tcp_sync_mss(newsk, dst_mtu(dst));
1497         newtp->advmss = dst_metric_advmss(dst);
1498         if (tcp_sk(sk)->rx_opt.user_mss &&
1499             tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1500                 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1501
1502         tcp_initialize_rcv_mss(newsk);
1503         if (tcp_rsk(req)->snt_synack)
1504                 tcp_valid_rtt_meas(newsk,
1505                     tcp_time_stamp - tcp_rsk(req)->snt_synack);
1506         newtp->total_retrans = req->retrans;
1507
1508 #ifdef CONFIG_TCP_MD5SIG
1509         /* Copy over the MD5 key from the original socket */
1510         key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1511                                 AF_INET);
1512         if (key != NULL) {
1513                 /*
1514                  * We're using one, so create a matching key
1515                  * on the newsk structure. If we fail to get
1516                  * memory, then we end up not copying the key
1517                  * across. Shucks.
1518                  */
1519                 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1520                                AF_INET, key->key, key->keylen, GFP_ATOMIC);
1521                 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1522         }
1523 #endif
1524
1525         if (__inet_inherit_port(sk, newsk) < 0)
1526                 goto put_and_exit;
1527         __inet_hash_nolisten(newsk, NULL);
1528
1529         return newsk;
1530
1531 exit_overflow:
1532         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1533 exit_nonewsk:
1534         dst_release(dst);
1535 exit:
1536         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1537         return NULL;
1538 put_and_exit:
1539         tcp_clear_xmit_timers(newsk);
1540         tcp_cleanup_congestion_control(newsk);
1541         bh_unlock_sock(newsk);
1542         sock_put(newsk);
1543         goto exit;
1544 }
1545 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1546
1547 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1548 {
1549         struct tcphdr *th = tcp_hdr(skb);
1550         const struct iphdr *iph = ip_hdr(skb);
1551         struct sock *nsk;
1552         struct request_sock **prev;
1553         /* Find possible connection requests. */
1554         struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1555                                                        iph->saddr, iph->daddr);
1556         if (req)
1557                 return tcp_check_req(sk, skb, req, prev);
1558
1559         nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1560                         th->source, iph->daddr, th->dest, inet_iif(skb));
1561
1562         if (nsk) {
1563                 if (nsk->sk_state != TCP_TIME_WAIT) {
1564                         bh_lock_sock(nsk);
1565                         return nsk;
1566                 }
1567                 inet_twsk_put(inet_twsk(nsk));
1568                 return NULL;
1569         }
1570
1571 #ifdef CONFIG_SYN_COOKIES
1572         if (!th->syn)
1573                 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1574 #endif
1575         return sk;
1576 }
1577
1578 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1579 {
1580         const struct iphdr *iph = ip_hdr(skb);
1581
1582         if (skb->ip_summed == CHECKSUM_COMPLETE) {
1583                 if (!tcp_v4_check(skb->len, iph->saddr,
1584                                   iph->daddr, skb->csum)) {
1585                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1586                         return 0;
1587                 }
1588         }
1589
1590         skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1591                                        skb->len, IPPROTO_TCP, 0);
1592
1593         if (skb->len <= 76) {
1594                 return __skb_checksum_complete(skb);
1595         }
1596         return 0;
1597 }
1598
1599
1600 /* The socket must have it's spinlock held when we get
1601  * here.
1602  *
1603  * We have a potential double-lock case here, so even when
1604  * doing backlog processing we use the BH locking scheme.
1605  * This is because we cannot sleep with the original spinlock
1606  * held.
1607  */
1608 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1609 {
1610         struct sock *rsk;
1611 #ifdef CONFIG_TCP_MD5SIG
1612         /*
1613          * We really want to reject the packet as early as possible
1614          * if:
1615          *  o We're expecting an MD5'd packet and this is no MD5 tcp option
1616          *  o There is an MD5 option and we're not expecting one
1617          */
1618         if (tcp_v4_inbound_md5_hash(sk, skb))
1619                 goto discard;
1620 #endif
1621
1622         if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1623                 struct dst_entry *dst = sk->sk_rx_dst;
1624
1625                 sock_rps_save_rxhash(sk, skb);
1626                 if (dst) {
1627                         if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1628                             dst->ops->check(dst, 0) == NULL) {
1629                                 dst_release(dst);
1630                                 sk->sk_rx_dst = NULL;
1631                         }
1632                 }
1633                 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1634                         rsk = sk;
1635                         goto reset;
1636                 }
1637                 return 0;
1638         }
1639
1640         if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1641                 goto csum_err;
1642
1643         if (sk->sk_state == TCP_LISTEN) {
1644                 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1645                 if (!nsk)
1646                         goto discard;
1647
1648                 if (nsk != sk) {
1649                         sock_rps_save_rxhash(nsk, skb);
1650                         if (tcp_child_process(sk, nsk, skb)) {
1651                                 rsk = nsk;
1652                                 goto reset;
1653                         }
1654                         return 0;
1655                 }
1656         } else
1657                 sock_rps_save_rxhash(sk, skb);
1658
1659         if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1660                 rsk = sk;
1661                 goto reset;
1662         }
1663         return 0;
1664
1665 reset:
1666         tcp_v4_send_reset(rsk, skb);
1667 discard:
1668         kfree_skb(skb);
1669         /* Be careful here. If this function gets more complicated and
1670          * gcc suffers from register pressure on the x86, sk (in %ebx)
1671          * might be destroyed here. This current version compiles correctly,
1672          * but you have been warned.
1673          */
1674         return 0;
1675
1676 csum_err:
1677         TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1678         goto discard;
1679 }
1680 EXPORT_SYMBOL(tcp_v4_do_rcv);
1681
1682 void tcp_v4_early_demux(struct sk_buff *skb)
1683 {
1684         struct net *net = dev_net(skb->dev);
1685         const struct iphdr *iph;
1686         const struct tcphdr *th;
1687         struct sock *sk;
1688
1689         if (skb->pkt_type != PACKET_HOST)
1690                 return;
1691
1692         if (!pskb_may_pull(skb, ip_hdrlen(skb) + sizeof(struct tcphdr)))
1693                 return;
1694
1695         iph = ip_hdr(skb);
1696         th = (struct tcphdr *) ((char *)iph + ip_hdrlen(skb));
1697
1698         if (th->doff < sizeof(struct tcphdr) / 4)
1699                 return;
1700
1701         sk = __inet_lookup_established(net, &tcp_hashinfo,
1702                                        iph->saddr, th->source,
1703                                        iph->daddr, ntohs(th->dest),
1704                                        skb->skb_iif);
1705         if (sk) {
1706                 skb->sk = sk;
1707                 skb->destructor = sock_edemux;
1708                 if (sk->sk_state != TCP_TIME_WAIT) {
1709                         struct dst_entry *dst = sk->sk_rx_dst;
1710
1711                         if (dst)
1712                                 dst = dst_check(dst, 0);
1713                         if (dst &&
1714                             inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1715                                 skb_dst_set_noref(skb, dst);
1716                 }
1717         }
1718 }
1719
1720 /*
1721  *      From tcp_input.c
1722  */
1723
1724 int tcp_v4_rcv(struct sk_buff *skb)
1725 {
1726         const struct iphdr *iph;
1727         const struct tcphdr *th;
1728         struct sock *sk;
1729         int ret;
1730         struct net *net = dev_net(skb->dev);
1731
1732         if (skb->pkt_type != PACKET_HOST)
1733                 goto discard_it;
1734
1735         /* Count it even if it's bad */
1736         TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1737
1738         if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1739                 goto discard_it;
1740
1741         th = tcp_hdr(skb);
1742
1743         if (th->doff < sizeof(struct tcphdr) / 4)
1744                 goto bad_packet;
1745         if (!pskb_may_pull(skb, th->doff * 4))
1746                 goto discard_it;
1747
1748         /* An explanation is required here, I think.
1749          * Packet length and doff are validated by header prediction,
1750          * provided case of th->doff==0 is eliminated.
1751          * So, we defer the checks. */
1752         if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1753                 goto bad_packet;
1754
1755         th = tcp_hdr(skb);
1756         iph = ip_hdr(skb);
1757         TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1758         TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1759                                     skb->len - th->doff * 4);
1760         TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1761         TCP_SKB_CB(skb)->when    = 0;
1762         TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1763         TCP_SKB_CB(skb)->sacked  = 0;
1764
1765         sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1766         if (!sk)
1767                 goto no_tcp_socket;
1768
1769 process:
1770         if (sk->sk_state == TCP_TIME_WAIT)
1771                 goto do_time_wait;
1772
1773         if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1774                 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1775                 goto discard_and_relse;
1776         }
1777
1778         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1779                 goto discard_and_relse;
1780         nf_reset(skb);
1781
1782         if (sk_filter(sk, skb))
1783                 goto discard_and_relse;
1784
1785         skb->dev = NULL;
1786
1787         bh_lock_sock_nested(sk);
1788         ret = 0;
1789         if (!sock_owned_by_user(sk)) {
1790 #ifdef CONFIG_NET_DMA
1791                 struct tcp_sock *tp = tcp_sk(sk);
1792                 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1793                         tp->ucopy.dma_chan = net_dma_find_channel();
1794                 if (tp->ucopy.dma_chan)
1795                         ret = tcp_v4_do_rcv(sk, skb);
1796                 else
1797 #endif
1798                 {
1799                         if (!tcp_prequeue(sk, skb))
1800                                 ret = tcp_v4_do_rcv(sk, skb);
1801                 }
1802         } else if (unlikely(sk_add_backlog(sk, skb,
1803                                            sk->sk_rcvbuf + sk->sk_sndbuf))) {
1804                 bh_unlock_sock(sk);
1805                 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1806                 goto discard_and_relse;
1807         }
1808         bh_unlock_sock(sk);
1809
1810         sock_put(sk);
1811
1812         return ret;
1813
1814 no_tcp_socket:
1815         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1816                 goto discard_it;
1817
1818         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1819 bad_packet:
1820                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1821         } else {
1822                 tcp_v4_send_reset(NULL, skb);
1823         }
1824
1825 discard_it:
1826         /* Discard frame. */
1827         kfree_skb(skb);
1828         return 0;
1829
1830 discard_and_relse:
1831         sock_put(sk);
1832         goto discard_it;
1833
1834 do_time_wait:
1835         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1836                 inet_twsk_put(inet_twsk(sk));
1837                 goto discard_it;
1838         }
1839
1840         if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1841                 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1842                 inet_twsk_put(inet_twsk(sk));
1843                 goto discard_it;
1844         }
1845         switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1846         case TCP_TW_SYN: {
1847                 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1848                                                         &tcp_hashinfo,
1849                                                         iph->daddr, th->dest,
1850                                                         inet_iif(skb));
1851                 if (sk2) {
1852                         inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1853                         inet_twsk_put(inet_twsk(sk));
1854                         sk = sk2;
1855                         goto process;
1856                 }
1857                 /* Fall through to ACK */
1858         }
1859         case TCP_TW_ACK:
1860                 tcp_v4_timewait_ack(sk, skb);
1861                 break;
1862         case TCP_TW_RST:
1863                 goto no_tcp_socket;
1864         case TCP_TW_SUCCESS:;
1865         }
1866         goto discard_it;
1867 }
1868
1869 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1870         .twsk_obj_size  = sizeof(struct tcp_timewait_sock),
1871         .twsk_unique    = tcp_twsk_unique,
1872         .twsk_destructor= tcp_twsk_destructor,
1873 };
1874
1875 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1876 {
1877         struct dst_entry *dst = skb_dst(skb);
1878
1879         dst_hold(dst);
1880         sk->sk_rx_dst = dst;
1881         inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1882 }
1883 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1884
1885 const struct inet_connection_sock_af_ops ipv4_specific = {
1886         .queue_xmit        = ip_queue_xmit,
1887         .send_check        = tcp_v4_send_check,
1888         .rebuild_header    = inet_sk_rebuild_header,
1889         .sk_rx_dst_set     = inet_sk_rx_dst_set,
1890         .conn_request      = tcp_v4_conn_request,
1891         .syn_recv_sock     = tcp_v4_syn_recv_sock,
1892         .net_header_len    = sizeof(struct iphdr),
1893         .setsockopt        = ip_setsockopt,
1894         .getsockopt        = ip_getsockopt,
1895         .addr2sockaddr     = inet_csk_addr2sockaddr,
1896         .sockaddr_len      = sizeof(struct sockaddr_in),
1897         .bind_conflict     = inet_csk_bind_conflict,
1898 #ifdef CONFIG_COMPAT
1899         .compat_setsockopt = compat_ip_setsockopt,
1900         .compat_getsockopt = compat_ip_getsockopt,
1901 #endif
1902 };
1903 EXPORT_SYMBOL(ipv4_specific);
1904
1905 #ifdef CONFIG_TCP_MD5SIG
1906 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1907         .md5_lookup             = tcp_v4_md5_lookup,
1908         .calc_md5_hash          = tcp_v4_md5_hash_skb,
1909         .md5_parse              = tcp_v4_parse_md5_keys,
1910 };
1911 #endif
1912
1913 /* NOTE: A lot of things set to zero explicitly by call to
1914  *       sk_alloc() so need not be done here.
1915  */
1916 static int tcp_v4_init_sock(struct sock *sk)
1917 {
1918         struct inet_connection_sock *icsk = inet_csk(sk);
1919
1920         tcp_init_sock(sk);
1921
1922         icsk->icsk_af_ops = &ipv4_specific;
1923
1924 #ifdef CONFIG_TCP_MD5SIG
1925         tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1926 #endif
1927
1928         return 0;
1929 }
1930
1931 void tcp_v4_destroy_sock(struct sock *sk)
1932 {
1933         struct tcp_sock *tp = tcp_sk(sk);
1934
1935         tcp_clear_xmit_timers(sk);
1936
1937         tcp_cleanup_congestion_control(sk);
1938
1939         /* Cleanup up the write buffer. */
1940         tcp_write_queue_purge(sk);
1941
1942         /* Cleans up our, hopefully empty, out_of_order_queue. */
1943         __skb_queue_purge(&tp->out_of_order_queue);
1944
1945 #ifdef CONFIG_TCP_MD5SIG
1946         /* Clean up the MD5 key list, if any */
1947         if (tp->md5sig_info) {
1948                 tcp_clear_md5_list(sk);
1949                 kfree_rcu(tp->md5sig_info, rcu);
1950                 tp->md5sig_info = NULL;
1951         }
1952 #endif
1953
1954 #ifdef CONFIG_NET_DMA
1955         /* Cleans up our sk_async_wait_queue */
1956         __skb_queue_purge(&sk->sk_async_wait_queue);
1957 #endif
1958
1959         /* Clean prequeue, it must be empty really */
1960         __skb_queue_purge(&tp->ucopy.prequeue);
1961
1962         /* Clean up a referenced TCP bind bucket. */
1963         if (inet_csk(sk)->icsk_bind_hash)
1964                 inet_put_port(sk);
1965
1966         /*
1967          * If sendmsg cached page exists, toss it.
1968          */
1969         if (sk->sk_sndmsg_page) {
1970                 __free_page(sk->sk_sndmsg_page);
1971                 sk->sk_sndmsg_page = NULL;
1972         }
1973
1974         /* TCP Cookie Transactions */
1975         if (tp->cookie_values != NULL) {
1976                 kref_put(&tp->cookie_values->kref,
1977                          tcp_cookie_values_release);
1978                 tp->cookie_values = NULL;
1979         }
1980
1981         /* If socket is aborted during connect operation */
1982         tcp_free_fastopen_req(tp);
1983
1984         sk_sockets_allocated_dec(sk);
1985         sock_release_memcg(sk);
1986 }
1987 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1988
1989 #ifdef CONFIG_PROC_FS
1990 /* Proc filesystem TCP sock list dumping. */
1991
1992 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1993 {
1994         return hlist_nulls_empty(head) ? NULL :
1995                 list_entry(head->first, struct inet_timewait_sock, tw_node);
1996 }
1997
1998 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1999 {
2000         return !is_a_nulls(tw->tw_node.next) ?
2001                 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
2002 }
2003
2004 /*
2005  * Get next listener socket follow cur.  If cur is NULL, get first socket
2006  * starting from bucket given in st->bucket; when st->bucket is zero the
2007  * very first socket in the hash table is returned.
2008  */
2009 static void *listening_get_next(struct seq_file *seq, void *cur)
2010 {
2011         struct inet_connection_sock *icsk;
2012         struct hlist_nulls_node *node;
2013         struct sock *sk = cur;
2014         struct inet_listen_hashbucket *ilb;
2015         struct tcp_iter_state *st = seq->private;
2016         struct net *net = seq_file_net(seq);
2017
2018         if (!sk) {
2019                 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2020                 spin_lock_bh(&ilb->lock);
2021                 sk = sk_nulls_head(&ilb->head);
2022                 st->offset = 0;
2023                 goto get_sk;
2024         }
2025         ilb = &tcp_hashinfo.listening_hash[st->bucket];
2026         ++st->num;
2027         ++st->offset;
2028
2029         if (st->state == TCP_SEQ_STATE_OPENREQ) {
2030                 struct request_sock *req = cur;
2031
2032                 icsk = inet_csk(st->syn_wait_sk);
2033                 req = req->dl_next;
2034                 while (1) {
2035                         while (req) {
2036                                 if (req->rsk_ops->family == st->family) {
2037                                         cur = req;
2038                                         goto out;
2039                                 }
2040                                 req = req->dl_next;
2041                         }
2042                         if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2043                                 break;
2044 get_req:
2045                         req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2046                 }
2047                 sk        = sk_nulls_next(st->syn_wait_sk);
2048                 st->state = TCP_SEQ_STATE_LISTENING;
2049                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2050         } else {
2051                 icsk = inet_csk(sk);
2052                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2053                 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2054                         goto start_req;
2055                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2056                 sk = sk_nulls_next(sk);
2057         }
2058 get_sk:
2059         sk_nulls_for_each_from(sk, node) {
2060                 if (!net_eq(sock_net(sk), net))
2061                         continue;
2062                 if (sk->sk_family == st->family) {
2063                         cur = sk;
2064                         goto out;
2065                 }
2066                 icsk = inet_csk(sk);
2067                 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2068                 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2069 start_req:
2070                         st->uid         = sock_i_uid(sk);
2071                         st->syn_wait_sk = sk;
2072                         st->state       = TCP_SEQ_STATE_OPENREQ;
2073                         st->sbucket     = 0;
2074                         goto get_req;
2075                 }
2076                 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2077         }
2078         spin_unlock_bh(&ilb->lock);
2079         st->offset = 0;
2080         if (++st->bucket < INET_LHTABLE_SIZE) {
2081                 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2082                 spin_lock_bh(&ilb->lock);
2083                 sk = sk_nulls_head(&ilb->head);
2084                 goto get_sk;
2085         }
2086         cur = NULL;
2087 out:
2088         return cur;
2089 }
2090
2091 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2092 {
2093         struct tcp_iter_state *st = seq->private;
2094         void *rc;
2095
2096         st->bucket = 0;
2097         st->offset = 0;
2098         rc = listening_get_next(seq, NULL);
2099
2100         while (rc && *pos) {
2101                 rc = listening_get_next(seq, rc);
2102                 --*pos;
2103         }
2104         return rc;
2105 }
2106
2107 static inline bool empty_bucket(struct tcp_iter_state *st)
2108 {
2109         return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2110                 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2111 }
2112
2113 /*
2114  * Get first established socket starting from bucket given in st->bucket.
2115  * If st->bucket is zero, the very first socket in the hash is returned.
2116  */
2117 static void *established_get_first(struct seq_file *seq)
2118 {
2119         struct tcp_iter_state *st = seq->private;
2120         struct net *net = seq_file_net(seq);
2121         void *rc = NULL;
2122
2123         st->offset = 0;
2124         for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2125                 struct sock *sk;
2126                 struct hlist_nulls_node *node;
2127                 struct inet_timewait_sock *tw;
2128                 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2129
2130                 /* Lockless fast path for the common case of empty buckets */
2131                 if (empty_bucket(st))
2132                         continue;
2133
2134                 spin_lock_bh(lock);
2135                 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2136                         if (sk->sk_family != st->family ||
2137                             !net_eq(sock_net(sk), net)) {
2138                                 continue;
2139                         }
2140                         rc = sk;
2141                         goto out;
2142                 }
2143                 st->state = TCP_SEQ_STATE_TIME_WAIT;
2144                 inet_twsk_for_each(tw, node,
2145                                    &tcp_hashinfo.ehash[st->bucket].twchain) {
2146                         if (tw->tw_family != st->family ||
2147                             !net_eq(twsk_net(tw), net)) {
2148                                 continue;
2149                         }
2150                         rc = tw;
2151                         goto out;
2152                 }
2153                 spin_unlock_bh(lock);
2154                 st->state = TCP_SEQ_STATE_ESTABLISHED;
2155         }
2156 out:
2157         return rc;
2158 }
2159
2160 static void *established_get_next(struct seq_file *seq, void *cur)
2161 {
2162         struct sock *sk = cur;
2163         struct inet_timewait_sock *tw;
2164         struct hlist_nulls_node *node;
2165         struct tcp_iter_state *st = seq->private;
2166         struct net *net = seq_file_net(seq);
2167
2168         ++st->num;
2169         ++st->offset;
2170
2171         if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2172                 tw = cur;
2173                 tw = tw_next(tw);
2174 get_tw:
2175                 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2176                         tw = tw_next(tw);
2177                 }
2178                 if (tw) {
2179                         cur = tw;
2180                         goto out;
2181                 }
2182                 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2183                 st->state = TCP_SEQ_STATE_ESTABLISHED;
2184
2185                 /* Look for next non empty bucket */
2186                 st->offset = 0;
2187                 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2188                                 empty_bucket(st))
2189                         ;
2190                 if (st->bucket > tcp_hashinfo.ehash_mask)
2191                         return NULL;
2192
2193                 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2194                 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2195         } else
2196                 sk = sk_nulls_next(sk);
2197
2198         sk_nulls_for_each_from(sk, node) {
2199                 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2200                         goto found;
2201         }
2202
2203         st->state = TCP_SEQ_STATE_TIME_WAIT;
2204         tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2205         goto get_tw;
2206 found:
2207         cur = sk;
2208 out:
2209         return cur;
2210 }
2211
2212 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2213 {
2214         struct tcp_iter_state *st = seq->private;
2215         void *rc;
2216
2217         st->bucket = 0;
2218         rc = established_get_first(seq);
2219
2220         while (rc && pos) {
2221                 rc = established_get_next(seq, rc);
2222                 --pos;
2223         }
2224         return rc;
2225 }
2226
2227 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2228 {
2229         void *rc;
2230         struct tcp_iter_state *st = seq->private;
2231
2232         st->state = TCP_SEQ_STATE_LISTENING;
2233         rc        = listening_get_idx(seq, &pos);
2234
2235         if (!rc) {
2236                 st->state = TCP_SEQ_STATE_ESTABLISHED;
2237                 rc        = established_get_idx(seq, pos);
2238         }
2239
2240         return rc;
2241 }
2242
2243 static void *tcp_seek_last_pos(struct seq_file *seq)
2244 {
2245         struct tcp_iter_state *st = seq->private;
2246         int offset = st->offset;
2247         int orig_num = st->num;
2248         void *rc = NULL;
2249
2250         switch (st->state) {
2251         case TCP_SEQ_STATE_OPENREQ:
2252         case TCP_SEQ_STATE_LISTENING:
2253                 if (st->bucket >= INET_LHTABLE_SIZE)
2254                         break;
2255                 st->state = TCP_SEQ_STATE_LISTENING;
2256                 rc = listening_get_next(seq, NULL);
2257                 while (offset-- && rc)
2258                         rc = listening_get_next(seq, rc);
2259                 if (rc)
2260                         break;
2261                 st->bucket = 0;
2262                 /* Fallthrough */
2263         case TCP_SEQ_STATE_ESTABLISHED:
2264         case TCP_SEQ_STATE_TIME_WAIT:
2265                 st->state = TCP_SEQ_STATE_ESTABLISHED;
2266                 if (st->bucket > tcp_hashinfo.ehash_mask)
2267                         break;
2268                 rc = established_get_first(seq);
2269                 while (offset-- && rc)
2270                         rc = established_get_next(seq, rc);
2271         }
2272
2273         st->num = orig_num;
2274
2275         return rc;
2276 }
2277
2278 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2279 {
2280         struct tcp_iter_state *st = seq->private;
2281         void *rc;
2282
2283         if (*pos && *pos == st->last_pos) {
2284                 rc = tcp_seek_last_pos(seq);
2285                 if (rc)
2286                         goto out;
2287         }
2288
2289         st->state = TCP_SEQ_STATE_LISTENING;
2290         st->num = 0;
2291         st->bucket = 0;
2292         st->offset = 0;
2293         rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2294
2295 out:
2296         st->last_pos = *pos;
2297         return rc;
2298 }
2299
2300 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2301 {
2302         struct tcp_iter_state *st = seq->private;
2303         void *rc = NULL;
2304
2305         if (v == SEQ_START_TOKEN) {
2306                 rc = tcp_get_idx(seq, 0);
2307                 goto out;
2308         }
2309
2310         switch (st->state) {
2311         case TCP_SEQ_STATE_OPENREQ:
2312         case TCP_SEQ_STATE_LISTENING:
2313                 rc = listening_get_next(seq, v);
2314                 if (!rc) {
2315                         st->state = TCP_SEQ_STATE_ESTABLISHED;
2316                         st->bucket = 0;
2317                         st->offset = 0;
2318                         rc        = established_get_first(seq);
2319                 }
2320                 break;
2321         case TCP_SEQ_STATE_ESTABLISHED:
2322         case TCP_SEQ_STATE_TIME_WAIT:
2323                 rc = established_get_next(seq, v);
2324                 break;
2325         }
2326 out:
2327         ++*pos;
2328         st->last_pos = *pos;
2329         return rc;
2330 }
2331
2332 static void tcp_seq_stop(struct seq_file *seq, void *v)
2333 {
2334         struct tcp_iter_state *st = seq->private;
2335
2336         switch (st->state) {
2337         case TCP_SEQ_STATE_OPENREQ:
2338                 if (v) {
2339                         struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2340                         read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2341                 }
2342         case TCP_SEQ_STATE_LISTENING:
2343                 if (v != SEQ_START_TOKEN)
2344                         spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2345                 break;
2346         case TCP_SEQ_STATE_TIME_WAIT:
2347         case TCP_SEQ_STATE_ESTABLISHED:
2348                 if (v)
2349                         spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2350                 break;
2351         }
2352 }
2353
2354 int tcp_seq_open(struct inode *inode, struct file *file)
2355 {
2356         struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2357         struct tcp_iter_state *s;
2358         int err;
2359
2360         err = seq_open_net(inode, file, &afinfo->seq_ops,
2361                           sizeof(struct tcp_iter_state));
2362         if (err < 0)
2363                 return err;
2364
2365         s = ((struct seq_file *)file->private_data)->private;
2366         s->family               = afinfo->family;
2367         s->last_pos             = 0;
2368         return 0;
2369 }
2370 EXPORT_SYMBOL(tcp_seq_open);
2371
2372 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2373 {
2374         int rc = 0;
2375         struct proc_dir_entry *p;
2376
2377         afinfo->seq_ops.start           = tcp_seq_start;
2378         afinfo->seq_ops.next            = tcp_seq_next;
2379         afinfo->seq_ops.stop            = tcp_seq_stop;
2380
2381         p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2382                              afinfo->seq_fops, afinfo);
2383         if (!p)
2384                 rc = -ENOMEM;
2385         return rc;
2386 }
2387 EXPORT_SYMBOL(tcp_proc_register);
2388
2389 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2390 {
2391         proc_net_remove(net, afinfo->name);
2392 }
2393 EXPORT_SYMBOL(tcp_proc_unregister);
2394
2395 static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2396                          struct seq_file *f, int i, int uid, int *len)
2397 {
2398         const struct inet_request_sock *ireq = inet_rsk(req);
2399         int ttd = req->expires - jiffies;
2400
2401         seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2402                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2403                 i,
2404                 ireq->loc_addr,
2405                 ntohs(inet_sk(sk)->inet_sport),
2406                 ireq->rmt_addr,
2407                 ntohs(ireq->rmt_port),
2408                 TCP_SYN_RECV,
2409                 0, 0, /* could print option size, but that is af dependent. */
2410                 1,    /* timers active (only the expire timer) */
2411                 jiffies_to_clock_t(ttd),
2412                 req->retrans,
2413                 uid,
2414                 0,  /* non standard timer */
2415                 0, /* open_requests have no inode */
2416                 atomic_read(&sk->sk_refcnt),
2417                 req,
2418                 len);
2419 }
2420
2421 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2422 {
2423         int timer_active;
2424         unsigned long timer_expires;
2425         const struct tcp_sock *tp = tcp_sk(sk);
2426         const struct inet_connection_sock *icsk = inet_csk(sk);
2427         const struct inet_sock *inet = inet_sk(sk);
2428         __be32 dest = inet->inet_daddr;
2429         __be32 src = inet->inet_rcv_saddr;
2430         __u16 destp = ntohs(inet->inet_dport);
2431         __u16 srcp = ntohs(inet->inet_sport);
2432         int rx_queue;
2433
2434         if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2435                 timer_active    = 1;
2436                 timer_expires   = icsk->icsk_timeout;
2437         } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2438                 timer_active    = 4;
2439                 timer_expires   = icsk->icsk_timeout;
2440         } else if (timer_pending(&sk->sk_timer)) {
2441                 timer_active    = 2;
2442                 timer_expires   = sk->sk_timer.expires;
2443         } else {
2444                 timer_active    = 0;
2445                 timer_expires = jiffies;
2446         }
2447
2448         if (sk->sk_state == TCP_LISTEN)
2449                 rx_queue = sk->sk_ack_backlog;
2450         else
2451                 /*
2452                  * because we dont lock socket, we might find a transient negative value
2453                  */
2454                 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2455
2456         seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2457                         "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2458                 i, src, srcp, dest, destp, sk->sk_state,
2459                 tp->write_seq - tp->snd_una,
2460                 rx_queue,
2461                 timer_active,
2462                 jiffies_to_clock_t(timer_expires - jiffies),
2463                 icsk->icsk_retransmits,
2464                 sock_i_uid(sk),
2465                 icsk->icsk_probes_out,
2466                 sock_i_ino(sk),
2467                 atomic_read(&sk->sk_refcnt), sk,
2468                 jiffies_to_clock_t(icsk->icsk_rto),
2469                 jiffies_to_clock_t(icsk->icsk_ack.ato),
2470                 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2471                 tp->snd_cwnd,
2472                 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2473                 len);
2474 }
2475
2476 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2477                                struct seq_file *f, int i, int *len)
2478 {
2479         __be32 dest, src;
2480         __u16 destp, srcp;
2481         int ttd = tw->tw_ttd - jiffies;
2482
2483         if (ttd < 0)
2484                 ttd = 0;
2485
2486         dest  = tw->tw_daddr;
2487         src   = tw->tw_rcv_saddr;
2488         destp = ntohs(tw->tw_dport);
2489         srcp  = ntohs(tw->tw_sport);
2490
2491         seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2492                 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2493                 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2494                 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2495                 atomic_read(&tw->tw_refcnt), tw, len);
2496 }
2497
2498 #define TMPSZ 150
2499
2500 static int tcp4_seq_show(struct seq_file *seq, void *v)
2501 {
2502         struct tcp_iter_state *st;
2503         int len;
2504
2505         if (v == SEQ_START_TOKEN) {
2506                 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2507                            "  sl  local_address rem_address   st tx_queue "
2508                            "rx_queue tr tm->when retrnsmt   uid  timeout "
2509                            "inode");
2510                 goto out;
2511         }
2512         st = seq->private;
2513
2514         switch (st->state) {
2515         case TCP_SEQ_STATE_LISTENING:
2516         case TCP_SEQ_STATE_ESTABLISHED:
2517                 get_tcp4_sock(v, seq, st->num, &len);
2518                 break;
2519         case TCP_SEQ_STATE_OPENREQ:
2520                 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2521                 break;
2522         case TCP_SEQ_STATE_TIME_WAIT:
2523                 get_timewait4_sock(v, seq, st->num, &len);
2524                 break;
2525         }
2526         seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2527 out:
2528         return 0;
2529 }
2530
2531 static const struct file_operations tcp_afinfo_seq_fops = {
2532         .owner   = THIS_MODULE,
2533         .open    = tcp_seq_open,
2534         .read    = seq_read,
2535         .llseek  = seq_lseek,
2536         .release = seq_release_net
2537 };
2538
2539 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2540         .name           = "tcp",
2541         .family         = AF_INET,
2542         .seq_fops       = &tcp_afinfo_seq_fops,
2543         .seq_ops        = {
2544                 .show           = tcp4_seq_show,
2545         },
2546 };
2547
2548 static int __net_init tcp4_proc_init_net(struct net *net)
2549 {
2550         return tcp_proc_register(net, &tcp4_seq_afinfo);
2551 }
2552
2553 static void __net_exit tcp4_proc_exit_net(struct net *net)
2554 {
2555         tcp_proc_unregister(net, &tcp4_seq_afinfo);
2556 }
2557
2558 static struct pernet_operations tcp4_net_ops = {
2559         .init = tcp4_proc_init_net,
2560         .exit = tcp4_proc_exit_net,
2561 };
2562
2563 int __init tcp4_proc_init(void)
2564 {
2565         return register_pernet_subsys(&tcp4_net_ops);
2566 }
2567
2568 void tcp4_proc_exit(void)
2569 {
2570         unregister_pernet_subsys(&tcp4_net_ops);
2571 }
2572 #endif /* CONFIG_PROC_FS */
2573
2574 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2575 {
2576         const struct iphdr *iph = skb_gro_network_header(skb);
2577
2578         switch (skb->ip_summed) {
2579         case CHECKSUM_COMPLETE:
2580                 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2581                                   skb->csum)) {
2582                         skb->ip_summed = CHECKSUM_UNNECESSARY;
2583                         break;
2584                 }
2585
2586                 /* fall through */
2587         case CHECKSUM_NONE:
2588                 NAPI_GRO_CB(skb)->flush = 1;
2589                 return NULL;
2590         }
2591
2592         return tcp_gro_receive(head, skb);
2593 }
2594
2595 int tcp4_gro_complete(struct sk_buff *skb)
2596 {
2597         const struct iphdr *iph = ip_hdr(skb);
2598         struct tcphdr *th = tcp_hdr(skb);
2599
2600         th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2601                                   iph->saddr, iph->daddr, 0);
2602         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2603
2604         return tcp_gro_complete(skb);
2605 }
2606
2607 struct proto tcp_prot = {
2608         .name                   = "TCP",
2609         .owner                  = THIS_MODULE,
2610         .close                  = tcp_close,
2611         .connect                = tcp_v4_connect,
2612         .disconnect             = tcp_disconnect,
2613         .accept                 = inet_csk_accept,
2614         .ioctl                  = tcp_ioctl,
2615         .init                   = tcp_v4_init_sock,
2616         .destroy                = tcp_v4_destroy_sock,
2617         .shutdown               = tcp_shutdown,
2618         .setsockopt             = tcp_setsockopt,
2619         .getsockopt             = tcp_getsockopt,
2620         .recvmsg                = tcp_recvmsg,
2621         .sendmsg                = tcp_sendmsg,
2622         .sendpage               = tcp_sendpage,
2623         .backlog_rcv            = tcp_v4_do_rcv,
2624         .release_cb             = tcp_release_cb,
2625         .mtu_reduced            = tcp_v4_mtu_reduced,
2626         .hash                   = inet_hash,
2627         .unhash                 = inet_unhash,
2628         .get_port               = inet_csk_get_port,
2629         .enter_memory_pressure  = tcp_enter_memory_pressure,
2630         .sockets_allocated      = &tcp_sockets_allocated,
2631         .orphan_count           = &tcp_orphan_count,
2632         .memory_allocated       = &tcp_memory_allocated,
2633         .memory_pressure        = &tcp_memory_pressure,
2634         .sysctl_wmem            = sysctl_tcp_wmem,
2635         .sysctl_rmem            = sysctl_tcp_rmem,
2636         .max_header             = MAX_TCP_HEADER,
2637         .obj_size               = sizeof(struct tcp_sock),
2638         .slab_flags             = SLAB_DESTROY_BY_RCU,
2639         .twsk_prot              = &tcp_timewait_sock_ops,
2640         .rsk_prot               = &tcp_request_sock_ops,
2641         .h.hashinfo             = &tcp_hashinfo,
2642         .no_autobind            = true,
2643 #ifdef CONFIG_COMPAT
2644         .compat_setsockopt      = compat_tcp_setsockopt,
2645         .compat_getsockopt      = compat_tcp_getsockopt,
2646 #endif
2647 #ifdef CONFIG_MEMCG_KMEM
2648         .init_cgroup            = tcp_init_cgroup,
2649         .destroy_cgroup         = tcp_destroy_cgroup,
2650         .proto_cgroup           = tcp_proto_cgroup,
2651 #endif
2652 };
2653 EXPORT_SYMBOL(tcp_prot);
2654
2655 static int __net_init tcp_sk_init(struct net *net)
2656 {
2657         return 0;
2658 }
2659
2660 static void __net_exit tcp_sk_exit(struct net *net)
2661 {
2662 }
2663
2664 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2665 {
2666         inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2667 }
2668
2669 static struct pernet_operations __net_initdata tcp_sk_ops = {
2670        .init       = tcp_sk_init,
2671        .exit       = tcp_sk_exit,
2672        .exit_batch = tcp_sk_exit_batch,
2673 };
2674
2675 void __init tcp_v4_init(void)
2676 {
2677         inet_hashinfo_init(&tcp_hashinfo);
2678         if (register_pernet_subsys(&tcp_sk_ops))
2679                 panic("Failed to create the TCP control socket.\n");
2680 }