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.
6 * Implementation of the Transmission Control Protocol(TCP).
8 * IPv4 specific functions
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
16 * See tcp.c for author information
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.
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
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
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.
54 #include <linux/bottom_half.h>
55 #include <linux/types.h>
56 #include <linux/fcntl.h>
57 #include <linux/module.h>
58 #include <linux/random.h>
59 #include <linux/cache.h>
60 #include <linux/jhash.h>
61 #include <linux/init.h>
62 #include <linux/times.h>
64 #include <net/net_namespace.h>
66 #include <net/inet_hashtables.h>
68 #include <net/transp_v6.h>
70 #include <net/inet_common.h>
71 #include <net/timewait_sock.h>
73 #include <net/netdma.h>
75 #include <linux/inet.h>
76 #include <linux/ipv6.h>
77 #include <linux/stddef.h>
78 #include <linux/proc_fs.h>
79 #include <linux/seq_file.h>
81 #include <linux/crypto.h>
82 #include <linux/scatterlist.h>
84 int sysctl_tcp_tw_reuse __read_mostly;
85 int sysctl_tcp_low_latency __read_mostly;
88 #ifdef CONFIG_TCP_MD5SIG
89 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
91 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
92 __be32 daddr, __be32 saddr, struct tcphdr *th);
95 struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
101 struct inet_hashinfo tcp_hashinfo;
103 static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
105 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
108 tcp_hdr(skb)->source);
111 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
113 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
114 struct tcp_sock *tp = tcp_sk(sk);
116 /* With PAWS, it is safe from the viewpoint
117 of data integrity. Even without PAWS it is safe provided sequence
118 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
120 Actually, the idea is close to VJ's one, only timestamp cache is
121 held not per host, but per port pair and TW bucket is used as state
124 If TW bucket has been already destroyed we fall back to VJ's scheme
125 and use initial timestamp retrieved from peer table.
127 if (tcptw->tw_ts_recent_stamp &&
128 (twp == NULL || (sysctl_tcp_tw_reuse &&
129 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
130 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
131 if (tp->write_seq == 0)
133 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
134 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
142 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
144 /* This will initiate an outgoing connection. */
145 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
147 struct inet_sock *inet = inet_sk(sk);
148 struct tcp_sock *tp = tcp_sk(sk);
149 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
151 __be32 daddr, nexthop;
155 if (addr_len < sizeof(struct sockaddr_in))
158 if (usin->sin_family != AF_INET)
159 return -EAFNOSUPPORT;
161 nexthop = daddr = usin->sin_addr.s_addr;
162 if (inet->opt && inet->opt->srr) {
165 nexthop = inet->opt->faddr;
168 tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
169 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
171 inet->inet_sport, usin->sin_port, sk, 1);
173 if (tmp == -ENETUNREACH)
174 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
178 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
183 if (!inet->opt || !inet->opt->srr)
186 if (!inet->inet_saddr)
187 inet->inet_saddr = rt->rt_src;
188 inet->inet_rcv_saddr = inet->inet_saddr;
190 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
191 /* Reset inherited state */
192 tp->rx_opt.ts_recent = 0;
193 tp->rx_opt.ts_recent_stamp = 0;
197 if (tcp_death_row.sysctl_tw_recycle &&
198 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
199 struct inet_peer *peer = rt_get_peer(rt);
201 * VJ's idea. We save last timestamp seen from
202 * the destination in peer table, when entering state
203 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
204 * when trying new connection.
207 (u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
208 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
209 tp->rx_opt.ts_recent = peer->tcp_ts;
213 inet->inet_dport = usin->sin_port;
214 inet->inet_daddr = daddr;
216 inet_csk(sk)->icsk_ext_hdr_len = 0;
218 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
220 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
222 /* Socket identity is still unknown (sport may be zero).
223 * However we set state to SYN-SENT and not releasing socket
224 * lock select source port, enter ourselves into the hash tables and
225 * complete initialization after this.
227 tcp_set_state(sk, TCP_SYN_SENT);
228 err = inet_hash_connect(&tcp_death_row, sk);
232 err = ip_route_newports(&rt, IPPROTO_TCP,
233 inet->inet_sport, inet->inet_dport, sk);
237 /* OK, now commit destination to socket. */
238 sk->sk_gso_type = SKB_GSO_TCPV4;
239 sk_setup_caps(sk, &rt->u.dst);
242 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
247 inet->inet_id = tp->write_seq ^ jiffies;
249 err = tcp_connect(sk);
258 * This unhashes the socket and releases the local port,
261 tcp_set_state(sk, TCP_CLOSE);
263 sk->sk_route_caps = 0;
264 inet->inet_dport = 0;
269 * This routine does path mtu discovery as defined in RFC1191.
271 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
276 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
277 * send out by Linux are always <576bytes so they should go through
280 if (sk->sk_state == TCP_LISTEN)
283 /* We don't check in the destentry if pmtu discovery is forbidden
284 * on this route. We just assume that no packet_to_big packets
285 * are send back when pmtu discovery is not active.
286 * There is a small race when the user changes this flag in the
287 * route, but I think that's acceptable.
289 if ((dst = __sk_dst_check(sk, 0)) == NULL)
292 dst->ops->update_pmtu(dst, mtu);
294 /* Something is about to be wrong... Remember soft error
295 * for the case, if this connection will not able to recover.
297 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
298 sk->sk_err_soft = EMSGSIZE;
302 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
303 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
304 tcp_sync_mss(sk, mtu);
306 /* Resend the TCP packet because it's
307 * clear that the old packet has been
308 * dropped. This is the new "fast" path mtu
311 tcp_simple_retransmit(sk);
312 } /* else let the usual retransmit timer handle it */
316 * This routine is called by the ICMP module when it gets some
317 * sort of error condition. If err < 0 then the socket should
318 * be closed and the error returned to the user. If err > 0
319 * it's just the icmp type << 8 | icmp code. After adjustment
320 * header points to the first 8 bytes of the tcp header. We need
321 * to find the appropriate port.
323 * The locking strategy used here is very "optimistic". When
324 * someone else accesses the socket the ICMP is just dropped
325 * and for some paths there is no check at all.
326 * A more general error queue to queue errors for later handling
327 * is probably better.
331 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
333 struct iphdr *iph = (struct iphdr *)icmp_skb->data;
334 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
335 struct inet_connection_sock *icsk;
337 struct inet_sock *inet;
338 const int type = icmp_hdr(icmp_skb)->type;
339 const int code = icmp_hdr(icmp_skb)->code;
345 struct net *net = dev_net(icmp_skb->dev);
347 if (icmp_skb->len < (iph->ihl << 2) + 8) {
348 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
352 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
353 iph->saddr, th->source, inet_iif(icmp_skb));
355 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
358 if (sk->sk_state == TCP_TIME_WAIT) {
359 inet_twsk_put(inet_twsk(sk));
364 /* If too many ICMPs get dropped on busy
365 * servers this needs to be solved differently.
367 if (sock_owned_by_user(sk))
368 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
370 if (sk->sk_state == TCP_CLOSE)
375 seq = ntohl(th->seq);
376 if (sk->sk_state != TCP_LISTEN &&
377 !between(seq, tp->snd_una, tp->snd_nxt)) {
378 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
383 case ICMP_SOURCE_QUENCH:
384 /* Just silently ignore these. */
386 case ICMP_PARAMETERPROB:
389 case ICMP_DEST_UNREACH:
390 if (code > NR_ICMP_UNREACH)
393 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
394 if (!sock_owned_by_user(sk))
395 do_pmtu_discovery(sk, iph, info);
399 err = icmp_err_convert[code].errno;
400 /* check if icmp_skb allows revert of backoff
401 * (see draft-zimmermann-tcp-lcd) */
402 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
404 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
408 icsk->icsk_backoff--;
409 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
413 skb = tcp_write_queue_head(sk);
416 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
417 tcp_time_stamp - TCP_SKB_CB(skb)->when);
420 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
421 remaining, TCP_RTO_MAX);
422 } else if (sock_owned_by_user(sk)) {
423 /* RTO revert clocked out retransmission,
424 * but socket is locked. Will defer. */
425 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
428 /* RTO revert clocked out retransmission.
429 * Will retransmit now */
430 tcp_retransmit_timer(sk);
434 case ICMP_TIME_EXCEEDED:
441 switch (sk->sk_state) {
442 struct request_sock *req, **prev;
444 if (sock_owned_by_user(sk))
447 req = inet_csk_search_req(sk, &prev, th->dest,
448 iph->daddr, iph->saddr);
452 /* ICMPs are not backlogged, hence we cannot get
453 an established socket here.
457 if (seq != tcp_rsk(req)->snt_isn) {
458 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
463 * Still in SYN_RECV, just remove it silently.
464 * There is no good way to pass the error to the newly
465 * created socket, and POSIX does not want network
466 * errors returned from accept().
468 inet_csk_reqsk_queue_drop(sk, req, prev);
472 case TCP_SYN_RECV: /* Cannot happen.
473 It can f.e. if SYNs crossed.
475 if (!sock_owned_by_user(sk)) {
478 sk->sk_error_report(sk);
482 sk->sk_err_soft = err;
487 /* If we've already connected we will keep trying
488 * until we time out, or the user gives up.
490 * rfc1122 4.2.3.9 allows to consider as hard errors
491 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
492 * but it is obsoleted by pmtu discovery).
494 * Note, that in modern internet, where routing is unreliable
495 * and in each dark corner broken firewalls sit, sending random
496 * errors ordered by their masters even this two messages finally lose
497 * their original sense (even Linux sends invalid PORT_UNREACHs)
499 * Now we are in compliance with RFCs.
504 if (!sock_owned_by_user(sk) && inet->recverr) {
506 sk->sk_error_report(sk);
507 } else { /* Only an error on timeout */
508 sk->sk_err_soft = err;
516 /* This routine computes an IPv4 TCP checksum. */
517 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
519 struct inet_sock *inet = inet_sk(sk);
520 struct tcphdr *th = tcp_hdr(skb);
522 if (skb->ip_summed == CHECKSUM_PARTIAL) {
523 th->check = ~tcp_v4_check(len, inet->inet_saddr,
524 inet->inet_daddr, 0);
525 skb->csum_start = skb_transport_header(skb) - skb->head;
526 skb->csum_offset = offsetof(struct tcphdr, check);
528 th->check = tcp_v4_check(len, inet->inet_saddr,
536 int tcp_v4_gso_send_check(struct sk_buff *skb)
538 const struct iphdr *iph;
541 if (!pskb_may_pull(skb, sizeof(*th)))
548 th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0);
549 skb->csum_start = skb_transport_header(skb) - skb->head;
550 skb->csum_offset = offsetof(struct tcphdr, check);
551 skb->ip_summed = CHECKSUM_PARTIAL;
556 * This routine will send an RST to the other tcp.
558 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
560 * Answer: if a packet caused RST, it is not for a socket
561 * existing in our system, if it is matched to a socket,
562 * it is just duplicate segment or bug in other side's TCP.
563 * So that we build reply only basing on parameters
564 * arrived with segment.
565 * Exception: precedence violation. We do not implement it in any case.
568 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
570 struct tcphdr *th = tcp_hdr(skb);
573 #ifdef CONFIG_TCP_MD5SIG
574 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
577 struct ip_reply_arg arg;
578 #ifdef CONFIG_TCP_MD5SIG
579 struct tcp_md5sig_key *key;
583 /* Never send a reset in response to a reset. */
587 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
590 /* Swap the send and the receive. */
591 memset(&rep, 0, sizeof(rep));
592 rep.th.dest = th->source;
593 rep.th.source = th->dest;
594 rep.th.doff = sizeof(struct tcphdr) / 4;
598 rep.th.seq = th->ack_seq;
601 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
602 skb->len - (th->doff << 2));
605 memset(&arg, 0, sizeof(arg));
606 arg.iov[0].iov_base = (unsigned char *)&rep;
607 arg.iov[0].iov_len = sizeof(rep.th);
609 #ifdef CONFIG_TCP_MD5SIG
610 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
612 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
614 (TCPOPT_MD5SIG << 8) |
616 /* Update length and the length the header thinks exists */
617 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
618 rep.th.doff = arg.iov[0].iov_len / 4;
620 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
621 key, ip_hdr(skb)->saddr,
622 ip_hdr(skb)->daddr, &rep.th);
625 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
626 ip_hdr(skb)->saddr, /* XXX */
627 arg.iov[0].iov_len, IPPROTO_TCP, 0);
628 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
629 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
631 net = dev_net(skb_dst(skb)->dev);
632 ip_send_reply(net->ipv4.tcp_sock, skb,
633 &arg, arg.iov[0].iov_len);
635 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
636 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
639 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
640 outside socket context is ugly, certainly. What can I do?
643 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
644 u32 win, u32 ts, int oif,
645 struct tcp_md5sig_key *key,
648 struct tcphdr *th = tcp_hdr(skb);
651 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
652 #ifdef CONFIG_TCP_MD5SIG
653 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
657 struct ip_reply_arg arg;
658 struct net *net = dev_net(skb_dst(skb)->dev);
660 memset(&rep.th, 0, sizeof(struct tcphdr));
661 memset(&arg, 0, sizeof(arg));
663 arg.iov[0].iov_base = (unsigned char *)&rep;
664 arg.iov[0].iov_len = sizeof(rep.th);
666 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
667 (TCPOPT_TIMESTAMP << 8) |
669 rep.opt[1] = htonl(tcp_time_stamp);
670 rep.opt[2] = htonl(ts);
671 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
674 /* Swap the send and the receive. */
675 rep.th.dest = th->source;
676 rep.th.source = th->dest;
677 rep.th.doff = arg.iov[0].iov_len / 4;
678 rep.th.seq = htonl(seq);
679 rep.th.ack_seq = htonl(ack);
681 rep.th.window = htons(win);
683 #ifdef CONFIG_TCP_MD5SIG
685 int offset = (ts) ? 3 : 0;
687 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
689 (TCPOPT_MD5SIG << 8) |
691 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
692 rep.th.doff = arg.iov[0].iov_len/4;
694 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
695 key, ip_hdr(skb)->saddr,
696 ip_hdr(skb)->daddr, &rep.th);
699 arg.flags = reply_flags;
700 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
701 ip_hdr(skb)->saddr, /* XXX */
702 arg.iov[0].iov_len, IPPROTO_TCP, 0);
703 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
705 arg.bound_dev_if = oif;
707 ip_send_reply(net->ipv4.tcp_sock, skb,
708 &arg, arg.iov[0].iov_len);
710 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
713 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
715 struct inet_timewait_sock *tw = inet_twsk(sk);
716 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
718 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
719 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
722 tcp_twsk_md5_key(tcptw),
723 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
729 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
730 struct request_sock *req)
732 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
733 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
736 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
737 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
741 * Send a SYN-ACK after having received a SYN.
742 * This still operates on a request_sock only, not on a big
745 static int __tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
746 struct dst_entry *dst)
748 const struct inet_request_sock *ireq = inet_rsk(req);
750 struct sk_buff * skb;
752 /* First, grab a route. */
753 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
756 skb = tcp_make_synack(sk, dst, req);
759 struct tcphdr *th = tcp_hdr(skb);
761 th->check = tcp_v4_check(skb->len,
764 csum_partial(th, skb->len,
767 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
770 err = net_xmit_eval(err);
777 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req)
779 return __tcp_v4_send_synack(sk, req, NULL);
783 * IPv4 request_sock destructor.
785 static void tcp_v4_reqsk_destructor(struct request_sock *req)
787 kfree(inet_rsk(req)->opt);
790 #ifdef CONFIG_SYN_COOKIES
791 static void syn_flood_warning(struct sk_buff *skb)
793 static unsigned long warntime;
795 if (time_after(jiffies, (warntime + HZ * 60))) {
798 "possible SYN flooding on port %d. Sending cookies.\n",
799 ntohs(tcp_hdr(skb)->dest));
805 * Save and compile IPv4 options into the request_sock if needed.
807 static struct ip_options *tcp_v4_save_options(struct sock *sk,
810 struct ip_options *opt = &(IPCB(skb)->opt);
811 struct ip_options *dopt = NULL;
813 if (opt && opt->optlen) {
814 int opt_size = optlength(opt);
815 dopt = kmalloc(opt_size, GFP_ATOMIC);
817 if (ip_options_echo(dopt, skb)) {
826 #ifdef CONFIG_TCP_MD5SIG
828 * RFC2385 MD5 checksumming requires a mapping of
829 * IP address->MD5 Key.
830 * We need to maintain these in the sk structure.
833 /* Find the Key structure for an address. */
834 static struct tcp_md5sig_key *
835 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
837 struct tcp_sock *tp = tcp_sk(sk);
840 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
842 for (i = 0; i < tp->md5sig_info->entries4; i++) {
843 if (tp->md5sig_info->keys4[i].addr == addr)
844 return &tp->md5sig_info->keys4[i].base;
849 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
850 struct sock *addr_sk)
852 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
855 EXPORT_SYMBOL(tcp_v4_md5_lookup);
857 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
858 struct request_sock *req)
860 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
863 /* This can be called on a newly created socket, from other files */
864 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
865 u8 *newkey, u8 newkeylen)
867 /* Add Key to the list */
868 struct tcp_md5sig_key *key;
869 struct tcp_sock *tp = tcp_sk(sk);
870 struct tcp4_md5sig_key *keys;
872 key = tcp_v4_md5_do_lookup(sk, addr);
874 /* Pre-existing entry - just update that one. */
877 key->keylen = newkeylen;
879 struct tcp_md5sig_info *md5sig;
881 if (!tp->md5sig_info) {
882 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
884 if (!tp->md5sig_info) {
888 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
890 if (tcp_alloc_md5sig_pool(sk) == NULL) {
894 md5sig = tp->md5sig_info;
896 if (md5sig->alloced4 == md5sig->entries4) {
897 keys = kmalloc((sizeof(*keys) *
898 (md5sig->entries4 + 1)), GFP_ATOMIC);
901 tcp_free_md5sig_pool();
905 if (md5sig->entries4)
906 memcpy(keys, md5sig->keys4,
907 sizeof(*keys) * md5sig->entries4);
909 /* Free old key list, and reference new one */
910 kfree(md5sig->keys4);
911 md5sig->keys4 = keys;
915 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
916 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
917 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
922 EXPORT_SYMBOL(tcp_v4_md5_do_add);
924 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
925 u8 *newkey, u8 newkeylen)
927 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
931 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
933 struct tcp_sock *tp = tcp_sk(sk);
936 for (i = 0; i < tp->md5sig_info->entries4; i++) {
937 if (tp->md5sig_info->keys4[i].addr == addr) {
939 kfree(tp->md5sig_info->keys4[i].base.key);
940 tp->md5sig_info->entries4--;
942 if (tp->md5sig_info->entries4 == 0) {
943 kfree(tp->md5sig_info->keys4);
944 tp->md5sig_info->keys4 = NULL;
945 tp->md5sig_info->alloced4 = 0;
946 } else if (tp->md5sig_info->entries4 != i) {
947 /* Need to do some manipulation */
948 memmove(&tp->md5sig_info->keys4[i],
949 &tp->md5sig_info->keys4[i+1],
950 (tp->md5sig_info->entries4 - i) *
951 sizeof(struct tcp4_md5sig_key));
953 tcp_free_md5sig_pool();
960 EXPORT_SYMBOL(tcp_v4_md5_do_del);
962 static void tcp_v4_clear_md5_list(struct sock *sk)
964 struct tcp_sock *tp = tcp_sk(sk);
966 /* Free each key, then the set of key keys,
967 * the crypto element, and then decrement our
968 * hold on the last resort crypto.
970 if (tp->md5sig_info->entries4) {
972 for (i = 0; i < tp->md5sig_info->entries4; i++)
973 kfree(tp->md5sig_info->keys4[i].base.key);
974 tp->md5sig_info->entries4 = 0;
975 tcp_free_md5sig_pool();
977 if (tp->md5sig_info->keys4) {
978 kfree(tp->md5sig_info->keys4);
979 tp->md5sig_info->keys4 = NULL;
980 tp->md5sig_info->alloced4 = 0;
984 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
987 struct tcp_md5sig cmd;
988 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
991 if (optlen < sizeof(cmd))
994 if (copy_from_user(&cmd, optval, sizeof(cmd)))
997 if (sin->sin_family != AF_INET)
1000 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1001 if (!tcp_sk(sk)->md5sig_info)
1003 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1006 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1009 if (!tcp_sk(sk)->md5sig_info) {
1010 struct tcp_sock *tp = tcp_sk(sk);
1011 struct tcp_md5sig_info *p;
1013 p = kzalloc(sizeof(*p), sk->sk_allocation);
1017 tp->md5sig_info = p;
1018 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1021 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1024 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1025 newkey, cmd.tcpm_keylen);
1028 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1029 __be32 daddr, __be32 saddr, int nbytes)
1031 struct tcp4_pseudohdr *bp;
1032 struct scatterlist sg;
1034 bp = &hp->md5_blk.ip4;
1037 * 1. the TCP pseudo-header (in the order: source IP address,
1038 * destination IP address, zero-padded protocol number, and
1044 bp->protocol = IPPROTO_TCP;
1045 bp->len = cpu_to_be16(nbytes);
1047 sg_init_one(&sg, bp, sizeof(*bp));
1048 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1051 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1052 __be32 daddr, __be32 saddr, struct tcphdr *th)
1054 struct tcp_md5sig_pool *hp;
1055 struct hash_desc *desc;
1057 hp = tcp_get_md5sig_pool();
1059 goto clear_hash_noput;
1060 desc = &hp->md5_desc;
1062 if (crypto_hash_init(desc))
1064 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1066 if (tcp_md5_hash_header(hp, th))
1068 if (tcp_md5_hash_key(hp, key))
1070 if (crypto_hash_final(desc, md5_hash))
1073 tcp_put_md5sig_pool();
1077 tcp_put_md5sig_pool();
1079 memset(md5_hash, 0, 16);
1083 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1084 struct sock *sk, struct request_sock *req,
1085 struct sk_buff *skb)
1087 struct tcp_md5sig_pool *hp;
1088 struct hash_desc *desc;
1089 struct tcphdr *th = tcp_hdr(skb);
1090 __be32 saddr, daddr;
1093 saddr = inet_sk(sk)->inet_saddr;
1094 daddr = inet_sk(sk)->inet_daddr;
1096 saddr = inet_rsk(req)->loc_addr;
1097 daddr = inet_rsk(req)->rmt_addr;
1099 const struct iphdr *iph = ip_hdr(skb);
1104 hp = tcp_get_md5sig_pool();
1106 goto clear_hash_noput;
1107 desc = &hp->md5_desc;
1109 if (crypto_hash_init(desc))
1112 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1114 if (tcp_md5_hash_header(hp, th))
1116 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1118 if (tcp_md5_hash_key(hp, key))
1120 if (crypto_hash_final(desc, md5_hash))
1123 tcp_put_md5sig_pool();
1127 tcp_put_md5sig_pool();
1129 memset(md5_hash, 0, 16);
1133 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1135 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1138 * This gets called for each TCP segment that arrives
1139 * so we want to be efficient.
1140 * We have 3 drop cases:
1141 * o No MD5 hash and one expected.
1142 * o MD5 hash and we're not expecting one.
1143 * o MD5 hash and its wrong.
1145 __u8 *hash_location = NULL;
1146 struct tcp_md5sig_key *hash_expected;
1147 const struct iphdr *iph = ip_hdr(skb);
1148 struct tcphdr *th = tcp_hdr(skb);
1150 unsigned char newhash[16];
1152 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1153 hash_location = tcp_parse_md5sig_option(th);
1155 /* We've parsed the options - do we have a hash? */
1156 if (!hash_expected && !hash_location)
1159 if (hash_expected && !hash_location) {
1160 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1164 if (!hash_expected && hash_location) {
1165 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1169 /* Okay, so this is hash_expected and hash_location -
1170 * so we need to calculate the checksum.
1172 genhash = tcp_v4_md5_hash_skb(newhash,
1176 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1177 if (net_ratelimit()) {
1178 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1179 &iph->saddr, ntohs(th->source),
1180 &iph->daddr, ntohs(th->dest),
1181 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1190 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1192 .obj_size = sizeof(struct tcp_request_sock),
1193 .rtx_syn_ack = tcp_v4_send_synack,
1194 .send_ack = tcp_v4_reqsk_send_ack,
1195 .destructor = tcp_v4_reqsk_destructor,
1196 .send_reset = tcp_v4_send_reset,
1199 #ifdef CONFIG_TCP_MD5SIG
1200 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1201 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1202 .calc_md5_hash = tcp_v4_md5_hash_skb,
1206 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1207 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1208 .twsk_unique = tcp_twsk_unique,
1209 .twsk_destructor= tcp_twsk_destructor,
1212 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1214 struct inet_request_sock *ireq;
1215 struct tcp_options_received tmp_opt;
1216 struct request_sock *req;
1217 __be32 saddr = ip_hdr(skb)->saddr;
1218 __be32 daddr = ip_hdr(skb)->daddr;
1219 __u32 isn = TCP_SKB_CB(skb)->when;
1220 struct dst_entry *dst = NULL;
1221 #ifdef CONFIG_SYN_COOKIES
1222 int want_cookie = 0;
1224 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1227 /* Never answer to SYNs send to broadcast or multicast */
1228 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1231 /* TW buckets are converted to open requests without
1232 * limitations, they conserve resources and peer is
1233 * evidently real one.
1235 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1236 #ifdef CONFIG_SYN_COOKIES
1237 if (sysctl_tcp_syncookies) {
1244 /* Accept backlog is full. If we have already queued enough
1245 * of warm entries in syn queue, drop request. It is better than
1246 * clogging syn queue with openreqs with exponentially increasing
1249 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1252 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1256 #ifdef CONFIG_TCP_MD5SIG
1257 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1260 ireq = inet_rsk(req);
1261 ireq->loc_addr = daddr;
1262 ireq->rmt_addr = saddr;
1263 ireq->no_srccheck = inet_sk(sk)->transparent;
1264 ireq->opt = tcp_v4_save_options(sk, skb);
1266 dst = inet_csk_route_req(sk, req);
1270 tcp_clear_options(&tmp_opt);
1271 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1272 tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss;
1274 tcp_parse_options(skb, &tmp_opt, 0, dst);
1276 if (want_cookie && !tmp_opt.saw_tstamp)
1277 tcp_clear_options(&tmp_opt);
1279 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1281 tcp_openreq_init(req, &tmp_opt, skb);
1283 if (security_inet_conn_request(sk, skb, req))
1284 goto drop_and_release;
1287 TCP_ECN_create_request(req, tcp_hdr(skb));
1290 #ifdef CONFIG_SYN_COOKIES
1291 syn_flood_warning(skb);
1292 req->cookie_ts = tmp_opt.tstamp_ok;
1294 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1296 struct inet_peer *peer = NULL;
1298 /* VJ's idea. We save last timestamp seen
1299 * from the destination in peer table, when entering
1300 * state TIME-WAIT, and check against it before
1301 * accepting new connection request.
1303 * If "isn" is not zero, this request hit alive
1304 * timewait bucket, so that all the necessary checks
1305 * are made in the function processing timewait state.
1307 if (tmp_opt.saw_tstamp &&
1308 tcp_death_row.sysctl_tw_recycle &&
1309 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1310 peer->v4daddr == saddr) {
1311 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1312 (s32)(peer->tcp_ts - req->ts_recent) >
1314 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1315 goto drop_and_release;
1318 /* Kill the following clause, if you dislike this way. */
1319 else if (!sysctl_tcp_syncookies &&
1320 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1321 (sysctl_max_syn_backlog >> 2)) &&
1322 (!peer || !peer->tcp_ts_stamp) &&
1323 (!dst || !dst_metric(dst, RTAX_RTT))) {
1324 /* Without syncookies last quarter of
1325 * backlog is filled with destinations,
1326 * proven to be alive.
1327 * It means that we continue to communicate
1328 * to destinations, already remembered
1329 * to the moment of synflood.
1331 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1332 &saddr, ntohs(tcp_hdr(skb)->source));
1333 goto drop_and_release;
1336 isn = tcp_v4_init_sequence(skb);
1338 tcp_rsk(req)->snt_isn = isn;
1340 if (__tcp_v4_send_synack(sk, req, dst) || want_cookie)
1343 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1356 * The three way handshake has completed - we got a valid synack -
1357 * now create the new socket.
1359 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1360 struct request_sock *req,
1361 struct dst_entry *dst)
1363 struct inet_request_sock *ireq;
1364 struct inet_sock *newinet;
1365 struct tcp_sock *newtp;
1367 #ifdef CONFIG_TCP_MD5SIG
1368 struct tcp_md5sig_key *key;
1371 if (sk_acceptq_is_full(sk))
1374 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1377 newsk = tcp_create_openreq_child(sk, req, skb);
1381 newsk->sk_gso_type = SKB_GSO_TCPV4;
1382 sk_setup_caps(newsk, dst);
1384 newtp = tcp_sk(newsk);
1385 newinet = inet_sk(newsk);
1386 ireq = inet_rsk(req);
1387 newinet->inet_daddr = ireq->rmt_addr;
1388 newinet->inet_rcv_saddr = ireq->loc_addr;
1389 newinet->inet_saddr = ireq->loc_addr;
1390 newinet->opt = ireq->opt;
1392 newinet->mc_index = inet_iif(skb);
1393 newinet->mc_ttl = ip_hdr(skb)->ttl;
1394 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1396 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
1397 newinet->inet_id = newtp->write_seq ^ jiffies;
1399 tcp_mtup_init(newsk);
1400 tcp_sync_mss(newsk, dst_mtu(dst));
1401 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1402 if (tcp_sk(sk)->rx_opt.user_mss &&
1403 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1404 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1406 tcp_initialize_rcv_mss(newsk);
1408 #ifdef CONFIG_TCP_MD5SIG
1409 /* Copy over the MD5 key from the original socket */
1410 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1413 * We're using one, so create a matching key
1414 * on the newsk structure. If we fail to get
1415 * memory, then we end up not copying the key
1418 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1420 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1421 newkey, key->keylen);
1422 newsk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1426 __inet_hash_nolisten(newsk);
1427 __inet_inherit_port(sk, newsk);
1432 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1434 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1439 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1441 struct tcphdr *th = tcp_hdr(skb);
1442 const struct iphdr *iph = ip_hdr(skb);
1444 struct request_sock **prev;
1445 /* Find possible connection requests. */
1446 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1447 iph->saddr, iph->daddr);
1449 return tcp_check_req(sk, skb, req, prev);
1451 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1452 th->source, iph->daddr, th->dest, inet_iif(skb));
1455 if (nsk->sk_state != TCP_TIME_WAIT) {
1459 inet_twsk_put(inet_twsk(nsk));
1463 #ifdef CONFIG_SYN_COOKIES
1464 if (!th->rst && !th->syn && th->ack)
1465 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1470 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1472 const struct iphdr *iph = ip_hdr(skb);
1474 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1475 if (!tcp_v4_check(skb->len, iph->saddr,
1476 iph->daddr, skb->csum)) {
1477 skb->ip_summed = CHECKSUM_UNNECESSARY;
1482 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1483 skb->len, IPPROTO_TCP, 0);
1485 if (skb->len <= 76) {
1486 return __skb_checksum_complete(skb);
1492 /* The socket must have it's spinlock held when we get
1495 * We have a potential double-lock case here, so even when
1496 * doing backlog processing we use the BH locking scheme.
1497 * This is because we cannot sleep with the original spinlock
1500 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1503 #ifdef CONFIG_TCP_MD5SIG
1505 * We really want to reject the packet as early as possible
1507 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1508 * o There is an MD5 option and we're not expecting one
1510 if (tcp_v4_inbound_md5_hash(sk, skb))
1514 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1515 TCP_CHECK_TIMER(sk);
1516 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1520 TCP_CHECK_TIMER(sk);
1524 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1527 if (sk->sk_state == TCP_LISTEN) {
1528 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1533 if (tcp_child_process(sk, nsk, skb)) {
1541 TCP_CHECK_TIMER(sk);
1542 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1546 TCP_CHECK_TIMER(sk);
1550 tcp_v4_send_reset(rsk, skb);
1553 /* Be careful here. If this function gets more complicated and
1554 * gcc suffers from register pressure on the x86, sk (in %ebx)
1555 * might be destroyed here. This current version compiles correctly,
1556 * but you have been warned.
1561 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1569 int tcp_v4_rcv(struct sk_buff *skb)
1571 const struct iphdr *iph;
1575 struct net *net = dev_net(skb->dev);
1577 if (skb->pkt_type != PACKET_HOST)
1580 /* Count it even if it's bad */
1581 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1583 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1588 if (th->doff < sizeof(struct tcphdr) / 4)
1590 if (!pskb_may_pull(skb, th->doff * 4))
1593 /* An explanation is required here, I think.
1594 * Packet length and doff are validated by header prediction,
1595 * provided case of th->doff==0 is eliminated.
1596 * So, we defer the checks. */
1597 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1602 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1603 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1604 skb->len - th->doff * 4);
1605 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1606 TCP_SKB_CB(skb)->when = 0;
1607 TCP_SKB_CB(skb)->flags = iph->tos;
1608 TCP_SKB_CB(skb)->sacked = 0;
1610 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1615 if (sk->sk_state == TCP_TIME_WAIT)
1618 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1619 goto discard_and_relse;
1622 if (sk_filter(sk, skb))
1623 goto discard_and_relse;
1627 bh_lock_sock_nested(sk);
1629 if (!sock_owned_by_user(sk)) {
1630 #ifdef CONFIG_NET_DMA
1631 struct tcp_sock *tp = tcp_sk(sk);
1632 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1633 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1634 if (tp->ucopy.dma_chan)
1635 ret = tcp_v4_do_rcv(sk, skb);
1639 if (!tcp_prequeue(sk, skb))
1640 ret = tcp_v4_do_rcv(sk, skb);
1643 sk_add_backlog(sk, skb);
1651 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1654 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1656 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1658 tcp_v4_send_reset(NULL, skb);
1662 /* Discard frame. */
1671 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1672 inet_twsk_put(inet_twsk(sk));
1676 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1677 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1678 inet_twsk_put(inet_twsk(sk));
1681 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1683 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1685 iph->daddr, th->dest,
1688 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1689 inet_twsk_put(inet_twsk(sk));
1693 /* Fall through to ACK */
1696 tcp_v4_timewait_ack(sk, skb);
1700 case TCP_TW_SUCCESS:;
1705 /* VJ's idea. Save last timestamp seen from this destination
1706 * and hold it at least for normal timewait interval to use for duplicate
1707 * segment detection in subsequent connections, before they enter synchronized
1711 int tcp_v4_remember_stamp(struct sock *sk)
1713 struct inet_sock *inet = inet_sk(sk);
1714 struct tcp_sock *tp = tcp_sk(sk);
1715 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1716 struct inet_peer *peer = NULL;
1719 if (!rt || rt->rt_dst != inet->inet_daddr) {
1720 peer = inet_getpeer(inet->inet_daddr, 1);
1724 rt_bind_peer(rt, 1);
1729 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1730 ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1731 peer->tcp_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
1732 peer->tcp_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
1733 peer->tcp_ts = tp->rx_opt.ts_recent;
1743 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1745 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1748 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1750 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1751 ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1752 peer->tcp_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
1753 peer->tcp_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
1754 peer->tcp_ts = tcptw->tw_ts_recent;
1763 const struct inet_connection_sock_af_ops ipv4_specific = {
1764 .queue_xmit = ip_queue_xmit,
1765 .send_check = tcp_v4_send_check,
1766 .rebuild_header = inet_sk_rebuild_header,
1767 .conn_request = tcp_v4_conn_request,
1768 .syn_recv_sock = tcp_v4_syn_recv_sock,
1769 .remember_stamp = tcp_v4_remember_stamp,
1770 .net_header_len = sizeof(struct iphdr),
1771 .setsockopt = ip_setsockopt,
1772 .getsockopt = ip_getsockopt,
1773 .addr2sockaddr = inet_csk_addr2sockaddr,
1774 .sockaddr_len = sizeof(struct sockaddr_in),
1775 .bind_conflict = inet_csk_bind_conflict,
1776 #ifdef CONFIG_COMPAT
1777 .compat_setsockopt = compat_ip_setsockopt,
1778 .compat_getsockopt = compat_ip_getsockopt,
1782 #ifdef CONFIG_TCP_MD5SIG
1783 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1784 .md5_lookup = tcp_v4_md5_lookup,
1785 .calc_md5_hash = tcp_v4_md5_hash_skb,
1786 .md5_add = tcp_v4_md5_add_func,
1787 .md5_parse = tcp_v4_parse_md5_keys,
1791 /* NOTE: A lot of things set to zero explicitly by call to
1792 * sk_alloc() so need not be done here.
1794 static int tcp_v4_init_sock(struct sock *sk)
1796 struct inet_connection_sock *icsk = inet_csk(sk);
1797 struct tcp_sock *tp = tcp_sk(sk);
1799 skb_queue_head_init(&tp->out_of_order_queue);
1800 tcp_init_xmit_timers(sk);
1801 tcp_prequeue_init(tp);
1803 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1804 tp->mdev = TCP_TIMEOUT_INIT;
1806 /* So many TCP implementations out there (incorrectly) count the
1807 * initial SYN frame in their delayed-ACK and congestion control
1808 * algorithms that we must have the following bandaid to talk
1809 * efficiently to them. -DaveM
1813 /* See draft-stevens-tcpca-spec-01 for discussion of the
1814 * initialization of these values.
1816 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1817 tp->snd_cwnd_clamp = ~0;
1818 tp->mss_cache = TCP_MSS_DEFAULT;
1820 tp->reordering = sysctl_tcp_reordering;
1821 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1823 sk->sk_state = TCP_CLOSE;
1825 sk->sk_write_space = sk_stream_write_space;
1826 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1828 icsk->icsk_af_ops = &ipv4_specific;
1829 icsk->icsk_sync_mss = tcp_sync_mss;
1830 #ifdef CONFIG_TCP_MD5SIG
1831 tp->af_specific = &tcp_sock_ipv4_specific;
1834 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1835 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1838 percpu_counter_inc(&tcp_sockets_allocated);
1844 void tcp_v4_destroy_sock(struct sock *sk)
1846 struct tcp_sock *tp = tcp_sk(sk);
1848 tcp_clear_xmit_timers(sk);
1850 tcp_cleanup_congestion_control(sk);
1852 /* Cleanup up the write buffer. */
1853 tcp_write_queue_purge(sk);
1855 /* Cleans up our, hopefully empty, out_of_order_queue. */
1856 __skb_queue_purge(&tp->out_of_order_queue);
1858 #ifdef CONFIG_TCP_MD5SIG
1859 /* Clean up the MD5 key list, if any */
1860 if (tp->md5sig_info) {
1861 tcp_v4_clear_md5_list(sk);
1862 kfree(tp->md5sig_info);
1863 tp->md5sig_info = NULL;
1867 #ifdef CONFIG_NET_DMA
1868 /* Cleans up our sk_async_wait_queue */
1869 __skb_queue_purge(&sk->sk_async_wait_queue);
1872 /* Clean prequeue, it must be empty really */
1873 __skb_queue_purge(&tp->ucopy.prequeue);
1875 /* Clean up a referenced TCP bind bucket. */
1876 if (inet_csk(sk)->icsk_bind_hash)
1880 * If sendmsg cached page exists, toss it.
1882 if (sk->sk_sndmsg_page) {
1883 __free_page(sk->sk_sndmsg_page);
1884 sk->sk_sndmsg_page = NULL;
1887 percpu_counter_dec(&tcp_sockets_allocated);
1890 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1892 #ifdef CONFIG_PROC_FS
1893 /* Proc filesystem TCP sock list dumping. */
1895 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1897 return hlist_nulls_empty(head) ? NULL :
1898 list_entry(head->first, struct inet_timewait_sock, tw_node);
1901 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1903 return !is_a_nulls(tw->tw_node.next) ?
1904 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1907 static void *listening_get_next(struct seq_file *seq, void *cur)
1909 struct inet_connection_sock *icsk;
1910 struct hlist_nulls_node *node;
1911 struct sock *sk = cur;
1912 struct inet_listen_hashbucket *ilb;
1913 struct tcp_iter_state *st = seq->private;
1914 struct net *net = seq_file_net(seq);
1918 ilb = &tcp_hashinfo.listening_hash[0];
1919 spin_lock_bh(&ilb->lock);
1920 sk = sk_nulls_head(&ilb->head);
1923 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1926 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1927 struct request_sock *req = cur;
1929 icsk = inet_csk(st->syn_wait_sk);
1933 if (req->rsk_ops->family == st->family) {
1939 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1942 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1944 sk = sk_next(st->syn_wait_sk);
1945 st->state = TCP_SEQ_STATE_LISTENING;
1946 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1948 icsk = inet_csk(sk);
1949 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1950 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1952 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1956 sk_nulls_for_each_from(sk, node) {
1957 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) {
1961 icsk = inet_csk(sk);
1962 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1963 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1965 st->uid = sock_i_uid(sk);
1966 st->syn_wait_sk = sk;
1967 st->state = TCP_SEQ_STATE_OPENREQ;
1971 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1973 spin_unlock_bh(&ilb->lock);
1974 if (++st->bucket < INET_LHTABLE_SIZE) {
1975 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1976 spin_lock_bh(&ilb->lock);
1977 sk = sk_nulls_head(&ilb->head);
1985 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1987 void *rc = listening_get_next(seq, NULL);
1989 while (rc && *pos) {
1990 rc = listening_get_next(seq, rc);
1996 static inline int empty_bucket(struct tcp_iter_state *st)
1998 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
1999 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2002 static void *established_get_first(struct seq_file *seq)
2004 struct tcp_iter_state *st = seq->private;
2005 struct net *net = seq_file_net(seq);
2008 for (st->bucket = 0; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2010 struct hlist_nulls_node *node;
2011 struct inet_timewait_sock *tw;
2012 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2014 /* Lockless fast path for the common case of empty buckets */
2015 if (empty_bucket(st))
2019 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2020 if (sk->sk_family != st->family ||
2021 !net_eq(sock_net(sk), net)) {
2027 st->state = TCP_SEQ_STATE_TIME_WAIT;
2028 inet_twsk_for_each(tw, node,
2029 &tcp_hashinfo.ehash[st->bucket].twchain) {
2030 if (tw->tw_family != st->family ||
2031 !net_eq(twsk_net(tw), net)) {
2037 spin_unlock_bh(lock);
2038 st->state = TCP_SEQ_STATE_ESTABLISHED;
2044 static void *established_get_next(struct seq_file *seq, void *cur)
2046 struct sock *sk = cur;
2047 struct inet_timewait_sock *tw;
2048 struct hlist_nulls_node *node;
2049 struct tcp_iter_state *st = seq->private;
2050 struct net *net = seq_file_net(seq);
2054 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2058 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2065 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2066 st->state = TCP_SEQ_STATE_ESTABLISHED;
2068 /* Look for next non empty bucket */
2069 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2072 if (st->bucket > tcp_hashinfo.ehash_mask)
2075 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2076 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2078 sk = sk_nulls_next(sk);
2080 sk_nulls_for_each_from(sk, node) {
2081 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2085 st->state = TCP_SEQ_STATE_TIME_WAIT;
2086 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2094 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2096 void *rc = established_get_first(seq);
2099 rc = established_get_next(seq, rc);
2105 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2108 struct tcp_iter_state *st = seq->private;
2110 st->state = TCP_SEQ_STATE_LISTENING;
2111 rc = listening_get_idx(seq, &pos);
2114 st->state = TCP_SEQ_STATE_ESTABLISHED;
2115 rc = established_get_idx(seq, pos);
2121 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2123 struct tcp_iter_state *st = seq->private;
2124 st->state = TCP_SEQ_STATE_LISTENING;
2126 return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2129 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2132 struct tcp_iter_state *st;
2134 if (v == SEQ_START_TOKEN) {
2135 rc = tcp_get_idx(seq, 0);
2140 switch (st->state) {
2141 case TCP_SEQ_STATE_OPENREQ:
2142 case TCP_SEQ_STATE_LISTENING:
2143 rc = listening_get_next(seq, v);
2145 st->state = TCP_SEQ_STATE_ESTABLISHED;
2146 rc = established_get_first(seq);
2149 case TCP_SEQ_STATE_ESTABLISHED:
2150 case TCP_SEQ_STATE_TIME_WAIT:
2151 rc = established_get_next(seq, v);
2159 static void tcp_seq_stop(struct seq_file *seq, void *v)
2161 struct tcp_iter_state *st = seq->private;
2163 switch (st->state) {
2164 case TCP_SEQ_STATE_OPENREQ:
2166 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2167 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2169 case TCP_SEQ_STATE_LISTENING:
2170 if (v != SEQ_START_TOKEN)
2171 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2173 case TCP_SEQ_STATE_TIME_WAIT:
2174 case TCP_SEQ_STATE_ESTABLISHED:
2176 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2181 static int tcp_seq_open(struct inode *inode, struct file *file)
2183 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2184 struct tcp_iter_state *s;
2187 err = seq_open_net(inode, file, &afinfo->seq_ops,
2188 sizeof(struct tcp_iter_state));
2192 s = ((struct seq_file *)file->private_data)->private;
2193 s->family = afinfo->family;
2197 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2200 struct proc_dir_entry *p;
2202 afinfo->seq_fops.open = tcp_seq_open;
2203 afinfo->seq_fops.read = seq_read;
2204 afinfo->seq_fops.llseek = seq_lseek;
2205 afinfo->seq_fops.release = seq_release_net;
2207 afinfo->seq_ops.start = tcp_seq_start;
2208 afinfo->seq_ops.next = tcp_seq_next;
2209 afinfo->seq_ops.stop = tcp_seq_stop;
2211 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2212 &afinfo->seq_fops, afinfo);
2218 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2220 proc_net_remove(net, afinfo->name);
2223 static void get_openreq4(struct sock *sk, struct request_sock *req,
2224 struct seq_file *f, int i, int uid, int *len)
2226 const struct inet_request_sock *ireq = inet_rsk(req);
2227 int ttd = req->expires - jiffies;
2229 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2230 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
2233 ntohs(inet_sk(sk)->inet_sport),
2235 ntohs(ireq->rmt_port),
2237 0, 0, /* could print option size, but that is af dependent. */
2238 1, /* timers active (only the expire timer) */
2239 jiffies_to_clock_t(ttd),
2242 0, /* non standard timer */
2243 0, /* open_requests have no inode */
2244 atomic_read(&sk->sk_refcnt),
2249 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2252 unsigned long timer_expires;
2253 struct tcp_sock *tp = tcp_sk(sk);
2254 const struct inet_connection_sock *icsk = inet_csk(sk);
2255 struct inet_sock *inet = inet_sk(sk);
2256 __be32 dest = inet->inet_daddr;
2257 __be32 src = inet->inet_rcv_saddr;
2258 __u16 destp = ntohs(inet->inet_dport);
2259 __u16 srcp = ntohs(inet->inet_sport);
2261 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2263 timer_expires = icsk->icsk_timeout;
2264 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2266 timer_expires = icsk->icsk_timeout;
2267 } else if (timer_pending(&sk->sk_timer)) {
2269 timer_expires = sk->sk_timer.expires;
2272 timer_expires = jiffies;
2275 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2276 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
2277 i, src, srcp, dest, destp, sk->sk_state,
2278 tp->write_seq - tp->snd_una,
2279 sk->sk_state == TCP_LISTEN ? sk->sk_ack_backlog :
2280 (tp->rcv_nxt - tp->copied_seq),
2282 jiffies_to_clock_t(timer_expires - jiffies),
2283 icsk->icsk_retransmits,
2285 icsk->icsk_probes_out,
2287 atomic_read(&sk->sk_refcnt), sk,
2288 jiffies_to_clock_t(icsk->icsk_rto),
2289 jiffies_to_clock_t(icsk->icsk_ack.ato),
2290 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2292 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2296 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2297 struct seq_file *f, int i, int *len)
2301 int ttd = tw->tw_ttd - jiffies;
2306 dest = tw->tw_daddr;
2307 src = tw->tw_rcv_saddr;
2308 destp = ntohs(tw->tw_dport);
2309 srcp = ntohs(tw->tw_sport);
2311 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2312 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
2313 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2314 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2315 atomic_read(&tw->tw_refcnt), tw, len);
2320 static int tcp4_seq_show(struct seq_file *seq, void *v)
2322 struct tcp_iter_state *st;
2325 if (v == SEQ_START_TOKEN) {
2326 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2327 " sl local_address rem_address st tx_queue "
2328 "rx_queue tr tm->when retrnsmt uid timeout "
2334 switch (st->state) {
2335 case TCP_SEQ_STATE_LISTENING:
2336 case TCP_SEQ_STATE_ESTABLISHED:
2337 get_tcp4_sock(v, seq, st->num, &len);
2339 case TCP_SEQ_STATE_OPENREQ:
2340 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2342 case TCP_SEQ_STATE_TIME_WAIT:
2343 get_timewait4_sock(v, seq, st->num, &len);
2346 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2351 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2355 .owner = THIS_MODULE,
2358 .show = tcp4_seq_show,
2362 static int tcp4_proc_init_net(struct net *net)
2364 return tcp_proc_register(net, &tcp4_seq_afinfo);
2367 static void tcp4_proc_exit_net(struct net *net)
2369 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2372 static struct pernet_operations tcp4_net_ops = {
2373 .init = tcp4_proc_init_net,
2374 .exit = tcp4_proc_exit_net,
2377 int __init tcp4_proc_init(void)
2379 return register_pernet_subsys(&tcp4_net_ops);
2382 void tcp4_proc_exit(void)
2384 unregister_pernet_subsys(&tcp4_net_ops);
2386 #endif /* CONFIG_PROC_FS */
2388 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2390 struct iphdr *iph = skb_gro_network_header(skb);
2392 switch (skb->ip_summed) {
2393 case CHECKSUM_COMPLETE:
2394 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2396 skb->ip_summed = CHECKSUM_UNNECESSARY;
2402 NAPI_GRO_CB(skb)->flush = 1;
2406 return tcp_gro_receive(head, skb);
2408 EXPORT_SYMBOL(tcp4_gro_receive);
2410 int tcp4_gro_complete(struct sk_buff *skb)
2412 struct iphdr *iph = ip_hdr(skb);
2413 struct tcphdr *th = tcp_hdr(skb);
2415 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2416 iph->saddr, iph->daddr, 0);
2417 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2419 return tcp_gro_complete(skb);
2421 EXPORT_SYMBOL(tcp4_gro_complete);
2423 struct proto tcp_prot = {
2425 .owner = THIS_MODULE,
2427 .connect = tcp_v4_connect,
2428 .disconnect = tcp_disconnect,
2429 .accept = inet_csk_accept,
2431 .init = tcp_v4_init_sock,
2432 .destroy = tcp_v4_destroy_sock,
2433 .shutdown = tcp_shutdown,
2434 .setsockopt = tcp_setsockopt,
2435 .getsockopt = tcp_getsockopt,
2436 .recvmsg = tcp_recvmsg,
2437 .backlog_rcv = tcp_v4_do_rcv,
2439 .unhash = inet_unhash,
2440 .get_port = inet_csk_get_port,
2441 .enter_memory_pressure = tcp_enter_memory_pressure,
2442 .sockets_allocated = &tcp_sockets_allocated,
2443 .orphan_count = &tcp_orphan_count,
2444 .memory_allocated = &tcp_memory_allocated,
2445 .memory_pressure = &tcp_memory_pressure,
2446 .sysctl_mem = sysctl_tcp_mem,
2447 .sysctl_wmem = sysctl_tcp_wmem,
2448 .sysctl_rmem = sysctl_tcp_rmem,
2449 .max_header = MAX_TCP_HEADER,
2450 .obj_size = sizeof(struct tcp_sock),
2451 .slab_flags = SLAB_DESTROY_BY_RCU,
2452 .twsk_prot = &tcp_timewait_sock_ops,
2453 .rsk_prot = &tcp_request_sock_ops,
2454 .h.hashinfo = &tcp_hashinfo,
2455 #ifdef CONFIG_COMPAT
2456 .compat_setsockopt = compat_tcp_setsockopt,
2457 .compat_getsockopt = compat_tcp_getsockopt,
2462 static int __net_init tcp_sk_init(struct net *net)
2464 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2465 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2468 static void __net_exit tcp_sk_exit(struct net *net)
2470 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2471 inet_twsk_purge(net, &tcp_hashinfo, &tcp_death_row, AF_INET);
2474 static struct pernet_operations __net_initdata tcp_sk_ops = {
2475 .init = tcp_sk_init,
2476 .exit = tcp_sk_exit,
2479 void __init tcp_v4_init(void)
2481 inet_hashinfo_init(&tcp_hashinfo);
2482 if (register_pernet_subsys(&tcp_sk_ops))
2483 panic("Failed to create the TCP control socket.\n");
2486 EXPORT_SYMBOL(ipv4_specific);
2487 EXPORT_SYMBOL(tcp_hashinfo);
2488 EXPORT_SYMBOL(tcp_prot);
2489 EXPORT_SYMBOL(tcp_v4_conn_request);
2490 EXPORT_SYMBOL(tcp_v4_connect);
2491 EXPORT_SYMBOL(tcp_v4_do_rcv);
2492 EXPORT_SYMBOL(tcp_v4_remember_stamp);
2493 EXPORT_SYMBOL(tcp_v4_send_check);
2494 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2496 #ifdef CONFIG_PROC_FS
2497 EXPORT_SYMBOL(tcp_proc_register);
2498 EXPORT_SYMBOL(tcp_proc_unregister);
2500 EXPORT_SYMBOL(sysctl_tcp_low_latency);