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->saddr,
169 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
171 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)
187 inet->saddr = rt->rt_src;
188 inet->rcv_saddr = inet->saddr;
190 if (tp->rx_opt.ts_recent_stamp && 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 peer->tcp_ts_stamp + TCP_PAWS_MSL >= get_seconds()) {
208 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
209 tp->rx_opt.ts_recent = peer->tcp_ts;
213 inet->dport = usin->sin_port;
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 = 536;
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->sport, 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->saddr,
247 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;
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 *skb, u32 info)
333 struct iphdr *iph = (struct iphdr *)skb->data;
334 struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
336 struct inet_sock *inet;
337 const int type = icmp_hdr(skb)->type;
338 const int code = icmp_hdr(skb)->code;
342 struct net *net = dev_net(skb->dev);
344 if (skb->len < (iph->ihl << 2) + 8) {
345 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
349 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
350 iph->saddr, th->source, inet_iif(skb));
352 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
355 if (sk->sk_state == TCP_TIME_WAIT) {
356 inet_twsk_put(inet_twsk(sk));
361 /* If too many ICMPs get dropped on busy
362 * servers this needs to be solved differently.
364 if (sock_owned_by_user(sk))
365 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
367 if (sk->sk_state == TCP_CLOSE)
371 seq = ntohl(th->seq);
372 if (sk->sk_state != TCP_LISTEN &&
373 !between(seq, tp->snd_una, tp->snd_nxt)) {
374 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
379 case ICMP_SOURCE_QUENCH:
380 /* Just silently ignore these. */
382 case ICMP_PARAMETERPROB:
385 case ICMP_DEST_UNREACH:
386 if (code > NR_ICMP_UNREACH)
389 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
390 if (!sock_owned_by_user(sk))
391 do_pmtu_discovery(sk, iph, info);
395 err = icmp_err_convert[code].errno;
397 case ICMP_TIME_EXCEEDED:
404 switch (sk->sk_state) {
405 struct request_sock *req, **prev;
407 if (sock_owned_by_user(sk))
410 req = inet_csk_search_req(sk, &prev, th->dest,
411 iph->daddr, iph->saddr);
415 /* ICMPs are not backlogged, hence we cannot get
416 an established socket here.
420 if (seq != tcp_rsk(req)->snt_isn) {
421 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
426 * Still in SYN_RECV, just remove it silently.
427 * There is no good way to pass the error to the newly
428 * created socket, and POSIX does not want network
429 * errors returned from accept().
431 inet_csk_reqsk_queue_drop(sk, req, prev);
435 case TCP_SYN_RECV: /* Cannot happen.
436 It can f.e. if SYNs crossed.
438 if (!sock_owned_by_user(sk)) {
441 sk->sk_error_report(sk);
445 sk->sk_err_soft = err;
450 /* If we've already connected we will keep trying
451 * until we time out, or the user gives up.
453 * rfc1122 4.2.3.9 allows to consider as hard errors
454 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
455 * but it is obsoleted by pmtu discovery).
457 * Note, that in modern internet, where routing is unreliable
458 * and in each dark corner broken firewalls sit, sending random
459 * errors ordered by their masters even this two messages finally lose
460 * their original sense (even Linux sends invalid PORT_UNREACHs)
462 * Now we are in compliance with RFCs.
467 if (!sock_owned_by_user(sk) && inet->recverr) {
469 sk->sk_error_report(sk);
470 } else { /* Only an error on timeout */
471 sk->sk_err_soft = err;
479 /* This routine computes an IPv4 TCP checksum. */
480 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
482 struct inet_sock *inet = inet_sk(sk);
483 struct tcphdr *th = tcp_hdr(skb);
485 if (skb->ip_summed == CHECKSUM_PARTIAL) {
486 th->check = ~tcp_v4_check(len, inet->saddr,
488 skb->csum_start = skb_transport_header(skb) - skb->head;
489 skb->csum_offset = offsetof(struct tcphdr, check);
491 th->check = tcp_v4_check(len, inet->saddr, inet->daddr,
498 int tcp_v4_gso_send_check(struct sk_buff *skb)
500 const struct iphdr *iph;
503 if (!pskb_may_pull(skb, sizeof(*th)))
510 th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0);
511 skb->csum_start = skb_transport_header(skb) - skb->head;
512 skb->csum_offset = offsetof(struct tcphdr, check);
513 skb->ip_summed = CHECKSUM_PARTIAL;
518 * This routine will send an RST to the other tcp.
520 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
522 * Answer: if a packet caused RST, it is not for a socket
523 * existing in our system, if it is matched to a socket,
524 * it is just duplicate segment or bug in other side's TCP.
525 * So that we build reply only basing on parameters
526 * arrived with segment.
527 * Exception: precedence violation. We do not implement it in any case.
530 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
532 struct tcphdr *th = tcp_hdr(skb);
535 #ifdef CONFIG_TCP_MD5SIG
536 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
539 struct ip_reply_arg arg;
540 #ifdef CONFIG_TCP_MD5SIG
541 struct tcp_md5sig_key *key;
545 /* Never send a reset in response to a reset. */
549 if (skb->rtable->rt_type != RTN_LOCAL)
552 /* Swap the send and the receive. */
553 memset(&rep, 0, sizeof(rep));
554 rep.th.dest = th->source;
555 rep.th.source = th->dest;
556 rep.th.doff = sizeof(struct tcphdr) / 4;
560 rep.th.seq = th->ack_seq;
563 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
564 skb->len - (th->doff << 2));
567 memset(&arg, 0, sizeof(arg));
568 arg.iov[0].iov_base = (unsigned char *)&rep;
569 arg.iov[0].iov_len = sizeof(rep.th);
571 #ifdef CONFIG_TCP_MD5SIG
572 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
574 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
576 (TCPOPT_MD5SIG << 8) |
578 /* Update length and the length the header thinks exists */
579 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
580 rep.th.doff = arg.iov[0].iov_len / 4;
582 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
583 key, ip_hdr(skb)->saddr,
584 ip_hdr(skb)->daddr, &rep.th);
587 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
588 ip_hdr(skb)->saddr, /* XXX */
589 arg.iov[0].iov_len, IPPROTO_TCP, 0);
590 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
591 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
593 net = dev_net(skb->dst->dev);
594 ip_send_reply(net->ipv4.tcp_sock, skb,
595 &arg, arg.iov[0].iov_len);
597 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
598 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
601 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
602 outside socket context is ugly, certainly. What can I do?
605 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
606 u32 win, u32 ts, int oif,
607 struct tcp_md5sig_key *key,
610 struct tcphdr *th = tcp_hdr(skb);
613 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
614 #ifdef CONFIG_TCP_MD5SIG
615 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
619 struct ip_reply_arg arg;
620 struct net *net = dev_net(skb->dst->dev);
622 memset(&rep.th, 0, sizeof(struct tcphdr));
623 memset(&arg, 0, sizeof(arg));
625 arg.iov[0].iov_base = (unsigned char *)&rep;
626 arg.iov[0].iov_len = sizeof(rep.th);
628 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
629 (TCPOPT_TIMESTAMP << 8) |
631 rep.opt[1] = htonl(tcp_time_stamp);
632 rep.opt[2] = htonl(ts);
633 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
636 /* Swap the send and the receive. */
637 rep.th.dest = th->source;
638 rep.th.source = th->dest;
639 rep.th.doff = arg.iov[0].iov_len / 4;
640 rep.th.seq = htonl(seq);
641 rep.th.ack_seq = htonl(ack);
643 rep.th.window = htons(win);
645 #ifdef CONFIG_TCP_MD5SIG
647 int offset = (ts) ? 3 : 0;
649 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
651 (TCPOPT_MD5SIG << 8) |
653 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
654 rep.th.doff = arg.iov[0].iov_len/4;
656 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
657 key, ip_hdr(skb)->saddr,
658 ip_hdr(skb)->daddr, &rep.th);
661 arg.flags = reply_flags;
662 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
663 ip_hdr(skb)->saddr, /* XXX */
664 arg.iov[0].iov_len, IPPROTO_TCP, 0);
665 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
667 arg.bound_dev_if = oif;
669 ip_send_reply(net->ipv4.tcp_sock, skb,
670 &arg, arg.iov[0].iov_len);
672 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
675 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
677 struct inet_timewait_sock *tw = inet_twsk(sk);
678 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
680 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
681 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
684 tcp_twsk_md5_key(tcptw),
685 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
691 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
692 struct request_sock *req)
694 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
695 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
698 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
699 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
703 * Send a SYN-ACK after having received a SYN.
704 * This still operates on a request_sock only, not on a big
707 static int __tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
708 struct dst_entry *dst)
710 const struct inet_request_sock *ireq = inet_rsk(req);
712 struct sk_buff * skb;
714 /* First, grab a route. */
715 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
718 skb = tcp_make_synack(sk, dst, req);
721 struct tcphdr *th = tcp_hdr(skb);
723 th->check = tcp_v4_check(skb->len,
726 csum_partial(th, skb->len,
729 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
732 err = net_xmit_eval(err);
739 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req)
741 return __tcp_v4_send_synack(sk, req, NULL);
745 * IPv4 request_sock destructor.
747 static void tcp_v4_reqsk_destructor(struct request_sock *req)
749 kfree(inet_rsk(req)->opt);
752 #ifdef CONFIG_SYN_COOKIES
753 static void syn_flood_warning(struct sk_buff *skb)
755 static unsigned long warntime;
757 if (time_after(jiffies, (warntime + HZ * 60))) {
760 "possible SYN flooding on port %d. Sending cookies.\n",
761 ntohs(tcp_hdr(skb)->dest));
767 * Save and compile IPv4 options into the request_sock if needed.
769 static struct ip_options *tcp_v4_save_options(struct sock *sk,
772 struct ip_options *opt = &(IPCB(skb)->opt);
773 struct ip_options *dopt = NULL;
775 if (opt && opt->optlen) {
776 int opt_size = optlength(opt);
777 dopt = kmalloc(opt_size, GFP_ATOMIC);
779 if (ip_options_echo(dopt, skb)) {
788 #ifdef CONFIG_TCP_MD5SIG
790 * RFC2385 MD5 checksumming requires a mapping of
791 * IP address->MD5 Key.
792 * We need to maintain these in the sk structure.
795 /* Find the Key structure for an address. */
796 static struct tcp_md5sig_key *
797 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
799 struct tcp_sock *tp = tcp_sk(sk);
802 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
804 for (i = 0; i < tp->md5sig_info->entries4; i++) {
805 if (tp->md5sig_info->keys4[i].addr == addr)
806 return &tp->md5sig_info->keys4[i].base;
811 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
812 struct sock *addr_sk)
814 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->daddr);
817 EXPORT_SYMBOL(tcp_v4_md5_lookup);
819 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
820 struct request_sock *req)
822 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
825 /* This can be called on a newly created socket, from other files */
826 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
827 u8 *newkey, u8 newkeylen)
829 /* Add Key to the list */
830 struct tcp_md5sig_key *key;
831 struct tcp_sock *tp = tcp_sk(sk);
832 struct tcp4_md5sig_key *keys;
834 key = tcp_v4_md5_do_lookup(sk, addr);
836 /* Pre-existing entry - just update that one. */
839 key->keylen = newkeylen;
841 struct tcp_md5sig_info *md5sig;
843 if (!tp->md5sig_info) {
844 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
846 if (!tp->md5sig_info) {
850 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
852 if (tcp_alloc_md5sig_pool() == NULL) {
856 md5sig = tp->md5sig_info;
858 if (md5sig->alloced4 == md5sig->entries4) {
859 keys = kmalloc((sizeof(*keys) *
860 (md5sig->entries4 + 1)), GFP_ATOMIC);
863 tcp_free_md5sig_pool();
867 if (md5sig->entries4)
868 memcpy(keys, md5sig->keys4,
869 sizeof(*keys) * md5sig->entries4);
871 /* Free old key list, and reference new one */
872 kfree(md5sig->keys4);
873 md5sig->keys4 = keys;
877 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
878 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
879 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
884 EXPORT_SYMBOL(tcp_v4_md5_do_add);
886 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
887 u8 *newkey, u8 newkeylen)
889 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->daddr,
893 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
895 struct tcp_sock *tp = tcp_sk(sk);
898 for (i = 0; i < tp->md5sig_info->entries4; i++) {
899 if (tp->md5sig_info->keys4[i].addr == addr) {
901 kfree(tp->md5sig_info->keys4[i].base.key);
902 tp->md5sig_info->entries4--;
904 if (tp->md5sig_info->entries4 == 0) {
905 kfree(tp->md5sig_info->keys4);
906 tp->md5sig_info->keys4 = NULL;
907 tp->md5sig_info->alloced4 = 0;
908 } else if (tp->md5sig_info->entries4 != i) {
909 /* Need to do some manipulation */
910 memmove(&tp->md5sig_info->keys4[i],
911 &tp->md5sig_info->keys4[i+1],
912 (tp->md5sig_info->entries4 - i) *
913 sizeof(struct tcp4_md5sig_key));
915 tcp_free_md5sig_pool();
922 EXPORT_SYMBOL(tcp_v4_md5_do_del);
924 static void tcp_v4_clear_md5_list(struct sock *sk)
926 struct tcp_sock *tp = tcp_sk(sk);
928 /* Free each key, then the set of key keys,
929 * the crypto element, and then decrement our
930 * hold on the last resort crypto.
932 if (tp->md5sig_info->entries4) {
934 for (i = 0; i < tp->md5sig_info->entries4; i++)
935 kfree(tp->md5sig_info->keys4[i].base.key);
936 tp->md5sig_info->entries4 = 0;
937 tcp_free_md5sig_pool();
939 if (tp->md5sig_info->keys4) {
940 kfree(tp->md5sig_info->keys4);
941 tp->md5sig_info->keys4 = NULL;
942 tp->md5sig_info->alloced4 = 0;
946 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
949 struct tcp_md5sig cmd;
950 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
953 if (optlen < sizeof(cmd))
956 if (copy_from_user(&cmd, optval, sizeof(cmd)))
959 if (sin->sin_family != AF_INET)
962 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
963 if (!tcp_sk(sk)->md5sig_info)
965 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
968 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
971 if (!tcp_sk(sk)->md5sig_info) {
972 struct tcp_sock *tp = tcp_sk(sk);
973 struct tcp_md5sig_info *p = kzalloc(sizeof(*p), GFP_KERNEL);
979 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
982 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
985 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
986 newkey, cmd.tcpm_keylen);
989 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
990 __be32 daddr, __be32 saddr, int nbytes)
992 struct tcp4_pseudohdr *bp;
993 struct scatterlist sg;
995 bp = &hp->md5_blk.ip4;
998 * 1. the TCP pseudo-header (in the order: source IP address,
999 * destination IP address, zero-padded protocol number, and
1005 bp->protocol = IPPROTO_TCP;
1006 bp->len = cpu_to_be16(nbytes);
1008 sg_init_one(&sg, bp, sizeof(*bp));
1009 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1012 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1013 __be32 daddr, __be32 saddr, struct tcphdr *th)
1015 struct tcp_md5sig_pool *hp;
1016 struct hash_desc *desc;
1018 hp = tcp_get_md5sig_pool();
1020 goto clear_hash_noput;
1021 desc = &hp->md5_desc;
1023 if (crypto_hash_init(desc))
1025 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1027 if (tcp_md5_hash_header(hp, th))
1029 if (tcp_md5_hash_key(hp, key))
1031 if (crypto_hash_final(desc, md5_hash))
1034 tcp_put_md5sig_pool();
1038 tcp_put_md5sig_pool();
1040 memset(md5_hash, 0, 16);
1044 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1045 struct sock *sk, struct request_sock *req,
1046 struct sk_buff *skb)
1048 struct tcp_md5sig_pool *hp;
1049 struct hash_desc *desc;
1050 struct tcphdr *th = tcp_hdr(skb);
1051 __be32 saddr, daddr;
1054 saddr = inet_sk(sk)->saddr;
1055 daddr = inet_sk(sk)->daddr;
1057 saddr = inet_rsk(req)->loc_addr;
1058 daddr = inet_rsk(req)->rmt_addr;
1060 const struct iphdr *iph = ip_hdr(skb);
1065 hp = tcp_get_md5sig_pool();
1067 goto clear_hash_noput;
1068 desc = &hp->md5_desc;
1070 if (crypto_hash_init(desc))
1073 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1075 if (tcp_md5_hash_header(hp, th))
1077 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1079 if (tcp_md5_hash_key(hp, key))
1081 if (crypto_hash_final(desc, md5_hash))
1084 tcp_put_md5sig_pool();
1088 tcp_put_md5sig_pool();
1090 memset(md5_hash, 0, 16);
1094 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1096 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1099 * This gets called for each TCP segment that arrives
1100 * so we want to be efficient.
1101 * We have 3 drop cases:
1102 * o No MD5 hash and one expected.
1103 * o MD5 hash and we're not expecting one.
1104 * o MD5 hash and its wrong.
1106 __u8 *hash_location = NULL;
1107 struct tcp_md5sig_key *hash_expected;
1108 const struct iphdr *iph = ip_hdr(skb);
1109 struct tcphdr *th = tcp_hdr(skb);
1111 unsigned char newhash[16];
1113 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1114 hash_location = tcp_parse_md5sig_option(th);
1116 /* We've parsed the options - do we have a hash? */
1117 if (!hash_expected && !hash_location)
1120 if (hash_expected && !hash_location) {
1121 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1125 if (!hash_expected && hash_location) {
1126 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1130 /* Okay, so this is hash_expected and hash_location -
1131 * so we need to calculate the checksum.
1133 genhash = tcp_v4_md5_hash_skb(newhash,
1137 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1138 if (net_ratelimit()) {
1139 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1140 &iph->saddr, ntohs(th->source),
1141 &iph->daddr, ntohs(th->dest),
1142 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1151 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1153 .obj_size = sizeof(struct tcp_request_sock),
1154 .rtx_syn_ack = tcp_v4_send_synack,
1155 .send_ack = tcp_v4_reqsk_send_ack,
1156 .destructor = tcp_v4_reqsk_destructor,
1157 .send_reset = tcp_v4_send_reset,
1160 #ifdef CONFIG_TCP_MD5SIG
1161 static struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1162 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1166 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1167 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1168 .twsk_unique = tcp_twsk_unique,
1169 .twsk_destructor= tcp_twsk_destructor,
1172 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1174 struct inet_request_sock *ireq;
1175 struct tcp_options_received tmp_opt;
1176 struct request_sock *req;
1177 __be32 saddr = ip_hdr(skb)->saddr;
1178 __be32 daddr = ip_hdr(skb)->daddr;
1179 __u32 isn = TCP_SKB_CB(skb)->when;
1180 struct dst_entry *dst = NULL;
1181 #ifdef CONFIG_SYN_COOKIES
1182 int want_cookie = 0;
1184 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1187 /* Never answer to SYNs send to broadcast or multicast */
1188 if (skb->rtable->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1191 /* TW buckets are converted to open requests without
1192 * limitations, they conserve resources and peer is
1193 * evidently real one.
1195 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1196 #ifdef CONFIG_SYN_COOKIES
1197 if (sysctl_tcp_syncookies) {
1204 /* Accept backlog is full. If we have already queued enough
1205 * of warm entries in syn queue, drop request. It is better than
1206 * clogging syn queue with openreqs with exponentially increasing
1209 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1212 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1216 #ifdef CONFIG_TCP_MD5SIG
1217 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1220 tcp_clear_options(&tmp_opt);
1221 tmp_opt.mss_clamp = 536;
1222 tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss;
1224 tcp_parse_options(skb, &tmp_opt, 0);
1226 if (want_cookie && !tmp_opt.saw_tstamp)
1227 tcp_clear_options(&tmp_opt);
1229 if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) {
1230 /* Some OSes (unknown ones, but I see them on web server, which
1231 * contains information interesting only for windows'
1232 * users) do not send their stamp in SYN. It is easy case.
1233 * We simply do not advertise TS support.
1235 tmp_opt.saw_tstamp = 0;
1236 tmp_opt.tstamp_ok = 0;
1238 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1240 tcp_openreq_init(req, &tmp_opt, skb);
1242 if (security_inet_conn_request(sk, skb, req))
1245 ireq = inet_rsk(req);
1246 ireq->loc_addr = daddr;
1247 ireq->rmt_addr = saddr;
1248 ireq->no_srccheck = inet_sk(sk)->transparent;
1249 ireq->opt = tcp_v4_save_options(sk, skb);
1251 TCP_ECN_create_request(req, tcp_hdr(skb));
1254 #ifdef CONFIG_SYN_COOKIES
1255 syn_flood_warning(skb);
1256 req->cookie_ts = tmp_opt.tstamp_ok;
1258 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1260 struct inet_peer *peer = NULL;
1262 /* VJ's idea. We save last timestamp seen
1263 * from the destination in peer table, when entering
1264 * state TIME-WAIT, and check against it before
1265 * accepting new connection request.
1267 * If "isn" is not zero, this request hit alive
1268 * timewait bucket, so that all the necessary checks
1269 * are made in the function processing timewait state.
1271 if (tmp_opt.saw_tstamp &&
1272 tcp_death_row.sysctl_tw_recycle &&
1273 (dst = inet_csk_route_req(sk, req)) != NULL &&
1274 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1275 peer->v4daddr == saddr) {
1276 if (get_seconds() < peer->tcp_ts_stamp + TCP_PAWS_MSL &&
1277 (s32)(peer->tcp_ts - req->ts_recent) >
1279 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1280 goto drop_and_release;
1283 /* Kill the following clause, if you dislike this way. */
1284 else if (!sysctl_tcp_syncookies &&
1285 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1286 (sysctl_max_syn_backlog >> 2)) &&
1287 (!peer || !peer->tcp_ts_stamp) &&
1288 (!dst || !dst_metric(dst, RTAX_RTT))) {
1289 /* Without syncookies last quarter of
1290 * backlog is filled with destinations,
1291 * proven to be alive.
1292 * It means that we continue to communicate
1293 * to destinations, already remembered
1294 * to the moment of synflood.
1296 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1297 &saddr, ntohs(tcp_hdr(skb)->source));
1298 goto drop_and_release;
1301 isn = tcp_v4_init_sequence(skb);
1303 tcp_rsk(req)->snt_isn = isn;
1305 if (__tcp_v4_send_synack(sk, req, dst) || want_cookie)
1308 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1321 * The three way handshake has completed - we got a valid synack -
1322 * now create the new socket.
1324 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1325 struct request_sock *req,
1326 struct dst_entry *dst)
1328 struct inet_request_sock *ireq;
1329 struct inet_sock *newinet;
1330 struct tcp_sock *newtp;
1332 #ifdef CONFIG_TCP_MD5SIG
1333 struct tcp_md5sig_key *key;
1336 if (sk_acceptq_is_full(sk))
1339 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1342 newsk = tcp_create_openreq_child(sk, req, skb);
1346 newsk->sk_gso_type = SKB_GSO_TCPV4;
1347 sk_setup_caps(newsk, dst);
1349 newtp = tcp_sk(newsk);
1350 newinet = inet_sk(newsk);
1351 ireq = inet_rsk(req);
1352 newinet->daddr = ireq->rmt_addr;
1353 newinet->rcv_saddr = ireq->loc_addr;
1354 newinet->saddr = ireq->loc_addr;
1355 newinet->opt = ireq->opt;
1357 newinet->mc_index = inet_iif(skb);
1358 newinet->mc_ttl = ip_hdr(skb)->ttl;
1359 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1361 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
1362 newinet->id = newtp->write_seq ^ jiffies;
1364 tcp_mtup_init(newsk);
1365 tcp_sync_mss(newsk, dst_mtu(dst));
1366 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1367 if (tcp_sk(sk)->rx_opt.user_mss &&
1368 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1369 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1371 tcp_initialize_rcv_mss(newsk);
1373 #ifdef CONFIG_TCP_MD5SIG
1374 /* Copy over the MD5 key from the original socket */
1375 if ((key = tcp_v4_md5_do_lookup(sk, newinet->daddr)) != NULL) {
1377 * We're using one, so create a matching key
1378 * on the newsk structure. If we fail to get
1379 * memory, then we end up not copying the key
1382 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1384 tcp_v4_md5_do_add(newsk, inet_sk(sk)->daddr,
1385 newkey, key->keylen);
1386 newsk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1390 __inet_hash_nolisten(newsk);
1391 __inet_inherit_port(sk, newsk);
1396 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1398 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1403 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1405 struct tcphdr *th = tcp_hdr(skb);
1406 const struct iphdr *iph = ip_hdr(skb);
1408 struct request_sock **prev;
1409 /* Find possible connection requests. */
1410 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1411 iph->saddr, iph->daddr);
1413 return tcp_check_req(sk, skb, req, prev);
1415 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1416 th->source, iph->daddr, th->dest, inet_iif(skb));
1419 if (nsk->sk_state != TCP_TIME_WAIT) {
1423 inet_twsk_put(inet_twsk(nsk));
1427 #ifdef CONFIG_SYN_COOKIES
1428 if (!th->rst && !th->syn && th->ack)
1429 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1434 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1436 const struct iphdr *iph = ip_hdr(skb);
1438 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1439 if (!tcp_v4_check(skb->len, iph->saddr,
1440 iph->daddr, skb->csum)) {
1441 skb->ip_summed = CHECKSUM_UNNECESSARY;
1446 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1447 skb->len, IPPROTO_TCP, 0);
1449 if (skb->len <= 76) {
1450 return __skb_checksum_complete(skb);
1456 /* The socket must have it's spinlock held when we get
1459 * We have a potential double-lock case here, so even when
1460 * doing backlog processing we use the BH locking scheme.
1461 * This is because we cannot sleep with the original spinlock
1464 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1467 #ifdef CONFIG_TCP_MD5SIG
1469 * We really want to reject the packet as early as possible
1471 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1472 * o There is an MD5 option and we're not expecting one
1474 if (tcp_v4_inbound_md5_hash(sk, skb))
1478 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1479 TCP_CHECK_TIMER(sk);
1480 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1484 TCP_CHECK_TIMER(sk);
1488 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1491 if (sk->sk_state == TCP_LISTEN) {
1492 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1497 if (tcp_child_process(sk, nsk, skb)) {
1505 TCP_CHECK_TIMER(sk);
1506 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1510 TCP_CHECK_TIMER(sk);
1514 tcp_v4_send_reset(rsk, skb);
1517 /* Be careful here. If this function gets more complicated and
1518 * gcc suffers from register pressure on the x86, sk (in %ebx)
1519 * might be destroyed here. This current version compiles correctly,
1520 * but you have been warned.
1525 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1533 int tcp_v4_rcv(struct sk_buff *skb)
1535 const struct iphdr *iph;
1539 struct net *net = dev_net(skb->dev);
1541 if (skb->pkt_type != PACKET_HOST)
1544 /* Count it even if it's bad */
1545 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1547 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1552 if (th->doff < sizeof(struct tcphdr) / 4)
1554 if (!pskb_may_pull(skb, th->doff * 4))
1557 /* An explanation is required here, I think.
1558 * Packet length and doff are validated by header prediction,
1559 * provided case of th->doff==0 is eliminated.
1560 * So, we defer the checks. */
1561 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1566 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1567 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1568 skb->len - th->doff * 4);
1569 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1570 TCP_SKB_CB(skb)->when = 0;
1571 TCP_SKB_CB(skb)->flags = iph->tos;
1572 TCP_SKB_CB(skb)->sacked = 0;
1574 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1579 if (sk->sk_state == TCP_TIME_WAIT)
1582 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1583 goto discard_and_relse;
1586 if (sk_filter(sk, skb))
1587 goto discard_and_relse;
1591 bh_lock_sock_nested(sk);
1593 if (!sock_owned_by_user(sk)) {
1594 #ifdef CONFIG_NET_DMA
1595 struct tcp_sock *tp = tcp_sk(sk);
1596 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1597 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1598 if (tp->ucopy.dma_chan)
1599 ret = tcp_v4_do_rcv(sk, skb);
1603 if (!tcp_prequeue(sk, skb))
1604 ret = tcp_v4_do_rcv(sk, skb);
1607 sk_add_backlog(sk, skb);
1615 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1618 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1620 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1622 tcp_v4_send_reset(NULL, skb);
1626 /* Discard frame. */
1635 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1636 inet_twsk_put(inet_twsk(sk));
1640 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1641 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1642 inet_twsk_put(inet_twsk(sk));
1645 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1647 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1649 iph->daddr, th->dest,
1652 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1653 inet_twsk_put(inet_twsk(sk));
1657 /* Fall through to ACK */
1660 tcp_v4_timewait_ack(sk, skb);
1664 case TCP_TW_SUCCESS:;
1669 /* VJ's idea. Save last timestamp seen from this destination
1670 * and hold it at least for normal timewait interval to use for duplicate
1671 * segment detection in subsequent connections, before they enter synchronized
1675 int tcp_v4_remember_stamp(struct sock *sk)
1677 struct inet_sock *inet = inet_sk(sk);
1678 struct tcp_sock *tp = tcp_sk(sk);
1679 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1680 struct inet_peer *peer = NULL;
1683 if (!rt || rt->rt_dst != inet->daddr) {
1684 peer = inet_getpeer(inet->daddr, 1);
1688 rt_bind_peer(rt, 1);
1693 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1694 (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() &&
1695 peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) {
1696 peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp;
1697 peer->tcp_ts = tp->rx_opt.ts_recent;
1707 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1709 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1712 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1714 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1715 (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() &&
1716 peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) {
1717 peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp;
1718 peer->tcp_ts = tcptw->tw_ts_recent;
1727 struct inet_connection_sock_af_ops ipv4_specific = {
1728 .queue_xmit = ip_queue_xmit,
1729 .send_check = tcp_v4_send_check,
1730 .rebuild_header = inet_sk_rebuild_header,
1731 .conn_request = tcp_v4_conn_request,
1732 .syn_recv_sock = tcp_v4_syn_recv_sock,
1733 .remember_stamp = tcp_v4_remember_stamp,
1734 .net_header_len = sizeof(struct iphdr),
1735 .setsockopt = ip_setsockopt,
1736 .getsockopt = ip_getsockopt,
1737 .addr2sockaddr = inet_csk_addr2sockaddr,
1738 .sockaddr_len = sizeof(struct sockaddr_in),
1739 .bind_conflict = inet_csk_bind_conflict,
1740 #ifdef CONFIG_COMPAT
1741 .compat_setsockopt = compat_ip_setsockopt,
1742 .compat_getsockopt = compat_ip_getsockopt,
1746 #ifdef CONFIG_TCP_MD5SIG
1747 static struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1748 .md5_lookup = tcp_v4_md5_lookup,
1749 .calc_md5_hash = tcp_v4_md5_hash_skb,
1750 .md5_add = tcp_v4_md5_add_func,
1751 .md5_parse = tcp_v4_parse_md5_keys,
1755 /* NOTE: A lot of things set to zero explicitly by call to
1756 * sk_alloc() so need not be done here.
1758 static int tcp_v4_init_sock(struct sock *sk)
1760 struct inet_connection_sock *icsk = inet_csk(sk);
1761 struct tcp_sock *tp = tcp_sk(sk);
1763 skb_queue_head_init(&tp->out_of_order_queue);
1764 tcp_init_xmit_timers(sk);
1765 tcp_prequeue_init(tp);
1767 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1768 tp->mdev = TCP_TIMEOUT_INIT;
1770 /* So many TCP implementations out there (incorrectly) count the
1771 * initial SYN frame in their delayed-ACK and congestion control
1772 * algorithms that we must have the following bandaid to talk
1773 * efficiently to them. -DaveM
1777 /* See draft-stevens-tcpca-spec-01 for discussion of the
1778 * initialization of these values.
1780 tp->snd_ssthresh = 0x7fffffff; /* Infinity */
1781 tp->snd_cwnd_clamp = ~0;
1782 tp->mss_cache = 536;
1784 tp->reordering = sysctl_tcp_reordering;
1785 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1787 sk->sk_state = TCP_CLOSE;
1789 sk->sk_write_space = sk_stream_write_space;
1790 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1792 icsk->icsk_af_ops = &ipv4_specific;
1793 icsk->icsk_sync_mss = tcp_sync_mss;
1794 #ifdef CONFIG_TCP_MD5SIG
1795 tp->af_specific = &tcp_sock_ipv4_specific;
1798 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1799 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1802 percpu_counter_inc(&tcp_sockets_allocated);
1808 void tcp_v4_destroy_sock(struct sock *sk)
1810 struct tcp_sock *tp = tcp_sk(sk);
1812 tcp_clear_xmit_timers(sk);
1814 tcp_cleanup_congestion_control(sk);
1816 /* Cleanup up the write buffer. */
1817 tcp_write_queue_purge(sk);
1819 /* Cleans up our, hopefully empty, out_of_order_queue. */
1820 __skb_queue_purge(&tp->out_of_order_queue);
1822 #ifdef CONFIG_TCP_MD5SIG
1823 /* Clean up the MD5 key list, if any */
1824 if (tp->md5sig_info) {
1825 tcp_v4_clear_md5_list(sk);
1826 kfree(tp->md5sig_info);
1827 tp->md5sig_info = NULL;
1831 #ifdef CONFIG_NET_DMA
1832 /* Cleans up our sk_async_wait_queue */
1833 __skb_queue_purge(&sk->sk_async_wait_queue);
1836 /* Clean prequeue, it must be empty really */
1837 __skb_queue_purge(&tp->ucopy.prequeue);
1839 /* Clean up a referenced TCP bind bucket. */
1840 if (inet_csk(sk)->icsk_bind_hash)
1844 * If sendmsg cached page exists, toss it.
1846 if (sk->sk_sndmsg_page) {
1847 __free_page(sk->sk_sndmsg_page);
1848 sk->sk_sndmsg_page = NULL;
1851 percpu_counter_dec(&tcp_sockets_allocated);
1854 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1857 * tcp_v4_nuke_addr - destroy all sockets on the given local address
1859 void tcp_v4_nuke_addr(__u32 saddr)
1861 unsigned int bucket;
1863 for (bucket = 0; bucket < tcp_hashinfo.ehash_size; bucket++) {
1864 struct hlist_nulls_node *node;
1866 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, bucket);
1869 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[bucket].chain) {
1870 struct inet_sock *inet = inet_sk(sk);
1872 if (inet->rcv_saddr != saddr)
1874 if (sysctl_ip_dynaddr && sk->sk_state == TCP_SYN_SENT)
1876 if (sock_flag(sk, SOCK_DEAD))
1879 sk->sk_err = ETIMEDOUT;
1880 sk->sk_error_report(sk);
1883 spin_unlock_bh(lock);
1887 #ifdef CONFIG_PROC_FS
1888 /* Proc filesystem TCP sock list dumping. */
1890 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1892 return hlist_nulls_empty(head) ? NULL :
1893 list_entry(head->first, struct inet_timewait_sock, tw_node);
1896 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1898 return !is_a_nulls(tw->tw_node.next) ?
1899 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1902 static void *listening_get_next(struct seq_file *seq, void *cur)
1904 struct inet_connection_sock *icsk;
1905 struct hlist_nulls_node *node;
1906 struct sock *sk = cur;
1907 struct inet_listen_hashbucket *ilb;
1908 struct tcp_iter_state *st = seq->private;
1909 struct net *net = seq_file_net(seq);
1913 ilb = &tcp_hashinfo.listening_hash[0];
1914 spin_lock_bh(&ilb->lock);
1915 sk = sk_nulls_head(&ilb->head);
1918 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1921 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1922 struct request_sock *req = cur;
1924 icsk = inet_csk(st->syn_wait_sk);
1928 if (req->rsk_ops->family == st->family) {
1934 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1937 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1939 sk = sk_next(st->syn_wait_sk);
1940 st->state = TCP_SEQ_STATE_LISTENING;
1941 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1943 icsk = inet_csk(sk);
1944 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1945 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1947 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1951 sk_nulls_for_each_from(sk, node) {
1952 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) {
1956 icsk = inet_csk(sk);
1957 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1958 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1960 st->uid = sock_i_uid(sk);
1961 st->syn_wait_sk = sk;
1962 st->state = TCP_SEQ_STATE_OPENREQ;
1966 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1968 spin_unlock_bh(&ilb->lock);
1969 if (++st->bucket < INET_LHTABLE_SIZE) {
1970 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1971 spin_lock_bh(&ilb->lock);
1972 sk = sk_nulls_head(&ilb->head);
1980 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1982 void *rc = listening_get_next(seq, NULL);
1984 while (rc && *pos) {
1985 rc = listening_get_next(seq, rc);
1991 static inline int empty_bucket(struct tcp_iter_state *st)
1993 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
1994 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
1997 static void *established_get_first(struct seq_file *seq)
1999 struct tcp_iter_state *st = seq->private;
2000 struct net *net = seq_file_net(seq);
2003 for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
2005 struct hlist_nulls_node *node;
2006 struct inet_timewait_sock *tw;
2007 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2009 /* Lockless fast path for the common case of empty buckets */
2010 if (empty_bucket(st))
2014 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2015 if (sk->sk_family != st->family ||
2016 !net_eq(sock_net(sk), net)) {
2022 st->state = TCP_SEQ_STATE_TIME_WAIT;
2023 inet_twsk_for_each(tw, node,
2024 &tcp_hashinfo.ehash[st->bucket].twchain) {
2025 if (tw->tw_family != st->family ||
2026 !net_eq(twsk_net(tw), net)) {
2032 spin_unlock_bh(lock);
2033 st->state = TCP_SEQ_STATE_ESTABLISHED;
2039 static void *established_get_next(struct seq_file *seq, void *cur)
2041 struct sock *sk = cur;
2042 struct inet_timewait_sock *tw;
2043 struct hlist_nulls_node *node;
2044 struct tcp_iter_state *st = seq->private;
2045 struct net *net = seq_file_net(seq);
2049 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2053 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2060 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2061 st->state = TCP_SEQ_STATE_ESTABLISHED;
2063 /* Look for next non empty bucket */
2064 while (++st->bucket < tcp_hashinfo.ehash_size &&
2067 if (st->bucket >= tcp_hashinfo.ehash_size)
2070 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2071 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2073 sk = sk_nulls_next(sk);
2075 sk_nulls_for_each_from(sk, node) {
2076 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2080 st->state = TCP_SEQ_STATE_TIME_WAIT;
2081 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2089 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2091 void *rc = established_get_first(seq);
2094 rc = established_get_next(seq, rc);
2100 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2103 struct tcp_iter_state *st = seq->private;
2105 st->state = TCP_SEQ_STATE_LISTENING;
2106 rc = listening_get_idx(seq, &pos);
2109 st->state = TCP_SEQ_STATE_ESTABLISHED;
2110 rc = established_get_idx(seq, pos);
2116 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2118 struct tcp_iter_state *st = seq->private;
2119 st->state = TCP_SEQ_STATE_LISTENING;
2121 return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2124 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2127 struct tcp_iter_state *st;
2129 if (v == SEQ_START_TOKEN) {
2130 rc = tcp_get_idx(seq, 0);
2135 switch (st->state) {
2136 case TCP_SEQ_STATE_OPENREQ:
2137 case TCP_SEQ_STATE_LISTENING:
2138 rc = listening_get_next(seq, v);
2140 st->state = TCP_SEQ_STATE_ESTABLISHED;
2141 rc = established_get_first(seq);
2144 case TCP_SEQ_STATE_ESTABLISHED:
2145 case TCP_SEQ_STATE_TIME_WAIT:
2146 rc = established_get_next(seq, v);
2154 static void tcp_seq_stop(struct seq_file *seq, void *v)
2156 struct tcp_iter_state *st = seq->private;
2158 switch (st->state) {
2159 case TCP_SEQ_STATE_OPENREQ:
2161 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2162 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2164 case TCP_SEQ_STATE_LISTENING:
2165 if (v != SEQ_START_TOKEN)
2166 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2168 case TCP_SEQ_STATE_TIME_WAIT:
2169 case TCP_SEQ_STATE_ESTABLISHED:
2171 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2176 static int tcp_seq_open(struct inode *inode, struct file *file)
2178 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2179 struct tcp_iter_state *s;
2182 err = seq_open_net(inode, file, &afinfo->seq_ops,
2183 sizeof(struct tcp_iter_state));
2187 s = ((struct seq_file *)file->private_data)->private;
2188 s->family = afinfo->family;
2192 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2195 struct proc_dir_entry *p;
2197 afinfo->seq_fops.open = tcp_seq_open;
2198 afinfo->seq_fops.read = seq_read;
2199 afinfo->seq_fops.llseek = seq_lseek;
2200 afinfo->seq_fops.release = seq_release_net;
2202 afinfo->seq_ops.start = tcp_seq_start;
2203 afinfo->seq_ops.next = tcp_seq_next;
2204 afinfo->seq_ops.stop = tcp_seq_stop;
2206 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2207 &afinfo->seq_fops, afinfo);
2213 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2215 proc_net_remove(net, afinfo->name);
2218 static void get_openreq4(struct sock *sk, struct request_sock *req,
2219 struct seq_file *f, int i, int uid, int *len)
2221 const struct inet_request_sock *ireq = inet_rsk(req);
2222 int ttd = req->expires - jiffies;
2224 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2225 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
2228 ntohs(inet_sk(sk)->sport),
2230 ntohs(ireq->rmt_port),
2232 0, 0, /* could print option size, but that is af dependent. */
2233 1, /* timers active (only the expire timer) */
2234 jiffies_to_clock_t(ttd),
2237 0, /* non standard timer */
2238 0, /* open_requests have no inode */
2239 atomic_read(&sk->sk_refcnt),
2244 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2247 unsigned long timer_expires;
2248 struct tcp_sock *tp = tcp_sk(sk);
2249 const struct inet_connection_sock *icsk = inet_csk(sk);
2250 struct inet_sock *inet = inet_sk(sk);
2251 __be32 dest = inet->daddr;
2252 __be32 src = inet->rcv_saddr;
2253 __u16 destp = ntohs(inet->dport);
2254 __u16 srcp = ntohs(inet->sport);
2256 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2258 timer_expires = icsk->icsk_timeout;
2259 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2261 timer_expires = icsk->icsk_timeout;
2262 } else if (timer_pending(&sk->sk_timer)) {
2264 timer_expires = sk->sk_timer.expires;
2267 timer_expires = jiffies;
2270 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2271 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
2272 i, src, srcp, dest, destp, sk->sk_state,
2273 tp->write_seq - tp->snd_una,
2274 sk->sk_state == TCP_LISTEN ? sk->sk_ack_backlog :
2275 (tp->rcv_nxt - tp->copied_seq),
2277 jiffies_to_clock_t(timer_expires - jiffies),
2278 icsk->icsk_retransmits,
2280 icsk->icsk_probes_out,
2282 atomic_read(&sk->sk_refcnt), sk,
2283 jiffies_to_clock_t(icsk->icsk_rto),
2284 jiffies_to_clock_t(icsk->icsk_ack.ato),
2285 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2287 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh,
2291 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2292 struct seq_file *f, int i, int *len)
2296 int ttd = tw->tw_ttd - jiffies;
2301 dest = tw->tw_daddr;
2302 src = tw->tw_rcv_saddr;
2303 destp = ntohs(tw->tw_dport);
2304 srcp = ntohs(tw->tw_sport);
2306 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2307 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
2308 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2309 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2310 atomic_read(&tw->tw_refcnt), tw, len);
2315 static int tcp4_seq_show(struct seq_file *seq, void *v)
2317 struct tcp_iter_state *st;
2320 if (v == SEQ_START_TOKEN) {
2321 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2322 " sl local_address rem_address st tx_queue "
2323 "rx_queue tr tm->when retrnsmt uid timeout "
2329 switch (st->state) {
2330 case TCP_SEQ_STATE_LISTENING:
2331 case TCP_SEQ_STATE_ESTABLISHED:
2332 get_tcp4_sock(v, seq, st->num, &len);
2334 case TCP_SEQ_STATE_OPENREQ:
2335 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2337 case TCP_SEQ_STATE_TIME_WAIT:
2338 get_timewait4_sock(v, seq, st->num, &len);
2341 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2346 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2350 .owner = THIS_MODULE,
2353 .show = tcp4_seq_show,
2357 static int tcp4_proc_init_net(struct net *net)
2359 return tcp_proc_register(net, &tcp4_seq_afinfo);
2362 static void tcp4_proc_exit_net(struct net *net)
2364 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2367 static struct pernet_operations tcp4_net_ops = {
2368 .init = tcp4_proc_init_net,
2369 .exit = tcp4_proc_exit_net,
2372 int __init tcp4_proc_init(void)
2374 return register_pernet_subsys(&tcp4_net_ops);
2377 void tcp4_proc_exit(void)
2379 unregister_pernet_subsys(&tcp4_net_ops);
2381 #endif /* CONFIG_PROC_FS */
2383 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2385 struct iphdr *iph = ip_hdr(skb);
2387 switch (skb->ip_summed) {
2388 case CHECKSUM_COMPLETE:
2389 if (!tcp_v4_check(skb->len, iph->saddr, iph->daddr,
2391 skb->ip_summed = CHECKSUM_UNNECESSARY;
2397 NAPI_GRO_CB(skb)->flush = 1;
2401 return tcp_gro_receive(head, skb);
2403 EXPORT_SYMBOL(tcp4_gro_receive);
2405 int tcp4_gro_complete(struct sk_buff *skb)
2407 struct iphdr *iph = ip_hdr(skb);
2408 struct tcphdr *th = tcp_hdr(skb);
2410 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2411 iph->saddr, iph->daddr, 0);
2412 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2414 return tcp_gro_complete(skb);
2416 EXPORT_SYMBOL(tcp4_gro_complete);
2418 struct proto tcp_prot = {
2420 .owner = THIS_MODULE,
2422 .connect = tcp_v4_connect,
2423 .disconnect = tcp_disconnect,
2424 .accept = inet_csk_accept,
2426 .init = tcp_v4_init_sock,
2427 .destroy = tcp_v4_destroy_sock,
2428 .shutdown = tcp_shutdown,
2429 .setsockopt = tcp_setsockopt,
2430 .getsockopt = tcp_getsockopt,
2431 .recvmsg = tcp_recvmsg,
2432 .backlog_rcv = tcp_v4_do_rcv,
2434 .unhash = inet_unhash,
2435 .get_port = inet_csk_get_port,
2436 .enter_memory_pressure = tcp_enter_memory_pressure,
2437 .sockets_allocated = &tcp_sockets_allocated,
2438 .orphan_count = &tcp_orphan_count,
2439 .memory_allocated = &tcp_memory_allocated,
2440 .memory_pressure = &tcp_memory_pressure,
2441 .sysctl_mem = sysctl_tcp_mem,
2442 .sysctl_wmem = sysctl_tcp_wmem,
2443 .sysctl_rmem = sysctl_tcp_rmem,
2444 .max_header = MAX_TCP_HEADER,
2445 .obj_size = sizeof(struct tcp_sock),
2446 .slab_flags = SLAB_DESTROY_BY_RCU,
2447 .twsk_prot = &tcp_timewait_sock_ops,
2448 .rsk_prot = &tcp_request_sock_ops,
2449 .h.hashinfo = &tcp_hashinfo,
2450 #ifdef CONFIG_COMPAT
2451 .compat_setsockopt = compat_tcp_setsockopt,
2452 .compat_getsockopt = compat_tcp_getsockopt,
2457 static int __net_init tcp_sk_init(struct net *net)
2459 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2460 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2463 static void __net_exit tcp_sk_exit(struct net *net)
2465 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2466 inet_twsk_purge(net, &tcp_hashinfo, &tcp_death_row, AF_INET);
2469 static struct pernet_operations __net_initdata tcp_sk_ops = {
2470 .init = tcp_sk_init,
2471 .exit = tcp_sk_exit,
2474 void __init tcp_v4_init(void)
2476 inet_hashinfo_init(&tcp_hashinfo);
2477 if (register_pernet_subsys(&tcp_sk_ops))
2478 panic("Failed to create the TCP control socket.\n");
2481 EXPORT_SYMBOL(ipv4_specific);
2482 EXPORT_SYMBOL(tcp_hashinfo);
2483 EXPORT_SYMBOL(tcp_prot);
2484 EXPORT_SYMBOL(tcp_v4_conn_request);
2485 EXPORT_SYMBOL(tcp_v4_connect);
2486 EXPORT_SYMBOL(tcp_v4_do_rcv);
2487 EXPORT_SYMBOL(tcp_v4_remember_stamp);
2488 EXPORT_SYMBOL(tcp_v4_send_check);
2489 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2491 #ifdef CONFIG_PROC_FS
2492 EXPORT_SYMBOL(tcp_proc_register);
2493 EXPORT_SYMBOL(tcp_proc_unregister);
2495 EXPORT_SYMBOL(sysctl_tcp_low_latency);