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 * Version: $Id: tcp_ipv4.c,v 1.240 2002/02/01 22:01:04 davem Exp $
10 * IPv4 specific functions
15 * linux/ipv4/tcp_input.c
16 * linux/ipv4/tcp_output.c
18 * See tcp.c for author information
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
28 * David S. Miller : New socket lookup architecture.
29 * This code is dedicated to John Dyson.
30 * David S. Miller : Change semantics of established hash,
31 * half is devoted to TIME_WAIT sockets
32 * and the rest go in the other half.
33 * Andi Kleen : Add support for syncookies and fixed
34 * some bugs: ip options weren't passed to
35 * the TCP layer, missed a check for an
37 * Andi Kleen : Implemented fast path mtu discovery.
38 * Fixed many serious bugs in the
39 * request_sock handling and moved
40 * most of it into the af independent code.
41 * Added tail drop and some other bugfixes.
42 * Added new listen sematics.
43 * Mike McLagan : Routing by source
44 * Juan Jose Ciarlante: ip_dynaddr bits
45 * Andi Kleen: various fixes.
46 * Vitaly E. Lavrov : Transparent proxy revived after year
48 * Andi Kleen : Fix new listen.
49 * Andi Kleen : Fix accept error reporting.
50 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
51 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
52 * a single port at the same time.
55 #include <linux/config.h>
57 #include <linux/types.h>
58 #include <linux/fcntl.h>
59 #include <linux/module.h>
60 #include <linux/random.h>
61 #include <linux/cache.h>
62 #include <linux/jhash.h>
63 #include <linux/init.h>
64 #include <linux/times.h>
67 #include <net/inet_hashtables.h>
70 #include <net/inet_common.h>
73 #include <linux/inet.h>
74 #include <linux/ipv6.h>
75 #include <linux/stddef.h>
76 #include <linux/proc_fs.h>
77 #include <linux/seq_file.h>
79 extern int sysctl_ip_dynaddr;
80 int sysctl_tcp_tw_reuse;
81 int sysctl_tcp_low_latency;
83 /* Check TCP sequence numbers in ICMP packets. */
84 #define ICMP_MIN_LENGTH 8
86 /* Socket used for sending RSTs */
87 static struct socket *tcp_socket;
89 void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len,
92 struct inet_hashinfo __cacheline_aligned tcp_hashinfo = {
93 .lhash_lock = RW_LOCK_UNLOCKED,
94 .lhash_users = ATOMIC_INIT(0),
95 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait),
96 .portalloc_lock = SPIN_LOCK_UNLOCKED,
97 .port_rover = 1024 - 1,
100 static int tcp_v4_get_port(struct sock *sk, unsigned short snum)
102 return inet_csk_get_port(&tcp_hashinfo, sk, snum);
105 static void tcp_v4_hash(struct sock *sk)
107 inet_hash(&tcp_hashinfo, sk);
110 void tcp_unhash(struct sock *sk)
112 inet_unhash(&tcp_hashinfo, sk);
115 static inline __u32 tcp_v4_init_sequence(struct sock *sk, struct sk_buff *skb)
117 return secure_tcp_sequence_number(skb->nh.iph->daddr,
123 /* called with local bh disabled */
124 static int __tcp_v4_check_established(struct sock *sk, __u16 lport,
125 struct inet_timewait_sock **twp)
127 struct inet_sock *inet = inet_sk(sk);
128 u32 daddr = inet->rcv_saddr;
129 u32 saddr = inet->daddr;
130 int dif = sk->sk_bound_dev_if;
131 INET_ADDR_COOKIE(acookie, saddr, daddr)
132 const __u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
133 const int hash = inet_ehashfn(daddr, lport, saddr, inet->dport, tcp_hashinfo.ehash_size);
134 struct inet_ehash_bucket *head = &tcp_hashinfo.ehash[hash];
136 const struct hlist_node *node;
137 struct inet_timewait_sock *tw;
139 write_lock(&head->lock);
141 /* Check TIME-WAIT sockets first. */
142 sk_for_each(sk2, node, &(head + tcp_hashinfo.ehash_size)->chain) {
145 if (INET_TW_MATCH(sk2, acookie, saddr, daddr, ports, dif)) {
146 const struct tcp_timewait_sock *tcptw = tcp_twsk(sk2);
147 struct tcp_sock *tp = tcp_sk(sk);
149 /* With PAWS, it is safe from the viewpoint
150 of data integrity. Even without PAWS it
151 is safe provided sequence spaces do not
152 overlap i.e. at data rates <= 80Mbit/sec.
154 Actually, the idea is close to VJ's one,
155 only timestamp cache is held not per host,
156 but per port pair and TW bucket is used
159 If TW bucket has been already destroyed we
160 fall back to VJ's scheme and use initial
161 timestamp retrieved from peer table.
163 if (tcptw->tw_ts_recent_stamp &&
164 (!twp || (sysctl_tcp_tw_reuse &&
166 tcptw->tw_ts_recent_stamp > 1))) {
167 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
168 if (tp->write_seq == 0)
170 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
171 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
180 /* And established part... */
181 sk_for_each(sk2, node, &head->chain) {
182 if (INET_MATCH(sk2, acookie, saddr, daddr, ports, dif))
187 /* Must record num and sport now. Otherwise we will see
188 * in hash table socket with a funny identity. */
190 inet->sport = htons(lport);
191 sk->sk_hashent = hash;
192 BUG_TRAP(sk_unhashed(sk));
193 __sk_add_node(sk, &head->chain);
194 sock_prot_inc_use(sk->sk_prot);
195 write_unlock(&head->lock);
199 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
201 /* Silly. Should hash-dance instead... */
202 tcp_tw_deschedule(tw);
203 NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
211 write_unlock(&head->lock);
212 return -EADDRNOTAVAIL;
215 static inline u32 connect_port_offset(const struct sock *sk)
217 const struct inet_sock *inet = inet_sk(sk);
219 return secure_tcp_port_ephemeral(inet->rcv_saddr, inet->daddr,
224 * Bind a port for a connect operation and hash it.
226 static inline int tcp_v4_hash_connect(struct sock *sk)
228 const unsigned short snum = inet_sk(sk)->num;
229 struct inet_bind_hashbucket *head;
230 struct inet_bind_bucket *tb;
234 int low = sysctl_local_port_range[0];
235 int high = sysctl_local_port_range[1];
236 int range = high - low;
240 u32 offset = hint + connect_port_offset(sk);
241 struct hlist_node *node;
242 struct inet_timewait_sock *tw = NULL;
245 for (i = 1; i <= range; i++) {
246 port = low + (i + offset) % range;
247 head = &tcp_hashinfo.bhash[inet_bhashfn(port, tcp_hashinfo.bhash_size)];
248 spin_lock(&head->lock);
250 /* Does not bother with rcv_saddr checks,
251 * because the established check is already
254 inet_bind_bucket_for_each(tb, node, &head->chain) {
255 if (tb->port == port) {
256 BUG_TRAP(!hlist_empty(&tb->owners));
257 if (tb->fastreuse >= 0)
259 if (!__tcp_v4_check_established(sk,
267 tb = inet_bind_bucket_create(tcp_hashinfo.bind_bucket_cachep, head, port);
269 spin_unlock(&head->lock);
276 spin_unlock(&head->lock);
280 return -EADDRNOTAVAIL;
285 /* Head lock still held and bh's disabled */
286 inet_bind_hash(sk, tb, port);
287 if (sk_unhashed(sk)) {
288 inet_sk(sk)->sport = htons(port);
289 __inet_hash(&tcp_hashinfo, sk, 0);
291 spin_unlock(&head->lock);
294 tcp_tw_deschedule(tw);
302 head = &tcp_hashinfo.bhash[inet_bhashfn(snum, tcp_hashinfo.bhash_size)];
303 tb = inet_csk(sk)->icsk_bind_hash;
304 spin_lock_bh(&head->lock);
305 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
306 __inet_hash(&tcp_hashinfo, sk, 0);
307 spin_unlock_bh(&head->lock);
310 spin_unlock(&head->lock);
311 /* No definite answer... Walk to established hash table */
312 ret = __tcp_v4_check_established(sk, snum, NULL);
319 /* This will initiate an outgoing connection. */
320 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
322 struct inet_sock *inet = inet_sk(sk);
323 struct tcp_sock *tp = tcp_sk(sk);
324 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
330 if (addr_len < sizeof(struct sockaddr_in))
333 if (usin->sin_family != AF_INET)
334 return -EAFNOSUPPORT;
336 nexthop = daddr = usin->sin_addr.s_addr;
337 if (inet->opt && inet->opt->srr) {
340 nexthop = inet->opt->faddr;
343 tmp = ip_route_connect(&rt, nexthop, inet->saddr,
344 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
346 inet->sport, usin->sin_port, sk);
350 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
355 if (!inet->opt || !inet->opt->srr)
359 inet->saddr = rt->rt_src;
360 inet->rcv_saddr = inet->saddr;
362 if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) {
363 /* Reset inherited state */
364 tp->rx_opt.ts_recent = 0;
365 tp->rx_opt.ts_recent_stamp = 0;
369 if (sysctl_tcp_tw_recycle &&
370 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
371 struct inet_peer *peer = rt_get_peer(rt);
373 /* VJ's idea. We save last timestamp seen from
374 * the destination in peer table, when entering state TIME-WAIT
375 * and initialize rx_opt.ts_recent from it, when trying new connection.
378 if (peer && peer->tcp_ts_stamp + TCP_PAWS_MSL >= xtime.tv_sec) {
379 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
380 tp->rx_opt.ts_recent = peer->tcp_ts;
384 inet->dport = usin->sin_port;
387 tp->ext_header_len = 0;
389 tp->ext_header_len = inet->opt->optlen;
391 tp->rx_opt.mss_clamp = 536;
393 /* Socket identity is still unknown (sport may be zero).
394 * However we set state to SYN-SENT and not releasing socket
395 * lock select source port, enter ourselves into the hash tables and
396 * complete initialization after this.
398 tcp_set_state(sk, TCP_SYN_SENT);
399 err = tcp_v4_hash_connect(sk);
403 err = ip_route_newports(&rt, inet->sport, inet->dport, sk);
407 /* OK, now commit destination to socket. */
408 sk_setup_caps(sk, &rt->u.dst);
411 tp->write_seq = secure_tcp_sequence_number(inet->saddr,
416 inet->id = tp->write_seq ^ jiffies;
418 err = tcp_connect(sk);
426 /* This unhashes the socket and releases the local port, if necessary. */
427 tcp_set_state(sk, TCP_CLOSE);
429 sk->sk_route_caps = 0;
435 * This routine does path mtu discovery as defined in RFC1191.
437 static inline void do_pmtu_discovery(struct sock *sk, struct iphdr *iph,
440 struct dst_entry *dst;
441 struct inet_sock *inet = inet_sk(sk);
442 struct tcp_sock *tp = tcp_sk(sk);
444 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
445 * send out by Linux are always <576bytes so they should go through
448 if (sk->sk_state == TCP_LISTEN)
451 /* We don't check in the destentry if pmtu discovery is forbidden
452 * on this route. We just assume that no packet_to_big packets
453 * are send back when pmtu discovery is not active.
454 * There is a small race when the user changes this flag in the
455 * route, but I think that's acceptable.
457 if ((dst = __sk_dst_check(sk, 0)) == NULL)
460 dst->ops->update_pmtu(dst, mtu);
462 /* Something is about to be wrong... Remember soft error
463 * for the case, if this connection will not able to recover.
465 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
466 sk->sk_err_soft = EMSGSIZE;
470 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
471 tp->pmtu_cookie > mtu) {
472 tcp_sync_mss(sk, mtu);
474 /* Resend the TCP packet because it's
475 * clear that the old packet has been
476 * dropped. This is the new "fast" path mtu
479 tcp_simple_retransmit(sk);
480 } /* else let the usual retransmit timer handle it */
484 * This routine is called by the ICMP module when it gets some
485 * sort of error condition. If err < 0 then the socket should
486 * be closed and the error returned to the user. If err > 0
487 * it's just the icmp type << 8 | icmp code. After adjustment
488 * header points to the first 8 bytes of the tcp header. We need
489 * to find the appropriate port.
491 * The locking strategy used here is very "optimistic". When
492 * someone else accesses the socket the ICMP is just dropped
493 * and for some paths there is no check at all.
494 * A more general error queue to queue errors for later handling
495 * is probably better.
499 void tcp_v4_err(struct sk_buff *skb, u32 info)
501 struct iphdr *iph = (struct iphdr *)skb->data;
502 struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
504 struct inet_sock *inet;
505 int type = skb->h.icmph->type;
506 int code = skb->h.icmph->code;
511 if (skb->len < (iph->ihl << 2) + 8) {
512 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
516 sk = inet_lookup(&tcp_hashinfo, iph->daddr, th->dest, iph->saddr,
517 th->source, inet_iif(skb));
519 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
522 if (sk->sk_state == TCP_TIME_WAIT) {
523 inet_twsk_put((struct inet_timewait_sock *)sk);
528 /* If too many ICMPs get dropped on busy
529 * servers this needs to be solved differently.
531 if (sock_owned_by_user(sk))
532 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
534 if (sk->sk_state == TCP_CLOSE)
538 seq = ntohl(th->seq);
539 if (sk->sk_state != TCP_LISTEN &&
540 !between(seq, tp->snd_una, tp->snd_nxt)) {
541 NET_INC_STATS(LINUX_MIB_OUTOFWINDOWICMPS);
546 case ICMP_SOURCE_QUENCH:
547 /* Just silently ignore these. */
549 case ICMP_PARAMETERPROB:
552 case ICMP_DEST_UNREACH:
553 if (code > NR_ICMP_UNREACH)
556 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
557 if (!sock_owned_by_user(sk))
558 do_pmtu_discovery(sk, iph, info);
562 err = icmp_err_convert[code].errno;
564 case ICMP_TIME_EXCEEDED:
571 switch (sk->sk_state) {
572 struct request_sock *req, **prev;
574 if (sock_owned_by_user(sk))
577 req = inet_csk_search_req(sk, &prev, th->dest,
578 iph->daddr, iph->saddr);
582 /* ICMPs are not backlogged, hence we cannot get
583 an established socket here.
587 if (seq != tcp_rsk(req)->snt_isn) {
588 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
593 * Still in SYN_RECV, just remove it silently.
594 * There is no good way to pass the error to the newly
595 * created socket, and POSIX does not want network
596 * errors returned from accept().
598 inet_csk_reqsk_queue_drop(sk, req, prev);
602 case TCP_SYN_RECV: /* Cannot happen.
603 It can f.e. if SYNs crossed.
605 if (!sock_owned_by_user(sk)) {
606 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
609 sk->sk_error_report(sk);
613 sk->sk_err_soft = err;
618 /* If we've already connected we will keep trying
619 * until we time out, or the user gives up.
621 * rfc1122 4.2.3.9 allows to consider as hard errors
622 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
623 * but it is obsoleted by pmtu discovery).
625 * Note, that in modern internet, where routing is unreliable
626 * and in each dark corner broken firewalls sit, sending random
627 * errors ordered by their masters even this two messages finally lose
628 * their original sense (even Linux sends invalid PORT_UNREACHs)
630 * Now we are in compliance with RFCs.
635 if (!sock_owned_by_user(sk) && inet->recverr) {
637 sk->sk_error_report(sk);
638 } else { /* Only an error on timeout */
639 sk->sk_err_soft = err;
647 /* This routine computes an IPv4 TCP checksum. */
648 void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len,
651 struct inet_sock *inet = inet_sk(sk);
653 if (skb->ip_summed == CHECKSUM_HW) {
654 th->check = ~tcp_v4_check(th, len, inet->saddr, inet->daddr, 0);
655 skb->csum = offsetof(struct tcphdr, check);
657 th->check = tcp_v4_check(th, len, inet->saddr, inet->daddr,
658 csum_partial((char *)th,
665 * This routine will send an RST to the other tcp.
667 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
669 * Answer: if a packet caused RST, it is not for a socket
670 * existing in our system, if it is matched to a socket,
671 * it is just duplicate segment or bug in other side's TCP.
672 * So that we build reply only basing on parameters
673 * arrived with segment.
674 * Exception: precedence violation. We do not implement it in any case.
677 static void tcp_v4_send_reset(struct sk_buff *skb)
679 struct tcphdr *th = skb->h.th;
681 struct ip_reply_arg arg;
683 /* Never send a reset in response to a reset. */
687 if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL)
690 /* Swap the send and the receive. */
691 memset(&rth, 0, sizeof(struct tcphdr));
692 rth.dest = th->source;
693 rth.source = th->dest;
694 rth.doff = sizeof(struct tcphdr) / 4;
698 rth.seq = th->ack_seq;
701 rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
702 skb->len - (th->doff << 2));
705 memset(&arg, 0, sizeof arg);
706 arg.iov[0].iov_base = (unsigned char *)&rth;
707 arg.iov[0].iov_len = sizeof rth;
708 arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr,
709 skb->nh.iph->saddr, /*XXX*/
710 sizeof(struct tcphdr), IPPROTO_TCP, 0);
711 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
713 ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth);
715 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
716 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
719 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
720 outside socket context is ugly, certainly. What can I do?
723 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
726 struct tcphdr *th = skb->h.th;
731 struct ip_reply_arg arg;
733 memset(&rep.th, 0, sizeof(struct tcphdr));
734 memset(&arg, 0, sizeof arg);
736 arg.iov[0].iov_base = (unsigned char *)&rep;
737 arg.iov[0].iov_len = sizeof(rep.th);
739 rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
740 (TCPOPT_TIMESTAMP << 8) |
742 rep.tsopt[1] = htonl(tcp_time_stamp);
743 rep.tsopt[2] = htonl(ts);
744 arg.iov[0].iov_len = sizeof(rep);
747 /* Swap the send and the receive. */
748 rep.th.dest = th->source;
749 rep.th.source = th->dest;
750 rep.th.doff = arg.iov[0].iov_len / 4;
751 rep.th.seq = htonl(seq);
752 rep.th.ack_seq = htonl(ack);
754 rep.th.window = htons(win);
756 arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr,
757 skb->nh.iph->saddr, /*XXX*/
758 arg.iov[0].iov_len, IPPROTO_TCP, 0);
759 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
761 ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len);
763 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
766 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
768 struct inet_timewait_sock *tw = inet_twsk(sk);
769 const struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
771 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
772 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, tcptw->tw_ts_recent);
777 static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
779 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
784 * Send a SYN-ACK after having received an ACK.
785 * This still operates on a request_sock only, not on a big
788 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
789 struct dst_entry *dst)
791 const struct inet_request_sock *ireq = inet_rsk(req);
793 struct sk_buff * skb;
795 /* First, grab a route. */
796 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
799 skb = tcp_make_synack(sk, dst, req);
802 struct tcphdr *th = skb->h.th;
804 th->check = tcp_v4_check(th, skb->len,
807 csum_partial((char *)th, skb->len,
810 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
813 if (err == NET_XMIT_CN)
823 * IPv4 request_sock destructor.
825 static void tcp_v4_reqsk_destructor(struct request_sock *req)
827 if (inet_rsk(req)->opt)
828 kfree(inet_rsk(req)->opt);
831 static inline void syn_flood_warning(struct sk_buff *skb)
833 static unsigned long warntime;
835 if (time_after(jiffies, (warntime + HZ * 60))) {
838 "possible SYN flooding on port %d. Sending cookies.\n",
839 ntohs(skb->h.th->dest));
844 * Save and compile IPv4 options into the request_sock if needed.
846 static inline struct ip_options *tcp_v4_save_options(struct sock *sk,
849 struct ip_options *opt = &(IPCB(skb)->opt);
850 struct ip_options *dopt = NULL;
852 if (opt && opt->optlen) {
853 int opt_size = optlength(opt);
854 dopt = kmalloc(opt_size, GFP_ATOMIC);
856 if (ip_options_echo(dopt, skb)) {
865 struct request_sock_ops tcp_request_sock_ops = {
867 .obj_size = sizeof(struct tcp_request_sock),
868 .rtx_syn_ack = tcp_v4_send_synack,
869 .send_ack = tcp_v4_reqsk_send_ack,
870 .destructor = tcp_v4_reqsk_destructor,
871 .send_reset = tcp_v4_send_reset,
874 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
876 struct inet_request_sock *ireq;
877 struct tcp_options_received tmp_opt;
878 struct request_sock *req;
879 __u32 saddr = skb->nh.iph->saddr;
880 __u32 daddr = skb->nh.iph->daddr;
881 __u32 isn = TCP_SKB_CB(skb)->when;
882 struct dst_entry *dst = NULL;
883 #ifdef CONFIG_SYN_COOKIES
886 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
889 /* Never answer to SYNs send to broadcast or multicast */
890 if (((struct rtable *)skb->dst)->rt_flags &
891 (RTCF_BROADCAST | RTCF_MULTICAST))
894 /* TW buckets are converted to open requests without
895 * limitations, they conserve resources and peer is
896 * evidently real one.
898 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
899 #ifdef CONFIG_SYN_COOKIES
900 if (sysctl_tcp_syncookies) {
907 /* Accept backlog is full. If we have already queued enough
908 * of warm entries in syn queue, drop request. It is better than
909 * clogging syn queue with openreqs with exponentially increasing
912 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
915 req = reqsk_alloc(&tcp_request_sock_ops);
919 tcp_clear_options(&tmp_opt);
920 tmp_opt.mss_clamp = 536;
921 tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss;
923 tcp_parse_options(skb, &tmp_opt, 0);
926 tcp_clear_options(&tmp_opt);
927 tmp_opt.saw_tstamp = 0;
930 if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) {
931 /* Some OSes (unknown ones, but I see them on web server, which
932 * contains information interesting only for windows'
933 * users) do not send their stamp in SYN. It is easy case.
934 * We simply do not advertise TS support.
936 tmp_opt.saw_tstamp = 0;
937 tmp_opt.tstamp_ok = 0;
939 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
941 tcp_openreq_init(req, &tmp_opt, skb);
943 ireq = inet_rsk(req);
944 ireq->loc_addr = daddr;
945 ireq->rmt_addr = saddr;
946 ireq->opt = tcp_v4_save_options(sk, skb);
948 TCP_ECN_create_request(req, skb->h.th);
951 #ifdef CONFIG_SYN_COOKIES
952 syn_flood_warning(skb);
954 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
956 struct inet_peer *peer = NULL;
958 /* VJ's idea. We save last timestamp seen
959 * from the destination in peer table, when entering
960 * state TIME-WAIT, and check against it before
961 * accepting new connection request.
963 * If "isn" is not zero, this request hit alive
964 * timewait bucket, so that all the necessary checks
965 * are made in the function processing timewait state.
967 if (tmp_opt.saw_tstamp &&
968 sysctl_tcp_tw_recycle &&
969 (dst = inet_csk_route_req(sk, req)) != NULL &&
970 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
971 peer->v4daddr == saddr) {
972 if (xtime.tv_sec < peer->tcp_ts_stamp + TCP_PAWS_MSL &&
973 (s32)(peer->tcp_ts - req->ts_recent) >
975 NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED);
980 /* Kill the following clause, if you dislike this way. */
981 else if (!sysctl_tcp_syncookies &&
982 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
983 (sysctl_max_syn_backlog >> 2)) &&
984 (!peer || !peer->tcp_ts_stamp) &&
985 (!dst || !dst_metric(dst, RTAX_RTT))) {
986 /* Without syncookies last quarter of
987 * backlog is filled with destinations,
988 * proven to be alive.
989 * It means that we continue to communicate
990 * to destinations, already remembered
991 * to the moment of synflood.
993 LIMIT_NETDEBUG(printk(KERN_DEBUG "TCP: drop open "
994 "request from %u.%u."
997 ntohs(skb->h.th->source)));
1002 isn = tcp_v4_init_sequence(sk, skb);
1004 tcp_rsk(req)->snt_isn = isn;
1006 if (tcp_v4_send_synack(sk, req, dst))
1012 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1019 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
1025 * The three way handshake has completed - we got a valid synack -
1026 * now create the new socket.
1028 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1029 struct request_sock *req,
1030 struct dst_entry *dst)
1032 struct inet_request_sock *ireq;
1033 struct inet_sock *newinet;
1034 struct tcp_sock *newtp;
1037 if (sk_acceptq_is_full(sk))
1040 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1043 newsk = tcp_create_openreq_child(sk, req, skb);
1047 sk_setup_caps(newsk, dst);
1049 newtp = tcp_sk(newsk);
1050 newinet = inet_sk(newsk);
1051 ireq = inet_rsk(req);
1052 newinet->daddr = ireq->rmt_addr;
1053 newinet->rcv_saddr = ireq->loc_addr;
1054 newinet->saddr = ireq->loc_addr;
1055 newinet->opt = ireq->opt;
1057 newinet->mc_index = inet_iif(skb);
1058 newinet->mc_ttl = skb->nh.iph->ttl;
1059 newtp->ext_header_len = 0;
1061 newtp->ext_header_len = newinet->opt->optlen;
1062 newinet->id = newtp->write_seq ^ jiffies;
1064 tcp_sync_mss(newsk, dst_mtu(dst));
1065 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1066 tcp_initialize_rcv_mss(newsk);
1068 __inet_hash(&tcp_hashinfo, newsk, 0);
1069 __inet_inherit_port(&tcp_hashinfo, sk, newsk);
1074 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1076 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1081 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1083 struct tcphdr *th = skb->h.th;
1084 struct iphdr *iph = skb->nh.iph;
1086 struct request_sock **prev;
1087 /* Find possible connection requests. */
1088 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1089 iph->saddr, iph->daddr);
1091 return tcp_check_req(sk, skb, req, prev);
1093 nsk = __inet_lookup_established(&tcp_hashinfo, skb->nh.iph->saddr,
1094 th->source, skb->nh.iph->daddr,
1095 ntohs(th->dest), inet_iif(skb));
1098 if (nsk->sk_state != TCP_TIME_WAIT) {
1102 inet_twsk_put((struct inet_timewait_sock *)nsk);
1106 #ifdef CONFIG_SYN_COOKIES
1107 if (!th->rst && !th->syn && th->ack)
1108 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1113 static int tcp_v4_checksum_init(struct sk_buff *skb)
1115 if (skb->ip_summed == CHECKSUM_HW) {
1116 skb->ip_summed = CHECKSUM_UNNECESSARY;
1117 if (!tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr,
1118 skb->nh.iph->daddr, skb->csum))
1121 LIMIT_NETDEBUG(printk(KERN_DEBUG "hw tcp v4 csum failed\n"));
1122 skb->ip_summed = CHECKSUM_NONE;
1124 if (skb->len <= 76) {
1125 if (tcp_v4_check(skb->h.th, skb->len, skb->nh.iph->saddr,
1127 skb_checksum(skb, 0, skb->len, 0)))
1129 skb->ip_summed = CHECKSUM_UNNECESSARY;
1131 skb->csum = ~tcp_v4_check(skb->h.th, skb->len,
1133 skb->nh.iph->daddr, 0);
1139 /* The socket must have it's spinlock held when we get
1142 * We have a potential double-lock case here, so even when
1143 * doing backlog processing we use the BH locking scheme.
1144 * This is because we cannot sleep with the original spinlock
1147 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1149 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1150 TCP_CHECK_TIMER(sk);
1151 if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
1153 TCP_CHECK_TIMER(sk);
1157 if (skb->len < (skb->h.th->doff << 2) || tcp_checksum_complete(skb))
1160 if (sk->sk_state == TCP_LISTEN) {
1161 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1166 if (tcp_child_process(sk, nsk, skb))
1172 TCP_CHECK_TIMER(sk);
1173 if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
1175 TCP_CHECK_TIMER(sk);
1179 tcp_v4_send_reset(skb);
1182 /* Be careful here. If this function gets more complicated and
1183 * gcc suffers from register pressure on the x86, sk (in %ebx)
1184 * might be destroyed here. This current version compiles correctly,
1185 * but you have been warned.
1190 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1198 int tcp_v4_rcv(struct sk_buff *skb)
1204 if (skb->pkt_type != PACKET_HOST)
1207 /* Count it even if it's bad */
1208 TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1210 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1215 if (th->doff < sizeof(struct tcphdr) / 4)
1217 if (!pskb_may_pull(skb, th->doff * 4))
1220 /* An explanation is required here, I think.
1221 * Packet length and doff are validated by header prediction,
1222 * provided case of th->doff==0 is elimineted.
1223 * So, we defer the checks. */
1224 if ((skb->ip_summed != CHECKSUM_UNNECESSARY &&
1225 tcp_v4_checksum_init(skb) < 0))
1229 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1230 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1231 skb->len - th->doff * 4);
1232 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1233 TCP_SKB_CB(skb)->when = 0;
1234 TCP_SKB_CB(skb)->flags = skb->nh.iph->tos;
1235 TCP_SKB_CB(skb)->sacked = 0;
1237 sk = __inet_lookup(&tcp_hashinfo, skb->nh.iph->saddr, th->source,
1238 skb->nh.iph->daddr, ntohs(th->dest),
1245 if (sk->sk_state == TCP_TIME_WAIT)
1248 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1249 goto discard_and_relse;
1251 if (sk_filter(sk, skb, 0))
1252 goto discard_and_relse;
1258 if (!sock_owned_by_user(sk)) {
1259 if (!tcp_prequeue(sk, skb))
1260 ret = tcp_v4_do_rcv(sk, skb);
1262 sk_add_backlog(sk, skb);
1270 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1273 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1275 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1277 tcp_v4_send_reset(skb);
1281 /* Discard frame. */
1290 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1291 inet_twsk_put((struct inet_timewait_sock *) sk);
1295 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1296 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1297 inet_twsk_put((struct inet_timewait_sock *) sk);
1300 switch (tcp_timewait_state_process((struct inet_timewait_sock *)sk,
1303 struct sock *sk2 = inet_lookup_listener(&tcp_hashinfo,
1308 tcp_tw_deschedule((struct inet_timewait_sock *)sk);
1309 inet_twsk_put((struct inet_timewait_sock *)sk);
1313 /* Fall through to ACK */
1316 tcp_v4_timewait_ack(sk, skb);
1320 case TCP_TW_SUCCESS:;
1325 static void v4_addr2sockaddr(struct sock *sk, struct sockaddr * uaddr)
1327 struct sockaddr_in *sin = (struct sockaddr_in *) uaddr;
1328 struct inet_sock *inet = inet_sk(sk);
1330 sin->sin_family = AF_INET;
1331 sin->sin_addr.s_addr = inet->daddr;
1332 sin->sin_port = inet->dport;
1335 /* VJ's idea. Save last timestamp seen from this destination
1336 * and hold it at least for normal timewait interval to use for duplicate
1337 * segment detection in subsequent connections, before they enter synchronized
1341 int tcp_v4_remember_stamp(struct sock *sk)
1343 struct inet_sock *inet = inet_sk(sk);
1344 struct tcp_sock *tp = tcp_sk(sk);
1345 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1346 struct inet_peer *peer = NULL;
1349 if (!rt || rt->rt_dst != inet->daddr) {
1350 peer = inet_getpeer(inet->daddr, 1);
1354 rt_bind_peer(rt, 1);
1359 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1360 (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec &&
1361 peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) {
1362 peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp;
1363 peer->tcp_ts = tp->rx_opt.ts_recent;
1373 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1375 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1378 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1380 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1381 (peer->tcp_ts_stamp + TCP_PAWS_MSL < xtime.tv_sec &&
1382 peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) {
1383 peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp;
1384 peer->tcp_ts = tcptw->tw_ts_recent;
1393 struct tcp_func ipv4_specific = {
1394 .queue_xmit = ip_queue_xmit,
1395 .send_check = tcp_v4_send_check,
1396 .rebuild_header = inet_sk_rebuild_header,
1397 .conn_request = tcp_v4_conn_request,
1398 .syn_recv_sock = tcp_v4_syn_recv_sock,
1399 .remember_stamp = tcp_v4_remember_stamp,
1400 .net_header_len = sizeof(struct iphdr),
1401 .setsockopt = ip_setsockopt,
1402 .getsockopt = ip_getsockopt,
1403 .addr2sockaddr = v4_addr2sockaddr,
1404 .sockaddr_len = sizeof(struct sockaddr_in),
1407 /* NOTE: A lot of things set to zero explicitly by call to
1408 * sk_alloc() so need not be done here.
1410 static int tcp_v4_init_sock(struct sock *sk)
1412 struct tcp_sock *tp = tcp_sk(sk);
1414 skb_queue_head_init(&tp->out_of_order_queue);
1415 tcp_init_xmit_timers(sk);
1416 tcp_prequeue_init(tp);
1418 inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
1419 tp->mdev = TCP_TIMEOUT_INIT;
1421 /* So many TCP implementations out there (incorrectly) count the
1422 * initial SYN frame in their delayed-ACK and congestion control
1423 * algorithms that we must have the following bandaid to talk
1424 * efficiently to them. -DaveM
1428 /* See draft-stevens-tcpca-spec-01 for discussion of the
1429 * initialization of these values.
1431 tp->snd_ssthresh = 0x7fffffff; /* Infinity */
1432 tp->snd_cwnd_clamp = ~0;
1433 tp->mss_cache = 536;
1435 tp->reordering = sysctl_tcp_reordering;
1436 tp->ca_ops = &tcp_init_congestion_ops;
1438 sk->sk_state = TCP_CLOSE;
1440 sk->sk_write_space = sk_stream_write_space;
1441 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1443 tp->af_specific = &ipv4_specific;
1445 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1446 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1448 atomic_inc(&tcp_sockets_allocated);
1453 int tcp_v4_destroy_sock(struct sock *sk)
1455 struct tcp_sock *tp = tcp_sk(sk);
1457 tcp_clear_xmit_timers(sk);
1459 tcp_cleanup_congestion_control(tp);
1461 /* Cleanup up the write buffer. */
1462 sk_stream_writequeue_purge(sk);
1464 /* Cleans up our, hopefully empty, out_of_order_queue. */
1465 __skb_queue_purge(&tp->out_of_order_queue);
1467 /* Clean prequeue, it must be empty really */
1468 __skb_queue_purge(&tp->ucopy.prequeue);
1470 /* Clean up a referenced TCP bind bucket. */
1471 if (inet_csk(sk)->icsk_bind_hash)
1472 inet_put_port(&tcp_hashinfo, sk);
1475 * If sendmsg cached page exists, toss it.
1477 if (sk->sk_sndmsg_page) {
1478 __free_page(sk->sk_sndmsg_page);
1479 sk->sk_sndmsg_page = NULL;
1482 atomic_dec(&tcp_sockets_allocated);
1487 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1489 #ifdef CONFIG_PROC_FS
1490 /* Proc filesystem TCP sock list dumping. */
1492 static inline struct inet_timewait_sock *tw_head(struct hlist_head *head)
1494 return hlist_empty(head) ? NULL :
1495 list_entry(head->first, struct inet_timewait_sock, tw_node);
1498 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1500 return tw->tw_node.next ?
1501 hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1504 static void *listening_get_next(struct seq_file *seq, void *cur)
1506 struct inet_connection_sock *icsk;
1507 struct hlist_node *node;
1508 struct sock *sk = cur;
1509 struct tcp_iter_state* st = seq->private;
1513 sk = sk_head(&tcp_hashinfo.listening_hash[0]);
1519 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1520 struct request_sock *req = cur;
1522 icsk = inet_csk(st->syn_wait_sk);
1526 if (req->rsk_ops->family == st->family) {
1532 if (++st->sbucket >= TCP_SYNQ_HSIZE)
1535 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1537 sk = sk_next(st->syn_wait_sk);
1538 st->state = TCP_SEQ_STATE_LISTENING;
1539 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1541 icsk = inet_csk(sk);
1542 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1543 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1545 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1549 sk_for_each_from(sk, node) {
1550 if (sk->sk_family == st->family) {
1554 icsk = inet_csk(sk);
1555 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1556 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1558 st->uid = sock_i_uid(sk);
1559 st->syn_wait_sk = sk;
1560 st->state = TCP_SEQ_STATE_OPENREQ;
1564 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1566 if (++st->bucket < INET_LHTABLE_SIZE) {
1567 sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]);
1575 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1577 void *rc = listening_get_next(seq, NULL);
1579 while (rc && *pos) {
1580 rc = listening_get_next(seq, rc);
1586 static void *established_get_first(struct seq_file *seq)
1588 struct tcp_iter_state* st = seq->private;
1591 for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
1593 struct hlist_node *node;
1594 struct inet_timewait_sock *tw;
1596 /* We can reschedule _before_ having picked the target: */
1597 cond_resched_softirq();
1599 read_lock(&tcp_hashinfo.ehash[st->bucket].lock);
1600 sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1601 if (sk->sk_family != st->family) {
1607 st->state = TCP_SEQ_STATE_TIME_WAIT;
1608 inet_twsk_for_each(tw, node,
1609 &tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain) {
1610 if (tw->tw_family != st->family) {
1616 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1617 st->state = TCP_SEQ_STATE_ESTABLISHED;
1623 static void *established_get_next(struct seq_file *seq, void *cur)
1625 struct sock *sk = cur;
1626 struct inet_timewait_sock *tw;
1627 struct hlist_node *node;
1628 struct tcp_iter_state* st = seq->private;
1632 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
1636 while (tw && tw->tw_family != st->family) {
1643 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1644 st->state = TCP_SEQ_STATE_ESTABLISHED;
1646 /* We can reschedule between buckets: */
1647 cond_resched_softirq();
1649 if (++st->bucket < tcp_hashinfo.ehash_size) {
1650 read_lock(&tcp_hashinfo.ehash[st->bucket].lock);
1651 sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain);
1659 sk_for_each_from(sk, node) {
1660 if (sk->sk_family == st->family)
1664 st->state = TCP_SEQ_STATE_TIME_WAIT;
1665 tw = tw_head(&tcp_hashinfo.ehash[st->bucket + tcp_hashinfo.ehash_size].chain);
1673 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1675 void *rc = established_get_first(seq);
1678 rc = established_get_next(seq, rc);
1684 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1687 struct tcp_iter_state* st = seq->private;
1689 inet_listen_lock(&tcp_hashinfo);
1690 st->state = TCP_SEQ_STATE_LISTENING;
1691 rc = listening_get_idx(seq, &pos);
1694 inet_listen_unlock(&tcp_hashinfo);
1696 st->state = TCP_SEQ_STATE_ESTABLISHED;
1697 rc = established_get_idx(seq, pos);
1703 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
1705 struct tcp_iter_state* st = seq->private;
1706 st->state = TCP_SEQ_STATE_LISTENING;
1708 return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1711 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1714 struct tcp_iter_state* st;
1716 if (v == SEQ_START_TOKEN) {
1717 rc = tcp_get_idx(seq, 0);
1722 switch (st->state) {
1723 case TCP_SEQ_STATE_OPENREQ:
1724 case TCP_SEQ_STATE_LISTENING:
1725 rc = listening_get_next(seq, v);
1727 inet_listen_unlock(&tcp_hashinfo);
1729 st->state = TCP_SEQ_STATE_ESTABLISHED;
1730 rc = established_get_first(seq);
1733 case TCP_SEQ_STATE_ESTABLISHED:
1734 case TCP_SEQ_STATE_TIME_WAIT:
1735 rc = established_get_next(seq, v);
1743 static void tcp_seq_stop(struct seq_file *seq, void *v)
1745 struct tcp_iter_state* st = seq->private;
1747 switch (st->state) {
1748 case TCP_SEQ_STATE_OPENREQ:
1750 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
1751 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1753 case TCP_SEQ_STATE_LISTENING:
1754 if (v != SEQ_START_TOKEN)
1755 inet_listen_unlock(&tcp_hashinfo);
1757 case TCP_SEQ_STATE_TIME_WAIT:
1758 case TCP_SEQ_STATE_ESTABLISHED:
1760 read_unlock(&tcp_hashinfo.ehash[st->bucket].lock);
1766 static int tcp_seq_open(struct inode *inode, struct file *file)
1768 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
1769 struct seq_file *seq;
1770 struct tcp_iter_state *s;
1773 if (unlikely(afinfo == NULL))
1776 s = kmalloc(sizeof(*s), GFP_KERNEL);
1779 memset(s, 0, sizeof(*s));
1780 s->family = afinfo->family;
1781 s->seq_ops.start = tcp_seq_start;
1782 s->seq_ops.next = tcp_seq_next;
1783 s->seq_ops.show = afinfo->seq_show;
1784 s->seq_ops.stop = tcp_seq_stop;
1786 rc = seq_open(file, &s->seq_ops);
1789 seq = file->private_data;
1798 int tcp_proc_register(struct tcp_seq_afinfo *afinfo)
1801 struct proc_dir_entry *p;
1805 afinfo->seq_fops->owner = afinfo->owner;
1806 afinfo->seq_fops->open = tcp_seq_open;
1807 afinfo->seq_fops->read = seq_read;
1808 afinfo->seq_fops->llseek = seq_lseek;
1809 afinfo->seq_fops->release = seq_release_private;
1811 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1819 void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo)
1823 proc_net_remove(afinfo->name);
1824 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1827 static void get_openreq4(struct sock *sk, struct request_sock *req,
1828 char *tmpbuf, int i, int uid)
1830 const struct inet_request_sock *ireq = inet_rsk(req);
1831 int ttd = req->expires - jiffies;
1833 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1834 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p",
1837 ntohs(inet_sk(sk)->sport),
1839 ntohs(ireq->rmt_port),
1841 0, 0, /* could print option size, but that is af dependent. */
1842 1, /* timers active (only the expire timer) */
1843 jiffies_to_clock_t(ttd),
1846 0, /* non standard timer */
1847 0, /* open_requests have no inode */
1848 atomic_read(&sk->sk_refcnt),
1852 static void get_tcp4_sock(struct sock *sp, char *tmpbuf, int i)
1855 unsigned long timer_expires;
1856 struct tcp_sock *tp = tcp_sk(sp);
1857 const struct inet_connection_sock *icsk = inet_csk(sp);
1858 struct inet_sock *inet = inet_sk(sp);
1859 unsigned int dest = inet->daddr;
1860 unsigned int src = inet->rcv_saddr;
1861 __u16 destp = ntohs(inet->dport);
1862 __u16 srcp = ntohs(inet->sport);
1864 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1866 timer_expires = icsk->icsk_timeout;
1867 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1869 timer_expires = icsk->icsk_timeout;
1870 } else if (timer_pending(&sp->sk_timer)) {
1872 timer_expires = sp->sk_timer.expires;
1875 timer_expires = jiffies;
1878 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
1879 "%08X %5d %8d %lu %d %p %u %u %u %u %d",
1880 i, src, srcp, dest, destp, sp->sk_state,
1881 tp->write_seq - tp->snd_una, tp->rcv_nxt - tp->copied_seq,
1883 jiffies_to_clock_t(timer_expires - jiffies),
1884 icsk->icsk_retransmits,
1888 atomic_read(&sp->sk_refcnt), sp,
1891 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
1893 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh);
1896 static void get_timewait4_sock(struct inet_timewait_sock *tw, char *tmpbuf, int i)
1898 unsigned int dest, src;
1900 int ttd = tw->tw_ttd - jiffies;
1905 dest = tw->tw_daddr;
1906 src = tw->tw_rcv_saddr;
1907 destp = ntohs(tw->tw_dport);
1908 srcp = ntohs(tw->tw_sport);
1910 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1911 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p",
1912 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
1913 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
1914 atomic_read(&tw->tw_refcnt), tw);
1919 static int tcp4_seq_show(struct seq_file *seq, void *v)
1921 struct tcp_iter_state* st;
1922 char tmpbuf[TMPSZ + 1];
1924 if (v == SEQ_START_TOKEN) {
1925 seq_printf(seq, "%-*s\n", TMPSZ - 1,
1926 " sl local_address rem_address st tx_queue "
1927 "rx_queue tr tm->when retrnsmt uid timeout "
1933 switch (st->state) {
1934 case TCP_SEQ_STATE_LISTENING:
1935 case TCP_SEQ_STATE_ESTABLISHED:
1936 get_tcp4_sock(v, tmpbuf, st->num);
1938 case TCP_SEQ_STATE_OPENREQ:
1939 get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid);
1941 case TCP_SEQ_STATE_TIME_WAIT:
1942 get_timewait4_sock(v, tmpbuf, st->num);
1945 seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf);
1950 static struct file_operations tcp4_seq_fops;
1951 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
1952 .owner = THIS_MODULE,
1955 .seq_show = tcp4_seq_show,
1956 .seq_fops = &tcp4_seq_fops,
1959 int __init tcp4_proc_init(void)
1961 return tcp_proc_register(&tcp4_seq_afinfo);
1964 void tcp4_proc_exit(void)
1966 tcp_proc_unregister(&tcp4_seq_afinfo);
1968 #endif /* CONFIG_PROC_FS */
1970 struct proto tcp_prot = {
1972 .owner = THIS_MODULE,
1974 .connect = tcp_v4_connect,
1975 .disconnect = tcp_disconnect,
1976 .accept = inet_csk_accept,
1978 .init = tcp_v4_init_sock,
1979 .destroy = tcp_v4_destroy_sock,
1980 .shutdown = tcp_shutdown,
1981 .setsockopt = tcp_setsockopt,
1982 .getsockopt = tcp_getsockopt,
1983 .sendmsg = tcp_sendmsg,
1984 .recvmsg = tcp_recvmsg,
1985 .backlog_rcv = tcp_v4_do_rcv,
1986 .hash = tcp_v4_hash,
1987 .unhash = tcp_unhash,
1988 .get_port = tcp_v4_get_port,
1989 .enter_memory_pressure = tcp_enter_memory_pressure,
1990 .sockets_allocated = &tcp_sockets_allocated,
1991 .orphan_count = &tcp_orphan_count,
1992 .memory_allocated = &tcp_memory_allocated,
1993 .memory_pressure = &tcp_memory_pressure,
1994 .sysctl_mem = sysctl_tcp_mem,
1995 .sysctl_wmem = sysctl_tcp_wmem,
1996 .sysctl_rmem = sysctl_tcp_rmem,
1997 .max_header = MAX_TCP_HEADER,
1998 .obj_size = sizeof(struct tcp_sock),
1999 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
2000 .rsk_prot = &tcp_request_sock_ops,
2005 void __init tcp_v4_init(struct net_proto_family *ops)
2007 int err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_TCP, &tcp_socket);
2009 panic("Failed to create the TCP control socket.\n");
2010 tcp_socket->sk->sk_allocation = GFP_ATOMIC;
2011 inet_sk(tcp_socket->sk)->uc_ttl = -1;
2013 /* Unhash it so that IP input processing does not even
2014 * see it, we do not wish this socket to see incoming
2017 tcp_socket->sk->sk_prot->unhash(tcp_socket->sk);
2020 EXPORT_SYMBOL(ipv4_specific);
2021 EXPORT_SYMBOL(inet_bind_bucket_create);
2022 EXPORT_SYMBOL(tcp_hashinfo);
2023 EXPORT_SYMBOL(tcp_prot);
2024 EXPORT_SYMBOL(tcp_unhash);
2025 EXPORT_SYMBOL(tcp_v4_conn_request);
2026 EXPORT_SYMBOL(tcp_v4_connect);
2027 EXPORT_SYMBOL(tcp_v4_do_rcv);
2028 EXPORT_SYMBOL(tcp_v4_remember_stamp);
2029 EXPORT_SYMBOL(tcp_v4_send_check);
2030 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2032 #ifdef CONFIG_PROC_FS
2033 EXPORT_SYMBOL(tcp_proc_register);
2034 EXPORT_SYMBOL(tcp_proc_unregister);
2036 EXPORT_SYMBOL(sysctl_local_port_range);
2037 EXPORT_SYMBOL(sysctl_tcp_low_latency);
2038 EXPORT_SYMBOL(sysctl_tcp_tw_reuse);