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>
63 #include <linux/slab.h>
65 #include <net/net_namespace.h>
67 #include <net/inet_hashtables.h>
69 #include <net/transp_v6.h>
71 #include <net/inet_common.h>
72 #include <net/timewait_sock.h>
74 #include <net/netdma.h>
76 #include <linux/inet.h>
77 #include <linux/ipv6.h>
78 #include <linux/stddef.h>
79 #include <linux/proc_fs.h>
80 #include <linux/seq_file.h>
82 #include <linux/crypto.h>
83 #include <linux/scatterlist.h>
85 int sysctl_tcp_tw_reuse __read_mostly;
86 int sysctl_tcp_low_latency __read_mostly;
89 #ifdef CONFIG_TCP_MD5SIG
90 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
92 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
93 __be32 daddr, __be32 saddr, struct tcphdr *th);
96 struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
102 struct inet_hashinfo tcp_hashinfo;
104 static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
106 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
109 tcp_hdr(skb)->source);
112 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
114 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
115 struct tcp_sock *tp = tcp_sk(sk);
117 /* With PAWS, it is safe from the viewpoint
118 of data integrity. Even without PAWS it is safe provided sequence
119 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
121 Actually, the idea is close to VJ's one, only timestamp cache is
122 held not per host, but per port pair and TW bucket is used as state
125 If TW bucket has been already destroyed we fall back to VJ's scheme
126 and use initial timestamp retrieved from peer table.
128 if (tcptw->tw_ts_recent_stamp &&
129 (twp == NULL || (sysctl_tcp_tw_reuse &&
130 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
131 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
132 if (tp->write_seq == 0)
134 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
135 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
143 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
145 /* This will initiate an outgoing connection. */
146 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
148 struct inet_sock *inet = inet_sk(sk);
149 struct tcp_sock *tp = tcp_sk(sk);
150 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
152 __be32 daddr, nexthop;
156 if (addr_len < sizeof(struct sockaddr_in))
159 if (usin->sin_family != AF_INET)
160 return -EAFNOSUPPORT;
162 nexthop = daddr = usin->sin_addr.s_addr;
163 if (inet->opt && inet->opt->srr) {
166 nexthop = inet->opt->faddr;
169 tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
170 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
172 inet->inet_sport, usin->sin_port, sk, 1);
174 if (tmp == -ENETUNREACH)
175 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
179 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
184 if (!inet->opt || !inet->opt->srr)
187 if (!inet->inet_saddr)
188 inet->inet_saddr = rt->rt_src;
189 inet->inet_rcv_saddr = inet->inet_saddr;
191 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
192 /* Reset inherited state */
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
198 if (tcp_death_row.sysctl_tw_recycle &&
199 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
200 struct inet_peer *peer = rt_get_peer(rt);
202 * VJ's idea. We save last timestamp seen from
203 * the destination in peer table, when entering state
204 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
205 * when trying new connection.
208 inet_peer_refcheck(peer);
209 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
210 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
211 tp->rx_opt.ts_recent = peer->tcp_ts;
216 inet->inet_dport = usin->sin_port;
217 inet->inet_daddr = daddr;
219 inet_csk(sk)->icsk_ext_hdr_len = 0;
221 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
223 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
225 /* Socket identity is still unknown (sport may be zero).
226 * However we set state to SYN-SENT and not releasing socket
227 * lock select source port, enter ourselves into the hash tables and
228 * complete initialization after this.
230 tcp_set_state(sk, TCP_SYN_SENT);
231 err = inet_hash_connect(&tcp_death_row, sk);
235 err = ip_route_newports(&rt, IPPROTO_TCP,
236 inet->inet_sport, inet->inet_dport, sk);
240 /* OK, now commit destination to socket. */
241 sk->sk_gso_type = SKB_GSO_TCPV4;
242 sk_setup_caps(sk, &rt->dst);
245 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
250 inet->inet_id = tp->write_seq ^ jiffies;
252 err = tcp_connect(sk);
261 * This unhashes the socket and releases the local port,
264 tcp_set_state(sk, TCP_CLOSE);
266 sk->sk_route_caps = 0;
267 inet->inet_dport = 0;
272 * This routine does path mtu discovery as defined in RFC1191.
274 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
276 struct dst_entry *dst;
277 struct inet_sock *inet = inet_sk(sk);
279 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
280 * send out by Linux are always <576bytes so they should go through
283 if (sk->sk_state == TCP_LISTEN)
286 /* We don't check in the destentry if pmtu discovery is forbidden
287 * on this route. We just assume that no packet_to_big packets
288 * are send back when pmtu discovery is not active.
289 * There is a small race when the user changes this flag in the
290 * route, but I think that's acceptable.
292 if ((dst = __sk_dst_check(sk, 0)) == NULL)
295 dst->ops->update_pmtu(dst, mtu);
297 /* Something is about to be wrong... Remember soft error
298 * for the case, if this connection will not able to recover.
300 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
301 sk->sk_err_soft = EMSGSIZE;
305 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
306 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
307 tcp_sync_mss(sk, mtu);
309 /* Resend the TCP packet because it's
310 * clear that the old packet has been
311 * dropped. This is the new "fast" path mtu
314 tcp_simple_retransmit(sk);
315 } /* else let the usual retransmit timer handle it */
319 * This routine is called by the ICMP module when it gets some
320 * sort of error condition. If err < 0 then the socket should
321 * be closed and the error returned to the user. If err > 0
322 * it's just the icmp type << 8 | icmp code. After adjustment
323 * header points to the first 8 bytes of the tcp header. We need
324 * to find the appropriate port.
326 * The locking strategy used here is very "optimistic". When
327 * someone else accesses the socket the ICMP is just dropped
328 * and for some paths there is no check at all.
329 * A more general error queue to queue errors for later handling
330 * is probably better.
334 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
336 struct iphdr *iph = (struct iphdr *)icmp_skb->data;
337 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
338 struct inet_connection_sock *icsk;
340 struct inet_sock *inet;
341 const int type = icmp_hdr(icmp_skb)->type;
342 const int code = icmp_hdr(icmp_skb)->code;
348 struct net *net = dev_net(icmp_skb->dev);
350 if (icmp_skb->len < (iph->ihl << 2) + 8) {
351 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
355 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
356 iph->saddr, th->source, inet_iif(icmp_skb));
358 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
361 if (sk->sk_state == TCP_TIME_WAIT) {
362 inet_twsk_put(inet_twsk(sk));
367 /* If too many ICMPs get dropped on busy
368 * servers this needs to be solved differently.
370 if (sock_owned_by_user(sk))
371 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
373 if (sk->sk_state == TCP_CLOSE)
376 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
377 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
383 seq = ntohl(th->seq);
384 if (sk->sk_state != TCP_LISTEN &&
385 !between(seq, tp->snd_una, tp->snd_nxt)) {
386 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
391 case ICMP_SOURCE_QUENCH:
392 /* Just silently ignore these. */
394 case ICMP_PARAMETERPROB:
397 case ICMP_DEST_UNREACH:
398 if (code > NR_ICMP_UNREACH)
401 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
402 if (!sock_owned_by_user(sk))
403 do_pmtu_discovery(sk, iph, info);
407 err = icmp_err_convert[code].errno;
408 /* check if icmp_skb allows revert of backoff
409 * (see draft-zimmermann-tcp-lcd) */
410 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
412 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
416 icsk->icsk_backoff--;
417 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
421 skb = tcp_write_queue_head(sk);
424 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
425 tcp_time_stamp - TCP_SKB_CB(skb)->when);
428 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
429 remaining, TCP_RTO_MAX);
430 } else if (sock_owned_by_user(sk)) {
431 /* RTO revert clocked out retransmission,
432 * but socket is locked. Will defer. */
433 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
436 /* RTO revert clocked out retransmission.
437 * Will retransmit now */
438 tcp_retransmit_timer(sk);
442 case ICMP_TIME_EXCEEDED:
449 switch (sk->sk_state) {
450 struct request_sock *req, **prev;
452 if (sock_owned_by_user(sk))
455 req = inet_csk_search_req(sk, &prev, th->dest,
456 iph->daddr, iph->saddr);
460 /* ICMPs are not backlogged, hence we cannot get
461 an established socket here.
465 if (seq != tcp_rsk(req)->snt_isn) {
466 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
471 * Still in SYN_RECV, just remove it silently.
472 * There is no good way to pass the error to the newly
473 * created socket, and POSIX does not want network
474 * errors returned from accept().
476 inet_csk_reqsk_queue_drop(sk, req, prev);
480 case TCP_SYN_RECV: /* Cannot happen.
481 It can f.e. if SYNs crossed.
483 if (!sock_owned_by_user(sk)) {
486 sk->sk_error_report(sk);
490 sk->sk_err_soft = err;
495 /* If we've already connected we will keep trying
496 * until we time out, or the user gives up.
498 * rfc1122 4.2.3.9 allows to consider as hard errors
499 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
500 * but it is obsoleted by pmtu discovery).
502 * Note, that in modern internet, where routing is unreliable
503 * and in each dark corner broken firewalls sit, sending random
504 * errors ordered by their masters even this two messages finally lose
505 * their original sense (even Linux sends invalid PORT_UNREACHs)
507 * Now we are in compliance with RFCs.
512 if (!sock_owned_by_user(sk) && inet->recverr) {
514 sk->sk_error_report(sk);
515 } else { /* Only an error on timeout */
516 sk->sk_err_soft = err;
524 static void __tcp_v4_send_check(struct sk_buff *skb,
525 __be32 saddr, __be32 daddr)
527 struct tcphdr *th = tcp_hdr(skb);
529 if (skb->ip_summed == CHECKSUM_PARTIAL) {
530 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
531 skb->csum_start = skb_transport_header(skb) - skb->head;
532 skb->csum_offset = offsetof(struct tcphdr, check);
534 th->check = tcp_v4_check(skb->len, saddr, daddr,
541 /* This routine computes an IPv4 TCP checksum. */
542 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
544 struct inet_sock *inet = inet_sk(sk);
546 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
549 int tcp_v4_gso_send_check(struct sk_buff *skb)
551 const struct iphdr *iph;
554 if (!pskb_may_pull(skb, sizeof(*th)))
561 skb->ip_summed = CHECKSUM_PARTIAL;
562 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
567 * This routine will send an RST to the other tcp.
569 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
571 * Answer: if a packet caused RST, it is not for a socket
572 * existing in our system, if it is matched to a socket,
573 * it is just duplicate segment or bug in other side's TCP.
574 * So that we build reply only basing on parameters
575 * arrived with segment.
576 * Exception: precedence violation. We do not implement it in any case.
579 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
581 struct tcphdr *th = tcp_hdr(skb);
584 #ifdef CONFIG_TCP_MD5SIG
585 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
588 struct ip_reply_arg arg;
589 #ifdef CONFIG_TCP_MD5SIG
590 struct tcp_md5sig_key *key;
594 /* Never send a reset in response to a reset. */
598 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
601 /* Swap the send and the receive. */
602 memset(&rep, 0, sizeof(rep));
603 rep.th.dest = th->source;
604 rep.th.source = th->dest;
605 rep.th.doff = sizeof(struct tcphdr) / 4;
609 rep.th.seq = th->ack_seq;
612 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
613 skb->len - (th->doff << 2));
616 memset(&arg, 0, sizeof(arg));
617 arg.iov[0].iov_base = (unsigned char *)&rep;
618 arg.iov[0].iov_len = sizeof(rep.th);
620 #ifdef CONFIG_TCP_MD5SIG
621 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
623 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
625 (TCPOPT_MD5SIG << 8) |
627 /* Update length and the length the header thinks exists */
628 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
629 rep.th.doff = arg.iov[0].iov_len / 4;
631 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
632 key, ip_hdr(skb)->saddr,
633 ip_hdr(skb)->daddr, &rep.th);
636 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
637 ip_hdr(skb)->saddr, /* XXX */
638 arg.iov[0].iov_len, IPPROTO_TCP, 0);
639 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
640 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
642 net = dev_net(skb_dst(skb)->dev);
643 ip_send_reply(net->ipv4.tcp_sock, skb,
644 &arg, arg.iov[0].iov_len);
646 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
647 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
650 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
651 outside socket context is ugly, certainly. What can I do?
654 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
655 u32 win, u32 ts, int oif,
656 struct tcp_md5sig_key *key,
659 struct tcphdr *th = tcp_hdr(skb);
662 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
663 #ifdef CONFIG_TCP_MD5SIG
664 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
668 struct ip_reply_arg arg;
669 struct net *net = dev_net(skb_dst(skb)->dev);
671 memset(&rep.th, 0, sizeof(struct tcphdr));
672 memset(&arg, 0, sizeof(arg));
674 arg.iov[0].iov_base = (unsigned char *)&rep;
675 arg.iov[0].iov_len = sizeof(rep.th);
677 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
678 (TCPOPT_TIMESTAMP << 8) |
680 rep.opt[1] = htonl(tcp_time_stamp);
681 rep.opt[2] = htonl(ts);
682 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
685 /* Swap the send and the receive. */
686 rep.th.dest = th->source;
687 rep.th.source = th->dest;
688 rep.th.doff = arg.iov[0].iov_len / 4;
689 rep.th.seq = htonl(seq);
690 rep.th.ack_seq = htonl(ack);
692 rep.th.window = htons(win);
694 #ifdef CONFIG_TCP_MD5SIG
696 int offset = (ts) ? 3 : 0;
698 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
700 (TCPOPT_MD5SIG << 8) |
702 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
703 rep.th.doff = arg.iov[0].iov_len/4;
705 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
706 key, ip_hdr(skb)->saddr,
707 ip_hdr(skb)->daddr, &rep.th);
710 arg.flags = reply_flags;
711 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
712 ip_hdr(skb)->saddr, /* XXX */
713 arg.iov[0].iov_len, IPPROTO_TCP, 0);
714 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
716 arg.bound_dev_if = oif;
718 ip_send_reply(net->ipv4.tcp_sock, skb,
719 &arg, arg.iov[0].iov_len);
721 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
724 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
726 struct inet_timewait_sock *tw = inet_twsk(sk);
727 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
729 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
730 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
733 tcp_twsk_md5_key(tcptw),
734 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
740 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
741 struct request_sock *req)
743 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
744 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
747 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
748 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
752 * Send a SYN-ACK after having received a SYN.
753 * This still operates on a request_sock only, not on a big
756 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
757 struct request_sock *req,
758 struct request_values *rvp)
760 const struct inet_request_sock *ireq = inet_rsk(req);
762 struct sk_buff * skb;
764 /* First, grab a route. */
765 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
768 skb = tcp_make_synack(sk, dst, req, rvp);
771 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
773 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
776 err = net_xmit_eval(err);
783 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
784 struct request_values *rvp)
786 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
787 return tcp_v4_send_synack(sk, NULL, req, rvp);
791 * IPv4 request_sock destructor.
793 static void tcp_v4_reqsk_destructor(struct request_sock *req)
795 kfree(inet_rsk(req)->opt);
798 static void syn_flood_warning(const struct sk_buff *skb)
802 #ifdef CONFIG_SYN_COOKIES
803 if (sysctl_tcp_syncookies)
804 msg = "Sending cookies";
807 msg = "Dropping request";
809 pr_info("TCP: Possible SYN flooding on port %d. %s.\n",
810 ntohs(tcp_hdr(skb)->dest), msg);
814 * Save and compile IPv4 options into the request_sock if needed.
816 static struct ip_options *tcp_v4_save_options(struct sock *sk,
819 struct ip_options *opt = &(IPCB(skb)->opt);
820 struct ip_options *dopt = NULL;
822 if (opt && opt->optlen) {
823 int opt_size = optlength(opt);
824 dopt = kmalloc(opt_size, GFP_ATOMIC);
826 if (ip_options_echo(dopt, skb)) {
835 #ifdef CONFIG_TCP_MD5SIG
837 * RFC2385 MD5 checksumming requires a mapping of
838 * IP address->MD5 Key.
839 * We need to maintain these in the sk structure.
842 /* Find the Key structure for an address. */
843 static struct tcp_md5sig_key *
844 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
846 struct tcp_sock *tp = tcp_sk(sk);
849 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
851 for (i = 0; i < tp->md5sig_info->entries4; i++) {
852 if (tp->md5sig_info->keys4[i].addr == addr)
853 return &tp->md5sig_info->keys4[i].base;
858 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
859 struct sock *addr_sk)
861 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
864 EXPORT_SYMBOL(tcp_v4_md5_lookup);
866 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
867 struct request_sock *req)
869 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
872 /* This can be called on a newly created socket, from other files */
873 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
874 u8 *newkey, u8 newkeylen)
876 /* Add Key to the list */
877 struct tcp_md5sig_key *key;
878 struct tcp_sock *tp = tcp_sk(sk);
879 struct tcp4_md5sig_key *keys;
881 key = tcp_v4_md5_do_lookup(sk, addr);
883 /* Pre-existing entry - just update that one. */
886 key->keylen = newkeylen;
888 struct tcp_md5sig_info *md5sig;
890 if (!tp->md5sig_info) {
891 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
893 if (!tp->md5sig_info) {
897 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
899 if (tcp_alloc_md5sig_pool(sk) == NULL) {
903 md5sig = tp->md5sig_info;
905 if (md5sig->alloced4 == md5sig->entries4) {
906 keys = kmalloc((sizeof(*keys) *
907 (md5sig->entries4 + 1)), GFP_ATOMIC);
910 tcp_free_md5sig_pool();
914 if (md5sig->entries4)
915 memcpy(keys, md5sig->keys4,
916 sizeof(*keys) * md5sig->entries4);
918 /* Free old key list, and reference new one */
919 kfree(md5sig->keys4);
920 md5sig->keys4 = keys;
924 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
925 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
926 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
931 EXPORT_SYMBOL(tcp_v4_md5_do_add);
933 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
934 u8 *newkey, u8 newkeylen)
936 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
940 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
942 struct tcp_sock *tp = tcp_sk(sk);
945 for (i = 0; i < tp->md5sig_info->entries4; i++) {
946 if (tp->md5sig_info->keys4[i].addr == addr) {
948 kfree(tp->md5sig_info->keys4[i].base.key);
949 tp->md5sig_info->entries4--;
951 if (tp->md5sig_info->entries4 == 0) {
952 kfree(tp->md5sig_info->keys4);
953 tp->md5sig_info->keys4 = NULL;
954 tp->md5sig_info->alloced4 = 0;
955 } else if (tp->md5sig_info->entries4 != i) {
956 /* Need to do some manipulation */
957 memmove(&tp->md5sig_info->keys4[i],
958 &tp->md5sig_info->keys4[i+1],
959 (tp->md5sig_info->entries4 - i) *
960 sizeof(struct tcp4_md5sig_key));
962 tcp_free_md5sig_pool();
969 EXPORT_SYMBOL(tcp_v4_md5_do_del);
971 static void tcp_v4_clear_md5_list(struct sock *sk)
973 struct tcp_sock *tp = tcp_sk(sk);
975 /* Free each key, then the set of key keys,
976 * the crypto element, and then decrement our
977 * hold on the last resort crypto.
979 if (tp->md5sig_info->entries4) {
981 for (i = 0; i < tp->md5sig_info->entries4; i++)
982 kfree(tp->md5sig_info->keys4[i].base.key);
983 tp->md5sig_info->entries4 = 0;
984 tcp_free_md5sig_pool();
986 if (tp->md5sig_info->keys4) {
987 kfree(tp->md5sig_info->keys4);
988 tp->md5sig_info->keys4 = NULL;
989 tp->md5sig_info->alloced4 = 0;
993 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
996 struct tcp_md5sig cmd;
997 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1000 if (optlen < sizeof(cmd))
1003 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1006 if (sin->sin_family != AF_INET)
1009 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1010 if (!tcp_sk(sk)->md5sig_info)
1012 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1015 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1018 if (!tcp_sk(sk)->md5sig_info) {
1019 struct tcp_sock *tp = tcp_sk(sk);
1020 struct tcp_md5sig_info *p;
1022 p = kzalloc(sizeof(*p), sk->sk_allocation);
1026 tp->md5sig_info = p;
1027 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1030 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1033 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1034 newkey, cmd.tcpm_keylen);
1037 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1038 __be32 daddr, __be32 saddr, int nbytes)
1040 struct tcp4_pseudohdr *bp;
1041 struct scatterlist sg;
1043 bp = &hp->md5_blk.ip4;
1046 * 1. the TCP pseudo-header (in the order: source IP address,
1047 * destination IP address, zero-padded protocol number, and
1053 bp->protocol = IPPROTO_TCP;
1054 bp->len = cpu_to_be16(nbytes);
1056 sg_init_one(&sg, bp, sizeof(*bp));
1057 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1060 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1061 __be32 daddr, __be32 saddr, struct tcphdr *th)
1063 struct tcp_md5sig_pool *hp;
1064 struct hash_desc *desc;
1066 hp = tcp_get_md5sig_pool();
1068 goto clear_hash_noput;
1069 desc = &hp->md5_desc;
1071 if (crypto_hash_init(desc))
1073 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1075 if (tcp_md5_hash_header(hp, th))
1077 if (tcp_md5_hash_key(hp, key))
1079 if (crypto_hash_final(desc, md5_hash))
1082 tcp_put_md5sig_pool();
1086 tcp_put_md5sig_pool();
1088 memset(md5_hash, 0, 16);
1092 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1093 struct sock *sk, struct request_sock *req,
1094 struct sk_buff *skb)
1096 struct tcp_md5sig_pool *hp;
1097 struct hash_desc *desc;
1098 struct tcphdr *th = tcp_hdr(skb);
1099 __be32 saddr, daddr;
1102 saddr = inet_sk(sk)->inet_saddr;
1103 daddr = inet_sk(sk)->inet_daddr;
1105 saddr = inet_rsk(req)->loc_addr;
1106 daddr = inet_rsk(req)->rmt_addr;
1108 const struct iphdr *iph = ip_hdr(skb);
1113 hp = tcp_get_md5sig_pool();
1115 goto clear_hash_noput;
1116 desc = &hp->md5_desc;
1118 if (crypto_hash_init(desc))
1121 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1123 if (tcp_md5_hash_header(hp, th))
1125 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1127 if (tcp_md5_hash_key(hp, key))
1129 if (crypto_hash_final(desc, md5_hash))
1132 tcp_put_md5sig_pool();
1136 tcp_put_md5sig_pool();
1138 memset(md5_hash, 0, 16);
1142 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1144 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1147 * This gets called for each TCP segment that arrives
1148 * so we want to be efficient.
1149 * We have 3 drop cases:
1150 * o No MD5 hash and one expected.
1151 * o MD5 hash and we're not expecting one.
1152 * o MD5 hash and its wrong.
1154 __u8 *hash_location = NULL;
1155 struct tcp_md5sig_key *hash_expected;
1156 const struct iphdr *iph = ip_hdr(skb);
1157 struct tcphdr *th = tcp_hdr(skb);
1159 unsigned char newhash[16];
1161 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1162 hash_location = tcp_parse_md5sig_option(th);
1164 /* We've parsed the options - do we have a hash? */
1165 if (!hash_expected && !hash_location)
1168 if (hash_expected && !hash_location) {
1169 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1173 if (!hash_expected && hash_location) {
1174 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1178 /* Okay, so this is hash_expected and hash_location -
1179 * so we need to calculate the checksum.
1181 genhash = tcp_v4_md5_hash_skb(newhash,
1185 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1186 if (net_ratelimit()) {
1187 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1188 &iph->saddr, ntohs(th->source),
1189 &iph->daddr, ntohs(th->dest),
1190 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1199 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1201 .obj_size = sizeof(struct tcp_request_sock),
1202 .rtx_syn_ack = tcp_v4_rtx_synack,
1203 .send_ack = tcp_v4_reqsk_send_ack,
1204 .destructor = tcp_v4_reqsk_destructor,
1205 .send_reset = tcp_v4_send_reset,
1206 .syn_ack_timeout = tcp_syn_ack_timeout,
1209 #ifdef CONFIG_TCP_MD5SIG
1210 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1211 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1212 .calc_md5_hash = tcp_v4_md5_hash_skb,
1216 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1217 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1218 .twsk_unique = tcp_twsk_unique,
1219 .twsk_destructor= tcp_twsk_destructor,
1222 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1224 struct tcp_extend_values tmp_ext;
1225 struct tcp_options_received tmp_opt;
1227 struct request_sock *req;
1228 struct inet_request_sock *ireq;
1229 struct tcp_sock *tp = tcp_sk(sk);
1230 struct dst_entry *dst = NULL;
1231 __be32 saddr = ip_hdr(skb)->saddr;
1232 __be32 daddr = ip_hdr(skb)->daddr;
1233 __u32 isn = TCP_SKB_CB(skb)->when;
1234 #ifdef CONFIG_SYN_COOKIES
1235 int want_cookie = 0;
1237 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1240 /* Never answer to SYNs send to broadcast or multicast */
1241 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1244 /* TW buckets are converted to open requests without
1245 * limitations, they conserve resources and peer is
1246 * evidently real one.
1248 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1249 if (net_ratelimit())
1250 syn_flood_warning(skb);
1251 #ifdef CONFIG_SYN_COOKIES
1252 if (sysctl_tcp_syncookies) {
1259 /* Accept backlog is full. If we have already queued enough
1260 * of warm entries in syn queue, drop request. It is better than
1261 * clogging syn queue with openreqs with exponentially increasing
1264 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1267 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1271 #ifdef CONFIG_TCP_MD5SIG
1272 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1275 tcp_clear_options(&tmp_opt);
1276 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1277 tmp_opt.user_mss = tp->rx_opt.user_mss;
1278 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1280 if (tmp_opt.cookie_plus > 0 &&
1281 tmp_opt.saw_tstamp &&
1282 !tp->rx_opt.cookie_out_never &&
1283 (sysctl_tcp_cookie_size > 0 ||
1284 (tp->cookie_values != NULL &&
1285 tp->cookie_values->cookie_desired > 0))) {
1287 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1288 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1290 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1291 goto drop_and_release;
1293 /* Secret recipe starts with IP addresses */
1294 *mess++ ^= (__force u32)daddr;
1295 *mess++ ^= (__force u32)saddr;
1297 /* plus variable length Initiator Cookie */
1300 *c++ ^= *hash_location++;
1302 #ifdef CONFIG_SYN_COOKIES
1303 want_cookie = 0; /* not our kind of cookie */
1305 tmp_ext.cookie_out_never = 0; /* false */
1306 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1307 } else if (!tp->rx_opt.cookie_in_always) {
1308 /* redundant indications, but ensure initialization. */
1309 tmp_ext.cookie_out_never = 1; /* true */
1310 tmp_ext.cookie_plus = 0;
1312 goto drop_and_release;
1314 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1316 if (want_cookie && !tmp_opt.saw_tstamp)
1317 tcp_clear_options(&tmp_opt);
1319 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1320 tcp_openreq_init(req, &tmp_opt, skb);
1322 ireq = inet_rsk(req);
1323 ireq->loc_addr = daddr;
1324 ireq->rmt_addr = saddr;
1325 ireq->no_srccheck = inet_sk(sk)->transparent;
1326 ireq->opt = tcp_v4_save_options(sk, skb);
1328 if (security_inet_conn_request(sk, skb, req))
1332 TCP_ECN_create_request(req, tcp_hdr(skb));
1335 #ifdef CONFIG_SYN_COOKIES
1336 req->cookie_ts = tmp_opt.tstamp_ok;
1338 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1340 struct inet_peer *peer = NULL;
1342 /* VJ's idea. We save last timestamp seen
1343 * from the destination in peer table, when entering
1344 * state TIME-WAIT, and check against it before
1345 * accepting new connection request.
1347 * If "isn" is not zero, this request hit alive
1348 * timewait bucket, so that all the necessary checks
1349 * are made in the function processing timewait state.
1351 if (tmp_opt.saw_tstamp &&
1352 tcp_death_row.sysctl_tw_recycle &&
1353 (dst = inet_csk_route_req(sk, req)) != NULL &&
1354 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1355 peer->v4daddr == saddr) {
1356 inet_peer_refcheck(peer);
1357 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1358 (s32)(peer->tcp_ts - req->ts_recent) >
1360 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1361 goto drop_and_release;
1364 /* Kill the following clause, if you dislike this way. */
1365 else if (!sysctl_tcp_syncookies &&
1366 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1367 (sysctl_max_syn_backlog >> 2)) &&
1368 (!peer || !peer->tcp_ts_stamp) &&
1369 (!dst || !dst_metric(dst, RTAX_RTT))) {
1370 /* Without syncookies last quarter of
1371 * backlog is filled with destinations,
1372 * proven to be alive.
1373 * It means that we continue to communicate
1374 * to destinations, already remembered
1375 * to the moment of synflood.
1377 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1378 &saddr, ntohs(tcp_hdr(skb)->source));
1379 goto drop_and_release;
1382 isn = tcp_v4_init_sequence(skb);
1384 tcp_rsk(req)->snt_isn = isn;
1386 if (tcp_v4_send_synack(sk, dst, req,
1387 (struct request_values *)&tmp_ext) ||
1391 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1404 * The three way handshake has completed - we got a valid synack -
1405 * now create the new socket.
1407 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1408 struct request_sock *req,
1409 struct dst_entry *dst)
1411 struct inet_request_sock *ireq;
1412 struct inet_sock *newinet;
1413 struct tcp_sock *newtp;
1415 #ifdef CONFIG_TCP_MD5SIG
1416 struct tcp_md5sig_key *key;
1419 if (sk_acceptq_is_full(sk))
1422 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1425 newsk = tcp_create_openreq_child(sk, req, skb);
1429 newsk->sk_gso_type = SKB_GSO_TCPV4;
1430 sk_setup_caps(newsk, dst);
1432 newtp = tcp_sk(newsk);
1433 newinet = inet_sk(newsk);
1434 ireq = inet_rsk(req);
1435 newinet->inet_daddr = ireq->rmt_addr;
1436 newinet->inet_rcv_saddr = ireq->loc_addr;
1437 newinet->inet_saddr = ireq->loc_addr;
1438 newinet->opt = ireq->opt;
1440 newinet->mc_index = inet_iif(skb);
1441 newinet->mc_ttl = ip_hdr(skb)->ttl;
1442 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1444 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
1445 newinet->inet_id = newtp->write_seq ^ jiffies;
1447 tcp_mtup_init(newsk);
1448 tcp_sync_mss(newsk, dst_mtu(dst));
1449 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1450 if (tcp_sk(sk)->rx_opt.user_mss &&
1451 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1452 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1454 tcp_initialize_rcv_mss(newsk);
1456 #ifdef CONFIG_TCP_MD5SIG
1457 /* Copy over the MD5 key from the original socket */
1458 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1461 * We're using one, so create a matching key
1462 * on the newsk structure. If we fail to get
1463 * memory, then we end up not copying the key
1466 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1468 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1469 newkey, key->keylen);
1470 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1474 __inet_hash_nolisten(newsk, NULL);
1475 __inet_inherit_port(sk, newsk);
1480 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1482 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1487 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1489 struct tcphdr *th = tcp_hdr(skb);
1490 const struct iphdr *iph = ip_hdr(skb);
1492 struct request_sock **prev;
1493 /* Find possible connection requests. */
1494 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1495 iph->saddr, iph->daddr);
1497 return tcp_check_req(sk, skb, req, prev);
1499 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1500 th->source, iph->daddr, th->dest, inet_iif(skb));
1503 if (nsk->sk_state != TCP_TIME_WAIT) {
1507 inet_twsk_put(inet_twsk(nsk));
1511 #ifdef CONFIG_SYN_COOKIES
1513 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1518 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1520 const struct iphdr *iph = ip_hdr(skb);
1522 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1523 if (!tcp_v4_check(skb->len, iph->saddr,
1524 iph->daddr, skb->csum)) {
1525 skb->ip_summed = CHECKSUM_UNNECESSARY;
1530 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1531 skb->len, IPPROTO_TCP, 0);
1533 if (skb->len <= 76) {
1534 return __skb_checksum_complete(skb);
1540 /* The socket must have it's spinlock held when we get
1543 * We have a potential double-lock case here, so even when
1544 * doing backlog processing we use the BH locking scheme.
1545 * This is because we cannot sleep with the original spinlock
1548 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1551 #ifdef CONFIG_TCP_MD5SIG
1553 * We really want to reject the packet as early as possible
1555 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1556 * o There is an MD5 option and we're not expecting one
1558 if (tcp_v4_inbound_md5_hash(sk, skb))
1562 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1563 sock_rps_save_rxhash(sk, skb->rxhash);
1564 TCP_CHECK_TIMER(sk);
1565 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1569 TCP_CHECK_TIMER(sk);
1573 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1576 if (sk->sk_state == TCP_LISTEN) {
1577 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1582 if (tcp_child_process(sk, nsk, skb)) {
1589 sock_rps_save_rxhash(sk, skb->rxhash);
1592 TCP_CHECK_TIMER(sk);
1593 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1597 TCP_CHECK_TIMER(sk);
1601 tcp_v4_send_reset(rsk, skb);
1604 /* Be careful here. If this function gets more complicated and
1605 * gcc suffers from register pressure on the x86, sk (in %ebx)
1606 * might be destroyed here. This current version compiles correctly,
1607 * but you have been warned.
1612 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1620 int tcp_v4_rcv(struct sk_buff *skb)
1622 const struct iphdr *iph;
1626 struct net *net = dev_net(skb->dev);
1628 if (skb->pkt_type != PACKET_HOST)
1631 /* Count it even if it's bad */
1632 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1634 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1639 if (th->doff < sizeof(struct tcphdr) / 4)
1641 if (!pskb_may_pull(skb, th->doff * 4))
1644 /* An explanation is required here, I think.
1645 * Packet length and doff are validated by header prediction,
1646 * provided case of th->doff==0 is eliminated.
1647 * So, we defer the checks. */
1648 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1653 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1654 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1655 skb->len - th->doff * 4);
1656 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1657 TCP_SKB_CB(skb)->when = 0;
1658 TCP_SKB_CB(skb)->flags = iph->tos;
1659 TCP_SKB_CB(skb)->sacked = 0;
1661 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1666 if (sk->sk_state == TCP_TIME_WAIT)
1669 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1670 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1671 goto discard_and_relse;
1674 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1675 goto discard_and_relse;
1678 if (sk_filter(sk, skb))
1679 goto discard_and_relse;
1683 bh_lock_sock_nested(sk);
1685 if (!sock_owned_by_user(sk)) {
1686 #ifdef CONFIG_NET_DMA
1687 struct tcp_sock *tp = tcp_sk(sk);
1688 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1689 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1690 if (tp->ucopy.dma_chan)
1691 ret = tcp_v4_do_rcv(sk, skb);
1695 if (!tcp_prequeue(sk, skb))
1696 ret = tcp_v4_do_rcv(sk, skb);
1698 } else if (unlikely(sk_add_backlog(sk, skb))) {
1700 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1701 goto discard_and_relse;
1710 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1713 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1715 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1717 tcp_v4_send_reset(NULL, skb);
1721 /* Discard frame. */
1730 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1731 inet_twsk_put(inet_twsk(sk));
1735 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1736 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1737 inet_twsk_put(inet_twsk(sk));
1740 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1742 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1744 iph->daddr, th->dest,
1747 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1748 inet_twsk_put(inet_twsk(sk));
1752 /* Fall through to ACK */
1755 tcp_v4_timewait_ack(sk, skb);
1759 case TCP_TW_SUCCESS:;
1764 /* VJ's idea. Save last timestamp seen from this destination
1765 * and hold it at least for normal timewait interval to use for duplicate
1766 * segment detection in subsequent connections, before they enter synchronized
1770 int tcp_v4_remember_stamp(struct sock *sk)
1772 struct inet_sock *inet = inet_sk(sk);
1773 struct tcp_sock *tp = tcp_sk(sk);
1774 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1775 struct inet_peer *peer = NULL;
1778 if (!rt || rt->rt_dst != inet->inet_daddr) {
1779 peer = inet_getpeer(inet->inet_daddr, 1);
1783 rt_bind_peer(rt, 1);
1788 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1789 ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1790 peer->tcp_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
1791 peer->tcp_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
1792 peer->tcp_ts = tp->rx_opt.ts_recent;
1802 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1804 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1807 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1809 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1810 ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1811 peer->tcp_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
1812 peer->tcp_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
1813 peer->tcp_ts = tcptw->tw_ts_recent;
1822 const struct inet_connection_sock_af_ops ipv4_specific = {
1823 .queue_xmit = ip_queue_xmit,
1824 .send_check = tcp_v4_send_check,
1825 .rebuild_header = inet_sk_rebuild_header,
1826 .conn_request = tcp_v4_conn_request,
1827 .syn_recv_sock = tcp_v4_syn_recv_sock,
1828 .remember_stamp = tcp_v4_remember_stamp,
1829 .net_header_len = sizeof(struct iphdr),
1830 .setsockopt = ip_setsockopt,
1831 .getsockopt = ip_getsockopt,
1832 .addr2sockaddr = inet_csk_addr2sockaddr,
1833 .sockaddr_len = sizeof(struct sockaddr_in),
1834 .bind_conflict = inet_csk_bind_conflict,
1835 #ifdef CONFIG_COMPAT
1836 .compat_setsockopt = compat_ip_setsockopt,
1837 .compat_getsockopt = compat_ip_getsockopt,
1841 #ifdef CONFIG_TCP_MD5SIG
1842 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1843 .md5_lookup = tcp_v4_md5_lookup,
1844 .calc_md5_hash = tcp_v4_md5_hash_skb,
1845 .md5_add = tcp_v4_md5_add_func,
1846 .md5_parse = tcp_v4_parse_md5_keys,
1850 /* NOTE: A lot of things set to zero explicitly by call to
1851 * sk_alloc() so need not be done here.
1853 static int tcp_v4_init_sock(struct sock *sk)
1855 struct inet_connection_sock *icsk = inet_csk(sk);
1856 struct tcp_sock *tp = tcp_sk(sk);
1858 skb_queue_head_init(&tp->out_of_order_queue);
1859 tcp_init_xmit_timers(sk);
1860 tcp_prequeue_init(tp);
1862 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1863 tp->mdev = TCP_TIMEOUT_INIT;
1865 /* So many TCP implementations out there (incorrectly) count the
1866 * initial SYN frame in their delayed-ACK and congestion control
1867 * algorithms that we must have the following bandaid to talk
1868 * efficiently to them. -DaveM
1872 /* See draft-stevens-tcpca-spec-01 for discussion of the
1873 * initialization of these values.
1875 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1876 tp->snd_cwnd_clamp = ~0;
1877 tp->mss_cache = TCP_MSS_DEFAULT;
1879 tp->reordering = sysctl_tcp_reordering;
1880 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1882 sk->sk_state = TCP_CLOSE;
1884 sk->sk_write_space = sk_stream_write_space;
1885 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1887 icsk->icsk_af_ops = &ipv4_specific;
1888 icsk->icsk_sync_mss = tcp_sync_mss;
1889 #ifdef CONFIG_TCP_MD5SIG
1890 tp->af_specific = &tcp_sock_ipv4_specific;
1893 /* TCP Cookie Transactions */
1894 if (sysctl_tcp_cookie_size > 0) {
1895 /* Default, cookies without s_data_payload. */
1897 kzalloc(sizeof(*tp->cookie_values),
1899 if (tp->cookie_values != NULL)
1900 kref_init(&tp->cookie_values->kref);
1902 /* Presumed zeroed, in order of appearance:
1903 * cookie_in_always, cookie_out_never,
1904 * s_data_constant, s_data_in, s_data_out
1906 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1907 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1910 percpu_counter_inc(&tcp_sockets_allocated);
1916 void tcp_v4_destroy_sock(struct sock *sk)
1918 struct tcp_sock *tp = tcp_sk(sk);
1920 tcp_clear_xmit_timers(sk);
1922 tcp_cleanup_congestion_control(sk);
1924 /* Cleanup up the write buffer. */
1925 tcp_write_queue_purge(sk);
1927 /* Cleans up our, hopefully empty, out_of_order_queue. */
1928 __skb_queue_purge(&tp->out_of_order_queue);
1930 #ifdef CONFIG_TCP_MD5SIG
1931 /* Clean up the MD5 key list, if any */
1932 if (tp->md5sig_info) {
1933 tcp_v4_clear_md5_list(sk);
1934 kfree(tp->md5sig_info);
1935 tp->md5sig_info = NULL;
1939 #ifdef CONFIG_NET_DMA
1940 /* Cleans up our sk_async_wait_queue */
1941 __skb_queue_purge(&sk->sk_async_wait_queue);
1944 /* Clean prequeue, it must be empty really */
1945 __skb_queue_purge(&tp->ucopy.prequeue);
1947 /* Clean up a referenced TCP bind bucket. */
1948 if (inet_csk(sk)->icsk_bind_hash)
1952 * If sendmsg cached page exists, toss it.
1954 if (sk->sk_sndmsg_page) {
1955 __free_page(sk->sk_sndmsg_page);
1956 sk->sk_sndmsg_page = NULL;
1959 /* TCP Cookie Transactions */
1960 if (tp->cookie_values != NULL) {
1961 kref_put(&tp->cookie_values->kref,
1962 tcp_cookie_values_release);
1963 tp->cookie_values = NULL;
1966 percpu_counter_dec(&tcp_sockets_allocated);
1969 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1971 #ifdef CONFIG_PROC_FS
1972 /* Proc filesystem TCP sock list dumping. */
1974 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1976 return hlist_nulls_empty(head) ? NULL :
1977 list_entry(head->first, struct inet_timewait_sock, tw_node);
1980 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1982 return !is_a_nulls(tw->tw_node.next) ?
1983 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1987 * Get next listener socket follow cur. If cur is NULL, get first socket
1988 * starting from bucket given in st->bucket; when st->bucket is zero the
1989 * very first socket in the hash table is returned.
1991 static void *listening_get_next(struct seq_file *seq, void *cur)
1993 struct inet_connection_sock *icsk;
1994 struct hlist_nulls_node *node;
1995 struct sock *sk = cur;
1996 struct inet_listen_hashbucket *ilb;
1997 struct tcp_iter_state *st = seq->private;
1998 struct net *net = seq_file_net(seq);
2001 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2002 spin_lock_bh(&ilb->lock);
2003 sk = sk_nulls_head(&ilb->head);
2007 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2011 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2012 struct request_sock *req = cur;
2014 icsk = inet_csk(st->syn_wait_sk);
2018 if (req->rsk_ops->family == st->family) {
2025 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2028 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2030 sk = sk_next(st->syn_wait_sk);
2031 st->state = TCP_SEQ_STATE_LISTENING;
2032 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2034 icsk = inet_csk(sk);
2035 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2036 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2038 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2042 sk_nulls_for_each_from(sk, node) {
2043 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) {
2047 icsk = inet_csk(sk);
2048 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2049 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2051 st->uid = sock_i_uid(sk);
2052 st->syn_wait_sk = sk;
2053 st->state = TCP_SEQ_STATE_OPENREQ;
2057 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2059 spin_unlock_bh(&ilb->lock);
2061 if (++st->bucket < INET_LHTABLE_SIZE) {
2062 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2063 spin_lock_bh(&ilb->lock);
2064 sk = sk_nulls_head(&ilb->head);
2072 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2074 struct tcp_iter_state *st = seq->private;
2079 rc = listening_get_next(seq, NULL);
2081 while (rc && *pos) {
2082 rc = listening_get_next(seq, rc);
2088 static inline int empty_bucket(struct tcp_iter_state *st)
2090 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2091 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2095 * Get first established socket starting from bucket given in st->bucket.
2096 * If st->bucket is zero, the very first socket in the hash is returned.
2098 static void *established_get_first(struct seq_file *seq)
2100 struct tcp_iter_state *st = seq->private;
2101 struct net *net = seq_file_net(seq);
2105 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2107 struct hlist_nulls_node *node;
2108 struct inet_timewait_sock *tw;
2109 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2111 /* Lockless fast path for the common case of empty buckets */
2112 if (empty_bucket(st))
2116 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2117 if (sk->sk_family != st->family ||
2118 !net_eq(sock_net(sk), net)) {
2124 st->state = TCP_SEQ_STATE_TIME_WAIT;
2125 inet_twsk_for_each(tw, node,
2126 &tcp_hashinfo.ehash[st->bucket].twchain) {
2127 if (tw->tw_family != st->family ||
2128 !net_eq(twsk_net(tw), net)) {
2134 spin_unlock_bh(lock);
2135 st->state = TCP_SEQ_STATE_ESTABLISHED;
2141 static void *established_get_next(struct seq_file *seq, void *cur)
2143 struct sock *sk = cur;
2144 struct inet_timewait_sock *tw;
2145 struct hlist_nulls_node *node;
2146 struct tcp_iter_state *st = seq->private;
2147 struct net *net = seq_file_net(seq);
2152 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2156 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2163 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2164 st->state = TCP_SEQ_STATE_ESTABLISHED;
2166 /* Look for next non empty bucket */
2168 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2171 if (st->bucket > tcp_hashinfo.ehash_mask)
2174 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2175 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2177 sk = sk_nulls_next(sk);
2179 sk_nulls_for_each_from(sk, node) {
2180 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2184 st->state = TCP_SEQ_STATE_TIME_WAIT;
2185 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2193 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2195 struct tcp_iter_state *st = seq->private;
2199 rc = established_get_first(seq);
2202 rc = established_get_next(seq, rc);
2208 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2211 struct tcp_iter_state *st = seq->private;
2213 st->state = TCP_SEQ_STATE_LISTENING;
2214 rc = listening_get_idx(seq, &pos);
2217 st->state = TCP_SEQ_STATE_ESTABLISHED;
2218 rc = established_get_idx(seq, pos);
2224 static void *tcp_seek_last_pos(struct seq_file *seq)
2226 struct tcp_iter_state *st = seq->private;
2227 int offset = st->offset;
2228 int orig_num = st->num;
2231 switch (st->state) {
2232 case TCP_SEQ_STATE_OPENREQ:
2233 case TCP_SEQ_STATE_LISTENING:
2234 if (st->bucket >= INET_LHTABLE_SIZE)
2236 st->state = TCP_SEQ_STATE_LISTENING;
2237 rc = listening_get_next(seq, NULL);
2238 while (offset-- && rc)
2239 rc = listening_get_next(seq, rc);
2244 case TCP_SEQ_STATE_ESTABLISHED:
2245 case TCP_SEQ_STATE_TIME_WAIT:
2246 st->state = TCP_SEQ_STATE_ESTABLISHED;
2247 if (st->bucket > tcp_hashinfo.ehash_mask)
2249 rc = established_get_first(seq);
2250 while (offset-- && rc)
2251 rc = established_get_next(seq, rc);
2259 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2261 struct tcp_iter_state *st = seq->private;
2264 if (*pos && *pos == st->last_pos) {
2265 rc = tcp_seek_last_pos(seq);
2270 st->state = TCP_SEQ_STATE_LISTENING;
2274 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2277 st->last_pos = *pos;
2281 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2283 struct tcp_iter_state *st = seq->private;
2286 if (v == SEQ_START_TOKEN) {
2287 rc = tcp_get_idx(seq, 0);
2291 switch (st->state) {
2292 case TCP_SEQ_STATE_OPENREQ:
2293 case TCP_SEQ_STATE_LISTENING:
2294 rc = listening_get_next(seq, v);
2296 st->state = TCP_SEQ_STATE_ESTABLISHED;
2299 rc = established_get_first(seq);
2302 case TCP_SEQ_STATE_ESTABLISHED:
2303 case TCP_SEQ_STATE_TIME_WAIT:
2304 rc = established_get_next(seq, v);
2309 st->last_pos = *pos;
2313 static void tcp_seq_stop(struct seq_file *seq, void *v)
2315 struct tcp_iter_state *st = seq->private;
2317 switch (st->state) {
2318 case TCP_SEQ_STATE_OPENREQ:
2320 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2321 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2323 case TCP_SEQ_STATE_LISTENING:
2324 if (v != SEQ_START_TOKEN)
2325 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2327 case TCP_SEQ_STATE_TIME_WAIT:
2328 case TCP_SEQ_STATE_ESTABLISHED:
2330 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2335 static int tcp_seq_open(struct inode *inode, struct file *file)
2337 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2338 struct tcp_iter_state *s;
2341 err = seq_open_net(inode, file, &afinfo->seq_ops,
2342 sizeof(struct tcp_iter_state));
2346 s = ((struct seq_file *)file->private_data)->private;
2347 s->family = afinfo->family;
2352 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2355 struct proc_dir_entry *p;
2357 afinfo->seq_fops.open = tcp_seq_open;
2358 afinfo->seq_fops.read = seq_read;
2359 afinfo->seq_fops.llseek = seq_lseek;
2360 afinfo->seq_fops.release = seq_release_net;
2362 afinfo->seq_ops.start = tcp_seq_start;
2363 afinfo->seq_ops.next = tcp_seq_next;
2364 afinfo->seq_ops.stop = tcp_seq_stop;
2366 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2367 &afinfo->seq_fops, afinfo);
2373 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2375 proc_net_remove(net, afinfo->name);
2378 static void get_openreq4(struct sock *sk, struct request_sock *req,
2379 struct seq_file *f, int i, int uid, int *len)
2381 const struct inet_request_sock *ireq = inet_rsk(req);
2382 int ttd = req->expires - jiffies;
2384 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2385 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
2388 ntohs(inet_sk(sk)->inet_sport),
2390 ntohs(ireq->rmt_port),
2392 0, 0, /* could print option size, but that is af dependent. */
2393 1, /* timers active (only the expire timer) */
2394 jiffies_to_clock_t(ttd),
2397 0, /* non standard timer */
2398 0, /* open_requests have no inode */
2399 atomic_read(&sk->sk_refcnt),
2404 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2407 unsigned long timer_expires;
2408 struct tcp_sock *tp = tcp_sk(sk);
2409 const struct inet_connection_sock *icsk = inet_csk(sk);
2410 struct inet_sock *inet = inet_sk(sk);
2411 __be32 dest = inet->inet_daddr;
2412 __be32 src = inet->inet_rcv_saddr;
2413 __u16 destp = ntohs(inet->inet_dport);
2414 __u16 srcp = ntohs(inet->inet_sport);
2417 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2419 timer_expires = icsk->icsk_timeout;
2420 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2422 timer_expires = icsk->icsk_timeout;
2423 } else if (timer_pending(&sk->sk_timer)) {
2425 timer_expires = sk->sk_timer.expires;
2428 timer_expires = jiffies;
2431 if (sk->sk_state == TCP_LISTEN)
2432 rx_queue = sk->sk_ack_backlog;
2435 * because we dont lock socket, we might find a transient negative value
2437 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2439 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2440 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
2441 i, src, srcp, dest, destp, sk->sk_state,
2442 tp->write_seq - tp->snd_una,
2445 jiffies_to_clock_t(timer_expires - jiffies),
2446 icsk->icsk_retransmits,
2448 icsk->icsk_probes_out,
2450 atomic_read(&sk->sk_refcnt), sk,
2451 jiffies_to_clock_t(icsk->icsk_rto),
2452 jiffies_to_clock_t(icsk->icsk_ack.ato),
2453 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2455 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2459 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2460 struct seq_file *f, int i, int *len)
2464 int ttd = tw->tw_ttd - jiffies;
2469 dest = tw->tw_daddr;
2470 src = tw->tw_rcv_saddr;
2471 destp = ntohs(tw->tw_dport);
2472 srcp = ntohs(tw->tw_sport);
2474 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2475 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
2476 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2477 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2478 atomic_read(&tw->tw_refcnt), tw, len);
2483 static int tcp4_seq_show(struct seq_file *seq, void *v)
2485 struct tcp_iter_state *st;
2488 if (v == SEQ_START_TOKEN) {
2489 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2490 " sl local_address rem_address st tx_queue "
2491 "rx_queue tr tm->when retrnsmt uid timeout "
2497 switch (st->state) {
2498 case TCP_SEQ_STATE_LISTENING:
2499 case TCP_SEQ_STATE_ESTABLISHED:
2500 get_tcp4_sock(v, seq, st->num, &len);
2502 case TCP_SEQ_STATE_OPENREQ:
2503 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2505 case TCP_SEQ_STATE_TIME_WAIT:
2506 get_timewait4_sock(v, seq, st->num, &len);
2509 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2514 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2518 .owner = THIS_MODULE,
2521 .show = tcp4_seq_show,
2525 static int __net_init tcp4_proc_init_net(struct net *net)
2527 return tcp_proc_register(net, &tcp4_seq_afinfo);
2530 static void __net_exit tcp4_proc_exit_net(struct net *net)
2532 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2535 static struct pernet_operations tcp4_net_ops = {
2536 .init = tcp4_proc_init_net,
2537 .exit = tcp4_proc_exit_net,
2540 int __init tcp4_proc_init(void)
2542 return register_pernet_subsys(&tcp4_net_ops);
2545 void tcp4_proc_exit(void)
2547 unregister_pernet_subsys(&tcp4_net_ops);
2549 #endif /* CONFIG_PROC_FS */
2551 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2553 struct iphdr *iph = skb_gro_network_header(skb);
2555 switch (skb->ip_summed) {
2556 case CHECKSUM_COMPLETE:
2557 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2559 skb->ip_summed = CHECKSUM_UNNECESSARY;
2565 NAPI_GRO_CB(skb)->flush = 1;
2569 return tcp_gro_receive(head, skb);
2571 EXPORT_SYMBOL(tcp4_gro_receive);
2573 int tcp4_gro_complete(struct sk_buff *skb)
2575 struct iphdr *iph = ip_hdr(skb);
2576 struct tcphdr *th = tcp_hdr(skb);
2578 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2579 iph->saddr, iph->daddr, 0);
2580 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2582 return tcp_gro_complete(skb);
2584 EXPORT_SYMBOL(tcp4_gro_complete);
2586 struct proto tcp_prot = {
2588 .owner = THIS_MODULE,
2590 .connect = tcp_v4_connect,
2591 .disconnect = tcp_disconnect,
2592 .accept = inet_csk_accept,
2594 .init = tcp_v4_init_sock,
2595 .destroy = tcp_v4_destroy_sock,
2596 .shutdown = tcp_shutdown,
2597 .setsockopt = tcp_setsockopt,
2598 .getsockopt = tcp_getsockopt,
2599 .recvmsg = tcp_recvmsg,
2600 .backlog_rcv = tcp_v4_do_rcv,
2602 .unhash = inet_unhash,
2603 .get_port = inet_csk_get_port,
2604 .enter_memory_pressure = tcp_enter_memory_pressure,
2605 .sockets_allocated = &tcp_sockets_allocated,
2606 .orphan_count = &tcp_orphan_count,
2607 .memory_allocated = &tcp_memory_allocated,
2608 .memory_pressure = &tcp_memory_pressure,
2609 .sysctl_mem = sysctl_tcp_mem,
2610 .sysctl_wmem = sysctl_tcp_wmem,
2611 .sysctl_rmem = sysctl_tcp_rmem,
2612 .max_header = MAX_TCP_HEADER,
2613 .obj_size = sizeof(struct tcp_sock),
2614 .slab_flags = SLAB_DESTROY_BY_RCU,
2615 .twsk_prot = &tcp_timewait_sock_ops,
2616 .rsk_prot = &tcp_request_sock_ops,
2617 .h.hashinfo = &tcp_hashinfo,
2618 #ifdef CONFIG_COMPAT
2619 .compat_setsockopt = compat_tcp_setsockopt,
2620 .compat_getsockopt = compat_tcp_getsockopt,
2625 static int __net_init tcp_sk_init(struct net *net)
2627 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2628 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2631 static void __net_exit tcp_sk_exit(struct net *net)
2633 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2636 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2638 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2641 static struct pernet_operations __net_initdata tcp_sk_ops = {
2642 .init = tcp_sk_init,
2643 .exit = tcp_sk_exit,
2644 .exit_batch = tcp_sk_exit_batch,
2647 void __init tcp_v4_init(void)
2649 inet_hashinfo_init(&tcp_hashinfo);
2650 if (register_pernet_subsys(&tcp_sk_ops))
2651 panic("Failed to create the TCP control socket.\n");
2654 EXPORT_SYMBOL(ipv4_specific);
2655 EXPORT_SYMBOL(tcp_hashinfo);
2656 EXPORT_SYMBOL(tcp_prot);
2657 EXPORT_SYMBOL(tcp_v4_conn_request);
2658 EXPORT_SYMBOL(tcp_v4_connect);
2659 EXPORT_SYMBOL(tcp_v4_do_rcv);
2660 EXPORT_SYMBOL(tcp_v4_remember_stamp);
2661 EXPORT_SYMBOL(tcp_v4_send_check);
2662 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2664 #ifdef CONFIG_PROC_FS
2665 EXPORT_SYMBOL(tcp_proc_register);
2666 EXPORT_SYMBOL(tcp_proc_unregister);
2668 EXPORT_SYMBOL(sysctl_tcp_low_latency);