2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Implementation of the Transmission Control Protocol(TCP).
8 * IPv4 specific functions
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
16 * See tcp.c for author information
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
54 #include <linux/bottom_half.h>
55 #include <linux/types.h>
56 #include <linux/fcntl.h>
57 #include <linux/module.h>
58 #include <linux/random.h>
59 #include <linux/cache.h>
60 #include <linux/jhash.h>
61 #include <linux/init.h>
62 #include <linux/times.h>
64 #include <net/net_namespace.h>
66 #include <net/inet_hashtables.h>
68 #include <net/transp_v6.h>
70 #include <net/inet_common.h>
71 #include <net/timewait_sock.h>
73 #include <net/netdma.h>
75 #include <linux/inet.h>
76 #include <linux/ipv6.h>
77 #include <linux/stddef.h>
78 #include <linux/proc_fs.h>
79 #include <linux/seq_file.h>
81 #include <linux/crypto.h>
82 #include <linux/scatterlist.h>
84 int sysctl_tcp_tw_reuse __read_mostly;
85 int sysctl_tcp_low_latency __read_mostly;
88 #ifdef CONFIG_TCP_MD5SIG
89 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
91 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
92 __be32 daddr, __be32 saddr, struct tcphdr *th);
95 struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
101 struct inet_hashinfo tcp_hashinfo;
103 static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
105 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
108 tcp_hdr(skb)->source);
111 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
113 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
114 struct tcp_sock *tp = tcp_sk(sk);
116 /* With PAWS, it is safe from the viewpoint
117 of data integrity. Even without PAWS it is safe provided sequence
118 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
120 Actually, the idea is close to VJ's one, only timestamp cache is
121 held not per host, but per port pair and TW bucket is used as state
124 If TW bucket has been already destroyed we fall back to VJ's scheme
125 and use initial timestamp retrieved from peer table.
127 if (tcptw->tw_ts_recent_stamp &&
128 (twp == NULL || (sysctl_tcp_tw_reuse &&
129 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
130 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
131 if (tp->write_seq == 0)
133 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
134 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
142 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
144 /* This will initiate an outgoing connection. */
145 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
147 struct inet_sock *inet = inet_sk(sk);
148 struct tcp_sock *tp = tcp_sk(sk);
149 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
151 __be32 daddr, nexthop;
155 if (addr_len < sizeof(struct sockaddr_in))
158 if (usin->sin_family != AF_INET)
159 return -EAFNOSUPPORT;
161 nexthop = daddr = usin->sin_addr.s_addr;
162 if (inet->opt && inet->opt->srr) {
165 nexthop = inet->opt->faddr;
168 tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr,
169 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
171 inet->inet_sport, usin->sin_port, sk, 1);
173 if (tmp == -ENETUNREACH)
174 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
178 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
183 if (!inet->opt || !inet->opt->srr)
186 if (!inet->inet_saddr)
187 inet->inet_saddr = rt->rt_src;
188 inet->inet_rcv_saddr = inet->inet_saddr;
190 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
191 /* Reset inherited state */
192 tp->rx_opt.ts_recent = 0;
193 tp->rx_opt.ts_recent_stamp = 0;
197 if (tcp_death_row.sysctl_tw_recycle &&
198 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
199 struct inet_peer *peer = rt_get_peer(rt);
201 * VJ's idea. We save last timestamp seen from
202 * the destination in peer table, when entering state
203 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
204 * when trying new connection.
207 (u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
208 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
209 tp->rx_opt.ts_recent = peer->tcp_ts;
213 inet->inet_dport = usin->sin_port;
214 inet->inet_daddr = daddr;
216 inet_csk(sk)->icsk_ext_hdr_len = 0;
218 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
220 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
222 /* Socket identity is still unknown (sport may be zero).
223 * However we set state to SYN-SENT and not releasing socket
224 * lock select source port, enter ourselves into the hash tables and
225 * complete initialization after this.
227 tcp_set_state(sk, TCP_SYN_SENT);
228 err = inet_hash_connect(&tcp_death_row, sk);
232 err = ip_route_newports(&rt, IPPROTO_TCP,
233 inet->inet_sport, inet->inet_dport, sk);
237 /* OK, now commit destination to socket. */
238 sk->sk_gso_type = SKB_GSO_TCPV4;
239 sk_setup_caps(sk, &rt->u.dst);
242 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
247 inet->inet_id = tp->write_seq ^ jiffies;
249 err = tcp_connect(sk);
258 * This unhashes the socket and releases the local port,
261 tcp_set_state(sk, TCP_CLOSE);
263 sk->sk_route_caps = 0;
264 inet->inet_dport = 0;
269 * This routine does path mtu discovery as defined in RFC1191.
271 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
276 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
277 * send out by Linux are always <576bytes so they should go through
280 if (sk->sk_state == TCP_LISTEN)
283 /* We don't check in the destentry if pmtu discovery is forbidden
284 * on this route. We just assume that no packet_to_big packets
285 * are send back when pmtu discovery is not active.
286 * There is a small race when the user changes this flag in the
287 * route, but I think that's acceptable.
289 if ((dst = __sk_dst_check(sk, 0)) == NULL)
292 dst->ops->update_pmtu(dst, mtu);
294 /* Something is about to be wrong... Remember soft error
295 * for the case, if this connection will not able to recover.
297 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
298 sk->sk_err_soft = EMSGSIZE;
302 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
303 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
304 tcp_sync_mss(sk, mtu);
306 /* Resend the TCP packet because it's
307 * clear that the old packet has been
308 * dropped. This is the new "fast" path mtu
311 tcp_simple_retransmit(sk);
312 } /* else let the usual retransmit timer handle it */
316 * This routine is called by the ICMP module when it gets some
317 * sort of error condition. If err < 0 then the socket should
318 * be closed and the error returned to the user. If err > 0
319 * it's just the icmp type << 8 | icmp code. After adjustment
320 * header points to the first 8 bytes of the tcp header. We need
321 * to find the appropriate port.
323 * The locking strategy used here is very "optimistic". When
324 * someone else accesses the socket the ICMP is just dropped
325 * and for some paths there is no check at all.
326 * A more general error queue to queue errors for later handling
327 * is probably better.
331 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
333 struct iphdr *iph = (struct iphdr *)icmp_skb->data;
334 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
335 struct inet_connection_sock *icsk;
337 struct inet_sock *inet;
338 const int type = icmp_hdr(icmp_skb)->type;
339 const int code = icmp_hdr(icmp_skb)->code;
345 struct net *net = dev_net(icmp_skb->dev);
347 if (icmp_skb->len < (iph->ihl << 2) + 8) {
348 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
352 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
353 iph->saddr, th->source, inet_iif(icmp_skb));
355 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
358 if (sk->sk_state == TCP_TIME_WAIT) {
359 inet_twsk_put(inet_twsk(sk));
364 /* If too many ICMPs get dropped on busy
365 * servers this needs to be solved differently.
367 if (sock_owned_by_user(sk))
368 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
370 if (sk->sk_state == TCP_CLOSE)
375 seq = ntohl(th->seq);
376 if (sk->sk_state != TCP_LISTEN &&
377 !between(seq, tp->snd_una, tp->snd_nxt)) {
378 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
383 case ICMP_SOURCE_QUENCH:
384 /* Just silently ignore these. */
386 case ICMP_PARAMETERPROB:
389 case ICMP_DEST_UNREACH:
390 if (code > NR_ICMP_UNREACH)
393 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
394 if (!sock_owned_by_user(sk))
395 do_pmtu_discovery(sk, iph, info);
399 err = icmp_err_convert[code].errno;
400 /* check if icmp_skb allows revert of backoff
401 * (see draft-zimmermann-tcp-lcd) */
402 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
404 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
408 icsk->icsk_backoff--;
409 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
413 skb = tcp_write_queue_head(sk);
416 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
417 tcp_time_stamp - TCP_SKB_CB(skb)->when);
420 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
421 remaining, TCP_RTO_MAX);
422 } else if (sock_owned_by_user(sk)) {
423 /* RTO revert clocked out retransmission,
424 * but socket is locked. Will defer. */
425 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
428 /* RTO revert clocked out retransmission.
429 * Will retransmit now */
430 tcp_retransmit_timer(sk);
434 case ICMP_TIME_EXCEEDED:
441 switch (sk->sk_state) {
442 struct request_sock *req, **prev;
444 if (sock_owned_by_user(sk))
447 req = inet_csk_search_req(sk, &prev, th->dest,
448 iph->daddr, iph->saddr);
452 /* ICMPs are not backlogged, hence we cannot get
453 an established socket here.
457 if (seq != tcp_rsk(req)->snt_isn) {
458 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
463 * Still in SYN_RECV, just remove it silently.
464 * There is no good way to pass the error to the newly
465 * created socket, and POSIX does not want network
466 * errors returned from accept().
468 inet_csk_reqsk_queue_drop(sk, req, prev);
472 case TCP_SYN_RECV: /* Cannot happen.
473 It can f.e. if SYNs crossed.
475 if (!sock_owned_by_user(sk)) {
478 sk->sk_error_report(sk);
482 sk->sk_err_soft = err;
487 /* If we've already connected we will keep trying
488 * until we time out, or the user gives up.
490 * rfc1122 4.2.3.9 allows to consider as hard errors
491 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
492 * but it is obsoleted by pmtu discovery).
494 * Note, that in modern internet, where routing is unreliable
495 * and in each dark corner broken firewalls sit, sending random
496 * errors ordered by their masters even this two messages finally lose
497 * their original sense (even Linux sends invalid PORT_UNREACHs)
499 * Now we are in compliance with RFCs.
504 if (!sock_owned_by_user(sk) && inet->recverr) {
506 sk->sk_error_report(sk);
507 } else { /* Only an error on timeout */
508 sk->sk_err_soft = err;
516 /* This routine computes an IPv4 TCP checksum. */
517 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
519 struct inet_sock *inet = inet_sk(sk);
520 struct tcphdr *th = tcp_hdr(skb);
522 if (skb->ip_summed == CHECKSUM_PARTIAL) {
523 th->check = ~tcp_v4_check(len, inet->inet_saddr,
524 inet->inet_daddr, 0);
525 skb->csum_start = skb_transport_header(skb) - skb->head;
526 skb->csum_offset = offsetof(struct tcphdr, check);
528 th->check = tcp_v4_check(len, inet->inet_saddr,
536 int tcp_v4_gso_send_check(struct sk_buff *skb)
538 const struct iphdr *iph;
541 if (!pskb_may_pull(skb, sizeof(*th)))
548 th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0);
549 skb->csum_start = skb_transport_header(skb) - skb->head;
550 skb->csum_offset = offsetof(struct tcphdr, check);
551 skb->ip_summed = CHECKSUM_PARTIAL;
556 * This routine will send an RST to the other tcp.
558 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
560 * Answer: if a packet caused RST, it is not for a socket
561 * existing in our system, if it is matched to a socket,
562 * it is just duplicate segment or bug in other side's TCP.
563 * So that we build reply only basing on parameters
564 * arrived with segment.
565 * Exception: precedence violation. We do not implement it in any case.
568 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
570 struct tcphdr *th = tcp_hdr(skb);
573 #ifdef CONFIG_TCP_MD5SIG
574 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
577 struct ip_reply_arg arg;
578 #ifdef CONFIG_TCP_MD5SIG
579 struct tcp_md5sig_key *key;
583 /* Never send a reset in response to a reset. */
587 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
590 /* Swap the send and the receive. */
591 memset(&rep, 0, sizeof(rep));
592 rep.th.dest = th->source;
593 rep.th.source = th->dest;
594 rep.th.doff = sizeof(struct tcphdr) / 4;
598 rep.th.seq = th->ack_seq;
601 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
602 skb->len - (th->doff << 2));
605 memset(&arg, 0, sizeof(arg));
606 arg.iov[0].iov_base = (unsigned char *)&rep;
607 arg.iov[0].iov_len = sizeof(rep.th);
609 #ifdef CONFIG_TCP_MD5SIG
610 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
612 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
614 (TCPOPT_MD5SIG << 8) |
616 /* Update length and the length the header thinks exists */
617 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
618 rep.th.doff = arg.iov[0].iov_len / 4;
620 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
621 key, ip_hdr(skb)->saddr,
622 ip_hdr(skb)->daddr, &rep.th);
625 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
626 ip_hdr(skb)->saddr, /* XXX */
627 arg.iov[0].iov_len, IPPROTO_TCP, 0);
628 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
629 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
631 net = dev_net(skb_dst(skb)->dev);
632 ip_send_reply(net->ipv4.tcp_sock, skb,
633 &arg, arg.iov[0].iov_len);
635 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
636 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
639 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
640 outside socket context is ugly, certainly. What can I do?
643 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
644 u32 win, u32 ts, int oif,
645 struct tcp_md5sig_key *key,
648 struct tcphdr *th = tcp_hdr(skb);
651 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
652 #ifdef CONFIG_TCP_MD5SIG
653 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
657 struct ip_reply_arg arg;
658 struct net *net = dev_net(skb_dst(skb)->dev);
660 memset(&rep.th, 0, sizeof(struct tcphdr));
661 memset(&arg, 0, sizeof(arg));
663 arg.iov[0].iov_base = (unsigned char *)&rep;
664 arg.iov[0].iov_len = sizeof(rep.th);
666 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
667 (TCPOPT_TIMESTAMP << 8) |
669 rep.opt[1] = htonl(tcp_time_stamp);
670 rep.opt[2] = htonl(ts);
671 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
674 /* Swap the send and the receive. */
675 rep.th.dest = th->source;
676 rep.th.source = th->dest;
677 rep.th.doff = arg.iov[0].iov_len / 4;
678 rep.th.seq = htonl(seq);
679 rep.th.ack_seq = htonl(ack);
681 rep.th.window = htons(win);
683 #ifdef CONFIG_TCP_MD5SIG
685 int offset = (ts) ? 3 : 0;
687 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
689 (TCPOPT_MD5SIG << 8) |
691 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
692 rep.th.doff = arg.iov[0].iov_len/4;
694 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
695 key, ip_hdr(skb)->saddr,
696 ip_hdr(skb)->daddr, &rep.th);
699 arg.flags = reply_flags;
700 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
701 ip_hdr(skb)->saddr, /* XXX */
702 arg.iov[0].iov_len, IPPROTO_TCP, 0);
703 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
705 arg.bound_dev_if = oif;
707 ip_send_reply(net->ipv4.tcp_sock, skb,
708 &arg, arg.iov[0].iov_len);
710 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
713 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
715 struct inet_timewait_sock *tw = inet_twsk(sk);
716 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
718 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
719 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
722 tcp_twsk_md5_key(tcptw),
723 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
729 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
730 struct request_sock *req)
732 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
733 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
736 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
737 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
741 * Send a SYN-ACK after having received a SYN.
742 * This still operates on a request_sock only, not on a big
745 static int __tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
746 struct request_sock *req,
747 struct request_values *rvp)
749 const struct inet_request_sock *ireq = inet_rsk(req);
751 struct sk_buff * skb;
753 /* First, grab a route. */
754 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
757 skb = tcp_make_synack(sk, dst, req, rvp);
760 struct tcphdr *th = tcp_hdr(skb);
762 th->check = tcp_v4_check(skb->len,
765 csum_partial(th, skb->len,
768 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
771 err = net_xmit_eval(err);
778 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
779 struct request_values *rvp)
781 return __tcp_v4_send_synack(sk, NULL, req, rvp);
785 * IPv4 request_sock destructor.
787 static void tcp_v4_reqsk_destructor(struct request_sock *req)
789 kfree(inet_rsk(req)->opt);
792 #ifdef CONFIG_SYN_COOKIES
793 static void syn_flood_warning(struct sk_buff *skb)
795 static unsigned long warntime;
797 if (time_after(jiffies, (warntime + HZ * 60))) {
800 "possible SYN flooding on port %d. Sending cookies.\n",
801 ntohs(tcp_hdr(skb)->dest));
807 * Save and compile IPv4 options into the request_sock if needed.
809 static struct ip_options *tcp_v4_save_options(struct sock *sk,
812 struct ip_options *opt = &(IPCB(skb)->opt);
813 struct ip_options *dopt = NULL;
815 if (opt && opt->optlen) {
816 int opt_size = optlength(opt);
817 dopt = kmalloc(opt_size, GFP_ATOMIC);
819 if (ip_options_echo(dopt, skb)) {
828 #ifdef CONFIG_TCP_MD5SIG
830 * RFC2385 MD5 checksumming requires a mapping of
831 * IP address->MD5 Key.
832 * We need to maintain these in the sk structure.
835 /* Find the Key structure for an address. */
836 static struct tcp_md5sig_key *
837 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
839 struct tcp_sock *tp = tcp_sk(sk);
842 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
844 for (i = 0; i < tp->md5sig_info->entries4; i++) {
845 if (tp->md5sig_info->keys4[i].addr == addr)
846 return &tp->md5sig_info->keys4[i].base;
851 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
852 struct sock *addr_sk)
854 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
857 EXPORT_SYMBOL(tcp_v4_md5_lookup);
859 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
860 struct request_sock *req)
862 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
865 /* This can be called on a newly created socket, from other files */
866 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
867 u8 *newkey, u8 newkeylen)
869 /* Add Key to the list */
870 struct tcp_md5sig_key *key;
871 struct tcp_sock *tp = tcp_sk(sk);
872 struct tcp4_md5sig_key *keys;
874 key = tcp_v4_md5_do_lookup(sk, addr);
876 /* Pre-existing entry - just update that one. */
879 key->keylen = newkeylen;
881 struct tcp_md5sig_info *md5sig;
883 if (!tp->md5sig_info) {
884 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
886 if (!tp->md5sig_info) {
890 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
892 if (tcp_alloc_md5sig_pool(sk) == NULL) {
896 md5sig = tp->md5sig_info;
898 if (md5sig->alloced4 == md5sig->entries4) {
899 keys = kmalloc((sizeof(*keys) *
900 (md5sig->entries4 + 1)), GFP_ATOMIC);
903 tcp_free_md5sig_pool();
907 if (md5sig->entries4)
908 memcpy(keys, md5sig->keys4,
909 sizeof(*keys) * md5sig->entries4);
911 /* Free old key list, and reference new one */
912 kfree(md5sig->keys4);
913 md5sig->keys4 = keys;
917 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
918 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
919 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
924 EXPORT_SYMBOL(tcp_v4_md5_do_add);
926 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
927 u8 *newkey, u8 newkeylen)
929 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
933 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
935 struct tcp_sock *tp = tcp_sk(sk);
938 for (i = 0; i < tp->md5sig_info->entries4; i++) {
939 if (tp->md5sig_info->keys4[i].addr == addr) {
941 kfree(tp->md5sig_info->keys4[i].base.key);
942 tp->md5sig_info->entries4--;
944 if (tp->md5sig_info->entries4 == 0) {
945 kfree(tp->md5sig_info->keys4);
946 tp->md5sig_info->keys4 = NULL;
947 tp->md5sig_info->alloced4 = 0;
948 } else if (tp->md5sig_info->entries4 != i) {
949 /* Need to do some manipulation */
950 memmove(&tp->md5sig_info->keys4[i],
951 &tp->md5sig_info->keys4[i+1],
952 (tp->md5sig_info->entries4 - i) *
953 sizeof(struct tcp4_md5sig_key));
955 tcp_free_md5sig_pool();
962 EXPORT_SYMBOL(tcp_v4_md5_do_del);
964 static void tcp_v4_clear_md5_list(struct sock *sk)
966 struct tcp_sock *tp = tcp_sk(sk);
968 /* Free each key, then the set of key keys,
969 * the crypto element, and then decrement our
970 * hold on the last resort crypto.
972 if (tp->md5sig_info->entries4) {
974 for (i = 0; i < tp->md5sig_info->entries4; i++)
975 kfree(tp->md5sig_info->keys4[i].base.key);
976 tp->md5sig_info->entries4 = 0;
977 tcp_free_md5sig_pool();
979 if (tp->md5sig_info->keys4) {
980 kfree(tp->md5sig_info->keys4);
981 tp->md5sig_info->keys4 = NULL;
982 tp->md5sig_info->alloced4 = 0;
986 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
989 struct tcp_md5sig cmd;
990 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
993 if (optlen < sizeof(cmd))
996 if (copy_from_user(&cmd, optval, sizeof(cmd)))
999 if (sin->sin_family != AF_INET)
1002 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1003 if (!tcp_sk(sk)->md5sig_info)
1005 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1008 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1011 if (!tcp_sk(sk)->md5sig_info) {
1012 struct tcp_sock *tp = tcp_sk(sk);
1013 struct tcp_md5sig_info *p;
1015 p = kzalloc(sizeof(*p), sk->sk_allocation);
1019 tp->md5sig_info = p;
1020 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1023 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1026 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1027 newkey, cmd.tcpm_keylen);
1030 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1031 __be32 daddr, __be32 saddr, int nbytes)
1033 struct tcp4_pseudohdr *bp;
1034 struct scatterlist sg;
1036 bp = &hp->md5_blk.ip4;
1039 * 1. the TCP pseudo-header (in the order: source IP address,
1040 * destination IP address, zero-padded protocol number, and
1046 bp->protocol = IPPROTO_TCP;
1047 bp->len = cpu_to_be16(nbytes);
1049 sg_init_one(&sg, bp, sizeof(*bp));
1050 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1053 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1054 __be32 daddr, __be32 saddr, struct tcphdr *th)
1056 struct tcp_md5sig_pool *hp;
1057 struct hash_desc *desc;
1059 hp = tcp_get_md5sig_pool();
1061 goto clear_hash_noput;
1062 desc = &hp->md5_desc;
1064 if (crypto_hash_init(desc))
1066 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1068 if (tcp_md5_hash_header(hp, th))
1070 if (tcp_md5_hash_key(hp, key))
1072 if (crypto_hash_final(desc, md5_hash))
1075 tcp_put_md5sig_pool();
1079 tcp_put_md5sig_pool();
1081 memset(md5_hash, 0, 16);
1085 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1086 struct sock *sk, struct request_sock *req,
1087 struct sk_buff *skb)
1089 struct tcp_md5sig_pool *hp;
1090 struct hash_desc *desc;
1091 struct tcphdr *th = tcp_hdr(skb);
1092 __be32 saddr, daddr;
1095 saddr = inet_sk(sk)->inet_saddr;
1096 daddr = inet_sk(sk)->inet_daddr;
1098 saddr = inet_rsk(req)->loc_addr;
1099 daddr = inet_rsk(req)->rmt_addr;
1101 const struct iphdr *iph = ip_hdr(skb);
1106 hp = tcp_get_md5sig_pool();
1108 goto clear_hash_noput;
1109 desc = &hp->md5_desc;
1111 if (crypto_hash_init(desc))
1114 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1116 if (tcp_md5_hash_header(hp, th))
1118 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1120 if (tcp_md5_hash_key(hp, key))
1122 if (crypto_hash_final(desc, md5_hash))
1125 tcp_put_md5sig_pool();
1129 tcp_put_md5sig_pool();
1131 memset(md5_hash, 0, 16);
1135 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1137 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1140 * This gets called for each TCP segment that arrives
1141 * so we want to be efficient.
1142 * We have 3 drop cases:
1143 * o No MD5 hash and one expected.
1144 * o MD5 hash and we're not expecting one.
1145 * o MD5 hash and its wrong.
1147 __u8 *hash_location = NULL;
1148 struct tcp_md5sig_key *hash_expected;
1149 const struct iphdr *iph = ip_hdr(skb);
1150 struct tcphdr *th = tcp_hdr(skb);
1152 unsigned char newhash[16];
1154 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1155 hash_location = tcp_parse_md5sig_option(th);
1157 /* We've parsed the options - do we have a hash? */
1158 if (!hash_expected && !hash_location)
1161 if (hash_expected && !hash_location) {
1162 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1166 if (!hash_expected && hash_location) {
1167 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1171 /* Okay, so this is hash_expected and hash_location -
1172 * so we need to calculate the checksum.
1174 genhash = tcp_v4_md5_hash_skb(newhash,
1178 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1179 if (net_ratelimit()) {
1180 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1181 &iph->saddr, ntohs(th->source),
1182 &iph->daddr, ntohs(th->dest),
1183 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1192 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1194 .obj_size = sizeof(struct tcp_request_sock),
1195 .rtx_syn_ack = tcp_v4_send_synack,
1196 .send_ack = tcp_v4_reqsk_send_ack,
1197 .destructor = tcp_v4_reqsk_destructor,
1198 .send_reset = tcp_v4_send_reset,
1201 #ifdef CONFIG_TCP_MD5SIG
1202 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1203 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1204 .calc_md5_hash = tcp_v4_md5_hash_skb,
1208 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1209 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1210 .twsk_unique = tcp_twsk_unique,
1211 .twsk_destructor= tcp_twsk_destructor,
1214 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1216 struct tcp_extend_values tmp_ext;
1217 struct tcp_options_received tmp_opt;
1219 struct request_sock *req;
1220 struct inet_request_sock *ireq;
1221 struct tcp_sock *tp = tcp_sk(sk);
1222 struct dst_entry *dst = NULL;
1223 __be32 saddr = ip_hdr(skb)->saddr;
1224 __be32 daddr = ip_hdr(skb)->daddr;
1225 __u32 isn = TCP_SKB_CB(skb)->when;
1226 #ifdef CONFIG_SYN_COOKIES
1227 int want_cookie = 0;
1229 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1232 /* Never answer to SYNs send to broadcast or multicast */
1233 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1236 /* TW buckets are converted to open requests without
1237 * limitations, they conserve resources and peer is
1238 * evidently real one.
1240 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1241 #ifdef CONFIG_SYN_COOKIES
1242 if (sysctl_tcp_syncookies) {
1249 /* Accept backlog is full. If we have already queued enough
1250 * of warm entries in syn queue, drop request. It is better than
1251 * clogging syn queue with openreqs with exponentially increasing
1254 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1257 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1261 #ifdef CONFIG_TCP_MD5SIG
1262 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1265 tcp_clear_options(&tmp_opt);
1266 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1267 tmp_opt.user_mss = tp->rx_opt.user_mss;
1268 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1270 if (tmp_opt.cookie_plus > 0 &&
1271 tmp_opt.saw_tstamp &&
1272 !tp->rx_opt.cookie_out_never &&
1273 (sysctl_tcp_cookie_size > 0 ||
1274 (tp->cookie_values != NULL &&
1275 tp->cookie_values->cookie_desired > 0))) {
1277 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1278 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1280 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1281 goto drop_and_release;
1283 /* Secret recipe starts with IP addresses */
1287 /* plus variable length Initiator Cookie */
1290 *c++ ^= *hash_location++;
1292 #ifdef CONFIG_SYN_COOKIES
1293 want_cookie = 0; /* not our kind of cookie */
1295 tmp_ext.cookie_out_never = 0; /* false */
1296 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1297 } else if (!tp->rx_opt.cookie_in_always) {
1298 /* redundant indications, but ensure initialization. */
1299 tmp_ext.cookie_out_never = 1; /* true */
1300 tmp_ext.cookie_plus = 0;
1302 goto drop_and_release;
1304 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1306 if (want_cookie && !tmp_opt.saw_tstamp)
1307 tcp_clear_options(&tmp_opt);
1309 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1310 tcp_openreq_init(req, &tmp_opt, skb);
1312 ireq = inet_rsk(req);
1313 ireq->loc_addr = daddr;
1314 ireq->rmt_addr = saddr;
1315 ireq->no_srccheck = inet_sk(sk)->transparent;
1316 ireq->opt = tcp_v4_save_options(sk, skb);
1318 if (security_inet_conn_request(sk, skb, req))
1322 TCP_ECN_create_request(req, tcp_hdr(skb));
1325 #ifdef CONFIG_SYN_COOKIES
1326 syn_flood_warning(skb);
1327 req->cookie_ts = tmp_opt.tstamp_ok;
1329 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1331 struct inet_peer *peer = NULL;
1333 /* VJ's idea. We save last timestamp seen
1334 * from the destination in peer table, when entering
1335 * state TIME-WAIT, and check against it before
1336 * accepting new connection request.
1338 * If "isn" is not zero, this request hit alive
1339 * timewait bucket, so that all the necessary checks
1340 * are made in the function processing timewait state.
1342 if (tmp_opt.saw_tstamp &&
1343 tcp_death_row.sysctl_tw_recycle &&
1344 (dst = inet_csk_route_req(sk, req)) != NULL &&
1345 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1346 peer->v4daddr == saddr) {
1347 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1348 (s32)(peer->tcp_ts - req->ts_recent) >
1350 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1351 goto drop_and_release;
1354 /* Kill the following clause, if you dislike this way. */
1355 else if (!sysctl_tcp_syncookies &&
1356 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1357 (sysctl_max_syn_backlog >> 2)) &&
1358 (!peer || !peer->tcp_ts_stamp) &&
1359 (!dst || !dst_metric(dst, RTAX_RTT))) {
1360 /* Without syncookies last quarter of
1361 * backlog is filled with destinations,
1362 * proven to be alive.
1363 * It means that we continue to communicate
1364 * to destinations, already remembered
1365 * to the moment of synflood.
1367 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1368 &saddr, ntohs(tcp_hdr(skb)->source));
1369 goto drop_and_release;
1372 isn = tcp_v4_init_sequence(skb);
1374 tcp_rsk(req)->snt_isn = isn;
1376 if (__tcp_v4_send_synack(sk, dst, req,
1377 (struct request_values *)&tmp_ext) ||
1381 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1394 * The three way handshake has completed - we got a valid synack -
1395 * now create the new socket.
1397 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1398 struct request_sock *req,
1399 struct dst_entry *dst)
1401 struct inet_request_sock *ireq;
1402 struct inet_sock *newinet;
1403 struct tcp_sock *newtp;
1405 #ifdef CONFIG_TCP_MD5SIG
1406 struct tcp_md5sig_key *key;
1409 if (sk_acceptq_is_full(sk))
1412 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1415 newsk = tcp_create_openreq_child(sk, req, skb);
1419 newsk->sk_gso_type = SKB_GSO_TCPV4;
1420 sk_setup_caps(newsk, dst);
1422 newtp = tcp_sk(newsk);
1423 newinet = inet_sk(newsk);
1424 ireq = inet_rsk(req);
1425 newinet->inet_daddr = ireq->rmt_addr;
1426 newinet->inet_rcv_saddr = ireq->loc_addr;
1427 newinet->inet_saddr = ireq->loc_addr;
1428 newinet->opt = ireq->opt;
1430 newinet->mc_index = inet_iif(skb);
1431 newinet->mc_ttl = ip_hdr(skb)->ttl;
1432 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1434 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
1435 newinet->inet_id = newtp->write_seq ^ jiffies;
1437 tcp_mtup_init(newsk);
1438 tcp_sync_mss(newsk, dst_mtu(dst));
1439 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1440 if (tcp_sk(sk)->rx_opt.user_mss &&
1441 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1442 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1444 tcp_initialize_rcv_mss(newsk);
1446 #ifdef CONFIG_TCP_MD5SIG
1447 /* Copy over the MD5 key from the original socket */
1448 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1451 * We're using one, so create a matching key
1452 * on the newsk structure. If we fail to get
1453 * memory, then we end up not copying the key
1456 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1458 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1459 newkey, key->keylen);
1460 newsk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1464 __inet_hash_nolisten(newsk, NULL);
1465 __inet_inherit_port(sk, newsk);
1470 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1472 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1477 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1479 struct tcphdr *th = tcp_hdr(skb);
1480 const struct iphdr *iph = ip_hdr(skb);
1482 struct request_sock **prev;
1483 /* Find possible connection requests. */
1484 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1485 iph->saddr, iph->daddr);
1487 return tcp_check_req(sk, skb, req, prev);
1489 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1490 th->source, iph->daddr, th->dest, inet_iif(skb));
1493 if (nsk->sk_state != TCP_TIME_WAIT) {
1497 inet_twsk_put(inet_twsk(nsk));
1501 #ifdef CONFIG_SYN_COOKIES
1502 if (!th->rst && !th->syn && th->ack)
1503 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1508 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1510 const struct iphdr *iph = ip_hdr(skb);
1512 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1513 if (!tcp_v4_check(skb->len, iph->saddr,
1514 iph->daddr, skb->csum)) {
1515 skb->ip_summed = CHECKSUM_UNNECESSARY;
1520 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1521 skb->len, IPPROTO_TCP, 0);
1523 if (skb->len <= 76) {
1524 return __skb_checksum_complete(skb);
1530 /* The socket must have it's spinlock held when we get
1533 * We have a potential double-lock case here, so even when
1534 * doing backlog processing we use the BH locking scheme.
1535 * This is because we cannot sleep with the original spinlock
1538 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1541 #ifdef CONFIG_TCP_MD5SIG
1543 * We really want to reject the packet as early as possible
1545 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1546 * o There is an MD5 option and we're not expecting one
1548 if (tcp_v4_inbound_md5_hash(sk, skb))
1552 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1553 TCP_CHECK_TIMER(sk);
1554 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1558 TCP_CHECK_TIMER(sk);
1562 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1565 if (sk->sk_state == TCP_LISTEN) {
1566 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1571 if (tcp_child_process(sk, nsk, skb)) {
1579 TCP_CHECK_TIMER(sk);
1580 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1584 TCP_CHECK_TIMER(sk);
1588 tcp_v4_send_reset(rsk, skb);
1591 /* Be careful here. If this function gets more complicated and
1592 * gcc suffers from register pressure on the x86, sk (in %ebx)
1593 * might be destroyed here. This current version compiles correctly,
1594 * but you have been warned.
1599 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1607 int tcp_v4_rcv(struct sk_buff *skb)
1609 const struct iphdr *iph;
1613 struct net *net = dev_net(skb->dev);
1615 if (skb->pkt_type != PACKET_HOST)
1618 /* Count it even if it's bad */
1619 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1621 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1626 if (th->doff < sizeof(struct tcphdr) / 4)
1628 if (!pskb_may_pull(skb, th->doff * 4))
1631 /* An explanation is required here, I think.
1632 * Packet length and doff are validated by header prediction,
1633 * provided case of th->doff==0 is eliminated.
1634 * So, we defer the checks. */
1635 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1640 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1641 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1642 skb->len - th->doff * 4);
1643 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1644 TCP_SKB_CB(skb)->when = 0;
1645 TCP_SKB_CB(skb)->flags = iph->tos;
1646 TCP_SKB_CB(skb)->sacked = 0;
1648 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1653 if (sk->sk_state == TCP_TIME_WAIT)
1656 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1657 goto discard_and_relse;
1660 if (sk_filter(sk, skb))
1661 goto discard_and_relse;
1665 bh_lock_sock_nested(sk);
1667 if (!sock_owned_by_user(sk)) {
1668 #ifdef CONFIG_NET_DMA
1669 struct tcp_sock *tp = tcp_sk(sk);
1670 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1671 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1672 if (tp->ucopy.dma_chan)
1673 ret = tcp_v4_do_rcv(sk, skb);
1677 if (!tcp_prequeue(sk, skb))
1678 ret = tcp_v4_do_rcv(sk, skb);
1681 sk_add_backlog(sk, skb);
1689 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1692 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1694 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1696 tcp_v4_send_reset(NULL, skb);
1700 /* Discard frame. */
1709 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1710 inet_twsk_put(inet_twsk(sk));
1714 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1715 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1716 inet_twsk_put(inet_twsk(sk));
1719 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1721 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1723 iph->daddr, th->dest,
1726 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1727 inet_twsk_put(inet_twsk(sk));
1731 /* Fall through to ACK */
1734 tcp_v4_timewait_ack(sk, skb);
1738 case TCP_TW_SUCCESS:;
1743 /* VJ's idea. Save last timestamp seen from this destination
1744 * and hold it at least for normal timewait interval to use for duplicate
1745 * segment detection in subsequent connections, before they enter synchronized
1749 int tcp_v4_remember_stamp(struct sock *sk)
1751 struct inet_sock *inet = inet_sk(sk);
1752 struct tcp_sock *tp = tcp_sk(sk);
1753 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1754 struct inet_peer *peer = NULL;
1757 if (!rt || rt->rt_dst != inet->inet_daddr) {
1758 peer = inet_getpeer(inet->inet_daddr, 1);
1762 rt_bind_peer(rt, 1);
1767 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1768 ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1769 peer->tcp_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
1770 peer->tcp_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
1771 peer->tcp_ts = tp->rx_opt.ts_recent;
1781 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1783 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1786 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1788 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1789 ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL &&
1790 peer->tcp_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
1791 peer->tcp_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
1792 peer->tcp_ts = tcptw->tw_ts_recent;
1801 const struct inet_connection_sock_af_ops ipv4_specific = {
1802 .queue_xmit = ip_queue_xmit,
1803 .send_check = tcp_v4_send_check,
1804 .rebuild_header = inet_sk_rebuild_header,
1805 .conn_request = tcp_v4_conn_request,
1806 .syn_recv_sock = tcp_v4_syn_recv_sock,
1807 .remember_stamp = tcp_v4_remember_stamp,
1808 .net_header_len = sizeof(struct iphdr),
1809 .setsockopt = ip_setsockopt,
1810 .getsockopt = ip_getsockopt,
1811 .addr2sockaddr = inet_csk_addr2sockaddr,
1812 .sockaddr_len = sizeof(struct sockaddr_in),
1813 .bind_conflict = inet_csk_bind_conflict,
1814 #ifdef CONFIG_COMPAT
1815 .compat_setsockopt = compat_ip_setsockopt,
1816 .compat_getsockopt = compat_ip_getsockopt,
1820 #ifdef CONFIG_TCP_MD5SIG
1821 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1822 .md5_lookup = tcp_v4_md5_lookup,
1823 .calc_md5_hash = tcp_v4_md5_hash_skb,
1824 .md5_add = tcp_v4_md5_add_func,
1825 .md5_parse = tcp_v4_parse_md5_keys,
1829 /* NOTE: A lot of things set to zero explicitly by call to
1830 * sk_alloc() so need not be done here.
1832 static int tcp_v4_init_sock(struct sock *sk)
1834 struct inet_connection_sock *icsk = inet_csk(sk);
1835 struct tcp_sock *tp = tcp_sk(sk);
1837 skb_queue_head_init(&tp->out_of_order_queue);
1838 tcp_init_xmit_timers(sk);
1839 tcp_prequeue_init(tp);
1841 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1842 tp->mdev = TCP_TIMEOUT_INIT;
1844 /* So many TCP implementations out there (incorrectly) count the
1845 * initial SYN frame in their delayed-ACK and congestion control
1846 * algorithms that we must have the following bandaid to talk
1847 * efficiently to them. -DaveM
1851 /* See draft-stevens-tcpca-spec-01 for discussion of the
1852 * initialization of these values.
1854 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1855 tp->snd_cwnd_clamp = ~0;
1856 tp->mss_cache = TCP_MSS_DEFAULT;
1858 tp->reordering = sysctl_tcp_reordering;
1859 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1861 sk->sk_state = TCP_CLOSE;
1863 sk->sk_write_space = sk_stream_write_space;
1864 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1866 icsk->icsk_af_ops = &ipv4_specific;
1867 icsk->icsk_sync_mss = tcp_sync_mss;
1868 #ifdef CONFIG_TCP_MD5SIG
1869 tp->af_specific = &tcp_sock_ipv4_specific;
1872 /* TCP Cookie Transactions */
1873 if (sysctl_tcp_cookie_size > 0) {
1874 /* Default, cookies without s_data_payload. */
1876 kzalloc(sizeof(*tp->cookie_values),
1878 if (tp->cookie_values != NULL)
1879 kref_init(&tp->cookie_values->kref);
1881 /* Presumed zeroed, in order of appearance:
1882 * cookie_in_always, cookie_out_never,
1883 * s_data_constant, s_data_in, s_data_out
1885 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1886 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1889 percpu_counter_inc(&tcp_sockets_allocated);
1895 void tcp_v4_destroy_sock(struct sock *sk)
1897 struct tcp_sock *tp = tcp_sk(sk);
1899 tcp_clear_xmit_timers(sk);
1901 tcp_cleanup_congestion_control(sk);
1903 /* Cleanup up the write buffer. */
1904 tcp_write_queue_purge(sk);
1906 /* Cleans up our, hopefully empty, out_of_order_queue. */
1907 __skb_queue_purge(&tp->out_of_order_queue);
1909 #ifdef CONFIG_TCP_MD5SIG
1910 /* Clean up the MD5 key list, if any */
1911 if (tp->md5sig_info) {
1912 tcp_v4_clear_md5_list(sk);
1913 kfree(tp->md5sig_info);
1914 tp->md5sig_info = NULL;
1918 #ifdef CONFIG_NET_DMA
1919 /* Cleans up our sk_async_wait_queue */
1920 __skb_queue_purge(&sk->sk_async_wait_queue);
1923 /* Clean prequeue, it must be empty really */
1924 __skb_queue_purge(&tp->ucopy.prequeue);
1926 /* Clean up a referenced TCP bind bucket. */
1927 if (inet_csk(sk)->icsk_bind_hash)
1931 * If sendmsg cached page exists, toss it.
1933 if (sk->sk_sndmsg_page) {
1934 __free_page(sk->sk_sndmsg_page);
1935 sk->sk_sndmsg_page = NULL;
1938 /* TCP Cookie Transactions */
1939 if (tp->cookie_values != NULL) {
1940 kref_put(&tp->cookie_values->kref,
1941 tcp_cookie_values_release);
1942 tp->cookie_values = NULL;
1945 percpu_counter_dec(&tcp_sockets_allocated);
1948 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1950 #ifdef CONFIG_PROC_FS
1951 /* Proc filesystem TCP sock list dumping. */
1953 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1955 return hlist_nulls_empty(head) ? NULL :
1956 list_entry(head->first, struct inet_timewait_sock, tw_node);
1959 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1961 return !is_a_nulls(tw->tw_node.next) ?
1962 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1965 static void *listening_get_next(struct seq_file *seq, void *cur)
1967 struct inet_connection_sock *icsk;
1968 struct hlist_nulls_node *node;
1969 struct sock *sk = cur;
1970 struct inet_listen_hashbucket *ilb;
1971 struct tcp_iter_state *st = seq->private;
1972 struct net *net = seq_file_net(seq);
1976 ilb = &tcp_hashinfo.listening_hash[0];
1977 spin_lock_bh(&ilb->lock);
1978 sk = sk_nulls_head(&ilb->head);
1981 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1984 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1985 struct request_sock *req = cur;
1987 icsk = inet_csk(st->syn_wait_sk);
1991 if (req->rsk_ops->family == st->family) {
1997 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2000 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2002 sk = sk_next(st->syn_wait_sk);
2003 st->state = TCP_SEQ_STATE_LISTENING;
2004 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2006 icsk = inet_csk(sk);
2007 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2008 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2010 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2014 sk_nulls_for_each_from(sk, node) {
2015 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) {
2019 icsk = inet_csk(sk);
2020 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2021 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2023 st->uid = sock_i_uid(sk);
2024 st->syn_wait_sk = sk;
2025 st->state = TCP_SEQ_STATE_OPENREQ;
2029 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2031 spin_unlock_bh(&ilb->lock);
2032 if (++st->bucket < INET_LHTABLE_SIZE) {
2033 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2034 spin_lock_bh(&ilb->lock);
2035 sk = sk_nulls_head(&ilb->head);
2043 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2045 void *rc = listening_get_next(seq, NULL);
2047 while (rc && *pos) {
2048 rc = listening_get_next(seq, rc);
2054 static inline int empty_bucket(struct tcp_iter_state *st)
2056 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2057 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2060 static void *established_get_first(struct seq_file *seq)
2062 struct tcp_iter_state *st = seq->private;
2063 struct net *net = seq_file_net(seq);
2066 for (st->bucket = 0; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2068 struct hlist_nulls_node *node;
2069 struct inet_timewait_sock *tw;
2070 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2072 /* Lockless fast path for the common case of empty buckets */
2073 if (empty_bucket(st))
2077 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2078 if (sk->sk_family != st->family ||
2079 !net_eq(sock_net(sk), net)) {
2085 st->state = TCP_SEQ_STATE_TIME_WAIT;
2086 inet_twsk_for_each(tw, node,
2087 &tcp_hashinfo.ehash[st->bucket].twchain) {
2088 if (tw->tw_family != st->family ||
2089 !net_eq(twsk_net(tw), net)) {
2095 spin_unlock_bh(lock);
2096 st->state = TCP_SEQ_STATE_ESTABLISHED;
2102 static void *established_get_next(struct seq_file *seq, void *cur)
2104 struct sock *sk = cur;
2105 struct inet_timewait_sock *tw;
2106 struct hlist_nulls_node *node;
2107 struct tcp_iter_state *st = seq->private;
2108 struct net *net = seq_file_net(seq);
2112 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2116 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2123 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2124 st->state = TCP_SEQ_STATE_ESTABLISHED;
2126 /* Look for next non empty bucket */
2127 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2130 if (st->bucket > tcp_hashinfo.ehash_mask)
2133 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2134 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2136 sk = sk_nulls_next(sk);
2138 sk_nulls_for_each_from(sk, node) {
2139 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2143 st->state = TCP_SEQ_STATE_TIME_WAIT;
2144 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2152 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2154 void *rc = established_get_first(seq);
2157 rc = established_get_next(seq, rc);
2163 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2166 struct tcp_iter_state *st = seq->private;
2168 st->state = TCP_SEQ_STATE_LISTENING;
2169 rc = listening_get_idx(seq, &pos);
2172 st->state = TCP_SEQ_STATE_ESTABLISHED;
2173 rc = established_get_idx(seq, pos);
2179 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2181 struct tcp_iter_state *st = seq->private;
2182 st->state = TCP_SEQ_STATE_LISTENING;
2184 return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2187 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2190 struct tcp_iter_state *st;
2192 if (v == SEQ_START_TOKEN) {
2193 rc = tcp_get_idx(seq, 0);
2198 switch (st->state) {
2199 case TCP_SEQ_STATE_OPENREQ:
2200 case TCP_SEQ_STATE_LISTENING:
2201 rc = listening_get_next(seq, v);
2203 st->state = TCP_SEQ_STATE_ESTABLISHED;
2204 rc = established_get_first(seq);
2207 case TCP_SEQ_STATE_ESTABLISHED:
2208 case TCP_SEQ_STATE_TIME_WAIT:
2209 rc = established_get_next(seq, v);
2217 static void tcp_seq_stop(struct seq_file *seq, void *v)
2219 struct tcp_iter_state *st = seq->private;
2221 switch (st->state) {
2222 case TCP_SEQ_STATE_OPENREQ:
2224 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2225 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2227 case TCP_SEQ_STATE_LISTENING:
2228 if (v != SEQ_START_TOKEN)
2229 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2231 case TCP_SEQ_STATE_TIME_WAIT:
2232 case TCP_SEQ_STATE_ESTABLISHED:
2234 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2239 static int tcp_seq_open(struct inode *inode, struct file *file)
2241 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2242 struct tcp_iter_state *s;
2245 err = seq_open_net(inode, file, &afinfo->seq_ops,
2246 sizeof(struct tcp_iter_state));
2250 s = ((struct seq_file *)file->private_data)->private;
2251 s->family = afinfo->family;
2255 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2258 struct proc_dir_entry *p;
2260 afinfo->seq_fops.open = tcp_seq_open;
2261 afinfo->seq_fops.read = seq_read;
2262 afinfo->seq_fops.llseek = seq_lseek;
2263 afinfo->seq_fops.release = seq_release_net;
2265 afinfo->seq_ops.start = tcp_seq_start;
2266 afinfo->seq_ops.next = tcp_seq_next;
2267 afinfo->seq_ops.stop = tcp_seq_stop;
2269 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2270 &afinfo->seq_fops, afinfo);
2276 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2278 proc_net_remove(net, afinfo->name);
2281 static void get_openreq4(struct sock *sk, struct request_sock *req,
2282 struct seq_file *f, int i, int uid, int *len)
2284 const struct inet_request_sock *ireq = inet_rsk(req);
2285 int ttd = req->expires - jiffies;
2287 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2288 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
2291 ntohs(inet_sk(sk)->inet_sport),
2293 ntohs(ireq->rmt_port),
2295 0, 0, /* could print option size, but that is af dependent. */
2296 1, /* timers active (only the expire timer) */
2297 jiffies_to_clock_t(ttd),
2300 0, /* non standard timer */
2301 0, /* open_requests have no inode */
2302 atomic_read(&sk->sk_refcnt),
2307 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2310 unsigned long timer_expires;
2311 struct tcp_sock *tp = tcp_sk(sk);
2312 const struct inet_connection_sock *icsk = inet_csk(sk);
2313 struct inet_sock *inet = inet_sk(sk);
2314 __be32 dest = inet->inet_daddr;
2315 __be32 src = inet->inet_rcv_saddr;
2316 __u16 destp = ntohs(inet->inet_dport);
2317 __u16 srcp = ntohs(inet->inet_sport);
2320 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2322 timer_expires = icsk->icsk_timeout;
2323 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2325 timer_expires = icsk->icsk_timeout;
2326 } else if (timer_pending(&sk->sk_timer)) {
2328 timer_expires = sk->sk_timer.expires;
2331 timer_expires = jiffies;
2334 if (sk->sk_state == TCP_LISTEN)
2335 rx_queue = sk->sk_ack_backlog;
2338 * because we dont lock socket, we might find a transient negative value
2340 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2342 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2343 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
2344 i, src, srcp, dest, destp, sk->sk_state,
2345 tp->write_seq - tp->snd_una,
2348 jiffies_to_clock_t(timer_expires - jiffies),
2349 icsk->icsk_retransmits,
2351 icsk->icsk_probes_out,
2353 atomic_read(&sk->sk_refcnt), sk,
2354 jiffies_to_clock_t(icsk->icsk_rto),
2355 jiffies_to_clock_t(icsk->icsk_ack.ato),
2356 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2358 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2362 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2363 struct seq_file *f, int i, int *len)
2367 int ttd = tw->tw_ttd - jiffies;
2372 dest = tw->tw_daddr;
2373 src = tw->tw_rcv_saddr;
2374 destp = ntohs(tw->tw_dport);
2375 srcp = ntohs(tw->tw_sport);
2377 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2378 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
2379 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2380 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2381 atomic_read(&tw->tw_refcnt), tw, len);
2386 static int tcp4_seq_show(struct seq_file *seq, void *v)
2388 struct tcp_iter_state *st;
2391 if (v == SEQ_START_TOKEN) {
2392 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2393 " sl local_address rem_address st tx_queue "
2394 "rx_queue tr tm->when retrnsmt uid timeout "
2400 switch (st->state) {
2401 case TCP_SEQ_STATE_LISTENING:
2402 case TCP_SEQ_STATE_ESTABLISHED:
2403 get_tcp4_sock(v, seq, st->num, &len);
2405 case TCP_SEQ_STATE_OPENREQ:
2406 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2408 case TCP_SEQ_STATE_TIME_WAIT:
2409 get_timewait4_sock(v, seq, st->num, &len);
2412 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2417 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2421 .owner = THIS_MODULE,
2424 .show = tcp4_seq_show,
2428 static int tcp4_proc_init_net(struct net *net)
2430 return tcp_proc_register(net, &tcp4_seq_afinfo);
2433 static void tcp4_proc_exit_net(struct net *net)
2435 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2438 static struct pernet_operations tcp4_net_ops = {
2439 .init = tcp4_proc_init_net,
2440 .exit = tcp4_proc_exit_net,
2443 int __init tcp4_proc_init(void)
2445 return register_pernet_subsys(&tcp4_net_ops);
2448 void tcp4_proc_exit(void)
2450 unregister_pernet_subsys(&tcp4_net_ops);
2452 #endif /* CONFIG_PROC_FS */
2454 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2456 struct iphdr *iph = skb_gro_network_header(skb);
2458 switch (skb->ip_summed) {
2459 case CHECKSUM_COMPLETE:
2460 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2462 skb->ip_summed = CHECKSUM_UNNECESSARY;
2468 NAPI_GRO_CB(skb)->flush = 1;
2472 return tcp_gro_receive(head, skb);
2474 EXPORT_SYMBOL(tcp4_gro_receive);
2476 int tcp4_gro_complete(struct sk_buff *skb)
2478 struct iphdr *iph = ip_hdr(skb);
2479 struct tcphdr *th = tcp_hdr(skb);
2481 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2482 iph->saddr, iph->daddr, 0);
2483 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2485 return tcp_gro_complete(skb);
2487 EXPORT_SYMBOL(tcp4_gro_complete);
2489 struct proto tcp_prot = {
2491 .owner = THIS_MODULE,
2493 .connect = tcp_v4_connect,
2494 .disconnect = tcp_disconnect,
2495 .accept = inet_csk_accept,
2497 .init = tcp_v4_init_sock,
2498 .destroy = tcp_v4_destroy_sock,
2499 .shutdown = tcp_shutdown,
2500 .setsockopt = tcp_setsockopt,
2501 .getsockopt = tcp_getsockopt,
2502 .recvmsg = tcp_recvmsg,
2503 .backlog_rcv = tcp_v4_do_rcv,
2505 .unhash = inet_unhash,
2506 .get_port = inet_csk_get_port,
2507 .enter_memory_pressure = tcp_enter_memory_pressure,
2508 .sockets_allocated = &tcp_sockets_allocated,
2509 .orphan_count = &tcp_orphan_count,
2510 .memory_allocated = &tcp_memory_allocated,
2511 .memory_pressure = &tcp_memory_pressure,
2512 .sysctl_mem = sysctl_tcp_mem,
2513 .sysctl_wmem = sysctl_tcp_wmem,
2514 .sysctl_rmem = sysctl_tcp_rmem,
2515 .max_header = MAX_TCP_HEADER,
2516 .obj_size = sizeof(struct tcp_sock),
2517 .slab_flags = SLAB_DESTROY_BY_RCU,
2518 .twsk_prot = &tcp_timewait_sock_ops,
2519 .rsk_prot = &tcp_request_sock_ops,
2520 .h.hashinfo = &tcp_hashinfo,
2521 #ifdef CONFIG_COMPAT
2522 .compat_setsockopt = compat_tcp_setsockopt,
2523 .compat_getsockopt = compat_tcp_getsockopt,
2528 static int __net_init tcp_sk_init(struct net *net)
2530 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2531 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2534 static void __net_exit tcp_sk_exit(struct net *net)
2536 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2539 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2541 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2544 static struct pernet_operations __net_initdata tcp_sk_ops = {
2545 .init = tcp_sk_init,
2546 .exit = tcp_sk_exit,
2547 .exit_batch = tcp_sk_exit_batch,
2550 void __init tcp_v4_init(void)
2552 inet_hashinfo_init(&tcp_hashinfo);
2553 if (register_pernet_subsys(&tcp_sk_ops))
2554 panic("Failed to create the TCP control socket.\n");
2557 EXPORT_SYMBOL(ipv4_specific);
2558 EXPORT_SYMBOL(tcp_hashinfo);
2559 EXPORT_SYMBOL(tcp_prot);
2560 EXPORT_SYMBOL(tcp_v4_conn_request);
2561 EXPORT_SYMBOL(tcp_v4_connect);
2562 EXPORT_SYMBOL(tcp_v4_do_rcv);
2563 EXPORT_SYMBOL(tcp_v4_remember_stamp);
2564 EXPORT_SYMBOL(tcp_v4_send_check);
2565 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2567 #ifdef CONFIG_PROC_FS
2568 EXPORT_SYMBOL(tcp_proc_register);
2569 EXPORT_SYMBOL(tcp_proc_unregister);
2571 EXPORT_SYMBOL(sysctl_tcp_low_latency);