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>
75 #include <net/secure_seq.h>
77 #include <linux/inet.h>
78 #include <linux/ipv6.h>
79 #include <linux/stddef.h>
80 #include <linux/proc_fs.h>
81 #include <linux/seq_file.h>
83 #include <linux/crypto.h>
84 #include <linux/scatterlist.h>
86 int sysctl_tcp_tw_reuse __read_mostly;
87 int sysctl_tcp_low_latency __read_mostly;
88 EXPORT_SYMBOL(sysctl_tcp_low_latency);
91 #ifdef CONFIG_TCP_MD5SIG
92 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
94 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
95 __be32 daddr, __be32 saddr, struct tcphdr *th);
98 struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
104 struct inet_hashinfo tcp_hashinfo;
105 EXPORT_SYMBOL(tcp_hashinfo);
107 static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
109 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
112 tcp_hdr(skb)->source);
115 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
117 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
118 struct tcp_sock *tp = tcp_sk(sk);
120 /* With PAWS, it is safe from the viewpoint
121 of data integrity. Even without PAWS it is safe provided sequence
122 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
124 Actually, the idea is close to VJ's one, only timestamp cache is
125 held not per host, but per port pair and TW bucket is used as state
128 If TW bucket has been already destroyed we fall back to VJ's scheme
129 and use initial timestamp retrieved from peer table.
131 if (tcptw->tw_ts_recent_stamp &&
132 (twp == NULL || (sysctl_tcp_tw_reuse &&
133 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
134 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
135 if (tp->write_seq == 0)
137 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
138 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
145 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
147 /* This will initiate an outgoing connection. */
148 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
150 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
151 struct inet_sock *inet = inet_sk(sk);
152 struct tcp_sock *tp = tcp_sk(sk);
153 __be16 orig_sport, orig_dport;
154 __be32 daddr, nexthop;
158 struct ip_options_rcu *inet_opt;
160 if (addr_len < sizeof(struct sockaddr_in))
163 if (usin->sin_family != AF_INET)
164 return -EAFNOSUPPORT;
166 nexthop = daddr = usin->sin_addr.s_addr;
167 inet_opt = rcu_dereference_protected(inet->inet_opt,
168 sock_owned_by_user(sk));
169 if (inet_opt && inet_opt->opt.srr) {
172 nexthop = inet_opt->opt.faddr;
175 orig_sport = inet->inet_sport;
176 orig_dport = usin->sin_port;
177 fl4 = &inet->cork.fl.u.ip4;
178 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
179 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
181 orig_sport, orig_dport, sk, true);
184 if (err == -ENETUNREACH)
185 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
189 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
194 if (!inet_opt || !inet_opt->opt.srr)
197 if (!inet->inet_saddr)
198 inet->inet_saddr = fl4->saddr;
199 inet->inet_rcv_saddr = inet->inet_saddr;
201 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
202 /* Reset inherited state */
203 tp->rx_opt.ts_recent = 0;
204 tp->rx_opt.ts_recent_stamp = 0;
208 if (tcp_death_row.sysctl_tw_recycle &&
209 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) {
210 struct inet_peer *peer = rt_get_peer(rt, fl4->daddr);
212 * VJ's idea. We save last timestamp seen from
213 * the destination in peer table, when entering state
214 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
215 * when trying new connection.
218 inet_peer_refcheck(peer);
219 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
220 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
221 tp->rx_opt.ts_recent = peer->tcp_ts;
226 inet->inet_dport = usin->sin_port;
227 inet->inet_daddr = daddr;
229 inet_csk(sk)->icsk_ext_hdr_len = 0;
231 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
233 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
235 /* Socket identity is still unknown (sport may be zero).
236 * However we set state to SYN-SENT and not releasing socket
237 * lock select source port, enter ourselves into the hash tables and
238 * complete initialization after this.
240 tcp_set_state(sk, TCP_SYN_SENT);
241 err = inet_hash_connect(&tcp_death_row, sk);
245 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
246 inet->inet_sport, inet->inet_dport, sk);
252 /* OK, now commit destination to socket. */
253 sk->sk_gso_type = SKB_GSO_TCPV4;
254 sk_setup_caps(sk, &rt->dst);
257 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
262 inet->inet_id = tp->write_seq ^ jiffies;
264 err = tcp_connect(sk);
273 * This unhashes the socket and releases the local port,
276 tcp_set_state(sk, TCP_CLOSE);
278 sk->sk_route_caps = 0;
279 inet->inet_dport = 0;
282 EXPORT_SYMBOL(tcp_v4_connect);
285 * This routine does path mtu discovery as defined in RFC1191.
287 static void do_pmtu_discovery(struct sock *sk, const struct iphdr *iph, u32 mtu)
289 struct dst_entry *dst;
290 struct inet_sock *inet = inet_sk(sk);
292 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
293 * send out by Linux are always <576bytes so they should go through
296 if (sk->sk_state == TCP_LISTEN)
299 /* We don't check in the destentry if pmtu discovery is forbidden
300 * on this route. We just assume that no packet_to_big packets
301 * are send back when pmtu discovery is not active.
302 * There is a small race when the user changes this flag in the
303 * route, but I think that's acceptable.
305 if ((dst = __sk_dst_check(sk, 0)) == NULL)
308 dst->ops->update_pmtu(dst, mtu);
310 /* Something is about to be wrong... Remember soft error
311 * for the case, if this connection will not able to recover.
313 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
314 sk->sk_err_soft = EMSGSIZE;
318 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
319 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
320 tcp_sync_mss(sk, mtu);
322 /* Resend the TCP packet because it's
323 * clear that the old packet has been
324 * dropped. This is the new "fast" path mtu
327 tcp_simple_retransmit(sk);
328 } /* else let the usual retransmit timer handle it */
332 * This routine is called by the ICMP module when it gets some
333 * sort of error condition. If err < 0 then the socket should
334 * be closed and the error returned to the user. If err > 0
335 * it's just the icmp type << 8 | icmp code. After adjustment
336 * header points to the first 8 bytes of the tcp header. We need
337 * to find the appropriate port.
339 * The locking strategy used here is very "optimistic". When
340 * someone else accesses the socket the ICMP is just dropped
341 * and for some paths there is no check at all.
342 * A more general error queue to queue errors for later handling
343 * is probably better.
347 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
349 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
350 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
351 struct inet_connection_sock *icsk;
353 struct inet_sock *inet;
354 const int type = icmp_hdr(icmp_skb)->type;
355 const int code = icmp_hdr(icmp_skb)->code;
361 struct net *net = dev_net(icmp_skb->dev);
363 if (icmp_skb->len < (iph->ihl << 2) + 8) {
364 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
368 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
369 iph->saddr, th->source, inet_iif(icmp_skb));
371 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
374 if (sk->sk_state == TCP_TIME_WAIT) {
375 inet_twsk_put(inet_twsk(sk));
380 /* If too many ICMPs get dropped on busy
381 * servers this needs to be solved differently.
383 if (sock_owned_by_user(sk))
384 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
386 if (sk->sk_state == TCP_CLOSE)
389 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
390 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
396 seq = ntohl(th->seq);
397 if (sk->sk_state != TCP_LISTEN &&
398 !between(seq, tp->snd_una, tp->snd_nxt)) {
399 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
404 case ICMP_SOURCE_QUENCH:
405 /* Just silently ignore these. */
407 case ICMP_PARAMETERPROB:
410 case ICMP_DEST_UNREACH:
411 if (code > NR_ICMP_UNREACH)
414 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
415 if (!sock_owned_by_user(sk))
416 do_pmtu_discovery(sk, iph, info);
420 err = icmp_err_convert[code].errno;
421 /* check if icmp_skb allows revert of backoff
422 * (see draft-zimmermann-tcp-lcd) */
423 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
425 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
429 if (sock_owned_by_user(sk))
432 icsk->icsk_backoff--;
433 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
437 skb = tcp_write_queue_head(sk);
440 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
441 tcp_time_stamp - TCP_SKB_CB(skb)->when);
444 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
445 remaining, TCP_RTO_MAX);
447 /* RTO revert clocked out retransmission.
448 * Will retransmit now */
449 tcp_retransmit_timer(sk);
453 case ICMP_TIME_EXCEEDED:
460 switch (sk->sk_state) {
461 struct request_sock *req, **prev;
463 if (sock_owned_by_user(sk))
466 req = inet_csk_search_req(sk, &prev, th->dest,
467 iph->daddr, iph->saddr);
471 /* ICMPs are not backlogged, hence we cannot get
472 an established socket here.
476 if (seq != tcp_rsk(req)->snt_isn) {
477 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
482 * Still in SYN_RECV, just remove it silently.
483 * There is no good way to pass the error to the newly
484 * created socket, and POSIX does not want network
485 * errors returned from accept().
487 inet_csk_reqsk_queue_drop(sk, req, prev);
491 case TCP_SYN_RECV: /* Cannot happen.
492 It can f.e. if SYNs crossed.
494 if (!sock_owned_by_user(sk)) {
497 sk->sk_error_report(sk);
501 sk->sk_err_soft = err;
506 /* If we've already connected we will keep trying
507 * until we time out, or the user gives up.
509 * rfc1122 4.2.3.9 allows to consider as hard errors
510 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
511 * but it is obsoleted by pmtu discovery).
513 * Note, that in modern internet, where routing is unreliable
514 * and in each dark corner broken firewalls sit, sending random
515 * errors ordered by their masters even this two messages finally lose
516 * their original sense (even Linux sends invalid PORT_UNREACHs)
518 * Now we are in compliance with RFCs.
523 if (!sock_owned_by_user(sk) && inet->recverr) {
525 sk->sk_error_report(sk);
526 } else { /* Only an error on timeout */
527 sk->sk_err_soft = err;
535 static void __tcp_v4_send_check(struct sk_buff *skb,
536 __be32 saddr, __be32 daddr)
538 struct tcphdr *th = tcp_hdr(skb);
540 if (skb->ip_summed == CHECKSUM_PARTIAL) {
541 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
542 skb->csum_start = skb_transport_header(skb) - skb->head;
543 skb->csum_offset = offsetof(struct tcphdr, check);
545 th->check = tcp_v4_check(skb->len, saddr, daddr,
552 /* This routine computes an IPv4 TCP checksum. */
553 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
555 struct inet_sock *inet = inet_sk(sk);
557 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
559 EXPORT_SYMBOL(tcp_v4_send_check);
561 int tcp_v4_gso_send_check(struct sk_buff *skb)
563 const struct iphdr *iph;
566 if (!pskb_may_pull(skb, sizeof(*th)))
573 skb->ip_summed = CHECKSUM_PARTIAL;
574 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
579 * This routine will send an RST to the other tcp.
581 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
583 * Answer: if a packet caused RST, it is not for a socket
584 * existing in our system, if it is matched to a socket,
585 * it is just duplicate segment or bug in other side's TCP.
586 * So that we build reply only basing on parameters
587 * arrived with segment.
588 * Exception: precedence violation. We do not implement it in any case.
591 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
593 struct tcphdr *th = tcp_hdr(skb);
596 #ifdef CONFIG_TCP_MD5SIG
597 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
600 struct ip_reply_arg arg;
601 #ifdef CONFIG_TCP_MD5SIG
602 struct tcp_md5sig_key *key;
606 /* Never send a reset in response to a reset. */
610 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
613 /* Swap the send and the receive. */
614 memset(&rep, 0, sizeof(rep));
615 rep.th.dest = th->source;
616 rep.th.source = th->dest;
617 rep.th.doff = sizeof(struct tcphdr) / 4;
621 rep.th.seq = th->ack_seq;
624 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
625 skb->len - (th->doff << 2));
628 memset(&arg, 0, sizeof(arg));
629 arg.iov[0].iov_base = (unsigned char *)&rep;
630 arg.iov[0].iov_len = sizeof(rep.th);
632 #ifdef CONFIG_TCP_MD5SIG
633 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->saddr) : NULL;
635 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
637 (TCPOPT_MD5SIG << 8) |
639 /* Update length and the length the header thinks exists */
640 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
641 rep.th.doff = arg.iov[0].iov_len / 4;
643 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
644 key, ip_hdr(skb)->saddr,
645 ip_hdr(skb)->daddr, &rep.th);
648 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
649 ip_hdr(skb)->saddr, /* XXX */
650 arg.iov[0].iov_len, IPPROTO_TCP, 0);
651 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
652 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
653 /* When socket is gone, all binding information is lost.
654 * routing might fail in this case. No choice here, if we choose to force
655 * input interface, we will misroute in case of asymmetric route.
658 arg.bound_dev_if = sk->sk_bound_dev_if;
660 net = dev_net(skb_dst(skb)->dev);
661 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
662 &arg, arg.iov[0].iov_len);
664 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
665 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
668 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
669 outside socket context is ugly, certainly. What can I do?
672 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
673 u32 win, u32 ts, int oif,
674 struct tcp_md5sig_key *key,
677 struct tcphdr *th = tcp_hdr(skb);
680 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
681 #ifdef CONFIG_TCP_MD5SIG
682 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
686 struct ip_reply_arg arg;
687 struct net *net = dev_net(skb_dst(skb)->dev);
689 memset(&rep.th, 0, sizeof(struct tcphdr));
690 memset(&arg, 0, sizeof(arg));
692 arg.iov[0].iov_base = (unsigned char *)&rep;
693 arg.iov[0].iov_len = sizeof(rep.th);
695 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
696 (TCPOPT_TIMESTAMP << 8) |
698 rep.opt[1] = htonl(tcp_time_stamp);
699 rep.opt[2] = htonl(ts);
700 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
703 /* Swap the send and the receive. */
704 rep.th.dest = th->source;
705 rep.th.source = th->dest;
706 rep.th.doff = arg.iov[0].iov_len / 4;
707 rep.th.seq = htonl(seq);
708 rep.th.ack_seq = htonl(ack);
710 rep.th.window = htons(win);
712 #ifdef CONFIG_TCP_MD5SIG
714 int offset = (ts) ? 3 : 0;
716 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
718 (TCPOPT_MD5SIG << 8) |
720 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
721 rep.th.doff = arg.iov[0].iov_len/4;
723 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
724 key, ip_hdr(skb)->saddr,
725 ip_hdr(skb)->daddr, &rep.th);
728 arg.flags = reply_flags;
729 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
730 ip_hdr(skb)->saddr, /* XXX */
731 arg.iov[0].iov_len, IPPROTO_TCP, 0);
732 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
734 arg.bound_dev_if = oif;
736 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
737 &arg, arg.iov[0].iov_len);
739 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
742 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
744 struct inet_timewait_sock *tw = inet_twsk(sk);
745 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
747 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
748 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
751 tcp_twsk_md5_key(tcptw),
752 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
758 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
759 struct request_sock *req)
761 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
762 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
765 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
766 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
770 * Send a SYN-ACK after having received a SYN.
771 * This still operates on a request_sock only, not on a big
774 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
775 struct request_sock *req,
776 struct request_values *rvp)
778 const struct inet_request_sock *ireq = inet_rsk(req);
781 struct sk_buff * skb;
783 /* First, grab a route. */
784 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
787 skb = tcp_make_synack(sk, dst, req, rvp);
790 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
792 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
795 err = net_xmit_eval(err);
802 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
803 struct request_values *rvp)
805 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
806 return tcp_v4_send_synack(sk, NULL, req, rvp);
810 * IPv4 request_sock destructor.
812 static void tcp_v4_reqsk_destructor(struct request_sock *req)
814 kfree(inet_rsk(req)->opt);
817 static void syn_flood_warning(const struct sk_buff *skb)
821 #ifdef CONFIG_SYN_COOKIES
822 if (sysctl_tcp_syncookies)
823 msg = "Sending cookies";
826 msg = "Dropping request";
828 pr_info("TCP: Possible SYN flooding on port %d. %s.\n",
829 ntohs(tcp_hdr(skb)->dest), msg);
833 * Save and compile IPv4 options into the request_sock if needed.
835 static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk,
838 const struct ip_options *opt = &(IPCB(skb)->opt);
839 struct ip_options_rcu *dopt = NULL;
841 if (opt && opt->optlen) {
842 int opt_size = sizeof(*dopt) + opt->optlen;
844 dopt = kmalloc(opt_size, GFP_ATOMIC);
846 if (ip_options_echo(&dopt->opt, skb)) {
855 #ifdef CONFIG_TCP_MD5SIG
857 * RFC2385 MD5 checksumming requires a mapping of
858 * IP address->MD5 Key.
859 * We need to maintain these in the sk structure.
862 /* Find the Key structure for an address. */
863 static struct tcp_md5sig_key *
864 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
866 struct tcp_sock *tp = tcp_sk(sk);
869 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
871 for (i = 0; i < tp->md5sig_info->entries4; i++) {
872 if (tp->md5sig_info->keys4[i].addr == addr)
873 return &tp->md5sig_info->keys4[i].base;
878 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
879 struct sock *addr_sk)
881 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
883 EXPORT_SYMBOL(tcp_v4_md5_lookup);
885 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
886 struct request_sock *req)
888 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
891 /* This can be called on a newly created socket, from other files */
892 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
893 u8 *newkey, u8 newkeylen)
895 /* Add Key to the list */
896 struct tcp_md5sig_key *key;
897 struct tcp_sock *tp = tcp_sk(sk);
898 struct tcp4_md5sig_key *keys;
900 key = tcp_v4_md5_do_lookup(sk, addr);
902 /* Pre-existing entry - just update that one. */
905 key->keylen = newkeylen;
907 struct tcp_md5sig_info *md5sig;
909 if (!tp->md5sig_info) {
910 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
912 if (!tp->md5sig_info) {
916 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
919 md5sig = tp->md5sig_info;
920 if (md5sig->entries4 == 0 &&
921 tcp_alloc_md5sig_pool(sk) == NULL) {
926 if (md5sig->alloced4 == md5sig->entries4) {
927 keys = kmalloc((sizeof(*keys) *
928 (md5sig->entries4 + 1)), GFP_ATOMIC);
931 if (md5sig->entries4 == 0)
932 tcp_free_md5sig_pool();
936 if (md5sig->entries4)
937 memcpy(keys, md5sig->keys4,
938 sizeof(*keys) * md5sig->entries4);
940 /* Free old key list, and reference new one */
941 kfree(md5sig->keys4);
942 md5sig->keys4 = keys;
946 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
947 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
948 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
952 EXPORT_SYMBOL(tcp_v4_md5_do_add);
954 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
955 u8 *newkey, u8 newkeylen)
957 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
961 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
963 struct tcp_sock *tp = tcp_sk(sk);
966 for (i = 0; i < tp->md5sig_info->entries4; i++) {
967 if (tp->md5sig_info->keys4[i].addr == addr) {
969 kfree(tp->md5sig_info->keys4[i].base.key);
970 tp->md5sig_info->entries4--;
972 if (tp->md5sig_info->entries4 == 0) {
973 kfree(tp->md5sig_info->keys4);
974 tp->md5sig_info->keys4 = NULL;
975 tp->md5sig_info->alloced4 = 0;
976 tcp_free_md5sig_pool();
977 } else if (tp->md5sig_info->entries4 != i) {
978 /* Need to do some manipulation */
979 memmove(&tp->md5sig_info->keys4[i],
980 &tp->md5sig_info->keys4[i+1],
981 (tp->md5sig_info->entries4 - i) *
982 sizeof(struct tcp4_md5sig_key));
989 EXPORT_SYMBOL(tcp_v4_md5_do_del);
991 static void tcp_v4_clear_md5_list(struct sock *sk)
993 struct tcp_sock *tp = tcp_sk(sk);
995 /* Free each key, then the set of key keys,
996 * the crypto element, and then decrement our
997 * hold on the last resort crypto.
999 if (tp->md5sig_info->entries4) {
1001 for (i = 0; i < tp->md5sig_info->entries4; i++)
1002 kfree(tp->md5sig_info->keys4[i].base.key);
1003 tp->md5sig_info->entries4 = 0;
1004 tcp_free_md5sig_pool();
1006 if (tp->md5sig_info->keys4) {
1007 kfree(tp->md5sig_info->keys4);
1008 tp->md5sig_info->keys4 = NULL;
1009 tp->md5sig_info->alloced4 = 0;
1013 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1016 struct tcp_md5sig cmd;
1017 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1020 if (optlen < sizeof(cmd))
1023 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1026 if (sin->sin_family != AF_INET)
1029 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1030 if (!tcp_sk(sk)->md5sig_info)
1032 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1035 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1038 if (!tcp_sk(sk)->md5sig_info) {
1039 struct tcp_sock *tp = tcp_sk(sk);
1040 struct tcp_md5sig_info *p;
1042 p = kzalloc(sizeof(*p), sk->sk_allocation);
1046 tp->md5sig_info = p;
1047 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1050 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1053 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1054 newkey, cmd.tcpm_keylen);
1057 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1058 __be32 daddr, __be32 saddr, int nbytes)
1060 struct tcp4_pseudohdr *bp;
1061 struct scatterlist sg;
1063 bp = &hp->md5_blk.ip4;
1066 * 1. the TCP pseudo-header (in the order: source IP address,
1067 * destination IP address, zero-padded protocol number, and
1073 bp->protocol = IPPROTO_TCP;
1074 bp->len = cpu_to_be16(nbytes);
1076 sg_init_one(&sg, bp, sizeof(*bp));
1077 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1080 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1081 __be32 daddr, __be32 saddr, struct tcphdr *th)
1083 struct tcp_md5sig_pool *hp;
1084 struct hash_desc *desc;
1086 hp = tcp_get_md5sig_pool();
1088 goto clear_hash_noput;
1089 desc = &hp->md5_desc;
1091 if (crypto_hash_init(desc))
1093 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1095 if (tcp_md5_hash_header(hp, th))
1097 if (tcp_md5_hash_key(hp, key))
1099 if (crypto_hash_final(desc, md5_hash))
1102 tcp_put_md5sig_pool();
1106 tcp_put_md5sig_pool();
1108 memset(md5_hash, 0, 16);
1112 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1113 struct sock *sk, struct request_sock *req,
1114 struct sk_buff *skb)
1116 struct tcp_md5sig_pool *hp;
1117 struct hash_desc *desc;
1118 struct tcphdr *th = tcp_hdr(skb);
1119 __be32 saddr, daddr;
1122 saddr = inet_sk(sk)->inet_saddr;
1123 daddr = inet_sk(sk)->inet_daddr;
1125 saddr = inet_rsk(req)->loc_addr;
1126 daddr = inet_rsk(req)->rmt_addr;
1128 const struct iphdr *iph = ip_hdr(skb);
1133 hp = tcp_get_md5sig_pool();
1135 goto clear_hash_noput;
1136 desc = &hp->md5_desc;
1138 if (crypto_hash_init(desc))
1141 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1143 if (tcp_md5_hash_header(hp, th))
1145 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1147 if (tcp_md5_hash_key(hp, key))
1149 if (crypto_hash_final(desc, md5_hash))
1152 tcp_put_md5sig_pool();
1156 tcp_put_md5sig_pool();
1158 memset(md5_hash, 0, 16);
1161 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1163 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1166 * This gets called for each TCP segment that arrives
1167 * so we want to be efficient.
1168 * We have 3 drop cases:
1169 * o No MD5 hash and one expected.
1170 * o MD5 hash and we're not expecting one.
1171 * o MD5 hash and its wrong.
1173 __u8 *hash_location = NULL;
1174 struct tcp_md5sig_key *hash_expected;
1175 const struct iphdr *iph = ip_hdr(skb);
1176 struct tcphdr *th = tcp_hdr(skb);
1178 unsigned char newhash[16];
1180 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1181 hash_location = tcp_parse_md5sig_option(th);
1183 /* We've parsed the options - do we have a hash? */
1184 if (!hash_expected && !hash_location)
1187 if (hash_expected && !hash_location) {
1188 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1192 if (!hash_expected && hash_location) {
1193 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1197 /* Okay, so this is hash_expected and hash_location -
1198 * so we need to calculate the checksum.
1200 genhash = tcp_v4_md5_hash_skb(newhash,
1204 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1205 if (net_ratelimit()) {
1206 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1207 &iph->saddr, ntohs(th->source),
1208 &iph->daddr, ntohs(th->dest),
1209 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1218 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1220 .obj_size = sizeof(struct tcp_request_sock),
1221 .rtx_syn_ack = tcp_v4_rtx_synack,
1222 .send_ack = tcp_v4_reqsk_send_ack,
1223 .destructor = tcp_v4_reqsk_destructor,
1224 .send_reset = tcp_v4_send_reset,
1225 .syn_ack_timeout = tcp_syn_ack_timeout,
1228 #ifdef CONFIG_TCP_MD5SIG
1229 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1230 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1231 .calc_md5_hash = tcp_v4_md5_hash_skb,
1235 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1237 struct tcp_extend_values tmp_ext;
1238 struct tcp_options_received tmp_opt;
1240 struct request_sock *req;
1241 struct inet_request_sock *ireq;
1242 struct tcp_sock *tp = tcp_sk(sk);
1243 struct dst_entry *dst = NULL;
1244 __be32 saddr = ip_hdr(skb)->saddr;
1245 __be32 daddr = ip_hdr(skb)->daddr;
1246 __u32 isn = TCP_SKB_CB(skb)->when;
1247 #ifdef CONFIG_SYN_COOKIES
1248 int want_cookie = 0;
1250 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1253 /* Never answer to SYNs send to broadcast or multicast */
1254 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1257 /* TW buckets are converted to open requests without
1258 * limitations, they conserve resources and peer is
1259 * evidently real one.
1261 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1262 if (net_ratelimit())
1263 syn_flood_warning(skb);
1264 #ifdef CONFIG_SYN_COOKIES
1265 if (sysctl_tcp_syncookies) {
1272 /* Accept backlog is full. If we have already queued enough
1273 * of warm entries in syn queue, drop request. It is better than
1274 * clogging syn queue with openreqs with exponentially increasing
1277 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1280 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1284 #ifdef CONFIG_TCP_MD5SIG
1285 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1288 tcp_clear_options(&tmp_opt);
1289 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1290 tmp_opt.user_mss = tp->rx_opt.user_mss;
1291 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1293 if (tmp_opt.cookie_plus > 0 &&
1294 tmp_opt.saw_tstamp &&
1295 !tp->rx_opt.cookie_out_never &&
1296 (sysctl_tcp_cookie_size > 0 ||
1297 (tp->cookie_values != NULL &&
1298 tp->cookie_values->cookie_desired > 0))) {
1300 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1301 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1303 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1304 goto drop_and_release;
1306 /* Secret recipe starts with IP addresses */
1307 *mess++ ^= (__force u32)daddr;
1308 *mess++ ^= (__force u32)saddr;
1310 /* plus variable length Initiator Cookie */
1313 *c++ ^= *hash_location++;
1315 #ifdef CONFIG_SYN_COOKIES
1316 want_cookie = 0; /* not our kind of cookie */
1318 tmp_ext.cookie_out_never = 0; /* false */
1319 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1320 } else if (!tp->rx_opt.cookie_in_always) {
1321 /* redundant indications, but ensure initialization. */
1322 tmp_ext.cookie_out_never = 1; /* true */
1323 tmp_ext.cookie_plus = 0;
1325 goto drop_and_release;
1327 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1329 if (want_cookie && !tmp_opt.saw_tstamp)
1330 tcp_clear_options(&tmp_opt);
1332 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1333 tcp_openreq_init(req, &tmp_opt, skb);
1335 ireq = inet_rsk(req);
1336 ireq->loc_addr = daddr;
1337 ireq->rmt_addr = saddr;
1338 ireq->no_srccheck = inet_sk(sk)->transparent;
1339 ireq->opt = tcp_v4_save_options(sk, skb);
1341 if (security_inet_conn_request(sk, skb, req))
1344 if (!want_cookie || tmp_opt.tstamp_ok)
1345 TCP_ECN_create_request(req, tcp_hdr(skb));
1348 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1349 req->cookie_ts = tmp_opt.tstamp_ok;
1351 struct inet_peer *peer = NULL;
1354 /* VJ's idea. We save last timestamp seen
1355 * from the destination in peer table, when entering
1356 * state TIME-WAIT, and check against it before
1357 * accepting new connection request.
1359 * If "isn" is not zero, this request hit alive
1360 * timewait bucket, so that all the necessary checks
1361 * are made in the function processing timewait state.
1363 if (tmp_opt.saw_tstamp &&
1364 tcp_death_row.sysctl_tw_recycle &&
1365 (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1366 fl4.daddr == saddr &&
1367 (peer = rt_get_peer((struct rtable *)dst, fl4.daddr)) != NULL) {
1368 inet_peer_refcheck(peer);
1369 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1370 (s32)(peer->tcp_ts - req->ts_recent) >
1372 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1373 goto drop_and_release;
1376 /* Kill the following clause, if you dislike this way. */
1377 else if (!sysctl_tcp_syncookies &&
1378 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1379 (sysctl_max_syn_backlog >> 2)) &&
1380 (!peer || !peer->tcp_ts_stamp) &&
1381 (!dst || !dst_metric(dst, RTAX_RTT))) {
1382 /* Without syncookies last quarter of
1383 * backlog is filled with destinations,
1384 * proven to be alive.
1385 * It means that we continue to communicate
1386 * to destinations, already remembered
1387 * to the moment of synflood.
1389 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1390 &saddr, ntohs(tcp_hdr(skb)->source));
1391 goto drop_and_release;
1394 isn = tcp_v4_init_sequence(skb);
1396 tcp_rsk(req)->snt_isn = isn;
1398 if (tcp_v4_send_synack(sk, dst, req,
1399 (struct request_values *)&tmp_ext) ||
1403 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1413 EXPORT_SYMBOL(tcp_v4_conn_request);
1417 * The three way handshake has completed - we got a valid synack -
1418 * now create the new socket.
1420 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1421 struct request_sock *req,
1422 struct dst_entry *dst)
1424 struct inet_request_sock *ireq;
1425 struct inet_sock *newinet;
1426 struct tcp_sock *newtp;
1428 #ifdef CONFIG_TCP_MD5SIG
1429 struct tcp_md5sig_key *key;
1431 struct ip_options_rcu *inet_opt;
1433 if (sk_acceptq_is_full(sk))
1436 newsk = tcp_create_openreq_child(sk, req, skb);
1440 newsk->sk_gso_type = SKB_GSO_TCPV4;
1442 newtp = tcp_sk(newsk);
1443 newinet = inet_sk(newsk);
1444 ireq = inet_rsk(req);
1445 newinet->inet_daddr = ireq->rmt_addr;
1446 newinet->inet_rcv_saddr = ireq->loc_addr;
1447 newinet->inet_saddr = ireq->loc_addr;
1448 inet_opt = ireq->opt;
1449 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1451 newinet->mc_index = inet_iif(skb);
1452 newinet->mc_ttl = ip_hdr(skb)->ttl;
1453 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1455 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1456 newinet->inet_id = newtp->write_seq ^ jiffies;
1459 dst = inet_csk_route_child_sock(sk, newsk, req);
1463 /* syncookie case : see end of cookie_v4_check() */
1465 sk_setup_caps(newsk, dst);
1467 tcp_mtup_init(newsk);
1468 tcp_sync_mss(newsk, dst_mtu(dst));
1469 newtp->advmss = dst_metric_advmss(dst);
1470 if (tcp_sk(sk)->rx_opt.user_mss &&
1471 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1472 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1474 tcp_initialize_rcv_mss(newsk);
1476 #ifdef CONFIG_TCP_MD5SIG
1477 /* Copy over the MD5 key from the original socket */
1478 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1481 * We're using one, so create a matching key
1482 * on the newsk structure. If we fail to get
1483 * memory, then we end up not copying the key
1486 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1488 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1489 newkey, key->keylen);
1490 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1494 if (__inet_inherit_port(sk, newsk) < 0)
1496 __inet_hash_nolisten(newsk, NULL);
1501 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1505 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1511 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1513 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1515 struct tcphdr *th = tcp_hdr(skb);
1516 const struct iphdr *iph = ip_hdr(skb);
1518 struct request_sock **prev;
1519 /* Find possible connection requests. */
1520 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1521 iph->saddr, iph->daddr);
1523 return tcp_check_req(sk, skb, req, prev);
1525 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1526 th->source, iph->daddr, th->dest, inet_iif(skb));
1529 if (nsk->sk_state != TCP_TIME_WAIT) {
1533 inet_twsk_put(inet_twsk(nsk));
1537 #ifdef CONFIG_SYN_COOKIES
1539 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1544 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1546 const struct iphdr *iph = ip_hdr(skb);
1548 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1549 if (!tcp_v4_check(skb->len, iph->saddr,
1550 iph->daddr, skb->csum)) {
1551 skb->ip_summed = CHECKSUM_UNNECESSARY;
1556 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1557 skb->len, IPPROTO_TCP, 0);
1559 if (skb->len <= 76) {
1560 return __skb_checksum_complete(skb);
1566 /* The socket must have it's spinlock held when we get
1569 * We have a potential double-lock case here, so even when
1570 * doing backlog processing we use the BH locking scheme.
1571 * This is because we cannot sleep with the original spinlock
1574 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1577 #ifdef CONFIG_TCP_MD5SIG
1579 * We really want to reject the packet as early as possible
1581 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1582 * o There is an MD5 option and we're not expecting one
1584 if (tcp_v4_inbound_md5_hash(sk, skb))
1588 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1589 sock_rps_save_rxhash(sk, skb->rxhash);
1590 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1597 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1600 if (sk->sk_state == TCP_LISTEN) {
1601 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1606 sock_rps_save_rxhash(nsk, skb->rxhash);
1607 if (tcp_child_process(sk, nsk, skb)) {
1614 sock_rps_save_rxhash(sk, skb->rxhash);
1616 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1623 tcp_v4_send_reset(rsk, skb);
1626 /* Be careful here. If this function gets more complicated and
1627 * gcc suffers from register pressure on the x86, sk (in %ebx)
1628 * might be destroyed here. This current version compiles correctly,
1629 * but you have been warned.
1634 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1637 EXPORT_SYMBOL(tcp_v4_do_rcv);
1643 int tcp_v4_rcv(struct sk_buff *skb)
1645 const struct iphdr *iph;
1649 struct net *net = dev_net(skb->dev);
1651 if (skb->pkt_type != PACKET_HOST)
1654 /* Count it even if it's bad */
1655 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1657 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1662 if (th->doff < sizeof(struct tcphdr) / 4)
1664 if (!pskb_may_pull(skb, th->doff * 4))
1667 /* An explanation is required here, I think.
1668 * Packet length and doff are validated by header prediction,
1669 * provided case of th->doff==0 is eliminated.
1670 * So, we defer the checks. */
1671 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1676 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1677 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1678 skb->len - th->doff * 4);
1679 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1680 TCP_SKB_CB(skb)->when = 0;
1681 TCP_SKB_CB(skb)->flags = iph->tos;
1682 TCP_SKB_CB(skb)->sacked = 0;
1684 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1689 if (sk->sk_state == TCP_TIME_WAIT)
1692 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1693 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1694 goto discard_and_relse;
1697 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1698 goto discard_and_relse;
1701 if (sk_filter(sk, skb))
1702 goto discard_and_relse;
1706 bh_lock_sock_nested(sk);
1708 if (!sock_owned_by_user(sk)) {
1709 #ifdef CONFIG_NET_DMA
1710 struct tcp_sock *tp = tcp_sk(sk);
1711 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1712 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1713 if (tp->ucopy.dma_chan)
1714 ret = tcp_v4_do_rcv(sk, skb);
1718 if (!tcp_prequeue(sk, skb))
1719 ret = tcp_v4_do_rcv(sk, skb);
1721 } else if (unlikely(sk_add_backlog(sk, skb))) {
1723 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1724 goto discard_and_relse;
1733 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1736 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1738 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1740 tcp_v4_send_reset(NULL, skb);
1744 /* Discard frame. */
1753 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1754 inet_twsk_put(inet_twsk(sk));
1758 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1759 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1760 inet_twsk_put(inet_twsk(sk));
1763 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1765 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1767 iph->daddr, th->dest,
1770 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1771 inet_twsk_put(inet_twsk(sk));
1775 /* Fall through to ACK */
1778 tcp_v4_timewait_ack(sk, skb);
1782 case TCP_TW_SUCCESS:;
1787 struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it)
1789 struct rtable *rt = (struct rtable *) __sk_dst_get(sk);
1790 struct inet_sock *inet = inet_sk(sk);
1791 struct inet_peer *peer;
1794 inet->cork.fl.u.ip4.daddr != inet->inet_daddr) {
1795 peer = inet_getpeer_v4(inet->inet_daddr, 1);
1799 rt_bind_peer(rt, inet->inet_daddr, 1);
1801 *release_it = false;
1806 EXPORT_SYMBOL(tcp_v4_get_peer);
1808 void *tcp_v4_tw_get_peer(struct sock *sk)
1810 struct inet_timewait_sock *tw = inet_twsk(sk);
1812 return inet_getpeer_v4(tw->tw_daddr, 1);
1814 EXPORT_SYMBOL(tcp_v4_tw_get_peer);
1816 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1817 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1818 .twsk_unique = tcp_twsk_unique,
1819 .twsk_destructor= tcp_twsk_destructor,
1820 .twsk_getpeer = tcp_v4_tw_get_peer,
1823 const struct inet_connection_sock_af_ops ipv4_specific = {
1824 .queue_xmit = ip_queue_xmit,
1825 .send_check = tcp_v4_send_check,
1826 .rebuild_header = inet_sk_rebuild_header,
1827 .conn_request = tcp_v4_conn_request,
1828 .syn_recv_sock = tcp_v4_syn_recv_sock,
1829 .get_peer = tcp_v4_get_peer,
1830 .net_header_len = sizeof(struct iphdr),
1831 .setsockopt = ip_setsockopt,
1832 .getsockopt = ip_getsockopt,
1833 .addr2sockaddr = inet_csk_addr2sockaddr,
1834 .sockaddr_len = sizeof(struct sockaddr_in),
1835 .bind_conflict = inet_csk_bind_conflict,
1836 #ifdef CONFIG_COMPAT
1837 .compat_setsockopt = compat_ip_setsockopt,
1838 .compat_getsockopt = compat_ip_getsockopt,
1841 EXPORT_SYMBOL(ipv4_specific);
1843 #ifdef CONFIG_TCP_MD5SIG
1844 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1845 .md5_lookup = tcp_v4_md5_lookup,
1846 .calc_md5_hash = tcp_v4_md5_hash_skb,
1847 .md5_add = tcp_v4_md5_add_func,
1848 .md5_parse = tcp_v4_parse_md5_keys,
1852 /* NOTE: A lot of things set to zero explicitly by call to
1853 * sk_alloc() so need not be done here.
1855 static int tcp_v4_init_sock(struct sock *sk)
1857 struct inet_connection_sock *icsk = inet_csk(sk);
1858 struct tcp_sock *tp = tcp_sk(sk);
1860 skb_queue_head_init(&tp->out_of_order_queue);
1861 tcp_init_xmit_timers(sk);
1862 tcp_prequeue_init(tp);
1864 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1865 tp->mdev = TCP_TIMEOUT_INIT;
1867 /* So many TCP implementations out there (incorrectly) count the
1868 * initial SYN frame in their delayed-ACK and congestion control
1869 * algorithms that we must have the following bandaid to talk
1870 * efficiently to them. -DaveM
1874 /* See draft-stevens-tcpca-spec-01 for discussion of the
1875 * initialization of these values.
1877 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1878 tp->snd_cwnd_clamp = ~0;
1879 tp->mss_cache = TCP_MSS_DEFAULT;
1881 tp->reordering = sysctl_tcp_reordering;
1882 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1884 sk->sk_state = TCP_CLOSE;
1886 sk->sk_write_space = sk_stream_write_space;
1887 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1889 icsk->icsk_af_ops = &ipv4_specific;
1890 icsk->icsk_sync_mss = tcp_sync_mss;
1891 #ifdef CONFIG_TCP_MD5SIG
1892 tp->af_specific = &tcp_sock_ipv4_specific;
1895 /* TCP Cookie Transactions */
1896 if (sysctl_tcp_cookie_size > 0) {
1897 /* Default, cookies without s_data_payload. */
1899 kzalloc(sizeof(*tp->cookie_values),
1901 if (tp->cookie_values != NULL)
1902 kref_init(&tp->cookie_values->kref);
1904 /* Presumed zeroed, in order of appearance:
1905 * cookie_in_always, cookie_out_never,
1906 * s_data_constant, s_data_in, s_data_out
1908 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1909 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1912 percpu_counter_inc(&tcp_sockets_allocated);
1918 void tcp_v4_destroy_sock(struct sock *sk)
1920 struct tcp_sock *tp = tcp_sk(sk);
1922 tcp_clear_xmit_timers(sk);
1924 tcp_cleanup_congestion_control(sk);
1926 /* Cleanup up the write buffer. */
1927 tcp_write_queue_purge(sk);
1929 /* Cleans up our, hopefully empty, out_of_order_queue. */
1930 __skb_queue_purge(&tp->out_of_order_queue);
1932 #ifdef CONFIG_TCP_MD5SIG
1933 /* Clean up the MD5 key list, if any */
1934 if (tp->md5sig_info) {
1935 tcp_v4_clear_md5_list(sk);
1936 kfree(tp->md5sig_info);
1937 tp->md5sig_info = NULL;
1941 #ifdef CONFIG_NET_DMA
1942 /* Cleans up our sk_async_wait_queue */
1943 __skb_queue_purge(&sk->sk_async_wait_queue);
1946 /* Clean prequeue, it must be empty really */
1947 __skb_queue_purge(&tp->ucopy.prequeue);
1949 /* Clean up a referenced TCP bind bucket. */
1950 if (inet_csk(sk)->icsk_bind_hash)
1954 * If sendmsg cached page exists, toss it.
1956 if (sk->sk_sndmsg_page) {
1957 __free_page(sk->sk_sndmsg_page);
1958 sk->sk_sndmsg_page = NULL;
1961 /* TCP Cookie Transactions */
1962 if (tp->cookie_values != NULL) {
1963 kref_put(&tp->cookie_values->kref,
1964 tcp_cookie_values_release);
1965 tp->cookie_values = NULL;
1968 percpu_counter_dec(&tcp_sockets_allocated);
1970 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1972 #ifdef CONFIG_PROC_FS
1973 /* Proc filesystem TCP sock list dumping. */
1975 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1977 return hlist_nulls_empty(head) ? NULL :
1978 list_entry(head->first, struct inet_timewait_sock, tw_node);
1981 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1983 return !is_a_nulls(tw->tw_node.next) ?
1984 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1988 * Get next listener socket follow cur. If cur is NULL, get first socket
1989 * starting from bucket given in st->bucket; when st->bucket is zero the
1990 * very first socket in the hash table is returned.
1992 static void *listening_get_next(struct seq_file *seq, void *cur)
1994 struct inet_connection_sock *icsk;
1995 struct hlist_nulls_node *node;
1996 struct sock *sk = cur;
1997 struct inet_listen_hashbucket *ilb;
1998 struct tcp_iter_state *st = seq->private;
1999 struct net *net = seq_file_net(seq);
2002 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2003 spin_lock_bh(&ilb->lock);
2004 sk = sk_nulls_head(&ilb->head);
2008 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2012 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2013 struct request_sock *req = cur;
2015 icsk = inet_csk(st->syn_wait_sk);
2019 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_nulls_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);
2039 sk = sk_nulls_next(sk);
2042 sk_nulls_for_each_from(sk, node) {
2043 if (!net_eq(sock_net(sk), net))
2045 if (sk->sk_family == st->family) {
2049 icsk = inet_csk(sk);
2050 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2051 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2053 st->uid = sock_i_uid(sk);
2054 st->syn_wait_sk = sk;
2055 st->state = TCP_SEQ_STATE_OPENREQ;
2059 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2061 spin_unlock_bh(&ilb->lock);
2063 if (++st->bucket < INET_LHTABLE_SIZE) {
2064 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2065 spin_lock_bh(&ilb->lock);
2066 sk = sk_nulls_head(&ilb->head);
2074 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2076 struct tcp_iter_state *st = seq->private;
2081 rc = listening_get_next(seq, NULL);
2083 while (rc && *pos) {
2084 rc = listening_get_next(seq, rc);
2090 static inline int empty_bucket(struct tcp_iter_state *st)
2092 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2093 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2097 * Get first established socket starting from bucket given in st->bucket.
2098 * If st->bucket is zero, the very first socket in the hash is returned.
2100 static void *established_get_first(struct seq_file *seq)
2102 struct tcp_iter_state *st = seq->private;
2103 struct net *net = seq_file_net(seq);
2107 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2109 struct hlist_nulls_node *node;
2110 struct inet_timewait_sock *tw;
2111 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2113 /* Lockless fast path for the common case of empty buckets */
2114 if (empty_bucket(st))
2118 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2119 if (sk->sk_family != st->family ||
2120 !net_eq(sock_net(sk), net)) {
2126 st->state = TCP_SEQ_STATE_TIME_WAIT;
2127 inet_twsk_for_each(tw, node,
2128 &tcp_hashinfo.ehash[st->bucket].twchain) {
2129 if (tw->tw_family != st->family ||
2130 !net_eq(twsk_net(tw), net)) {
2136 spin_unlock_bh(lock);
2137 st->state = TCP_SEQ_STATE_ESTABLISHED;
2143 static void *established_get_next(struct seq_file *seq, void *cur)
2145 struct sock *sk = cur;
2146 struct inet_timewait_sock *tw;
2147 struct hlist_nulls_node *node;
2148 struct tcp_iter_state *st = seq->private;
2149 struct net *net = seq_file_net(seq);
2154 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2158 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2165 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2166 st->state = TCP_SEQ_STATE_ESTABLISHED;
2168 /* Look for next non empty bucket */
2170 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2173 if (st->bucket > tcp_hashinfo.ehash_mask)
2176 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2177 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2179 sk = sk_nulls_next(sk);
2181 sk_nulls_for_each_from(sk, node) {
2182 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2186 st->state = TCP_SEQ_STATE_TIME_WAIT;
2187 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2195 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2197 struct tcp_iter_state *st = seq->private;
2201 rc = established_get_first(seq);
2204 rc = established_get_next(seq, rc);
2210 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2213 struct tcp_iter_state *st = seq->private;
2215 st->state = TCP_SEQ_STATE_LISTENING;
2216 rc = listening_get_idx(seq, &pos);
2219 st->state = TCP_SEQ_STATE_ESTABLISHED;
2220 rc = established_get_idx(seq, pos);
2226 static void *tcp_seek_last_pos(struct seq_file *seq)
2228 struct tcp_iter_state *st = seq->private;
2229 int offset = st->offset;
2230 int orig_num = st->num;
2233 switch (st->state) {
2234 case TCP_SEQ_STATE_OPENREQ:
2235 case TCP_SEQ_STATE_LISTENING:
2236 if (st->bucket >= INET_LHTABLE_SIZE)
2238 st->state = TCP_SEQ_STATE_LISTENING;
2239 rc = listening_get_next(seq, NULL);
2240 while (offset-- && rc)
2241 rc = listening_get_next(seq, rc);
2246 case TCP_SEQ_STATE_ESTABLISHED:
2247 case TCP_SEQ_STATE_TIME_WAIT:
2248 st->state = TCP_SEQ_STATE_ESTABLISHED;
2249 if (st->bucket > tcp_hashinfo.ehash_mask)
2251 rc = established_get_first(seq);
2252 while (offset-- && rc)
2253 rc = established_get_next(seq, rc);
2261 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2263 struct tcp_iter_state *st = seq->private;
2266 if (*pos && *pos == st->last_pos) {
2267 rc = tcp_seek_last_pos(seq);
2272 st->state = TCP_SEQ_STATE_LISTENING;
2276 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2279 st->last_pos = *pos;
2283 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2285 struct tcp_iter_state *st = seq->private;
2288 if (v == SEQ_START_TOKEN) {
2289 rc = tcp_get_idx(seq, 0);
2293 switch (st->state) {
2294 case TCP_SEQ_STATE_OPENREQ:
2295 case TCP_SEQ_STATE_LISTENING:
2296 rc = listening_get_next(seq, v);
2298 st->state = TCP_SEQ_STATE_ESTABLISHED;
2301 rc = established_get_first(seq);
2304 case TCP_SEQ_STATE_ESTABLISHED:
2305 case TCP_SEQ_STATE_TIME_WAIT:
2306 rc = established_get_next(seq, v);
2311 st->last_pos = *pos;
2315 static void tcp_seq_stop(struct seq_file *seq, void *v)
2317 struct tcp_iter_state *st = seq->private;
2319 switch (st->state) {
2320 case TCP_SEQ_STATE_OPENREQ:
2322 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2323 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2325 case TCP_SEQ_STATE_LISTENING:
2326 if (v != SEQ_START_TOKEN)
2327 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2329 case TCP_SEQ_STATE_TIME_WAIT:
2330 case TCP_SEQ_STATE_ESTABLISHED:
2332 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2337 static int tcp_seq_open(struct inode *inode, struct file *file)
2339 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2340 struct tcp_iter_state *s;
2343 err = seq_open_net(inode, file, &afinfo->seq_ops,
2344 sizeof(struct tcp_iter_state));
2348 s = ((struct seq_file *)file->private_data)->private;
2349 s->family = afinfo->family;
2354 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2357 struct proc_dir_entry *p;
2359 afinfo->seq_fops.open = tcp_seq_open;
2360 afinfo->seq_fops.read = seq_read;
2361 afinfo->seq_fops.llseek = seq_lseek;
2362 afinfo->seq_fops.release = seq_release_net;
2364 afinfo->seq_ops.start = tcp_seq_start;
2365 afinfo->seq_ops.next = tcp_seq_next;
2366 afinfo->seq_ops.stop = tcp_seq_stop;
2368 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2369 &afinfo->seq_fops, afinfo);
2374 EXPORT_SYMBOL(tcp_proc_register);
2376 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2378 proc_net_remove(net, afinfo->name);
2380 EXPORT_SYMBOL(tcp_proc_unregister);
2382 static void get_openreq4(struct sock *sk, struct request_sock *req,
2383 struct seq_file *f, int i, int uid, int *len)
2385 const struct inet_request_sock *ireq = inet_rsk(req);
2386 int ttd = req->expires - jiffies;
2388 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2389 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2392 ntohs(inet_sk(sk)->inet_sport),
2394 ntohs(ireq->rmt_port),
2396 0, 0, /* could print option size, but that is af dependent. */
2397 1, /* timers active (only the expire timer) */
2398 jiffies_to_clock_t(ttd),
2401 0, /* non standard timer */
2402 0, /* open_requests have no inode */
2403 atomic_read(&sk->sk_refcnt),
2408 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2411 unsigned long timer_expires;
2412 struct tcp_sock *tp = tcp_sk(sk);
2413 const struct inet_connection_sock *icsk = inet_csk(sk);
2414 struct inet_sock *inet = inet_sk(sk);
2415 __be32 dest = inet->inet_daddr;
2416 __be32 src = inet->inet_rcv_saddr;
2417 __u16 destp = ntohs(inet->inet_dport);
2418 __u16 srcp = ntohs(inet->inet_sport);
2421 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2423 timer_expires = icsk->icsk_timeout;
2424 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2426 timer_expires = icsk->icsk_timeout;
2427 } else if (timer_pending(&sk->sk_timer)) {
2429 timer_expires = sk->sk_timer.expires;
2432 timer_expires = jiffies;
2435 if (sk->sk_state == TCP_LISTEN)
2436 rx_queue = sk->sk_ack_backlog;
2439 * because we dont lock socket, we might find a transient negative value
2441 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2443 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2444 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2445 i, src, srcp, dest, destp, sk->sk_state,
2446 tp->write_seq - tp->snd_una,
2449 jiffies_to_clock_t(timer_expires - jiffies),
2450 icsk->icsk_retransmits,
2452 icsk->icsk_probes_out,
2454 atomic_read(&sk->sk_refcnt), sk,
2455 jiffies_to_clock_t(icsk->icsk_rto),
2456 jiffies_to_clock_t(icsk->icsk_ack.ato),
2457 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2459 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2463 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2464 struct seq_file *f, int i, int *len)
2468 int ttd = tw->tw_ttd - jiffies;
2473 dest = tw->tw_daddr;
2474 src = tw->tw_rcv_saddr;
2475 destp = ntohs(tw->tw_dport);
2476 srcp = ntohs(tw->tw_sport);
2478 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2479 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2480 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2481 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2482 atomic_read(&tw->tw_refcnt), tw, len);
2487 static int tcp4_seq_show(struct seq_file *seq, void *v)
2489 struct tcp_iter_state *st;
2492 if (v == SEQ_START_TOKEN) {
2493 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2494 " sl local_address rem_address st tx_queue "
2495 "rx_queue tr tm->when retrnsmt uid timeout "
2501 switch (st->state) {
2502 case TCP_SEQ_STATE_LISTENING:
2503 case TCP_SEQ_STATE_ESTABLISHED:
2504 get_tcp4_sock(v, seq, st->num, &len);
2506 case TCP_SEQ_STATE_OPENREQ:
2507 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2509 case TCP_SEQ_STATE_TIME_WAIT:
2510 get_timewait4_sock(v, seq, st->num, &len);
2513 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2518 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2522 .owner = THIS_MODULE,
2525 .show = tcp4_seq_show,
2529 static int __net_init tcp4_proc_init_net(struct net *net)
2531 return tcp_proc_register(net, &tcp4_seq_afinfo);
2534 static void __net_exit tcp4_proc_exit_net(struct net *net)
2536 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2539 static struct pernet_operations tcp4_net_ops = {
2540 .init = tcp4_proc_init_net,
2541 .exit = tcp4_proc_exit_net,
2544 int __init tcp4_proc_init(void)
2546 return register_pernet_subsys(&tcp4_net_ops);
2549 void tcp4_proc_exit(void)
2551 unregister_pernet_subsys(&tcp4_net_ops);
2553 #endif /* CONFIG_PROC_FS */
2555 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2557 const struct iphdr *iph = skb_gro_network_header(skb);
2559 switch (skb->ip_summed) {
2560 case CHECKSUM_COMPLETE:
2561 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2563 skb->ip_summed = CHECKSUM_UNNECESSARY;
2569 NAPI_GRO_CB(skb)->flush = 1;
2573 return tcp_gro_receive(head, skb);
2576 int tcp4_gro_complete(struct sk_buff *skb)
2578 const struct iphdr *iph = ip_hdr(skb);
2579 struct tcphdr *th = tcp_hdr(skb);
2581 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2582 iph->saddr, iph->daddr, 0);
2583 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2585 return tcp_gro_complete(skb);
2588 struct proto tcp_prot = {
2590 .owner = THIS_MODULE,
2592 .connect = tcp_v4_connect,
2593 .disconnect = tcp_disconnect,
2594 .accept = inet_csk_accept,
2596 .init = tcp_v4_init_sock,
2597 .destroy = tcp_v4_destroy_sock,
2598 .shutdown = tcp_shutdown,
2599 .setsockopt = tcp_setsockopt,
2600 .getsockopt = tcp_getsockopt,
2601 .recvmsg = tcp_recvmsg,
2602 .sendmsg = tcp_sendmsg,
2603 .sendpage = tcp_sendpage,
2604 .backlog_rcv = tcp_v4_do_rcv,
2606 .unhash = inet_unhash,
2607 .get_port = inet_csk_get_port,
2608 .enter_memory_pressure = tcp_enter_memory_pressure,
2609 .sockets_allocated = &tcp_sockets_allocated,
2610 .orphan_count = &tcp_orphan_count,
2611 .memory_allocated = &tcp_memory_allocated,
2612 .memory_pressure = &tcp_memory_pressure,
2613 .sysctl_mem = sysctl_tcp_mem,
2614 .sysctl_wmem = sysctl_tcp_wmem,
2615 .sysctl_rmem = sysctl_tcp_rmem,
2616 .max_header = MAX_TCP_HEADER,
2617 .obj_size = sizeof(struct tcp_sock),
2618 .slab_flags = SLAB_DESTROY_BY_RCU,
2619 .twsk_prot = &tcp_timewait_sock_ops,
2620 .rsk_prot = &tcp_request_sock_ops,
2621 .h.hashinfo = &tcp_hashinfo,
2622 .no_autobind = true,
2623 #ifdef CONFIG_COMPAT
2624 .compat_setsockopt = compat_tcp_setsockopt,
2625 .compat_getsockopt = compat_tcp_getsockopt,
2628 EXPORT_SYMBOL(tcp_prot);
2631 static int __net_init tcp_sk_init(struct net *net)
2633 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2634 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2637 static void __net_exit tcp_sk_exit(struct net *net)
2639 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2642 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2644 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2647 static struct pernet_operations __net_initdata tcp_sk_ops = {
2648 .init = tcp_sk_init,
2649 .exit = tcp_sk_exit,
2650 .exit_batch = tcp_sk_exit_batch,
2653 void __init tcp_v4_init(void)
2655 inet_hashinfo_init(&tcp_hashinfo);
2656 if (register_pernet_subsys(&tcp_sk_ops))
2657 panic("Failed to create the TCP control socket.\n");