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;
654 net = dev_net(skb_dst(skb)->dev);
655 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
656 &arg, arg.iov[0].iov_len);
658 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
659 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
662 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
663 outside socket context is ugly, certainly. What can I do?
666 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
667 u32 win, u32 ts, int oif,
668 struct tcp_md5sig_key *key,
671 struct tcphdr *th = tcp_hdr(skb);
674 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
675 #ifdef CONFIG_TCP_MD5SIG
676 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
680 struct ip_reply_arg arg;
681 struct net *net = dev_net(skb_dst(skb)->dev);
683 memset(&rep.th, 0, sizeof(struct tcphdr));
684 memset(&arg, 0, sizeof(arg));
686 arg.iov[0].iov_base = (unsigned char *)&rep;
687 arg.iov[0].iov_len = sizeof(rep.th);
689 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
690 (TCPOPT_TIMESTAMP << 8) |
692 rep.opt[1] = htonl(tcp_time_stamp);
693 rep.opt[2] = htonl(ts);
694 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
697 /* Swap the send and the receive. */
698 rep.th.dest = th->source;
699 rep.th.source = th->dest;
700 rep.th.doff = arg.iov[0].iov_len / 4;
701 rep.th.seq = htonl(seq);
702 rep.th.ack_seq = htonl(ack);
704 rep.th.window = htons(win);
706 #ifdef CONFIG_TCP_MD5SIG
708 int offset = (ts) ? 3 : 0;
710 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
712 (TCPOPT_MD5SIG << 8) |
714 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
715 rep.th.doff = arg.iov[0].iov_len/4;
717 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
718 key, ip_hdr(skb)->saddr,
719 ip_hdr(skb)->daddr, &rep.th);
722 arg.flags = reply_flags;
723 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
724 ip_hdr(skb)->saddr, /* XXX */
725 arg.iov[0].iov_len, IPPROTO_TCP, 0);
726 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
728 arg.bound_dev_if = oif;
730 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
731 &arg, arg.iov[0].iov_len);
733 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
736 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
738 struct inet_timewait_sock *tw = inet_twsk(sk);
739 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
741 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
742 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
745 tcp_twsk_md5_key(tcptw),
746 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
752 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
753 struct request_sock *req)
755 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
756 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
759 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
760 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
764 * Send a SYN-ACK after having received a SYN.
765 * This still operates on a request_sock only, not on a big
768 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
769 struct request_sock *req,
770 struct request_values *rvp)
772 const struct inet_request_sock *ireq = inet_rsk(req);
775 struct sk_buff * skb;
777 /* First, grab a route. */
778 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
781 skb = tcp_make_synack(sk, dst, req, rvp);
784 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
786 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
789 err = net_xmit_eval(err);
796 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
797 struct request_values *rvp)
799 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
800 return tcp_v4_send_synack(sk, NULL, req, rvp);
804 * IPv4 request_sock destructor.
806 static void tcp_v4_reqsk_destructor(struct request_sock *req)
808 kfree(inet_rsk(req)->opt);
811 static void syn_flood_warning(const struct sk_buff *skb)
815 #ifdef CONFIG_SYN_COOKIES
816 if (sysctl_tcp_syncookies)
817 msg = "Sending cookies";
820 msg = "Dropping request";
822 pr_info("TCP: Possible SYN flooding on port %d. %s.\n",
823 ntohs(tcp_hdr(skb)->dest), msg);
827 * Save and compile IPv4 options into the request_sock if needed.
829 static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk,
832 const struct ip_options *opt = &(IPCB(skb)->opt);
833 struct ip_options_rcu *dopt = NULL;
835 if (opt && opt->optlen) {
836 int opt_size = sizeof(*dopt) + opt->optlen;
838 dopt = kmalloc(opt_size, GFP_ATOMIC);
840 if (ip_options_echo(&dopt->opt, skb)) {
849 #ifdef CONFIG_TCP_MD5SIG
851 * RFC2385 MD5 checksumming requires a mapping of
852 * IP address->MD5 Key.
853 * We need to maintain these in the sk structure.
856 /* Find the Key structure for an address. */
857 static struct tcp_md5sig_key *
858 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
860 struct tcp_sock *tp = tcp_sk(sk);
863 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
865 for (i = 0; i < tp->md5sig_info->entries4; i++) {
866 if (tp->md5sig_info->keys4[i].addr == addr)
867 return &tp->md5sig_info->keys4[i].base;
872 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
873 struct sock *addr_sk)
875 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
877 EXPORT_SYMBOL(tcp_v4_md5_lookup);
879 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
880 struct request_sock *req)
882 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
885 /* This can be called on a newly created socket, from other files */
886 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
887 u8 *newkey, u8 newkeylen)
889 /* Add Key to the list */
890 struct tcp_md5sig_key *key;
891 struct tcp_sock *tp = tcp_sk(sk);
892 struct tcp4_md5sig_key *keys;
894 key = tcp_v4_md5_do_lookup(sk, addr);
896 /* Pre-existing entry - just update that one. */
899 key->keylen = newkeylen;
901 struct tcp_md5sig_info *md5sig;
903 if (!tp->md5sig_info) {
904 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
906 if (!tp->md5sig_info) {
910 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
913 md5sig = tp->md5sig_info;
914 if (md5sig->entries4 == 0 &&
915 tcp_alloc_md5sig_pool(sk) == NULL) {
920 if (md5sig->alloced4 == md5sig->entries4) {
921 keys = kmalloc((sizeof(*keys) *
922 (md5sig->entries4 + 1)), GFP_ATOMIC);
925 if (md5sig->entries4 == 0)
926 tcp_free_md5sig_pool();
930 if (md5sig->entries4)
931 memcpy(keys, md5sig->keys4,
932 sizeof(*keys) * md5sig->entries4);
934 /* Free old key list, and reference new one */
935 kfree(md5sig->keys4);
936 md5sig->keys4 = keys;
940 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
941 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
942 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
946 EXPORT_SYMBOL(tcp_v4_md5_do_add);
948 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
949 u8 *newkey, u8 newkeylen)
951 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
955 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
957 struct tcp_sock *tp = tcp_sk(sk);
960 for (i = 0; i < tp->md5sig_info->entries4; i++) {
961 if (tp->md5sig_info->keys4[i].addr == addr) {
963 kfree(tp->md5sig_info->keys4[i].base.key);
964 tp->md5sig_info->entries4--;
966 if (tp->md5sig_info->entries4 == 0) {
967 kfree(tp->md5sig_info->keys4);
968 tp->md5sig_info->keys4 = NULL;
969 tp->md5sig_info->alloced4 = 0;
970 tcp_free_md5sig_pool();
971 } else if (tp->md5sig_info->entries4 != i) {
972 /* Need to do some manipulation */
973 memmove(&tp->md5sig_info->keys4[i],
974 &tp->md5sig_info->keys4[i+1],
975 (tp->md5sig_info->entries4 - i) *
976 sizeof(struct tcp4_md5sig_key));
983 EXPORT_SYMBOL(tcp_v4_md5_do_del);
985 static void tcp_v4_clear_md5_list(struct sock *sk)
987 struct tcp_sock *tp = tcp_sk(sk);
989 /* Free each key, then the set of key keys,
990 * the crypto element, and then decrement our
991 * hold on the last resort crypto.
993 if (tp->md5sig_info->entries4) {
995 for (i = 0; i < tp->md5sig_info->entries4; i++)
996 kfree(tp->md5sig_info->keys4[i].base.key);
997 tp->md5sig_info->entries4 = 0;
998 tcp_free_md5sig_pool();
1000 if (tp->md5sig_info->keys4) {
1001 kfree(tp->md5sig_info->keys4);
1002 tp->md5sig_info->keys4 = NULL;
1003 tp->md5sig_info->alloced4 = 0;
1007 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1010 struct tcp_md5sig cmd;
1011 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1014 if (optlen < sizeof(cmd))
1017 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1020 if (sin->sin_family != AF_INET)
1023 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1024 if (!tcp_sk(sk)->md5sig_info)
1026 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1029 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1032 if (!tcp_sk(sk)->md5sig_info) {
1033 struct tcp_sock *tp = tcp_sk(sk);
1034 struct tcp_md5sig_info *p;
1036 p = kzalloc(sizeof(*p), sk->sk_allocation);
1040 tp->md5sig_info = p;
1041 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1044 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1047 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1048 newkey, cmd.tcpm_keylen);
1051 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1052 __be32 daddr, __be32 saddr, int nbytes)
1054 struct tcp4_pseudohdr *bp;
1055 struct scatterlist sg;
1057 bp = &hp->md5_blk.ip4;
1060 * 1. the TCP pseudo-header (in the order: source IP address,
1061 * destination IP address, zero-padded protocol number, and
1067 bp->protocol = IPPROTO_TCP;
1068 bp->len = cpu_to_be16(nbytes);
1070 sg_init_one(&sg, bp, sizeof(*bp));
1071 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1074 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1075 __be32 daddr, __be32 saddr, struct tcphdr *th)
1077 struct tcp_md5sig_pool *hp;
1078 struct hash_desc *desc;
1080 hp = tcp_get_md5sig_pool();
1082 goto clear_hash_noput;
1083 desc = &hp->md5_desc;
1085 if (crypto_hash_init(desc))
1087 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1089 if (tcp_md5_hash_header(hp, th))
1091 if (tcp_md5_hash_key(hp, key))
1093 if (crypto_hash_final(desc, md5_hash))
1096 tcp_put_md5sig_pool();
1100 tcp_put_md5sig_pool();
1102 memset(md5_hash, 0, 16);
1106 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1107 struct sock *sk, struct request_sock *req,
1108 struct sk_buff *skb)
1110 struct tcp_md5sig_pool *hp;
1111 struct hash_desc *desc;
1112 struct tcphdr *th = tcp_hdr(skb);
1113 __be32 saddr, daddr;
1116 saddr = inet_sk(sk)->inet_saddr;
1117 daddr = inet_sk(sk)->inet_daddr;
1119 saddr = inet_rsk(req)->loc_addr;
1120 daddr = inet_rsk(req)->rmt_addr;
1122 const struct iphdr *iph = ip_hdr(skb);
1127 hp = tcp_get_md5sig_pool();
1129 goto clear_hash_noput;
1130 desc = &hp->md5_desc;
1132 if (crypto_hash_init(desc))
1135 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1137 if (tcp_md5_hash_header(hp, th))
1139 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1141 if (tcp_md5_hash_key(hp, key))
1143 if (crypto_hash_final(desc, md5_hash))
1146 tcp_put_md5sig_pool();
1150 tcp_put_md5sig_pool();
1152 memset(md5_hash, 0, 16);
1155 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1157 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1160 * This gets called for each TCP segment that arrives
1161 * so we want to be efficient.
1162 * We have 3 drop cases:
1163 * o No MD5 hash and one expected.
1164 * o MD5 hash and we're not expecting one.
1165 * o MD5 hash and its wrong.
1167 __u8 *hash_location = NULL;
1168 struct tcp_md5sig_key *hash_expected;
1169 const struct iphdr *iph = ip_hdr(skb);
1170 struct tcphdr *th = tcp_hdr(skb);
1172 unsigned char newhash[16];
1174 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1175 hash_location = tcp_parse_md5sig_option(th);
1177 /* We've parsed the options - do we have a hash? */
1178 if (!hash_expected && !hash_location)
1181 if (hash_expected && !hash_location) {
1182 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1186 if (!hash_expected && hash_location) {
1187 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1191 /* Okay, so this is hash_expected and hash_location -
1192 * so we need to calculate the checksum.
1194 genhash = tcp_v4_md5_hash_skb(newhash,
1198 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1199 if (net_ratelimit()) {
1200 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1201 &iph->saddr, ntohs(th->source),
1202 &iph->daddr, ntohs(th->dest),
1203 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1212 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1214 .obj_size = sizeof(struct tcp_request_sock),
1215 .rtx_syn_ack = tcp_v4_rtx_synack,
1216 .send_ack = tcp_v4_reqsk_send_ack,
1217 .destructor = tcp_v4_reqsk_destructor,
1218 .send_reset = tcp_v4_send_reset,
1219 .syn_ack_timeout = tcp_syn_ack_timeout,
1222 #ifdef CONFIG_TCP_MD5SIG
1223 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1224 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1225 .calc_md5_hash = tcp_v4_md5_hash_skb,
1229 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1231 struct tcp_extend_values tmp_ext;
1232 struct tcp_options_received tmp_opt;
1234 struct request_sock *req;
1235 struct inet_request_sock *ireq;
1236 struct tcp_sock *tp = tcp_sk(sk);
1237 struct dst_entry *dst = NULL;
1238 __be32 saddr = ip_hdr(skb)->saddr;
1239 __be32 daddr = ip_hdr(skb)->daddr;
1240 __u32 isn = TCP_SKB_CB(skb)->when;
1241 #ifdef CONFIG_SYN_COOKIES
1242 int want_cookie = 0;
1244 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1247 /* Never answer to SYNs send to broadcast or multicast */
1248 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1251 /* TW buckets are converted to open requests without
1252 * limitations, they conserve resources and peer is
1253 * evidently real one.
1255 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1256 if (net_ratelimit())
1257 syn_flood_warning(skb);
1258 #ifdef CONFIG_SYN_COOKIES
1259 if (sysctl_tcp_syncookies) {
1266 /* Accept backlog is full. If we have already queued enough
1267 * of warm entries in syn queue, drop request. It is better than
1268 * clogging syn queue with openreqs with exponentially increasing
1271 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1274 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1278 #ifdef CONFIG_TCP_MD5SIG
1279 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1282 tcp_clear_options(&tmp_opt);
1283 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1284 tmp_opt.user_mss = tp->rx_opt.user_mss;
1285 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1287 if (tmp_opt.cookie_plus > 0 &&
1288 tmp_opt.saw_tstamp &&
1289 !tp->rx_opt.cookie_out_never &&
1290 (sysctl_tcp_cookie_size > 0 ||
1291 (tp->cookie_values != NULL &&
1292 tp->cookie_values->cookie_desired > 0))) {
1294 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1295 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1297 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1298 goto drop_and_release;
1300 /* Secret recipe starts with IP addresses */
1301 *mess++ ^= (__force u32)daddr;
1302 *mess++ ^= (__force u32)saddr;
1304 /* plus variable length Initiator Cookie */
1307 *c++ ^= *hash_location++;
1309 #ifdef CONFIG_SYN_COOKIES
1310 want_cookie = 0; /* not our kind of cookie */
1312 tmp_ext.cookie_out_never = 0; /* false */
1313 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1314 } else if (!tp->rx_opt.cookie_in_always) {
1315 /* redundant indications, but ensure initialization. */
1316 tmp_ext.cookie_out_never = 1; /* true */
1317 tmp_ext.cookie_plus = 0;
1319 goto drop_and_release;
1321 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1323 if (want_cookie && !tmp_opt.saw_tstamp)
1324 tcp_clear_options(&tmp_opt);
1326 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1327 tcp_openreq_init(req, &tmp_opt, skb);
1329 ireq = inet_rsk(req);
1330 ireq->loc_addr = daddr;
1331 ireq->rmt_addr = saddr;
1332 ireq->no_srccheck = inet_sk(sk)->transparent;
1333 ireq->opt = tcp_v4_save_options(sk, skb);
1335 if (security_inet_conn_request(sk, skb, req))
1338 if (!want_cookie || tmp_opt.tstamp_ok)
1339 TCP_ECN_create_request(req, tcp_hdr(skb));
1342 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1343 req->cookie_ts = tmp_opt.tstamp_ok;
1345 struct inet_peer *peer = NULL;
1348 /* VJ's idea. We save last timestamp seen
1349 * from the destination in peer table, when entering
1350 * state TIME-WAIT, and check against it before
1351 * accepting new connection request.
1353 * If "isn" is not zero, this request hit alive
1354 * timewait bucket, so that all the necessary checks
1355 * are made in the function processing timewait state.
1357 if (tmp_opt.saw_tstamp &&
1358 tcp_death_row.sysctl_tw_recycle &&
1359 (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1360 fl4.daddr == saddr &&
1361 (peer = rt_get_peer((struct rtable *)dst, fl4.daddr)) != NULL) {
1362 inet_peer_refcheck(peer);
1363 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1364 (s32)(peer->tcp_ts - req->ts_recent) >
1366 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1367 goto drop_and_release;
1370 /* Kill the following clause, if you dislike this way. */
1371 else if (!sysctl_tcp_syncookies &&
1372 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1373 (sysctl_max_syn_backlog >> 2)) &&
1374 (!peer || !peer->tcp_ts_stamp) &&
1375 (!dst || !dst_metric(dst, RTAX_RTT))) {
1376 /* Without syncookies last quarter of
1377 * backlog is filled with destinations,
1378 * proven to be alive.
1379 * It means that we continue to communicate
1380 * to destinations, already remembered
1381 * to the moment of synflood.
1383 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1384 &saddr, ntohs(tcp_hdr(skb)->source));
1385 goto drop_and_release;
1388 isn = tcp_v4_init_sequence(skb);
1390 tcp_rsk(req)->snt_isn = isn;
1392 if (tcp_v4_send_synack(sk, dst, req,
1393 (struct request_values *)&tmp_ext) ||
1397 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1407 EXPORT_SYMBOL(tcp_v4_conn_request);
1411 * The three way handshake has completed - we got a valid synack -
1412 * now create the new socket.
1414 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1415 struct request_sock *req,
1416 struct dst_entry *dst)
1418 struct inet_request_sock *ireq;
1419 struct inet_sock *newinet;
1420 struct tcp_sock *newtp;
1422 #ifdef CONFIG_TCP_MD5SIG
1423 struct tcp_md5sig_key *key;
1425 struct ip_options_rcu *inet_opt;
1427 if (sk_acceptq_is_full(sk))
1430 newsk = tcp_create_openreq_child(sk, req, skb);
1434 newsk->sk_gso_type = SKB_GSO_TCPV4;
1436 newtp = tcp_sk(newsk);
1437 newinet = inet_sk(newsk);
1438 ireq = inet_rsk(req);
1439 newinet->inet_daddr = ireq->rmt_addr;
1440 newinet->inet_rcv_saddr = ireq->loc_addr;
1441 newinet->inet_saddr = ireq->loc_addr;
1442 inet_opt = ireq->opt;
1443 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1445 newinet->mc_index = inet_iif(skb);
1446 newinet->mc_ttl = ip_hdr(skb)->ttl;
1447 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1449 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1450 newinet->inet_id = newtp->write_seq ^ jiffies;
1452 if (!dst && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
1455 sk_setup_caps(newsk, dst);
1457 tcp_mtup_init(newsk);
1458 tcp_sync_mss(newsk, dst_mtu(dst));
1459 newtp->advmss = dst_metric_advmss(dst);
1460 if (tcp_sk(sk)->rx_opt.user_mss &&
1461 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1462 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1464 tcp_initialize_rcv_mss(newsk);
1466 #ifdef CONFIG_TCP_MD5SIG
1467 /* Copy over the MD5 key from the original socket */
1468 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1471 * We're using one, so create a matching key
1472 * on the newsk structure. If we fail to get
1473 * memory, then we end up not copying the key
1476 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1478 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1479 newkey, key->keylen);
1480 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1484 if (__inet_inherit_port(sk, newsk) < 0)
1486 __inet_hash_nolisten(newsk, NULL);
1491 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1495 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1501 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1503 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1505 struct tcphdr *th = tcp_hdr(skb);
1506 const struct iphdr *iph = ip_hdr(skb);
1508 struct request_sock **prev;
1509 /* Find possible connection requests. */
1510 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1511 iph->saddr, iph->daddr);
1513 return tcp_check_req(sk, skb, req, prev);
1515 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1516 th->source, iph->daddr, th->dest, inet_iif(skb));
1519 if (nsk->sk_state != TCP_TIME_WAIT) {
1523 inet_twsk_put(inet_twsk(nsk));
1527 #ifdef CONFIG_SYN_COOKIES
1529 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1534 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1536 const struct iphdr *iph = ip_hdr(skb);
1538 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1539 if (!tcp_v4_check(skb->len, iph->saddr,
1540 iph->daddr, skb->csum)) {
1541 skb->ip_summed = CHECKSUM_UNNECESSARY;
1546 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1547 skb->len, IPPROTO_TCP, 0);
1549 if (skb->len <= 76) {
1550 return __skb_checksum_complete(skb);
1556 /* The socket must have it's spinlock held when we get
1559 * We have a potential double-lock case here, so even when
1560 * doing backlog processing we use the BH locking scheme.
1561 * This is because we cannot sleep with the original spinlock
1564 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1567 #ifdef CONFIG_TCP_MD5SIG
1569 * We really want to reject the packet as early as possible
1571 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1572 * o There is an MD5 option and we're not expecting one
1574 if (tcp_v4_inbound_md5_hash(sk, skb))
1578 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1579 sock_rps_save_rxhash(sk, skb->rxhash);
1580 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1587 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1590 if (sk->sk_state == TCP_LISTEN) {
1591 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1596 sock_rps_save_rxhash(nsk, skb->rxhash);
1597 if (tcp_child_process(sk, nsk, skb)) {
1604 sock_rps_save_rxhash(sk, skb->rxhash);
1606 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1613 tcp_v4_send_reset(rsk, skb);
1616 /* Be careful here. If this function gets more complicated and
1617 * gcc suffers from register pressure on the x86, sk (in %ebx)
1618 * might be destroyed here. This current version compiles correctly,
1619 * but you have been warned.
1624 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1627 EXPORT_SYMBOL(tcp_v4_do_rcv);
1633 int tcp_v4_rcv(struct sk_buff *skb)
1635 const struct iphdr *iph;
1639 struct net *net = dev_net(skb->dev);
1641 if (skb->pkt_type != PACKET_HOST)
1644 /* Count it even if it's bad */
1645 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1647 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1652 if (th->doff < sizeof(struct tcphdr) / 4)
1654 if (!pskb_may_pull(skb, th->doff * 4))
1657 /* An explanation is required here, I think.
1658 * Packet length and doff are validated by header prediction,
1659 * provided case of th->doff==0 is eliminated.
1660 * So, we defer the checks. */
1661 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1666 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1667 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1668 skb->len - th->doff * 4);
1669 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1670 TCP_SKB_CB(skb)->when = 0;
1671 TCP_SKB_CB(skb)->flags = iph->tos;
1672 TCP_SKB_CB(skb)->sacked = 0;
1674 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1679 if (sk->sk_state == TCP_TIME_WAIT)
1682 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1683 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1684 goto discard_and_relse;
1687 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1688 goto discard_and_relse;
1691 if (sk_filter(sk, skb))
1692 goto discard_and_relse;
1696 bh_lock_sock_nested(sk);
1698 if (!sock_owned_by_user(sk)) {
1699 #ifdef CONFIG_NET_DMA
1700 struct tcp_sock *tp = tcp_sk(sk);
1701 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1702 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1703 if (tp->ucopy.dma_chan)
1704 ret = tcp_v4_do_rcv(sk, skb);
1708 if (!tcp_prequeue(sk, skb))
1709 ret = tcp_v4_do_rcv(sk, skb);
1711 } else if (unlikely(sk_add_backlog(sk, skb))) {
1713 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1714 goto discard_and_relse;
1723 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1726 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1728 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1730 tcp_v4_send_reset(NULL, skb);
1734 /* Discard frame. */
1743 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1744 inet_twsk_put(inet_twsk(sk));
1748 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1749 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1750 inet_twsk_put(inet_twsk(sk));
1753 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1755 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1757 iph->daddr, th->dest,
1760 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1761 inet_twsk_put(inet_twsk(sk));
1765 /* Fall through to ACK */
1768 tcp_v4_timewait_ack(sk, skb);
1772 case TCP_TW_SUCCESS:;
1777 struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it)
1779 struct rtable *rt = (struct rtable *) __sk_dst_get(sk);
1780 struct inet_sock *inet = inet_sk(sk);
1781 struct inet_peer *peer;
1784 inet->cork.fl.u.ip4.daddr != inet->inet_daddr) {
1785 peer = inet_getpeer_v4(inet->inet_daddr, 1);
1789 rt_bind_peer(rt, inet->inet_daddr, 1);
1791 *release_it = false;
1796 EXPORT_SYMBOL(tcp_v4_get_peer);
1798 void *tcp_v4_tw_get_peer(struct sock *sk)
1800 struct inet_timewait_sock *tw = inet_twsk(sk);
1802 return inet_getpeer_v4(tw->tw_daddr, 1);
1804 EXPORT_SYMBOL(tcp_v4_tw_get_peer);
1806 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1807 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1808 .twsk_unique = tcp_twsk_unique,
1809 .twsk_destructor= tcp_twsk_destructor,
1810 .twsk_getpeer = tcp_v4_tw_get_peer,
1813 const struct inet_connection_sock_af_ops ipv4_specific = {
1814 .queue_xmit = ip_queue_xmit,
1815 .send_check = tcp_v4_send_check,
1816 .rebuild_header = inet_sk_rebuild_header,
1817 .conn_request = tcp_v4_conn_request,
1818 .syn_recv_sock = tcp_v4_syn_recv_sock,
1819 .get_peer = tcp_v4_get_peer,
1820 .net_header_len = sizeof(struct iphdr),
1821 .setsockopt = ip_setsockopt,
1822 .getsockopt = ip_getsockopt,
1823 .addr2sockaddr = inet_csk_addr2sockaddr,
1824 .sockaddr_len = sizeof(struct sockaddr_in),
1825 .bind_conflict = inet_csk_bind_conflict,
1826 #ifdef CONFIG_COMPAT
1827 .compat_setsockopt = compat_ip_setsockopt,
1828 .compat_getsockopt = compat_ip_getsockopt,
1831 EXPORT_SYMBOL(ipv4_specific);
1833 #ifdef CONFIG_TCP_MD5SIG
1834 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1835 .md5_lookup = tcp_v4_md5_lookup,
1836 .calc_md5_hash = tcp_v4_md5_hash_skb,
1837 .md5_add = tcp_v4_md5_add_func,
1838 .md5_parse = tcp_v4_parse_md5_keys,
1842 /* NOTE: A lot of things set to zero explicitly by call to
1843 * sk_alloc() so need not be done here.
1845 static int tcp_v4_init_sock(struct sock *sk)
1847 struct inet_connection_sock *icsk = inet_csk(sk);
1848 struct tcp_sock *tp = tcp_sk(sk);
1850 skb_queue_head_init(&tp->out_of_order_queue);
1851 tcp_init_xmit_timers(sk);
1852 tcp_prequeue_init(tp);
1854 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1855 tp->mdev = TCP_TIMEOUT_INIT;
1857 /* So many TCP implementations out there (incorrectly) count the
1858 * initial SYN frame in their delayed-ACK and congestion control
1859 * algorithms that we must have the following bandaid to talk
1860 * efficiently to them. -DaveM
1864 /* See draft-stevens-tcpca-spec-01 for discussion of the
1865 * initialization of these values.
1867 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1868 tp->snd_cwnd_clamp = ~0;
1869 tp->mss_cache = TCP_MSS_DEFAULT;
1871 tp->reordering = sysctl_tcp_reordering;
1872 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1874 sk->sk_state = TCP_CLOSE;
1876 sk->sk_write_space = sk_stream_write_space;
1877 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1879 icsk->icsk_af_ops = &ipv4_specific;
1880 icsk->icsk_sync_mss = tcp_sync_mss;
1881 #ifdef CONFIG_TCP_MD5SIG
1882 tp->af_specific = &tcp_sock_ipv4_specific;
1885 /* TCP Cookie Transactions */
1886 if (sysctl_tcp_cookie_size > 0) {
1887 /* Default, cookies without s_data_payload. */
1889 kzalloc(sizeof(*tp->cookie_values),
1891 if (tp->cookie_values != NULL)
1892 kref_init(&tp->cookie_values->kref);
1894 /* Presumed zeroed, in order of appearance:
1895 * cookie_in_always, cookie_out_never,
1896 * s_data_constant, s_data_in, s_data_out
1898 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1899 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1902 percpu_counter_inc(&tcp_sockets_allocated);
1908 void tcp_v4_destroy_sock(struct sock *sk)
1910 struct tcp_sock *tp = tcp_sk(sk);
1912 tcp_clear_xmit_timers(sk);
1914 tcp_cleanup_congestion_control(sk);
1916 /* Cleanup up the write buffer. */
1917 tcp_write_queue_purge(sk);
1919 /* Cleans up our, hopefully empty, out_of_order_queue. */
1920 __skb_queue_purge(&tp->out_of_order_queue);
1922 #ifdef CONFIG_TCP_MD5SIG
1923 /* Clean up the MD5 key list, if any */
1924 if (tp->md5sig_info) {
1925 tcp_v4_clear_md5_list(sk);
1926 kfree(tp->md5sig_info);
1927 tp->md5sig_info = NULL;
1931 #ifdef CONFIG_NET_DMA
1932 /* Cleans up our sk_async_wait_queue */
1933 __skb_queue_purge(&sk->sk_async_wait_queue);
1936 /* Clean prequeue, it must be empty really */
1937 __skb_queue_purge(&tp->ucopy.prequeue);
1939 /* Clean up a referenced TCP bind bucket. */
1940 if (inet_csk(sk)->icsk_bind_hash)
1944 * If sendmsg cached page exists, toss it.
1946 if (sk->sk_sndmsg_page) {
1947 __free_page(sk->sk_sndmsg_page);
1948 sk->sk_sndmsg_page = NULL;
1951 /* TCP Cookie Transactions */
1952 if (tp->cookie_values != NULL) {
1953 kref_put(&tp->cookie_values->kref,
1954 tcp_cookie_values_release);
1955 tp->cookie_values = NULL;
1958 percpu_counter_dec(&tcp_sockets_allocated);
1960 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1962 #ifdef CONFIG_PROC_FS
1963 /* Proc filesystem TCP sock list dumping. */
1965 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1967 return hlist_nulls_empty(head) ? NULL :
1968 list_entry(head->first, struct inet_timewait_sock, tw_node);
1971 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1973 return !is_a_nulls(tw->tw_node.next) ?
1974 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1978 * Get next listener socket follow cur. If cur is NULL, get first socket
1979 * starting from bucket given in st->bucket; when st->bucket is zero the
1980 * very first socket in the hash table is returned.
1982 static void *listening_get_next(struct seq_file *seq, void *cur)
1984 struct inet_connection_sock *icsk;
1985 struct hlist_nulls_node *node;
1986 struct sock *sk = cur;
1987 struct inet_listen_hashbucket *ilb;
1988 struct tcp_iter_state *st = seq->private;
1989 struct net *net = seq_file_net(seq);
1992 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1993 spin_lock_bh(&ilb->lock);
1994 sk = sk_nulls_head(&ilb->head);
1998 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2002 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2003 struct request_sock *req = cur;
2005 icsk = inet_csk(st->syn_wait_sk);
2009 if (req->rsk_ops->family == st->family) {
2015 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2018 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2020 sk = sk_nulls_next(st->syn_wait_sk);
2021 st->state = TCP_SEQ_STATE_LISTENING;
2022 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2024 icsk = inet_csk(sk);
2025 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2026 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2028 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2029 sk = sk_nulls_next(sk);
2032 sk_nulls_for_each_from(sk, node) {
2033 if (!net_eq(sock_net(sk), net))
2035 if (sk->sk_family == st->family) {
2039 icsk = inet_csk(sk);
2040 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2041 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2043 st->uid = sock_i_uid(sk);
2044 st->syn_wait_sk = sk;
2045 st->state = TCP_SEQ_STATE_OPENREQ;
2049 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2051 spin_unlock_bh(&ilb->lock);
2053 if (++st->bucket < INET_LHTABLE_SIZE) {
2054 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2055 spin_lock_bh(&ilb->lock);
2056 sk = sk_nulls_head(&ilb->head);
2064 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2066 struct tcp_iter_state *st = seq->private;
2071 rc = listening_get_next(seq, NULL);
2073 while (rc && *pos) {
2074 rc = listening_get_next(seq, rc);
2080 static inline int empty_bucket(struct tcp_iter_state *st)
2082 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2083 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2087 * Get first established socket starting from bucket given in st->bucket.
2088 * If st->bucket is zero, the very first socket in the hash is returned.
2090 static void *established_get_first(struct seq_file *seq)
2092 struct tcp_iter_state *st = seq->private;
2093 struct net *net = seq_file_net(seq);
2097 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2099 struct hlist_nulls_node *node;
2100 struct inet_timewait_sock *tw;
2101 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2103 /* Lockless fast path for the common case of empty buckets */
2104 if (empty_bucket(st))
2108 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2109 if (sk->sk_family != st->family ||
2110 !net_eq(sock_net(sk), net)) {
2116 st->state = TCP_SEQ_STATE_TIME_WAIT;
2117 inet_twsk_for_each(tw, node,
2118 &tcp_hashinfo.ehash[st->bucket].twchain) {
2119 if (tw->tw_family != st->family ||
2120 !net_eq(twsk_net(tw), net)) {
2126 spin_unlock_bh(lock);
2127 st->state = TCP_SEQ_STATE_ESTABLISHED;
2133 static void *established_get_next(struct seq_file *seq, void *cur)
2135 struct sock *sk = cur;
2136 struct inet_timewait_sock *tw;
2137 struct hlist_nulls_node *node;
2138 struct tcp_iter_state *st = seq->private;
2139 struct net *net = seq_file_net(seq);
2144 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2148 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2155 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2156 st->state = TCP_SEQ_STATE_ESTABLISHED;
2158 /* Look for next non empty bucket */
2160 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2163 if (st->bucket > tcp_hashinfo.ehash_mask)
2166 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2167 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2169 sk = sk_nulls_next(sk);
2171 sk_nulls_for_each_from(sk, node) {
2172 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2176 st->state = TCP_SEQ_STATE_TIME_WAIT;
2177 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2185 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2187 struct tcp_iter_state *st = seq->private;
2191 rc = established_get_first(seq);
2194 rc = established_get_next(seq, rc);
2200 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2203 struct tcp_iter_state *st = seq->private;
2205 st->state = TCP_SEQ_STATE_LISTENING;
2206 rc = listening_get_idx(seq, &pos);
2209 st->state = TCP_SEQ_STATE_ESTABLISHED;
2210 rc = established_get_idx(seq, pos);
2216 static void *tcp_seek_last_pos(struct seq_file *seq)
2218 struct tcp_iter_state *st = seq->private;
2219 int offset = st->offset;
2220 int orig_num = st->num;
2223 switch (st->state) {
2224 case TCP_SEQ_STATE_OPENREQ:
2225 case TCP_SEQ_STATE_LISTENING:
2226 if (st->bucket >= INET_LHTABLE_SIZE)
2228 st->state = TCP_SEQ_STATE_LISTENING;
2229 rc = listening_get_next(seq, NULL);
2230 while (offset-- && rc)
2231 rc = listening_get_next(seq, rc);
2236 case TCP_SEQ_STATE_ESTABLISHED:
2237 case TCP_SEQ_STATE_TIME_WAIT:
2238 st->state = TCP_SEQ_STATE_ESTABLISHED;
2239 if (st->bucket > tcp_hashinfo.ehash_mask)
2241 rc = established_get_first(seq);
2242 while (offset-- && rc)
2243 rc = established_get_next(seq, rc);
2251 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2253 struct tcp_iter_state *st = seq->private;
2256 if (*pos && *pos == st->last_pos) {
2257 rc = tcp_seek_last_pos(seq);
2262 st->state = TCP_SEQ_STATE_LISTENING;
2266 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2269 st->last_pos = *pos;
2273 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2275 struct tcp_iter_state *st = seq->private;
2278 if (v == SEQ_START_TOKEN) {
2279 rc = tcp_get_idx(seq, 0);
2283 switch (st->state) {
2284 case TCP_SEQ_STATE_OPENREQ:
2285 case TCP_SEQ_STATE_LISTENING:
2286 rc = listening_get_next(seq, v);
2288 st->state = TCP_SEQ_STATE_ESTABLISHED;
2291 rc = established_get_first(seq);
2294 case TCP_SEQ_STATE_ESTABLISHED:
2295 case TCP_SEQ_STATE_TIME_WAIT:
2296 rc = established_get_next(seq, v);
2301 st->last_pos = *pos;
2305 static void tcp_seq_stop(struct seq_file *seq, void *v)
2307 struct tcp_iter_state *st = seq->private;
2309 switch (st->state) {
2310 case TCP_SEQ_STATE_OPENREQ:
2312 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2313 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2315 case TCP_SEQ_STATE_LISTENING:
2316 if (v != SEQ_START_TOKEN)
2317 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2319 case TCP_SEQ_STATE_TIME_WAIT:
2320 case TCP_SEQ_STATE_ESTABLISHED:
2322 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2327 static int tcp_seq_open(struct inode *inode, struct file *file)
2329 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2330 struct tcp_iter_state *s;
2333 err = seq_open_net(inode, file, &afinfo->seq_ops,
2334 sizeof(struct tcp_iter_state));
2338 s = ((struct seq_file *)file->private_data)->private;
2339 s->family = afinfo->family;
2344 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2347 struct proc_dir_entry *p;
2349 afinfo->seq_fops.open = tcp_seq_open;
2350 afinfo->seq_fops.read = seq_read;
2351 afinfo->seq_fops.llseek = seq_lseek;
2352 afinfo->seq_fops.release = seq_release_net;
2354 afinfo->seq_ops.start = tcp_seq_start;
2355 afinfo->seq_ops.next = tcp_seq_next;
2356 afinfo->seq_ops.stop = tcp_seq_stop;
2358 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2359 &afinfo->seq_fops, afinfo);
2364 EXPORT_SYMBOL(tcp_proc_register);
2366 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2368 proc_net_remove(net, afinfo->name);
2370 EXPORT_SYMBOL(tcp_proc_unregister);
2372 static void get_openreq4(struct sock *sk, struct request_sock *req,
2373 struct seq_file *f, int i, int uid, int *len)
2375 const struct inet_request_sock *ireq = inet_rsk(req);
2376 int ttd = req->expires - jiffies;
2378 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2379 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2382 ntohs(inet_sk(sk)->inet_sport),
2384 ntohs(ireq->rmt_port),
2386 0, 0, /* could print option size, but that is af dependent. */
2387 1, /* timers active (only the expire timer) */
2388 jiffies_to_clock_t(ttd),
2391 0, /* non standard timer */
2392 0, /* open_requests have no inode */
2393 atomic_read(&sk->sk_refcnt),
2398 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2401 unsigned long timer_expires;
2402 struct tcp_sock *tp = tcp_sk(sk);
2403 const struct inet_connection_sock *icsk = inet_csk(sk);
2404 struct inet_sock *inet = inet_sk(sk);
2405 __be32 dest = inet->inet_daddr;
2406 __be32 src = inet->inet_rcv_saddr;
2407 __u16 destp = ntohs(inet->inet_dport);
2408 __u16 srcp = ntohs(inet->inet_sport);
2411 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2413 timer_expires = icsk->icsk_timeout;
2414 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2416 timer_expires = icsk->icsk_timeout;
2417 } else if (timer_pending(&sk->sk_timer)) {
2419 timer_expires = sk->sk_timer.expires;
2422 timer_expires = jiffies;
2425 if (sk->sk_state == TCP_LISTEN)
2426 rx_queue = sk->sk_ack_backlog;
2429 * because we dont lock socket, we might find a transient negative value
2431 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2433 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2434 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2435 i, src, srcp, dest, destp, sk->sk_state,
2436 tp->write_seq - tp->snd_una,
2439 jiffies_to_clock_t(timer_expires - jiffies),
2440 icsk->icsk_retransmits,
2442 icsk->icsk_probes_out,
2444 atomic_read(&sk->sk_refcnt), sk,
2445 jiffies_to_clock_t(icsk->icsk_rto),
2446 jiffies_to_clock_t(icsk->icsk_ack.ato),
2447 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2449 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2453 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2454 struct seq_file *f, int i, int *len)
2458 int ttd = tw->tw_ttd - jiffies;
2463 dest = tw->tw_daddr;
2464 src = tw->tw_rcv_saddr;
2465 destp = ntohs(tw->tw_dport);
2466 srcp = ntohs(tw->tw_sport);
2468 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2469 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2470 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2471 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2472 atomic_read(&tw->tw_refcnt), tw, len);
2477 static int tcp4_seq_show(struct seq_file *seq, void *v)
2479 struct tcp_iter_state *st;
2482 if (v == SEQ_START_TOKEN) {
2483 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2484 " sl local_address rem_address st tx_queue "
2485 "rx_queue tr tm->when retrnsmt uid timeout "
2491 switch (st->state) {
2492 case TCP_SEQ_STATE_LISTENING:
2493 case TCP_SEQ_STATE_ESTABLISHED:
2494 get_tcp4_sock(v, seq, st->num, &len);
2496 case TCP_SEQ_STATE_OPENREQ:
2497 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2499 case TCP_SEQ_STATE_TIME_WAIT:
2500 get_timewait4_sock(v, seq, st->num, &len);
2503 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2508 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2512 .owner = THIS_MODULE,
2515 .show = tcp4_seq_show,
2519 static int __net_init tcp4_proc_init_net(struct net *net)
2521 return tcp_proc_register(net, &tcp4_seq_afinfo);
2524 static void __net_exit tcp4_proc_exit_net(struct net *net)
2526 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2529 static struct pernet_operations tcp4_net_ops = {
2530 .init = tcp4_proc_init_net,
2531 .exit = tcp4_proc_exit_net,
2534 int __init tcp4_proc_init(void)
2536 return register_pernet_subsys(&tcp4_net_ops);
2539 void tcp4_proc_exit(void)
2541 unregister_pernet_subsys(&tcp4_net_ops);
2543 #endif /* CONFIG_PROC_FS */
2545 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2547 const struct iphdr *iph = skb_gro_network_header(skb);
2549 switch (skb->ip_summed) {
2550 case CHECKSUM_COMPLETE:
2551 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2553 skb->ip_summed = CHECKSUM_UNNECESSARY;
2559 NAPI_GRO_CB(skb)->flush = 1;
2563 return tcp_gro_receive(head, skb);
2566 int tcp4_gro_complete(struct sk_buff *skb)
2568 const struct iphdr *iph = ip_hdr(skb);
2569 struct tcphdr *th = tcp_hdr(skb);
2571 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2572 iph->saddr, iph->daddr, 0);
2573 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2575 return tcp_gro_complete(skb);
2578 struct proto tcp_prot = {
2580 .owner = THIS_MODULE,
2582 .connect = tcp_v4_connect,
2583 .disconnect = tcp_disconnect,
2584 .accept = inet_csk_accept,
2586 .init = tcp_v4_init_sock,
2587 .destroy = tcp_v4_destroy_sock,
2588 .shutdown = tcp_shutdown,
2589 .setsockopt = tcp_setsockopt,
2590 .getsockopt = tcp_getsockopt,
2591 .recvmsg = tcp_recvmsg,
2592 .sendmsg = tcp_sendmsg,
2593 .sendpage = tcp_sendpage,
2594 .backlog_rcv = tcp_v4_do_rcv,
2596 .unhash = inet_unhash,
2597 .get_port = inet_csk_get_port,
2598 .enter_memory_pressure = tcp_enter_memory_pressure,
2599 .sockets_allocated = &tcp_sockets_allocated,
2600 .orphan_count = &tcp_orphan_count,
2601 .memory_allocated = &tcp_memory_allocated,
2602 .memory_pressure = &tcp_memory_pressure,
2603 .sysctl_mem = sysctl_tcp_mem,
2604 .sysctl_wmem = sysctl_tcp_wmem,
2605 .sysctl_rmem = sysctl_tcp_rmem,
2606 .max_header = MAX_TCP_HEADER,
2607 .obj_size = sizeof(struct tcp_sock),
2608 .slab_flags = SLAB_DESTROY_BY_RCU,
2609 .twsk_prot = &tcp_timewait_sock_ops,
2610 .rsk_prot = &tcp_request_sock_ops,
2611 .h.hashinfo = &tcp_hashinfo,
2612 .no_autobind = true,
2613 #ifdef CONFIG_COMPAT
2614 .compat_setsockopt = compat_tcp_setsockopt,
2615 .compat_getsockopt = compat_tcp_getsockopt,
2618 EXPORT_SYMBOL(tcp_prot);
2621 static int __net_init tcp_sk_init(struct net *net)
2623 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2624 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2627 static void __net_exit tcp_sk_exit(struct net *net)
2629 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2632 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2634 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2637 static struct pernet_operations __net_initdata tcp_sk_ops = {
2638 .init = tcp_sk_init,
2639 .exit = tcp_sk_exit,
2640 .exit_batch = tcp_sk_exit_batch,
2643 void __init tcp_v4_init(void)
2645 inet_hashinfo_init(&tcp_hashinfo);
2646 if (register_pernet_subsys(&tcp_sk_ops))
2647 panic("Failed to create the TCP control socket.\n");