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.
53 #define pr_fmt(fmt) "TCP: " fmt
55 #include <linux/bottom_half.h>
56 #include <linux/types.h>
57 #include <linux/fcntl.h>
58 #include <linux/module.h>
59 #include <linux/random.h>
60 #include <linux/cache.h>
61 #include <linux/jhash.h>
62 #include <linux/init.h>
63 #include <linux/times.h>
64 #include <linux/slab.h>
66 #include <net/net_namespace.h>
68 #include <net/inet_hashtables.h>
70 #include <net/transp_v6.h>
72 #include <net/inet_common.h>
73 #include <net/timewait_sock.h>
75 #include <net/secure_seq.h>
76 #include <net/busy_poll.h>
78 #include <linux/inet.h>
79 #include <linux/ipv6.h>
80 #include <linux/stddef.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
84 #include <linux/crypto.h>
85 #include <linux/scatterlist.h>
87 int sysctl_tcp_tw_reuse __read_mostly;
88 int sysctl_tcp_low_latency __read_mostly;
89 EXPORT_SYMBOL(sysctl_tcp_low_latency);
91 #ifdef CONFIG_TCP_MD5SIG
92 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
93 __be32 daddr, __be32 saddr, const struct tcphdr *th);
96 struct inet_hashinfo tcp_hashinfo;
97 EXPORT_SYMBOL(tcp_hashinfo);
99 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
101 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
104 tcp_hdr(skb)->source);
107 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
109 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
110 struct tcp_sock *tp = tcp_sk(sk);
112 /* With PAWS, it is safe from the viewpoint
113 of data integrity. Even without PAWS it is safe provided sequence
114 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
116 Actually, the idea is close to VJ's one, only timestamp cache is
117 held not per host, but per port pair and TW bucket is used as state
120 If TW bucket has been already destroyed we fall back to VJ's scheme
121 and use initial timestamp retrieved from peer table.
123 if (tcptw->tw_ts_recent_stamp &&
124 (!twp || (sysctl_tcp_tw_reuse &&
125 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
126 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
127 if (tp->write_seq == 0)
129 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
130 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
137 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
139 /* This will initiate an outgoing connection. */
140 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
142 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
143 struct inet_sock *inet = inet_sk(sk);
144 struct tcp_sock *tp = tcp_sk(sk);
145 __be16 orig_sport, orig_dport;
146 __be32 daddr, nexthop;
150 struct ip_options_rcu *inet_opt;
152 if (addr_len < sizeof(struct sockaddr_in))
155 if (usin->sin_family != AF_INET)
156 return -EAFNOSUPPORT;
158 nexthop = daddr = usin->sin_addr.s_addr;
159 inet_opt = rcu_dereference_protected(inet->inet_opt,
160 sock_owned_by_user(sk));
161 if (inet_opt && inet_opt->opt.srr) {
164 nexthop = inet_opt->opt.faddr;
167 orig_sport = inet->inet_sport;
168 orig_dport = usin->sin_port;
169 fl4 = &inet->cork.fl.u.ip4;
170 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
171 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
173 orig_sport, orig_dport, sk);
176 if (err == -ENETUNREACH)
177 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
181 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
186 if (!inet_opt || !inet_opt->opt.srr)
189 if (!inet->inet_saddr)
190 inet->inet_saddr = fl4->saddr;
191 sk_rcv_saddr_set(sk, inet->inet_saddr);
193 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
194 /* Reset inherited state */
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
197 if (likely(!tp->repair))
201 if (tcp_death_row.sysctl_tw_recycle &&
202 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
203 tcp_fetch_timewait_stamp(sk, &rt->dst);
205 inet->inet_dport = usin->sin_port;
206 sk_daddr_set(sk, daddr);
208 inet_csk(sk)->icsk_ext_hdr_len = 0;
210 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
212 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
214 /* Socket identity is still unknown (sport may be zero).
215 * However we set state to SYN-SENT and not releasing socket
216 * lock select source port, enter ourselves into the hash tables and
217 * complete initialization after this.
219 tcp_set_state(sk, TCP_SYN_SENT);
220 err = inet_hash_connect(&tcp_death_row, sk);
226 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
227 inet->inet_sport, inet->inet_dport, sk);
233 /* OK, now commit destination to socket. */
234 sk->sk_gso_type = SKB_GSO_TCPV4;
235 sk_setup_caps(sk, &rt->dst);
237 if (!tp->write_seq && likely(!tp->repair))
238 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
243 inet->inet_id = tp->write_seq ^ jiffies;
245 err = tcp_connect(sk);
255 * This unhashes the socket and releases the local port,
258 tcp_set_state(sk, TCP_CLOSE);
260 sk->sk_route_caps = 0;
261 inet->inet_dport = 0;
264 EXPORT_SYMBOL(tcp_v4_connect);
267 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
268 * It can be called through tcp_release_cb() if socket was owned by user
269 * at the time tcp_v4_err() was called to handle ICMP message.
271 void tcp_v4_mtu_reduced(struct sock *sk)
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
275 u32 mtu = tcp_sk(sk)->mtu_info;
277 dst = inet_csk_update_pmtu(sk, mtu);
281 /* Something is about to be wrong... Remember soft error
282 * for the case, if this connection will not able to recover.
284 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
285 sk->sk_err_soft = EMSGSIZE;
289 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
290 ip_sk_accept_pmtu(sk) &&
291 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
292 tcp_sync_mss(sk, mtu);
294 /* Resend the TCP packet because it's
295 * clear that the old packet has been
296 * dropped. This is the new "fast" path mtu
299 tcp_simple_retransmit(sk);
300 } /* else let the usual retransmit timer handle it */
302 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
304 static void do_redirect(struct sk_buff *skb, struct sock *sk)
306 struct dst_entry *dst = __sk_dst_check(sk, 0);
309 dst->ops->redirect(dst, sk, skb);
313 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
314 void tcp_req_err(struct sock *sk, u32 seq)
316 struct request_sock *req = inet_reqsk(sk);
317 struct net *net = sock_net(sk);
319 /* ICMPs are not backlogged, hence we cannot get
320 * an established socket here.
324 if (seq != tcp_rsk(req)->snt_isn) {
325 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
328 * Still in SYN_RECV, just remove it silently.
329 * There is no good way to pass the error to the newly
330 * created socket, and POSIX does not want network
331 * errors returned from accept().
333 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
334 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS);
338 EXPORT_SYMBOL(tcp_req_err);
341 * This routine is called by the ICMP module when it gets some
342 * sort of error condition. If err < 0 then the socket should
343 * be closed and the error returned to the user. If err > 0
344 * it's just the icmp type << 8 | icmp code. After adjustment
345 * header points to the first 8 bytes of the tcp header. We need
346 * to find the appropriate port.
348 * The locking strategy used here is very "optimistic". When
349 * someone else accesses the socket the ICMP is just dropped
350 * and for some paths there is no check at all.
351 * A more general error queue to queue errors for later handling
352 * is probably better.
356 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
358 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
359 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
360 struct inet_connection_sock *icsk;
362 struct inet_sock *inet;
363 const int type = icmp_hdr(icmp_skb)->type;
364 const int code = icmp_hdr(icmp_skb)->code;
367 struct request_sock *fastopen;
371 struct net *net = dev_net(icmp_skb->dev);
373 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
374 th->dest, iph->saddr, ntohs(th->source),
377 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
380 if (sk->sk_state == TCP_TIME_WAIT) {
381 inet_twsk_put(inet_twsk(sk));
384 seq = ntohl(th->seq);
385 if (sk->sk_state == TCP_NEW_SYN_RECV)
386 return tcp_req_err(sk, seq);
389 /* If too many ICMPs get dropped on busy
390 * servers this needs to be solved differently.
391 * We do take care of PMTU discovery (RFC1191) special case :
392 * we can receive locally generated ICMP messages while socket is held.
394 if (sock_owned_by_user(sk)) {
395 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
396 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
398 if (sk->sk_state == TCP_CLOSE)
401 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
402 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
408 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
409 fastopen = tp->fastopen_rsk;
410 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
411 if (sk->sk_state != TCP_LISTEN &&
412 !between(seq, snd_una, tp->snd_nxt)) {
413 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
419 do_redirect(icmp_skb, sk);
421 case ICMP_SOURCE_QUENCH:
422 /* Just silently ignore these. */
424 case ICMP_PARAMETERPROB:
427 case ICMP_DEST_UNREACH:
428 if (code > NR_ICMP_UNREACH)
431 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
432 /* We are not interested in TCP_LISTEN and open_requests
433 * (SYN-ACKs send out by Linux are always <576bytes so
434 * they should go through unfragmented).
436 if (sk->sk_state == TCP_LISTEN)
440 if (!sock_owned_by_user(sk)) {
441 tcp_v4_mtu_reduced(sk);
443 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
449 err = icmp_err_convert[code].errno;
450 /* check if icmp_skb allows revert of backoff
451 * (see draft-zimmermann-tcp-lcd) */
452 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
454 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
455 !icsk->icsk_backoff || fastopen)
458 if (sock_owned_by_user(sk))
461 icsk->icsk_backoff--;
462 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
464 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
466 skb = tcp_write_queue_head(sk);
469 remaining = icsk->icsk_rto -
471 tcp_time_stamp - tcp_skb_timestamp(skb));
474 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
475 remaining, TCP_RTO_MAX);
477 /* RTO revert clocked out retransmission.
478 * Will retransmit now */
479 tcp_retransmit_timer(sk);
483 case ICMP_TIME_EXCEEDED:
490 switch (sk->sk_state) {
493 /* Only in fast or simultaneous open. If a fast open socket is
494 * is already accepted it is treated as a connected one below.
496 if (fastopen && !fastopen->sk)
499 if (!sock_owned_by_user(sk)) {
502 sk->sk_error_report(sk);
506 sk->sk_err_soft = err;
511 /* If we've already connected we will keep trying
512 * until we time out, or the user gives up.
514 * rfc1122 4.2.3.9 allows to consider as hard errors
515 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
516 * but it is obsoleted by pmtu discovery).
518 * Note, that in modern internet, where routing is unreliable
519 * and in each dark corner broken firewalls sit, sending random
520 * errors ordered by their masters even this two messages finally lose
521 * their original sense (even Linux sends invalid PORT_UNREACHs)
523 * Now we are in compliance with RFCs.
528 if (!sock_owned_by_user(sk) && inet->recverr) {
530 sk->sk_error_report(sk);
531 } else { /* Only an error on timeout */
532 sk->sk_err_soft = err;
540 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
542 struct tcphdr *th = tcp_hdr(skb);
544 if (skb->ip_summed == CHECKSUM_PARTIAL) {
545 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
546 skb->csum_start = skb_transport_header(skb) - skb->head;
547 skb->csum_offset = offsetof(struct tcphdr, check);
549 th->check = tcp_v4_check(skb->len, saddr, daddr,
556 /* This routine computes an IPv4 TCP checksum. */
557 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
559 const struct inet_sock *inet = inet_sk(sk);
561 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
563 EXPORT_SYMBOL(tcp_v4_send_check);
566 * This routine will send an RST to the other tcp.
568 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
570 * Answer: if a packet caused RST, it is not for a socket
571 * existing in our system, if it is matched to a socket,
572 * it is just duplicate segment or bug in other side's TCP.
573 * So that we build reply only basing on parameters
574 * arrived with segment.
575 * Exception: precedence violation. We do not implement it in any case.
578 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
580 const struct tcphdr *th = tcp_hdr(skb);
583 #ifdef CONFIG_TCP_MD5SIG
584 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
587 struct ip_reply_arg arg;
588 #ifdef CONFIG_TCP_MD5SIG
589 struct tcp_md5sig_key *key = NULL;
590 const __u8 *hash_location = NULL;
591 unsigned char newhash[16];
593 struct sock *sk1 = NULL;
597 /* Never send a reset in response to a reset. */
601 /* If sk not NULL, it means we did a successful lookup and incoming
602 * route had to be correct. prequeue might have dropped our dst.
604 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
607 /* Swap the send and the receive. */
608 memset(&rep, 0, sizeof(rep));
609 rep.th.dest = th->source;
610 rep.th.source = th->dest;
611 rep.th.doff = sizeof(struct tcphdr) / 4;
615 rep.th.seq = th->ack_seq;
618 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
619 skb->len - (th->doff << 2));
622 memset(&arg, 0, sizeof(arg));
623 arg.iov[0].iov_base = (unsigned char *)&rep;
624 arg.iov[0].iov_len = sizeof(rep.th);
626 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
627 #ifdef CONFIG_TCP_MD5SIG
628 hash_location = tcp_parse_md5sig_option(th);
629 if (sk && sk_fullsock(sk)) {
630 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
631 &ip_hdr(skb)->saddr, AF_INET);
632 } else if (hash_location) {
634 * active side is lost. Try to find listening socket through
635 * source port, and then find md5 key through listening socket.
636 * we are not loose security here:
637 * Incoming packet is checked with md5 hash with finding key,
638 * no RST generated if md5 hash doesn't match.
640 sk1 = __inet_lookup_listener(net,
641 &tcp_hashinfo, ip_hdr(skb)->saddr,
642 th->source, ip_hdr(skb)->daddr,
643 ntohs(th->source), inet_iif(skb));
644 /* don't send rst if it can't find key */
648 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
649 &ip_hdr(skb)->saddr, AF_INET);
653 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
654 if (genhash || memcmp(hash_location, newhash, 16) != 0)
659 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
661 (TCPOPT_MD5SIG << 8) |
663 /* Update length and the length the header thinks exists */
664 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
665 rep.th.doff = arg.iov[0].iov_len / 4;
667 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
668 key, ip_hdr(skb)->saddr,
669 ip_hdr(skb)->daddr, &rep.th);
672 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
673 ip_hdr(skb)->saddr, /* XXX */
674 arg.iov[0].iov_len, IPPROTO_TCP, 0);
675 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
676 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
678 /* When socket is gone, all binding information is lost.
679 * routing might fail in this case. No choice here, if we choose to force
680 * input interface, we will misroute in case of asymmetric route.
683 arg.bound_dev_if = sk->sk_bound_dev_if;
685 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
686 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
688 arg.tos = ip_hdr(skb)->tos;
689 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
690 skb, &TCP_SKB_CB(skb)->header.h4.opt,
691 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
692 &arg, arg.iov[0].iov_len);
694 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
695 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
697 #ifdef CONFIG_TCP_MD5SIG
706 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
707 outside socket context is ugly, certainly. What can I do?
710 static void tcp_v4_send_ack(struct net *net,
711 struct sk_buff *skb, u32 seq, u32 ack,
712 u32 win, u32 tsval, u32 tsecr, int oif,
713 struct tcp_md5sig_key *key,
714 int reply_flags, u8 tos)
716 const struct tcphdr *th = tcp_hdr(skb);
719 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
720 #ifdef CONFIG_TCP_MD5SIG
721 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
725 struct ip_reply_arg arg;
727 memset(&rep.th, 0, sizeof(struct tcphdr));
728 memset(&arg, 0, sizeof(arg));
730 arg.iov[0].iov_base = (unsigned char *)&rep;
731 arg.iov[0].iov_len = sizeof(rep.th);
733 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
734 (TCPOPT_TIMESTAMP << 8) |
736 rep.opt[1] = htonl(tsval);
737 rep.opt[2] = htonl(tsecr);
738 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
741 /* Swap the send and the receive. */
742 rep.th.dest = th->source;
743 rep.th.source = th->dest;
744 rep.th.doff = arg.iov[0].iov_len / 4;
745 rep.th.seq = htonl(seq);
746 rep.th.ack_seq = htonl(ack);
748 rep.th.window = htons(win);
750 #ifdef CONFIG_TCP_MD5SIG
752 int offset = (tsecr) ? 3 : 0;
754 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
756 (TCPOPT_MD5SIG << 8) |
758 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
759 rep.th.doff = arg.iov[0].iov_len/4;
761 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
762 key, ip_hdr(skb)->saddr,
763 ip_hdr(skb)->daddr, &rep.th);
766 arg.flags = reply_flags;
767 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
768 ip_hdr(skb)->saddr, /* XXX */
769 arg.iov[0].iov_len, IPPROTO_TCP, 0);
770 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
772 arg.bound_dev_if = oif;
774 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
775 skb, &TCP_SKB_CB(skb)->header.h4.opt,
776 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
777 &arg, arg.iov[0].iov_len);
779 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
782 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
784 struct inet_timewait_sock *tw = inet_twsk(sk);
785 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
787 tcp_v4_send_ack(sock_net(sk), skb,
788 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
789 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
790 tcp_time_stamp + tcptw->tw_ts_offset,
793 tcp_twsk_md5_key(tcptw),
794 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
801 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
802 struct request_sock *req)
804 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
805 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
807 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
810 tcp_v4_send_ack(sock_net(sk), skb, seq,
811 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
815 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
817 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
822 * Send a SYN-ACK after having received a SYN.
823 * This still operates on a request_sock only, not on a big
826 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
828 struct request_sock *req,
829 struct tcp_fastopen_cookie *foc,
832 const struct inet_request_sock *ireq = inet_rsk(req);
837 /* First, grab a route. */
838 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
841 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
844 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
846 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
849 err = net_xmit_eval(err);
856 * IPv4 request_sock destructor.
858 static void tcp_v4_reqsk_destructor(struct request_sock *req)
860 kfree(inet_rsk(req)->opt);
864 #ifdef CONFIG_TCP_MD5SIG
866 * RFC2385 MD5 checksumming requires a mapping of
867 * IP address->MD5 Key.
868 * We need to maintain these in the sk structure.
871 /* Find the Key structure for an address. */
872 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
873 const union tcp_md5_addr *addr,
876 const struct tcp_sock *tp = tcp_sk(sk);
877 struct tcp_md5sig_key *key;
878 unsigned int size = sizeof(struct in_addr);
879 const struct tcp_md5sig_info *md5sig;
881 /* caller either holds rcu_read_lock() or socket lock */
882 md5sig = rcu_dereference_check(tp->md5sig_info,
883 sock_owned_by_user(sk) ||
884 lockdep_is_held((spinlock_t *)&sk->sk_lock.slock));
887 #if IS_ENABLED(CONFIG_IPV6)
888 if (family == AF_INET6)
889 size = sizeof(struct in6_addr);
891 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
892 if (key->family != family)
894 if (!memcmp(&key->addr, addr, size))
899 EXPORT_SYMBOL(tcp_md5_do_lookup);
901 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
902 const struct sock *addr_sk)
904 const union tcp_md5_addr *addr;
906 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
907 return tcp_md5_do_lookup(sk, addr, AF_INET);
909 EXPORT_SYMBOL(tcp_v4_md5_lookup);
911 /* This can be called on a newly created socket, from other files */
912 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
913 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
915 /* Add Key to the list */
916 struct tcp_md5sig_key *key;
917 struct tcp_sock *tp = tcp_sk(sk);
918 struct tcp_md5sig_info *md5sig;
920 key = tcp_md5_do_lookup(sk, addr, family);
922 /* Pre-existing entry - just update that one. */
923 memcpy(key->key, newkey, newkeylen);
924 key->keylen = newkeylen;
928 md5sig = rcu_dereference_protected(tp->md5sig_info,
929 sock_owned_by_user(sk) ||
930 lockdep_is_held(&sk->sk_lock.slock));
932 md5sig = kmalloc(sizeof(*md5sig), gfp);
936 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
937 INIT_HLIST_HEAD(&md5sig->head);
938 rcu_assign_pointer(tp->md5sig_info, md5sig);
941 key = sock_kmalloc(sk, sizeof(*key), gfp);
944 if (!tcp_alloc_md5sig_pool()) {
945 sock_kfree_s(sk, key, sizeof(*key));
949 memcpy(key->key, newkey, newkeylen);
950 key->keylen = newkeylen;
951 key->family = family;
952 memcpy(&key->addr, addr,
953 (family == AF_INET6) ? sizeof(struct in6_addr) :
954 sizeof(struct in_addr));
955 hlist_add_head_rcu(&key->node, &md5sig->head);
958 EXPORT_SYMBOL(tcp_md5_do_add);
960 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
962 struct tcp_md5sig_key *key;
964 key = tcp_md5_do_lookup(sk, addr, family);
967 hlist_del_rcu(&key->node);
968 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
972 EXPORT_SYMBOL(tcp_md5_do_del);
974 static void tcp_clear_md5_list(struct sock *sk)
976 struct tcp_sock *tp = tcp_sk(sk);
977 struct tcp_md5sig_key *key;
978 struct hlist_node *n;
979 struct tcp_md5sig_info *md5sig;
981 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
983 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
984 hlist_del_rcu(&key->node);
985 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
990 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
993 struct tcp_md5sig cmd;
994 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
996 if (optlen < sizeof(cmd))
999 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1002 if (sin->sin_family != AF_INET)
1005 if (!cmd.tcpm_keylen)
1006 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1009 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1012 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1013 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1017 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1018 __be32 daddr, __be32 saddr, int nbytes)
1020 struct tcp4_pseudohdr *bp;
1021 struct scatterlist sg;
1023 bp = &hp->md5_blk.ip4;
1026 * 1. the TCP pseudo-header (in the order: source IP address,
1027 * destination IP address, zero-padded protocol number, and
1033 bp->protocol = IPPROTO_TCP;
1034 bp->len = cpu_to_be16(nbytes);
1036 sg_init_one(&sg, bp, sizeof(*bp));
1037 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1040 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1041 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1043 struct tcp_md5sig_pool *hp;
1044 struct hash_desc *desc;
1046 hp = tcp_get_md5sig_pool();
1048 goto clear_hash_noput;
1049 desc = &hp->md5_desc;
1051 if (crypto_hash_init(desc))
1053 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1055 if (tcp_md5_hash_header(hp, th))
1057 if (tcp_md5_hash_key(hp, key))
1059 if (crypto_hash_final(desc, md5_hash))
1062 tcp_put_md5sig_pool();
1066 tcp_put_md5sig_pool();
1068 memset(md5_hash, 0, 16);
1072 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1073 const struct sock *sk,
1074 const struct sk_buff *skb)
1076 struct tcp_md5sig_pool *hp;
1077 struct hash_desc *desc;
1078 const struct tcphdr *th = tcp_hdr(skb);
1079 __be32 saddr, daddr;
1081 if (sk) { /* valid for establish/request sockets */
1082 saddr = sk->sk_rcv_saddr;
1083 daddr = sk->sk_daddr;
1085 const struct iphdr *iph = ip_hdr(skb);
1090 hp = tcp_get_md5sig_pool();
1092 goto clear_hash_noput;
1093 desc = &hp->md5_desc;
1095 if (crypto_hash_init(desc))
1098 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1100 if (tcp_md5_hash_header(hp, th))
1102 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1104 if (tcp_md5_hash_key(hp, key))
1106 if (crypto_hash_final(desc, md5_hash))
1109 tcp_put_md5sig_pool();
1113 tcp_put_md5sig_pool();
1115 memset(md5_hash, 0, 16);
1118 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1122 /* Called with rcu_read_lock() */
1123 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1124 const struct sk_buff *skb)
1126 #ifdef CONFIG_TCP_MD5SIG
1128 * This gets called for each TCP segment that arrives
1129 * so we want to be efficient.
1130 * We have 3 drop cases:
1131 * o No MD5 hash and one expected.
1132 * o MD5 hash and we're not expecting one.
1133 * o MD5 hash and its wrong.
1135 const __u8 *hash_location = NULL;
1136 struct tcp_md5sig_key *hash_expected;
1137 const struct iphdr *iph = ip_hdr(skb);
1138 const struct tcphdr *th = tcp_hdr(skb);
1140 unsigned char newhash[16];
1142 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1144 hash_location = tcp_parse_md5sig_option(th);
1146 /* We've parsed the options - do we have a hash? */
1147 if (!hash_expected && !hash_location)
1150 if (hash_expected && !hash_location) {
1151 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1155 if (!hash_expected && hash_location) {
1156 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1160 /* Okay, so this is hash_expected and hash_location -
1161 * so we need to calculate the checksum.
1163 genhash = tcp_v4_md5_hash_skb(newhash,
1167 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1168 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1169 &iph->saddr, ntohs(th->source),
1170 &iph->daddr, ntohs(th->dest),
1171 genhash ? " tcp_v4_calc_md5_hash failed"
1180 static void tcp_v4_init_req(struct request_sock *req,
1181 const struct sock *sk_listener,
1182 struct sk_buff *skb)
1184 struct inet_request_sock *ireq = inet_rsk(req);
1186 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1187 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1188 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
1189 ireq->opt = tcp_v4_save_options(skb);
1192 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1194 const struct request_sock *req,
1197 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1200 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1209 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1211 .obj_size = sizeof(struct tcp_request_sock),
1212 .rtx_syn_ack = tcp_rtx_synack,
1213 .send_ack = tcp_v4_reqsk_send_ack,
1214 .destructor = tcp_v4_reqsk_destructor,
1215 .send_reset = tcp_v4_send_reset,
1216 .syn_ack_timeout = tcp_syn_ack_timeout,
1219 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1220 .mss_clamp = TCP_MSS_DEFAULT,
1221 #ifdef CONFIG_TCP_MD5SIG
1222 .req_md5_lookup = tcp_v4_md5_lookup,
1223 .calc_md5_hash = tcp_v4_md5_hash_skb,
1225 .init_req = tcp_v4_init_req,
1226 #ifdef CONFIG_SYN_COOKIES
1227 .cookie_init_seq = cookie_v4_init_sequence,
1229 .route_req = tcp_v4_route_req,
1230 .init_seq = tcp_v4_init_sequence,
1231 .send_synack = tcp_v4_send_synack,
1234 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1236 /* Never answer to SYNs send to broadcast or multicast */
1237 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1240 return tcp_conn_request(&tcp_request_sock_ops,
1241 &tcp_request_sock_ipv4_ops, sk, skb);
1244 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1247 EXPORT_SYMBOL(tcp_v4_conn_request);
1251 * The three way handshake has completed - we got a valid synack -
1252 * now create the new socket.
1254 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1255 struct request_sock *req,
1256 struct dst_entry *dst,
1257 struct request_sock *req_unhash,
1260 struct inet_request_sock *ireq;
1261 struct inet_sock *newinet;
1262 struct tcp_sock *newtp;
1264 #ifdef CONFIG_TCP_MD5SIG
1265 struct tcp_md5sig_key *key;
1267 struct ip_options_rcu *inet_opt;
1269 if (sk_acceptq_is_full(sk))
1272 newsk = tcp_create_openreq_child(sk, req, skb);
1276 newsk->sk_gso_type = SKB_GSO_TCPV4;
1277 inet_sk_rx_dst_set(newsk, skb);
1279 newtp = tcp_sk(newsk);
1280 newinet = inet_sk(newsk);
1281 ireq = inet_rsk(req);
1282 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1283 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1284 newsk->sk_bound_dev_if = ireq->ir_iif;
1285 newinet->inet_saddr = ireq->ir_loc_addr;
1286 inet_opt = ireq->opt;
1287 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1289 newinet->mc_index = inet_iif(skb);
1290 newinet->mc_ttl = ip_hdr(skb)->ttl;
1291 newinet->rcv_tos = ip_hdr(skb)->tos;
1292 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1294 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1295 newinet->inet_id = newtp->write_seq ^ jiffies;
1298 dst = inet_csk_route_child_sock(sk, newsk, req);
1302 /* syncookie case : see end of cookie_v4_check() */
1304 sk_setup_caps(newsk, dst);
1306 tcp_ca_openreq_child(newsk, dst);
1308 tcp_sync_mss(newsk, dst_mtu(dst));
1309 newtp->advmss = dst_metric_advmss(dst);
1310 if (tcp_sk(sk)->rx_opt.user_mss &&
1311 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1312 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1314 tcp_initialize_rcv_mss(newsk);
1316 #ifdef CONFIG_TCP_MD5SIG
1317 /* Copy over the MD5 key from the original socket */
1318 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1322 * We're using one, so create a matching key
1323 * on the newsk structure. If we fail to get
1324 * memory, then we end up not copying the key
1327 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1328 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1329 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1333 if (__inet_inherit_port(sk, newsk) < 0)
1335 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1337 tcp_move_syn(newtp, req);
1342 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1346 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1349 inet_csk_prepare_forced_close(newsk);
1353 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1355 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1357 #ifdef CONFIG_SYN_COOKIES
1358 const struct tcphdr *th = tcp_hdr(skb);
1361 sk = cookie_v4_check(sk, skb);
1366 /* The socket must have it's spinlock held when we get
1367 * here, unless it is a TCP_LISTEN socket.
1369 * We have a potential double-lock case here, so even when
1370 * doing backlog processing we use the BH locking scheme.
1371 * This is because we cannot sleep with the original spinlock
1374 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1378 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1379 struct dst_entry *dst = sk->sk_rx_dst;
1381 sock_rps_save_rxhash(sk, skb);
1382 sk_mark_napi_id(sk, skb);
1384 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1385 !dst->ops->check(dst, 0)) {
1387 sk->sk_rx_dst = NULL;
1390 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1394 if (tcp_checksum_complete(skb))
1397 if (sk->sk_state == TCP_LISTEN) {
1398 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1403 sock_rps_save_rxhash(nsk, skb);
1404 sk_mark_napi_id(nsk, skb);
1405 if (tcp_child_process(sk, nsk, skb)) {
1412 sock_rps_save_rxhash(sk, skb);
1414 if (tcp_rcv_state_process(sk, skb)) {
1421 tcp_v4_send_reset(rsk, skb);
1424 /* Be careful here. If this function gets more complicated and
1425 * gcc suffers from register pressure on the x86, sk (in %ebx)
1426 * might be destroyed here. This current version compiles correctly,
1427 * but you have been warned.
1432 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1433 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1436 EXPORT_SYMBOL(tcp_v4_do_rcv);
1438 void tcp_v4_early_demux(struct sk_buff *skb)
1440 const struct iphdr *iph;
1441 const struct tcphdr *th;
1444 if (skb->pkt_type != PACKET_HOST)
1447 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1453 if (th->doff < sizeof(struct tcphdr) / 4)
1456 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1457 iph->saddr, th->source,
1458 iph->daddr, ntohs(th->dest),
1462 skb->destructor = sock_edemux;
1463 if (sk_fullsock(sk)) {
1464 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1467 dst = dst_check(dst, 0);
1469 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1470 skb_dst_set_noref(skb, dst);
1475 /* Packet is added to VJ-style prequeue for processing in process
1476 * context, if a reader task is waiting. Apparently, this exciting
1477 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1478 * failed somewhere. Latency? Burstiness? Well, at least now we will
1479 * see, why it failed. 8)8) --ANK
1482 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1484 struct tcp_sock *tp = tcp_sk(sk);
1486 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1489 if (skb->len <= tcp_hdrlen(skb) &&
1490 skb_queue_len(&tp->ucopy.prequeue) == 0)
1493 /* Before escaping RCU protected region, we need to take care of skb
1494 * dst. Prequeue is only enabled for established sockets.
1495 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1496 * Instead of doing full sk_rx_dst validity here, let's perform
1497 * an optimistic check.
1499 if (likely(sk->sk_rx_dst))
1502 skb_dst_force_safe(skb);
1504 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1505 tp->ucopy.memory += skb->truesize;
1506 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1507 struct sk_buff *skb1;
1509 BUG_ON(sock_owned_by_user(sk));
1511 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1512 sk_backlog_rcv(sk, skb1);
1513 NET_INC_STATS_BH(sock_net(sk),
1514 LINUX_MIB_TCPPREQUEUEDROPPED);
1517 tp->ucopy.memory = 0;
1518 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1519 wake_up_interruptible_sync_poll(sk_sleep(sk),
1520 POLLIN | POLLRDNORM | POLLRDBAND);
1521 if (!inet_csk_ack_scheduled(sk))
1522 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1523 (3 * tcp_rto_min(sk)) / 4,
1528 EXPORT_SYMBOL(tcp_prequeue);
1534 int tcp_v4_rcv(struct sk_buff *skb)
1536 const struct iphdr *iph;
1537 const struct tcphdr *th;
1540 struct net *net = dev_net(skb->dev);
1542 if (skb->pkt_type != PACKET_HOST)
1545 /* Count it even if it's bad */
1546 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1548 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1553 if (th->doff < sizeof(struct tcphdr) / 4)
1555 if (!pskb_may_pull(skb, th->doff * 4))
1558 /* An explanation is required here, I think.
1559 * Packet length and doff are validated by header prediction,
1560 * provided case of th->doff==0 is eliminated.
1561 * So, we defer the checks. */
1563 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1568 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1569 * barrier() makes sure compiler wont play fool^Waliasing games.
1571 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1572 sizeof(struct inet_skb_parm));
1575 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1576 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1577 skb->len - th->doff * 4);
1578 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1579 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1580 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1581 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1582 TCP_SKB_CB(skb)->sacked = 0;
1585 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1590 if (sk->sk_state == TCP_TIME_WAIT)
1593 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1594 struct request_sock *req = inet_reqsk(sk);
1595 struct sock *nsk = NULL;
1597 sk = req->rsk_listener;
1598 if (tcp_v4_inbound_md5_hash(sk, skb))
1599 goto discard_and_relse;
1600 if (likely(sk->sk_state == TCP_LISTEN)) {
1601 nsk = tcp_check_req(sk, skb, req, false);
1603 inet_csk_reqsk_queue_drop_and_put(sk, req);
1613 } else if (tcp_child_process(sk, nsk, skb)) {
1614 tcp_v4_send_reset(nsk, skb);
1620 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1621 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1622 goto discard_and_relse;
1625 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1626 goto discard_and_relse;
1628 if (tcp_v4_inbound_md5_hash(sk, skb))
1629 goto discard_and_relse;
1633 if (sk_filter(sk, skb))
1634 goto discard_and_relse;
1638 if (sk->sk_state == TCP_LISTEN) {
1639 ret = tcp_v4_do_rcv(sk, skb);
1640 goto put_and_return;
1643 sk_incoming_cpu_update(sk);
1645 bh_lock_sock_nested(sk);
1646 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1648 if (!sock_owned_by_user(sk)) {
1649 if (!tcp_prequeue(sk, skb))
1650 ret = tcp_v4_do_rcv(sk, skb);
1651 } else if (unlikely(sk_add_backlog(sk, skb,
1652 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1654 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1655 goto discard_and_relse;
1665 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1668 if (tcp_checksum_complete(skb)) {
1670 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1672 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1674 tcp_v4_send_reset(NULL, skb);
1678 /* Discard frame. */
1687 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1688 inet_twsk_put(inet_twsk(sk));
1692 if (tcp_checksum_complete(skb)) {
1693 inet_twsk_put(inet_twsk(sk));
1696 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1698 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1700 iph->saddr, th->source,
1701 iph->daddr, th->dest,
1704 inet_twsk_deschedule_put(inet_twsk(sk));
1708 /* Fall through to ACK */
1711 tcp_v4_timewait_ack(sk, skb);
1714 tcp_v4_send_reset(sk, skb);
1715 inet_twsk_deschedule_put(inet_twsk(sk));
1717 case TCP_TW_SUCCESS:;
1722 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1723 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1724 .twsk_unique = tcp_twsk_unique,
1725 .twsk_destructor= tcp_twsk_destructor,
1728 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1730 struct dst_entry *dst = skb_dst(skb);
1732 if (dst && dst_hold_safe(dst)) {
1733 sk->sk_rx_dst = dst;
1734 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1737 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1739 const struct inet_connection_sock_af_ops ipv4_specific = {
1740 .queue_xmit = ip_queue_xmit,
1741 .send_check = tcp_v4_send_check,
1742 .rebuild_header = inet_sk_rebuild_header,
1743 .sk_rx_dst_set = inet_sk_rx_dst_set,
1744 .conn_request = tcp_v4_conn_request,
1745 .syn_recv_sock = tcp_v4_syn_recv_sock,
1746 .net_header_len = sizeof(struct iphdr),
1747 .setsockopt = ip_setsockopt,
1748 .getsockopt = ip_getsockopt,
1749 .addr2sockaddr = inet_csk_addr2sockaddr,
1750 .sockaddr_len = sizeof(struct sockaddr_in),
1751 .bind_conflict = inet_csk_bind_conflict,
1752 #ifdef CONFIG_COMPAT
1753 .compat_setsockopt = compat_ip_setsockopt,
1754 .compat_getsockopt = compat_ip_getsockopt,
1756 .mtu_reduced = tcp_v4_mtu_reduced,
1758 EXPORT_SYMBOL(ipv4_specific);
1760 #ifdef CONFIG_TCP_MD5SIG
1761 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1762 .md5_lookup = tcp_v4_md5_lookup,
1763 .calc_md5_hash = tcp_v4_md5_hash_skb,
1764 .md5_parse = tcp_v4_parse_md5_keys,
1768 /* NOTE: A lot of things set to zero explicitly by call to
1769 * sk_alloc() so need not be done here.
1771 static int tcp_v4_init_sock(struct sock *sk)
1773 struct inet_connection_sock *icsk = inet_csk(sk);
1777 icsk->icsk_af_ops = &ipv4_specific;
1779 #ifdef CONFIG_TCP_MD5SIG
1780 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1786 void tcp_v4_destroy_sock(struct sock *sk)
1788 struct tcp_sock *tp = tcp_sk(sk);
1790 tcp_clear_xmit_timers(sk);
1792 tcp_cleanup_congestion_control(sk);
1794 /* Cleanup up the write buffer. */
1795 tcp_write_queue_purge(sk);
1797 /* Cleans up our, hopefully empty, out_of_order_queue. */
1798 __skb_queue_purge(&tp->out_of_order_queue);
1800 #ifdef CONFIG_TCP_MD5SIG
1801 /* Clean up the MD5 key list, if any */
1802 if (tp->md5sig_info) {
1803 tcp_clear_md5_list(sk);
1804 kfree_rcu(tp->md5sig_info, rcu);
1805 tp->md5sig_info = NULL;
1809 /* Clean prequeue, it must be empty really */
1810 __skb_queue_purge(&tp->ucopy.prequeue);
1812 /* Clean up a referenced TCP bind bucket. */
1813 if (inet_csk(sk)->icsk_bind_hash)
1816 BUG_ON(tp->fastopen_rsk);
1818 /* If socket is aborted during connect operation */
1819 tcp_free_fastopen_req(tp);
1820 tcp_saved_syn_free(tp);
1822 sk_sockets_allocated_dec(sk);
1824 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1825 sock_release_memcg(sk);
1827 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1829 #ifdef CONFIG_PROC_FS
1830 /* Proc filesystem TCP sock list dumping. */
1833 * Get next listener socket follow cur. If cur is NULL, get first socket
1834 * starting from bucket given in st->bucket; when st->bucket is zero the
1835 * very first socket in the hash table is returned.
1837 static void *listening_get_next(struct seq_file *seq, void *cur)
1839 struct inet_connection_sock *icsk;
1840 struct hlist_nulls_node *node;
1841 struct sock *sk = cur;
1842 struct inet_listen_hashbucket *ilb;
1843 struct tcp_iter_state *st = seq->private;
1844 struct net *net = seq_file_net(seq);
1847 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1848 spin_lock_bh(&ilb->lock);
1849 sk = sk_nulls_head(&ilb->head);
1853 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1857 sk = sk_nulls_next(sk);
1859 sk_nulls_for_each_from(sk, node) {
1860 if (!net_eq(sock_net(sk), net))
1862 if (sk->sk_family == st->family) {
1866 icsk = inet_csk(sk);
1868 spin_unlock_bh(&ilb->lock);
1870 if (++st->bucket < INET_LHTABLE_SIZE) {
1871 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1872 spin_lock_bh(&ilb->lock);
1873 sk = sk_nulls_head(&ilb->head);
1881 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1883 struct tcp_iter_state *st = seq->private;
1888 rc = listening_get_next(seq, NULL);
1890 while (rc && *pos) {
1891 rc = listening_get_next(seq, rc);
1897 static inline bool empty_bucket(const struct tcp_iter_state *st)
1899 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1903 * Get first established socket starting from bucket given in st->bucket.
1904 * If st->bucket is zero, the very first socket in the hash is returned.
1906 static void *established_get_first(struct seq_file *seq)
1908 struct tcp_iter_state *st = seq->private;
1909 struct net *net = seq_file_net(seq);
1913 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1915 struct hlist_nulls_node *node;
1916 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1918 /* Lockless fast path for the common case of empty buckets */
1919 if (empty_bucket(st))
1923 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1924 if (sk->sk_family != st->family ||
1925 !net_eq(sock_net(sk), net)) {
1931 spin_unlock_bh(lock);
1937 static void *established_get_next(struct seq_file *seq, void *cur)
1939 struct sock *sk = cur;
1940 struct hlist_nulls_node *node;
1941 struct tcp_iter_state *st = seq->private;
1942 struct net *net = seq_file_net(seq);
1947 sk = sk_nulls_next(sk);
1949 sk_nulls_for_each_from(sk, node) {
1950 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
1954 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1956 return established_get_first(seq);
1959 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1961 struct tcp_iter_state *st = seq->private;
1965 rc = established_get_first(seq);
1968 rc = established_get_next(seq, rc);
1974 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1977 struct tcp_iter_state *st = seq->private;
1979 st->state = TCP_SEQ_STATE_LISTENING;
1980 rc = listening_get_idx(seq, &pos);
1983 st->state = TCP_SEQ_STATE_ESTABLISHED;
1984 rc = established_get_idx(seq, pos);
1990 static void *tcp_seek_last_pos(struct seq_file *seq)
1992 struct tcp_iter_state *st = seq->private;
1993 int offset = st->offset;
1994 int orig_num = st->num;
1997 switch (st->state) {
1998 case TCP_SEQ_STATE_LISTENING:
1999 if (st->bucket >= INET_LHTABLE_SIZE)
2001 st->state = TCP_SEQ_STATE_LISTENING;
2002 rc = listening_get_next(seq, NULL);
2003 while (offset-- && rc)
2004 rc = listening_get_next(seq, rc);
2008 st->state = TCP_SEQ_STATE_ESTABLISHED;
2010 case TCP_SEQ_STATE_ESTABLISHED:
2011 if (st->bucket > tcp_hashinfo.ehash_mask)
2013 rc = established_get_first(seq);
2014 while (offset-- && rc)
2015 rc = established_get_next(seq, rc);
2023 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2025 struct tcp_iter_state *st = seq->private;
2028 if (*pos && *pos == st->last_pos) {
2029 rc = tcp_seek_last_pos(seq);
2034 st->state = TCP_SEQ_STATE_LISTENING;
2038 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2041 st->last_pos = *pos;
2045 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2047 struct tcp_iter_state *st = seq->private;
2050 if (v == SEQ_START_TOKEN) {
2051 rc = tcp_get_idx(seq, 0);
2055 switch (st->state) {
2056 case TCP_SEQ_STATE_LISTENING:
2057 rc = listening_get_next(seq, v);
2059 st->state = TCP_SEQ_STATE_ESTABLISHED;
2062 rc = established_get_first(seq);
2065 case TCP_SEQ_STATE_ESTABLISHED:
2066 rc = established_get_next(seq, v);
2071 st->last_pos = *pos;
2075 static void tcp_seq_stop(struct seq_file *seq, void *v)
2077 struct tcp_iter_state *st = seq->private;
2079 switch (st->state) {
2080 case TCP_SEQ_STATE_LISTENING:
2081 if (v != SEQ_START_TOKEN)
2082 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2084 case TCP_SEQ_STATE_ESTABLISHED:
2086 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2091 int tcp_seq_open(struct inode *inode, struct file *file)
2093 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2094 struct tcp_iter_state *s;
2097 err = seq_open_net(inode, file, &afinfo->seq_ops,
2098 sizeof(struct tcp_iter_state));
2102 s = ((struct seq_file *)file->private_data)->private;
2103 s->family = afinfo->family;
2107 EXPORT_SYMBOL(tcp_seq_open);
2109 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2112 struct proc_dir_entry *p;
2114 afinfo->seq_ops.start = tcp_seq_start;
2115 afinfo->seq_ops.next = tcp_seq_next;
2116 afinfo->seq_ops.stop = tcp_seq_stop;
2118 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2119 afinfo->seq_fops, afinfo);
2124 EXPORT_SYMBOL(tcp_proc_register);
2126 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2128 remove_proc_entry(afinfo->name, net->proc_net);
2130 EXPORT_SYMBOL(tcp_proc_unregister);
2132 static void get_openreq4(const struct request_sock *req,
2133 struct seq_file *f, int i)
2135 const struct inet_request_sock *ireq = inet_rsk(req);
2136 long delta = req->rsk_timer.expires - jiffies;
2138 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2139 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2144 ntohs(ireq->ir_rmt_port),
2146 0, 0, /* could print option size, but that is af dependent. */
2147 1, /* timers active (only the expire timer) */
2148 jiffies_delta_to_clock_t(delta),
2150 from_kuid_munged(seq_user_ns(f),
2151 sock_i_uid(req->rsk_listener)),
2152 0, /* non standard timer */
2153 0, /* open_requests have no inode */
2158 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2161 unsigned long timer_expires;
2162 const struct tcp_sock *tp = tcp_sk(sk);
2163 const struct inet_connection_sock *icsk = inet_csk(sk);
2164 const struct inet_sock *inet = inet_sk(sk);
2165 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2166 __be32 dest = inet->inet_daddr;
2167 __be32 src = inet->inet_rcv_saddr;
2168 __u16 destp = ntohs(inet->inet_dport);
2169 __u16 srcp = ntohs(inet->inet_sport);
2173 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2174 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2175 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2177 timer_expires = icsk->icsk_timeout;
2178 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2180 timer_expires = icsk->icsk_timeout;
2181 } else if (timer_pending(&sk->sk_timer)) {
2183 timer_expires = sk->sk_timer.expires;
2186 timer_expires = jiffies;
2189 state = sk_state_load(sk);
2190 if (state == TCP_LISTEN)
2191 rx_queue = sk->sk_ack_backlog;
2193 /* Because we don't lock the socket,
2194 * we might find a transient negative value.
2196 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2198 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2199 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2200 i, src, srcp, dest, destp, state,
2201 tp->write_seq - tp->snd_una,
2204 jiffies_delta_to_clock_t(timer_expires - jiffies),
2205 icsk->icsk_retransmits,
2206 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2207 icsk->icsk_probes_out,
2209 atomic_read(&sk->sk_refcnt), sk,
2210 jiffies_to_clock_t(icsk->icsk_rto),
2211 jiffies_to_clock_t(icsk->icsk_ack.ato),
2212 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2214 state == TCP_LISTEN ?
2215 fastopenq->max_qlen :
2216 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2219 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2220 struct seq_file *f, int i)
2222 long delta = tw->tw_timer.expires - jiffies;
2226 dest = tw->tw_daddr;
2227 src = tw->tw_rcv_saddr;
2228 destp = ntohs(tw->tw_dport);
2229 srcp = ntohs(tw->tw_sport);
2231 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2232 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2233 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2234 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2235 atomic_read(&tw->tw_refcnt), tw);
2240 static int tcp4_seq_show(struct seq_file *seq, void *v)
2242 struct tcp_iter_state *st;
2243 struct sock *sk = v;
2245 seq_setwidth(seq, TMPSZ - 1);
2246 if (v == SEQ_START_TOKEN) {
2247 seq_puts(seq, " sl local_address rem_address st tx_queue "
2248 "rx_queue tr tm->when retrnsmt uid timeout "
2254 if (sk->sk_state == TCP_TIME_WAIT)
2255 get_timewait4_sock(v, seq, st->num);
2256 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2257 get_openreq4(v, seq, st->num);
2259 get_tcp4_sock(v, seq, st->num);
2265 static const struct file_operations tcp_afinfo_seq_fops = {
2266 .owner = THIS_MODULE,
2267 .open = tcp_seq_open,
2269 .llseek = seq_lseek,
2270 .release = seq_release_net
2273 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2276 .seq_fops = &tcp_afinfo_seq_fops,
2278 .show = tcp4_seq_show,
2282 static int __net_init tcp4_proc_init_net(struct net *net)
2284 return tcp_proc_register(net, &tcp4_seq_afinfo);
2287 static void __net_exit tcp4_proc_exit_net(struct net *net)
2289 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2292 static struct pernet_operations tcp4_net_ops = {
2293 .init = tcp4_proc_init_net,
2294 .exit = tcp4_proc_exit_net,
2297 int __init tcp4_proc_init(void)
2299 return register_pernet_subsys(&tcp4_net_ops);
2302 void tcp4_proc_exit(void)
2304 unregister_pernet_subsys(&tcp4_net_ops);
2306 #endif /* CONFIG_PROC_FS */
2308 struct proto tcp_prot = {
2310 .owner = THIS_MODULE,
2312 .connect = tcp_v4_connect,
2313 .disconnect = tcp_disconnect,
2314 .accept = inet_csk_accept,
2316 .init = tcp_v4_init_sock,
2317 .destroy = tcp_v4_destroy_sock,
2318 .shutdown = tcp_shutdown,
2319 .setsockopt = tcp_setsockopt,
2320 .getsockopt = tcp_getsockopt,
2321 .recvmsg = tcp_recvmsg,
2322 .sendmsg = tcp_sendmsg,
2323 .sendpage = tcp_sendpage,
2324 .backlog_rcv = tcp_v4_do_rcv,
2325 .release_cb = tcp_release_cb,
2327 .unhash = inet_unhash,
2328 .get_port = inet_csk_get_port,
2329 .enter_memory_pressure = tcp_enter_memory_pressure,
2330 .stream_memory_free = tcp_stream_memory_free,
2331 .sockets_allocated = &tcp_sockets_allocated,
2332 .orphan_count = &tcp_orphan_count,
2333 .memory_allocated = &tcp_memory_allocated,
2334 .memory_pressure = &tcp_memory_pressure,
2335 .sysctl_mem = sysctl_tcp_mem,
2336 .sysctl_wmem = sysctl_tcp_wmem,
2337 .sysctl_rmem = sysctl_tcp_rmem,
2338 .max_header = MAX_TCP_HEADER,
2339 .obj_size = sizeof(struct tcp_sock),
2340 .slab_flags = SLAB_DESTROY_BY_RCU,
2341 .twsk_prot = &tcp_timewait_sock_ops,
2342 .rsk_prot = &tcp_request_sock_ops,
2343 .h.hashinfo = &tcp_hashinfo,
2344 .no_autobind = true,
2345 #ifdef CONFIG_COMPAT
2346 .compat_setsockopt = compat_tcp_setsockopt,
2347 .compat_getsockopt = compat_tcp_getsockopt,
2349 .diag_destroy = tcp_abort,
2351 EXPORT_SYMBOL(tcp_prot);
2353 static void __net_exit tcp_sk_exit(struct net *net)
2357 for_each_possible_cpu(cpu)
2358 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2359 free_percpu(net->ipv4.tcp_sk);
2362 static int __net_init tcp_sk_init(struct net *net)
2366 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2367 if (!net->ipv4.tcp_sk)
2370 for_each_possible_cpu(cpu) {
2373 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2377 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2380 net->ipv4.sysctl_tcp_ecn = 2;
2381 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2383 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2384 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2385 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2387 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2388 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2389 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2398 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2400 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2403 static struct pernet_operations __net_initdata tcp_sk_ops = {
2404 .init = tcp_sk_init,
2405 .exit = tcp_sk_exit,
2406 .exit_batch = tcp_sk_exit_batch,
2409 void __init tcp_v4_init(void)
2411 inet_hashinfo_init(&tcp_hashinfo);
2412 if (register_pernet_subsys(&tcp_sk_ops))
2413 panic("Failed to create the TCP control socket.\n");