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 <crypto/hash.h>
85 #include <linux/scatterlist.h>
87 int sysctl_tcp_low_latency __read_mostly;
89 #ifdef CONFIG_TCP_MD5SIG
90 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
91 __be32 daddr, __be32 saddr, const struct tcphdr *th);
94 struct inet_hashinfo tcp_hashinfo;
95 EXPORT_SYMBOL(tcp_hashinfo);
97 static u32 tcp_v4_init_seq_and_tsoff(const struct sk_buff *skb, u32 *tsoff)
99 return secure_tcp_seq_and_tsoff(ip_hdr(skb)->daddr,
102 tcp_hdr(skb)->source, tsoff);
105 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
107 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
108 struct tcp_sock *tp = tcp_sk(sk);
110 /* With PAWS, it is safe from the viewpoint
111 of data integrity. Even without PAWS it is safe provided sequence
112 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
114 Actually, the idea is close to VJ's one, only timestamp cache is
115 held not per host, but per port pair and TW bucket is used as state
118 If TW bucket has been already destroyed we fall back to VJ's scheme
119 and use initial timestamp retrieved from peer table.
121 if (tcptw->tw_ts_recent_stamp &&
122 (!twp || (sock_net(sk)->ipv4.sysctl_tcp_tw_reuse &&
123 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
124 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
125 if (tp->write_seq == 0)
127 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
128 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
135 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
137 /* This will initiate an outgoing connection. */
138 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
140 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
141 struct inet_sock *inet = inet_sk(sk);
142 struct tcp_sock *tp = tcp_sk(sk);
143 __be16 orig_sport, orig_dport;
144 __be32 daddr, nexthop;
149 struct ip_options_rcu *inet_opt;
150 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
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 lockdep_sock_is_held(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 inet->inet_dport = usin->sin_port;
202 sk_daddr_set(sk, daddr);
204 inet_csk(sk)->icsk_ext_hdr_len = 0;
206 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
208 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
210 /* Socket identity is still unknown (sport may be zero).
211 * However we set state to SYN-SENT and not releasing socket
212 * lock select source port, enter ourselves into the hash tables and
213 * complete initialization after this.
215 tcp_set_state(sk, TCP_SYN_SENT);
216 err = inet_hash_connect(tcp_death_row, sk);
222 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
223 inet->inet_sport, inet->inet_dport, sk);
229 /* OK, now commit destination to socket. */
230 sk->sk_gso_type = SKB_GSO_TCPV4;
231 sk_setup_caps(sk, &rt->dst);
234 if (likely(!tp->repair)) {
235 seq = secure_tcp_seq_and_tsoff(inet->inet_saddr,
244 inet->inet_id = tp->write_seq ^ jiffies;
246 if (tcp_fastopen_defer_connect(sk, &err))
251 err = tcp_connect(sk);
260 * This unhashes the socket and releases the local port,
263 tcp_set_state(sk, TCP_CLOSE);
265 sk->sk_route_caps = 0;
266 inet->inet_dport = 0;
269 EXPORT_SYMBOL(tcp_v4_connect);
272 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
273 * It can be called through tcp_release_cb() if socket was owned by user
274 * at the time tcp_v4_err() was called to handle ICMP message.
276 void tcp_v4_mtu_reduced(struct sock *sk)
278 struct inet_sock *inet = inet_sk(sk);
279 struct dst_entry *dst;
282 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
284 mtu = tcp_sk(sk)->mtu_info;
285 dst = inet_csk_update_pmtu(sk, mtu);
289 /* Something is about to be wrong... Remember soft error
290 * for the case, if this connection will not able to recover.
292 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
293 sk->sk_err_soft = EMSGSIZE;
297 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
298 ip_sk_accept_pmtu(sk) &&
299 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
300 tcp_sync_mss(sk, mtu);
302 /* Resend the TCP packet because it's
303 * clear that the old packet has been
304 * dropped. This is the new "fast" path mtu
307 tcp_simple_retransmit(sk);
308 } /* else let the usual retransmit timer handle it */
310 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
312 static void do_redirect(struct sk_buff *skb, struct sock *sk)
314 struct dst_entry *dst = __sk_dst_check(sk, 0);
317 dst->ops->redirect(dst, sk, skb);
321 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
322 void tcp_req_err(struct sock *sk, u32 seq, bool abort)
324 struct request_sock *req = inet_reqsk(sk);
325 struct net *net = sock_net(sk);
327 /* ICMPs are not backlogged, hence we cannot get
328 * an established socket here.
330 if (seq != tcp_rsk(req)->snt_isn) {
331 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
334 * Still in SYN_RECV, just remove it silently.
335 * There is no good way to pass the error to the newly
336 * created socket, and POSIX does not want network
337 * errors returned from accept().
339 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
340 tcp_listendrop(req->rsk_listener);
344 EXPORT_SYMBOL(tcp_req_err);
347 * This routine is called by the ICMP module when it gets some
348 * sort of error condition. If err < 0 then the socket should
349 * be closed and the error returned to the user. If err > 0
350 * it's just the icmp type << 8 | icmp code. After adjustment
351 * header points to the first 8 bytes of the tcp header. We need
352 * to find the appropriate port.
354 * The locking strategy used here is very "optimistic". When
355 * someone else accesses the socket the ICMP is just dropped
356 * and for some paths there is no check at all.
357 * A more general error queue to queue errors for later handling
358 * is probably better.
362 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
364 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
365 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
366 struct inet_connection_sock *icsk;
368 struct inet_sock *inet;
369 const int type = icmp_hdr(icmp_skb)->type;
370 const int code = icmp_hdr(icmp_skb)->code;
373 struct request_sock *fastopen;
377 struct net *net = dev_net(icmp_skb->dev);
379 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
380 th->dest, iph->saddr, ntohs(th->source),
383 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
386 if (sk->sk_state == TCP_TIME_WAIT) {
387 inet_twsk_put(inet_twsk(sk));
390 seq = ntohl(th->seq);
391 if (sk->sk_state == TCP_NEW_SYN_RECV)
392 return tcp_req_err(sk, seq,
393 type == ICMP_PARAMETERPROB ||
394 type == ICMP_TIME_EXCEEDED ||
395 (type == ICMP_DEST_UNREACH &&
396 (code == ICMP_NET_UNREACH ||
397 code == ICMP_HOST_UNREACH)));
400 /* If too many ICMPs get dropped on busy
401 * servers this needs to be solved differently.
402 * We do take care of PMTU discovery (RFC1191) special case :
403 * we can receive locally generated ICMP messages while socket is held.
405 if (sock_owned_by_user(sk)) {
406 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
407 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
409 if (sk->sk_state == TCP_CLOSE)
412 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
413 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
419 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
420 fastopen = tp->fastopen_rsk;
421 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
422 if (sk->sk_state != TCP_LISTEN &&
423 !between(seq, snd_una, tp->snd_nxt)) {
424 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
430 if (!sock_owned_by_user(sk))
431 do_redirect(icmp_skb, sk);
433 case ICMP_SOURCE_QUENCH:
434 /* Just silently ignore these. */
436 case ICMP_PARAMETERPROB:
439 case ICMP_DEST_UNREACH:
440 if (code > NR_ICMP_UNREACH)
443 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
444 /* We are not interested in TCP_LISTEN and open_requests
445 * (SYN-ACKs send out by Linux are always <576bytes so
446 * they should go through unfragmented).
448 if (sk->sk_state == TCP_LISTEN)
452 if (!sock_owned_by_user(sk)) {
453 tcp_v4_mtu_reduced(sk);
455 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
461 err = icmp_err_convert[code].errno;
462 /* check if icmp_skb allows revert of backoff
463 * (see draft-zimmermann-tcp-lcd) */
464 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
466 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
467 !icsk->icsk_backoff || fastopen)
470 if (sock_owned_by_user(sk))
473 icsk->icsk_backoff--;
474 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
476 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
478 skb = tcp_write_queue_head(sk);
481 remaining = icsk->icsk_rto -
483 tcp_time_stamp - tcp_skb_timestamp(skb));
486 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
487 remaining, TCP_RTO_MAX);
489 /* RTO revert clocked out retransmission.
490 * Will retransmit now */
491 tcp_retransmit_timer(sk);
495 case ICMP_TIME_EXCEEDED:
502 switch (sk->sk_state) {
505 /* Only in fast or simultaneous open. If a fast open socket is
506 * is already accepted it is treated as a connected one below.
508 if (fastopen && !fastopen->sk)
511 if (!sock_owned_by_user(sk)) {
514 sk->sk_error_report(sk);
518 sk->sk_err_soft = err;
523 /* If we've already connected we will keep trying
524 * until we time out, or the user gives up.
526 * rfc1122 4.2.3.9 allows to consider as hard errors
527 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
528 * but it is obsoleted by pmtu discovery).
530 * Note, that in modern internet, where routing is unreliable
531 * and in each dark corner broken firewalls sit, sending random
532 * errors ordered by their masters even this two messages finally lose
533 * their original sense (even Linux sends invalid PORT_UNREACHs)
535 * Now we are in compliance with RFCs.
540 if (!sock_owned_by_user(sk) && inet->recverr) {
542 sk->sk_error_report(sk);
543 } else { /* Only an error on timeout */
544 sk->sk_err_soft = err;
552 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
554 struct tcphdr *th = tcp_hdr(skb);
556 if (skb->ip_summed == CHECKSUM_PARTIAL) {
557 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
558 skb->csum_start = skb_transport_header(skb) - skb->head;
559 skb->csum_offset = offsetof(struct tcphdr, check);
561 th->check = tcp_v4_check(skb->len, saddr, daddr,
568 /* This routine computes an IPv4 TCP checksum. */
569 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
571 const struct inet_sock *inet = inet_sk(sk);
573 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
575 EXPORT_SYMBOL(tcp_v4_send_check);
578 * This routine will send an RST to the other tcp.
580 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
582 * Answer: if a packet caused RST, it is not for a socket
583 * existing in our system, if it is matched to a socket,
584 * it is just duplicate segment or bug in other side's TCP.
585 * So that we build reply only basing on parameters
586 * arrived with segment.
587 * Exception: precedence violation. We do not implement it in any case.
590 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
592 const struct tcphdr *th = tcp_hdr(skb);
595 #ifdef CONFIG_TCP_MD5SIG
596 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
599 struct ip_reply_arg arg;
600 #ifdef CONFIG_TCP_MD5SIG
601 struct tcp_md5sig_key *key = NULL;
602 const __u8 *hash_location = NULL;
603 unsigned char newhash[16];
605 struct sock *sk1 = NULL;
609 /* Never send a reset in response to a reset. */
613 /* If sk not NULL, it means we did a successful lookup and incoming
614 * route had to be correct. prequeue might have dropped our dst.
616 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
619 /* Swap the send and the receive. */
620 memset(&rep, 0, sizeof(rep));
621 rep.th.dest = th->source;
622 rep.th.source = th->dest;
623 rep.th.doff = sizeof(struct tcphdr) / 4;
627 rep.th.seq = th->ack_seq;
630 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
631 skb->len - (th->doff << 2));
634 memset(&arg, 0, sizeof(arg));
635 arg.iov[0].iov_base = (unsigned char *)&rep;
636 arg.iov[0].iov_len = sizeof(rep.th);
638 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
639 #ifdef CONFIG_TCP_MD5SIG
641 hash_location = tcp_parse_md5sig_option(th);
642 if (sk && sk_fullsock(sk)) {
643 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
644 &ip_hdr(skb)->saddr, AF_INET);
645 } else if (hash_location) {
647 * active side is lost. Try to find listening socket through
648 * source port, and then find md5 key through listening socket.
649 * we are not loose security here:
650 * Incoming packet is checked with md5 hash with finding key,
651 * no RST generated if md5 hash doesn't match.
653 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
655 th->source, ip_hdr(skb)->daddr,
656 ntohs(th->source), inet_iif(skb));
657 /* don't send rst if it can't find key */
661 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
662 &ip_hdr(skb)->saddr, AF_INET);
667 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
668 if (genhash || memcmp(hash_location, newhash, 16) != 0)
674 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
676 (TCPOPT_MD5SIG << 8) |
678 /* Update length and the length the header thinks exists */
679 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
680 rep.th.doff = arg.iov[0].iov_len / 4;
682 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
683 key, ip_hdr(skb)->saddr,
684 ip_hdr(skb)->daddr, &rep.th);
687 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
688 ip_hdr(skb)->saddr, /* XXX */
689 arg.iov[0].iov_len, IPPROTO_TCP, 0);
690 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
691 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
693 /* When socket is gone, all binding information is lost.
694 * routing might fail in this case. No choice here, if we choose to force
695 * input interface, we will misroute in case of asymmetric route.
698 arg.bound_dev_if = sk->sk_bound_dev_if;
700 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
701 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
703 arg.tos = ip_hdr(skb)->tos;
704 arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
706 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
707 skb, &TCP_SKB_CB(skb)->header.h4.opt,
708 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
709 &arg, arg.iov[0].iov_len);
711 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
712 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
715 #ifdef CONFIG_TCP_MD5SIG
721 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
722 outside socket context is ugly, certainly. What can I do?
725 static void tcp_v4_send_ack(const struct sock *sk,
726 struct sk_buff *skb, u32 seq, u32 ack,
727 u32 win, u32 tsval, u32 tsecr, int oif,
728 struct tcp_md5sig_key *key,
729 int reply_flags, u8 tos)
731 const struct tcphdr *th = tcp_hdr(skb);
734 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
735 #ifdef CONFIG_TCP_MD5SIG
736 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
740 struct net *net = sock_net(sk);
741 struct ip_reply_arg arg;
743 memset(&rep.th, 0, sizeof(struct tcphdr));
744 memset(&arg, 0, sizeof(arg));
746 arg.iov[0].iov_base = (unsigned char *)&rep;
747 arg.iov[0].iov_len = sizeof(rep.th);
749 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
750 (TCPOPT_TIMESTAMP << 8) |
752 rep.opt[1] = htonl(tsval);
753 rep.opt[2] = htonl(tsecr);
754 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
757 /* Swap the send and the receive. */
758 rep.th.dest = th->source;
759 rep.th.source = th->dest;
760 rep.th.doff = arg.iov[0].iov_len / 4;
761 rep.th.seq = htonl(seq);
762 rep.th.ack_seq = htonl(ack);
764 rep.th.window = htons(win);
766 #ifdef CONFIG_TCP_MD5SIG
768 int offset = (tsecr) ? 3 : 0;
770 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
772 (TCPOPT_MD5SIG << 8) |
774 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
775 rep.th.doff = arg.iov[0].iov_len/4;
777 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
778 key, ip_hdr(skb)->saddr,
779 ip_hdr(skb)->daddr, &rep.th);
782 arg.flags = reply_flags;
783 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
784 ip_hdr(skb)->saddr, /* XXX */
785 arg.iov[0].iov_len, IPPROTO_TCP, 0);
786 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
788 arg.bound_dev_if = oif;
790 arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
792 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
793 skb, &TCP_SKB_CB(skb)->header.h4.opt,
794 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
795 &arg, arg.iov[0].iov_len);
797 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
801 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
803 struct inet_timewait_sock *tw = inet_twsk(sk);
804 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
806 tcp_v4_send_ack(sk, skb,
807 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
808 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
809 tcp_time_stamp + tcptw->tw_ts_offset,
812 tcp_twsk_md5_key(tcptw),
813 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
820 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
821 struct request_sock *req)
823 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
824 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
826 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
830 * The window field (SEG.WND) of every outgoing segment, with the
831 * exception of <SYN> segments, MUST be right-shifted by
832 * Rcv.Wind.Shift bits:
834 tcp_v4_send_ack(sk, skb, seq,
835 tcp_rsk(req)->rcv_nxt,
836 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
837 tcp_time_stamp + tcp_rsk(req)->ts_off,
840 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
842 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
847 * Send a SYN-ACK after having received a SYN.
848 * This still operates on a request_sock only, not on a big
851 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
853 struct request_sock *req,
854 struct tcp_fastopen_cookie *foc,
855 enum tcp_synack_type synack_type)
857 const struct inet_request_sock *ireq = inet_rsk(req);
862 /* First, grab a route. */
863 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
866 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
869 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
871 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
874 err = net_xmit_eval(err);
881 * IPv4 request_sock destructor.
883 static void tcp_v4_reqsk_destructor(struct request_sock *req)
885 kfree(inet_rsk(req)->opt);
888 #ifdef CONFIG_TCP_MD5SIG
890 * RFC2385 MD5 checksumming requires a mapping of
891 * IP address->MD5 Key.
892 * We need to maintain these in the sk structure.
895 /* Find the Key structure for an address. */
896 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
897 const union tcp_md5_addr *addr,
900 const struct tcp_sock *tp = tcp_sk(sk);
901 struct tcp_md5sig_key *key;
902 unsigned int size = sizeof(struct in_addr);
903 const struct tcp_md5sig_info *md5sig;
905 /* caller either holds rcu_read_lock() or socket lock */
906 md5sig = rcu_dereference_check(tp->md5sig_info,
907 lockdep_sock_is_held(sk));
910 #if IS_ENABLED(CONFIG_IPV6)
911 if (family == AF_INET6)
912 size = sizeof(struct in6_addr);
914 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
915 if (key->family != family)
917 if (!memcmp(&key->addr, addr, size))
922 EXPORT_SYMBOL(tcp_md5_do_lookup);
924 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
925 const struct sock *addr_sk)
927 const union tcp_md5_addr *addr;
929 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
930 return tcp_md5_do_lookup(sk, addr, AF_INET);
932 EXPORT_SYMBOL(tcp_v4_md5_lookup);
934 /* This can be called on a newly created socket, from other files */
935 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
936 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
938 /* Add Key to the list */
939 struct tcp_md5sig_key *key;
940 struct tcp_sock *tp = tcp_sk(sk);
941 struct tcp_md5sig_info *md5sig;
943 key = tcp_md5_do_lookup(sk, addr, family);
945 /* Pre-existing entry - just update that one. */
946 memcpy(key->key, newkey, newkeylen);
947 key->keylen = newkeylen;
951 md5sig = rcu_dereference_protected(tp->md5sig_info,
952 lockdep_sock_is_held(sk));
954 md5sig = kmalloc(sizeof(*md5sig), gfp);
958 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
959 INIT_HLIST_HEAD(&md5sig->head);
960 rcu_assign_pointer(tp->md5sig_info, md5sig);
963 key = sock_kmalloc(sk, sizeof(*key), gfp);
966 if (!tcp_alloc_md5sig_pool()) {
967 sock_kfree_s(sk, key, sizeof(*key));
971 memcpy(key->key, newkey, newkeylen);
972 key->keylen = newkeylen;
973 key->family = family;
974 memcpy(&key->addr, addr,
975 (family == AF_INET6) ? sizeof(struct in6_addr) :
976 sizeof(struct in_addr));
977 hlist_add_head_rcu(&key->node, &md5sig->head);
980 EXPORT_SYMBOL(tcp_md5_do_add);
982 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
984 struct tcp_md5sig_key *key;
986 key = tcp_md5_do_lookup(sk, addr, family);
989 hlist_del_rcu(&key->node);
990 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
994 EXPORT_SYMBOL(tcp_md5_do_del);
996 static void tcp_clear_md5_list(struct sock *sk)
998 struct tcp_sock *tp = tcp_sk(sk);
999 struct tcp_md5sig_key *key;
1000 struct hlist_node *n;
1001 struct tcp_md5sig_info *md5sig;
1003 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1005 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1006 hlist_del_rcu(&key->node);
1007 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1008 kfree_rcu(key, rcu);
1012 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1015 struct tcp_md5sig cmd;
1016 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1018 if (optlen < sizeof(cmd))
1021 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1024 if (sin->sin_family != AF_INET)
1027 if (!cmd.tcpm_keylen)
1028 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1031 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1034 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1035 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1039 static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1040 __be32 daddr, __be32 saddr,
1041 const struct tcphdr *th, int nbytes)
1043 struct tcp4_pseudohdr *bp;
1044 struct scatterlist sg;
1051 bp->protocol = IPPROTO_TCP;
1052 bp->len = cpu_to_be16(nbytes);
1054 _th = (struct tcphdr *)(bp + 1);
1055 memcpy(_th, th, sizeof(*th));
1058 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1059 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1060 sizeof(*bp) + sizeof(*th));
1061 return crypto_ahash_update(hp->md5_req);
1064 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1065 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1067 struct tcp_md5sig_pool *hp;
1068 struct ahash_request *req;
1070 hp = tcp_get_md5sig_pool();
1072 goto clear_hash_noput;
1075 if (crypto_ahash_init(req))
1077 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1079 if (tcp_md5_hash_key(hp, key))
1081 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1082 if (crypto_ahash_final(req))
1085 tcp_put_md5sig_pool();
1089 tcp_put_md5sig_pool();
1091 memset(md5_hash, 0, 16);
1095 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1096 const struct sock *sk,
1097 const struct sk_buff *skb)
1099 struct tcp_md5sig_pool *hp;
1100 struct ahash_request *req;
1101 const struct tcphdr *th = tcp_hdr(skb);
1102 __be32 saddr, daddr;
1104 if (sk) { /* valid for establish/request sockets */
1105 saddr = sk->sk_rcv_saddr;
1106 daddr = sk->sk_daddr;
1108 const struct iphdr *iph = ip_hdr(skb);
1113 hp = tcp_get_md5sig_pool();
1115 goto clear_hash_noput;
1118 if (crypto_ahash_init(req))
1121 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1123 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1125 if (tcp_md5_hash_key(hp, key))
1127 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1128 if (crypto_ahash_final(req))
1131 tcp_put_md5sig_pool();
1135 tcp_put_md5sig_pool();
1137 memset(md5_hash, 0, 16);
1140 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1144 /* Called with rcu_read_lock() */
1145 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1146 const struct sk_buff *skb)
1148 #ifdef CONFIG_TCP_MD5SIG
1150 * This gets called for each TCP segment that arrives
1151 * so we want to be efficient.
1152 * We have 3 drop cases:
1153 * o No MD5 hash and one expected.
1154 * o MD5 hash and we're not expecting one.
1155 * o MD5 hash and its wrong.
1157 const __u8 *hash_location = NULL;
1158 struct tcp_md5sig_key *hash_expected;
1159 const struct iphdr *iph = ip_hdr(skb);
1160 const struct tcphdr *th = tcp_hdr(skb);
1162 unsigned char newhash[16];
1164 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1166 hash_location = tcp_parse_md5sig_option(th);
1168 /* We've parsed the options - do we have a hash? */
1169 if (!hash_expected && !hash_location)
1172 if (hash_expected && !hash_location) {
1173 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1177 if (!hash_expected && hash_location) {
1178 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1182 /* Okay, so this is hash_expected and hash_location -
1183 * so we need to calculate the checksum.
1185 genhash = tcp_v4_md5_hash_skb(newhash,
1189 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1190 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1191 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1192 &iph->saddr, ntohs(th->source),
1193 &iph->daddr, ntohs(th->dest),
1194 genhash ? " tcp_v4_calc_md5_hash failed"
1203 static void tcp_v4_init_req(struct request_sock *req,
1204 const struct sock *sk_listener,
1205 struct sk_buff *skb)
1207 struct inet_request_sock *ireq = inet_rsk(req);
1209 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1210 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1211 ireq->opt = tcp_v4_save_options(skb);
1214 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1216 const struct request_sock *req)
1218 return inet_csk_route_req(sk, &fl->u.ip4, req);
1221 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1223 .obj_size = sizeof(struct tcp_request_sock),
1224 .rtx_syn_ack = tcp_rtx_synack,
1225 .send_ack = tcp_v4_reqsk_send_ack,
1226 .destructor = tcp_v4_reqsk_destructor,
1227 .send_reset = tcp_v4_send_reset,
1228 .syn_ack_timeout = tcp_syn_ack_timeout,
1231 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1232 .mss_clamp = TCP_MSS_DEFAULT,
1233 #ifdef CONFIG_TCP_MD5SIG
1234 .req_md5_lookup = tcp_v4_md5_lookup,
1235 .calc_md5_hash = tcp_v4_md5_hash_skb,
1237 .init_req = tcp_v4_init_req,
1238 #ifdef CONFIG_SYN_COOKIES
1239 .cookie_init_seq = cookie_v4_init_sequence,
1241 .route_req = tcp_v4_route_req,
1242 .init_seq_tsoff = tcp_v4_init_seq_and_tsoff,
1243 .send_synack = tcp_v4_send_synack,
1246 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1248 /* Never answer to SYNs send to broadcast or multicast */
1249 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1252 return tcp_conn_request(&tcp_request_sock_ops,
1253 &tcp_request_sock_ipv4_ops, sk, skb);
1259 EXPORT_SYMBOL(tcp_v4_conn_request);
1263 * The three way handshake has completed - we got a valid synack -
1264 * now create the new socket.
1266 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1267 struct request_sock *req,
1268 struct dst_entry *dst,
1269 struct request_sock *req_unhash,
1272 struct inet_request_sock *ireq;
1273 struct inet_sock *newinet;
1274 struct tcp_sock *newtp;
1276 #ifdef CONFIG_TCP_MD5SIG
1277 struct tcp_md5sig_key *key;
1279 struct ip_options_rcu *inet_opt;
1281 if (sk_acceptq_is_full(sk))
1284 newsk = tcp_create_openreq_child(sk, req, skb);
1288 newsk->sk_gso_type = SKB_GSO_TCPV4;
1289 inet_sk_rx_dst_set(newsk, skb);
1291 newtp = tcp_sk(newsk);
1292 newinet = inet_sk(newsk);
1293 ireq = inet_rsk(req);
1294 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1295 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1296 newsk->sk_bound_dev_if = ireq->ir_iif;
1297 newinet->inet_saddr = ireq->ir_loc_addr;
1298 inet_opt = ireq->opt;
1299 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1301 newinet->mc_index = inet_iif(skb);
1302 newinet->mc_ttl = ip_hdr(skb)->ttl;
1303 newinet->rcv_tos = ip_hdr(skb)->tos;
1304 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1306 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1307 newinet->inet_id = newtp->write_seq ^ jiffies;
1310 dst = inet_csk_route_child_sock(sk, newsk, req);
1314 /* syncookie case : see end of cookie_v4_check() */
1316 sk_setup_caps(newsk, dst);
1318 tcp_ca_openreq_child(newsk, dst);
1320 tcp_sync_mss(newsk, dst_mtu(dst));
1321 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1323 tcp_initialize_rcv_mss(newsk);
1325 #ifdef CONFIG_TCP_MD5SIG
1326 /* Copy over the MD5 key from the original socket */
1327 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1331 * We're using one, so create a matching key
1332 * on the newsk structure. If we fail to get
1333 * memory, then we end up not copying the key
1336 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1337 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1338 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1342 if (__inet_inherit_port(sk, newsk) < 0)
1344 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1346 tcp_move_syn(newtp, req);
1351 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1358 inet_csk_prepare_forced_close(newsk);
1362 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1364 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1366 #ifdef CONFIG_SYN_COOKIES
1367 const struct tcphdr *th = tcp_hdr(skb);
1370 sk = cookie_v4_check(sk, skb);
1375 /* The socket must have it's spinlock held when we get
1376 * here, unless it is a TCP_LISTEN socket.
1378 * We have a potential double-lock case here, so even when
1379 * doing backlog processing we use the BH locking scheme.
1380 * This is because we cannot sleep with the original spinlock
1383 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1387 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1388 struct dst_entry *dst = sk->sk_rx_dst;
1390 sock_rps_save_rxhash(sk, skb);
1391 sk_mark_napi_id(sk, skb);
1393 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1394 !dst->ops->check(dst, 0)) {
1396 sk->sk_rx_dst = NULL;
1399 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1403 if (tcp_checksum_complete(skb))
1406 if (sk->sk_state == TCP_LISTEN) {
1407 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1412 sock_rps_save_rxhash(nsk, skb);
1413 sk_mark_napi_id(nsk, skb);
1414 if (tcp_child_process(sk, nsk, skb)) {
1421 sock_rps_save_rxhash(sk, skb);
1423 if (tcp_rcv_state_process(sk, skb)) {
1430 tcp_v4_send_reset(rsk, skb);
1433 /* Be careful here. If this function gets more complicated and
1434 * gcc suffers from register pressure on the x86, sk (in %ebx)
1435 * might be destroyed here. This current version compiles correctly,
1436 * but you have been warned.
1441 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1442 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1445 EXPORT_SYMBOL(tcp_v4_do_rcv);
1447 void tcp_v4_early_demux(struct sk_buff *skb)
1449 const struct iphdr *iph;
1450 const struct tcphdr *th;
1453 if (skb->pkt_type != PACKET_HOST)
1456 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1462 if (th->doff < sizeof(struct tcphdr) / 4)
1465 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1466 iph->saddr, th->source,
1467 iph->daddr, ntohs(th->dest),
1471 skb->destructor = sock_edemux;
1472 if (sk_fullsock(sk)) {
1473 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1476 dst = dst_check(dst, 0);
1478 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1479 skb_dst_set_noref(skb, dst);
1484 /* Packet is added to VJ-style prequeue for processing in process
1485 * context, if a reader task is waiting. Apparently, this exciting
1486 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1487 * failed somewhere. Latency? Burstiness? Well, at least now we will
1488 * see, why it failed. 8)8) --ANK
1491 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1493 struct tcp_sock *tp = tcp_sk(sk);
1495 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1498 if (skb->len <= tcp_hdrlen(skb) &&
1499 skb_queue_len(&tp->ucopy.prequeue) == 0)
1502 /* Before escaping RCU protected region, we need to take care of skb
1503 * dst. Prequeue is only enabled for established sockets.
1504 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1505 * Instead of doing full sk_rx_dst validity here, let's perform
1506 * an optimistic check.
1508 if (likely(sk->sk_rx_dst))
1511 skb_dst_force_safe(skb);
1513 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1514 tp->ucopy.memory += skb->truesize;
1515 if (skb_queue_len(&tp->ucopy.prequeue) >= 32 ||
1516 tp->ucopy.memory + atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) {
1517 struct sk_buff *skb1;
1519 BUG_ON(sock_owned_by_user(sk));
1520 __NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED,
1521 skb_queue_len(&tp->ucopy.prequeue));
1523 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
1524 sk_backlog_rcv(sk, skb1);
1526 tp->ucopy.memory = 0;
1527 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1528 wake_up_interruptible_sync_poll(sk_sleep(sk),
1529 POLLIN | POLLRDNORM | POLLRDBAND);
1530 if (!inet_csk_ack_scheduled(sk))
1531 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1532 (3 * tcp_rto_min(sk)) / 4,
1537 EXPORT_SYMBOL(tcp_prequeue);
1539 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1541 u32 limit = sk->sk_rcvbuf + sk->sk_sndbuf;
1543 /* Only socket owner can try to collapse/prune rx queues
1544 * to reduce memory overhead, so add a little headroom here.
1545 * Few sockets backlog are possibly concurrently non empty.
1549 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1550 * we can fix skb->truesize to its real value to avoid future drops.
1551 * This is valid because skb is not yet charged to the socket.
1552 * It has been noticed pure SACK packets were sometimes dropped
1553 * (if cooked by drivers without copybreak feature).
1557 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1559 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1564 EXPORT_SYMBOL(tcp_add_backlog);
1566 int tcp_filter(struct sock *sk, struct sk_buff *skb)
1568 struct tcphdr *th = (struct tcphdr *)skb->data;
1569 unsigned int eaten = skb->len;
1572 err = sk_filter_trim_cap(sk, skb, th->doff * 4);
1575 TCP_SKB_CB(skb)->end_seq -= eaten;
1579 EXPORT_SYMBOL(tcp_filter);
1585 int tcp_v4_rcv(struct sk_buff *skb)
1587 struct net *net = dev_net(skb->dev);
1588 const struct iphdr *iph;
1589 const struct tcphdr *th;
1594 if (skb->pkt_type != PACKET_HOST)
1597 /* Count it even if it's bad */
1598 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1600 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1603 th = (const struct tcphdr *)skb->data;
1605 if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1607 if (!pskb_may_pull(skb, th->doff * 4))
1610 /* An explanation is required here, I think.
1611 * Packet length and doff are validated by header prediction,
1612 * provided case of th->doff==0 is eliminated.
1613 * So, we defer the checks. */
1615 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1618 th = (const struct tcphdr *)skb->data;
1620 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1621 * barrier() makes sure compiler wont play fool^Waliasing games.
1623 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1624 sizeof(struct inet_skb_parm));
1627 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1628 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1629 skb->len - th->doff * 4);
1630 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1631 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1632 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1633 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1634 TCP_SKB_CB(skb)->sacked = 0;
1637 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1638 th->dest, &refcounted);
1643 if (sk->sk_state == TCP_TIME_WAIT)
1646 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1647 struct request_sock *req = inet_reqsk(sk);
1650 sk = req->rsk_listener;
1651 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1652 sk_drops_add(sk, skb);
1656 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1657 inet_csk_reqsk_queue_drop_and_put(sk, req);
1660 /* We own a reference on the listener, increase it again
1661 * as we might lose it too soon.
1665 nsk = tcp_check_req(sk, skb, req, false);
1668 goto discard_and_relse;
1672 } else if (tcp_child_process(sk, nsk, skb)) {
1673 tcp_v4_send_reset(nsk, skb);
1674 goto discard_and_relse;
1680 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1681 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1682 goto discard_and_relse;
1685 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1686 goto discard_and_relse;
1688 if (tcp_v4_inbound_md5_hash(sk, skb))
1689 goto discard_and_relse;
1693 if (tcp_filter(sk, skb))
1694 goto discard_and_relse;
1695 th = (const struct tcphdr *)skb->data;
1700 if (sk->sk_state == TCP_LISTEN) {
1701 ret = tcp_v4_do_rcv(sk, skb);
1702 goto put_and_return;
1705 sk_incoming_cpu_update(sk);
1707 bh_lock_sock_nested(sk);
1708 tcp_segs_in(tcp_sk(sk), skb);
1710 if (!sock_owned_by_user(sk)) {
1711 if (!tcp_prequeue(sk, skb))
1712 ret = tcp_v4_do_rcv(sk, skb);
1713 } else if (tcp_add_backlog(sk, skb)) {
1714 goto discard_and_relse;
1725 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1728 if (tcp_checksum_complete(skb)) {
1730 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1732 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1734 tcp_v4_send_reset(NULL, skb);
1738 /* Discard frame. */
1743 sk_drops_add(sk, skb);
1749 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1750 inet_twsk_put(inet_twsk(sk));
1754 if (tcp_checksum_complete(skb)) {
1755 inet_twsk_put(inet_twsk(sk));
1758 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1760 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1763 iph->saddr, th->source,
1764 iph->daddr, th->dest,
1767 inet_twsk_deschedule_put(inet_twsk(sk));
1772 /* Fall through to ACK */
1775 tcp_v4_timewait_ack(sk, skb);
1778 tcp_v4_send_reset(sk, skb);
1779 inet_twsk_deschedule_put(inet_twsk(sk));
1781 case TCP_TW_SUCCESS:;
1786 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1787 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1788 .twsk_unique = tcp_twsk_unique,
1789 .twsk_destructor= tcp_twsk_destructor,
1792 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1794 struct dst_entry *dst = skb_dst(skb);
1796 if (dst && dst_hold_safe(dst)) {
1797 sk->sk_rx_dst = dst;
1798 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1801 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1803 const struct inet_connection_sock_af_ops ipv4_specific = {
1804 .queue_xmit = ip_queue_xmit,
1805 .send_check = tcp_v4_send_check,
1806 .rebuild_header = inet_sk_rebuild_header,
1807 .sk_rx_dst_set = inet_sk_rx_dst_set,
1808 .conn_request = tcp_v4_conn_request,
1809 .syn_recv_sock = tcp_v4_syn_recv_sock,
1810 .net_header_len = sizeof(struct iphdr),
1811 .setsockopt = ip_setsockopt,
1812 .getsockopt = ip_getsockopt,
1813 .addr2sockaddr = inet_csk_addr2sockaddr,
1814 .sockaddr_len = sizeof(struct sockaddr_in),
1815 #ifdef CONFIG_COMPAT
1816 .compat_setsockopt = compat_ip_setsockopt,
1817 .compat_getsockopt = compat_ip_getsockopt,
1819 .mtu_reduced = tcp_v4_mtu_reduced,
1821 EXPORT_SYMBOL(ipv4_specific);
1823 #ifdef CONFIG_TCP_MD5SIG
1824 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1825 .md5_lookup = tcp_v4_md5_lookup,
1826 .calc_md5_hash = tcp_v4_md5_hash_skb,
1827 .md5_parse = tcp_v4_parse_md5_keys,
1831 /* NOTE: A lot of things set to zero explicitly by call to
1832 * sk_alloc() so need not be done here.
1834 static int tcp_v4_init_sock(struct sock *sk)
1836 struct inet_connection_sock *icsk = inet_csk(sk);
1840 icsk->icsk_af_ops = &ipv4_specific;
1842 #ifdef CONFIG_TCP_MD5SIG
1843 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1849 void tcp_v4_destroy_sock(struct sock *sk)
1851 struct tcp_sock *tp = tcp_sk(sk);
1853 tcp_clear_xmit_timers(sk);
1855 tcp_cleanup_congestion_control(sk);
1857 /* Cleanup up the write buffer. */
1858 tcp_write_queue_purge(sk);
1860 /* Cleans up our, hopefully empty, out_of_order_queue. */
1861 skb_rbtree_purge(&tp->out_of_order_queue);
1863 #ifdef CONFIG_TCP_MD5SIG
1864 /* Clean up the MD5 key list, if any */
1865 if (tp->md5sig_info) {
1866 tcp_clear_md5_list(sk);
1867 kfree_rcu(tp->md5sig_info, rcu);
1868 tp->md5sig_info = NULL;
1872 /* Clean prequeue, it must be empty really */
1873 __skb_queue_purge(&tp->ucopy.prequeue);
1875 /* Clean up a referenced TCP bind bucket. */
1876 if (inet_csk(sk)->icsk_bind_hash)
1879 BUG_ON(tp->fastopen_rsk);
1881 /* If socket is aborted during connect operation */
1882 tcp_free_fastopen_req(tp);
1883 tcp_saved_syn_free(tp);
1885 sk_sockets_allocated_dec(sk);
1887 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1889 #ifdef CONFIG_PROC_FS
1890 /* Proc filesystem TCP sock list dumping. */
1893 * Get next listener socket follow cur. If cur is NULL, get first socket
1894 * starting from bucket given in st->bucket; when st->bucket is zero the
1895 * very first socket in the hash table is returned.
1897 static void *listening_get_next(struct seq_file *seq, void *cur)
1899 struct tcp_iter_state *st = seq->private;
1900 struct net *net = seq_file_net(seq);
1901 struct inet_listen_hashbucket *ilb;
1902 struct sock *sk = cur;
1906 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1907 spin_lock(&ilb->lock);
1908 sk = sk_head(&ilb->head);
1912 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1918 sk_for_each_from(sk) {
1919 if (!net_eq(sock_net(sk), net))
1921 if (sk->sk_family == st->family)
1924 spin_unlock(&ilb->lock);
1926 if (++st->bucket < INET_LHTABLE_SIZE)
1931 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1933 struct tcp_iter_state *st = seq->private;
1938 rc = listening_get_next(seq, NULL);
1940 while (rc && *pos) {
1941 rc = listening_get_next(seq, rc);
1947 static inline bool empty_bucket(const struct tcp_iter_state *st)
1949 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1953 * Get first established socket starting from bucket given in st->bucket.
1954 * If st->bucket is zero, the very first socket in the hash is returned.
1956 static void *established_get_first(struct seq_file *seq)
1958 struct tcp_iter_state *st = seq->private;
1959 struct net *net = seq_file_net(seq);
1963 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1965 struct hlist_nulls_node *node;
1966 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1968 /* Lockless fast path for the common case of empty buckets */
1969 if (empty_bucket(st))
1973 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1974 if (sk->sk_family != st->family ||
1975 !net_eq(sock_net(sk), net)) {
1981 spin_unlock_bh(lock);
1987 static void *established_get_next(struct seq_file *seq, void *cur)
1989 struct sock *sk = cur;
1990 struct hlist_nulls_node *node;
1991 struct tcp_iter_state *st = seq->private;
1992 struct net *net = seq_file_net(seq);
1997 sk = sk_nulls_next(sk);
1999 sk_nulls_for_each_from(sk, node) {
2000 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2004 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2006 return established_get_first(seq);
2009 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2011 struct tcp_iter_state *st = seq->private;
2015 rc = established_get_first(seq);
2018 rc = established_get_next(seq, rc);
2024 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2027 struct tcp_iter_state *st = seq->private;
2029 st->state = TCP_SEQ_STATE_LISTENING;
2030 rc = listening_get_idx(seq, &pos);
2033 st->state = TCP_SEQ_STATE_ESTABLISHED;
2034 rc = established_get_idx(seq, pos);
2040 static void *tcp_seek_last_pos(struct seq_file *seq)
2042 struct tcp_iter_state *st = seq->private;
2043 int offset = st->offset;
2044 int orig_num = st->num;
2047 switch (st->state) {
2048 case TCP_SEQ_STATE_LISTENING:
2049 if (st->bucket >= INET_LHTABLE_SIZE)
2051 st->state = TCP_SEQ_STATE_LISTENING;
2052 rc = listening_get_next(seq, NULL);
2053 while (offset-- && rc)
2054 rc = listening_get_next(seq, rc);
2058 st->state = TCP_SEQ_STATE_ESTABLISHED;
2060 case TCP_SEQ_STATE_ESTABLISHED:
2061 if (st->bucket > tcp_hashinfo.ehash_mask)
2063 rc = established_get_first(seq);
2064 while (offset-- && rc)
2065 rc = established_get_next(seq, rc);
2073 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2075 struct tcp_iter_state *st = seq->private;
2078 if (*pos && *pos == st->last_pos) {
2079 rc = tcp_seek_last_pos(seq);
2084 st->state = TCP_SEQ_STATE_LISTENING;
2088 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2091 st->last_pos = *pos;
2095 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2097 struct tcp_iter_state *st = seq->private;
2100 if (v == SEQ_START_TOKEN) {
2101 rc = tcp_get_idx(seq, 0);
2105 switch (st->state) {
2106 case TCP_SEQ_STATE_LISTENING:
2107 rc = listening_get_next(seq, v);
2109 st->state = TCP_SEQ_STATE_ESTABLISHED;
2112 rc = established_get_first(seq);
2115 case TCP_SEQ_STATE_ESTABLISHED:
2116 rc = established_get_next(seq, v);
2121 st->last_pos = *pos;
2125 static void tcp_seq_stop(struct seq_file *seq, void *v)
2127 struct tcp_iter_state *st = seq->private;
2129 switch (st->state) {
2130 case TCP_SEQ_STATE_LISTENING:
2131 if (v != SEQ_START_TOKEN)
2132 spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2134 case TCP_SEQ_STATE_ESTABLISHED:
2136 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2141 int tcp_seq_open(struct inode *inode, struct file *file)
2143 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2144 struct tcp_iter_state *s;
2147 err = seq_open_net(inode, file, &afinfo->seq_ops,
2148 sizeof(struct tcp_iter_state));
2152 s = ((struct seq_file *)file->private_data)->private;
2153 s->family = afinfo->family;
2157 EXPORT_SYMBOL(tcp_seq_open);
2159 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2162 struct proc_dir_entry *p;
2164 afinfo->seq_ops.start = tcp_seq_start;
2165 afinfo->seq_ops.next = tcp_seq_next;
2166 afinfo->seq_ops.stop = tcp_seq_stop;
2168 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2169 afinfo->seq_fops, afinfo);
2174 EXPORT_SYMBOL(tcp_proc_register);
2176 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2178 remove_proc_entry(afinfo->name, net->proc_net);
2180 EXPORT_SYMBOL(tcp_proc_unregister);
2182 static void get_openreq4(const struct request_sock *req,
2183 struct seq_file *f, int i)
2185 const struct inet_request_sock *ireq = inet_rsk(req);
2186 long delta = req->rsk_timer.expires - jiffies;
2188 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2189 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2194 ntohs(ireq->ir_rmt_port),
2196 0, 0, /* could print option size, but that is af dependent. */
2197 1, /* timers active (only the expire timer) */
2198 jiffies_delta_to_clock_t(delta),
2200 from_kuid_munged(seq_user_ns(f),
2201 sock_i_uid(req->rsk_listener)),
2202 0, /* non standard timer */
2203 0, /* open_requests have no inode */
2208 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2211 unsigned long timer_expires;
2212 const struct tcp_sock *tp = tcp_sk(sk);
2213 const struct inet_connection_sock *icsk = inet_csk(sk);
2214 const struct inet_sock *inet = inet_sk(sk);
2215 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2216 __be32 dest = inet->inet_daddr;
2217 __be32 src = inet->inet_rcv_saddr;
2218 __u16 destp = ntohs(inet->inet_dport);
2219 __u16 srcp = ntohs(inet->inet_sport);
2223 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2224 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2225 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2227 timer_expires = icsk->icsk_timeout;
2228 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2230 timer_expires = icsk->icsk_timeout;
2231 } else if (timer_pending(&sk->sk_timer)) {
2233 timer_expires = sk->sk_timer.expires;
2236 timer_expires = jiffies;
2239 state = sk_state_load(sk);
2240 if (state == TCP_LISTEN)
2241 rx_queue = sk->sk_ack_backlog;
2243 /* Because we don't lock the socket,
2244 * we might find a transient negative value.
2246 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2248 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2249 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2250 i, src, srcp, dest, destp, state,
2251 tp->write_seq - tp->snd_una,
2254 jiffies_delta_to_clock_t(timer_expires - jiffies),
2255 icsk->icsk_retransmits,
2256 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2257 icsk->icsk_probes_out,
2259 atomic_read(&sk->sk_refcnt), sk,
2260 jiffies_to_clock_t(icsk->icsk_rto),
2261 jiffies_to_clock_t(icsk->icsk_ack.ato),
2262 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2264 state == TCP_LISTEN ?
2265 fastopenq->max_qlen :
2266 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2269 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2270 struct seq_file *f, int i)
2272 long delta = tw->tw_timer.expires - jiffies;
2276 dest = tw->tw_daddr;
2277 src = tw->tw_rcv_saddr;
2278 destp = ntohs(tw->tw_dport);
2279 srcp = ntohs(tw->tw_sport);
2281 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2282 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2283 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2284 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2285 atomic_read(&tw->tw_refcnt), tw);
2290 static int tcp4_seq_show(struct seq_file *seq, void *v)
2292 struct tcp_iter_state *st;
2293 struct sock *sk = v;
2295 seq_setwidth(seq, TMPSZ - 1);
2296 if (v == SEQ_START_TOKEN) {
2297 seq_puts(seq, " sl local_address rem_address st tx_queue "
2298 "rx_queue tr tm->when retrnsmt uid timeout "
2304 if (sk->sk_state == TCP_TIME_WAIT)
2305 get_timewait4_sock(v, seq, st->num);
2306 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2307 get_openreq4(v, seq, st->num);
2309 get_tcp4_sock(v, seq, st->num);
2315 static const struct file_operations tcp_afinfo_seq_fops = {
2316 .owner = THIS_MODULE,
2317 .open = tcp_seq_open,
2319 .llseek = seq_lseek,
2320 .release = seq_release_net
2323 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2326 .seq_fops = &tcp_afinfo_seq_fops,
2328 .show = tcp4_seq_show,
2332 static int __net_init tcp4_proc_init_net(struct net *net)
2334 return tcp_proc_register(net, &tcp4_seq_afinfo);
2337 static void __net_exit tcp4_proc_exit_net(struct net *net)
2339 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2342 static struct pernet_operations tcp4_net_ops = {
2343 .init = tcp4_proc_init_net,
2344 .exit = tcp4_proc_exit_net,
2347 int __init tcp4_proc_init(void)
2349 return register_pernet_subsys(&tcp4_net_ops);
2352 void tcp4_proc_exit(void)
2354 unregister_pernet_subsys(&tcp4_net_ops);
2356 #endif /* CONFIG_PROC_FS */
2358 struct proto tcp_prot = {
2360 .owner = THIS_MODULE,
2362 .connect = tcp_v4_connect,
2363 .disconnect = tcp_disconnect,
2364 .accept = inet_csk_accept,
2366 .init = tcp_v4_init_sock,
2367 .destroy = tcp_v4_destroy_sock,
2368 .shutdown = tcp_shutdown,
2369 .setsockopt = tcp_setsockopt,
2370 .getsockopt = tcp_getsockopt,
2371 .keepalive = tcp_set_keepalive,
2372 .recvmsg = tcp_recvmsg,
2373 .sendmsg = tcp_sendmsg,
2374 .sendpage = tcp_sendpage,
2375 .backlog_rcv = tcp_v4_do_rcv,
2376 .release_cb = tcp_release_cb,
2378 .unhash = inet_unhash,
2379 .get_port = inet_csk_get_port,
2380 .enter_memory_pressure = tcp_enter_memory_pressure,
2381 .stream_memory_free = tcp_stream_memory_free,
2382 .sockets_allocated = &tcp_sockets_allocated,
2383 .orphan_count = &tcp_orphan_count,
2384 .memory_allocated = &tcp_memory_allocated,
2385 .memory_pressure = &tcp_memory_pressure,
2386 .sysctl_mem = sysctl_tcp_mem,
2387 .sysctl_wmem = sysctl_tcp_wmem,
2388 .sysctl_rmem = sysctl_tcp_rmem,
2389 .max_header = MAX_TCP_HEADER,
2390 .obj_size = sizeof(struct tcp_sock),
2391 .slab_flags = SLAB_DESTROY_BY_RCU,
2392 .twsk_prot = &tcp_timewait_sock_ops,
2393 .rsk_prot = &tcp_request_sock_ops,
2394 .h.hashinfo = &tcp_hashinfo,
2395 .no_autobind = true,
2396 #ifdef CONFIG_COMPAT
2397 .compat_setsockopt = compat_tcp_setsockopt,
2398 .compat_getsockopt = compat_tcp_getsockopt,
2400 .diag_destroy = tcp_abort,
2402 EXPORT_SYMBOL(tcp_prot);
2404 static void __net_exit tcp_sk_exit(struct net *net)
2408 for_each_possible_cpu(cpu)
2409 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2410 free_percpu(net->ipv4.tcp_sk);
2413 static int __net_init tcp_sk_init(struct net *net)
2417 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2418 if (!net->ipv4.tcp_sk)
2421 for_each_possible_cpu(cpu) {
2424 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2428 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2429 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2432 net->ipv4.sysctl_tcp_ecn = 2;
2433 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2435 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2436 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2437 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2439 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2440 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2441 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2443 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2444 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2445 net->ipv4.sysctl_tcp_syncookies = 1;
2446 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2447 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2448 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2449 net->ipv4.sysctl_tcp_orphan_retries = 0;
2450 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2451 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2452 net->ipv4.sysctl_tcp_tw_reuse = 0;
2454 cnt = tcp_hashinfo.ehash_mask + 1;
2455 net->ipv4.tcp_death_row.sysctl_max_tw_buckets = (cnt + 1) / 2;
2456 net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo;
2458 net->ipv4.sysctl_max_syn_backlog = max(128, cnt / 256);
2467 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2469 inet_twsk_purge(&tcp_hashinfo, AF_INET);
2472 static struct pernet_operations __net_initdata tcp_sk_ops = {
2473 .init = tcp_sk_init,
2474 .exit = tcp_sk_exit,
2475 .exit_batch = tcp_sk_exit_batch,
2478 void __init tcp_v4_init(void)
2480 if (register_pernet_subsys(&tcp_sk_ops))
2481 panic("Failed to create the TCP control socket.\n");