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_tw_reuse __read_mostly;
88 int sysctl_tcp_low_latency __read_mostly;
90 #ifdef CONFIG_TCP_MD5SIG
91 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
92 __be32 daddr, __be32 saddr, const struct tcphdr *th);
95 struct inet_hashinfo tcp_hashinfo;
96 EXPORT_SYMBOL(tcp_hashinfo);
98 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
100 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
103 tcp_hdr(skb)->source);
106 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
108 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
109 struct tcp_sock *tp = tcp_sk(sk);
111 /* With PAWS, it is safe from the viewpoint
112 of data integrity. Even without PAWS it is safe provided sequence
113 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
115 Actually, the idea is close to VJ's one, only timestamp cache is
116 held not per host, but per port pair and TW bucket is used as state
119 If TW bucket has been already destroyed we fall back to VJ's scheme
120 and use initial timestamp retrieved from peer table.
122 if (tcptw->tw_ts_recent_stamp &&
123 (!twp || (sysctl_tcp_tw_reuse &&
124 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
125 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
126 if (tp->write_seq == 0)
128 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
129 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
136 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
138 /* This will initiate an outgoing connection. */
139 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
141 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
142 struct inet_sock *inet = inet_sk(sk);
143 struct tcp_sock *tp = tcp_sk(sk);
144 __be16 orig_sport, orig_dport;
145 __be32 daddr, nexthop;
149 struct ip_options_rcu *inet_opt;
151 if (addr_len < sizeof(struct sockaddr_in))
154 if (usin->sin_family != AF_INET)
155 return -EAFNOSUPPORT;
157 nexthop = daddr = usin->sin_addr.s_addr;
158 inet_opt = rcu_dereference_protected(inet->inet_opt,
159 lockdep_sock_is_held(sk));
160 if (inet_opt && inet_opt->opt.srr) {
163 nexthop = inet_opt->opt.faddr;
166 orig_sport = inet->inet_sport;
167 orig_dport = usin->sin_port;
168 fl4 = &inet->cork.fl.u.ip4;
169 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
170 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
172 orig_sport, orig_dport, sk);
175 if (err == -ENETUNREACH)
176 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
180 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
185 if (!inet_opt || !inet_opt->opt.srr)
188 if (!inet->inet_saddr)
189 inet->inet_saddr = fl4->saddr;
190 sk_rcv_saddr_set(sk, inet->inet_saddr);
192 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
193 /* Reset inherited state */
194 tp->rx_opt.ts_recent = 0;
195 tp->rx_opt.ts_recent_stamp = 0;
196 if (likely(!tp->repair))
200 if (tcp_death_row.sysctl_tw_recycle &&
201 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
202 tcp_fetch_timewait_stamp(sk, &rt->dst);
204 inet->inet_dport = usin->sin_port;
205 sk_daddr_set(sk, daddr);
207 inet_csk(sk)->icsk_ext_hdr_len = 0;
209 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
211 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
213 /* Socket identity is still unknown (sport may be zero).
214 * However we set state to SYN-SENT and not releasing socket
215 * lock select source port, enter ourselves into the hash tables and
216 * complete initialization after this.
218 tcp_set_state(sk, TCP_SYN_SENT);
219 err = inet_hash_connect(&tcp_death_row, sk);
225 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
226 inet->inet_sport, inet->inet_dport, sk);
232 /* OK, now commit destination to socket. */
233 sk->sk_gso_type = SKB_GSO_TCPV4;
234 sk_setup_caps(sk, &rt->dst);
236 if (!tp->write_seq && likely(!tp->repair))
237 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
242 inet->inet_id = tp->write_seq ^ jiffies;
244 err = tcp_connect(sk);
254 * This unhashes the socket and releases the local port,
257 tcp_set_state(sk, TCP_CLOSE);
259 sk->sk_route_caps = 0;
260 inet->inet_dport = 0;
263 EXPORT_SYMBOL(tcp_v4_connect);
266 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
267 * It can be called through tcp_release_cb() if socket was owned by user
268 * at the time tcp_v4_err() was called to handle ICMP message.
270 void tcp_v4_mtu_reduced(struct sock *sk)
272 struct dst_entry *dst;
273 struct inet_sock *inet = inet_sk(sk);
274 u32 mtu = tcp_sk(sk)->mtu_info;
276 dst = inet_csk_update_pmtu(sk, mtu);
280 /* Something is about to be wrong... Remember soft error
281 * for the case, if this connection will not able to recover.
283 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
284 sk->sk_err_soft = EMSGSIZE;
288 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
289 ip_sk_accept_pmtu(sk) &&
290 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
291 tcp_sync_mss(sk, mtu);
293 /* Resend the TCP packet because it's
294 * clear that the old packet has been
295 * dropped. This is the new "fast" path mtu
298 tcp_simple_retransmit(sk);
299 } /* else let the usual retransmit timer handle it */
301 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
303 static void do_redirect(struct sk_buff *skb, struct sock *sk)
305 struct dst_entry *dst = __sk_dst_check(sk, 0);
308 dst->ops->redirect(dst, sk, skb);
312 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
313 void tcp_req_err(struct sock *sk, u32 seq, bool abort)
315 struct request_sock *req = inet_reqsk(sk);
316 struct net *net = sock_net(sk);
318 /* ICMPs are not backlogged, hence we cannot get
319 * an established socket here.
321 if (seq != tcp_rsk(req)->snt_isn) {
322 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
325 * Still in SYN_RECV, just remove it silently.
326 * There is no good way to pass the error to the newly
327 * created socket, and POSIX does not want network
328 * errors returned from accept().
330 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
331 tcp_listendrop(req->rsk_listener);
335 EXPORT_SYMBOL(tcp_req_err);
338 * This routine is called by the ICMP module when it gets some
339 * sort of error condition. If err < 0 then the socket should
340 * be closed and the error returned to the user. If err > 0
341 * it's just the icmp type << 8 | icmp code. After adjustment
342 * header points to the first 8 bytes of the tcp header. We need
343 * to find the appropriate port.
345 * The locking strategy used here is very "optimistic". When
346 * someone else accesses the socket the ICMP is just dropped
347 * and for some paths there is no check at all.
348 * A more general error queue to queue errors for later handling
349 * is probably better.
353 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
355 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
356 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
357 struct inet_connection_sock *icsk;
359 struct inet_sock *inet;
360 const int type = icmp_hdr(icmp_skb)->type;
361 const int code = icmp_hdr(icmp_skb)->code;
364 struct request_sock *fastopen;
368 struct net *net = dev_net(icmp_skb->dev);
370 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
371 th->dest, iph->saddr, ntohs(th->source),
374 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
377 if (sk->sk_state == TCP_TIME_WAIT) {
378 inet_twsk_put(inet_twsk(sk));
381 seq = ntohl(th->seq);
382 if (sk->sk_state == TCP_NEW_SYN_RECV)
383 return tcp_req_err(sk, seq,
384 type == ICMP_PARAMETERPROB ||
385 type == ICMP_TIME_EXCEEDED ||
386 (type == ICMP_DEST_UNREACH &&
387 (code == ICMP_NET_UNREACH ||
388 code == ICMP_HOST_UNREACH)));
391 /* If too many ICMPs get dropped on busy
392 * servers this needs to be solved differently.
393 * We do take care of PMTU discovery (RFC1191) special case :
394 * we can receive locally generated ICMP messages while socket is held.
396 if (sock_owned_by_user(sk)) {
397 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
398 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
400 if (sk->sk_state == TCP_CLOSE)
403 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
404 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
410 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
411 fastopen = tp->fastopen_rsk;
412 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
413 if (sk->sk_state != TCP_LISTEN &&
414 !between(seq, snd_una, tp->snd_nxt)) {
415 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
421 do_redirect(icmp_skb, sk);
423 case ICMP_SOURCE_QUENCH:
424 /* Just silently ignore these. */
426 case ICMP_PARAMETERPROB:
429 case ICMP_DEST_UNREACH:
430 if (code > NR_ICMP_UNREACH)
433 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
434 /* We are not interested in TCP_LISTEN and open_requests
435 * (SYN-ACKs send out by Linux are always <576bytes so
436 * they should go through unfragmented).
438 if (sk->sk_state == TCP_LISTEN)
442 if (!sock_owned_by_user(sk)) {
443 tcp_v4_mtu_reduced(sk);
445 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
451 err = icmp_err_convert[code].errno;
452 /* check if icmp_skb allows revert of backoff
453 * (see draft-zimmermann-tcp-lcd) */
454 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
456 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
457 !icsk->icsk_backoff || fastopen)
460 if (sock_owned_by_user(sk))
463 icsk->icsk_backoff--;
464 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
466 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
468 skb = tcp_write_queue_head(sk);
471 remaining = icsk->icsk_rto -
473 tcp_time_stamp - tcp_skb_timestamp(skb));
476 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
477 remaining, TCP_RTO_MAX);
479 /* RTO revert clocked out retransmission.
480 * Will retransmit now */
481 tcp_retransmit_timer(sk);
485 case ICMP_TIME_EXCEEDED:
492 switch (sk->sk_state) {
495 /* Only in fast or simultaneous open. If a fast open socket is
496 * is already accepted it is treated as a connected one below.
498 if (fastopen && !fastopen->sk)
501 if (!sock_owned_by_user(sk)) {
504 sk->sk_error_report(sk);
508 sk->sk_err_soft = err;
513 /* If we've already connected we will keep trying
514 * until we time out, or the user gives up.
516 * rfc1122 4.2.3.9 allows to consider as hard errors
517 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
518 * but it is obsoleted by pmtu discovery).
520 * Note, that in modern internet, where routing is unreliable
521 * and in each dark corner broken firewalls sit, sending random
522 * errors ordered by their masters even this two messages finally lose
523 * their original sense (even Linux sends invalid PORT_UNREACHs)
525 * Now we are in compliance with RFCs.
530 if (!sock_owned_by_user(sk) && inet->recverr) {
532 sk->sk_error_report(sk);
533 } else { /* Only an error on timeout */
534 sk->sk_err_soft = err;
542 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
544 struct tcphdr *th = tcp_hdr(skb);
546 if (skb->ip_summed == CHECKSUM_PARTIAL) {
547 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
548 skb->csum_start = skb_transport_header(skb) - skb->head;
549 skb->csum_offset = offsetof(struct tcphdr, check);
551 th->check = tcp_v4_check(skb->len, saddr, daddr,
558 /* This routine computes an IPv4 TCP checksum. */
559 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
561 const struct inet_sock *inet = inet_sk(sk);
563 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
565 EXPORT_SYMBOL(tcp_v4_send_check);
568 * This routine will send an RST to the other tcp.
570 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
572 * Answer: if a packet caused RST, it is not for a socket
573 * existing in our system, if it is matched to a socket,
574 * it is just duplicate segment or bug in other side's TCP.
575 * So that we build reply only basing on parameters
576 * arrived with segment.
577 * Exception: precedence violation. We do not implement it in any case.
580 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
582 const struct tcphdr *th = tcp_hdr(skb);
585 #ifdef CONFIG_TCP_MD5SIG
586 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
589 struct ip_reply_arg arg;
590 #ifdef CONFIG_TCP_MD5SIG
591 struct tcp_md5sig_key *key = NULL;
592 const __u8 *hash_location = NULL;
593 unsigned char newhash[16];
595 struct sock *sk1 = NULL;
599 /* Never send a reset in response to a reset. */
603 /* If sk not NULL, it means we did a successful lookup and incoming
604 * route had to be correct. prequeue might have dropped our dst.
606 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
609 /* Swap the send and the receive. */
610 memset(&rep, 0, sizeof(rep));
611 rep.th.dest = th->source;
612 rep.th.source = th->dest;
613 rep.th.doff = sizeof(struct tcphdr) / 4;
617 rep.th.seq = th->ack_seq;
620 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
621 skb->len - (th->doff << 2));
624 memset(&arg, 0, sizeof(arg));
625 arg.iov[0].iov_base = (unsigned char *)&rep;
626 arg.iov[0].iov_len = sizeof(rep.th);
628 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
629 #ifdef CONFIG_TCP_MD5SIG
631 hash_location = tcp_parse_md5sig_option(th);
632 if (sk && sk_fullsock(sk)) {
633 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
634 &ip_hdr(skb)->saddr, AF_INET);
635 } else if (hash_location) {
637 * active side is lost. Try to find listening socket through
638 * source port, and then find md5 key through listening socket.
639 * we are not loose security here:
640 * Incoming packet is checked with md5 hash with finding key,
641 * no RST generated if md5 hash doesn't match.
643 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
645 th->source, ip_hdr(skb)->daddr,
646 ntohs(th->source), inet_iif(skb));
647 /* don't send rst if it can't find key */
651 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
652 &ip_hdr(skb)->saddr, AF_INET);
657 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
658 if (genhash || memcmp(hash_location, newhash, 16) != 0)
664 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
666 (TCPOPT_MD5SIG << 8) |
668 /* Update length and the length the header thinks exists */
669 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
670 rep.th.doff = arg.iov[0].iov_len / 4;
672 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
673 key, ip_hdr(skb)->saddr,
674 ip_hdr(skb)->daddr, &rep.th);
677 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
678 ip_hdr(skb)->saddr, /* XXX */
679 arg.iov[0].iov_len, IPPROTO_TCP, 0);
680 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
681 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
683 /* When socket is gone, all binding information is lost.
684 * routing might fail in this case. No choice here, if we choose to force
685 * input interface, we will misroute in case of asymmetric route.
688 arg.bound_dev_if = sk->sk_bound_dev_if;
690 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
691 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
693 arg.tos = ip_hdr(skb)->tos;
695 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
696 skb, &TCP_SKB_CB(skb)->header.h4.opt,
697 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
698 &arg, arg.iov[0].iov_len);
700 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
701 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
704 #ifdef CONFIG_TCP_MD5SIG
710 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
711 outside socket context is ugly, certainly. What can I do?
714 static void tcp_v4_send_ack(struct net *net,
715 struct sk_buff *skb, u32 seq, u32 ack,
716 u32 win, u32 tsval, u32 tsecr, int oif,
717 struct tcp_md5sig_key *key,
718 int reply_flags, u8 tos)
720 const struct tcphdr *th = tcp_hdr(skb);
723 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
724 #ifdef CONFIG_TCP_MD5SIG
725 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
729 struct ip_reply_arg arg;
731 memset(&rep.th, 0, sizeof(struct tcphdr));
732 memset(&arg, 0, sizeof(arg));
734 arg.iov[0].iov_base = (unsigned char *)&rep;
735 arg.iov[0].iov_len = sizeof(rep.th);
737 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
738 (TCPOPT_TIMESTAMP << 8) |
740 rep.opt[1] = htonl(tsval);
741 rep.opt[2] = htonl(tsecr);
742 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
745 /* Swap the send and the receive. */
746 rep.th.dest = th->source;
747 rep.th.source = th->dest;
748 rep.th.doff = arg.iov[0].iov_len / 4;
749 rep.th.seq = htonl(seq);
750 rep.th.ack_seq = htonl(ack);
752 rep.th.window = htons(win);
754 #ifdef CONFIG_TCP_MD5SIG
756 int offset = (tsecr) ? 3 : 0;
758 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
760 (TCPOPT_MD5SIG << 8) |
762 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
763 rep.th.doff = arg.iov[0].iov_len/4;
765 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
766 key, ip_hdr(skb)->saddr,
767 ip_hdr(skb)->daddr, &rep.th);
770 arg.flags = reply_flags;
771 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
772 ip_hdr(skb)->saddr, /* XXX */
773 arg.iov[0].iov_len, IPPROTO_TCP, 0);
774 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
776 arg.bound_dev_if = oif;
779 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
780 skb, &TCP_SKB_CB(skb)->header.h4.opt,
781 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
782 &arg, arg.iov[0].iov_len);
784 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
788 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
790 struct inet_timewait_sock *tw = inet_twsk(sk);
791 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
793 tcp_v4_send_ack(sock_net(sk), skb,
794 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
795 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
796 tcp_time_stamp + tcptw->tw_ts_offset,
799 tcp_twsk_md5_key(tcptw),
800 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
807 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
808 struct request_sock *req)
810 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
811 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
813 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
817 * The window field (SEG.WND) of every outgoing segment, with the
818 * exception of <SYN> segments, MUST be right-shifted by
819 * Rcv.Wind.Shift bits:
821 tcp_v4_send_ack(sock_net(sk), skb, seq,
822 tcp_rsk(req)->rcv_nxt,
823 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
827 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
829 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
834 * Send a SYN-ACK after having received a SYN.
835 * This still operates on a request_sock only, not on a big
838 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
840 struct request_sock *req,
841 struct tcp_fastopen_cookie *foc,
842 enum tcp_synack_type synack_type)
844 const struct inet_request_sock *ireq = inet_rsk(req);
849 /* First, grab a route. */
850 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
853 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
856 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
858 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
861 err = net_xmit_eval(err);
868 * IPv4 request_sock destructor.
870 static void tcp_v4_reqsk_destructor(struct request_sock *req)
872 kfree(inet_rsk(req)->opt);
875 #ifdef CONFIG_TCP_MD5SIG
877 * RFC2385 MD5 checksumming requires a mapping of
878 * IP address->MD5 Key.
879 * We need to maintain these in the sk structure.
882 /* Find the Key structure for an address. */
883 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
884 const union tcp_md5_addr *addr,
887 const struct tcp_sock *tp = tcp_sk(sk);
888 struct tcp_md5sig_key *key;
889 unsigned int size = sizeof(struct in_addr);
890 const struct tcp_md5sig_info *md5sig;
892 /* caller either holds rcu_read_lock() or socket lock */
893 md5sig = rcu_dereference_check(tp->md5sig_info,
894 lockdep_sock_is_held(sk));
897 #if IS_ENABLED(CONFIG_IPV6)
898 if (family == AF_INET6)
899 size = sizeof(struct in6_addr);
901 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
902 if (key->family != family)
904 if (!memcmp(&key->addr, addr, size))
909 EXPORT_SYMBOL(tcp_md5_do_lookup);
911 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
912 const struct sock *addr_sk)
914 const union tcp_md5_addr *addr;
916 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
917 return tcp_md5_do_lookup(sk, addr, AF_INET);
919 EXPORT_SYMBOL(tcp_v4_md5_lookup);
921 /* This can be called on a newly created socket, from other files */
922 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
923 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
925 /* Add Key to the list */
926 struct tcp_md5sig_key *key;
927 struct tcp_sock *tp = tcp_sk(sk);
928 struct tcp_md5sig_info *md5sig;
930 key = tcp_md5_do_lookup(sk, addr, family);
932 /* Pre-existing entry - just update that one. */
933 memcpy(key->key, newkey, newkeylen);
934 key->keylen = newkeylen;
938 md5sig = rcu_dereference_protected(tp->md5sig_info,
939 lockdep_sock_is_held(sk));
941 md5sig = kmalloc(sizeof(*md5sig), gfp);
945 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
946 INIT_HLIST_HEAD(&md5sig->head);
947 rcu_assign_pointer(tp->md5sig_info, md5sig);
950 key = sock_kmalloc(sk, sizeof(*key), gfp);
953 if (!tcp_alloc_md5sig_pool()) {
954 sock_kfree_s(sk, key, sizeof(*key));
958 memcpy(key->key, newkey, newkeylen);
959 key->keylen = newkeylen;
960 key->family = family;
961 memcpy(&key->addr, addr,
962 (family == AF_INET6) ? sizeof(struct in6_addr) :
963 sizeof(struct in_addr));
964 hlist_add_head_rcu(&key->node, &md5sig->head);
967 EXPORT_SYMBOL(tcp_md5_do_add);
969 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
971 struct tcp_md5sig_key *key;
973 key = tcp_md5_do_lookup(sk, addr, family);
976 hlist_del_rcu(&key->node);
977 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
981 EXPORT_SYMBOL(tcp_md5_do_del);
983 static void tcp_clear_md5_list(struct sock *sk)
985 struct tcp_sock *tp = tcp_sk(sk);
986 struct tcp_md5sig_key *key;
987 struct hlist_node *n;
988 struct tcp_md5sig_info *md5sig;
990 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
992 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
993 hlist_del_rcu(&key->node);
994 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
999 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1002 struct tcp_md5sig cmd;
1003 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1005 if (optlen < sizeof(cmd))
1008 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1011 if (sin->sin_family != AF_INET)
1014 if (!cmd.tcpm_keylen)
1015 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1018 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1021 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1022 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1026 static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1027 __be32 daddr, __be32 saddr,
1028 const struct tcphdr *th, int nbytes)
1030 struct tcp4_pseudohdr *bp;
1031 struct scatterlist sg;
1038 bp->protocol = IPPROTO_TCP;
1039 bp->len = cpu_to_be16(nbytes);
1041 _th = (struct tcphdr *)(bp + 1);
1042 memcpy(_th, th, sizeof(*th));
1045 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1046 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1047 sizeof(*bp) + sizeof(*th));
1048 return crypto_ahash_update(hp->md5_req);
1051 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1052 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1054 struct tcp_md5sig_pool *hp;
1055 struct ahash_request *req;
1057 hp = tcp_get_md5sig_pool();
1059 goto clear_hash_noput;
1062 if (crypto_ahash_init(req))
1064 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1066 if (tcp_md5_hash_key(hp, key))
1068 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1069 if (crypto_ahash_final(req))
1072 tcp_put_md5sig_pool();
1076 tcp_put_md5sig_pool();
1078 memset(md5_hash, 0, 16);
1082 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1083 const struct sock *sk,
1084 const struct sk_buff *skb)
1086 struct tcp_md5sig_pool *hp;
1087 struct ahash_request *req;
1088 const struct tcphdr *th = tcp_hdr(skb);
1089 __be32 saddr, daddr;
1091 if (sk) { /* valid for establish/request sockets */
1092 saddr = sk->sk_rcv_saddr;
1093 daddr = sk->sk_daddr;
1095 const struct iphdr *iph = ip_hdr(skb);
1100 hp = tcp_get_md5sig_pool();
1102 goto clear_hash_noput;
1105 if (crypto_ahash_init(req))
1108 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1110 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1112 if (tcp_md5_hash_key(hp, key))
1114 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1115 if (crypto_ahash_final(req))
1118 tcp_put_md5sig_pool();
1122 tcp_put_md5sig_pool();
1124 memset(md5_hash, 0, 16);
1127 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1131 /* Called with rcu_read_lock() */
1132 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1133 const struct sk_buff *skb)
1135 #ifdef CONFIG_TCP_MD5SIG
1137 * This gets called for each TCP segment that arrives
1138 * so we want to be efficient.
1139 * We have 3 drop cases:
1140 * o No MD5 hash and one expected.
1141 * o MD5 hash and we're not expecting one.
1142 * o MD5 hash and its wrong.
1144 const __u8 *hash_location = NULL;
1145 struct tcp_md5sig_key *hash_expected;
1146 const struct iphdr *iph = ip_hdr(skb);
1147 const struct tcphdr *th = tcp_hdr(skb);
1149 unsigned char newhash[16];
1151 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1153 hash_location = tcp_parse_md5sig_option(th);
1155 /* We've parsed the options - do we have a hash? */
1156 if (!hash_expected && !hash_location)
1159 if (hash_expected && !hash_location) {
1160 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1164 if (!hash_expected && hash_location) {
1165 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1169 /* Okay, so this is hash_expected and hash_location -
1170 * so we need to calculate the checksum.
1172 genhash = tcp_v4_md5_hash_skb(newhash,
1176 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1177 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1178 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1179 &iph->saddr, ntohs(th->source),
1180 &iph->daddr, ntohs(th->dest),
1181 genhash ? " tcp_v4_calc_md5_hash failed"
1190 static void tcp_v4_init_req(struct request_sock *req,
1191 const struct sock *sk_listener,
1192 struct sk_buff *skb)
1194 struct inet_request_sock *ireq = inet_rsk(req);
1196 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1197 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1198 ireq->opt = tcp_v4_save_options(skb);
1201 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1203 const struct request_sock *req,
1206 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1209 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1218 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1220 .obj_size = sizeof(struct tcp_request_sock),
1221 .rtx_syn_ack = tcp_rtx_synack,
1222 .send_ack = tcp_v4_reqsk_send_ack,
1223 .destructor = tcp_v4_reqsk_destructor,
1224 .send_reset = tcp_v4_send_reset,
1225 .syn_ack_timeout = tcp_syn_ack_timeout,
1228 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1229 .mss_clamp = TCP_MSS_DEFAULT,
1230 #ifdef CONFIG_TCP_MD5SIG
1231 .req_md5_lookup = tcp_v4_md5_lookup,
1232 .calc_md5_hash = tcp_v4_md5_hash_skb,
1234 .init_req = tcp_v4_init_req,
1235 #ifdef CONFIG_SYN_COOKIES
1236 .cookie_init_seq = cookie_v4_init_sequence,
1238 .route_req = tcp_v4_route_req,
1239 .init_seq = tcp_v4_init_sequence,
1240 .send_synack = tcp_v4_send_synack,
1243 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1245 /* Never answer to SYNs send to broadcast or multicast */
1246 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1249 return tcp_conn_request(&tcp_request_sock_ops,
1250 &tcp_request_sock_ipv4_ops, sk, skb);
1256 EXPORT_SYMBOL(tcp_v4_conn_request);
1260 * The three way handshake has completed - we got a valid synack -
1261 * now create the new socket.
1263 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1264 struct request_sock *req,
1265 struct dst_entry *dst,
1266 struct request_sock *req_unhash,
1269 struct inet_request_sock *ireq;
1270 struct inet_sock *newinet;
1271 struct tcp_sock *newtp;
1273 #ifdef CONFIG_TCP_MD5SIG
1274 struct tcp_md5sig_key *key;
1276 struct ip_options_rcu *inet_opt;
1278 if (sk_acceptq_is_full(sk))
1281 newsk = tcp_create_openreq_child(sk, req, skb);
1285 newsk->sk_gso_type = SKB_GSO_TCPV4;
1286 inet_sk_rx_dst_set(newsk, skb);
1288 newtp = tcp_sk(newsk);
1289 newinet = inet_sk(newsk);
1290 ireq = inet_rsk(req);
1291 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1292 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1293 newsk->sk_bound_dev_if = ireq->ir_iif;
1294 newinet->inet_saddr = ireq->ir_loc_addr;
1295 inet_opt = ireq->opt;
1296 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1298 newinet->mc_index = inet_iif(skb);
1299 newinet->mc_ttl = ip_hdr(skb)->ttl;
1300 newinet->rcv_tos = ip_hdr(skb)->tos;
1301 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1303 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1304 newinet->inet_id = newtp->write_seq ^ jiffies;
1307 dst = inet_csk_route_child_sock(sk, newsk, req);
1311 /* syncookie case : see end of cookie_v4_check() */
1313 sk_setup_caps(newsk, dst);
1315 tcp_ca_openreq_child(newsk, dst);
1317 tcp_sync_mss(newsk, dst_mtu(dst));
1318 newtp->advmss = dst_metric_advmss(dst);
1319 if (tcp_sk(sk)->rx_opt.user_mss &&
1320 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1321 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
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).
1556 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
1558 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1560 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1565 EXPORT_SYMBOL(tcp_add_backlog);
1571 int tcp_v4_rcv(struct sk_buff *skb)
1573 struct net *net = dev_net(skb->dev);
1574 const struct iphdr *iph;
1575 const struct tcphdr *th;
1580 if (skb->pkt_type != PACKET_HOST)
1583 /* Count it even if it's bad */
1584 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1586 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1589 th = (const struct tcphdr *)skb->data;
1591 if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1593 if (!pskb_may_pull(skb, th->doff * 4))
1596 /* An explanation is required here, I think.
1597 * Packet length and doff are validated by header prediction,
1598 * provided case of th->doff==0 is eliminated.
1599 * So, we defer the checks. */
1601 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1604 th = (const struct tcphdr *)skb->data;
1606 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1607 * barrier() makes sure compiler wont play fool^Waliasing games.
1609 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1610 sizeof(struct inet_skb_parm));
1613 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1614 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1615 skb->len - th->doff * 4);
1616 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1617 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1618 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1619 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1620 TCP_SKB_CB(skb)->sacked = 0;
1623 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1624 th->dest, &refcounted);
1629 if (sk->sk_state == TCP_TIME_WAIT)
1632 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1633 struct request_sock *req = inet_reqsk(sk);
1636 sk = req->rsk_listener;
1637 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1638 sk_drops_add(sk, skb);
1642 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1643 inet_csk_reqsk_queue_drop_and_put(sk, req);
1646 /* We own a reference on the listener, increase it again
1647 * as we might lose it too soon.
1651 nsk = tcp_check_req(sk, skb, req, false);
1654 goto discard_and_relse;
1658 } else if (tcp_child_process(sk, nsk, skb)) {
1659 tcp_v4_send_reset(nsk, skb);
1660 goto discard_and_relse;
1666 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1667 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1668 goto discard_and_relse;
1671 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1672 goto discard_and_relse;
1674 if (tcp_v4_inbound_md5_hash(sk, skb))
1675 goto discard_and_relse;
1679 if (sk_filter(sk, skb))
1680 goto discard_and_relse;
1684 if (sk->sk_state == TCP_LISTEN) {
1685 ret = tcp_v4_do_rcv(sk, skb);
1686 goto put_and_return;
1689 sk_incoming_cpu_update(sk);
1691 bh_lock_sock_nested(sk);
1692 tcp_segs_in(tcp_sk(sk), skb);
1694 if (!sock_owned_by_user(sk)) {
1695 if (!tcp_prequeue(sk, skb))
1696 ret = tcp_v4_do_rcv(sk, skb);
1697 } else if (tcp_add_backlog(sk, skb)) {
1698 goto discard_and_relse;
1709 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1712 if (tcp_checksum_complete(skb)) {
1714 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1716 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1718 tcp_v4_send_reset(NULL, skb);
1722 /* Discard frame. */
1727 sk_drops_add(sk, skb);
1733 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1734 inet_twsk_put(inet_twsk(sk));
1738 if (tcp_checksum_complete(skb)) {
1739 inet_twsk_put(inet_twsk(sk));
1742 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1744 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1747 iph->saddr, th->source,
1748 iph->daddr, th->dest,
1751 inet_twsk_deschedule_put(inet_twsk(sk));
1756 /* Fall through to ACK */
1759 tcp_v4_timewait_ack(sk, skb);
1762 tcp_v4_send_reset(sk, skb);
1763 inet_twsk_deschedule_put(inet_twsk(sk));
1765 case TCP_TW_SUCCESS:;
1770 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1771 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1772 .twsk_unique = tcp_twsk_unique,
1773 .twsk_destructor= tcp_twsk_destructor,
1776 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1778 struct dst_entry *dst = skb_dst(skb);
1780 if (dst && dst_hold_safe(dst)) {
1781 sk->sk_rx_dst = dst;
1782 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1785 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1787 const struct inet_connection_sock_af_ops ipv4_specific = {
1788 .queue_xmit = ip_queue_xmit,
1789 .send_check = tcp_v4_send_check,
1790 .rebuild_header = inet_sk_rebuild_header,
1791 .sk_rx_dst_set = inet_sk_rx_dst_set,
1792 .conn_request = tcp_v4_conn_request,
1793 .syn_recv_sock = tcp_v4_syn_recv_sock,
1794 .net_header_len = sizeof(struct iphdr),
1795 .setsockopt = ip_setsockopt,
1796 .getsockopt = ip_getsockopt,
1797 .addr2sockaddr = inet_csk_addr2sockaddr,
1798 .sockaddr_len = sizeof(struct sockaddr_in),
1799 .bind_conflict = inet_csk_bind_conflict,
1800 #ifdef CONFIG_COMPAT
1801 .compat_setsockopt = compat_ip_setsockopt,
1802 .compat_getsockopt = compat_ip_getsockopt,
1804 .mtu_reduced = tcp_v4_mtu_reduced,
1806 EXPORT_SYMBOL(ipv4_specific);
1808 #ifdef CONFIG_TCP_MD5SIG
1809 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1810 .md5_lookup = tcp_v4_md5_lookup,
1811 .calc_md5_hash = tcp_v4_md5_hash_skb,
1812 .md5_parse = tcp_v4_parse_md5_keys,
1816 /* NOTE: A lot of things set to zero explicitly by call to
1817 * sk_alloc() so need not be done here.
1819 static int tcp_v4_init_sock(struct sock *sk)
1821 struct inet_connection_sock *icsk = inet_csk(sk);
1825 icsk->icsk_af_ops = &ipv4_specific;
1827 #ifdef CONFIG_TCP_MD5SIG
1828 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1834 void tcp_v4_destroy_sock(struct sock *sk)
1836 struct tcp_sock *tp = tcp_sk(sk);
1838 tcp_clear_xmit_timers(sk);
1840 tcp_cleanup_congestion_control(sk);
1842 /* Cleanup up the write buffer. */
1843 tcp_write_queue_purge(sk);
1845 /* Cleans up our, hopefully empty, out_of_order_queue. */
1846 skb_rbtree_purge(&tp->out_of_order_queue);
1848 #ifdef CONFIG_TCP_MD5SIG
1849 /* Clean up the MD5 key list, if any */
1850 if (tp->md5sig_info) {
1851 tcp_clear_md5_list(sk);
1852 kfree_rcu(tp->md5sig_info, rcu);
1853 tp->md5sig_info = NULL;
1857 /* Clean prequeue, it must be empty really */
1858 __skb_queue_purge(&tp->ucopy.prequeue);
1860 /* Clean up a referenced TCP bind bucket. */
1861 if (inet_csk(sk)->icsk_bind_hash)
1864 BUG_ON(tp->fastopen_rsk);
1866 /* If socket is aborted during connect operation */
1867 tcp_free_fastopen_req(tp);
1868 tcp_saved_syn_free(tp);
1871 sk_sockets_allocated_dec(sk);
1874 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1876 #ifdef CONFIG_PROC_FS
1877 /* Proc filesystem TCP sock list dumping. */
1880 * Get next listener socket follow cur. If cur is NULL, get first socket
1881 * starting from bucket given in st->bucket; when st->bucket is zero the
1882 * very first socket in the hash table is returned.
1884 static void *listening_get_next(struct seq_file *seq, void *cur)
1886 struct tcp_iter_state *st = seq->private;
1887 struct net *net = seq_file_net(seq);
1888 struct inet_listen_hashbucket *ilb;
1889 struct sock *sk = cur;
1893 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1894 spin_lock_bh(&ilb->lock);
1895 sk = sk_head(&ilb->head);
1899 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1905 sk_for_each_from(sk) {
1906 if (!net_eq(sock_net(sk), net))
1908 if (sk->sk_family == st->family)
1911 spin_unlock_bh(&ilb->lock);
1913 if (++st->bucket < INET_LHTABLE_SIZE)
1918 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1920 struct tcp_iter_state *st = seq->private;
1925 rc = listening_get_next(seq, NULL);
1927 while (rc && *pos) {
1928 rc = listening_get_next(seq, rc);
1934 static inline bool empty_bucket(const struct tcp_iter_state *st)
1936 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1940 * Get first established socket starting from bucket given in st->bucket.
1941 * If st->bucket is zero, the very first socket in the hash is returned.
1943 static void *established_get_first(struct seq_file *seq)
1945 struct tcp_iter_state *st = seq->private;
1946 struct net *net = seq_file_net(seq);
1950 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1952 struct hlist_nulls_node *node;
1953 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1955 /* Lockless fast path for the common case of empty buckets */
1956 if (empty_bucket(st))
1960 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1961 if (sk->sk_family != st->family ||
1962 !net_eq(sock_net(sk), net)) {
1968 spin_unlock_bh(lock);
1974 static void *established_get_next(struct seq_file *seq, void *cur)
1976 struct sock *sk = cur;
1977 struct hlist_nulls_node *node;
1978 struct tcp_iter_state *st = seq->private;
1979 struct net *net = seq_file_net(seq);
1984 sk = sk_nulls_next(sk);
1986 sk_nulls_for_each_from(sk, node) {
1987 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
1991 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1993 return established_get_first(seq);
1996 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1998 struct tcp_iter_state *st = seq->private;
2002 rc = established_get_first(seq);
2005 rc = established_get_next(seq, rc);
2011 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2014 struct tcp_iter_state *st = seq->private;
2016 st->state = TCP_SEQ_STATE_LISTENING;
2017 rc = listening_get_idx(seq, &pos);
2020 st->state = TCP_SEQ_STATE_ESTABLISHED;
2021 rc = established_get_idx(seq, pos);
2027 static void *tcp_seek_last_pos(struct seq_file *seq)
2029 struct tcp_iter_state *st = seq->private;
2030 int offset = st->offset;
2031 int orig_num = st->num;
2034 switch (st->state) {
2035 case TCP_SEQ_STATE_LISTENING:
2036 if (st->bucket >= INET_LHTABLE_SIZE)
2038 st->state = TCP_SEQ_STATE_LISTENING;
2039 rc = listening_get_next(seq, NULL);
2040 while (offset-- && rc)
2041 rc = listening_get_next(seq, rc);
2045 st->state = TCP_SEQ_STATE_ESTABLISHED;
2047 case TCP_SEQ_STATE_ESTABLISHED:
2048 if (st->bucket > tcp_hashinfo.ehash_mask)
2050 rc = established_get_first(seq);
2051 while (offset-- && rc)
2052 rc = established_get_next(seq, rc);
2060 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2062 struct tcp_iter_state *st = seq->private;
2065 if (*pos && *pos == st->last_pos) {
2066 rc = tcp_seek_last_pos(seq);
2071 st->state = TCP_SEQ_STATE_LISTENING;
2075 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2078 st->last_pos = *pos;
2082 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2084 struct tcp_iter_state *st = seq->private;
2087 if (v == SEQ_START_TOKEN) {
2088 rc = tcp_get_idx(seq, 0);
2092 switch (st->state) {
2093 case TCP_SEQ_STATE_LISTENING:
2094 rc = listening_get_next(seq, v);
2096 st->state = TCP_SEQ_STATE_ESTABLISHED;
2099 rc = established_get_first(seq);
2102 case TCP_SEQ_STATE_ESTABLISHED:
2103 rc = established_get_next(seq, v);
2108 st->last_pos = *pos;
2112 static void tcp_seq_stop(struct seq_file *seq, void *v)
2114 struct tcp_iter_state *st = seq->private;
2116 switch (st->state) {
2117 case TCP_SEQ_STATE_LISTENING:
2118 if (v != SEQ_START_TOKEN)
2119 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2121 case TCP_SEQ_STATE_ESTABLISHED:
2123 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2128 int tcp_seq_open(struct inode *inode, struct file *file)
2130 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2131 struct tcp_iter_state *s;
2134 err = seq_open_net(inode, file, &afinfo->seq_ops,
2135 sizeof(struct tcp_iter_state));
2139 s = ((struct seq_file *)file->private_data)->private;
2140 s->family = afinfo->family;
2144 EXPORT_SYMBOL(tcp_seq_open);
2146 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2149 struct proc_dir_entry *p;
2151 afinfo->seq_ops.start = tcp_seq_start;
2152 afinfo->seq_ops.next = tcp_seq_next;
2153 afinfo->seq_ops.stop = tcp_seq_stop;
2155 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2156 afinfo->seq_fops, afinfo);
2161 EXPORT_SYMBOL(tcp_proc_register);
2163 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2165 remove_proc_entry(afinfo->name, net->proc_net);
2167 EXPORT_SYMBOL(tcp_proc_unregister);
2169 static void get_openreq4(const struct request_sock *req,
2170 struct seq_file *f, int i)
2172 const struct inet_request_sock *ireq = inet_rsk(req);
2173 long delta = req->rsk_timer.expires - jiffies;
2175 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2176 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2181 ntohs(ireq->ir_rmt_port),
2183 0, 0, /* could print option size, but that is af dependent. */
2184 1, /* timers active (only the expire timer) */
2185 jiffies_delta_to_clock_t(delta),
2187 from_kuid_munged(seq_user_ns(f),
2188 sock_i_uid(req->rsk_listener)),
2189 0, /* non standard timer */
2190 0, /* open_requests have no inode */
2195 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2198 unsigned long timer_expires;
2199 const struct tcp_sock *tp = tcp_sk(sk);
2200 const struct inet_connection_sock *icsk = inet_csk(sk);
2201 const struct inet_sock *inet = inet_sk(sk);
2202 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2203 __be32 dest = inet->inet_daddr;
2204 __be32 src = inet->inet_rcv_saddr;
2205 __u16 destp = ntohs(inet->inet_dport);
2206 __u16 srcp = ntohs(inet->inet_sport);
2210 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2211 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2212 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2214 timer_expires = icsk->icsk_timeout;
2215 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2217 timer_expires = icsk->icsk_timeout;
2218 } else if (timer_pending(&sk->sk_timer)) {
2220 timer_expires = sk->sk_timer.expires;
2223 timer_expires = jiffies;
2226 state = sk_state_load(sk);
2227 if (state == TCP_LISTEN)
2228 rx_queue = sk->sk_ack_backlog;
2230 /* Because we don't lock the socket,
2231 * we might find a transient negative value.
2233 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2235 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2236 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2237 i, src, srcp, dest, destp, state,
2238 tp->write_seq - tp->snd_una,
2241 jiffies_delta_to_clock_t(timer_expires - jiffies),
2242 icsk->icsk_retransmits,
2243 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2244 icsk->icsk_probes_out,
2246 atomic_read(&sk->sk_refcnt), sk,
2247 jiffies_to_clock_t(icsk->icsk_rto),
2248 jiffies_to_clock_t(icsk->icsk_ack.ato),
2249 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2251 state == TCP_LISTEN ?
2252 fastopenq->max_qlen :
2253 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2256 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2257 struct seq_file *f, int i)
2259 long delta = tw->tw_timer.expires - jiffies;
2263 dest = tw->tw_daddr;
2264 src = tw->tw_rcv_saddr;
2265 destp = ntohs(tw->tw_dport);
2266 srcp = ntohs(tw->tw_sport);
2268 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2269 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2270 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2271 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2272 atomic_read(&tw->tw_refcnt), tw);
2277 static int tcp4_seq_show(struct seq_file *seq, void *v)
2279 struct tcp_iter_state *st;
2280 struct sock *sk = v;
2282 seq_setwidth(seq, TMPSZ - 1);
2283 if (v == SEQ_START_TOKEN) {
2284 seq_puts(seq, " sl local_address rem_address st tx_queue "
2285 "rx_queue tr tm->when retrnsmt uid timeout "
2291 if (sk->sk_state == TCP_TIME_WAIT)
2292 get_timewait4_sock(v, seq, st->num);
2293 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2294 get_openreq4(v, seq, st->num);
2296 get_tcp4_sock(v, seq, st->num);
2302 static const struct file_operations tcp_afinfo_seq_fops = {
2303 .owner = THIS_MODULE,
2304 .open = tcp_seq_open,
2306 .llseek = seq_lseek,
2307 .release = seq_release_net
2310 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2313 .seq_fops = &tcp_afinfo_seq_fops,
2315 .show = tcp4_seq_show,
2319 static int __net_init tcp4_proc_init_net(struct net *net)
2321 return tcp_proc_register(net, &tcp4_seq_afinfo);
2324 static void __net_exit tcp4_proc_exit_net(struct net *net)
2326 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2329 static struct pernet_operations tcp4_net_ops = {
2330 .init = tcp4_proc_init_net,
2331 .exit = tcp4_proc_exit_net,
2334 int __init tcp4_proc_init(void)
2336 return register_pernet_subsys(&tcp4_net_ops);
2339 void tcp4_proc_exit(void)
2341 unregister_pernet_subsys(&tcp4_net_ops);
2343 #endif /* CONFIG_PROC_FS */
2345 struct proto tcp_prot = {
2347 .owner = THIS_MODULE,
2349 .connect = tcp_v4_connect,
2350 .disconnect = tcp_disconnect,
2351 .accept = inet_csk_accept,
2353 .init = tcp_v4_init_sock,
2354 .destroy = tcp_v4_destroy_sock,
2355 .shutdown = tcp_shutdown,
2356 .setsockopt = tcp_setsockopt,
2357 .getsockopt = tcp_getsockopt,
2358 .recvmsg = tcp_recvmsg,
2359 .sendmsg = tcp_sendmsg,
2360 .sendpage = tcp_sendpage,
2361 .backlog_rcv = tcp_v4_do_rcv,
2362 .release_cb = tcp_release_cb,
2364 .unhash = inet_unhash,
2365 .get_port = inet_csk_get_port,
2366 .enter_memory_pressure = tcp_enter_memory_pressure,
2367 .stream_memory_free = tcp_stream_memory_free,
2368 .sockets_allocated = &tcp_sockets_allocated,
2369 .orphan_count = &tcp_orphan_count,
2370 .memory_allocated = &tcp_memory_allocated,
2371 .memory_pressure = &tcp_memory_pressure,
2372 .sysctl_mem = sysctl_tcp_mem,
2373 .sysctl_wmem = sysctl_tcp_wmem,
2374 .sysctl_rmem = sysctl_tcp_rmem,
2375 .max_header = MAX_TCP_HEADER,
2376 .obj_size = sizeof(struct tcp_sock),
2377 .slab_flags = SLAB_DESTROY_BY_RCU,
2378 .twsk_prot = &tcp_timewait_sock_ops,
2379 .rsk_prot = &tcp_request_sock_ops,
2380 .h.hashinfo = &tcp_hashinfo,
2381 .no_autobind = true,
2382 #ifdef CONFIG_COMPAT
2383 .compat_setsockopt = compat_tcp_setsockopt,
2384 .compat_getsockopt = compat_tcp_getsockopt,
2386 .diag_destroy = tcp_abort,
2388 EXPORT_SYMBOL(tcp_prot);
2390 static void __net_exit tcp_sk_exit(struct net *net)
2394 for_each_possible_cpu(cpu)
2395 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2396 free_percpu(net->ipv4.tcp_sk);
2399 static int __net_init tcp_sk_init(struct net *net)
2403 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2404 if (!net->ipv4.tcp_sk)
2407 for_each_possible_cpu(cpu) {
2410 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2414 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2415 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2418 net->ipv4.sysctl_tcp_ecn = 2;
2419 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2421 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2422 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2423 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2425 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2426 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2427 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2429 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2430 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2431 net->ipv4.sysctl_tcp_syncookies = 1;
2432 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2433 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2434 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2435 net->ipv4.sysctl_tcp_orphan_retries = 0;
2436 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2437 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2446 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2448 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2451 static struct pernet_operations __net_initdata tcp_sk_ops = {
2452 .init = tcp_sk_init,
2453 .exit = tcp_sk_exit,
2454 .exit_batch = tcp_sk_exit_batch,
2457 void __init tcp_v4_init(void)
2459 inet_hashinfo_init(&tcp_hashinfo);
2460 if (register_pernet_subsys(&tcp_sk_ops))
2461 panic("Failed to create the TCP control socket.\n");