2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Implementation of the Transmission Control Protocol(TCP).
8 * IPv4 specific functions
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
16 * See tcp.c for author information
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
53 #define pr_fmt(fmt) "TCP: " fmt
55 #include <linux/bottom_half.h>
56 #include <linux/types.h>
57 #include <linux/fcntl.h>
58 #include <linux/module.h>
59 #include <linux/random.h>
60 #include <linux/cache.h>
61 #include <linux/jhash.h>
62 #include <linux/init.h>
63 #include <linux/times.h>
64 #include <linux/slab.h>
66 #include <net/net_namespace.h>
68 #include <net/inet_hashtables.h>
70 #include <net/transp_v6.h>
72 #include <net/inet_common.h>
73 #include <net/timewait_sock.h>
75 #include <net/secure_seq.h>
76 #include <net/busy_poll.h>
78 #include <linux/inet.h>
79 #include <linux/ipv6.h>
80 #include <linux/stddef.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
84 #include <linux/crypto.h>
85 #include <linux/scatterlist.h>
87 int sysctl_tcp_tw_reuse __read_mostly;
88 int sysctl_tcp_low_latency __read_mostly;
89 EXPORT_SYMBOL(sysctl_tcp_low_latency);
91 #ifdef CONFIG_TCP_MD5SIG
92 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
93 __be32 daddr, __be32 saddr, const struct tcphdr *th);
96 struct inet_hashinfo tcp_hashinfo;
97 EXPORT_SYMBOL(tcp_hashinfo);
99 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
101 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
104 tcp_hdr(skb)->source);
107 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
109 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
110 struct tcp_sock *tp = tcp_sk(sk);
112 /* With PAWS, it is safe from the viewpoint
113 of data integrity. Even without PAWS it is safe provided sequence
114 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
116 Actually, the idea is close to VJ's one, only timestamp cache is
117 held not per host, but per port pair and TW bucket is used as state
120 If TW bucket has been already destroyed we fall back to VJ's scheme
121 and use initial timestamp retrieved from peer table.
123 if (tcptw->tw_ts_recent_stamp &&
124 (!twp || (sysctl_tcp_tw_reuse &&
125 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
126 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
127 if (tp->write_seq == 0)
129 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
130 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
137 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
139 /* This will initiate an outgoing connection. */
140 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
142 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
143 struct inet_sock *inet = inet_sk(sk);
144 struct tcp_sock *tp = tcp_sk(sk);
145 __be16 orig_sport, orig_dport;
146 __be32 daddr, nexthop;
150 struct ip_options_rcu *inet_opt;
152 if (addr_len < sizeof(struct sockaddr_in))
155 if (usin->sin_family != AF_INET)
156 return -EAFNOSUPPORT;
158 nexthop = daddr = usin->sin_addr.s_addr;
159 inet_opt = rcu_dereference_protected(inet->inet_opt,
160 sock_owned_by_user(sk));
161 if (inet_opt && inet_opt->opt.srr) {
164 nexthop = inet_opt->opt.faddr;
167 orig_sport = inet->inet_sport;
168 orig_dport = usin->sin_port;
169 fl4 = &inet->cork.fl.u.ip4;
170 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
171 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
173 orig_sport, orig_dport, sk);
176 if (err == -ENETUNREACH)
177 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
181 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
186 if (!inet_opt || !inet_opt->opt.srr)
189 if (!inet->inet_saddr)
190 inet->inet_saddr = fl4->saddr;
191 sk_rcv_saddr_set(sk, inet->inet_saddr);
193 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
194 /* Reset inherited state */
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
197 if (likely(!tp->repair))
201 if (tcp_death_row.sysctl_tw_recycle &&
202 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
203 tcp_fetch_timewait_stamp(sk, &rt->dst);
205 inet->inet_dport = usin->sin_port;
206 sk_daddr_set(sk, daddr);
208 inet_csk(sk)->icsk_ext_hdr_len = 0;
210 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
212 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
214 /* Socket identity is still unknown (sport may be zero).
215 * However we set state to SYN-SENT and not releasing socket
216 * lock select source port, enter ourselves into the hash tables and
217 * complete initialization after this.
219 tcp_set_state(sk, TCP_SYN_SENT);
220 err = inet_hash_connect(&tcp_death_row, sk);
226 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
227 inet->inet_sport, inet->inet_dport, sk);
233 /* OK, now commit destination to socket. */
234 sk->sk_gso_type = SKB_GSO_TCPV4;
235 sk_setup_caps(sk, &rt->dst);
237 if (!tp->write_seq && likely(!tp->repair))
238 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
243 inet->inet_id = tp->write_seq ^ jiffies;
245 err = tcp_connect(sk);
255 * This unhashes the socket and releases the local port,
258 tcp_set_state(sk, TCP_CLOSE);
260 sk->sk_route_caps = 0;
261 inet->inet_dport = 0;
264 EXPORT_SYMBOL(tcp_v4_connect);
267 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
268 * It can be called through tcp_release_cb() if socket was owned by user
269 * at the time tcp_v4_err() was called to handle ICMP message.
271 void tcp_v4_mtu_reduced(struct sock *sk)
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
275 u32 mtu = tcp_sk(sk)->mtu_info;
277 dst = inet_csk_update_pmtu(sk, mtu);
281 /* Something is about to be wrong... Remember soft error
282 * for the case, if this connection will not able to recover.
284 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
285 sk->sk_err_soft = EMSGSIZE;
289 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
290 ip_sk_accept_pmtu(sk) &&
291 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
292 tcp_sync_mss(sk, mtu);
294 /* Resend the TCP packet because it's
295 * clear that the old packet has been
296 * dropped. This is the new "fast" path mtu
299 tcp_simple_retransmit(sk);
300 } /* else let the usual retransmit timer handle it */
302 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
304 static void do_redirect(struct sk_buff *skb, struct sock *sk)
306 struct dst_entry *dst = __sk_dst_check(sk, 0);
309 dst->ops->redirect(dst, sk, skb);
313 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
314 void tcp_req_err(struct sock *sk, u32 seq, bool abort)
316 struct request_sock *req = inet_reqsk(sk);
317 struct net *net = sock_net(sk);
319 /* ICMPs are not backlogged, hence we cannot get
320 * an established socket here.
324 if (seq != tcp_rsk(req)->snt_isn) {
325 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
328 * Still in SYN_RECV, just remove it silently.
329 * There is no good way to pass the error to the newly
330 * created socket, and POSIX does not want network
331 * errors returned from accept().
333 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
334 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS);
338 EXPORT_SYMBOL(tcp_req_err);
341 * This routine is called by the ICMP module when it gets some
342 * sort of error condition. If err < 0 then the socket should
343 * be closed and the error returned to the user. If err > 0
344 * it's just the icmp type << 8 | icmp code. After adjustment
345 * header points to the first 8 bytes of the tcp header. We need
346 * to find the appropriate port.
348 * The locking strategy used here is very "optimistic". When
349 * someone else accesses the socket the ICMP is just dropped
350 * and for some paths there is no check at all.
351 * A more general error queue to queue errors for later handling
352 * is probably better.
356 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
358 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
359 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
360 struct inet_connection_sock *icsk;
362 struct inet_sock *inet;
363 const int type = icmp_hdr(icmp_skb)->type;
364 const int code = icmp_hdr(icmp_skb)->code;
367 struct request_sock *fastopen;
371 struct net *net = dev_net(icmp_skb->dev);
373 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
374 th->dest, iph->saddr, ntohs(th->source),
377 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
380 if (sk->sk_state == TCP_TIME_WAIT) {
381 inet_twsk_put(inet_twsk(sk));
384 seq = ntohl(th->seq);
385 if (sk->sk_state == TCP_NEW_SYN_RECV)
386 return tcp_req_err(sk, seq,
387 type == ICMP_PARAMETERPROB ||
388 type == ICMP_TIME_EXCEEDED ||
389 (type == ICMP_DEST_UNREACH &&
390 (code == ICMP_NET_UNREACH ||
391 code == ICMP_HOST_UNREACH)));
394 /* If too many ICMPs get dropped on busy
395 * servers this needs to be solved differently.
396 * We do take care of PMTU discovery (RFC1191) special case :
397 * we can receive locally generated ICMP messages while socket is held.
399 if (sock_owned_by_user(sk)) {
400 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
401 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
403 if (sk->sk_state == TCP_CLOSE)
406 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
407 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
413 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
414 fastopen = tp->fastopen_rsk;
415 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
416 if (sk->sk_state != TCP_LISTEN &&
417 !between(seq, snd_una, tp->snd_nxt)) {
418 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
424 do_redirect(icmp_skb, sk);
426 case ICMP_SOURCE_QUENCH:
427 /* Just silently ignore these. */
429 case ICMP_PARAMETERPROB:
432 case ICMP_DEST_UNREACH:
433 if (code > NR_ICMP_UNREACH)
436 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
437 /* We are not interested in TCP_LISTEN and open_requests
438 * (SYN-ACKs send out by Linux are always <576bytes so
439 * they should go through unfragmented).
441 if (sk->sk_state == TCP_LISTEN)
445 if (!sock_owned_by_user(sk)) {
446 tcp_v4_mtu_reduced(sk);
448 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
454 err = icmp_err_convert[code].errno;
455 /* check if icmp_skb allows revert of backoff
456 * (see draft-zimmermann-tcp-lcd) */
457 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
459 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
460 !icsk->icsk_backoff || fastopen)
463 if (sock_owned_by_user(sk))
466 icsk->icsk_backoff--;
467 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
469 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
471 skb = tcp_write_queue_head(sk);
474 remaining = icsk->icsk_rto -
476 tcp_time_stamp - tcp_skb_timestamp(skb));
479 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
480 remaining, TCP_RTO_MAX);
482 /* RTO revert clocked out retransmission.
483 * Will retransmit now */
484 tcp_retransmit_timer(sk);
488 case ICMP_TIME_EXCEEDED:
495 switch (sk->sk_state) {
498 /* Only in fast or simultaneous open. If a fast open socket is
499 * is already accepted it is treated as a connected one below.
501 if (fastopen && !fastopen->sk)
504 if (!sock_owned_by_user(sk)) {
507 sk->sk_error_report(sk);
511 sk->sk_err_soft = err;
516 /* If we've already connected we will keep trying
517 * until we time out, or the user gives up.
519 * rfc1122 4.2.3.9 allows to consider as hard errors
520 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
521 * but it is obsoleted by pmtu discovery).
523 * Note, that in modern internet, where routing is unreliable
524 * and in each dark corner broken firewalls sit, sending random
525 * errors ordered by their masters even this two messages finally lose
526 * their original sense (even Linux sends invalid PORT_UNREACHs)
528 * Now we are in compliance with RFCs.
533 if (!sock_owned_by_user(sk) && inet->recverr) {
535 sk->sk_error_report(sk);
536 } else { /* Only an error on timeout */
537 sk->sk_err_soft = err;
545 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
547 struct tcphdr *th = tcp_hdr(skb);
549 if (skb->ip_summed == CHECKSUM_PARTIAL) {
550 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
551 skb->csum_start = skb_transport_header(skb) - skb->head;
552 skb->csum_offset = offsetof(struct tcphdr, check);
554 th->check = tcp_v4_check(skb->len, saddr, daddr,
561 /* This routine computes an IPv4 TCP checksum. */
562 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
564 const struct inet_sock *inet = inet_sk(sk);
566 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
568 EXPORT_SYMBOL(tcp_v4_send_check);
571 * This routine will send an RST to the other tcp.
573 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
575 * Answer: if a packet caused RST, it is not for a socket
576 * existing in our system, if it is matched to a socket,
577 * it is just duplicate segment or bug in other side's TCP.
578 * So that we build reply only basing on parameters
579 * arrived with segment.
580 * Exception: precedence violation. We do not implement it in any case.
583 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
585 const struct tcphdr *th = tcp_hdr(skb);
588 #ifdef CONFIG_TCP_MD5SIG
589 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
592 struct ip_reply_arg arg;
593 #ifdef CONFIG_TCP_MD5SIG
594 struct tcp_md5sig_key *key = NULL;
595 const __u8 *hash_location = NULL;
596 unsigned char newhash[16];
598 struct sock *sk1 = NULL;
602 /* Never send a reset in response to a reset. */
606 /* If sk not NULL, it means we did a successful lookup and incoming
607 * route had to be correct. prequeue might have dropped our dst.
609 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
612 /* Swap the send and the receive. */
613 memset(&rep, 0, sizeof(rep));
614 rep.th.dest = th->source;
615 rep.th.source = th->dest;
616 rep.th.doff = sizeof(struct tcphdr) / 4;
620 rep.th.seq = th->ack_seq;
623 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
624 skb->len - (th->doff << 2));
627 memset(&arg, 0, sizeof(arg));
628 arg.iov[0].iov_base = (unsigned char *)&rep;
629 arg.iov[0].iov_len = sizeof(rep.th);
631 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
632 #ifdef CONFIG_TCP_MD5SIG
633 hash_location = tcp_parse_md5sig_option(th);
634 if (sk && sk_fullsock(sk)) {
635 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
636 &ip_hdr(skb)->saddr, AF_INET);
637 } else if (hash_location) {
639 * active side is lost. Try to find listening socket through
640 * source port, and then find md5 key through listening socket.
641 * we are not loose security here:
642 * Incoming packet is checked with md5 hash with finding key,
643 * no RST generated if md5 hash doesn't match.
645 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
647 th->source, ip_hdr(skb)->daddr,
648 ntohs(th->source), inet_iif(skb));
649 /* don't send rst if it can't find key */
653 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
654 &ip_hdr(skb)->saddr, AF_INET);
658 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
659 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;
694 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
695 skb, &TCP_SKB_CB(skb)->header.h4.opt,
696 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
697 &arg, arg.iov[0].iov_len);
699 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
700 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
702 #ifdef CONFIG_TCP_MD5SIG
711 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
712 outside socket context is ugly, certainly. What can I do?
715 static void tcp_v4_send_ack(struct net *net,
716 struct sk_buff *skb, u32 seq, u32 ack,
717 u32 win, u32 tsval, u32 tsecr, int oif,
718 struct tcp_md5sig_key *key,
719 int reply_flags, u8 tos)
721 const struct tcphdr *th = tcp_hdr(skb);
724 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
725 #ifdef CONFIG_TCP_MD5SIG
726 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
730 struct ip_reply_arg arg;
732 memset(&rep.th, 0, sizeof(struct tcphdr));
733 memset(&arg, 0, sizeof(arg));
735 arg.iov[0].iov_base = (unsigned char *)&rep;
736 arg.iov[0].iov_len = sizeof(rep.th);
738 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
739 (TCPOPT_TIMESTAMP << 8) |
741 rep.opt[1] = htonl(tsval);
742 rep.opt[2] = htonl(tsecr);
743 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
746 /* Swap the send and the receive. */
747 rep.th.dest = th->source;
748 rep.th.source = th->dest;
749 rep.th.doff = arg.iov[0].iov_len / 4;
750 rep.th.seq = htonl(seq);
751 rep.th.ack_seq = htonl(ack);
753 rep.th.window = htons(win);
755 #ifdef CONFIG_TCP_MD5SIG
757 int offset = (tsecr) ? 3 : 0;
759 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
761 (TCPOPT_MD5SIG << 8) |
763 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
764 rep.th.doff = arg.iov[0].iov_len/4;
766 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
767 key, ip_hdr(skb)->saddr,
768 ip_hdr(skb)->daddr, &rep.th);
771 arg.flags = reply_flags;
772 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
773 ip_hdr(skb)->saddr, /* XXX */
774 arg.iov[0].iov_len, IPPROTO_TCP, 0);
775 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
777 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_BH(net, TCP_MIB_OUTSEGS);
787 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
789 struct inet_timewait_sock *tw = inet_twsk(sk);
790 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
792 tcp_v4_send_ack(sock_net(sk), skb,
793 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
794 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
795 tcp_time_stamp + tcptw->tw_ts_offset,
798 tcp_twsk_md5_key(tcptw),
799 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
806 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
807 struct request_sock *req)
809 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
810 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
812 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
815 tcp_v4_send_ack(sock_net(sk), skb, seq,
816 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
820 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
822 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
827 * Send a SYN-ACK after having received a SYN.
828 * This still operates on a request_sock only, not on a big
831 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
833 struct request_sock *req,
834 struct tcp_fastopen_cookie *foc,
837 const struct inet_request_sock *ireq = inet_rsk(req);
842 /* First, grab a route. */
843 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
846 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
849 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
851 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
854 err = net_xmit_eval(err);
861 * IPv4 request_sock destructor.
863 static void tcp_v4_reqsk_destructor(struct request_sock *req)
865 kfree(inet_rsk(req)->opt);
868 #ifdef CONFIG_TCP_MD5SIG
870 * RFC2385 MD5 checksumming requires a mapping of
871 * IP address->MD5 Key.
872 * We need to maintain these in the sk structure.
875 /* Find the Key structure for an address. */
876 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
877 const union tcp_md5_addr *addr,
880 const struct tcp_sock *tp = tcp_sk(sk);
881 struct tcp_md5sig_key *key;
882 unsigned int size = sizeof(struct in_addr);
883 const struct tcp_md5sig_info *md5sig;
885 /* caller either holds rcu_read_lock() or socket lock */
886 md5sig = rcu_dereference_check(tp->md5sig_info,
887 sock_owned_by_user(sk) ||
888 lockdep_is_held((spinlock_t *)&sk->sk_lock.slock));
891 #if IS_ENABLED(CONFIG_IPV6)
892 if (family == AF_INET6)
893 size = sizeof(struct in6_addr);
895 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
896 if (key->family != family)
898 if (!memcmp(&key->addr, addr, size))
903 EXPORT_SYMBOL(tcp_md5_do_lookup);
905 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
906 const struct sock *addr_sk)
908 const union tcp_md5_addr *addr;
910 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
911 return tcp_md5_do_lookup(sk, addr, AF_INET);
913 EXPORT_SYMBOL(tcp_v4_md5_lookup);
915 /* This can be called on a newly created socket, from other files */
916 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
917 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
919 /* Add Key to the list */
920 struct tcp_md5sig_key *key;
921 struct tcp_sock *tp = tcp_sk(sk);
922 struct tcp_md5sig_info *md5sig;
924 key = tcp_md5_do_lookup(sk, addr, family);
926 /* Pre-existing entry - just update that one. */
927 memcpy(key->key, newkey, newkeylen);
928 key->keylen = newkeylen;
932 md5sig = rcu_dereference_protected(tp->md5sig_info,
933 sock_owned_by_user(sk) ||
934 lockdep_is_held(&sk->sk_lock.slock));
936 md5sig = kmalloc(sizeof(*md5sig), gfp);
940 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
941 INIT_HLIST_HEAD(&md5sig->head);
942 rcu_assign_pointer(tp->md5sig_info, md5sig);
945 key = sock_kmalloc(sk, sizeof(*key), gfp);
948 if (!tcp_alloc_md5sig_pool()) {
949 sock_kfree_s(sk, key, sizeof(*key));
953 memcpy(key->key, newkey, newkeylen);
954 key->keylen = newkeylen;
955 key->family = family;
956 memcpy(&key->addr, addr,
957 (family == AF_INET6) ? sizeof(struct in6_addr) :
958 sizeof(struct in_addr));
959 hlist_add_head_rcu(&key->node, &md5sig->head);
962 EXPORT_SYMBOL(tcp_md5_do_add);
964 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
966 struct tcp_md5sig_key *key;
968 key = tcp_md5_do_lookup(sk, addr, family);
971 hlist_del_rcu(&key->node);
972 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
976 EXPORT_SYMBOL(tcp_md5_do_del);
978 static void tcp_clear_md5_list(struct sock *sk)
980 struct tcp_sock *tp = tcp_sk(sk);
981 struct tcp_md5sig_key *key;
982 struct hlist_node *n;
983 struct tcp_md5sig_info *md5sig;
985 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
987 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
988 hlist_del_rcu(&key->node);
989 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
994 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
997 struct tcp_md5sig cmd;
998 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1000 if (optlen < sizeof(cmd))
1003 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1006 if (sin->sin_family != AF_INET)
1009 if (!cmd.tcpm_keylen)
1010 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1013 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1016 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1017 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1021 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1022 __be32 daddr, __be32 saddr, int nbytes)
1024 struct tcp4_pseudohdr *bp;
1025 struct scatterlist sg;
1027 bp = &hp->md5_blk.ip4;
1030 * 1. the TCP pseudo-header (in the order: source IP address,
1031 * destination IP address, zero-padded protocol number, and
1037 bp->protocol = IPPROTO_TCP;
1038 bp->len = cpu_to_be16(nbytes);
1040 sg_init_one(&sg, bp, sizeof(*bp));
1041 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1044 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1045 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1047 struct tcp_md5sig_pool *hp;
1048 struct hash_desc *desc;
1050 hp = tcp_get_md5sig_pool();
1052 goto clear_hash_noput;
1053 desc = &hp->md5_desc;
1055 if (crypto_hash_init(desc))
1057 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1059 if (tcp_md5_hash_header(hp, th))
1061 if (tcp_md5_hash_key(hp, key))
1063 if (crypto_hash_final(desc, md5_hash))
1066 tcp_put_md5sig_pool();
1070 tcp_put_md5sig_pool();
1072 memset(md5_hash, 0, 16);
1076 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1077 const struct sock *sk,
1078 const struct sk_buff *skb)
1080 struct tcp_md5sig_pool *hp;
1081 struct hash_desc *desc;
1082 const struct tcphdr *th = tcp_hdr(skb);
1083 __be32 saddr, daddr;
1085 if (sk) { /* valid for establish/request sockets */
1086 saddr = sk->sk_rcv_saddr;
1087 daddr = sk->sk_daddr;
1089 const struct iphdr *iph = ip_hdr(skb);
1094 hp = tcp_get_md5sig_pool();
1096 goto clear_hash_noput;
1097 desc = &hp->md5_desc;
1099 if (crypto_hash_init(desc))
1102 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1104 if (tcp_md5_hash_header(hp, th))
1106 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1108 if (tcp_md5_hash_key(hp, key))
1110 if (crypto_hash_final(desc, md5_hash))
1113 tcp_put_md5sig_pool();
1117 tcp_put_md5sig_pool();
1119 memset(md5_hash, 0, 16);
1122 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1126 /* Called with rcu_read_lock() */
1127 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1128 const struct sk_buff *skb)
1130 #ifdef CONFIG_TCP_MD5SIG
1132 * This gets called for each TCP segment that arrives
1133 * so we want to be efficient.
1134 * We have 3 drop cases:
1135 * o No MD5 hash and one expected.
1136 * o MD5 hash and we're not expecting one.
1137 * o MD5 hash and its wrong.
1139 const __u8 *hash_location = NULL;
1140 struct tcp_md5sig_key *hash_expected;
1141 const struct iphdr *iph = ip_hdr(skb);
1142 const struct tcphdr *th = tcp_hdr(skb);
1144 unsigned char newhash[16];
1146 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1148 hash_location = tcp_parse_md5sig_option(th);
1150 /* We've parsed the options - do we have a hash? */
1151 if (!hash_expected && !hash_location)
1154 if (hash_expected && !hash_location) {
1155 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1159 if (!hash_expected && hash_location) {
1160 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1164 /* Okay, so this is hash_expected and hash_location -
1165 * so we need to calculate the checksum.
1167 genhash = tcp_v4_md5_hash_skb(newhash,
1171 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1172 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1173 &iph->saddr, ntohs(th->source),
1174 &iph->daddr, ntohs(th->dest),
1175 genhash ? " tcp_v4_calc_md5_hash failed"
1184 static void tcp_v4_init_req(struct request_sock *req,
1185 const struct sock *sk_listener,
1186 struct sk_buff *skb)
1188 struct inet_request_sock *ireq = inet_rsk(req);
1190 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1191 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1192 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
1193 ireq->opt = tcp_v4_save_options(skb);
1196 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1198 const struct request_sock *req,
1201 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1204 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1213 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1215 .obj_size = sizeof(struct tcp_request_sock),
1216 .rtx_syn_ack = tcp_rtx_synack,
1217 .send_ack = tcp_v4_reqsk_send_ack,
1218 .destructor = tcp_v4_reqsk_destructor,
1219 .send_reset = tcp_v4_send_reset,
1220 .syn_ack_timeout = tcp_syn_ack_timeout,
1223 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1224 .mss_clamp = TCP_MSS_DEFAULT,
1225 #ifdef CONFIG_TCP_MD5SIG
1226 .req_md5_lookup = tcp_v4_md5_lookup,
1227 .calc_md5_hash = tcp_v4_md5_hash_skb,
1229 .init_req = tcp_v4_init_req,
1230 #ifdef CONFIG_SYN_COOKIES
1231 .cookie_init_seq = cookie_v4_init_sequence,
1233 .route_req = tcp_v4_route_req,
1234 .init_seq = tcp_v4_init_sequence,
1235 .send_synack = tcp_v4_send_synack,
1238 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1240 /* Never answer to SYNs send to broadcast or multicast */
1241 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1244 return tcp_conn_request(&tcp_request_sock_ops,
1245 &tcp_request_sock_ipv4_ops, sk, skb);
1248 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1251 EXPORT_SYMBOL(tcp_v4_conn_request);
1255 * The three way handshake has completed - we got a valid synack -
1256 * now create the new socket.
1258 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1259 struct request_sock *req,
1260 struct dst_entry *dst,
1261 struct request_sock *req_unhash,
1264 struct inet_request_sock *ireq;
1265 struct inet_sock *newinet;
1266 struct tcp_sock *newtp;
1268 #ifdef CONFIG_TCP_MD5SIG
1269 struct tcp_md5sig_key *key;
1271 struct ip_options_rcu *inet_opt;
1273 if (sk_acceptq_is_full(sk))
1276 newsk = tcp_create_openreq_child(sk, req, skb);
1280 newsk->sk_gso_type = SKB_GSO_TCPV4;
1281 inet_sk_rx_dst_set(newsk, skb);
1283 newtp = tcp_sk(newsk);
1284 newinet = inet_sk(newsk);
1285 ireq = inet_rsk(req);
1286 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1287 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1288 newsk->sk_bound_dev_if = ireq->ir_iif;
1289 newinet->inet_saddr = ireq->ir_loc_addr;
1290 inet_opt = ireq->opt;
1291 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1293 newinet->mc_index = inet_iif(skb);
1294 newinet->mc_ttl = ip_hdr(skb)->ttl;
1295 newinet->rcv_tos = ip_hdr(skb)->tos;
1296 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1298 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1299 newinet->inet_id = newtp->write_seq ^ jiffies;
1302 dst = inet_csk_route_child_sock(sk, newsk, req);
1306 /* syncookie case : see end of cookie_v4_check() */
1308 sk_setup_caps(newsk, dst);
1310 tcp_ca_openreq_child(newsk, dst);
1312 tcp_sync_mss(newsk, dst_mtu(dst));
1313 newtp->advmss = dst_metric_advmss(dst);
1314 if (tcp_sk(sk)->rx_opt.user_mss &&
1315 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1316 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1318 tcp_initialize_rcv_mss(newsk);
1320 #ifdef CONFIG_TCP_MD5SIG
1321 /* Copy over the MD5 key from the original socket */
1322 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1326 * We're using one, so create a matching key
1327 * on the newsk structure. If we fail to get
1328 * memory, then we end up not copying the key
1331 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1332 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1333 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1337 if (__inet_inherit_port(sk, newsk) < 0)
1339 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1341 tcp_move_syn(newtp, req);
1346 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1350 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1353 inet_csk_prepare_forced_close(newsk);
1357 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1359 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1361 #ifdef CONFIG_SYN_COOKIES
1362 const struct tcphdr *th = tcp_hdr(skb);
1365 sk = cookie_v4_check(sk, skb);
1370 /* The socket must have it's spinlock held when we get
1371 * here, unless it is a TCP_LISTEN socket.
1373 * We have a potential double-lock case here, so even when
1374 * doing backlog processing we use the BH locking scheme.
1375 * This is because we cannot sleep with the original spinlock
1378 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1382 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1383 struct dst_entry *dst = sk->sk_rx_dst;
1385 sock_rps_save_rxhash(sk, skb);
1386 sk_mark_napi_id(sk, skb);
1388 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1389 !dst->ops->check(dst, 0)) {
1391 sk->sk_rx_dst = NULL;
1394 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1398 if (tcp_checksum_complete(skb))
1401 if (sk->sk_state == TCP_LISTEN) {
1402 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1407 sock_rps_save_rxhash(nsk, skb);
1408 sk_mark_napi_id(nsk, skb);
1409 if (tcp_child_process(sk, nsk, skb)) {
1416 sock_rps_save_rxhash(sk, skb);
1418 if (tcp_rcv_state_process(sk, skb)) {
1425 tcp_v4_send_reset(rsk, skb);
1428 /* Be careful here. If this function gets more complicated and
1429 * gcc suffers from register pressure on the x86, sk (in %ebx)
1430 * might be destroyed here. This current version compiles correctly,
1431 * but you have been warned.
1436 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1437 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1440 EXPORT_SYMBOL(tcp_v4_do_rcv);
1442 void tcp_v4_early_demux(struct sk_buff *skb)
1444 const struct iphdr *iph;
1445 const struct tcphdr *th;
1448 if (skb->pkt_type != PACKET_HOST)
1451 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1457 if (th->doff < sizeof(struct tcphdr) / 4)
1460 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1461 iph->saddr, th->source,
1462 iph->daddr, ntohs(th->dest),
1466 skb->destructor = sock_edemux;
1467 if (sk_fullsock(sk)) {
1468 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1471 dst = dst_check(dst, 0);
1473 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1474 skb_dst_set_noref(skb, dst);
1479 /* Packet is added to VJ-style prequeue for processing in process
1480 * context, if a reader task is waiting. Apparently, this exciting
1481 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1482 * failed somewhere. Latency? Burstiness? Well, at least now we will
1483 * see, why it failed. 8)8) --ANK
1486 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1488 struct tcp_sock *tp = tcp_sk(sk);
1490 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1493 if (skb->len <= tcp_hdrlen(skb) &&
1494 skb_queue_len(&tp->ucopy.prequeue) == 0)
1497 /* Before escaping RCU protected region, we need to take care of skb
1498 * dst. Prequeue is only enabled for established sockets.
1499 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1500 * Instead of doing full sk_rx_dst validity here, let's perform
1501 * an optimistic check.
1503 if (likely(sk->sk_rx_dst))
1506 skb_dst_force_safe(skb);
1508 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1509 tp->ucopy.memory += skb->truesize;
1510 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1511 struct sk_buff *skb1;
1513 BUG_ON(sock_owned_by_user(sk));
1515 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1516 sk_backlog_rcv(sk, skb1);
1517 NET_INC_STATS_BH(sock_net(sk),
1518 LINUX_MIB_TCPPREQUEUEDROPPED);
1521 tp->ucopy.memory = 0;
1522 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1523 wake_up_interruptible_sync_poll(sk_sleep(sk),
1524 POLLIN | POLLRDNORM | POLLRDBAND);
1525 if (!inet_csk_ack_scheduled(sk))
1526 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1527 (3 * tcp_rto_min(sk)) / 4,
1532 EXPORT_SYMBOL(tcp_prequeue);
1538 int tcp_v4_rcv(struct sk_buff *skb)
1540 const struct iphdr *iph;
1541 const struct tcphdr *th;
1544 struct net *net = dev_net(skb->dev);
1546 if (skb->pkt_type != PACKET_HOST)
1549 /* Count it even if it's bad */
1550 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1552 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1557 if (th->doff < sizeof(struct tcphdr) / 4)
1559 if (!pskb_may_pull(skb, th->doff * 4))
1562 /* An explanation is required here, I think.
1563 * Packet length and doff are validated by header prediction,
1564 * provided case of th->doff==0 is eliminated.
1565 * So, we defer the checks. */
1567 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1572 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1573 * barrier() makes sure compiler wont play fool^Waliasing games.
1575 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1576 sizeof(struct inet_skb_parm));
1579 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1580 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1581 skb->len - th->doff * 4);
1582 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1583 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1584 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1585 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1586 TCP_SKB_CB(skb)->sacked = 0;
1589 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1595 if (sk->sk_state == TCP_TIME_WAIT)
1598 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1599 struct request_sock *req = inet_reqsk(sk);
1602 sk = req->rsk_listener;
1603 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1607 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1608 inet_csk_reqsk_queue_drop_and_put(sk, req);
1612 nsk = tcp_check_req(sk, skb, req, false);
1615 goto discard_and_relse;
1619 } else if (tcp_child_process(sk, nsk, skb)) {
1620 tcp_v4_send_reset(nsk, skb);
1621 goto discard_and_relse;
1627 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1628 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1629 goto discard_and_relse;
1632 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1633 goto discard_and_relse;
1635 if (tcp_v4_inbound_md5_hash(sk, skb))
1636 goto discard_and_relse;
1640 if (sk_filter(sk, skb))
1641 goto discard_and_relse;
1645 if (sk->sk_state == TCP_LISTEN) {
1646 ret = tcp_v4_do_rcv(sk, skb);
1647 goto put_and_return;
1650 sk_incoming_cpu_update(sk);
1652 bh_lock_sock_nested(sk);
1653 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1655 if (!sock_owned_by_user(sk)) {
1656 if (!tcp_prequeue(sk, skb))
1657 ret = tcp_v4_do_rcv(sk, skb);
1658 } else if (unlikely(sk_add_backlog(sk, skb,
1659 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1661 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1662 goto discard_and_relse;
1672 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1675 if (tcp_checksum_complete(skb)) {
1677 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1679 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1681 tcp_v4_send_reset(NULL, skb);
1685 /* Discard frame. */
1694 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1695 inet_twsk_put(inet_twsk(sk));
1699 if (tcp_checksum_complete(skb)) {
1700 inet_twsk_put(inet_twsk(sk));
1703 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1705 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1708 iph->saddr, th->source,
1709 iph->daddr, th->dest,
1712 inet_twsk_deschedule_put(inet_twsk(sk));
1716 /* Fall through to ACK */
1719 tcp_v4_timewait_ack(sk, skb);
1722 tcp_v4_send_reset(sk, skb);
1723 inet_twsk_deschedule_put(inet_twsk(sk));
1725 case TCP_TW_SUCCESS:;
1730 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1731 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1732 .twsk_unique = tcp_twsk_unique,
1733 .twsk_destructor= tcp_twsk_destructor,
1736 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1738 struct dst_entry *dst = skb_dst(skb);
1740 if (dst && dst_hold_safe(dst)) {
1741 sk->sk_rx_dst = dst;
1742 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1745 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1747 const struct inet_connection_sock_af_ops ipv4_specific = {
1748 .queue_xmit = ip_queue_xmit,
1749 .send_check = tcp_v4_send_check,
1750 .rebuild_header = inet_sk_rebuild_header,
1751 .sk_rx_dst_set = inet_sk_rx_dst_set,
1752 .conn_request = tcp_v4_conn_request,
1753 .syn_recv_sock = tcp_v4_syn_recv_sock,
1754 .net_header_len = sizeof(struct iphdr),
1755 .setsockopt = ip_setsockopt,
1756 .getsockopt = ip_getsockopt,
1757 .addr2sockaddr = inet_csk_addr2sockaddr,
1758 .sockaddr_len = sizeof(struct sockaddr_in),
1759 .bind_conflict = inet_csk_bind_conflict,
1760 #ifdef CONFIG_COMPAT
1761 .compat_setsockopt = compat_ip_setsockopt,
1762 .compat_getsockopt = compat_ip_getsockopt,
1764 .mtu_reduced = tcp_v4_mtu_reduced,
1766 EXPORT_SYMBOL(ipv4_specific);
1768 #ifdef CONFIG_TCP_MD5SIG
1769 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1770 .md5_lookup = tcp_v4_md5_lookup,
1771 .calc_md5_hash = tcp_v4_md5_hash_skb,
1772 .md5_parse = tcp_v4_parse_md5_keys,
1776 /* NOTE: A lot of things set to zero explicitly by call to
1777 * sk_alloc() so need not be done here.
1779 static int tcp_v4_init_sock(struct sock *sk)
1781 struct inet_connection_sock *icsk = inet_csk(sk);
1785 icsk->icsk_af_ops = &ipv4_specific;
1787 #ifdef CONFIG_TCP_MD5SIG
1788 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1794 void tcp_v4_destroy_sock(struct sock *sk)
1796 struct tcp_sock *tp = tcp_sk(sk);
1798 tcp_clear_xmit_timers(sk);
1800 tcp_cleanup_congestion_control(sk);
1802 /* Cleanup up the write buffer. */
1803 tcp_write_queue_purge(sk);
1805 /* Cleans up our, hopefully empty, out_of_order_queue. */
1806 __skb_queue_purge(&tp->out_of_order_queue);
1808 #ifdef CONFIG_TCP_MD5SIG
1809 /* Clean up the MD5 key list, if any */
1810 if (tp->md5sig_info) {
1811 tcp_clear_md5_list(sk);
1812 kfree_rcu(tp->md5sig_info, rcu);
1813 tp->md5sig_info = NULL;
1817 /* Clean prequeue, it must be empty really */
1818 __skb_queue_purge(&tp->ucopy.prequeue);
1820 /* Clean up a referenced TCP bind bucket. */
1821 if (inet_csk(sk)->icsk_bind_hash)
1824 BUG_ON(tp->fastopen_rsk);
1826 /* If socket is aborted during connect operation */
1827 tcp_free_fastopen_req(tp);
1828 tcp_saved_syn_free(tp);
1830 sk_sockets_allocated_dec(sk);
1832 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1833 sock_release_memcg(sk);
1835 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1837 #ifdef CONFIG_PROC_FS
1838 /* Proc filesystem TCP sock list dumping. */
1841 * Get next listener socket follow cur. If cur is NULL, get first socket
1842 * starting from bucket given in st->bucket; when st->bucket is zero the
1843 * very first socket in the hash table is returned.
1845 static void *listening_get_next(struct seq_file *seq, void *cur)
1847 struct inet_connection_sock *icsk;
1848 struct hlist_nulls_node *node;
1849 struct sock *sk = cur;
1850 struct inet_listen_hashbucket *ilb;
1851 struct tcp_iter_state *st = seq->private;
1852 struct net *net = seq_file_net(seq);
1855 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1856 spin_lock_bh(&ilb->lock);
1857 sk = sk_nulls_head(&ilb->head);
1861 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1865 sk = sk_nulls_next(sk);
1867 sk_nulls_for_each_from(sk, node) {
1868 if (!net_eq(sock_net(sk), net))
1870 if (sk->sk_family == st->family) {
1874 icsk = inet_csk(sk);
1876 spin_unlock_bh(&ilb->lock);
1878 if (++st->bucket < INET_LHTABLE_SIZE) {
1879 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1880 spin_lock_bh(&ilb->lock);
1881 sk = sk_nulls_head(&ilb->head);
1889 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1891 struct tcp_iter_state *st = seq->private;
1896 rc = listening_get_next(seq, NULL);
1898 while (rc && *pos) {
1899 rc = listening_get_next(seq, rc);
1905 static inline bool empty_bucket(const struct tcp_iter_state *st)
1907 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1911 * Get first established socket starting from bucket given in st->bucket.
1912 * If st->bucket is zero, the very first socket in the hash is returned.
1914 static void *established_get_first(struct seq_file *seq)
1916 struct tcp_iter_state *st = seq->private;
1917 struct net *net = seq_file_net(seq);
1921 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1923 struct hlist_nulls_node *node;
1924 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1926 /* Lockless fast path for the common case of empty buckets */
1927 if (empty_bucket(st))
1931 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1932 if (sk->sk_family != st->family ||
1933 !net_eq(sock_net(sk), net)) {
1939 spin_unlock_bh(lock);
1945 static void *established_get_next(struct seq_file *seq, void *cur)
1947 struct sock *sk = cur;
1948 struct hlist_nulls_node *node;
1949 struct tcp_iter_state *st = seq->private;
1950 struct net *net = seq_file_net(seq);
1955 sk = sk_nulls_next(sk);
1957 sk_nulls_for_each_from(sk, node) {
1958 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
1962 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1964 return established_get_first(seq);
1967 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1969 struct tcp_iter_state *st = seq->private;
1973 rc = established_get_first(seq);
1976 rc = established_get_next(seq, rc);
1982 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1985 struct tcp_iter_state *st = seq->private;
1987 st->state = TCP_SEQ_STATE_LISTENING;
1988 rc = listening_get_idx(seq, &pos);
1991 st->state = TCP_SEQ_STATE_ESTABLISHED;
1992 rc = established_get_idx(seq, pos);
1998 static void *tcp_seek_last_pos(struct seq_file *seq)
2000 struct tcp_iter_state *st = seq->private;
2001 int offset = st->offset;
2002 int orig_num = st->num;
2005 switch (st->state) {
2006 case TCP_SEQ_STATE_LISTENING:
2007 if (st->bucket >= INET_LHTABLE_SIZE)
2009 st->state = TCP_SEQ_STATE_LISTENING;
2010 rc = listening_get_next(seq, NULL);
2011 while (offset-- && rc)
2012 rc = listening_get_next(seq, rc);
2016 st->state = TCP_SEQ_STATE_ESTABLISHED;
2018 case TCP_SEQ_STATE_ESTABLISHED:
2019 if (st->bucket > tcp_hashinfo.ehash_mask)
2021 rc = established_get_first(seq);
2022 while (offset-- && rc)
2023 rc = established_get_next(seq, rc);
2031 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2033 struct tcp_iter_state *st = seq->private;
2036 if (*pos && *pos == st->last_pos) {
2037 rc = tcp_seek_last_pos(seq);
2042 st->state = TCP_SEQ_STATE_LISTENING;
2046 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2049 st->last_pos = *pos;
2053 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2055 struct tcp_iter_state *st = seq->private;
2058 if (v == SEQ_START_TOKEN) {
2059 rc = tcp_get_idx(seq, 0);
2063 switch (st->state) {
2064 case TCP_SEQ_STATE_LISTENING:
2065 rc = listening_get_next(seq, v);
2067 st->state = TCP_SEQ_STATE_ESTABLISHED;
2070 rc = established_get_first(seq);
2073 case TCP_SEQ_STATE_ESTABLISHED:
2074 rc = established_get_next(seq, v);
2079 st->last_pos = *pos;
2083 static void tcp_seq_stop(struct seq_file *seq, void *v)
2085 struct tcp_iter_state *st = seq->private;
2087 switch (st->state) {
2088 case TCP_SEQ_STATE_LISTENING:
2089 if (v != SEQ_START_TOKEN)
2090 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2092 case TCP_SEQ_STATE_ESTABLISHED:
2094 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2099 int tcp_seq_open(struct inode *inode, struct file *file)
2101 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2102 struct tcp_iter_state *s;
2105 err = seq_open_net(inode, file, &afinfo->seq_ops,
2106 sizeof(struct tcp_iter_state));
2110 s = ((struct seq_file *)file->private_data)->private;
2111 s->family = afinfo->family;
2115 EXPORT_SYMBOL(tcp_seq_open);
2117 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2120 struct proc_dir_entry *p;
2122 afinfo->seq_ops.start = tcp_seq_start;
2123 afinfo->seq_ops.next = tcp_seq_next;
2124 afinfo->seq_ops.stop = tcp_seq_stop;
2126 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2127 afinfo->seq_fops, afinfo);
2132 EXPORT_SYMBOL(tcp_proc_register);
2134 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2136 remove_proc_entry(afinfo->name, net->proc_net);
2138 EXPORT_SYMBOL(tcp_proc_unregister);
2140 static void get_openreq4(const struct request_sock *req,
2141 struct seq_file *f, int i)
2143 const struct inet_request_sock *ireq = inet_rsk(req);
2144 long delta = req->rsk_timer.expires - jiffies;
2146 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2147 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2152 ntohs(ireq->ir_rmt_port),
2154 0, 0, /* could print option size, but that is af dependent. */
2155 1, /* timers active (only the expire timer) */
2156 jiffies_delta_to_clock_t(delta),
2158 from_kuid_munged(seq_user_ns(f),
2159 sock_i_uid(req->rsk_listener)),
2160 0, /* non standard timer */
2161 0, /* open_requests have no inode */
2166 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2169 unsigned long timer_expires;
2170 const struct tcp_sock *tp = tcp_sk(sk);
2171 const struct inet_connection_sock *icsk = inet_csk(sk);
2172 const struct inet_sock *inet = inet_sk(sk);
2173 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2174 __be32 dest = inet->inet_daddr;
2175 __be32 src = inet->inet_rcv_saddr;
2176 __u16 destp = ntohs(inet->inet_dport);
2177 __u16 srcp = ntohs(inet->inet_sport);
2181 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2182 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2183 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2185 timer_expires = icsk->icsk_timeout;
2186 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2188 timer_expires = icsk->icsk_timeout;
2189 } else if (timer_pending(&sk->sk_timer)) {
2191 timer_expires = sk->sk_timer.expires;
2194 timer_expires = jiffies;
2197 state = sk_state_load(sk);
2198 if (state == TCP_LISTEN)
2199 rx_queue = sk->sk_ack_backlog;
2201 /* Because we don't lock the socket,
2202 * we might find a transient negative value.
2204 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2206 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2207 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2208 i, src, srcp, dest, destp, state,
2209 tp->write_seq - tp->snd_una,
2212 jiffies_delta_to_clock_t(timer_expires - jiffies),
2213 icsk->icsk_retransmits,
2214 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2215 icsk->icsk_probes_out,
2217 atomic_read(&sk->sk_refcnt), sk,
2218 jiffies_to_clock_t(icsk->icsk_rto),
2219 jiffies_to_clock_t(icsk->icsk_ack.ato),
2220 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2222 state == TCP_LISTEN ?
2223 fastopenq->max_qlen :
2224 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2227 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2228 struct seq_file *f, int i)
2230 long delta = tw->tw_timer.expires - jiffies;
2234 dest = tw->tw_daddr;
2235 src = tw->tw_rcv_saddr;
2236 destp = ntohs(tw->tw_dport);
2237 srcp = ntohs(tw->tw_sport);
2239 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2240 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2241 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2242 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2243 atomic_read(&tw->tw_refcnt), tw);
2248 static int tcp4_seq_show(struct seq_file *seq, void *v)
2250 struct tcp_iter_state *st;
2251 struct sock *sk = v;
2253 seq_setwidth(seq, TMPSZ - 1);
2254 if (v == SEQ_START_TOKEN) {
2255 seq_puts(seq, " sl local_address rem_address st tx_queue "
2256 "rx_queue tr tm->when retrnsmt uid timeout "
2262 if (sk->sk_state == TCP_TIME_WAIT)
2263 get_timewait4_sock(v, seq, st->num);
2264 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2265 get_openreq4(v, seq, st->num);
2267 get_tcp4_sock(v, seq, st->num);
2273 static const struct file_operations tcp_afinfo_seq_fops = {
2274 .owner = THIS_MODULE,
2275 .open = tcp_seq_open,
2277 .llseek = seq_lseek,
2278 .release = seq_release_net
2281 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2284 .seq_fops = &tcp_afinfo_seq_fops,
2286 .show = tcp4_seq_show,
2290 static int __net_init tcp4_proc_init_net(struct net *net)
2292 return tcp_proc_register(net, &tcp4_seq_afinfo);
2295 static void __net_exit tcp4_proc_exit_net(struct net *net)
2297 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2300 static struct pernet_operations tcp4_net_ops = {
2301 .init = tcp4_proc_init_net,
2302 .exit = tcp4_proc_exit_net,
2305 int __init tcp4_proc_init(void)
2307 return register_pernet_subsys(&tcp4_net_ops);
2310 void tcp4_proc_exit(void)
2312 unregister_pernet_subsys(&tcp4_net_ops);
2314 #endif /* CONFIG_PROC_FS */
2316 struct proto tcp_prot = {
2318 .owner = THIS_MODULE,
2320 .connect = tcp_v4_connect,
2321 .disconnect = tcp_disconnect,
2322 .accept = inet_csk_accept,
2324 .init = tcp_v4_init_sock,
2325 .destroy = tcp_v4_destroy_sock,
2326 .shutdown = tcp_shutdown,
2327 .setsockopt = tcp_setsockopt,
2328 .getsockopt = tcp_getsockopt,
2329 .recvmsg = tcp_recvmsg,
2330 .sendmsg = tcp_sendmsg,
2331 .sendpage = tcp_sendpage,
2332 .backlog_rcv = tcp_v4_do_rcv,
2333 .release_cb = tcp_release_cb,
2335 .unhash = inet_unhash,
2336 .get_port = inet_csk_get_port,
2337 .enter_memory_pressure = tcp_enter_memory_pressure,
2338 .stream_memory_free = tcp_stream_memory_free,
2339 .sockets_allocated = &tcp_sockets_allocated,
2340 .orphan_count = &tcp_orphan_count,
2341 .memory_allocated = &tcp_memory_allocated,
2342 .memory_pressure = &tcp_memory_pressure,
2343 .sysctl_mem = sysctl_tcp_mem,
2344 .sysctl_wmem = sysctl_tcp_wmem,
2345 .sysctl_rmem = sysctl_tcp_rmem,
2346 .max_header = MAX_TCP_HEADER,
2347 .obj_size = sizeof(struct tcp_sock),
2348 .slab_flags = SLAB_DESTROY_BY_RCU,
2349 .twsk_prot = &tcp_timewait_sock_ops,
2350 .rsk_prot = &tcp_request_sock_ops,
2351 .h.hashinfo = &tcp_hashinfo,
2352 .no_autobind = true,
2353 #ifdef CONFIG_COMPAT
2354 .compat_setsockopt = compat_tcp_setsockopt,
2355 .compat_getsockopt = compat_tcp_getsockopt,
2357 .diag_destroy = tcp_abort,
2359 EXPORT_SYMBOL(tcp_prot);
2361 static void __net_exit tcp_sk_exit(struct net *net)
2365 for_each_possible_cpu(cpu)
2366 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2367 free_percpu(net->ipv4.tcp_sk);
2370 static int __net_init tcp_sk_init(struct net *net)
2374 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2375 if (!net->ipv4.tcp_sk)
2378 for_each_possible_cpu(cpu) {
2381 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2385 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2388 net->ipv4.sysctl_tcp_ecn = 2;
2389 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2391 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2392 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2393 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2395 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2396 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2397 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2399 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2400 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2401 net->ipv4.sysctl_tcp_syncookies = 1;
2402 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2403 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2404 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2405 net->ipv4.sysctl_tcp_orphan_retries = 0;
2406 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2407 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2416 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2418 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2421 static struct pernet_operations __net_initdata tcp_sk_ops = {
2422 .init = tcp_sk_init,
2423 .exit = tcp_sk_exit,
2424 .exit_batch = tcp_sk_exit_batch,
2427 void __init tcp_v4_init(void)
2429 inet_hashinfo_init(&tcp_hashinfo);
2430 if (register_pernet_subsys(&tcp_sk_ops))
2431 panic("Failed to create the TCP control socket.\n");