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;
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 lockdep_sock_is_held(sk));
161 if (inet_opt && inet_opt->opt.srr) {
164 nexthop = inet_opt->opt.faddr;
167 orig_sport = inet->inet_sport;
168 orig_dport = usin->sin_port;
169 fl4 = &inet->cork.fl.u.ip4;
170 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
171 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
173 orig_sport, orig_dport, sk);
176 if (err == -ENETUNREACH)
177 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
181 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
186 if (!inet_opt || !inet_opt->opt.srr)
189 if (!inet->inet_saddr)
190 inet->inet_saddr = fl4->saddr;
191 sk_rcv_saddr_set(sk, inet->inet_saddr);
193 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
194 /* Reset inherited state */
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
197 if (likely(!tp->repair))
201 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.
322 if (seq != tcp_rsk(req)->snt_isn) {
323 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
326 * Still in SYN_RECV, just remove it silently.
327 * There is no good way to pass the error to the newly
328 * created socket, and POSIX does not want network
329 * errors returned from accept().
331 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
332 tcp_listendrop(req->rsk_listener);
336 EXPORT_SYMBOL(tcp_req_err);
339 * This routine is called by the ICMP module when it gets some
340 * sort of error condition. If err < 0 then the socket should
341 * be closed and the error returned to the user. If err > 0
342 * it's just the icmp type << 8 | icmp code. After adjustment
343 * header points to the first 8 bytes of the tcp header. We need
344 * to find the appropriate port.
346 * The locking strategy used here is very "optimistic". When
347 * someone else accesses the socket the ICMP is just dropped
348 * and for some paths there is no check at all.
349 * A more general error queue to queue errors for later handling
350 * is probably better.
354 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
356 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
357 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
358 struct inet_connection_sock *icsk;
360 struct inet_sock *inet;
361 const int type = icmp_hdr(icmp_skb)->type;
362 const int code = icmp_hdr(icmp_skb)->code;
365 struct request_sock *fastopen;
369 struct net *net = dev_net(icmp_skb->dev);
371 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
372 th->dest, iph->saddr, ntohs(th->source),
375 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
378 if (sk->sk_state == TCP_TIME_WAIT) {
379 inet_twsk_put(inet_twsk(sk));
382 seq = ntohl(th->seq);
383 if (sk->sk_state == TCP_NEW_SYN_RECV)
384 return tcp_req_err(sk, seq,
385 type == ICMP_PARAMETERPROB ||
386 type == ICMP_TIME_EXCEEDED ||
387 (type == ICMP_DEST_UNREACH &&
388 (code == ICMP_NET_UNREACH ||
389 code == ICMP_HOST_UNREACH)));
392 /* If too many ICMPs get dropped on busy
393 * servers this needs to be solved differently.
394 * We do take care of PMTU discovery (RFC1191) special case :
395 * we can receive locally generated ICMP messages while socket is held.
397 if (sock_owned_by_user(sk)) {
398 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
399 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
401 if (sk->sk_state == TCP_CLOSE)
404 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
405 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
411 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
412 fastopen = tp->fastopen_rsk;
413 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
414 if (sk->sk_state != TCP_LISTEN &&
415 !between(seq, snd_una, tp->snd_nxt)) {
416 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
422 do_redirect(icmp_skb, sk);
424 case ICMP_SOURCE_QUENCH:
425 /* Just silently ignore these. */
427 case ICMP_PARAMETERPROB:
430 case ICMP_DEST_UNREACH:
431 if (code > NR_ICMP_UNREACH)
434 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
435 /* We are not interested in TCP_LISTEN and open_requests
436 * (SYN-ACKs send out by Linux are always <576bytes so
437 * they should go through unfragmented).
439 if (sk->sk_state == TCP_LISTEN)
443 if (!sock_owned_by_user(sk)) {
444 tcp_v4_mtu_reduced(sk);
446 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
452 err = icmp_err_convert[code].errno;
453 /* check if icmp_skb allows revert of backoff
454 * (see draft-zimmermann-tcp-lcd) */
455 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
457 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
458 !icsk->icsk_backoff || fastopen)
461 if (sock_owned_by_user(sk))
464 icsk->icsk_backoff--;
465 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
467 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
469 skb = tcp_write_queue_head(sk);
472 remaining = icsk->icsk_rto -
474 tcp_time_stamp - tcp_skb_timestamp(skb));
477 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
478 remaining, TCP_RTO_MAX);
480 /* RTO revert clocked out retransmission.
481 * Will retransmit now */
482 tcp_retransmit_timer(sk);
486 case ICMP_TIME_EXCEEDED:
493 switch (sk->sk_state) {
496 /* Only in fast or simultaneous open. If a fast open socket is
497 * is already accepted it is treated as a connected one below.
499 if (fastopen && !fastopen->sk)
502 if (!sock_owned_by_user(sk)) {
505 sk->sk_error_report(sk);
509 sk->sk_err_soft = err;
514 /* If we've already connected we will keep trying
515 * until we time out, or the user gives up.
517 * rfc1122 4.2.3.9 allows to consider as hard errors
518 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
519 * but it is obsoleted by pmtu discovery).
521 * Note, that in modern internet, where routing is unreliable
522 * and in each dark corner broken firewalls sit, sending random
523 * errors ordered by their masters even this two messages finally lose
524 * their original sense (even Linux sends invalid PORT_UNREACHs)
526 * Now we are in compliance with RFCs.
531 if (!sock_owned_by_user(sk) && inet->recverr) {
533 sk->sk_error_report(sk);
534 } else { /* Only an error on timeout */
535 sk->sk_err_soft = err;
543 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
545 struct tcphdr *th = tcp_hdr(skb);
547 if (skb->ip_summed == CHECKSUM_PARTIAL) {
548 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
549 skb->csum_start = skb_transport_header(skb) - skb->head;
550 skb->csum_offset = offsetof(struct tcphdr, check);
552 th->check = tcp_v4_check(skb->len, saddr, daddr,
559 /* This routine computes an IPv4 TCP checksum. */
560 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
562 const struct inet_sock *inet = inet_sk(sk);
564 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
566 EXPORT_SYMBOL(tcp_v4_send_check);
569 * This routine will send an RST to the other tcp.
571 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
573 * Answer: if a packet caused RST, it is not for a socket
574 * existing in our system, if it is matched to a socket,
575 * it is just duplicate segment or bug in other side's TCP.
576 * So that we build reply only basing on parameters
577 * arrived with segment.
578 * Exception: precedence violation. We do not implement it in any case.
581 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
583 const struct tcphdr *th = tcp_hdr(skb);
586 #ifdef CONFIG_TCP_MD5SIG
587 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
590 struct ip_reply_arg arg;
591 #ifdef CONFIG_TCP_MD5SIG
592 struct tcp_md5sig_key *key = NULL;
593 const __u8 *hash_location = NULL;
594 unsigned char newhash[16];
596 struct sock *sk1 = NULL;
600 /* Never send a reset in response to a reset. */
604 /* If sk not NULL, it means we did a successful lookup and incoming
605 * route had to be correct. prequeue might have dropped our dst.
607 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
610 /* Swap the send and the receive. */
611 memset(&rep, 0, sizeof(rep));
612 rep.th.dest = th->source;
613 rep.th.source = th->dest;
614 rep.th.doff = sizeof(struct tcphdr) / 4;
618 rep.th.seq = th->ack_seq;
621 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
622 skb->len - (th->doff << 2));
625 memset(&arg, 0, sizeof(arg));
626 arg.iov[0].iov_base = (unsigned char *)&rep;
627 arg.iov[0].iov_len = sizeof(rep.th);
629 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
630 #ifdef CONFIG_TCP_MD5SIG
632 hash_location = tcp_parse_md5sig_option(th);
633 if (sk && sk_fullsock(sk)) {
634 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
635 &ip_hdr(skb)->saddr, AF_INET);
636 } else if (hash_location) {
638 * active side is lost. Try to find listening socket through
639 * source port, and then find md5 key through listening socket.
640 * we are not loose security here:
641 * Incoming packet is checked with md5 hash with finding key,
642 * no RST generated if md5 hash doesn't match.
644 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
646 th->source, ip_hdr(skb)->daddr,
647 ntohs(th->source), inet_iif(skb));
648 /* don't send rst if it can't find key */
652 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
653 &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)
665 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
667 (TCPOPT_MD5SIG << 8) |
669 /* Update length and the length the header thinks exists */
670 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
671 rep.th.doff = arg.iov[0].iov_len / 4;
673 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
674 key, ip_hdr(skb)->saddr,
675 ip_hdr(skb)->daddr, &rep.th);
678 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
679 ip_hdr(skb)->saddr, /* XXX */
680 arg.iov[0].iov_len, IPPROTO_TCP, 0);
681 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
682 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
684 /* When socket is gone, all binding information is lost.
685 * routing might fail in this case. No choice here, if we choose to force
686 * input interface, we will misroute in case of asymmetric route.
689 arg.bound_dev_if = sk->sk_bound_dev_if;
691 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
692 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
694 arg.tos = ip_hdr(skb)->tos;
696 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
697 skb, &TCP_SKB_CB(skb)->header.h4.opt,
698 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
699 &arg, arg.iov[0].iov_len);
701 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
702 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
705 #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;
780 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
781 skb, &TCP_SKB_CB(skb)->header.h4.opt,
782 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
783 &arg, arg.iov[0].iov_len);
785 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
789 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
791 struct inet_timewait_sock *tw = inet_twsk(sk);
792 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
794 tcp_v4_send_ack(sock_net(sk), skb,
795 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
796 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
797 tcp_time_stamp + tcptw->tw_ts_offset,
800 tcp_twsk_md5_key(tcptw),
801 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
808 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
809 struct request_sock *req)
811 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
812 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
814 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
818 * The window field (SEG.WND) of every outgoing segment, with the
819 * exception of <SYN> segments, MUST be right-shifted by
820 * Rcv.Wind.Shift bits:
822 tcp_v4_send_ack(sock_net(sk), skb, seq,
823 tcp_rsk(req)->rcv_nxt,
824 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
828 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
830 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
835 * Send a SYN-ACK after having received a SYN.
836 * This still operates on a request_sock only, not on a big
839 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
841 struct request_sock *req,
842 struct tcp_fastopen_cookie *foc,
843 enum tcp_synack_type synack_type)
845 const struct inet_request_sock *ireq = inet_rsk(req);
850 /* First, grab a route. */
851 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
854 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
857 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
859 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
862 err = net_xmit_eval(err);
869 * IPv4 request_sock destructor.
871 static void tcp_v4_reqsk_destructor(struct request_sock *req)
873 kfree(inet_rsk(req)->opt);
876 #ifdef CONFIG_TCP_MD5SIG
878 * RFC2385 MD5 checksumming requires a mapping of
879 * IP address->MD5 Key.
880 * We need to maintain these in the sk structure.
883 /* Find the Key structure for an address. */
884 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
885 const union tcp_md5_addr *addr,
888 const struct tcp_sock *tp = tcp_sk(sk);
889 struct tcp_md5sig_key *key;
890 unsigned int size = sizeof(struct in_addr);
891 const struct tcp_md5sig_info *md5sig;
893 /* caller either holds rcu_read_lock() or socket lock */
894 md5sig = rcu_dereference_check(tp->md5sig_info,
895 lockdep_sock_is_held(sk));
898 #if IS_ENABLED(CONFIG_IPV6)
899 if (family == AF_INET6)
900 size = sizeof(struct in6_addr);
902 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
903 if (key->family != family)
905 if (!memcmp(&key->addr, addr, size))
910 EXPORT_SYMBOL(tcp_md5_do_lookup);
912 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
913 const struct sock *addr_sk)
915 const union tcp_md5_addr *addr;
917 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
918 return tcp_md5_do_lookup(sk, addr, AF_INET);
920 EXPORT_SYMBOL(tcp_v4_md5_lookup);
922 /* This can be called on a newly created socket, from other files */
923 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
924 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
926 /* Add Key to the list */
927 struct tcp_md5sig_key *key;
928 struct tcp_sock *tp = tcp_sk(sk);
929 struct tcp_md5sig_info *md5sig;
931 key = tcp_md5_do_lookup(sk, addr, family);
933 /* Pre-existing entry - just update that one. */
934 memcpy(key->key, newkey, newkeylen);
935 key->keylen = newkeylen;
939 md5sig = rcu_dereference_protected(tp->md5sig_info,
940 lockdep_sock_is_held(sk));
942 md5sig = kmalloc(sizeof(*md5sig), gfp);
946 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
947 INIT_HLIST_HEAD(&md5sig->head);
948 rcu_assign_pointer(tp->md5sig_info, md5sig);
951 key = sock_kmalloc(sk, sizeof(*key), gfp);
954 if (!tcp_alloc_md5sig_pool()) {
955 sock_kfree_s(sk, key, sizeof(*key));
959 memcpy(key->key, newkey, newkeylen);
960 key->keylen = newkeylen;
961 key->family = family;
962 memcpy(&key->addr, addr,
963 (family == AF_INET6) ? sizeof(struct in6_addr) :
964 sizeof(struct in_addr));
965 hlist_add_head_rcu(&key->node, &md5sig->head);
968 EXPORT_SYMBOL(tcp_md5_do_add);
970 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
972 struct tcp_md5sig_key *key;
974 key = tcp_md5_do_lookup(sk, addr, family);
977 hlist_del_rcu(&key->node);
978 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
982 EXPORT_SYMBOL(tcp_md5_do_del);
984 static void tcp_clear_md5_list(struct sock *sk)
986 struct tcp_sock *tp = tcp_sk(sk);
987 struct tcp_md5sig_key *key;
988 struct hlist_node *n;
989 struct tcp_md5sig_info *md5sig;
991 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
993 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
994 hlist_del_rcu(&key->node);
995 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1000 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1003 struct tcp_md5sig cmd;
1004 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1006 if (optlen < sizeof(cmd))
1009 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1012 if (sin->sin_family != AF_INET)
1015 if (!cmd.tcpm_keylen)
1016 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1019 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1022 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1023 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1027 static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1028 __be32 daddr, __be32 saddr,
1029 const struct tcphdr *th, int nbytes)
1031 struct tcp4_pseudohdr *bp;
1032 struct scatterlist sg;
1039 bp->protocol = IPPROTO_TCP;
1040 bp->len = cpu_to_be16(nbytes);
1042 _th = (struct tcphdr *)(bp + 1);
1043 memcpy(_th, th, sizeof(*th));
1046 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1047 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1048 sizeof(*bp) + sizeof(*th));
1049 return crypto_ahash_update(hp->md5_req);
1052 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1053 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1055 struct tcp_md5sig_pool *hp;
1056 struct ahash_request *req;
1058 hp = tcp_get_md5sig_pool();
1060 goto clear_hash_noput;
1063 if (crypto_ahash_init(req))
1065 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1067 if (tcp_md5_hash_key(hp, key))
1069 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1070 if (crypto_ahash_final(req))
1073 tcp_put_md5sig_pool();
1077 tcp_put_md5sig_pool();
1079 memset(md5_hash, 0, 16);
1083 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1084 const struct sock *sk,
1085 const struct sk_buff *skb)
1087 struct tcp_md5sig_pool *hp;
1088 struct ahash_request *req;
1089 const struct tcphdr *th = tcp_hdr(skb);
1090 __be32 saddr, daddr;
1092 if (sk) { /* valid for establish/request sockets */
1093 saddr = sk->sk_rcv_saddr;
1094 daddr = sk->sk_daddr;
1096 const struct iphdr *iph = ip_hdr(skb);
1101 hp = tcp_get_md5sig_pool();
1103 goto clear_hash_noput;
1106 if (crypto_ahash_init(req))
1109 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1111 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1113 if (tcp_md5_hash_key(hp, key))
1115 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1116 if (crypto_ahash_final(req))
1119 tcp_put_md5sig_pool();
1123 tcp_put_md5sig_pool();
1125 memset(md5_hash, 0, 16);
1128 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1132 /* Called with rcu_read_lock() */
1133 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1134 const struct sk_buff *skb)
1136 #ifdef CONFIG_TCP_MD5SIG
1138 * This gets called for each TCP segment that arrives
1139 * so we want to be efficient.
1140 * We have 3 drop cases:
1141 * o No MD5 hash and one expected.
1142 * o MD5 hash and we're not expecting one.
1143 * o MD5 hash and its wrong.
1145 const __u8 *hash_location = NULL;
1146 struct tcp_md5sig_key *hash_expected;
1147 const struct iphdr *iph = ip_hdr(skb);
1148 const struct tcphdr *th = tcp_hdr(skb);
1150 unsigned char newhash[16];
1152 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1154 hash_location = tcp_parse_md5sig_option(th);
1156 /* We've parsed the options - do we have a hash? */
1157 if (!hash_expected && !hash_location)
1160 if (hash_expected && !hash_location) {
1161 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1165 if (!hash_expected && hash_location) {
1166 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1170 /* Okay, so this is hash_expected and hash_location -
1171 * so we need to calculate the checksum.
1173 genhash = tcp_v4_md5_hash_skb(newhash,
1177 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1178 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1179 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1180 &iph->saddr, ntohs(th->source),
1181 &iph->daddr, ntohs(th->dest),
1182 genhash ? " tcp_v4_calc_md5_hash failed"
1191 static void tcp_v4_init_req(struct request_sock *req,
1192 const struct sock *sk_listener,
1193 struct sk_buff *skb)
1195 struct inet_request_sock *ireq = inet_rsk(req);
1197 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1198 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1199 ireq->opt = tcp_v4_save_options(skb);
1202 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1204 const struct request_sock *req,
1207 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1210 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1219 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1221 .obj_size = sizeof(struct tcp_request_sock),
1222 .rtx_syn_ack = tcp_rtx_synack,
1223 .send_ack = tcp_v4_reqsk_send_ack,
1224 .destructor = tcp_v4_reqsk_destructor,
1225 .send_reset = tcp_v4_send_reset,
1226 .syn_ack_timeout = tcp_syn_ack_timeout,
1229 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1230 .mss_clamp = TCP_MSS_DEFAULT,
1231 #ifdef CONFIG_TCP_MD5SIG
1232 .req_md5_lookup = tcp_v4_md5_lookup,
1233 .calc_md5_hash = tcp_v4_md5_hash_skb,
1235 .init_req = tcp_v4_init_req,
1236 #ifdef CONFIG_SYN_COOKIES
1237 .cookie_init_seq = cookie_v4_init_sequence,
1239 .route_req = tcp_v4_route_req,
1240 .init_seq = tcp_v4_init_sequence,
1241 .send_synack = tcp_v4_send_synack,
1244 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1246 /* Never answer to SYNs send to broadcast or multicast */
1247 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1250 return tcp_conn_request(&tcp_request_sock_ops,
1251 &tcp_request_sock_ipv4_ops, sk, skb);
1257 EXPORT_SYMBOL(tcp_v4_conn_request);
1261 * The three way handshake has completed - we got a valid synack -
1262 * now create the new socket.
1264 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1265 struct request_sock *req,
1266 struct dst_entry *dst,
1267 struct request_sock *req_unhash,
1270 struct inet_request_sock *ireq;
1271 struct inet_sock *newinet;
1272 struct tcp_sock *newtp;
1274 #ifdef CONFIG_TCP_MD5SIG
1275 struct tcp_md5sig_key *key;
1277 struct ip_options_rcu *inet_opt;
1279 if (sk_acceptq_is_full(sk))
1282 newsk = tcp_create_openreq_child(sk, req, skb);
1286 newsk->sk_gso_type = SKB_GSO_TCPV4;
1287 inet_sk_rx_dst_set(newsk, skb);
1289 newtp = tcp_sk(newsk);
1290 newinet = inet_sk(newsk);
1291 ireq = inet_rsk(req);
1292 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1293 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1294 newsk->sk_bound_dev_if = ireq->ir_iif;
1295 newinet->inet_saddr = ireq->ir_loc_addr;
1296 inet_opt = ireq->opt;
1297 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1299 newinet->mc_index = inet_iif(skb);
1300 newinet->mc_ttl = ip_hdr(skb)->ttl;
1301 newinet->rcv_tos = ip_hdr(skb)->tos;
1302 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1304 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1305 newinet->inet_id = newtp->write_seq ^ jiffies;
1308 dst = inet_csk_route_child_sock(sk, newsk, req);
1312 /* syncookie case : see end of cookie_v4_check() */
1314 sk_setup_caps(newsk, dst);
1316 tcp_ca_openreq_child(newsk, dst);
1318 tcp_sync_mss(newsk, dst_mtu(dst));
1319 newtp->advmss = dst_metric_advmss(dst);
1320 if (tcp_sk(sk)->rx_opt.user_mss &&
1321 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1322 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1324 tcp_initialize_rcv_mss(newsk);
1326 #ifdef CONFIG_TCP_MD5SIG
1327 /* Copy over the MD5 key from the original socket */
1328 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1332 * We're using one, so create a matching key
1333 * on the newsk structure. If we fail to get
1334 * memory, then we end up not copying the key
1337 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1338 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1339 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1343 if (__inet_inherit_port(sk, newsk) < 0)
1345 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1347 tcp_move_syn(newtp, req);
1352 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1359 inet_csk_prepare_forced_close(newsk);
1363 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1365 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1367 #ifdef CONFIG_SYN_COOKIES
1368 const struct tcphdr *th = tcp_hdr(skb);
1371 sk = cookie_v4_check(sk, skb);
1376 /* The socket must have it's spinlock held when we get
1377 * here, unless it is a TCP_LISTEN socket.
1379 * We have a potential double-lock case here, so even when
1380 * doing backlog processing we use the BH locking scheme.
1381 * This is because we cannot sleep with the original spinlock
1384 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1388 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1389 struct dst_entry *dst = sk->sk_rx_dst;
1391 sock_rps_save_rxhash(sk, skb);
1392 sk_mark_napi_id(sk, skb);
1394 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1395 !dst->ops->check(dst, 0)) {
1397 sk->sk_rx_dst = NULL;
1400 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1404 if (tcp_checksum_complete(skb))
1407 if (sk->sk_state == TCP_LISTEN) {
1408 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1413 sock_rps_save_rxhash(nsk, skb);
1414 sk_mark_napi_id(nsk, skb);
1415 if (tcp_child_process(sk, nsk, skb)) {
1422 sock_rps_save_rxhash(sk, skb);
1424 if (tcp_rcv_state_process(sk, skb)) {
1431 tcp_v4_send_reset(rsk, skb);
1434 /* Be careful here. If this function gets more complicated and
1435 * gcc suffers from register pressure on the x86, sk (in %ebx)
1436 * might be destroyed here. This current version compiles correctly,
1437 * but you have been warned.
1442 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1443 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1446 EXPORT_SYMBOL(tcp_v4_do_rcv);
1448 void tcp_v4_early_demux(struct sk_buff *skb)
1450 const struct iphdr *iph;
1451 const struct tcphdr *th;
1454 if (skb->pkt_type != PACKET_HOST)
1457 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1463 if (th->doff < sizeof(struct tcphdr) / 4)
1466 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1467 iph->saddr, th->source,
1468 iph->daddr, ntohs(th->dest),
1472 skb->destructor = sock_edemux;
1473 if (sk_fullsock(sk)) {
1474 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1477 dst = dst_check(dst, 0);
1479 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1480 skb_dst_set_noref(skb, dst);
1485 /* Packet is added to VJ-style prequeue for processing in process
1486 * context, if a reader task is waiting. Apparently, this exciting
1487 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1488 * failed somewhere. Latency? Burstiness? Well, at least now we will
1489 * see, why it failed. 8)8) --ANK
1492 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1494 struct tcp_sock *tp = tcp_sk(sk);
1496 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1499 if (skb->len <= tcp_hdrlen(skb) &&
1500 skb_queue_len(&tp->ucopy.prequeue) == 0)
1503 /* Before escaping RCU protected region, we need to take care of skb
1504 * dst. Prequeue is only enabled for established sockets.
1505 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1506 * Instead of doing full sk_rx_dst validity here, let's perform
1507 * an optimistic check.
1509 if (likely(sk->sk_rx_dst))
1512 skb_dst_force_safe(skb);
1514 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1515 tp->ucopy.memory += skb->truesize;
1516 if (skb_queue_len(&tp->ucopy.prequeue) >= 32 ||
1517 tp->ucopy.memory + atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) {
1518 struct sk_buff *skb1;
1520 BUG_ON(sock_owned_by_user(sk));
1521 __NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED,
1522 skb_queue_len(&tp->ucopy.prequeue));
1524 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
1525 sk_backlog_rcv(sk, skb1);
1527 tp->ucopy.memory = 0;
1528 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1529 wake_up_interruptible_sync_poll(sk_sleep(sk),
1530 POLLIN | POLLRDNORM | POLLRDBAND);
1531 if (!inet_csk_ack_scheduled(sk))
1532 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1533 (3 * tcp_rto_min(sk)) / 4,
1538 EXPORT_SYMBOL(tcp_prequeue);
1540 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1542 u32 limit = sk->sk_rcvbuf + sk->sk_sndbuf;
1544 /* Only socket owner can try to collapse/prune rx queues
1545 * to reduce memory overhead, so add a little headroom here.
1546 * Few sockets backlog are possibly concurrently non empty.
1550 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1551 * we can fix skb->truesize to its real value to avoid future drops.
1552 * This is valid because skb is not yet charged to the socket.
1553 * It has been noticed pure SACK packets were sometimes dropped
1554 * (if cooked by drivers without copybreak feature).
1557 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
1559 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1561 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1566 EXPORT_SYMBOL(tcp_add_backlog);
1572 int tcp_v4_rcv(struct sk_buff *skb)
1574 struct net *net = dev_net(skb->dev);
1575 const struct iphdr *iph;
1576 const struct tcphdr *th;
1581 if (skb->pkt_type != PACKET_HOST)
1584 /* Count it even if it's bad */
1585 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1587 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1590 th = (const struct tcphdr *)skb->data;
1592 if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1594 if (!pskb_may_pull(skb, th->doff * 4))
1597 /* An explanation is required here, I think.
1598 * Packet length and doff are validated by header prediction,
1599 * provided case of th->doff==0 is eliminated.
1600 * So, we defer the checks. */
1602 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1605 th = (const struct tcphdr *)skb->data;
1607 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1608 * barrier() makes sure compiler wont play fool^Waliasing games.
1610 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1611 sizeof(struct inet_skb_parm));
1614 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1615 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1616 skb->len - th->doff * 4);
1617 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1618 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1619 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1620 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1621 TCP_SKB_CB(skb)->sacked = 0;
1624 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1625 th->dest, &refcounted);
1630 if (sk->sk_state == TCP_TIME_WAIT)
1633 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1634 struct request_sock *req = inet_reqsk(sk);
1637 sk = req->rsk_listener;
1638 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1639 sk_drops_add(sk, skb);
1643 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1644 inet_csk_reqsk_queue_drop_and_put(sk, req);
1647 /* We own a reference on the listener, increase it again
1648 * as we might lose it too soon.
1652 nsk = tcp_check_req(sk, skb, req, false);
1655 goto discard_and_relse;
1659 } else if (tcp_child_process(sk, nsk, skb)) {
1660 tcp_v4_send_reset(nsk, skb);
1661 goto discard_and_relse;
1667 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1668 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1669 goto discard_and_relse;
1672 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1673 goto discard_and_relse;
1675 if (tcp_v4_inbound_md5_hash(sk, skb))
1676 goto discard_and_relse;
1680 if (sk_filter(sk, skb))
1681 goto discard_and_relse;
1685 if (sk->sk_state == TCP_LISTEN) {
1686 ret = tcp_v4_do_rcv(sk, skb);
1687 goto put_and_return;
1690 sk_incoming_cpu_update(sk);
1692 bh_lock_sock_nested(sk);
1693 tcp_segs_in(tcp_sk(sk), skb);
1695 if (!sock_owned_by_user(sk)) {
1696 if (!tcp_prequeue(sk, skb))
1697 ret = tcp_v4_do_rcv(sk, skb);
1698 } else if (tcp_add_backlog(sk, skb)) {
1699 goto discard_and_relse;
1710 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1713 if (tcp_checksum_complete(skb)) {
1715 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1717 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1719 tcp_v4_send_reset(NULL, skb);
1723 /* Discard frame. */
1728 sk_drops_add(sk, skb);
1734 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1735 inet_twsk_put(inet_twsk(sk));
1739 if (tcp_checksum_complete(skb)) {
1740 inet_twsk_put(inet_twsk(sk));
1743 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1745 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1748 iph->saddr, th->source,
1749 iph->daddr, th->dest,
1752 inet_twsk_deschedule_put(inet_twsk(sk));
1757 /* Fall through to ACK */
1760 tcp_v4_timewait_ack(sk, skb);
1763 tcp_v4_send_reset(sk, skb);
1764 inet_twsk_deschedule_put(inet_twsk(sk));
1766 case TCP_TW_SUCCESS:;
1771 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1772 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1773 .twsk_unique = tcp_twsk_unique,
1774 .twsk_destructor= tcp_twsk_destructor,
1777 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1779 struct dst_entry *dst = skb_dst(skb);
1781 if (dst && dst_hold_safe(dst)) {
1782 sk->sk_rx_dst = dst;
1783 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1786 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1788 const struct inet_connection_sock_af_ops ipv4_specific = {
1789 .queue_xmit = ip_queue_xmit,
1790 .send_check = tcp_v4_send_check,
1791 .rebuild_header = inet_sk_rebuild_header,
1792 .sk_rx_dst_set = inet_sk_rx_dst_set,
1793 .conn_request = tcp_v4_conn_request,
1794 .syn_recv_sock = tcp_v4_syn_recv_sock,
1795 .net_header_len = sizeof(struct iphdr),
1796 .setsockopt = ip_setsockopt,
1797 .getsockopt = ip_getsockopt,
1798 .addr2sockaddr = inet_csk_addr2sockaddr,
1799 .sockaddr_len = sizeof(struct sockaddr_in),
1800 .bind_conflict = inet_csk_bind_conflict,
1801 #ifdef CONFIG_COMPAT
1802 .compat_setsockopt = compat_ip_setsockopt,
1803 .compat_getsockopt = compat_ip_getsockopt,
1805 .mtu_reduced = tcp_v4_mtu_reduced,
1807 EXPORT_SYMBOL(ipv4_specific);
1809 #ifdef CONFIG_TCP_MD5SIG
1810 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1811 .md5_lookup = tcp_v4_md5_lookup,
1812 .calc_md5_hash = tcp_v4_md5_hash_skb,
1813 .md5_parse = tcp_v4_parse_md5_keys,
1817 /* NOTE: A lot of things set to zero explicitly by call to
1818 * sk_alloc() so need not be done here.
1820 static int tcp_v4_init_sock(struct sock *sk)
1822 struct inet_connection_sock *icsk = inet_csk(sk);
1826 icsk->icsk_af_ops = &ipv4_specific;
1828 #ifdef CONFIG_TCP_MD5SIG
1829 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1835 void tcp_v4_destroy_sock(struct sock *sk)
1837 struct tcp_sock *tp = tcp_sk(sk);
1839 tcp_clear_xmit_timers(sk);
1841 tcp_cleanup_congestion_control(sk);
1843 /* Cleanup up the write buffer. */
1844 tcp_write_queue_purge(sk);
1846 /* Cleans up our, hopefully empty, out_of_order_queue. */
1847 skb_rbtree_purge(&tp->out_of_order_queue);
1849 #ifdef CONFIG_TCP_MD5SIG
1850 /* Clean up the MD5 key list, if any */
1851 if (tp->md5sig_info) {
1852 tcp_clear_md5_list(sk);
1853 kfree_rcu(tp->md5sig_info, rcu);
1854 tp->md5sig_info = NULL;
1858 /* Clean prequeue, it must be empty really */
1859 __skb_queue_purge(&tp->ucopy.prequeue);
1861 /* Clean up a referenced TCP bind bucket. */
1862 if (inet_csk(sk)->icsk_bind_hash)
1865 BUG_ON(tp->fastopen_rsk);
1867 /* If socket is aborted during connect operation */
1868 tcp_free_fastopen_req(tp);
1869 tcp_saved_syn_free(tp);
1872 sk_sockets_allocated_dec(sk);
1875 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1877 #ifdef CONFIG_PROC_FS
1878 /* Proc filesystem TCP sock list dumping. */
1881 * Get next listener socket follow cur. If cur is NULL, get first socket
1882 * starting from bucket given in st->bucket; when st->bucket is zero the
1883 * very first socket in the hash table is returned.
1885 static void *listening_get_next(struct seq_file *seq, void *cur)
1887 struct tcp_iter_state *st = seq->private;
1888 struct net *net = seq_file_net(seq);
1889 struct inet_listen_hashbucket *ilb;
1890 struct inet_connection_sock *icsk;
1891 struct sock *sk = cur;
1895 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1896 spin_lock_bh(&ilb->lock);
1897 sk = sk_head(&ilb->head);
1901 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1907 sk_for_each_from(sk) {
1908 if (!net_eq(sock_net(sk), net))
1910 if (sk->sk_family == st->family)
1912 icsk = inet_csk(sk);
1914 spin_unlock_bh(&ilb->lock);
1916 if (++st->bucket < INET_LHTABLE_SIZE)
1921 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1923 struct tcp_iter_state *st = seq->private;
1928 rc = listening_get_next(seq, NULL);
1930 while (rc && *pos) {
1931 rc = listening_get_next(seq, rc);
1937 static inline bool empty_bucket(const struct tcp_iter_state *st)
1939 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1943 * Get first established socket starting from bucket given in st->bucket.
1944 * If st->bucket is zero, the very first socket in the hash is returned.
1946 static void *established_get_first(struct seq_file *seq)
1948 struct tcp_iter_state *st = seq->private;
1949 struct net *net = seq_file_net(seq);
1953 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1955 struct hlist_nulls_node *node;
1956 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1958 /* Lockless fast path for the common case of empty buckets */
1959 if (empty_bucket(st))
1963 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1964 if (sk->sk_family != st->family ||
1965 !net_eq(sock_net(sk), net)) {
1971 spin_unlock_bh(lock);
1977 static void *established_get_next(struct seq_file *seq, void *cur)
1979 struct sock *sk = cur;
1980 struct hlist_nulls_node *node;
1981 struct tcp_iter_state *st = seq->private;
1982 struct net *net = seq_file_net(seq);
1987 sk = sk_nulls_next(sk);
1989 sk_nulls_for_each_from(sk, node) {
1990 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
1994 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1996 return established_get_first(seq);
1999 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2001 struct tcp_iter_state *st = seq->private;
2005 rc = established_get_first(seq);
2008 rc = established_get_next(seq, rc);
2014 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2017 struct tcp_iter_state *st = seq->private;
2019 st->state = TCP_SEQ_STATE_LISTENING;
2020 rc = listening_get_idx(seq, &pos);
2023 st->state = TCP_SEQ_STATE_ESTABLISHED;
2024 rc = established_get_idx(seq, pos);
2030 static void *tcp_seek_last_pos(struct seq_file *seq)
2032 struct tcp_iter_state *st = seq->private;
2033 int offset = st->offset;
2034 int orig_num = st->num;
2037 switch (st->state) {
2038 case TCP_SEQ_STATE_LISTENING:
2039 if (st->bucket >= INET_LHTABLE_SIZE)
2041 st->state = TCP_SEQ_STATE_LISTENING;
2042 rc = listening_get_next(seq, NULL);
2043 while (offset-- && rc)
2044 rc = listening_get_next(seq, rc);
2048 st->state = TCP_SEQ_STATE_ESTABLISHED;
2050 case TCP_SEQ_STATE_ESTABLISHED:
2051 if (st->bucket > tcp_hashinfo.ehash_mask)
2053 rc = established_get_first(seq);
2054 while (offset-- && rc)
2055 rc = established_get_next(seq, rc);
2063 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2065 struct tcp_iter_state *st = seq->private;
2068 if (*pos && *pos == st->last_pos) {
2069 rc = tcp_seek_last_pos(seq);
2074 st->state = TCP_SEQ_STATE_LISTENING;
2078 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2081 st->last_pos = *pos;
2085 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2087 struct tcp_iter_state *st = seq->private;
2090 if (v == SEQ_START_TOKEN) {
2091 rc = tcp_get_idx(seq, 0);
2095 switch (st->state) {
2096 case TCP_SEQ_STATE_LISTENING:
2097 rc = listening_get_next(seq, v);
2099 st->state = TCP_SEQ_STATE_ESTABLISHED;
2102 rc = established_get_first(seq);
2105 case TCP_SEQ_STATE_ESTABLISHED:
2106 rc = established_get_next(seq, v);
2111 st->last_pos = *pos;
2115 static void tcp_seq_stop(struct seq_file *seq, void *v)
2117 struct tcp_iter_state *st = seq->private;
2119 switch (st->state) {
2120 case TCP_SEQ_STATE_LISTENING:
2121 if (v != SEQ_START_TOKEN)
2122 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2124 case TCP_SEQ_STATE_ESTABLISHED:
2126 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2131 int tcp_seq_open(struct inode *inode, struct file *file)
2133 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2134 struct tcp_iter_state *s;
2137 err = seq_open_net(inode, file, &afinfo->seq_ops,
2138 sizeof(struct tcp_iter_state));
2142 s = ((struct seq_file *)file->private_data)->private;
2143 s->family = afinfo->family;
2147 EXPORT_SYMBOL(tcp_seq_open);
2149 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2152 struct proc_dir_entry *p;
2154 afinfo->seq_ops.start = tcp_seq_start;
2155 afinfo->seq_ops.next = tcp_seq_next;
2156 afinfo->seq_ops.stop = tcp_seq_stop;
2158 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2159 afinfo->seq_fops, afinfo);
2164 EXPORT_SYMBOL(tcp_proc_register);
2166 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2168 remove_proc_entry(afinfo->name, net->proc_net);
2170 EXPORT_SYMBOL(tcp_proc_unregister);
2172 static void get_openreq4(const struct request_sock *req,
2173 struct seq_file *f, int i)
2175 const struct inet_request_sock *ireq = inet_rsk(req);
2176 long delta = req->rsk_timer.expires - jiffies;
2178 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2179 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2184 ntohs(ireq->ir_rmt_port),
2186 0, 0, /* could print option size, but that is af dependent. */
2187 1, /* timers active (only the expire timer) */
2188 jiffies_delta_to_clock_t(delta),
2190 from_kuid_munged(seq_user_ns(f),
2191 sock_i_uid(req->rsk_listener)),
2192 0, /* non standard timer */
2193 0, /* open_requests have no inode */
2198 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2201 unsigned long timer_expires;
2202 const struct tcp_sock *tp = tcp_sk(sk);
2203 const struct inet_connection_sock *icsk = inet_csk(sk);
2204 const struct inet_sock *inet = inet_sk(sk);
2205 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2206 __be32 dest = inet->inet_daddr;
2207 __be32 src = inet->inet_rcv_saddr;
2208 __u16 destp = ntohs(inet->inet_dport);
2209 __u16 srcp = ntohs(inet->inet_sport);
2213 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2214 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2215 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2217 timer_expires = icsk->icsk_timeout;
2218 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2220 timer_expires = icsk->icsk_timeout;
2221 } else if (timer_pending(&sk->sk_timer)) {
2223 timer_expires = sk->sk_timer.expires;
2226 timer_expires = jiffies;
2229 state = sk_state_load(sk);
2230 if (state == TCP_LISTEN)
2231 rx_queue = sk->sk_ack_backlog;
2233 /* Because we don't lock the socket,
2234 * we might find a transient negative value.
2236 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2238 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2239 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2240 i, src, srcp, dest, destp, state,
2241 tp->write_seq - tp->snd_una,
2244 jiffies_delta_to_clock_t(timer_expires - jiffies),
2245 icsk->icsk_retransmits,
2246 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2247 icsk->icsk_probes_out,
2249 atomic_read(&sk->sk_refcnt), sk,
2250 jiffies_to_clock_t(icsk->icsk_rto),
2251 jiffies_to_clock_t(icsk->icsk_ack.ato),
2252 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2254 state == TCP_LISTEN ?
2255 fastopenq->max_qlen :
2256 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2259 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2260 struct seq_file *f, int i)
2262 long delta = tw->tw_timer.expires - jiffies;
2266 dest = tw->tw_daddr;
2267 src = tw->tw_rcv_saddr;
2268 destp = ntohs(tw->tw_dport);
2269 srcp = ntohs(tw->tw_sport);
2271 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2272 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2273 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2274 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2275 atomic_read(&tw->tw_refcnt), tw);
2280 static int tcp4_seq_show(struct seq_file *seq, void *v)
2282 struct tcp_iter_state *st;
2283 struct sock *sk = v;
2285 seq_setwidth(seq, TMPSZ - 1);
2286 if (v == SEQ_START_TOKEN) {
2287 seq_puts(seq, " sl local_address rem_address st tx_queue "
2288 "rx_queue tr tm->when retrnsmt uid timeout "
2294 if (sk->sk_state == TCP_TIME_WAIT)
2295 get_timewait4_sock(v, seq, st->num);
2296 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2297 get_openreq4(v, seq, st->num);
2299 get_tcp4_sock(v, seq, st->num);
2305 static const struct file_operations tcp_afinfo_seq_fops = {
2306 .owner = THIS_MODULE,
2307 .open = tcp_seq_open,
2309 .llseek = seq_lseek,
2310 .release = seq_release_net
2313 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2316 .seq_fops = &tcp_afinfo_seq_fops,
2318 .show = tcp4_seq_show,
2322 static int __net_init tcp4_proc_init_net(struct net *net)
2324 return tcp_proc_register(net, &tcp4_seq_afinfo);
2327 static void __net_exit tcp4_proc_exit_net(struct net *net)
2329 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2332 static struct pernet_operations tcp4_net_ops = {
2333 .init = tcp4_proc_init_net,
2334 .exit = tcp4_proc_exit_net,
2337 int __init tcp4_proc_init(void)
2339 return register_pernet_subsys(&tcp4_net_ops);
2342 void tcp4_proc_exit(void)
2344 unregister_pernet_subsys(&tcp4_net_ops);
2346 #endif /* CONFIG_PROC_FS */
2348 struct proto tcp_prot = {
2350 .owner = THIS_MODULE,
2352 .connect = tcp_v4_connect,
2353 .disconnect = tcp_disconnect,
2354 .accept = inet_csk_accept,
2356 .init = tcp_v4_init_sock,
2357 .destroy = tcp_v4_destroy_sock,
2358 .shutdown = tcp_shutdown,
2359 .setsockopt = tcp_setsockopt,
2360 .getsockopt = tcp_getsockopt,
2361 .recvmsg = tcp_recvmsg,
2362 .sendmsg = tcp_sendmsg,
2363 .sendpage = tcp_sendpage,
2364 .backlog_rcv = tcp_v4_do_rcv,
2365 .release_cb = tcp_release_cb,
2367 .unhash = inet_unhash,
2368 .get_port = inet_csk_get_port,
2369 .enter_memory_pressure = tcp_enter_memory_pressure,
2370 .stream_memory_free = tcp_stream_memory_free,
2371 .sockets_allocated = &tcp_sockets_allocated,
2372 .orphan_count = &tcp_orphan_count,
2373 .memory_allocated = &tcp_memory_allocated,
2374 .memory_pressure = &tcp_memory_pressure,
2375 .sysctl_mem = sysctl_tcp_mem,
2376 .sysctl_wmem = sysctl_tcp_wmem,
2377 .sysctl_rmem = sysctl_tcp_rmem,
2378 .max_header = MAX_TCP_HEADER,
2379 .obj_size = sizeof(struct tcp_sock),
2380 .slab_flags = SLAB_DESTROY_BY_RCU,
2381 .twsk_prot = &tcp_timewait_sock_ops,
2382 .rsk_prot = &tcp_request_sock_ops,
2383 .h.hashinfo = &tcp_hashinfo,
2384 .no_autobind = true,
2385 #ifdef CONFIG_COMPAT
2386 .compat_setsockopt = compat_tcp_setsockopt,
2387 .compat_getsockopt = compat_tcp_getsockopt,
2389 .diag_destroy = tcp_abort,
2391 EXPORT_SYMBOL(tcp_prot);
2393 static void __net_exit tcp_sk_exit(struct net *net)
2397 for_each_possible_cpu(cpu)
2398 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2399 free_percpu(net->ipv4.tcp_sk);
2402 static int __net_init tcp_sk_init(struct net *net)
2406 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2407 if (!net->ipv4.tcp_sk)
2410 for_each_possible_cpu(cpu) {
2413 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2417 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2418 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2421 net->ipv4.sysctl_tcp_ecn = 2;
2422 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2424 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2425 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2426 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2428 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2429 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2430 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2432 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2433 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2434 net->ipv4.sysctl_tcp_syncookies = 1;
2435 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2436 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2437 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2438 net->ipv4.sysctl_tcp_orphan_retries = 0;
2439 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2440 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2449 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2451 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2454 static struct pernet_operations __net_initdata tcp_sk_ops = {
2455 .init = tcp_sk_init,
2456 .exit = tcp_sk_exit,
2457 .exit_batch = tcp_sk_exit_batch,
2460 void __init tcp_v4_init(void)
2462 inet_hashinfo_init(&tcp_hashinfo);
2463 if (register_pernet_subsys(&tcp_sk_ops))
2464 panic("Failed to create the TCP control socket.\n");