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/tcp_memcontrol.h>
77 #include <net/busy_poll.h>
79 #include <linux/inet.h>
80 #include <linux/ipv6.h>
81 #include <linux/stddef.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
85 #include <linux/crypto.h>
86 #include <linux/scatterlist.h>
88 int sysctl_tcp_tw_reuse __read_mostly;
89 int sysctl_tcp_low_latency __read_mostly;
90 EXPORT_SYMBOL(sysctl_tcp_low_latency);
92 #ifdef CONFIG_TCP_MD5SIG
93 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
94 __be32 daddr, __be32 saddr, const struct tcphdr *th);
97 struct inet_hashinfo tcp_hashinfo;
98 EXPORT_SYMBOL(tcp_hashinfo);
100 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
102 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
105 tcp_hdr(skb)->source);
108 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
110 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
111 struct tcp_sock *tp = tcp_sk(sk);
113 /* With PAWS, it is safe from the viewpoint
114 of data integrity. Even without PAWS it is safe provided sequence
115 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
117 Actually, the idea is close to VJ's one, only timestamp cache is
118 held not per host, but per port pair and TW bucket is used as state
121 If TW bucket has been already destroyed we fall back to VJ's scheme
122 and use initial timestamp retrieved from peer table.
124 if (tcptw->tw_ts_recent_stamp &&
125 (twp == NULL || (sysctl_tcp_tw_reuse &&
126 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
127 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
128 if (tp->write_seq == 0)
130 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
131 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
138 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
140 /* This will initiate an outgoing connection. */
141 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
143 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
144 struct inet_sock *inet = inet_sk(sk);
145 struct tcp_sock *tp = tcp_sk(sk);
146 __be16 orig_sport, orig_dport;
147 __be32 daddr, nexthop;
151 struct ip_options_rcu *inet_opt;
153 if (addr_len < sizeof(struct sockaddr_in))
156 if (usin->sin_family != AF_INET)
157 return -EAFNOSUPPORT;
159 nexthop = daddr = usin->sin_addr.s_addr;
160 inet_opt = rcu_dereference_protected(inet->inet_opt,
161 sock_owned_by_user(sk));
162 if (inet_opt && inet_opt->opt.srr) {
165 nexthop = inet_opt->opt.faddr;
168 orig_sport = inet->inet_sport;
169 orig_dport = usin->sin_port;
170 fl4 = &inet->cork.fl.u.ip4;
171 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
172 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
174 orig_sport, orig_dport, sk);
177 if (err == -ENETUNREACH)
178 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
182 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
187 if (!inet_opt || !inet_opt->opt.srr)
190 if (!inet->inet_saddr)
191 inet->inet_saddr = fl4->saddr;
192 inet->inet_rcv_saddr = inet->inet_saddr;
194 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
195 /* Reset inherited state */
196 tp->rx_opt.ts_recent = 0;
197 tp->rx_opt.ts_recent_stamp = 0;
198 if (likely(!tp->repair))
202 if (tcp_death_row.sysctl_tw_recycle &&
203 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
204 tcp_fetch_timewait_stamp(sk, &rt->dst);
206 inet->inet_dport = usin->sin_port;
207 inet->inet_daddr = daddr;
209 inet_csk(sk)->icsk_ext_hdr_len = 0;
211 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
213 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
215 /* Socket identity is still unknown (sport may be zero).
216 * However we set state to SYN-SENT and not releasing socket
217 * lock select source port, enter ourselves into the hash tables and
218 * complete initialization after this.
220 tcp_set_state(sk, TCP_SYN_SENT);
221 err = inet_hash_connect(&tcp_death_row, sk);
227 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
228 inet->inet_sport, inet->inet_dport, sk);
234 /* OK, now commit destination to socket. */
235 sk->sk_gso_type = SKB_GSO_TCPV4;
236 sk_setup_caps(sk, &rt->dst);
238 if (!tp->write_seq && likely(!tp->repair))
239 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
244 inet->inet_id = tp->write_seq ^ jiffies;
246 err = tcp_connect(sk);
256 * This unhashes the socket and releases the local port,
259 tcp_set_state(sk, TCP_CLOSE);
261 sk->sk_route_caps = 0;
262 inet->inet_dport = 0;
265 EXPORT_SYMBOL(tcp_v4_connect);
268 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
269 * It can be called through tcp_release_cb() if socket was owned by user
270 * at the time tcp_v4_err() was called to handle ICMP message.
272 void tcp_v4_mtu_reduced(struct sock *sk)
274 struct dst_entry *dst;
275 struct inet_sock *inet = inet_sk(sk);
276 u32 mtu = tcp_sk(sk)->mtu_info;
278 dst = inet_csk_update_pmtu(sk, mtu);
282 /* Something is about to be wrong... Remember soft error
283 * for the case, if this connection will not able to recover.
285 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
286 sk->sk_err_soft = EMSGSIZE;
290 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
291 ip_sk_accept_pmtu(sk) &&
292 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
293 tcp_sync_mss(sk, mtu);
295 /* Resend the TCP packet because it's
296 * clear that the old packet has been
297 * dropped. This is the new "fast" path mtu
300 tcp_simple_retransmit(sk);
301 } /* else let the usual retransmit timer handle it */
303 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
305 static void do_redirect(struct sk_buff *skb, struct sock *sk)
307 struct dst_entry *dst = __sk_dst_check(sk, 0);
310 dst->ops->redirect(dst, sk, skb);
314 * This routine is called by the ICMP module when it gets some
315 * sort of error condition. If err < 0 then the socket should
316 * be closed and the error returned to the user. If err > 0
317 * it's just the icmp type << 8 | icmp code. After adjustment
318 * header points to the first 8 bytes of the tcp header. We need
319 * to find the appropriate port.
321 * The locking strategy used here is very "optimistic". When
322 * someone else accesses the socket the ICMP is just dropped
323 * and for some paths there is no check at all.
324 * A more general error queue to queue errors for later handling
325 * is probably better.
329 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
331 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
332 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
333 struct inet_connection_sock *icsk;
335 struct inet_sock *inet;
336 const int type = icmp_hdr(icmp_skb)->type;
337 const int code = icmp_hdr(icmp_skb)->code;
340 struct request_sock *fastopen;
344 struct net *net = dev_net(icmp_skb->dev);
346 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
347 iph->saddr, th->source, inet_iif(icmp_skb));
349 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
352 if (sk->sk_state == TCP_TIME_WAIT) {
353 inet_twsk_put(inet_twsk(sk));
358 /* If too many ICMPs get dropped on busy
359 * servers this needs to be solved differently.
360 * We do take care of PMTU discovery (RFC1191) special case :
361 * we can receive locally generated ICMP messages while socket is held.
363 if (sock_owned_by_user(sk)) {
364 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
365 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
367 if (sk->sk_state == TCP_CLOSE)
370 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
371 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
377 seq = ntohl(th->seq);
378 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
379 fastopen = tp->fastopen_rsk;
380 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
381 if (sk->sk_state != TCP_LISTEN &&
382 !between(seq, snd_una, tp->snd_nxt)) {
383 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
389 do_redirect(icmp_skb, sk);
391 case ICMP_SOURCE_QUENCH:
392 /* Just silently ignore these. */
394 case ICMP_PARAMETERPROB:
397 case ICMP_DEST_UNREACH:
398 if (code > NR_ICMP_UNREACH)
401 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
402 /* We are not interested in TCP_LISTEN and open_requests
403 * (SYN-ACKs send out by Linux are always <576bytes so
404 * they should go through unfragmented).
406 if (sk->sk_state == TCP_LISTEN)
410 if (!sock_owned_by_user(sk)) {
411 tcp_v4_mtu_reduced(sk);
413 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
419 err = icmp_err_convert[code].errno;
420 /* check if icmp_skb allows revert of backoff
421 * (see draft-zimmermann-tcp-lcd) */
422 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
424 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
425 !icsk->icsk_backoff || fastopen)
428 if (sock_owned_by_user(sk))
431 icsk->icsk_backoff--;
432 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
434 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
436 skb = tcp_write_queue_head(sk);
439 remaining = icsk->icsk_rto -
441 tcp_time_stamp - tcp_skb_timestamp(skb));
444 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
445 remaining, TCP_RTO_MAX);
447 /* RTO revert clocked out retransmission.
448 * Will retransmit now */
449 tcp_retransmit_timer(sk);
453 case ICMP_TIME_EXCEEDED:
460 switch (sk->sk_state) {
461 struct request_sock *req, **prev;
463 if (sock_owned_by_user(sk))
466 req = inet_csk_search_req(sk, &prev, th->dest,
467 iph->daddr, iph->saddr);
471 /* ICMPs are not backlogged, hence we cannot get
472 an established socket here.
476 if (seq != tcp_rsk(req)->snt_isn) {
477 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
482 * Still in SYN_RECV, just remove it silently.
483 * There is no good way to pass the error to the newly
484 * created socket, and POSIX does not want network
485 * errors returned from accept().
487 inet_csk_reqsk_queue_drop(sk, req, prev);
488 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
493 /* Only in fast or simultaneous open. If a fast open socket is
494 * is already accepted it is treated as a connected one below.
496 if (fastopen && fastopen->sk == NULL)
499 if (!sock_owned_by_user(sk)) {
502 sk->sk_error_report(sk);
506 sk->sk_err_soft = err;
511 /* If we've already connected we will keep trying
512 * until we time out, or the user gives up.
514 * rfc1122 4.2.3.9 allows to consider as hard errors
515 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
516 * but it is obsoleted by pmtu discovery).
518 * Note, that in modern internet, where routing is unreliable
519 * and in each dark corner broken firewalls sit, sending random
520 * errors ordered by their masters even this two messages finally lose
521 * their original sense (even Linux sends invalid PORT_UNREACHs)
523 * Now we are in compliance with RFCs.
528 if (!sock_owned_by_user(sk) && inet->recverr) {
530 sk->sk_error_report(sk);
531 } else { /* Only an error on timeout */
532 sk->sk_err_soft = err;
540 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
542 struct tcphdr *th = tcp_hdr(skb);
544 if (skb->ip_summed == CHECKSUM_PARTIAL) {
545 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
546 skb->csum_start = skb_transport_header(skb) - skb->head;
547 skb->csum_offset = offsetof(struct tcphdr, check);
549 th->check = tcp_v4_check(skb->len, saddr, daddr,
556 /* This routine computes an IPv4 TCP checksum. */
557 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
559 const struct inet_sock *inet = inet_sk(sk);
561 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
563 EXPORT_SYMBOL(tcp_v4_send_check);
566 * This routine will send an RST to the other tcp.
568 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
570 * Answer: if a packet caused RST, it is not for a socket
571 * existing in our system, if it is matched to a socket,
572 * it is just duplicate segment or bug in other side's TCP.
573 * So that we build reply only basing on parameters
574 * arrived with segment.
575 * Exception: precedence violation. We do not implement it in any case.
578 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
580 const struct tcphdr *th = tcp_hdr(skb);
583 #ifdef CONFIG_TCP_MD5SIG
584 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
587 struct ip_reply_arg arg;
588 #ifdef CONFIG_TCP_MD5SIG
589 struct tcp_md5sig_key *key;
590 const __u8 *hash_location = NULL;
591 unsigned char newhash[16];
593 struct sock *sk1 = NULL;
597 /* Never send a reset in response to a reset. */
601 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
604 /* Swap the send and the receive. */
605 memset(&rep, 0, sizeof(rep));
606 rep.th.dest = th->source;
607 rep.th.source = th->dest;
608 rep.th.doff = sizeof(struct tcphdr) / 4;
612 rep.th.seq = th->ack_seq;
615 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
616 skb->len - (th->doff << 2));
619 memset(&arg, 0, sizeof(arg));
620 arg.iov[0].iov_base = (unsigned char *)&rep;
621 arg.iov[0].iov_len = sizeof(rep.th);
623 #ifdef CONFIG_TCP_MD5SIG
624 hash_location = tcp_parse_md5sig_option(th);
625 if (!sk && hash_location) {
627 * active side is lost. Try to find listening socket through
628 * source port, and then find md5 key through listening socket.
629 * we are not loose security here:
630 * Incoming packet is checked with md5 hash with finding key,
631 * no RST generated if md5 hash doesn't match.
633 sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev),
634 &tcp_hashinfo, ip_hdr(skb)->saddr,
635 th->source, ip_hdr(skb)->daddr,
636 ntohs(th->source), inet_iif(skb));
637 /* don't send rst if it can't find key */
641 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
642 &ip_hdr(skb)->saddr, AF_INET);
646 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb);
647 if (genhash || memcmp(hash_location, newhash, 16) != 0)
650 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
656 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
658 (TCPOPT_MD5SIG << 8) |
660 /* Update length and the length the header thinks exists */
661 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
662 rep.th.doff = arg.iov[0].iov_len / 4;
664 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
665 key, ip_hdr(skb)->saddr,
666 ip_hdr(skb)->daddr, &rep.th);
669 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
670 ip_hdr(skb)->saddr, /* XXX */
671 arg.iov[0].iov_len, IPPROTO_TCP, 0);
672 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
673 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
674 /* When socket is gone, all binding information is lost.
675 * routing might fail in this case. No choice here, if we choose to force
676 * input interface, we will misroute in case of asymmetric route.
679 arg.bound_dev_if = sk->sk_bound_dev_if;
681 net = dev_net(skb_dst(skb)->dev);
682 arg.tos = ip_hdr(skb)->tos;
683 ip_send_unicast_reply(net, skb, &TCP_SKB_CB(skb)->header.h4.opt,
684 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
685 &arg, arg.iov[0].iov_len);
687 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
688 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
690 #ifdef CONFIG_TCP_MD5SIG
699 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
700 outside socket context is ugly, certainly. What can I do?
703 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
704 u32 win, u32 tsval, u32 tsecr, int oif,
705 struct tcp_md5sig_key *key,
706 int reply_flags, u8 tos)
708 const struct tcphdr *th = tcp_hdr(skb);
711 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
712 #ifdef CONFIG_TCP_MD5SIG
713 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
717 struct ip_reply_arg arg;
718 struct net *net = dev_net(skb_dst(skb)->dev);
720 memset(&rep.th, 0, sizeof(struct tcphdr));
721 memset(&arg, 0, sizeof(arg));
723 arg.iov[0].iov_base = (unsigned char *)&rep;
724 arg.iov[0].iov_len = sizeof(rep.th);
726 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
727 (TCPOPT_TIMESTAMP << 8) |
729 rep.opt[1] = htonl(tsval);
730 rep.opt[2] = htonl(tsecr);
731 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
734 /* Swap the send and the receive. */
735 rep.th.dest = th->source;
736 rep.th.source = th->dest;
737 rep.th.doff = arg.iov[0].iov_len / 4;
738 rep.th.seq = htonl(seq);
739 rep.th.ack_seq = htonl(ack);
741 rep.th.window = htons(win);
743 #ifdef CONFIG_TCP_MD5SIG
745 int offset = (tsecr) ? 3 : 0;
747 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
749 (TCPOPT_MD5SIG << 8) |
751 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
752 rep.th.doff = arg.iov[0].iov_len/4;
754 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
755 key, ip_hdr(skb)->saddr,
756 ip_hdr(skb)->daddr, &rep.th);
759 arg.flags = reply_flags;
760 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
761 ip_hdr(skb)->saddr, /* XXX */
762 arg.iov[0].iov_len, IPPROTO_TCP, 0);
763 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
765 arg.bound_dev_if = oif;
767 ip_send_unicast_reply(net, skb, &TCP_SKB_CB(skb)->header.h4.opt,
768 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
769 &arg, arg.iov[0].iov_len);
771 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
774 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
776 struct inet_timewait_sock *tw = inet_twsk(sk);
777 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
779 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
780 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
781 tcp_time_stamp + tcptw->tw_ts_offset,
784 tcp_twsk_md5_key(tcptw),
785 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
792 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
793 struct request_sock *req)
795 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
796 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
798 tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
799 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
800 tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
804 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
806 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
811 * Send a SYN-ACK after having received a SYN.
812 * This still operates on a request_sock only, not on a big
815 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
817 struct request_sock *req,
819 struct tcp_fastopen_cookie *foc)
821 const struct inet_request_sock *ireq = inet_rsk(req);
826 /* First, grab a route. */
827 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
830 skb = tcp_make_synack(sk, dst, req, foc);
833 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
835 skb_set_queue_mapping(skb, queue_mapping);
836 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
839 err = net_xmit_eval(err);
846 * IPv4 request_sock destructor.
848 static void tcp_v4_reqsk_destructor(struct request_sock *req)
850 kfree(inet_rsk(req)->opt);
854 * Return true if a syncookie should be sent
856 bool tcp_syn_flood_action(struct sock *sk,
857 const struct sk_buff *skb,
860 const char *msg = "Dropping request";
861 bool want_cookie = false;
862 struct listen_sock *lopt;
864 #ifdef CONFIG_SYN_COOKIES
865 if (sysctl_tcp_syncookies) {
866 msg = "Sending cookies";
868 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
871 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
873 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
874 if (!lopt->synflood_warned && sysctl_tcp_syncookies != 2) {
875 lopt->synflood_warned = 1;
876 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
877 proto, ntohs(tcp_hdr(skb)->dest), msg);
881 EXPORT_SYMBOL(tcp_syn_flood_action);
883 #ifdef CONFIG_TCP_MD5SIG
885 * RFC2385 MD5 checksumming requires a mapping of
886 * IP address->MD5 Key.
887 * We need to maintain these in the sk structure.
890 /* Find the Key structure for an address. */
891 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
892 const union tcp_md5_addr *addr,
895 struct tcp_sock *tp = tcp_sk(sk);
896 struct tcp_md5sig_key *key;
897 unsigned int size = sizeof(struct in_addr);
898 struct tcp_md5sig_info *md5sig;
900 /* caller either holds rcu_read_lock() or socket lock */
901 md5sig = rcu_dereference_check(tp->md5sig_info,
902 sock_owned_by_user(sk) ||
903 lockdep_is_held(&sk->sk_lock.slock));
906 #if IS_ENABLED(CONFIG_IPV6)
907 if (family == AF_INET6)
908 size = sizeof(struct in6_addr);
910 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
911 if (key->family != family)
913 if (!memcmp(&key->addr, addr, size))
918 EXPORT_SYMBOL(tcp_md5_do_lookup);
920 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
921 struct sock *addr_sk)
923 union tcp_md5_addr *addr;
925 addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
926 return tcp_md5_do_lookup(sk, addr, AF_INET);
928 EXPORT_SYMBOL(tcp_v4_md5_lookup);
930 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
931 struct request_sock *req)
933 union tcp_md5_addr *addr;
935 addr = (union tcp_md5_addr *)&inet_rsk(req)->ir_rmt_addr;
936 return tcp_md5_do_lookup(sk, addr, AF_INET);
939 /* This can be called on a newly created socket, from other files */
940 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
941 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
943 /* Add Key to the list */
944 struct tcp_md5sig_key *key;
945 struct tcp_sock *tp = tcp_sk(sk);
946 struct tcp_md5sig_info *md5sig;
948 key = tcp_md5_do_lookup(sk, addr, family);
950 /* Pre-existing entry - just update that one. */
951 memcpy(key->key, newkey, newkeylen);
952 key->keylen = newkeylen;
956 md5sig = rcu_dereference_protected(tp->md5sig_info,
957 sock_owned_by_user(sk));
959 md5sig = kmalloc(sizeof(*md5sig), gfp);
963 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
964 INIT_HLIST_HEAD(&md5sig->head);
965 rcu_assign_pointer(tp->md5sig_info, md5sig);
968 key = sock_kmalloc(sk, sizeof(*key), gfp);
971 if (!tcp_alloc_md5sig_pool()) {
972 sock_kfree_s(sk, key, sizeof(*key));
976 memcpy(key->key, newkey, newkeylen);
977 key->keylen = newkeylen;
978 key->family = family;
979 memcpy(&key->addr, addr,
980 (family == AF_INET6) ? sizeof(struct in6_addr) :
981 sizeof(struct in_addr));
982 hlist_add_head_rcu(&key->node, &md5sig->head);
985 EXPORT_SYMBOL(tcp_md5_do_add);
987 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
989 struct tcp_md5sig_key *key;
991 key = tcp_md5_do_lookup(sk, addr, family);
994 hlist_del_rcu(&key->node);
995 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
999 EXPORT_SYMBOL(tcp_md5_do_del);
1001 static void tcp_clear_md5_list(struct sock *sk)
1003 struct tcp_sock *tp = tcp_sk(sk);
1004 struct tcp_md5sig_key *key;
1005 struct hlist_node *n;
1006 struct tcp_md5sig_info *md5sig;
1008 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1010 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1011 hlist_del_rcu(&key->node);
1012 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1013 kfree_rcu(key, rcu);
1017 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1020 struct tcp_md5sig cmd;
1021 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1023 if (optlen < sizeof(cmd))
1026 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1029 if (sin->sin_family != AF_INET)
1032 if (!cmd.tcpm_keylen)
1033 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1036 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1039 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1040 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1044 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1045 __be32 daddr, __be32 saddr, int nbytes)
1047 struct tcp4_pseudohdr *bp;
1048 struct scatterlist sg;
1050 bp = &hp->md5_blk.ip4;
1053 * 1. the TCP pseudo-header (in the order: source IP address,
1054 * destination IP address, zero-padded protocol number, and
1060 bp->protocol = IPPROTO_TCP;
1061 bp->len = cpu_to_be16(nbytes);
1063 sg_init_one(&sg, bp, sizeof(*bp));
1064 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1067 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1068 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1070 struct tcp_md5sig_pool *hp;
1071 struct hash_desc *desc;
1073 hp = tcp_get_md5sig_pool();
1075 goto clear_hash_noput;
1076 desc = &hp->md5_desc;
1078 if (crypto_hash_init(desc))
1080 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1082 if (tcp_md5_hash_header(hp, th))
1084 if (tcp_md5_hash_key(hp, key))
1086 if (crypto_hash_final(desc, md5_hash))
1089 tcp_put_md5sig_pool();
1093 tcp_put_md5sig_pool();
1095 memset(md5_hash, 0, 16);
1099 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1100 const struct sock *sk, const struct request_sock *req,
1101 const struct sk_buff *skb)
1103 struct tcp_md5sig_pool *hp;
1104 struct hash_desc *desc;
1105 const struct tcphdr *th = tcp_hdr(skb);
1106 __be32 saddr, daddr;
1109 saddr = inet_sk(sk)->inet_saddr;
1110 daddr = inet_sk(sk)->inet_daddr;
1112 saddr = inet_rsk(req)->ir_loc_addr;
1113 daddr = inet_rsk(req)->ir_rmt_addr;
1115 const struct iphdr *iph = ip_hdr(skb);
1120 hp = tcp_get_md5sig_pool();
1122 goto clear_hash_noput;
1123 desc = &hp->md5_desc;
1125 if (crypto_hash_init(desc))
1128 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1130 if (tcp_md5_hash_header(hp, th))
1132 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1134 if (tcp_md5_hash_key(hp, key))
1136 if (crypto_hash_final(desc, md5_hash))
1139 tcp_put_md5sig_pool();
1143 tcp_put_md5sig_pool();
1145 memset(md5_hash, 0, 16);
1148 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1150 static bool __tcp_v4_inbound_md5_hash(struct sock *sk,
1151 const struct sk_buff *skb)
1154 * This gets called for each TCP segment that arrives
1155 * so we want to be efficient.
1156 * We have 3 drop cases:
1157 * o No MD5 hash and one expected.
1158 * o MD5 hash and we're not expecting one.
1159 * o MD5 hash and its wrong.
1161 const __u8 *hash_location = NULL;
1162 struct tcp_md5sig_key *hash_expected;
1163 const struct iphdr *iph = ip_hdr(skb);
1164 const struct tcphdr *th = tcp_hdr(skb);
1166 unsigned char newhash[16];
1168 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1170 hash_location = tcp_parse_md5sig_option(th);
1172 /* We've parsed the options - do we have a hash? */
1173 if (!hash_expected && !hash_location)
1176 if (hash_expected && !hash_location) {
1177 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1181 if (!hash_expected && hash_location) {
1182 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1186 /* Okay, so this is hash_expected and hash_location -
1187 * so we need to calculate the checksum.
1189 genhash = tcp_v4_md5_hash_skb(newhash,
1193 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1194 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1195 &iph->saddr, ntohs(th->source),
1196 &iph->daddr, ntohs(th->dest),
1197 genhash ? " tcp_v4_calc_md5_hash failed"
1204 static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1209 ret = __tcp_v4_inbound_md5_hash(sk, skb);
1217 static void tcp_v4_init_req(struct request_sock *req, struct sock *sk,
1218 struct sk_buff *skb)
1220 struct inet_request_sock *ireq = inet_rsk(req);
1222 ireq->ir_loc_addr = ip_hdr(skb)->daddr;
1223 ireq->ir_rmt_addr = ip_hdr(skb)->saddr;
1224 ireq->no_srccheck = inet_sk(sk)->transparent;
1225 ireq->opt = tcp_v4_save_options(skb);
1228 static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl,
1229 const struct request_sock *req,
1232 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1235 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1244 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1246 .obj_size = sizeof(struct tcp_request_sock),
1247 .rtx_syn_ack = tcp_rtx_synack,
1248 .send_ack = tcp_v4_reqsk_send_ack,
1249 .destructor = tcp_v4_reqsk_destructor,
1250 .send_reset = tcp_v4_send_reset,
1251 .syn_ack_timeout = tcp_syn_ack_timeout,
1254 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1255 .mss_clamp = TCP_MSS_DEFAULT,
1256 #ifdef CONFIG_TCP_MD5SIG
1257 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1258 .calc_md5_hash = tcp_v4_md5_hash_skb,
1260 .init_req = tcp_v4_init_req,
1261 #ifdef CONFIG_SYN_COOKIES
1262 .cookie_init_seq = cookie_v4_init_sequence,
1264 .route_req = tcp_v4_route_req,
1265 .init_seq = tcp_v4_init_sequence,
1266 .send_synack = tcp_v4_send_synack,
1267 .queue_hash_add = inet_csk_reqsk_queue_hash_add,
1270 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1272 /* Never answer to SYNs send to broadcast or multicast */
1273 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1276 return tcp_conn_request(&tcp_request_sock_ops,
1277 &tcp_request_sock_ipv4_ops, sk, skb);
1280 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1283 EXPORT_SYMBOL(tcp_v4_conn_request);
1287 * The three way handshake has completed - we got a valid synack -
1288 * now create the new socket.
1290 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1291 struct request_sock *req,
1292 struct dst_entry *dst)
1294 struct inet_request_sock *ireq;
1295 struct inet_sock *newinet;
1296 struct tcp_sock *newtp;
1298 #ifdef CONFIG_TCP_MD5SIG
1299 struct tcp_md5sig_key *key;
1301 struct ip_options_rcu *inet_opt;
1303 if (sk_acceptq_is_full(sk))
1306 newsk = tcp_create_openreq_child(sk, req, skb);
1310 newsk->sk_gso_type = SKB_GSO_TCPV4;
1311 inet_sk_rx_dst_set(newsk, skb);
1313 newtp = tcp_sk(newsk);
1314 newinet = inet_sk(newsk);
1315 ireq = inet_rsk(req);
1316 newinet->inet_daddr = ireq->ir_rmt_addr;
1317 newinet->inet_rcv_saddr = ireq->ir_loc_addr;
1318 newinet->inet_saddr = ireq->ir_loc_addr;
1319 inet_opt = ireq->opt;
1320 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1322 newinet->mc_index = inet_iif(skb);
1323 newinet->mc_ttl = ip_hdr(skb)->ttl;
1324 newinet->rcv_tos = ip_hdr(skb)->tos;
1325 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1326 inet_set_txhash(newsk);
1328 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1329 newinet->inet_id = newtp->write_seq ^ jiffies;
1332 dst = inet_csk_route_child_sock(sk, newsk, req);
1336 /* syncookie case : see end of cookie_v4_check() */
1338 sk_setup_caps(newsk, dst);
1340 tcp_sync_mss(newsk, dst_mtu(dst));
1341 newtp->advmss = dst_metric_advmss(dst);
1342 if (tcp_sk(sk)->rx_opt.user_mss &&
1343 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1344 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1346 tcp_initialize_rcv_mss(newsk);
1348 #ifdef CONFIG_TCP_MD5SIG
1349 /* Copy over the MD5 key from the original socket */
1350 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1354 * We're using one, so create a matching key
1355 * on the newsk structure. If we fail to get
1356 * memory, then we end up not copying the key
1359 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1360 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1361 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1365 if (__inet_inherit_port(sk, newsk) < 0)
1367 __inet_hash_nolisten(newsk, NULL);
1372 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1376 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1379 inet_csk_prepare_forced_close(newsk);
1383 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1385 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1387 struct tcphdr *th = tcp_hdr(skb);
1388 const struct iphdr *iph = ip_hdr(skb);
1390 struct request_sock **prev;
1391 /* Find possible connection requests. */
1392 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1393 iph->saddr, iph->daddr);
1395 return tcp_check_req(sk, skb, req, prev, false);
1397 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1398 th->source, iph->daddr, th->dest, inet_iif(skb));
1401 if (nsk->sk_state != TCP_TIME_WAIT) {
1405 inet_twsk_put(inet_twsk(nsk));
1409 #ifdef CONFIG_SYN_COOKIES
1411 sk = cookie_v4_check(sk, skb);
1416 /* The socket must have it's spinlock held when we get
1419 * We have a potential double-lock case here, so even when
1420 * doing backlog processing we use the BH locking scheme.
1421 * This is because we cannot sleep with the original spinlock
1424 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1428 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1429 struct dst_entry *dst = sk->sk_rx_dst;
1431 sock_rps_save_rxhash(sk, skb);
1432 sk_mark_napi_id(sk, skb);
1434 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1435 dst->ops->check(dst, 0) == NULL) {
1437 sk->sk_rx_dst = NULL;
1440 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1444 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1447 if (sk->sk_state == TCP_LISTEN) {
1448 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1453 sock_rps_save_rxhash(nsk, skb);
1454 sk_mark_napi_id(sk, skb);
1455 if (tcp_child_process(sk, nsk, skb)) {
1462 sock_rps_save_rxhash(sk, skb);
1464 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1471 tcp_v4_send_reset(rsk, skb);
1474 /* Be careful here. If this function gets more complicated and
1475 * gcc suffers from register pressure on the x86, sk (in %ebx)
1476 * might be destroyed here. This current version compiles correctly,
1477 * but you have been warned.
1482 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1483 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1486 EXPORT_SYMBOL(tcp_v4_do_rcv);
1488 void tcp_v4_early_demux(struct sk_buff *skb)
1490 const struct iphdr *iph;
1491 const struct tcphdr *th;
1494 if (skb->pkt_type != PACKET_HOST)
1497 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1503 if (th->doff < sizeof(struct tcphdr) / 4)
1506 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1507 iph->saddr, th->source,
1508 iph->daddr, ntohs(th->dest),
1512 skb->destructor = sock_edemux;
1513 if (sk->sk_state != TCP_TIME_WAIT) {
1514 struct dst_entry *dst = sk->sk_rx_dst;
1517 dst = dst_check(dst, 0);
1519 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1520 skb_dst_set_noref(skb, dst);
1525 /* Packet is added to VJ-style prequeue for processing in process
1526 * context, if a reader task is waiting. Apparently, this exciting
1527 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1528 * failed somewhere. Latency? Burstiness? Well, at least now we will
1529 * see, why it failed. 8)8) --ANK
1532 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1534 struct tcp_sock *tp = tcp_sk(sk);
1536 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1539 if (skb->len <= tcp_hdrlen(skb) &&
1540 skb_queue_len(&tp->ucopy.prequeue) == 0)
1543 /* Before escaping RCU protected region, we need to take care of skb
1544 * dst. Prequeue is only enabled for established sockets.
1545 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1546 * Instead of doing full sk_rx_dst validity here, let's perform
1547 * an optimistic check.
1549 if (likely(sk->sk_rx_dst))
1554 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1555 tp->ucopy.memory += skb->truesize;
1556 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1557 struct sk_buff *skb1;
1559 BUG_ON(sock_owned_by_user(sk));
1561 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1562 sk_backlog_rcv(sk, skb1);
1563 NET_INC_STATS_BH(sock_net(sk),
1564 LINUX_MIB_TCPPREQUEUEDROPPED);
1567 tp->ucopy.memory = 0;
1568 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1569 wake_up_interruptible_sync_poll(sk_sleep(sk),
1570 POLLIN | POLLRDNORM | POLLRDBAND);
1571 if (!inet_csk_ack_scheduled(sk))
1572 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1573 (3 * tcp_rto_min(sk)) / 4,
1578 EXPORT_SYMBOL(tcp_prequeue);
1584 int tcp_v4_rcv(struct sk_buff *skb)
1586 const struct iphdr *iph;
1587 const struct tcphdr *th;
1590 struct net *net = dev_net(skb->dev);
1592 if (skb->pkt_type != PACKET_HOST)
1595 /* Count it even if it's bad */
1596 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1598 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1603 if (th->doff < sizeof(struct tcphdr) / 4)
1605 if (!pskb_may_pull(skb, th->doff * 4))
1608 /* An explanation is required here, I think.
1609 * Packet length and doff are validated by header prediction,
1610 * provided case of th->doff==0 is eliminated.
1611 * So, we defer the checks. */
1613 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1618 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1619 * barrier() makes sure compiler wont play fool^Waliasing games.
1621 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1622 sizeof(struct inet_skb_parm));
1625 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1626 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1627 skb->len - th->doff * 4);
1628 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1629 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1630 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1631 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1632 TCP_SKB_CB(skb)->sacked = 0;
1634 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1639 if (sk->sk_state == TCP_TIME_WAIT)
1642 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1643 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1644 goto discard_and_relse;
1647 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1648 goto discard_and_relse;
1650 #ifdef CONFIG_TCP_MD5SIG
1652 * We really want to reject the packet as early as possible
1654 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1655 * o There is an MD5 option and we're not expecting one
1657 if (tcp_v4_inbound_md5_hash(sk, skb))
1658 goto discard_and_relse;
1663 if (sk_filter(sk, skb))
1664 goto discard_and_relse;
1668 bh_lock_sock_nested(sk);
1670 if (!sock_owned_by_user(sk)) {
1671 if (!tcp_prequeue(sk, skb))
1672 ret = tcp_v4_do_rcv(sk, skb);
1673 } else if (unlikely(sk_add_backlog(sk, skb,
1674 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1676 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1677 goto discard_and_relse;
1686 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1689 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1691 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1693 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1695 tcp_v4_send_reset(NULL, skb);
1699 /* Discard frame. */
1708 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1709 inet_twsk_put(inet_twsk(sk));
1713 if (skb->len < (th->doff << 2)) {
1714 inet_twsk_put(inet_twsk(sk));
1717 if (tcp_checksum_complete(skb)) {
1718 inet_twsk_put(inet_twsk(sk));
1721 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1723 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1725 iph->saddr, th->source,
1726 iph->daddr, th->dest,
1729 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1730 inet_twsk_put(inet_twsk(sk));
1734 /* Fall through to ACK */
1737 tcp_v4_timewait_ack(sk, skb);
1741 case TCP_TW_SUCCESS:;
1746 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1747 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1748 .twsk_unique = tcp_twsk_unique,
1749 .twsk_destructor= tcp_twsk_destructor,
1752 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1754 struct dst_entry *dst = skb_dst(skb);
1758 sk->sk_rx_dst = dst;
1759 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1762 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1764 const struct inet_connection_sock_af_ops ipv4_specific = {
1765 .queue_xmit = ip_queue_xmit,
1766 .send_check = tcp_v4_send_check,
1767 .rebuild_header = inet_sk_rebuild_header,
1768 .sk_rx_dst_set = inet_sk_rx_dst_set,
1769 .conn_request = tcp_v4_conn_request,
1770 .syn_recv_sock = tcp_v4_syn_recv_sock,
1771 .net_header_len = sizeof(struct iphdr),
1772 .setsockopt = ip_setsockopt,
1773 .getsockopt = ip_getsockopt,
1774 .addr2sockaddr = inet_csk_addr2sockaddr,
1775 .sockaddr_len = sizeof(struct sockaddr_in),
1776 .bind_conflict = inet_csk_bind_conflict,
1777 #ifdef CONFIG_COMPAT
1778 .compat_setsockopt = compat_ip_setsockopt,
1779 .compat_getsockopt = compat_ip_getsockopt,
1781 .mtu_reduced = tcp_v4_mtu_reduced,
1783 EXPORT_SYMBOL(ipv4_specific);
1785 #ifdef CONFIG_TCP_MD5SIG
1786 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1787 .md5_lookup = tcp_v4_md5_lookup,
1788 .calc_md5_hash = tcp_v4_md5_hash_skb,
1789 .md5_parse = tcp_v4_parse_md5_keys,
1793 /* NOTE: A lot of things set to zero explicitly by call to
1794 * sk_alloc() so need not be done here.
1796 static int tcp_v4_init_sock(struct sock *sk)
1798 struct inet_connection_sock *icsk = inet_csk(sk);
1802 icsk->icsk_af_ops = &ipv4_specific;
1804 #ifdef CONFIG_TCP_MD5SIG
1805 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1811 void tcp_v4_destroy_sock(struct sock *sk)
1813 struct tcp_sock *tp = tcp_sk(sk);
1815 tcp_clear_xmit_timers(sk);
1817 tcp_cleanup_congestion_control(sk);
1819 /* Cleanup up the write buffer. */
1820 tcp_write_queue_purge(sk);
1822 /* Cleans up our, hopefully empty, out_of_order_queue. */
1823 __skb_queue_purge(&tp->out_of_order_queue);
1825 #ifdef CONFIG_TCP_MD5SIG
1826 /* Clean up the MD5 key list, if any */
1827 if (tp->md5sig_info) {
1828 tcp_clear_md5_list(sk);
1829 kfree_rcu(tp->md5sig_info, rcu);
1830 tp->md5sig_info = NULL;
1834 /* Clean prequeue, it must be empty really */
1835 __skb_queue_purge(&tp->ucopy.prequeue);
1837 /* Clean up a referenced TCP bind bucket. */
1838 if (inet_csk(sk)->icsk_bind_hash)
1841 BUG_ON(tp->fastopen_rsk != NULL);
1843 /* If socket is aborted during connect operation */
1844 tcp_free_fastopen_req(tp);
1846 sk_sockets_allocated_dec(sk);
1847 sock_release_memcg(sk);
1849 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1851 #ifdef CONFIG_PROC_FS
1852 /* Proc filesystem TCP sock list dumping. */
1855 * Get next listener socket follow cur. If cur is NULL, get first socket
1856 * starting from bucket given in st->bucket; when st->bucket is zero the
1857 * very first socket in the hash table is returned.
1859 static void *listening_get_next(struct seq_file *seq, void *cur)
1861 struct inet_connection_sock *icsk;
1862 struct hlist_nulls_node *node;
1863 struct sock *sk = cur;
1864 struct inet_listen_hashbucket *ilb;
1865 struct tcp_iter_state *st = seq->private;
1866 struct net *net = seq_file_net(seq);
1869 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1870 spin_lock_bh(&ilb->lock);
1871 sk = sk_nulls_head(&ilb->head);
1875 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1879 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1880 struct request_sock *req = cur;
1882 icsk = inet_csk(st->syn_wait_sk);
1886 if (req->rsk_ops->family == st->family) {
1892 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1895 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1897 sk = sk_nulls_next(st->syn_wait_sk);
1898 st->state = TCP_SEQ_STATE_LISTENING;
1899 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1901 icsk = inet_csk(sk);
1902 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1903 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1905 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1906 sk = sk_nulls_next(sk);
1909 sk_nulls_for_each_from(sk, node) {
1910 if (!net_eq(sock_net(sk), net))
1912 if (sk->sk_family == st->family) {
1916 icsk = inet_csk(sk);
1917 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1918 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1920 st->uid = sock_i_uid(sk);
1921 st->syn_wait_sk = sk;
1922 st->state = TCP_SEQ_STATE_OPENREQ;
1926 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1928 spin_unlock_bh(&ilb->lock);
1930 if (++st->bucket < INET_LHTABLE_SIZE) {
1931 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1932 spin_lock_bh(&ilb->lock);
1933 sk = sk_nulls_head(&ilb->head);
1941 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1943 struct tcp_iter_state *st = seq->private;
1948 rc = listening_get_next(seq, NULL);
1950 while (rc && *pos) {
1951 rc = listening_get_next(seq, rc);
1957 static inline bool empty_bucket(const struct tcp_iter_state *st)
1959 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1963 * Get first established socket starting from bucket given in st->bucket.
1964 * If st->bucket is zero, the very first socket in the hash is returned.
1966 static void *established_get_first(struct seq_file *seq)
1968 struct tcp_iter_state *st = seq->private;
1969 struct net *net = seq_file_net(seq);
1973 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1975 struct hlist_nulls_node *node;
1976 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1978 /* Lockless fast path for the common case of empty buckets */
1979 if (empty_bucket(st))
1983 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1984 if (sk->sk_family != st->family ||
1985 !net_eq(sock_net(sk), net)) {
1991 spin_unlock_bh(lock);
1997 static void *established_get_next(struct seq_file *seq, void *cur)
1999 struct sock *sk = cur;
2000 struct hlist_nulls_node *node;
2001 struct tcp_iter_state *st = seq->private;
2002 struct net *net = seq_file_net(seq);
2007 sk = sk_nulls_next(sk);
2009 sk_nulls_for_each_from(sk, node) {
2010 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2014 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2016 return established_get_first(seq);
2019 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2021 struct tcp_iter_state *st = seq->private;
2025 rc = established_get_first(seq);
2028 rc = established_get_next(seq, rc);
2034 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2037 struct tcp_iter_state *st = seq->private;
2039 st->state = TCP_SEQ_STATE_LISTENING;
2040 rc = listening_get_idx(seq, &pos);
2043 st->state = TCP_SEQ_STATE_ESTABLISHED;
2044 rc = established_get_idx(seq, pos);
2050 static void *tcp_seek_last_pos(struct seq_file *seq)
2052 struct tcp_iter_state *st = seq->private;
2053 int offset = st->offset;
2054 int orig_num = st->num;
2057 switch (st->state) {
2058 case TCP_SEQ_STATE_OPENREQ:
2059 case TCP_SEQ_STATE_LISTENING:
2060 if (st->bucket >= INET_LHTABLE_SIZE)
2062 st->state = TCP_SEQ_STATE_LISTENING;
2063 rc = listening_get_next(seq, NULL);
2064 while (offset-- && rc)
2065 rc = listening_get_next(seq, rc);
2069 st->state = TCP_SEQ_STATE_ESTABLISHED;
2071 case TCP_SEQ_STATE_ESTABLISHED:
2072 if (st->bucket > tcp_hashinfo.ehash_mask)
2074 rc = established_get_first(seq);
2075 while (offset-- && rc)
2076 rc = established_get_next(seq, rc);
2084 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2086 struct tcp_iter_state *st = seq->private;
2089 if (*pos && *pos == st->last_pos) {
2090 rc = tcp_seek_last_pos(seq);
2095 st->state = TCP_SEQ_STATE_LISTENING;
2099 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2102 st->last_pos = *pos;
2106 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2108 struct tcp_iter_state *st = seq->private;
2111 if (v == SEQ_START_TOKEN) {
2112 rc = tcp_get_idx(seq, 0);
2116 switch (st->state) {
2117 case TCP_SEQ_STATE_OPENREQ:
2118 case TCP_SEQ_STATE_LISTENING:
2119 rc = listening_get_next(seq, v);
2121 st->state = TCP_SEQ_STATE_ESTABLISHED;
2124 rc = established_get_first(seq);
2127 case TCP_SEQ_STATE_ESTABLISHED:
2128 rc = established_get_next(seq, v);
2133 st->last_pos = *pos;
2137 static void tcp_seq_stop(struct seq_file *seq, void *v)
2139 struct tcp_iter_state *st = seq->private;
2141 switch (st->state) {
2142 case TCP_SEQ_STATE_OPENREQ:
2144 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2145 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2147 case TCP_SEQ_STATE_LISTENING:
2148 if (v != SEQ_START_TOKEN)
2149 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2151 case TCP_SEQ_STATE_ESTABLISHED:
2153 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2158 int tcp_seq_open(struct inode *inode, struct file *file)
2160 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2161 struct tcp_iter_state *s;
2164 err = seq_open_net(inode, file, &afinfo->seq_ops,
2165 sizeof(struct tcp_iter_state));
2169 s = ((struct seq_file *)file->private_data)->private;
2170 s->family = afinfo->family;
2174 EXPORT_SYMBOL(tcp_seq_open);
2176 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2179 struct proc_dir_entry *p;
2181 afinfo->seq_ops.start = tcp_seq_start;
2182 afinfo->seq_ops.next = tcp_seq_next;
2183 afinfo->seq_ops.stop = tcp_seq_stop;
2185 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2186 afinfo->seq_fops, afinfo);
2191 EXPORT_SYMBOL(tcp_proc_register);
2193 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2195 remove_proc_entry(afinfo->name, net->proc_net);
2197 EXPORT_SYMBOL(tcp_proc_unregister);
2199 static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2200 struct seq_file *f, int i, kuid_t uid)
2202 const struct inet_request_sock *ireq = inet_rsk(req);
2203 long delta = req->expires - jiffies;
2205 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2206 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2209 ntohs(inet_sk(sk)->inet_sport),
2211 ntohs(ireq->ir_rmt_port),
2213 0, 0, /* could print option size, but that is af dependent. */
2214 1, /* timers active (only the expire timer) */
2215 jiffies_delta_to_clock_t(delta),
2217 from_kuid_munged(seq_user_ns(f), uid),
2218 0, /* non standard timer */
2219 0, /* open_requests have no inode */
2220 atomic_read(&sk->sk_refcnt),
2224 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2227 unsigned long timer_expires;
2228 const struct tcp_sock *tp = tcp_sk(sk);
2229 const struct inet_connection_sock *icsk = inet_csk(sk);
2230 const struct inet_sock *inet = inet_sk(sk);
2231 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2232 __be32 dest = inet->inet_daddr;
2233 __be32 src = inet->inet_rcv_saddr;
2234 __u16 destp = ntohs(inet->inet_dport);
2235 __u16 srcp = ntohs(inet->inet_sport);
2238 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2239 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2240 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2242 timer_expires = icsk->icsk_timeout;
2243 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2245 timer_expires = icsk->icsk_timeout;
2246 } else if (timer_pending(&sk->sk_timer)) {
2248 timer_expires = sk->sk_timer.expires;
2251 timer_expires = jiffies;
2254 if (sk->sk_state == TCP_LISTEN)
2255 rx_queue = sk->sk_ack_backlog;
2258 * because we dont lock socket, we might find a transient negative value
2260 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2262 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2263 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2264 i, src, srcp, dest, destp, sk->sk_state,
2265 tp->write_seq - tp->snd_una,
2268 jiffies_delta_to_clock_t(timer_expires - jiffies),
2269 icsk->icsk_retransmits,
2270 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2271 icsk->icsk_probes_out,
2273 atomic_read(&sk->sk_refcnt), sk,
2274 jiffies_to_clock_t(icsk->icsk_rto),
2275 jiffies_to_clock_t(icsk->icsk_ack.ato),
2276 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2278 sk->sk_state == TCP_LISTEN ?
2279 (fastopenq ? fastopenq->max_qlen : 0) :
2280 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2283 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2284 struct seq_file *f, int i)
2288 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
2290 dest = tw->tw_daddr;
2291 src = tw->tw_rcv_saddr;
2292 destp = ntohs(tw->tw_dport);
2293 srcp = ntohs(tw->tw_sport);
2295 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2296 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2297 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2298 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2299 atomic_read(&tw->tw_refcnt), tw);
2304 static int tcp4_seq_show(struct seq_file *seq, void *v)
2306 struct tcp_iter_state *st;
2307 struct sock *sk = v;
2309 seq_setwidth(seq, TMPSZ - 1);
2310 if (v == SEQ_START_TOKEN) {
2311 seq_puts(seq, " sl local_address rem_address st tx_queue "
2312 "rx_queue tr tm->when retrnsmt uid timeout "
2318 switch (st->state) {
2319 case TCP_SEQ_STATE_LISTENING:
2320 case TCP_SEQ_STATE_ESTABLISHED:
2321 if (sk->sk_state == TCP_TIME_WAIT)
2322 get_timewait4_sock(v, seq, st->num);
2324 get_tcp4_sock(v, seq, st->num);
2326 case TCP_SEQ_STATE_OPENREQ:
2327 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid);
2335 static const struct file_operations tcp_afinfo_seq_fops = {
2336 .owner = THIS_MODULE,
2337 .open = tcp_seq_open,
2339 .llseek = seq_lseek,
2340 .release = seq_release_net
2343 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2346 .seq_fops = &tcp_afinfo_seq_fops,
2348 .show = tcp4_seq_show,
2352 static int __net_init tcp4_proc_init_net(struct net *net)
2354 return tcp_proc_register(net, &tcp4_seq_afinfo);
2357 static void __net_exit tcp4_proc_exit_net(struct net *net)
2359 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2362 static struct pernet_operations tcp4_net_ops = {
2363 .init = tcp4_proc_init_net,
2364 .exit = tcp4_proc_exit_net,
2367 int __init tcp4_proc_init(void)
2369 return register_pernet_subsys(&tcp4_net_ops);
2372 void tcp4_proc_exit(void)
2374 unregister_pernet_subsys(&tcp4_net_ops);
2376 #endif /* CONFIG_PROC_FS */
2378 struct proto tcp_prot = {
2380 .owner = THIS_MODULE,
2382 .connect = tcp_v4_connect,
2383 .disconnect = tcp_disconnect,
2384 .accept = inet_csk_accept,
2386 .init = tcp_v4_init_sock,
2387 .destroy = tcp_v4_destroy_sock,
2388 .shutdown = tcp_shutdown,
2389 .setsockopt = tcp_setsockopt,
2390 .getsockopt = tcp_getsockopt,
2391 .recvmsg = tcp_recvmsg,
2392 .sendmsg = tcp_sendmsg,
2393 .sendpage = tcp_sendpage,
2394 .backlog_rcv = tcp_v4_do_rcv,
2395 .release_cb = tcp_release_cb,
2397 .unhash = inet_unhash,
2398 .get_port = inet_csk_get_port,
2399 .enter_memory_pressure = tcp_enter_memory_pressure,
2400 .stream_memory_free = tcp_stream_memory_free,
2401 .sockets_allocated = &tcp_sockets_allocated,
2402 .orphan_count = &tcp_orphan_count,
2403 .memory_allocated = &tcp_memory_allocated,
2404 .memory_pressure = &tcp_memory_pressure,
2405 .sysctl_mem = sysctl_tcp_mem,
2406 .sysctl_wmem = sysctl_tcp_wmem,
2407 .sysctl_rmem = sysctl_tcp_rmem,
2408 .max_header = MAX_TCP_HEADER,
2409 .obj_size = sizeof(struct tcp_sock),
2410 .slab_flags = SLAB_DESTROY_BY_RCU,
2411 .twsk_prot = &tcp_timewait_sock_ops,
2412 .rsk_prot = &tcp_request_sock_ops,
2413 .h.hashinfo = &tcp_hashinfo,
2414 .no_autobind = true,
2415 #ifdef CONFIG_COMPAT
2416 .compat_setsockopt = compat_tcp_setsockopt,
2417 .compat_getsockopt = compat_tcp_getsockopt,
2419 #ifdef CONFIG_MEMCG_KMEM
2420 .init_cgroup = tcp_init_cgroup,
2421 .destroy_cgroup = tcp_destroy_cgroup,
2422 .proto_cgroup = tcp_proto_cgroup,
2425 EXPORT_SYMBOL(tcp_prot);
2427 static int __net_init tcp_sk_init(struct net *net)
2429 net->ipv4.sysctl_tcp_ecn = 2;
2433 static void __net_exit tcp_sk_exit(struct net *net)
2437 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2439 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2442 static struct pernet_operations __net_initdata tcp_sk_ops = {
2443 .init = tcp_sk_init,
2444 .exit = tcp_sk_exit,
2445 .exit_batch = tcp_sk_exit_batch,
2448 void __init tcp_v4_init(void)
2450 inet_hashinfo_init(&tcp_hashinfo);
2451 if (register_pernet_subsys(&tcp_sk_ops))
2452 panic("Failed to create the TCP control socket.\n");