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;
211 inet_csk(sk)->icsk_ext_hdr_len = 0;
213 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
215 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
217 /* Socket identity is still unknown (sport may be zero).
218 * However we set state to SYN-SENT and not releasing socket
219 * lock select source port, enter ourselves into the hash tables and
220 * complete initialization after this.
222 tcp_set_state(sk, TCP_SYN_SENT);
223 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);
884 * Save and compile IPv4 options into the request_sock if needed.
886 static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
888 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
889 struct ip_options_rcu *dopt = NULL;
891 if (opt && opt->optlen) {
892 int opt_size = sizeof(*dopt) + opt->optlen;
894 dopt = kmalloc(opt_size, GFP_ATOMIC);
895 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
903 #ifdef CONFIG_TCP_MD5SIG
905 * RFC2385 MD5 checksumming requires a mapping of
906 * IP address->MD5 Key.
907 * We need to maintain these in the sk structure.
910 /* Find the Key structure for an address. */
911 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
912 const union tcp_md5_addr *addr,
915 struct tcp_sock *tp = tcp_sk(sk);
916 struct tcp_md5sig_key *key;
917 unsigned int size = sizeof(struct in_addr);
918 struct tcp_md5sig_info *md5sig;
920 /* caller either holds rcu_read_lock() or socket lock */
921 md5sig = rcu_dereference_check(tp->md5sig_info,
922 sock_owned_by_user(sk) ||
923 lockdep_is_held(&sk->sk_lock.slock));
926 #if IS_ENABLED(CONFIG_IPV6)
927 if (family == AF_INET6)
928 size = sizeof(struct in6_addr);
930 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
931 if (key->family != family)
933 if (!memcmp(&key->addr, addr, size))
938 EXPORT_SYMBOL(tcp_md5_do_lookup);
940 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
941 struct sock *addr_sk)
943 union tcp_md5_addr *addr;
945 addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
946 return tcp_md5_do_lookup(sk, addr, AF_INET);
948 EXPORT_SYMBOL(tcp_v4_md5_lookup);
950 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
951 struct request_sock *req)
953 union tcp_md5_addr *addr;
955 addr = (union tcp_md5_addr *)&inet_rsk(req)->ir_rmt_addr;
956 return tcp_md5_do_lookup(sk, addr, AF_INET);
959 /* This can be called on a newly created socket, from other files */
960 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
961 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
963 /* Add Key to the list */
964 struct tcp_md5sig_key *key;
965 struct tcp_sock *tp = tcp_sk(sk);
966 struct tcp_md5sig_info *md5sig;
968 key = tcp_md5_do_lookup(sk, addr, family);
970 /* Pre-existing entry - just update that one. */
971 memcpy(key->key, newkey, newkeylen);
972 key->keylen = newkeylen;
976 md5sig = rcu_dereference_protected(tp->md5sig_info,
977 sock_owned_by_user(sk));
979 md5sig = kmalloc(sizeof(*md5sig), gfp);
983 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
984 INIT_HLIST_HEAD(&md5sig->head);
985 rcu_assign_pointer(tp->md5sig_info, md5sig);
988 key = sock_kmalloc(sk, sizeof(*key), gfp);
991 if (!tcp_alloc_md5sig_pool()) {
992 sock_kfree_s(sk, key, sizeof(*key));
996 memcpy(key->key, newkey, newkeylen);
997 key->keylen = newkeylen;
998 key->family = family;
999 memcpy(&key->addr, addr,
1000 (family == AF_INET6) ? sizeof(struct in6_addr) :
1001 sizeof(struct in_addr));
1002 hlist_add_head_rcu(&key->node, &md5sig->head);
1005 EXPORT_SYMBOL(tcp_md5_do_add);
1007 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
1009 struct tcp_md5sig_key *key;
1011 key = tcp_md5_do_lookup(sk, addr, family);
1014 hlist_del_rcu(&key->node);
1015 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1016 kfree_rcu(key, rcu);
1019 EXPORT_SYMBOL(tcp_md5_do_del);
1021 static void tcp_clear_md5_list(struct sock *sk)
1023 struct tcp_sock *tp = tcp_sk(sk);
1024 struct tcp_md5sig_key *key;
1025 struct hlist_node *n;
1026 struct tcp_md5sig_info *md5sig;
1028 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1030 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1031 hlist_del_rcu(&key->node);
1032 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1033 kfree_rcu(key, rcu);
1037 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1040 struct tcp_md5sig cmd;
1041 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1043 if (optlen < sizeof(cmd))
1046 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1049 if (sin->sin_family != AF_INET)
1052 if (!cmd.tcpm_keylen)
1053 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1056 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1059 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1060 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1064 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1065 __be32 daddr, __be32 saddr, int nbytes)
1067 struct tcp4_pseudohdr *bp;
1068 struct scatterlist sg;
1070 bp = &hp->md5_blk.ip4;
1073 * 1. the TCP pseudo-header (in the order: source IP address,
1074 * destination IP address, zero-padded protocol number, and
1080 bp->protocol = IPPROTO_TCP;
1081 bp->len = cpu_to_be16(nbytes);
1083 sg_init_one(&sg, bp, sizeof(*bp));
1084 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1087 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1088 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1090 struct tcp_md5sig_pool *hp;
1091 struct hash_desc *desc;
1093 hp = tcp_get_md5sig_pool();
1095 goto clear_hash_noput;
1096 desc = &hp->md5_desc;
1098 if (crypto_hash_init(desc))
1100 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1102 if (tcp_md5_hash_header(hp, th))
1104 if (tcp_md5_hash_key(hp, key))
1106 if (crypto_hash_final(desc, md5_hash))
1109 tcp_put_md5sig_pool();
1113 tcp_put_md5sig_pool();
1115 memset(md5_hash, 0, 16);
1119 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1120 const struct sock *sk, const struct request_sock *req,
1121 const struct sk_buff *skb)
1123 struct tcp_md5sig_pool *hp;
1124 struct hash_desc *desc;
1125 const struct tcphdr *th = tcp_hdr(skb);
1126 __be32 saddr, daddr;
1129 saddr = inet_sk(sk)->inet_saddr;
1130 daddr = inet_sk(sk)->inet_daddr;
1132 saddr = inet_rsk(req)->ir_loc_addr;
1133 daddr = inet_rsk(req)->ir_rmt_addr;
1135 const struct iphdr *iph = ip_hdr(skb);
1140 hp = tcp_get_md5sig_pool();
1142 goto clear_hash_noput;
1143 desc = &hp->md5_desc;
1145 if (crypto_hash_init(desc))
1148 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1150 if (tcp_md5_hash_header(hp, th))
1152 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1154 if (tcp_md5_hash_key(hp, key))
1156 if (crypto_hash_final(desc, md5_hash))
1159 tcp_put_md5sig_pool();
1163 tcp_put_md5sig_pool();
1165 memset(md5_hash, 0, 16);
1168 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1170 static bool __tcp_v4_inbound_md5_hash(struct sock *sk,
1171 const struct sk_buff *skb)
1174 * This gets called for each TCP segment that arrives
1175 * so we want to be efficient.
1176 * We have 3 drop cases:
1177 * o No MD5 hash and one expected.
1178 * o MD5 hash and we're not expecting one.
1179 * o MD5 hash and its wrong.
1181 const __u8 *hash_location = NULL;
1182 struct tcp_md5sig_key *hash_expected;
1183 const struct iphdr *iph = ip_hdr(skb);
1184 const struct tcphdr *th = tcp_hdr(skb);
1186 unsigned char newhash[16];
1188 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1190 hash_location = tcp_parse_md5sig_option(th);
1192 /* We've parsed the options - do we have a hash? */
1193 if (!hash_expected && !hash_location)
1196 if (hash_expected && !hash_location) {
1197 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1201 if (!hash_expected && hash_location) {
1202 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1206 /* Okay, so this is hash_expected and hash_location -
1207 * so we need to calculate the checksum.
1209 genhash = tcp_v4_md5_hash_skb(newhash,
1213 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1214 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1215 &iph->saddr, ntohs(th->source),
1216 &iph->daddr, ntohs(th->dest),
1217 genhash ? " tcp_v4_calc_md5_hash failed"
1224 static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1229 ret = __tcp_v4_inbound_md5_hash(sk, skb);
1237 static void tcp_v4_init_req(struct request_sock *req, struct sock *sk,
1238 struct sk_buff *skb)
1240 struct inet_request_sock *ireq = inet_rsk(req);
1242 ireq->ir_loc_addr = ip_hdr(skb)->daddr;
1243 ireq->ir_rmt_addr = ip_hdr(skb)->saddr;
1244 ireq->no_srccheck = inet_sk(sk)->transparent;
1245 ireq->opt = tcp_v4_save_options(skb);
1248 static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl,
1249 const struct request_sock *req,
1252 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1255 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1264 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1266 .obj_size = sizeof(struct tcp_request_sock),
1267 .rtx_syn_ack = tcp_rtx_synack,
1268 .send_ack = tcp_v4_reqsk_send_ack,
1269 .destructor = tcp_v4_reqsk_destructor,
1270 .send_reset = tcp_v4_send_reset,
1271 .syn_ack_timeout = tcp_syn_ack_timeout,
1274 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1275 .mss_clamp = TCP_MSS_DEFAULT,
1276 #ifdef CONFIG_TCP_MD5SIG
1277 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1278 .calc_md5_hash = tcp_v4_md5_hash_skb,
1280 .init_req = tcp_v4_init_req,
1281 #ifdef CONFIG_SYN_COOKIES
1282 .cookie_init_seq = cookie_v4_init_sequence,
1284 .route_req = tcp_v4_route_req,
1285 .init_seq = tcp_v4_init_sequence,
1286 .send_synack = tcp_v4_send_synack,
1287 .queue_hash_add = inet_csk_reqsk_queue_hash_add,
1290 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1292 /* Never answer to SYNs send to broadcast or multicast */
1293 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1296 return tcp_conn_request(&tcp_request_sock_ops,
1297 &tcp_request_sock_ipv4_ops, sk, skb);
1300 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1303 EXPORT_SYMBOL(tcp_v4_conn_request);
1307 * The three way handshake has completed - we got a valid synack -
1308 * now create the new socket.
1310 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1311 struct request_sock *req,
1312 struct dst_entry *dst)
1314 struct inet_request_sock *ireq;
1315 struct inet_sock *newinet;
1316 struct tcp_sock *newtp;
1318 #ifdef CONFIG_TCP_MD5SIG
1319 struct tcp_md5sig_key *key;
1321 struct ip_options_rcu *inet_opt;
1323 if (sk_acceptq_is_full(sk))
1326 newsk = tcp_create_openreq_child(sk, req, skb);
1330 newsk->sk_gso_type = SKB_GSO_TCPV4;
1331 inet_sk_rx_dst_set(newsk, skb);
1333 newtp = tcp_sk(newsk);
1334 newinet = inet_sk(newsk);
1335 ireq = inet_rsk(req);
1336 newinet->inet_daddr = ireq->ir_rmt_addr;
1337 newinet->inet_rcv_saddr = ireq->ir_loc_addr;
1338 newinet->inet_saddr = ireq->ir_loc_addr;
1339 inet_opt = ireq->opt;
1340 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1342 newinet->mc_index = inet_iif(skb);
1343 newinet->mc_ttl = ip_hdr(skb)->ttl;
1344 newinet->rcv_tos = ip_hdr(skb)->tos;
1345 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1346 inet_set_txhash(newsk);
1348 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1349 newinet->inet_id = newtp->write_seq ^ jiffies;
1352 dst = inet_csk_route_child_sock(sk, newsk, req);
1356 /* syncookie case : see end of cookie_v4_check() */
1358 sk_setup_caps(newsk, dst);
1360 tcp_sync_mss(newsk, dst_mtu(dst));
1361 newtp->advmss = dst_metric_advmss(dst);
1362 if (tcp_sk(sk)->rx_opt.user_mss &&
1363 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1364 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1366 tcp_initialize_rcv_mss(newsk);
1368 #ifdef CONFIG_TCP_MD5SIG
1369 /* Copy over the MD5 key from the original socket */
1370 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1374 * We're using one, so create a matching key
1375 * on the newsk structure. If we fail to get
1376 * memory, then we end up not copying the key
1379 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1380 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1381 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1385 if (__inet_inherit_port(sk, newsk) < 0)
1387 __inet_hash_nolisten(newsk, NULL);
1392 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1396 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1399 inet_csk_prepare_forced_close(newsk);
1403 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1405 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1407 struct tcphdr *th = tcp_hdr(skb);
1408 const struct iphdr *iph = ip_hdr(skb);
1410 struct request_sock **prev;
1411 /* Find possible connection requests. */
1412 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1413 iph->saddr, iph->daddr);
1415 return tcp_check_req(sk, skb, req, prev, false);
1417 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1418 th->source, iph->daddr, th->dest, inet_iif(skb));
1421 if (nsk->sk_state != TCP_TIME_WAIT) {
1425 inet_twsk_put(inet_twsk(nsk));
1429 #ifdef CONFIG_SYN_COOKIES
1431 sk = cookie_v4_check(sk, skb, &TCP_SKB_CB(skb)->header.h4.opt);
1436 /* The socket must have it's spinlock held when we get
1439 * We have a potential double-lock case here, so even when
1440 * doing backlog processing we use the BH locking scheme.
1441 * This is because we cannot sleep with the original spinlock
1444 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1448 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1449 struct dst_entry *dst = sk->sk_rx_dst;
1451 sock_rps_save_rxhash(sk, skb);
1453 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1454 dst->ops->check(dst, 0) == NULL) {
1456 sk->sk_rx_dst = NULL;
1459 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1463 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1466 if (sk->sk_state == TCP_LISTEN) {
1467 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1472 sock_rps_save_rxhash(nsk, skb);
1473 if (tcp_child_process(sk, nsk, skb)) {
1480 sock_rps_save_rxhash(sk, skb);
1482 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1489 tcp_v4_send_reset(rsk, skb);
1492 /* Be careful here. If this function gets more complicated and
1493 * gcc suffers from register pressure on the x86, sk (in %ebx)
1494 * might be destroyed here. This current version compiles correctly,
1495 * but you have been warned.
1500 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1501 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1504 EXPORT_SYMBOL(tcp_v4_do_rcv);
1506 void tcp_v4_early_demux(struct sk_buff *skb)
1508 const struct iphdr *iph;
1509 const struct tcphdr *th;
1512 if (skb->pkt_type != PACKET_HOST)
1515 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1521 if (th->doff < sizeof(struct tcphdr) / 4)
1524 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1525 iph->saddr, th->source,
1526 iph->daddr, ntohs(th->dest),
1530 skb->destructor = sock_edemux;
1531 if (sk->sk_state != TCP_TIME_WAIT) {
1532 struct dst_entry *dst = sk->sk_rx_dst;
1535 dst = dst_check(dst, 0);
1537 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1538 skb_dst_set_noref(skb, dst);
1543 /* Packet is added to VJ-style prequeue for processing in process
1544 * context, if a reader task is waiting. Apparently, this exciting
1545 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1546 * failed somewhere. Latency? Burstiness? Well, at least now we will
1547 * see, why it failed. 8)8) --ANK
1550 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1552 struct tcp_sock *tp = tcp_sk(sk);
1554 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1557 if (skb->len <= tcp_hdrlen(skb) &&
1558 skb_queue_len(&tp->ucopy.prequeue) == 0)
1561 /* Before escaping RCU protected region, we need to take care of skb
1562 * dst. Prequeue is only enabled for established sockets.
1563 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1564 * Instead of doing full sk_rx_dst validity here, let's perform
1565 * an optimistic check.
1567 if (likely(sk->sk_rx_dst))
1572 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1573 tp->ucopy.memory += skb->truesize;
1574 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1575 struct sk_buff *skb1;
1577 BUG_ON(sock_owned_by_user(sk));
1579 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1580 sk_backlog_rcv(sk, skb1);
1581 NET_INC_STATS_BH(sock_net(sk),
1582 LINUX_MIB_TCPPREQUEUEDROPPED);
1585 tp->ucopy.memory = 0;
1586 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1587 wake_up_interruptible_sync_poll(sk_sleep(sk),
1588 POLLIN | POLLRDNORM | POLLRDBAND);
1589 if (!inet_csk_ack_scheduled(sk))
1590 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1591 (3 * tcp_rto_min(sk)) / 4,
1596 EXPORT_SYMBOL(tcp_prequeue);
1602 int tcp_v4_rcv(struct sk_buff *skb)
1604 const struct iphdr *iph;
1605 const struct tcphdr *th;
1608 struct net *net = dev_net(skb->dev);
1610 if (skb->pkt_type != PACKET_HOST)
1613 /* Count it even if it's bad */
1614 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1616 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1621 if (th->doff < sizeof(struct tcphdr) / 4)
1623 if (!pskb_may_pull(skb, th->doff * 4))
1626 /* An explanation is required here, I think.
1627 * Packet length and doff are validated by header prediction,
1628 * provided case of th->doff==0 is eliminated.
1629 * So, we defer the checks. */
1631 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1636 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1637 * barrier() makes sure compiler wont play fool^Waliasing games.
1639 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1640 sizeof(struct inet_skb_parm));
1643 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1644 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1645 skb->len - th->doff * 4);
1646 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1647 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1648 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1649 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1650 TCP_SKB_CB(skb)->sacked = 0;
1652 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1657 if (sk->sk_state == TCP_TIME_WAIT)
1660 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1661 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1662 goto discard_and_relse;
1665 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1666 goto discard_and_relse;
1668 #ifdef CONFIG_TCP_MD5SIG
1670 * We really want to reject the packet as early as possible
1672 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1673 * o There is an MD5 option and we're not expecting one
1675 if (tcp_v4_inbound_md5_hash(sk, skb))
1676 goto discard_and_relse;
1681 if (sk_filter(sk, skb))
1682 goto discard_and_relse;
1684 sk_mark_napi_id(sk, skb);
1687 bh_lock_sock_nested(sk);
1689 if (!sock_owned_by_user(sk)) {
1690 if (!tcp_prequeue(sk, skb))
1691 ret = tcp_v4_do_rcv(sk, skb);
1692 } else if (unlikely(sk_add_backlog(sk, skb,
1693 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1695 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1696 goto discard_and_relse;
1705 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1708 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1710 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1712 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1714 tcp_v4_send_reset(NULL, skb);
1718 /* Discard frame. */
1727 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1728 inet_twsk_put(inet_twsk(sk));
1732 if (skb->len < (th->doff << 2)) {
1733 inet_twsk_put(inet_twsk(sk));
1736 if (tcp_checksum_complete(skb)) {
1737 inet_twsk_put(inet_twsk(sk));
1740 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1742 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1744 iph->saddr, th->source,
1745 iph->daddr, th->dest,
1748 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1749 inet_twsk_put(inet_twsk(sk));
1753 /* Fall through to ACK */
1756 tcp_v4_timewait_ack(sk, skb);
1760 case TCP_TW_SUCCESS:;
1765 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1766 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1767 .twsk_unique = tcp_twsk_unique,
1768 .twsk_destructor= tcp_twsk_destructor,
1771 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1773 struct dst_entry *dst = skb_dst(skb);
1777 sk->sk_rx_dst = dst;
1778 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1781 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1783 const struct inet_connection_sock_af_ops ipv4_specific = {
1784 .queue_xmit = ip_queue_xmit,
1785 .send_check = tcp_v4_send_check,
1786 .rebuild_header = inet_sk_rebuild_header,
1787 .sk_rx_dst_set = inet_sk_rx_dst_set,
1788 .conn_request = tcp_v4_conn_request,
1789 .syn_recv_sock = tcp_v4_syn_recv_sock,
1790 .net_header_len = sizeof(struct iphdr),
1791 .setsockopt = ip_setsockopt,
1792 .getsockopt = ip_getsockopt,
1793 .addr2sockaddr = inet_csk_addr2sockaddr,
1794 .sockaddr_len = sizeof(struct sockaddr_in),
1795 .bind_conflict = inet_csk_bind_conflict,
1796 #ifdef CONFIG_COMPAT
1797 .compat_setsockopt = compat_ip_setsockopt,
1798 .compat_getsockopt = compat_ip_getsockopt,
1800 .mtu_reduced = tcp_v4_mtu_reduced,
1802 EXPORT_SYMBOL(ipv4_specific);
1804 #ifdef CONFIG_TCP_MD5SIG
1805 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1806 .md5_lookup = tcp_v4_md5_lookup,
1807 .calc_md5_hash = tcp_v4_md5_hash_skb,
1808 .md5_parse = tcp_v4_parse_md5_keys,
1812 /* NOTE: A lot of things set to zero explicitly by call to
1813 * sk_alloc() so need not be done here.
1815 static int tcp_v4_init_sock(struct sock *sk)
1817 struct inet_connection_sock *icsk = inet_csk(sk);
1821 icsk->icsk_af_ops = &ipv4_specific;
1823 #ifdef CONFIG_TCP_MD5SIG
1824 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1830 void tcp_v4_destroy_sock(struct sock *sk)
1832 struct tcp_sock *tp = tcp_sk(sk);
1834 tcp_clear_xmit_timers(sk);
1836 tcp_cleanup_congestion_control(sk);
1838 /* Cleanup up the write buffer. */
1839 tcp_write_queue_purge(sk);
1841 /* Cleans up our, hopefully empty, out_of_order_queue. */
1842 __skb_queue_purge(&tp->out_of_order_queue);
1844 #ifdef CONFIG_TCP_MD5SIG
1845 /* Clean up the MD5 key list, if any */
1846 if (tp->md5sig_info) {
1847 tcp_clear_md5_list(sk);
1848 kfree_rcu(tp->md5sig_info, rcu);
1849 tp->md5sig_info = NULL;
1853 /* Clean prequeue, it must be empty really */
1854 __skb_queue_purge(&tp->ucopy.prequeue);
1856 /* Clean up a referenced TCP bind bucket. */
1857 if (inet_csk(sk)->icsk_bind_hash)
1860 BUG_ON(tp->fastopen_rsk != NULL);
1862 /* If socket is aborted during connect operation */
1863 tcp_free_fastopen_req(tp);
1865 sk_sockets_allocated_dec(sk);
1866 sock_release_memcg(sk);
1868 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1870 #ifdef CONFIG_PROC_FS
1871 /* Proc filesystem TCP sock list dumping. */
1874 * Get next listener socket follow cur. If cur is NULL, get first socket
1875 * starting from bucket given in st->bucket; when st->bucket is zero the
1876 * very first socket in the hash table is returned.
1878 static void *listening_get_next(struct seq_file *seq, void *cur)
1880 struct inet_connection_sock *icsk;
1881 struct hlist_nulls_node *node;
1882 struct sock *sk = cur;
1883 struct inet_listen_hashbucket *ilb;
1884 struct tcp_iter_state *st = seq->private;
1885 struct net *net = seq_file_net(seq);
1888 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1889 spin_lock_bh(&ilb->lock);
1890 sk = sk_nulls_head(&ilb->head);
1894 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1898 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1899 struct request_sock *req = cur;
1901 icsk = inet_csk(st->syn_wait_sk);
1905 if (req->rsk_ops->family == st->family) {
1911 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1914 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1916 sk = sk_nulls_next(st->syn_wait_sk);
1917 st->state = TCP_SEQ_STATE_LISTENING;
1918 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1920 icsk = inet_csk(sk);
1921 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1922 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1924 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1925 sk = sk_nulls_next(sk);
1928 sk_nulls_for_each_from(sk, node) {
1929 if (!net_eq(sock_net(sk), net))
1931 if (sk->sk_family == st->family) {
1935 icsk = inet_csk(sk);
1936 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1937 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1939 st->uid = sock_i_uid(sk);
1940 st->syn_wait_sk = sk;
1941 st->state = TCP_SEQ_STATE_OPENREQ;
1945 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1947 spin_unlock_bh(&ilb->lock);
1949 if (++st->bucket < INET_LHTABLE_SIZE) {
1950 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1951 spin_lock_bh(&ilb->lock);
1952 sk = sk_nulls_head(&ilb->head);
1960 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1962 struct tcp_iter_state *st = seq->private;
1967 rc = listening_get_next(seq, NULL);
1969 while (rc && *pos) {
1970 rc = listening_get_next(seq, rc);
1976 static inline bool empty_bucket(const struct tcp_iter_state *st)
1978 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1982 * Get first established socket starting from bucket given in st->bucket.
1983 * If st->bucket is zero, the very first socket in the hash is returned.
1985 static void *established_get_first(struct seq_file *seq)
1987 struct tcp_iter_state *st = seq->private;
1988 struct net *net = seq_file_net(seq);
1992 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1994 struct hlist_nulls_node *node;
1995 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1997 /* Lockless fast path for the common case of empty buckets */
1998 if (empty_bucket(st))
2002 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2003 if (sk->sk_family != st->family ||
2004 !net_eq(sock_net(sk), net)) {
2010 spin_unlock_bh(lock);
2016 static void *established_get_next(struct seq_file *seq, void *cur)
2018 struct sock *sk = cur;
2019 struct hlist_nulls_node *node;
2020 struct tcp_iter_state *st = seq->private;
2021 struct net *net = seq_file_net(seq);
2026 sk = sk_nulls_next(sk);
2028 sk_nulls_for_each_from(sk, node) {
2029 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2033 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2035 return established_get_first(seq);
2038 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2040 struct tcp_iter_state *st = seq->private;
2044 rc = established_get_first(seq);
2047 rc = established_get_next(seq, rc);
2053 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2056 struct tcp_iter_state *st = seq->private;
2058 st->state = TCP_SEQ_STATE_LISTENING;
2059 rc = listening_get_idx(seq, &pos);
2062 st->state = TCP_SEQ_STATE_ESTABLISHED;
2063 rc = established_get_idx(seq, pos);
2069 static void *tcp_seek_last_pos(struct seq_file *seq)
2071 struct tcp_iter_state *st = seq->private;
2072 int offset = st->offset;
2073 int orig_num = st->num;
2076 switch (st->state) {
2077 case TCP_SEQ_STATE_OPENREQ:
2078 case TCP_SEQ_STATE_LISTENING:
2079 if (st->bucket >= INET_LHTABLE_SIZE)
2081 st->state = TCP_SEQ_STATE_LISTENING;
2082 rc = listening_get_next(seq, NULL);
2083 while (offset-- && rc)
2084 rc = listening_get_next(seq, rc);
2088 st->state = TCP_SEQ_STATE_ESTABLISHED;
2090 case TCP_SEQ_STATE_ESTABLISHED:
2091 if (st->bucket > tcp_hashinfo.ehash_mask)
2093 rc = established_get_first(seq);
2094 while (offset-- && rc)
2095 rc = established_get_next(seq, rc);
2103 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2105 struct tcp_iter_state *st = seq->private;
2108 if (*pos && *pos == st->last_pos) {
2109 rc = tcp_seek_last_pos(seq);
2114 st->state = TCP_SEQ_STATE_LISTENING;
2118 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2121 st->last_pos = *pos;
2125 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2127 struct tcp_iter_state *st = seq->private;
2130 if (v == SEQ_START_TOKEN) {
2131 rc = tcp_get_idx(seq, 0);
2135 switch (st->state) {
2136 case TCP_SEQ_STATE_OPENREQ:
2137 case TCP_SEQ_STATE_LISTENING:
2138 rc = listening_get_next(seq, v);
2140 st->state = TCP_SEQ_STATE_ESTABLISHED;
2143 rc = established_get_first(seq);
2146 case TCP_SEQ_STATE_ESTABLISHED:
2147 rc = established_get_next(seq, v);
2152 st->last_pos = *pos;
2156 static void tcp_seq_stop(struct seq_file *seq, void *v)
2158 struct tcp_iter_state *st = seq->private;
2160 switch (st->state) {
2161 case TCP_SEQ_STATE_OPENREQ:
2163 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2164 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2166 case TCP_SEQ_STATE_LISTENING:
2167 if (v != SEQ_START_TOKEN)
2168 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2170 case TCP_SEQ_STATE_ESTABLISHED:
2172 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2177 int tcp_seq_open(struct inode *inode, struct file *file)
2179 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2180 struct tcp_iter_state *s;
2183 err = seq_open_net(inode, file, &afinfo->seq_ops,
2184 sizeof(struct tcp_iter_state));
2188 s = ((struct seq_file *)file->private_data)->private;
2189 s->family = afinfo->family;
2193 EXPORT_SYMBOL(tcp_seq_open);
2195 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2198 struct proc_dir_entry *p;
2200 afinfo->seq_ops.start = tcp_seq_start;
2201 afinfo->seq_ops.next = tcp_seq_next;
2202 afinfo->seq_ops.stop = tcp_seq_stop;
2204 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2205 afinfo->seq_fops, afinfo);
2210 EXPORT_SYMBOL(tcp_proc_register);
2212 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2214 remove_proc_entry(afinfo->name, net->proc_net);
2216 EXPORT_SYMBOL(tcp_proc_unregister);
2218 static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2219 struct seq_file *f, int i, kuid_t uid)
2221 const struct inet_request_sock *ireq = inet_rsk(req);
2222 long delta = req->expires - jiffies;
2224 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2225 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2228 ntohs(inet_sk(sk)->inet_sport),
2230 ntohs(ireq->ir_rmt_port),
2232 0, 0, /* could print option size, but that is af dependent. */
2233 1, /* timers active (only the expire timer) */
2234 jiffies_delta_to_clock_t(delta),
2236 from_kuid_munged(seq_user_ns(f), uid),
2237 0, /* non standard timer */
2238 0, /* open_requests have no inode */
2239 atomic_read(&sk->sk_refcnt),
2243 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2246 unsigned long timer_expires;
2247 const struct tcp_sock *tp = tcp_sk(sk);
2248 const struct inet_connection_sock *icsk = inet_csk(sk);
2249 const struct inet_sock *inet = inet_sk(sk);
2250 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2251 __be32 dest = inet->inet_daddr;
2252 __be32 src = inet->inet_rcv_saddr;
2253 __u16 destp = ntohs(inet->inet_dport);
2254 __u16 srcp = ntohs(inet->inet_sport);
2257 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2258 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2259 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2261 timer_expires = icsk->icsk_timeout;
2262 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2264 timer_expires = icsk->icsk_timeout;
2265 } else if (timer_pending(&sk->sk_timer)) {
2267 timer_expires = sk->sk_timer.expires;
2270 timer_expires = jiffies;
2273 if (sk->sk_state == TCP_LISTEN)
2274 rx_queue = sk->sk_ack_backlog;
2277 * because we dont lock socket, we might find a transient negative value
2279 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2281 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2282 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2283 i, src, srcp, dest, destp, sk->sk_state,
2284 tp->write_seq - tp->snd_una,
2287 jiffies_delta_to_clock_t(timer_expires - jiffies),
2288 icsk->icsk_retransmits,
2289 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2290 icsk->icsk_probes_out,
2292 atomic_read(&sk->sk_refcnt), sk,
2293 jiffies_to_clock_t(icsk->icsk_rto),
2294 jiffies_to_clock_t(icsk->icsk_ack.ato),
2295 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2297 sk->sk_state == TCP_LISTEN ?
2298 (fastopenq ? fastopenq->max_qlen : 0) :
2299 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2302 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2303 struct seq_file *f, int i)
2307 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
2309 dest = tw->tw_daddr;
2310 src = tw->tw_rcv_saddr;
2311 destp = ntohs(tw->tw_dport);
2312 srcp = ntohs(tw->tw_sport);
2314 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2315 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2316 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2317 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2318 atomic_read(&tw->tw_refcnt), tw);
2323 static int tcp4_seq_show(struct seq_file *seq, void *v)
2325 struct tcp_iter_state *st;
2326 struct sock *sk = v;
2328 seq_setwidth(seq, TMPSZ - 1);
2329 if (v == SEQ_START_TOKEN) {
2330 seq_puts(seq, " sl local_address rem_address st tx_queue "
2331 "rx_queue tr tm->when retrnsmt uid timeout "
2337 switch (st->state) {
2338 case TCP_SEQ_STATE_LISTENING:
2339 case TCP_SEQ_STATE_ESTABLISHED:
2340 if (sk->sk_state == TCP_TIME_WAIT)
2341 get_timewait4_sock(v, seq, st->num);
2343 get_tcp4_sock(v, seq, st->num);
2345 case TCP_SEQ_STATE_OPENREQ:
2346 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid);
2354 static const struct file_operations tcp_afinfo_seq_fops = {
2355 .owner = THIS_MODULE,
2356 .open = tcp_seq_open,
2358 .llseek = seq_lseek,
2359 .release = seq_release_net
2362 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2365 .seq_fops = &tcp_afinfo_seq_fops,
2367 .show = tcp4_seq_show,
2371 static int __net_init tcp4_proc_init_net(struct net *net)
2373 return tcp_proc_register(net, &tcp4_seq_afinfo);
2376 static void __net_exit tcp4_proc_exit_net(struct net *net)
2378 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2381 static struct pernet_operations tcp4_net_ops = {
2382 .init = tcp4_proc_init_net,
2383 .exit = tcp4_proc_exit_net,
2386 int __init tcp4_proc_init(void)
2388 return register_pernet_subsys(&tcp4_net_ops);
2391 void tcp4_proc_exit(void)
2393 unregister_pernet_subsys(&tcp4_net_ops);
2395 #endif /* CONFIG_PROC_FS */
2397 struct proto tcp_prot = {
2399 .owner = THIS_MODULE,
2401 .connect = tcp_v4_connect,
2402 .disconnect = tcp_disconnect,
2403 .accept = inet_csk_accept,
2405 .init = tcp_v4_init_sock,
2406 .destroy = tcp_v4_destroy_sock,
2407 .shutdown = tcp_shutdown,
2408 .setsockopt = tcp_setsockopt,
2409 .getsockopt = tcp_getsockopt,
2410 .recvmsg = tcp_recvmsg,
2411 .sendmsg = tcp_sendmsg,
2412 .sendpage = tcp_sendpage,
2413 .backlog_rcv = tcp_v4_do_rcv,
2414 .release_cb = tcp_release_cb,
2416 .unhash = inet_unhash,
2417 .get_port = inet_csk_get_port,
2418 .enter_memory_pressure = tcp_enter_memory_pressure,
2419 .stream_memory_free = tcp_stream_memory_free,
2420 .sockets_allocated = &tcp_sockets_allocated,
2421 .orphan_count = &tcp_orphan_count,
2422 .memory_allocated = &tcp_memory_allocated,
2423 .memory_pressure = &tcp_memory_pressure,
2424 .sysctl_mem = sysctl_tcp_mem,
2425 .sysctl_wmem = sysctl_tcp_wmem,
2426 .sysctl_rmem = sysctl_tcp_rmem,
2427 .max_header = MAX_TCP_HEADER,
2428 .obj_size = sizeof(struct tcp_sock),
2429 .slab_flags = SLAB_DESTROY_BY_RCU,
2430 .twsk_prot = &tcp_timewait_sock_ops,
2431 .rsk_prot = &tcp_request_sock_ops,
2432 .h.hashinfo = &tcp_hashinfo,
2433 .no_autobind = true,
2434 #ifdef CONFIG_COMPAT
2435 .compat_setsockopt = compat_tcp_setsockopt,
2436 .compat_getsockopt = compat_tcp_getsockopt,
2438 #ifdef CONFIG_MEMCG_KMEM
2439 .init_cgroup = tcp_init_cgroup,
2440 .destroy_cgroup = tcp_destroy_cgroup,
2441 .proto_cgroup = tcp_proto_cgroup,
2444 EXPORT_SYMBOL(tcp_prot);
2446 static int __net_init tcp_sk_init(struct net *net)
2448 net->ipv4.sysctl_tcp_ecn = 2;
2452 static void __net_exit tcp_sk_exit(struct net *net)
2456 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2458 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2461 static struct pernet_operations __net_initdata tcp_sk_ops = {
2462 .init = tcp_sk_init,
2463 .exit = tcp_sk_exit,
2464 .exit_batch = tcp_sk_exit_batch,
2467 void __init tcp_v4_init(void)
2469 inet_hashinfo_init(&tcp_hashinfo);
2470 if (register_pernet_subsys(&tcp_sk_ops))
2471 panic("Failed to create the TCP control socket.\n");