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 sk_rcv_saddr_set(sk, 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 sk_daddr_set(sk, 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 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
315 void tcp_req_err(struct sock *sk, u32 seq)
317 struct request_sock *req = inet_reqsk(sk);
318 struct net *net = sock_net(sk);
320 /* ICMPs are not backlogged, hence we cannot get
321 * an established socket here.
325 if (seq != tcp_rsk(req)->snt_isn) {
326 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
330 * Still in SYN_RECV, just remove it silently.
331 * There is no good way to pass the error to the newly
332 * created socket, and POSIX does not want network
333 * errors returned from accept().
335 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS);
336 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
339 EXPORT_SYMBOL(tcp_req_err);
342 * This routine is called by the ICMP module when it gets some
343 * sort of error condition. If err < 0 then the socket should
344 * be closed and the error returned to the user. If err > 0
345 * it's just the icmp type << 8 | icmp code. After adjustment
346 * header points to the first 8 bytes of the tcp header. We need
347 * to find the appropriate port.
349 * The locking strategy used here is very "optimistic". When
350 * someone else accesses the socket the ICMP is just dropped
351 * and for some paths there is no check at all.
352 * A more general error queue to queue errors for later handling
353 * is probably better.
357 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
359 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
360 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
361 struct inet_connection_sock *icsk;
363 struct inet_sock *inet;
364 const int type = icmp_hdr(icmp_skb)->type;
365 const int code = icmp_hdr(icmp_skb)->code;
368 struct request_sock *fastopen;
372 struct net *net = dev_net(icmp_skb->dev);
374 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
375 th->dest, iph->saddr, ntohs(th->source),
378 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
381 if (sk->sk_state == TCP_TIME_WAIT) {
382 inet_twsk_put(inet_twsk(sk));
385 seq = ntohl(th->seq);
386 if (sk->sk_state == TCP_NEW_SYN_RECV)
387 return tcp_req_err(sk, seq);
390 /* If too many ICMPs get dropped on busy
391 * servers this needs to be solved differently.
392 * We do take care of PMTU discovery (RFC1191) special case :
393 * we can receive locally generated ICMP messages while socket is held.
395 if (sock_owned_by_user(sk)) {
396 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
397 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
399 if (sk->sk_state == TCP_CLOSE)
402 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
403 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
409 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
410 fastopen = tp->fastopen_rsk;
411 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
412 if (sk->sk_state != TCP_LISTEN &&
413 !between(seq, snd_una, tp->snd_nxt)) {
414 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
420 do_redirect(icmp_skb, sk);
422 case ICMP_SOURCE_QUENCH:
423 /* Just silently ignore these. */
425 case ICMP_PARAMETERPROB:
428 case ICMP_DEST_UNREACH:
429 if (code > NR_ICMP_UNREACH)
432 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
433 /* We are not interested in TCP_LISTEN and open_requests
434 * (SYN-ACKs send out by Linux are always <576bytes so
435 * they should go through unfragmented).
437 if (sk->sk_state == TCP_LISTEN)
441 if (!sock_owned_by_user(sk)) {
442 tcp_v4_mtu_reduced(sk);
444 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
450 err = icmp_err_convert[code].errno;
451 /* check if icmp_skb allows revert of backoff
452 * (see draft-zimmermann-tcp-lcd) */
453 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
455 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
456 !icsk->icsk_backoff || fastopen)
459 if (sock_owned_by_user(sk))
462 icsk->icsk_backoff--;
463 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
465 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
467 skb = tcp_write_queue_head(sk);
470 remaining = icsk->icsk_rto -
472 tcp_time_stamp - tcp_skb_timestamp(skb));
475 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
476 remaining, TCP_RTO_MAX);
478 /* RTO revert clocked out retransmission.
479 * Will retransmit now */
480 tcp_retransmit_timer(sk);
484 case ICMP_TIME_EXCEEDED:
491 switch (sk->sk_state) {
494 /* Only in fast or simultaneous open. If a fast open socket is
495 * is already accepted it is treated as a connected one below.
497 if (fastopen && fastopen->sk == NULL)
500 if (!sock_owned_by_user(sk)) {
503 sk->sk_error_report(sk);
507 sk->sk_err_soft = err;
512 /* If we've already connected we will keep trying
513 * until we time out, or the user gives up.
515 * rfc1122 4.2.3.9 allows to consider as hard errors
516 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
517 * but it is obsoleted by pmtu discovery).
519 * Note, that in modern internet, where routing is unreliable
520 * and in each dark corner broken firewalls sit, sending random
521 * errors ordered by their masters even this two messages finally lose
522 * their original sense (even Linux sends invalid PORT_UNREACHs)
524 * Now we are in compliance with RFCs.
529 if (!sock_owned_by_user(sk) && inet->recverr) {
531 sk->sk_error_report(sk);
532 } else { /* Only an error on timeout */
533 sk->sk_err_soft = err;
541 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
543 struct tcphdr *th = tcp_hdr(skb);
545 if (skb->ip_summed == CHECKSUM_PARTIAL) {
546 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
547 skb->csum_start = skb_transport_header(skb) - skb->head;
548 skb->csum_offset = offsetof(struct tcphdr, check);
550 th->check = tcp_v4_check(skb->len, saddr, daddr,
557 /* This routine computes an IPv4 TCP checksum. */
558 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
560 const struct inet_sock *inet = inet_sk(sk);
562 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
564 EXPORT_SYMBOL(tcp_v4_send_check);
567 * This routine will send an RST to the other tcp.
569 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
571 * Answer: if a packet caused RST, it is not for a socket
572 * existing in our system, if it is matched to a socket,
573 * it is just duplicate segment or bug in other side's TCP.
574 * So that we build reply only basing on parameters
575 * arrived with segment.
576 * Exception: precedence violation. We do not implement it in any case.
579 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
581 const struct tcphdr *th = tcp_hdr(skb);
584 #ifdef CONFIG_TCP_MD5SIG
585 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
588 struct ip_reply_arg arg;
589 #ifdef CONFIG_TCP_MD5SIG
590 struct tcp_md5sig_key *key;
591 const __u8 *hash_location = NULL;
592 unsigned char newhash[16];
594 struct sock *sk1 = NULL;
598 /* Never send a reset in response to a reset. */
602 /* If sk not NULL, it means we did a successful lookup and incoming
603 * route had to be correct. prequeue might have dropped our dst.
605 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
608 /* Swap the send and the receive. */
609 memset(&rep, 0, sizeof(rep));
610 rep.th.dest = th->source;
611 rep.th.source = th->dest;
612 rep.th.doff = sizeof(struct tcphdr) / 4;
616 rep.th.seq = th->ack_seq;
619 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
620 skb->len - (th->doff << 2));
623 memset(&arg, 0, sizeof(arg));
624 arg.iov[0].iov_base = (unsigned char *)&rep;
625 arg.iov[0].iov_len = sizeof(rep.th);
627 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
628 #ifdef CONFIG_TCP_MD5SIG
629 hash_location = tcp_parse_md5sig_option(th);
630 if (!sk && hash_location) {
632 * active side is lost. Try to find listening socket through
633 * source port, and then find md5 key through listening socket.
634 * we are not loose security here:
635 * Incoming packet is checked with md5 hash with finding key,
636 * no RST generated if md5 hash doesn't match.
638 sk1 = __inet_lookup_listener(net,
639 &tcp_hashinfo, ip_hdr(skb)->saddr,
640 th->source, ip_hdr(skb)->daddr,
641 ntohs(th->source), inet_iif(skb));
642 /* don't send rst if it can't find key */
646 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
647 &ip_hdr(skb)->saddr, AF_INET);
651 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
652 if (genhash || memcmp(hash_location, newhash, 16) != 0)
655 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
661 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
663 (TCPOPT_MD5SIG << 8) |
665 /* Update length and the length the header thinks exists */
666 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
667 rep.th.doff = arg.iov[0].iov_len / 4;
669 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
670 key, ip_hdr(skb)->saddr,
671 ip_hdr(skb)->daddr, &rep.th);
674 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
675 ip_hdr(skb)->saddr, /* XXX */
676 arg.iov[0].iov_len, IPPROTO_TCP, 0);
677 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
678 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
679 /* When socket is gone, all binding information is lost.
680 * routing might fail in this case. No choice here, if we choose to force
681 * input interface, we will misroute in case of asymmetric route.
684 arg.bound_dev_if = sk->sk_bound_dev_if;
686 arg.tos = ip_hdr(skb)->tos;
687 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
688 skb, &TCP_SKB_CB(skb)->header.h4.opt,
689 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
690 &arg, arg.iov[0].iov_len);
692 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
693 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
695 #ifdef CONFIG_TCP_MD5SIG
704 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
705 outside socket context is ugly, certainly. What can I do?
708 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
709 u32 win, u32 tsval, u32 tsecr, int oif,
710 struct tcp_md5sig_key *key,
711 int reply_flags, u8 tos)
713 const struct tcphdr *th = tcp_hdr(skb);
716 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
717 #ifdef CONFIG_TCP_MD5SIG
718 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
722 struct ip_reply_arg arg;
723 struct net *net = dev_net(skb_dst(skb)->dev);
725 memset(&rep.th, 0, sizeof(struct tcphdr));
726 memset(&arg, 0, sizeof(arg));
728 arg.iov[0].iov_base = (unsigned char *)&rep;
729 arg.iov[0].iov_len = sizeof(rep.th);
731 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
732 (TCPOPT_TIMESTAMP << 8) |
734 rep.opt[1] = htonl(tsval);
735 rep.opt[2] = htonl(tsecr);
736 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
739 /* Swap the send and the receive. */
740 rep.th.dest = th->source;
741 rep.th.source = th->dest;
742 rep.th.doff = arg.iov[0].iov_len / 4;
743 rep.th.seq = htonl(seq);
744 rep.th.ack_seq = htonl(ack);
746 rep.th.window = htons(win);
748 #ifdef CONFIG_TCP_MD5SIG
750 int offset = (tsecr) ? 3 : 0;
752 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
754 (TCPOPT_MD5SIG << 8) |
756 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
757 rep.th.doff = arg.iov[0].iov_len/4;
759 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
760 key, ip_hdr(skb)->saddr,
761 ip_hdr(skb)->daddr, &rep.th);
764 arg.flags = reply_flags;
765 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
766 ip_hdr(skb)->saddr, /* XXX */
767 arg.iov[0].iov_len, IPPROTO_TCP, 0);
768 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
770 arg.bound_dev_if = oif;
772 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
773 skb, &TCP_SKB_CB(skb)->header.h4.opt,
774 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
775 &arg, arg.iov[0].iov_len);
777 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
780 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
782 struct inet_timewait_sock *tw = inet_twsk(sk);
783 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
785 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
786 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
787 tcp_time_stamp + tcptw->tw_ts_offset,
790 tcp_twsk_md5_key(tcptw),
791 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
798 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
799 struct request_sock *req)
801 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
802 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
804 tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
805 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
806 tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
810 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
812 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
817 * Send a SYN-ACK after having received a SYN.
818 * This still operates on a request_sock only, not on a big
821 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
823 struct request_sock *req,
825 struct tcp_fastopen_cookie *foc)
827 const struct inet_request_sock *ireq = inet_rsk(req);
832 /* First, grab a route. */
833 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
836 skb = tcp_make_synack(sk, dst, req, foc);
839 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
841 skb_set_queue_mapping(skb, queue_mapping);
842 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
845 err = net_xmit_eval(err);
852 * IPv4 request_sock destructor.
854 static void tcp_v4_reqsk_destructor(struct request_sock *req)
856 kfree(inet_rsk(req)->opt);
860 * Return true if a syncookie should be sent
862 bool tcp_syn_flood_action(struct sock *sk,
863 const struct sk_buff *skb,
866 const char *msg = "Dropping request";
867 bool want_cookie = false;
868 struct listen_sock *lopt;
870 #ifdef CONFIG_SYN_COOKIES
871 if (sysctl_tcp_syncookies) {
872 msg = "Sending cookies";
874 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
877 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
879 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
880 if (!lopt->synflood_warned && sysctl_tcp_syncookies != 2) {
881 lopt->synflood_warned = 1;
882 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
883 proto, ntohs(tcp_hdr(skb)->dest), msg);
887 EXPORT_SYMBOL(tcp_syn_flood_action);
889 #ifdef CONFIG_TCP_MD5SIG
891 * RFC2385 MD5 checksumming requires a mapping of
892 * IP address->MD5 Key.
893 * We need to maintain these in the sk structure.
896 /* Find the Key structure for an address. */
897 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
898 const union tcp_md5_addr *addr,
901 const struct tcp_sock *tp = tcp_sk(sk);
902 struct tcp_md5sig_key *key;
903 unsigned int size = sizeof(struct in_addr);
904 const struct tcp_md5sig_info *md5sig;
906 /* caller either holds rcu_read_lock() or socket lock */
907 md5sig = rcu_dereference_check(tp->md5sig_info,
908 sock_owned_by_user(sk) ||
909 lockdep_is_held(&sk->sk_lock.slock));
912 #if IS_ENABLED(CONFIG_IPV6)
913 if (family == AF_INET6)
914 size = sizeof(struct in6_addr);
916 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
917 if (key->family != family)
919 if (!memcmp(&key->addr, addr, size))
924 EXPORT_SYMBOL(tcp_md5_do_lookup);
926 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
927 const struct sock *addr_sk)
929 union tcp_md5_addr *addr;
931 addr = (union tcp_md5_addr *)&sk->sk_daddr;
932 return tcp_md5_do_lookup(sk, addr, AF_INET);
934 EXPORT_SYMBOL(tcp_v4_md5_lookup);
936 /* This can be called on a newly created socket, from other files */
937 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
938 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
940 /* Add Key to the list */
941 struct tcp_md5sig_key *key;
942 struct tcp_sock *tp = tcp_sk(sk);
943 struct tcp_md5sig_info *md5sig;
945 key = tcp_md5_do_lookup(sk, addr, family);
947 /* Pre-existing entry - just update that one. */
948 memcpy(key->key, newkey, newkeylen);
949 key->keylen = newkeylen;
953 md5sig = rcu_dereference_protected(tp->md5sig_info,
954 sock_owned_by_user(sk));
956 md5sig = kmalloc(sizeof(*md5sig), gfp);
960 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
961 INIT_HLIST_HEAD(&md5sig->head);
962 rcu_assign_pointer(tp->md5sig_info, md5sig);
965 key = sock_kmalloc(sk, sizeof(*key), gfp);
968 if (!tcp_alloc_md5sig_pool()) {
969 sock_kfree_s(sk, key, sizeof(*key));
973 memcpy(key->key, newkey, newkeylen);
974 key->keylen = newkeylen;
975 key->family = family;
976 memcpy(&key->addr, addr,
977 (family == AF_INET6) ? sizeof(struct in6_addr) :
978 sizeof(struct in_addr));
979 hlist_add_head_rcu(&key->node, &md5sig->head);
982 EXPORT_SYMBOL(tcp_md5_do_add);
984 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
986 struct tcp_md5sig_key *key;
988 key = tcp_md5_do_lookup(sk, addr, family);
991 hlist_del_rcu(&key->node);
992 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
996 EXPORT_SYMBOL(tcp_md5_do_del);
998 static void tcp_clear_md5_list(struct sock *sk)
1000 struct tcp_sock *tp = tcp_sk(sk);
1001 struct tcp_md5sig_key *key;
1002 struct hlist_node *n;
1003 struct tcp_md5sig_info *md5sig;
1005 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1007 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1008 hlist_del_rcu(&key->node);
1009 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1010 kfree_rcu(key, rcu);
1014 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1017 struct tcp_md5sig cmd;
1018 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1020 if (optlen < sizeof(cmd))
1023 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1026 if (sin->sin_family != AF_INET)
1029 if (!cmd.tcpm_keylen)
1030 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1033 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1036 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1037 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1041 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1042 __be32 daddr, __be32 saddr, int nbytes)
1044 struct tcp4_pseudohdr *bp;
1045 struct scatterlist sg;
1047 bp = &hp->md5_blk.ip4;
1050 * 1. the TCP pseudo-header (in the order: source IP address,
1051 * destination IP address, zero-padded protocol number, and
1057 bp->protocol = IPPROTO_TCP;
1058 bp->len = cpu_to_be16(nbytes);
1060 sg_init_one(&sg, bp, sizeof(*bp));
1061 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1064 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1065 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1067 struct tcp_md5sig_pool *hp;
1068 struct hash_desc *desc;
1070 hp = tcp_get_md5sig_pool();
1072 goto clear_hash_noput;
1073 desc = &hp->md5_desc;
1075 if (crypto_hash_init(desc))
1077 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1079 if (tcp_md5_hash_header(hp, th))
1081 if (tcp_md5_hash_key(hp, key))
1083 if (crypto_hash_final(desc, md5_hash))
1086 tcp_put_md5sig_pool();
1090 tcp_put_md5sig_pool();
1092 memset(md5_hash, 0, 16);
1096 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1097 const struct sock *sk,
1098 const struct sk_buff *skb)
1100 struct tcp_md5sig_pool *hp;
1101 struct hash_desc *desc;
1102 const struct tcphdr *th = tcp_hdr(skb);
1103 __be32 saddr, daddr;
1105 if (sk) { /* valid for establish/request sockets */
1106 saddr = sk->sk_rcv_saddr;
1107 daddr = sk->sk_daddr;
1109 const struct iphdr *iph = ip_hdr(skb);
1114 hp = tcp_get_md5sig_pool();
1116 goto clear_hash_noput;
1117 desc = &hp->md5_desc;
1119 if (crypto_hash_init(desc))
1122 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1124 if (tcp_md5_hash_header(hp, th))
1126 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1128 if (tcp_md5_hash_key(hp, key))
1130 if (crypto_hash_final(desc, md5_hash))
1133 tcp_put_md5sig_pool();
1137 tcp_put_md5sig_pool();
1139 memset(md5_hash, 0, 16);
1142 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1144 /* Called with rcu_read_lock() */
1145 static bool tcp_v4_inbound_md5_hash(struct sock *sk,
1146 const struct sk_buff *skb)
1149 * This gets called for each TCP segment that arrives
1150 * so we want to be efficient.
1151 * We have 3 drop cases:
1152 * o No MD5 hash and one expected.
1153 * o MD5 hash and we're not expecting one.
1154 * o MD5 hash and its wrong.
1156 const __u8 *hash_location = NULL;
1157 struct tcp_md5sig_key *hash_expected;
1158 const struct iphdr *iph = ip_hdr(skb);
1159 const struct tcphdr *th = tcp_hdr(skb);
1161 unsigned char newhash[16];
1163 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1165 hash_location = tcp_parse_md5sig_option(th);
1167 /* We've parsed the options - do we have a hash? */
1168 if (!hash_expected && !hash_location)
1171 if (hash_expected && !hash_location) {
1172 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1176 if (!hash_expected && hash_location) {
1177 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1181 /* Okay, so this is hash_expected and hash_location -
1182 * so we need to calculate the checksum.
1184 genhash = tcp_v4_md5_hash_skb(newhash,
1188 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1189 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1190 &iph->saddr, ntohs(th->source),
1191 &iph->daddr, ntohs(th->dest),
1192 genhash ? " tcp_v4_calc_md5_hash failed"
1200 static void tcp_v4_init_req(struct request_sock *req, struct sock *sk_listener,
1201 struct sk_buff *skb)
1203 struct inet_request_sock *ireq = inet_rsk(req);
1205 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1206 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1207 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
1208 ireq->opt = tcp_v4_save_options(skb);
1209 ireq->ireq_family = AF_INET;
1212 static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl,
1213 const struct request_sock *req,
1216 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1219 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1228 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1230 .obj_size = sizeof(struct tcp_request_sock),
1231 .rtx_syn_ack = tcp_rtx_synack,
1232 .send_ack = tcp_v4_reqsk_send_ack,
1233 .destructor = tcp_v4_reqsk_destructor,
1234 .send_reset = tcp_v4_send_reset,
1235 .syn_ack_timeout = tcp_syn_ack_timeout,
1238 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1239 .mss_clamp = TCP_MSS_DEFAULT,
1240 #ifdef CONFIG_TCP_MD5SIG
1241 .req_md5_lookup = tcp_v4_md5_lookup,
1242 .calc_md5_hash = tcp_v4_md5_hash_skb,
1244 .init_req = tcp_v4_init_req,
1245 #ifdef CONFIG_SYN_COOKIES
1246 .cookie_init_seq = cookie_v4_init_sequence,
1248 .route_req = tcp_v4_route_req,
1249 .init_seq = tcp_v4_init_sequence,
1250 .send_synack = tcp_v4_send_synack,
1251 .queue_hash_add = inet_csk_reqsk_queue_hash_add,
1254 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1256 /* Never answer to SYNs send to broadcast or multicast */
1257 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1260 return tcp_conn_request(&tcp_request_sock_ops,
1261 &tcp_request_sock_ipv4_ops, sk, skb);
1264 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1267 EXPORT_SYMBOL(tcp_v4_conn_request);
1271 * The three way handshake has completed - we got a valid synack -
1272 * now create the new socket.
1274 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1275 struct request_sock *req,
1276 struct dst_entry *dst)
1278 struct inet_request_sock *ireq;
1279 struct inet_sock *newinet;
1280 struct tcp_sock *newtp;
1282 #ifdef CONFIG_TCP_MD5SIG
1283 struct tcp_md5sig_key *key;
1285 struct ip_options_rcu *inet_opt;
1287 if (sk_acceptq_is_full(sk))
1290 newsk = tcp_create_openreq_child(sk, req, skb);
1294 newsk->sk_gso_type = SKB_GSO_TCPV4;
1295 inet_sk_rx_dst_set(newsk, skb);
1297 newtp = tcp_sk(newsk);
1298 newinet = inet_sk(newsk);
1299 ireq = inet_rsk(req);
1300 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1301 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1302 newinet->inet_saddr = ireq->ir_loc_addr;
1303 inet_opt = ireq->opt;
1304 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1306 newinet->mc_index = inet_iif(skb);
1307 newinet->mc_ttl = ip_hdr(skb)->ttl;
1308 newinet->rcv_tos = ip_hdr(skb)->tos;
1309 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1310 inet_set_txhash(newsk);
1312 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1313 newinet->inet_id = newtp->write_seq ^ jiffies;
1316 dst = inet_csk_route_child_sock(sk, newsk, req);
1320 /* syncookie case : see end of cookie_v4_check() */
1322 sk_setup_caps(newsk, dst);
1324 tcp_ca_openreq_child(newsk, dst);
1326 tcp_sync_mss(newsk, dst_mtu(dst));
1327 newtp->advmss = dst_metric_advmss(dst);
1328 if (tcp_sk(sk)->rx_opt.user_mss &&
1329 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1330 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1332 tcp_initialize_rcv_mss(newsk);
1334 #ifdef CONFIG_TCP_MD5SIG
1335 /* Copy over the MD5 key from the original socket */
1336 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1340 * We're using one, so create a matching key
1341 * on the newsk structure. If we fail to get
1342 * memory, then we end up not copying the key
1345 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1346 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1347 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1351 if (__inet_inherit_port(sk, newsk) < 0)
1353 __inet_hash_nolisten(newsk, NULL);
1358 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1362 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1365 inet_csk_prepare_forced_close(newsk);
1369 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1371 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1373 const struct tcphdr *th = tcp_hdr(skb);
1374 const struct iphdr *iph = ip_hdr(skb);
1375 struct request_sock *req;
1378 req = inet_csk_search_req(sk, th->source, iph->saddr, iph->daddr);
1380 nsk = tcp_check_req(sk, skb, req, false);
1385 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1386 th->source, iph->daddr, th->dest, inet_iif(skb));
1389 if (nsk->sk_state != TCP_TIME_WAIT) {
1393 inet_twsk_put(inet_twsk(nsk));
1397 #ifdef CONFIG_SYN_COOKIES
1399 sk = cookie_v4_check(sk, skb);
1404 /* The socket must have it's spinlock held when we get
1407 * We have a potential double-lock case here, so even when
1408 * doing backlog processing we use the BH locking scheme.
1409 * This is because we cannot sleep with the original spinlock
1412 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1416 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1417 struct dst_entry *dst = sk->sk_rx_dst;
1419 sock_rps_save_rxhash(sk, skb);
1420 sk_mark_napi_id(sk, skb);
1422 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1423 dst->ops->check(dst, 0) == NULL) {
1425 sk->sk_rx_dst = NULL;
1428 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1432 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1435 if (sk->sk_state == TCP_LISTEN) {
1436 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1441 sock_rps_save_rxhash(nsk, skb);
1442 sk_mark_napi_id(sk, skb);
1443 if (tcp_child_process(sk, nsk, skb)) {
1450 sock_rps_save_rxhash(sk, skb);
1452 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1459 tcp_v4_send_reset(rsk, skb);
1462 /* Be careful here. If this function gets more complicated and
1463 * gcc suffers from register pressure on the x86, sk (in %ebx)
1464 * might be destroyed here. This current version compiles correctly,
1465 * but you have been warned.
1470 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1471 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1474 EXPORT_SYMBOL(tcp_v4_do_rcv);
1476 void tcp_v4_early_demux(struct sk_buff *skb)
1478 const struct iphdr *iph;
1479 const struct tcphdr *th;
1482 if (skb->pkt_type != PACKET_HOST)
1485 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1491 if (th->doff < sizeof(struct tcphdr) / 4)
1494 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1495 iph->saddr, th->source,
1496 iph->daddr, ntohs(th->dest),
1500 skb->destructor = sock_edemux;
1501 if (sk_fullsock(sk)) {
1502 struct dst_entry *dst = sk->sk_rx_dst;
1505 dst = dst_check(dst, 0);
1507 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1508 skb_dst_set_noref(skb, dst);
1513 /* Packet is added to VJ-style prequeue for processing in process
1514 * context, if a reader task is waiting. Apparently, this exciting
1515 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1516 * failed somewhere. Latency? Burstiness? Well, at least now we will
1517 * see, why it failed. 8)8) --ANK
1520 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1522 struct tcp_sock *tp = tcp_sk(sk);
1524 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1527 if (skb->len <= tcp_hdrlen(skb) &&
1528 skb_queue_len(&tp->ucopy.prequeue) == 0)
1531 /* Before escaping RCU protected region, we need to take care of skb
1532 * dst. Prequeue is only enabled for established sockets.
1533 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1534 * Instead of doing full sk_rx_dst validity here, let's perform
1535 * an optimistic check.
1537 if (likely(sk->sk_rx_dst))
1542 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1543 tp->ucopy.memory += skb->truesize;
1544 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1545 struct sk_buff *skb1;
1547 BUG_ON(sock_owned_by_user(sk));
1549 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1550 sk_backlog_rcv(sk, skb1);
1551 NET_INC_STATS_BH(sock_net(sk),
1552 LINUX_MIB_TCPPREQUEUEDROPPED);
1555 tp->ucopy.memory = 0;
1556 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1557 wake_up_interruptible_sync_poll(sk_sleep(sk),
1558 POLLIN | POLLRDNORM | POLLRDBAND);
1559 if (!inet_csk_ack_scheduled(sk))
1560 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1561 (3 * tcp_rto_min(sk)) / 4,
1566 EXPORT_SYMBOL(tcp_prequeue);
1572 int tcp_v4_rcv(struct sk_buff *skb)
1574 const struct iphdr *iph;
1575 const struct tcphdr *th;
1578 struct net *net = dev_net(skb->dev);
1580 if (skb->pkt_type != PACKET_HOST)
1583 /* Count it even if it's bad */
1584 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1586 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1591 if (th->doff < sizeof(struct tcphdr) / 4)
1593 if (!pskb_may_pull(skb, th->doff * 4))
1596 /* An explanation is required here, I think.
1597 * Packet length and doff are validated by header prediction,
1598 * provided case of th->doff==0 is eliminated.
1599 * So, we defer the checks. */
1601 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1606 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1607 * barrier() makes sure compiler wont play fool^Waliasing games.
1609 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1610 sizeof(struct inet_skb_parm));
1613 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1614 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1615 skb->len - th->doff * 4);
1616 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1617 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1618 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1619 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1620 TCP_SKB_CB(skb)->sacked = 0;
1622 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1627 if (sk->sk_state == TCP_TIME_WAIT)
1630 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1631 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1632 goto discard_and_relse;
1635 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1636 goto discard_and_relse;
1638 #ifdef CONFIG_TCP_MD5SIG
1640 * We really want to reject the packet as early as possible
1642 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1643 * o There is an MD5 option and we're not expecting one
1645 if (tcp_v4_inbound_md5_hash(sk, skb))
1646 goto discard_and_relse;
1651 if (sk_filter(sk, skb))
1652 goto discard_and_relse;
1654 sk_incoming_cpu_update(sk);
1657 bh_lock_sock_nested(sk);
1659 if (!sock_owned_by_user(sk)) {
1660 if (!tcp_prequeue(sk, skb))
1661 ret = tcp_v4_do_rcv(sk, skb);
1662 } else if (unlikely(sk_add_backlog(sk, skb,
1663 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1665 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1666 goto discard_and_relse;
1675 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1678 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1680 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1682 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1684 tcp_v4_send_reset(NULL, skb);
1688 /* Discard frame. */
1697 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1698 inet_twsk_put(inet_twsk(sk));
1702 if (skb->len < (th->doff << 2)) {
1703 inet_twsk_put(inet_twsk(sk));
1706 if (tcp_checksum_complete(skb)) {
1707 inet_twsk_put(inet_twsk(sk));
1710 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1712 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1714 iph->saddr, th->source,
1715 iph->daddr, th->dest,
1718 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1719 inet_twsk_put(inet_twsk(sk));
1723 /* Fall through to ACK */
1726 tcp_v4_timewait_ack(sk, skb);
1730 case TCP_TW_SUCCESS:;
1735 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1736 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1737 .twsk_unique = tcp_twsk_unique,
1738 .twsk_destructor= tcp_twsk_destructor,
1741 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1743 struct dst_entry *dst = skb_dst(skb);
1747 sk->sk_rx_dst = dst;
1748 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1751 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1753 const struct inet_connection_sock_af_ops ipv4_specific = {
1754 .queue_xmit = ip_queue_xmit,
1755 .send_check = tcp_v4_send_check,
1756 .rebuild_header = inet_sk_rebuild_header,
1757 .sk_rx_dst_set = inet_sk_rx_dst_set,
1758 .conn_request = tcp_v4_conn_request,
1759 .syn_recv_sock = tcp_v4_syn_recv_sock,
1760 .net_header_len = sizeof(struct iphdr),
1761 .setsockopt = ip_setsockopt,
1762 .getsockopt = ip_getsockopt,
1763 .addr2sockaddr = inet_csk_addr2sockaddr,
1764 .sockaddr_len = sizeof(struct sockaddr_in),
1765 .bind_conflict = inet_csk_bind_conflict,
1766 #ifdef CONFIG_COMPAT
1767 .compat_setsockopt = compat_ip_setsockopt,
1768 .compat_getsockopt = compat_ip_getsockopt,
1770 .mtu_reduced = tcp_v4_mtu_reduced,
1772 EXPORT_SYMBOL(ipv4_specific);
1774 #ifdef CONFIG_TCP_MD5SIG
1775 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1776 .md5_lookup = tcp_v4_md5_lookup,
1777 .calc_md5_hash = tcp_v4_md5_hash_skb,
1778 .md5_parse = tcp_v4_parse_md5_keys,
1782 /* NOTE: A lot of things set to zero explicitly by call to
1783 * sk_alloc() so need not be done here.
1785 static int tcp_v4_init_sock(struct sock *sk)
1787 struct inet_connection_sock *icsk = inet_csk(sk);
1791 icsk->icsk_af_ops = &ipv4_specific;
1793 #ifdef CONFIG_TCP_MD5SIG
1794 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1800 void tcp_v4_destroy_sock(struct sock *sk)
1802 struct tcp_sock *tp = tcp_sk(sk);
1804 tcp_clear_xmit_timers(sk);
1806 tcp_cleanup_congestion_control(sk);
1808 /* Cleanup up the write buffer. */
1809 tcp_write_queue_purge(sk);
1811 /* Cleans up our, hopefully empty, out_of_order_queue. */
1812 __skb_queue_purge(&tp->out_of_order_queue);
1814 #ifdef CONFIG_TCP_MD5SIG
1815 /* Clean up the MD5 key list, if any */
1816 if (tp->md5sig_info) {
1817 tcp_clear_md5_list(sk);
1818 kfree_rcu(tp->md5sig_info, rcu);
1819 tp->md5sig_info = NULL;
1823 /* Clean prequeue, it must be empty really */
1824 __skb_queue_purge(&tp->ucopy.prequeue);
1826 /* Clean up a referenced TCP bind bucket. */
1827 if (inet_csk(sk)->icsk_bind_hash)
1830 BUG_ON(tp->fastopen_rsk != NULL);
1832 /* If socket is aborted during connect operation */
1833 tcp_free_fastopen_req(tp);
1835 sk_sockets_allocated_dec(sk);
1836 sock_release_memcg(sk);
1838 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1840 #ifdef CONFIG_PROC_FS
1841 /* Proc filesystem TCP sock list dumping. */
1844 * Get next listener socket follow cur. If cur is NULL, get first socket
1845 * starting from bucket given in st->bucket; when st->bucket is zero the
1846 * very first socket in the hash table is returned.
1848 static void *listening_get_next(struct seq_file *seq, void *cur)
1850 struct inet_connection_sock *icsk;
1851 struct hlist_nulls_node *node;
1852 struct sock *sk = cur;
1853 struct inet_listen_hashbucket *ilb;
1854 struct tcp_iter_state *st = seq->private;
1855 struct net *net = seq_file_net(seq);
1858 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1859 spin_lock_bh(&ilb->lock);
1860 sk = sk_nulls_head(&ilb->head);
1864 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1868 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1869 struct request_sock *req = cur;
1871 icsk = inet_csk(st->syn_wait_sk);
1875 if (req->rsk_ops->family == st->family) {
1881 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1884 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1886 sk = sk_nulls_next(st->syn_wait_sk);
1887 st->state = TCP_SEQ_STATE_LISTENING;
1888 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1890 icsk = inet_csk(sk);
1891 spin_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1892 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1894 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1895 sk = sk_nulls_next(sk);
1898 sk_nulls_for_each_from(sk, node) {
1899 if (!net_eq(sock_net(sk), net))
1901 if (sk->sk_family == st->family) {
1905 icsk = inet_csk(sk);
1906 spin_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1907 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1909 st->uid = sock_i_uid(sk);
1910 st->syn_wait_sk = sk;
1911 st->state = TCP_SEQ_STATE_OPENREQ;
1915 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1917 spin_unlock_bh(&ilb->lock);
1919 if (++st->bucket < INET_LHTABLE_SIZE) {
1920 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1921 spin_lock_bh(&ilb->lock);
1922 sk = sk_nulls_head(&ilb->head);
1930 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1932 struct tcp_iter_state *st = seq->private;
1937 rc = listening_get_next(seq, NULL);
1939 while (rc && *pos) {
1940 rc = listening_get_next(seq, rc);
1946 static inline bool empty_bucket(const struct tcp_iter_state *st)
1948 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1952 * Get first established socket starting from bucket given in st->bucket.
1953 * If st->bucket is zero, the very first socket in the hash is returned.
1955 static void *established_get_first(struct seq_file *seq)
1957 struct tcp_iter_state *st = seq->private;
1958 struct net *net = seq_file_net(seq);
1962 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1964 struct hlist_nulls_node *node;
1965 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1967 /* Lockless fast path for the common case of empty buckets */
1968 if (empty_bucket(st))
1972 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1973 if (sk->sk_family != st->family ||
1974 !net_eq(sock_net(sk), net)) {
1980 spin_unlock_bh(lock);
1986 static void *established_get_next(struct seq_file *seq, void *cur)
1988 struct sock *sk = cur;
1989 struct hlist_nulls_node *node;
1990 struct tcp_iter_state *st = seq->private;
1991 struct net *net = seq_file_net(seq);
1996 sk = sk_nulls_next(sk);
1998 sk_nulls_for_each_from(sk, node) {
1999 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2003 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2005 return established_get_first(seq);
2008 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2010 struct tcp_iter_state *st = seq->private;
2014 rc = established_get_first(seq);
2017 rc = established_get_next(seq, rc);
2023 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2026 struct tcp_iter_state *st = seq->private;
2028 st->state = TCP_SEQ_STATE_LISTENING;
2029 rc = listening_get_idx(seq, &pos);
2032 st->state = TCP_SEQ_STATE_ESTABLISHED;
2033 rc = established_get_idx(seq, pos);
2039 static void *tcp_seek_last_pos(struct seq_file *seq)
2041 struct tcp_iter_state *st = seq->private;
2042 int offset = st->offset;
2043 int orig_num = st->num;
2046 switch (st->state) {
2047 case TCP_SEQ_STATE_OPENREQ:
2048 case TCP_SEQ_STATE_LISTENING:
2049 if (st->bucket >= INET_LHTABLE_SIZE)
2051 st->state = TCP_SEQ_STATE_LISTENING;
2052 rc = listening_get_next(seq, NULL);
2053 while (offset-- && rc)
2054 rc = listening_get_next(seq, rc);
2058 st->state = TCP_SEQ_STATE_ESTABLISHED;
2060 case TCP_SEQ_STATE_ESTABLISHED:
2061 if (st->bucket > tcp_hashinfo.ehash_mask)
2063 rc = established_get_first(seq);
2064 while (offset-- && rc)
2065 rc = established_get_next(seq, rc);
2073 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2075 struct tcp_iter_state *st = seq->private;
2078 if (*pos && *pos == st->last_pos) {
2079 rc = tcp_seek_last_pos(seq);
2084 st->state = TCP_SEQ_STATE_LISTENING;
2088 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2091 st->last_pos = *pos;
2095 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2097 struct tcp_iter_state *st = seq->private;
2100 if (v == SEQ_START_TOKEN) {
2101 rc = tcp_get_idx(seq, 0);
2105 switch (st->state) {
2106 case TCP_SEQ_STATE_OPENREQ:
2107 case TCP_SEQ_STATE_LISTENING:
2108 rc = listening_get_next(seq, v);
2110 st->state = TCP_SEQ_STATE_ESTABLISHED;
2113 rc = established_get_first(seq);
2116 case TCP_SEQ_STATE_ESTABLISHED:
2117 rc = established_get_next(seq, v);
2122 st->last_pos = *pos;
2126 static void tcp_seq_stop(struct seq_file *seq, void *v)
2128 struct tcp_iter_state *st = seq->private;
2130 switch (st->state) {
2131 case TCP_SEQ_STATE_OPENREQ:
2133 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2134 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2136 case TCP_SEQ_STATE_LISTENING:
2137 if (v != SEQ_START_TOKEN)
2138 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2140 case TCP_SEQ_STATE_ESTABLISHED:
2142 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2147 int tcp_seq_open(struct inode *inode, struct file *file)
2149 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2150 struct tcp_iter_state *s;
2153 err = seq_open_net(inode, file, &afinfo->seq_ops,
2154 sizeof(struct tcp_iter_state));
2158 s = ((struct seq_file *)file->private_data)->private;
2159 s->family = afinfo->family;
2163 EXPORT_SYMBOL(tcp_seq_open);
2165 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2168 struct proc_dir_entry *p;
2170 afinfo->seq_ops.start = tcp_seq_start;
2171 afinfo->seq_ops.next = tcp_seq_next;
2172 afinfo->seq_ops.stop = tcp_seq_stop;
2174 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2175 afinfo->seq_fops, afinfo);
2180 EXPORT_SYMBOL(tcp_proc_register);
2182 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2184 remove_proc_entry(afinfo->name, net->proc_net);
2186 EXPORT_SYMBOL(tcp_proc_unregister);
2188 static void get_openreq4(const struct request_sock *req,
2189 struct seq_file *f, int i, kuid_t uid)
2191 const struct inet_request_sock *ireq = inet_rsk(req);
2192 long delta = req->rsk_timer.expires - jiffies;
2194 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2195 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2200 ntohs(ireq->ir_rmt_port),
2202 0, 0, /* could print option size, but that is af dependent. */
2203 1, /* timers active (only the expire timer) */
2204 jiffies_delta_to_clock_t(delta),
2206 from_kuid_munged(seq_user_ns(f), uid),
2207 0, /* non standard timer */
2208 0, /* open_requests have no inode */
2213 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2216 unsigned long timer_expires;
2217 const struct tcp_sock *tp = tcp_sk(sk);
2218 const struct inet_connection_sock *icsk = inet_csk(sk);
2219 const struct inet_sock *inet = inet_sk(sk);
2220 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2221 __be32 dest = inet->inet_daddr;
2222 __be32 src = inet->inet_rcv_saddr;
2223 __u16 destp = ntohs(inet->inet_dport);
2224 __u16 srcp = ntohs(inet->inet_sport);
2227 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2228 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2229 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2231 timer_expires = icsk->icsk_timeout;
2232 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2234 timer_expires = icsk->icsk_timeout;
2235 } else if (timer_pending(&sk->sk_timer)) {
2237 timer_expires = sk->sk_timer.expires;
2240 timer_expires = jiffies;
2243 if (sk->sk_state == TCP_LISTEN)
2244 rx_queue = sk->sk_ack_backlog;
2247 * because we dont lock socket, we might find a transient negative value
2249 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2251 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2252 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2253 i, src, srcp, dest, destp, sk->sk_state,
2254 tp->write_seq - tp->snd_una,
2257 jiffies_delta_to_clock_t(timer_expires - jiffies),
2258 icsk->icsk_retransmits,
2259 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2260 icsk->icsk_probes_out,
2262 atomic_read(&sk->sk_refcnt), sk,
2263 jiffies_to_clock_t(icsk->icsk_rto),
2264 jiffies_to_clock_t(icsk->icsk_ack.ato),
2265 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2267 sk->sk_state == TCP_LISTEN ?
2268 (fastopenq ? fastopenq->max_qlen : 0) :
2269 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2272 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2273 struct seq_file *f, int i)
2277 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
2279 dest = tw->tw_daddr;
2280 src = tw->tw_rcv_saddr;
2281 destp = ntohs(tw->tw_dport);
2282 srcp = ntohs(tw->tw_sport);
2284 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2285 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2286 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2287 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2288 atomic_read(&tw->tw_refcnt), tw);
2293 static int tcp4_seq_show(struct seq_file *seq, void *v)
2295 struct tcp_iter_state *st;
2296 struct sock *sk = v;
2298 seq_setwidth(seq, TMPSZ - 1);
2299 if (v == SEQ_START_TOKEN) {
2300 seq_puts(seq, " sl local_address rem_address st tx_queue "
2301 "rx_queue tr tm->when retrnsmt uid timeout "
2307 switch (st->state) {
2308 case TCP_SEQ_STATE_LISTENING:
2309 case TCP_SEQ_STATE_ESTABLISHED:
2310 if (sk->sk_state == TCP_TIME_WAIT)
2311 get_timewait4_sock(v, seq, st->num);
2313 get_tcp4_sock(v, seq, st->num);
2315 case TCP_SEQ_STATE_OPENREQ:
2316 get_openreq4(v, seq, st->num, st->uid);
2324 static const struct file_operations tcp_afinfo_seq_fops = {
2325 .owner = THIS_MODULE,
2326 .open = tcp_seq_open,
2328 .llseek = seq_lseek,
2329 .release = seq_release_net
2332 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2335 .seq_fops = &tcp_afinfo_seq_fops,
2337 .show = tcp4_seq_show,
2341 static int __net_init tcp4_proc_init_net(struct net *net)
2343 return tcp_proc_register(net, &tcp4_seq_afinfo);
2346 static void __net_exit tcp4_proc_exit_net(struct net *net)
2348 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2351 static struct pernet_operations tcp4_net_ops = {
2352 .init = tcp4_proc_init_net,
2353 .exit = tcp4_proc_exit_net,
2356 int __init tcp4_proc_init(void)
2358 return register_pernet_subsys(&tcp4_net_ops);
2361 void tcp4_proc_exit(void)
2363 unregister_pernet_subsys(&tcp4_net_ops);
2365 #endif /* CONFIG_PROC_FS */
2367 struct proto tcp_prot = {
2369 .owner = THIS_MODULE,
2371 .connect = tcp_v4_connect,
2372 .disconnect = tcp_disconnect,
2373 .accept = inet_csk_accept,
2375 .init = tcp_v4_init_sock,
2376 .destroy = tcp_v4_destroy_sock,
2377 .shutdown = tcp_shutdown,
2378 .setsockopt = tcp_setsockopt,
2379 .getsockopt = tcp_getsockopt,
2380 .recvmsg = tcp_recvmsg,
2381 .sendmsg = tcp_sendmsg,
2382 .sendpage = tcp_sendpage,
2383 .backlog_rcv = tcp_v4_do_rcv,
2384 .release_cb = tcp_release_cb,
2386 .unhash = inet_unhash,
2387 .get_port = inet_csk_get_port,
2388 .enter_memory_pressure = tcp_enter_memory_pressure,
2389 .stream_memory_free = tcp_stream_memory_free,
2390 .sockets_allocated = &tcp_sockets_allocated,
2391 .orphan_count = &tcp_orphan_count,
2392 .memory_allocated = &tcp_memory_allocated,
2393 .memory_pressure = &tcp_memory_pressure,
2394 .sysctl_mem = sysctl_tcp_mem,
2395 .sysctl_wmem = sysctl_tcp_wmem,
2396 .sysctl_rmem = sysctl_tcp_rmem,
2397 .max_header = MAX_TCP_HEADER,
2398 .obj_size = sizeof(struct tcp_sock),
2399 .slab_flags = SLAB_DESTROY_BY_RCU,
2400 .twsk_prot = &tcp_timewait_sock_ops,
2401 .rsk_prot = &tcp_request_sock_ops,
2402 .h.hashinfo = &tcp_hashinfo,
2403 .no_autobind = true,
2404 #ifdef CONFIG_COMPAT
2405 .compat_setsockopt = compat_tcp_setsockopt,
2406 .compat_getsockopt = compat_tcp_getsockopt,
2408 #ifdef CONFIG_MEMCG_KMEM
2409 .init_cgroup = tcp_init_cgroup,
2410 .destroy_cgroup = tcp_destroy_cgroup,
2411 .proto_cgroup = tcp_proto_cgroup,
2414 EXPORT_SYMBOL(tcp_prot);
2416 static void __net_exit tcp_sk_exit(struct net *net)
2420 for_each_possible_cpu(cpu)
2421 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2422 free_percpu(net->ipv4.tcp_sk);
2425 static int __net_init tcp_sk_init(struct net *net)
2429 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2430 if (!net->ipv4.tcp_sk)
2433 for_each_possible_cpu(cpu) {
2436 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2440 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2442 net->ipv4.sysctl_tcp_ecn = 2;
2443 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2444 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2445 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2454 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2456 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2459 static struct pernet_operations __net_initdata tcp_sk_ops = {
2460 .init = tcp_sk_init,
2461 .exit = tcp_sk_exit,
2462 .exit_batch = tcp_sk_exit_batch,
2465 void __init tcp_v4_init(void)
2467 inet_hashinfo_init(&tcp_hashinfo);
2468 if (register_pernet_subsys(&tcp_sk_ops))
2469 panic("Failed to create the TCP control socket.\n");