2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Implementation of the Transmission Control Protocol(TCP).
8 * IPv4 specific functions
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
16 * See tcp.c for author information
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
53 #define pr_fmt(fmt) "TCP: " fmt
55 #include <linux/bottom_half.h>
56 #include <linux/types.h>
57 #include <linux/fcntl.h>
58 #include <linux/module.h>
59 #include <linux/random.h>
60 #include <linux/cache.h>
61 #include <linux/jhash.h>
62 #include <linux/init.h>
63 #include <linux/times.h>
64 #include <linux/slab.h>
66 #include <net/net_namespace.h>
68 #include <net/inet_hashtables.h>
70 #include <net/transp_v6.h>
72 #include <net/inet_common.h>
73 #include <net/timewait_sock.h>
75 #include <net/secure_seq.h>
76 #include <net/busy_poll.h>
78 #include <linux/inet.h>
79 #include <linux/ipv6.h>
80 #include <linux/stddef.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/inetdevice.h>
85 #include <crypto/hash.h>
86 #include <linux/scatterlist.h>
88 #ifdef CONFIG_TCP_MD5SIG
89 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
90 __be32 daddr, __be32 saddr, const struct tcphdr *th);
93 struct inet_hashinfo tcp_hashinfo;
94 EXPORT_SYMBOL(tcp_hashinfo);
96 static u32 tcp_v4_init_seq(const struct sk_buff *skb)
98 return secure_tcp_seq(ip_hdr(skb)->daddr,
101 tcp_hdr(skb)->source);
104 static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
106 return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
109 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
111 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
112 struct tcp_sock *tp = tcp_sk(sk);
114 /* With PAWS, it is safe from the viewpoint
115 of data integrity. Even without PAWS it is safe provided sequence
116 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
118 Actually, the idea is close to VJ's one, only timestamp cache is
119 held not per host, but per port pair and TW bucket is used as state
122 If TW bucket has been already destroyed we fall back to VJ's scheme
123 and use initial timestamp retrieved from peer table.
125 if (tcptw->tw_ts_recent_stamp &&
126 (!twp || (sock_net(sk)->ipv4.sysctl_tcp_tw_reuse &&
127 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
128 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
129 if (tp->write_seq == 0)
131 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
132 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
139 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
141 /* This will initiate an outgoing connection. */
142 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
144 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
145 struct inet_sock *inet = inet_sk(sk);
146 struct tcp_sock *tp = tcp_sk(sk);
147 __be16 orig_sport, orig_dport;
148 __be32 daddr, nexthop;
152 struct ip_options_rcu *inet_opt;
153 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
155 if (addr_len < sizeof(struct sockaddr_in))
158 if (usin->sin_family != AF_INET)
159 return -EAFNOSUPPORT;
161 nexthop = daddr = usin->sin_addr.s_addr;
162 inet_opt = rcu_dereference_protected(inet->inet_opt,
163 lockdep_sock_is_held(sk));
164 if (inet_opt && inet_opt->opt.srr) {
167 nexthop = inet_opt->opt.faddr;
170 orig_sport = inet->inet_sport;
171 orig_dport = usin->sin_port;
172 fl4 = &inet->cork.fl.u.ip4;
173 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
174 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
176 orig_sport, orig_dport, sk);
179 if (err == -ENETUNREACH)
180 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
184 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
189 if (!inet_opt || !inet_opt->opt.srr)
192 if (!inet->inet_saddr)
193 inet->inet_saddr = fl4->saddr;
194 sk_rcv_saddr_set(sk, inet->inet_saddr);
196 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
197 /* Reset inherited state */
198 tp->rx_opt.ts_recent = 0;
199 tp->rx_opt.ts_recent_stamp = 0;
200 if (likely(!tp->repair))
204 inet->inet_dport = usin->sin_port;
205 sk_daddr_set(sk, daddr);
207 inet_csk(sk)->icsk_ext_hdr_len = 0;
209 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
211 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
213 /* Socket identity is still unknown (sport may be zero).
214 * However we set state to SYN-SENT and not releasing socket
215 * lock select source port, enter ourselves into the hash tables and
216 * complete initialization after this.
218 tcp_set_state(sk, TCP_SYN_SENT);
219 err = inet_hash_connect(tcp_death_row, sk);
225 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
226 inet->inet_sport, inet->inet_dport, sk);
232 /* OK, now commit destination to socket. */
233 sk->sk_gso_type = SKB_GSO_TCPV4;
234 sk_setup_caps(sk, &rt->dst);
237 if (likely(!tp->repair)) {
239 tp->write_seq = secure_tcp_seq(inet->inet_saddr,
243 tp->tsoffset = secure_tcp_ts_off(sock_net(sk),
248 inet->inet_id = tp->write_seq ^ jiffies;
250 if (tcp_fastopen_defer_connect(sk, &err))
255 err = tcp_connect(sk);
264 * This unhashes the socket and releases the local port,
267 tcp_set_state(sk, TCP_CLOSE);
269 sk->sk_route_caps = 0;
270 inet->inet_dport = 0;
273 EXPORT_SYMBOL(tcp_v4_connect);
276 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
277 * It can be called through tcp_release_cb() if socket was owned by user
278 * at the time tcp_v4_err() was called to handle ICMP message.
280 void tcp_v4_mtu_reduced(struct sock *sk)
282 struct inet_sock *inet = inet_sk(sk);
283 struct dst_entry *dst;
286 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
288 mtu = tcp_sk(sk)->mtu_info;
289 dst = inet_csk_update_pmtu(sk, mtu);
293 /* Something is about to be wrong... Remember soft error
294 * for the case, if this connection will not able to recover.
296 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
297 sk->sk_err_soft = EMSGSIZE;
301 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
302 ip_sk_accept_pmtu(sk) &&
303 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
304 tcp_sync_mss(sk, mtu);
306 /* Resend the TCP packet because it's
307 * clear that the old packet has been
308 * dropped. This is the new "fast" path mtu
311 tcp_simple_retransmit(sk);
312 } /* else let the usual retransmit timer handle it */
314 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
316 static void do_redirect(struct sk_buff *skb, struct sock *sk)
318 struct dst_entry *dst = __sk_dst_check(sk, 0);
321 dst->ops->redirect(dst, sk, skb);
325 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
326 void tcp_req_err(struct sock *sk, u32 seq, bool abort)
328 struct request_sock *req = inet_reqsk(sk);
329 struct net *net = sock_net(sk);
331 /* ICMPs are not backlogged, hence we cannot get
332 * an established socket here.
334 if (seq != tcp_rsk(req)->snt_isn) {
335 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
338 * Still in SYN_RECV, just remove it silently.
339 * There is no good way to pass the error to the newly
340 * created socket, and POSIX does not want network
341 * errors returned from accept().
343 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
344 tcp_listendrop(req->rsk_listener);
348 EXPORT_SYMBOL(tcp_req_err);
351 * This routine is called by the ICMP module when it gets some
352 * sort of error condition. If err < 0 then the socket should
353 * be closed and the error returned to the user. If err > 0
354 * it's just the icmp type << 8 | icmp code. After adjustment
355 * header points to the first 8 bytes of the tcp header. We need
356 * to find the appropriate port.
358 * The locking strategy used here is very "optimistic". When
359 * someone else accesses the socket the ICMP is just dropped
360 * and for some paths there is no check at all.
361 * A more general error queue to queue errors for later handling
362 * is probably better.
366 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
368 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
369 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
370 struct inet_connection_sock *icsk;
372 struct inet_sock *inet;
373 const int type = icmp_hdr(icmp_skb)->type;
374 const int code = icmp_hdr(icmp_skb)->code;
377 struct request_sock *fastopen;
382 struct net *net = dev_net(icmp_skb->dev);
384 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
385 th->dest, iph->saddr, ntohs(th->source),
386 inet_iif(icmp_skb), 0);
388 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
391 if (sk->sk_state == TCP_TIME_WAIT) {
392 inet_twsk_put(inet_twsk(sk));
395 seq = ntohl(th->seq);
396 if (sk->sk_state == TCP_NEW_SYN_RECV)
397 return tcp_req_err(sk, seq,
398 type == ICMP_PARAMETERPROB ||
399 type == ICMP_TIME_EXCEEDED ||
400 (type == ICMP_DEST_UNREACH &&
401 (code == ICMP_NET_UNREACH ||
402 code == ICMP_HOST_UNREACH)));
405 /* If too many ICMPs get dropped on busy
406 * servers this needs to be solved differently.
407 * We do take care of PMTU discovery (RFC1191) special case :
408 * we can receive locally generated ICMP messages while socket is held.
410 if (sock_owned_by_user(sk)) {
411 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
412 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
414 if (sk->sk_state == TCP_CLOSE)
417 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
418 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
424 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
425 fastopen = tp->fastopen_rsk;
426 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
427 if (sk->sk_state != TCP_LISTEN &&
428 !between(seq, snd_una, tp->snd_nxt)) {
429 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
435 if (!sock_owned_by_user(sk))
436 do_redirect(icmp_skb, sk);
438 case ICMP_SOURCE_QUENCH:
439 /* Just silently ignore these. */
441 case ICMP_PARAMETERPROB:
444 case ICMP_DEST_UNREACH:
445 if (code > NR_ICMP_UNREACH)
448 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
449 /* We are not interested in TCP_LISTEN and open_requests
450 * (SYN-ACKs send out by Linux are always <576bytes so
451 * they should go through unfragmented).
453 if (sk->sk_state == TCP_LISTEN)
457 if (!sock_owned_by_user(sk)) {
458 tcp_v4_mtu_reduced(sk);
460 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
466 err = icmp_err_convert[code].errno;
467 /* check if icmp_skb allows revert of backoff
468 * (see draft-zimmermann-tcp-lcd) */
469 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
471 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
472 !icsk->icsk_backoff || fastopen)
475 if (sock_owned_by_user(sk))
478 icsk->icsk_backoff--;
479 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
481 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
483 skb = tcp_rtx_queue_head(sk);
486 tcp_mstamp_refresh(tp);
487 delta_us = (u32)(tp->tcp_mstamp - skb->skb_mstamp);
488 remaining = icsk->icsk_rto -
489 usecs_to_jiffies(delta_us);
492 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
493 remaining, TCP_RTO_MAX);
495 /* RTO revert clocked out retransmission.
496 * Will retransmit now */
497 tcp_retransmit_timer(sk);
501 case ICMP_TIME_EXCEEDED:
508 switch (sk->sk_state) {
511 /* Only in fast or simultaneous open. If a fast open socket is
512 * is already accepted it is treated as a connected one below.
514 if (fastopen && !fastopen->sk)
517 if (!sock_owned_by_user(sk)) {
520 sk->sk_error_report(sk);
524 sk->sk_err_soft = err;
529 /* If we've already connected we will keep trying
530 * until we time out, or the user gives up.
532 * rfc1122 4.2.3.9 allows to consider as hard errors
533 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
534 * but it is obsoleted by pmtu discovery).
536 * Note, that in modern internet, where routing is unreliable
537 * and in each dark corner broken firewalls sit, sending random
538 * errors ordered by their masters even this two messages finally lose
539 * their original sense (even Linux sends invalid PORT_UNREACHs)
541 * Now we are in compliance with RFCs.
546 if (!sock_owned_by_user(sk) && inet->recverr) {
548 sk->sk_error_report(sk);
549 } else { /* Only an error on timeout */
550 sk->sk_err_soft = err;
558 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
560 struct tcphdr *th = tcp_hdr(skb);
562 if (skb->ip_summed == CHECKSUM_PARTIAL) {
563 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
564 skb->csum_start = skb_transport_header(skb) - skb->head;
565 skb->csum_offset = offsetof(struct tcphdr, check);
567 th->check = tcp_v4_check(skb->len, saddr, daddr,
574 /* This routine computes an IPv4 TCP checksum. */
575 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
577 const struct inet_sock *inet = inet_sk(sk);
579 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
581 EXPORT_SYMBOL(tcp_v4_send_check);
584 * This routine will send an RST to the other tcp.
586 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
588 * Answer: if a packet caused RST, it is not for a socket
589 * existing in our system, if it is matched to a socket,
590 * it is just duplicate segment or bug in other side's TCP.
591 * So that we build reply only basing on parameters
592 * arrived with segment.
593 * Exception: precedence violation. We do not implement it in any case.
596 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
598 const struct tcphdr *th = tcp_hdr(skb);
601 #ifdef CONFIG_TCP_MD5SIG
602 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
605 struct ip_reply_arg arg;
606 #ifdef CONFIG_TCP_MD5SIG
607 struct tcp_md5sig_key *key = NULL;
608 const __u8 *hash_location = NULL;
609 unsigned char newhash[16];
611 struct sock *sk1 = NULL;
615 /* Never send a reset in response to a reset. */
619 /* If sk not NULL, it means we did a successful lookup and incoming
620 * route had to be correct. prequeue might have dropped our dst.
622 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
625 /* Swap the send and the receive. */
626 memset(&rep, 0, sizeof(rep));
627 rep.th.dest = th->source;
628 rep.th.source = th->dest;
629 rep.th.doff = sizeof(struct tcphdr) / 4;
633 rep.th.seq = th->ack_seq;
636 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
637 skb->len - (th->doff << 2));
640 memset(&arg, 0, sizeof(arg));
641 arg.iov[0].iov_base = (unsigned char *)&rep;
642 arg.iov[0].iov_len = sizeof(rep.th);
644 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
645 #ifdef CONFIG_TCP_MD5SIG
647 hash_location = tcp_parse_md5sig_option(th);
648 if (sk && sk_fullsock(sk)) {
649 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
650 &ip_hdr(skb)->saddr, AF_INET);
651 } else if (hash_location) {
653 * active side is lost. Try to find listening socket through
654 * source port, and then find md5 key through listening socket.
655 * we are not loose security here:
656 * Incoming packet is checked with md5 hash with finding key,
657 * no RST generated if md5 hash doesn't match.
659 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
661 th->source, ip_hdr(skb)->daddr,
662 ntohs(th->source), inet_iif(skb),
664 /* don't send rst if it can't find key */
668 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
669 &ip_hdr(skb)->saddr, AF_INET);
674 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
675 if (genhash || memcmp(hash_location, newhash, 16) != 0)
681 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
683 (TCPOPT_MD5SIG << 8) |
685 /* Update length and the length the header thinks exists */
686 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
687 rep.th.doff = arg.iov[0].iov_len / 4;
689 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
690 key, ip_hdr(skb)->saddr,
691 ip_hdr(skb)->daddr, &rep.th);
694 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
695 ip_hdr(skb)->saddr, /* XXX */
696 arg.iov[0].iov_len, IPPROTO_TCP, 0);
697 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
698 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
700 /* When socket is gone, all binding information is lost.
701 * routing might fail in this case. No choice here, if we choose to force
702 * input interface, we will misroute in case of asymmetric route.
705 arg.bound_dev_if = sk->sk_bound_dev_if;
707 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
708 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
710 arg.tos = ip_hdr(skb)->tos;
711 arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
713 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
714 skb, &TCP_SKB_CB(skb)->header.h4.opt,
715 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
716 &arg, arg.iov[0].iov_len);
718 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
719 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
722 #ifdef CONFIG_TCP_MD5SIG
728 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
729 outside socket context is ugly, certainly. What can I do?
732 static void tcp_v4_send_ack(const struct sock *sk,
733 struct sk_buff *skb, u32 seq, u32 ack,
734 u32 win, u32 tsval, u32 tsecr, int oif,
735 struct tcp_md5sig_key *key,
736 int reply_flags, u8 tos)
738 const struct tcphdr *th = tcp_hdr(skb);
741 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
742 #ifdef CONFIG_TCP_MD5SIG
743 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
747 struct net *net = sock_net(sk);
748 struct ip_reply_arg arg;
750 memset(&rep.th, 0, sizeof(struct tcphdr));
751 memset(&arg, 0, sizeof(arg));
753 arg.iov[0].iov_base = (unsigned char *)&rep;
754 arg.iov[0].iov_len = sizeof(rep.th);
756 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
757 (TCPOPT_TIMESTAMP << 8) |
759 rep.opt[1] = htonl(tsval);
760 rep.opt[2] = htonl(tsecr);
761 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
764 /* Swap the send and the receive. */
765 rep.th.dest = th->source;
766 rep.th.source = th->dest;
767 rep.th.doff = arg.iov[0].iov_len / 4;
768 rep.th.seq = htonl(seq);
769 rep.th.ack_seq = htonl(ack);
771 rep.th.window = htons(win);
773 #ifdef CONFIG_TCP_MD5SIG
775 int offset = (tsecr) ? 3 : 0;
777 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
779 (TCPOPT_MD5SIG << 8) |
781 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
782 rep.th.doff = arg.iov[0].iov_len/4;
784 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
785 key, ip_hdr(skb)->saddr,
786 ip_hdr(skb)->daddr, &rep.th);
789 arg.flags = reply_flags;
790 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
791 ip_hdr(skb)->saddr, /* XXX */
792 arg.iov[0].iov_len, IPPROTO_TCP, 0);
793 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
795 arg.bound_dev_if = oif;
797 arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
799 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
800 skb, &TCP_SKB_CB(skb)->header.h4.opt,
801 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
802 &arg, arg.iov[0].iov_len);
804 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
808 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
810 struct inet_timewait_sock *tw = inet_twsk(sk);
811 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
813 tcp_v4_send_ack(sk, skb,
814 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
815 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
816 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
819 tcp_twsk_md5_key(tcptw),
820 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
827 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
828 struct request_sock *req)
830 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
831 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
833 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
837 * The window field (SEG.WND) of every outgoing segment, with the
838 * exception of <SYN> segments, MUST be right-shifted by
839 * Rcv.Wind.Shift bits:
841 tcp_v4_send_ack(sk, skb, seq,
842 tcp_rsk(req)->rcv_nxt,
843 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
844 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
847 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
849 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
854 * Send a SYN-ACK after having received a SYN.
855 * This still operates on a request_sock only, not on a big
858 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
860 struct request_sock *req,
861 struct tcp_fastopen_cookie *foc,
862 enum tcp_synack_type synack_type)
864 const struct inet_request_sock *ireq = inet_rsk(req);
869 /* First, grab a route. */
870 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
873 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
876 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
878 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
880 rcu_dereference(ireq->ireq_opt));
881 err = net_xmit_eval(err);
888 * IPv4 request_sock destructor.
890 static void tcp_v4_reqsk_destructor(struct request_sock *req)
892 kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
895 #ifdef CONFIG_TCP_MD5SIG
897 * RFC2385 MD5 checksumming requires a mapping of
898 * IP address->MD5 Key.
899 * We need to maintain these in the sk structure.
902 /* Find the Key structure for an address. */
903 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
904 const union tcp_md5_addr *addr,
907 const struct tcp_sock *tp = tcp_sk(sk);
908 struct tcp_md5sig_key *key;
909 const struct tcp_md5sig_info *md5sig;
911 struct tcp_md5sig_key *best_match = NULL;
914 /* caller either holds rcu_read_lock() or socket lock */
915 md5sig = rcu_dereference_check(tp->md5sig_info,
916 lockdep_sock_is_held(sk));
920 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
921 if (key->family != family)
924 if (family == AF_INET) {
925 mask = inet_make_mask(key->prefixlen);
926 match = (key->addr.a4.s_addr & mask) ==
927 (addr->a4.s_addr & mask);
928 #if IS_ENABLED(CONFIG_IPV6)
929 } else if (family == AF_INET6) {
930 match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
937 if (match && (!best_match ||
938 key->prefixlen > best_match->prefixlen))
943 EXPORT_SYMBOL(tcp_md5_do_lookup);
945 static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
946 const union tcp_md5_addr *addr,
947 int family, u8 prefixlen)
949 const struct tcp_sock *tp = tcp_sk(sk);
950 struct tcp_md5sig_key *key;
951 unsigned int size = sizeof(struct in_addr);
952 const struct tcp_md5sig_info *md5sig;
954 /* caller either holds rcu_read_lock() or socket lock */
955 md5sig = rcu_dereference_check(tp->md5sig_info,
956 lockdep_sock_is_held(sk));
959 #if IS_ENABLED(CONFIG_IPV6)
960 if (family == AF_INET6)
961 size = sizeof(struct in6_addr);
963 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
964 if (key->family != family)
966 if (!memcmp(&key->addr, addr, size) &&
967 key->prefixlen == prefixlen)
973 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
974 const struct sock *addr_sk)
976 const union tcp_md5_addr *addr;
978 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
979 return tcp_md5_do_lookup(sk, addr, AF_INET);
981 EXPORT_SYMBOL(tcp_v4_md5_lookup);
983 /* This can be called on a newly created socket, from other files */
984 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
985 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
988 /* Add Key to the list */
989 struct tcp_md5sig_key *key;
990 struct tcp_sock *tp = tcp_sk(sk);
991 struct tcp_md5sig_info *md5sig;
993 key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen);
995 /* Pre-existing entry - just update that one. */
996 memcpy(key->key, newkey, newkeylen);
997 key->keylen = newkeylen;
1001 md5sig = rcu_dereference_protected(tp->md5sig_info,
1002 lockdep_sock_is_held(sk));
1004 md5sig = kmalloc(sizeof(*md5sig), gfp);
1008 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1009 INIT_HLIST_HEAD(&md5sig->head);
1010 rcu_assign_pointer(tp->md5sig_info, md5sig);
1013 key = sock_kmalloc(sk, sizeof(*key), gfp);
1016 if (!tcp_alloc_md5sig_pool()) {
1017 sock_kfree_s(sk, key, sizeof(*key));
1021 memcpy(key->key, newkey, newkeylen);
1022 key->keylen = newkeylen;
1023 key->family = family;
1024 key->prefixlen = prefixlen;
1025 memcpy(&key->addr, addr,
1026 (family == AF_INET6) ? sizeof(struct in6_addr) :
1027 sizeof(struct in_addr));
1028 hlist_add_head_rcu(&key->node, &md5sig->head);
1031 EXPORT_SYMBOL(tcp_md5_do_add);
1033 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1036 struct tcp_md5sig_key *key;
1038 key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen);
1041 hlist_del_rcu(&key->node);
1042 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1043 kfree_rcu(key, rcu);
1046 EXPORT_SYMBOL(tcp_md5_do_del);
1048 static void tcp_clear_md5_list(struct sock *sk)
1050 struct tcp_sock *tp = tcp_sk(sk);
1051 struct tcp_md5sig_key *key;
1052 struct hlist_node *n;
1053 struct tcp_md5sig_info *md5sig;
1055 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1057 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1058 hlist_del_rcu(&key->node);
1059 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1060 kfree_rcu(key, rcu);
1064 static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1065 char __user *optval, int optlen)
1067 struct tcp_md5sig cmd;
1068 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1071 if (optlen < sizeof(cmd))
1074 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1077 if (sin->sin_family != AF_INET)
1080 if (optname == TCP_MD5SIG_EXT &&
1081 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1082 prefixlen = cmd.tcpm_prefixlen;
1087 if (!cmd.tcpm_keylen)
1088 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1089 AF_INET, prefixlen);
1091 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1094 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1095 AF_INET, prefixlen, cmd.tcpm_key, cmd.tcpm_keylen,
1099 static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1100 __be32 daddr, __be32 saddr,
1101 const struct tcphdr *th, int nbytes)
1103 struct tcp4_pseudohdr *bp;
1104 struct scatterlist sg;
1111 bp->protocol = IPPROTO_TCP;
1112 bp->len = cpu_to_be16(nbytes);
1114 _th = (struct tcphdr *)(bp + 1);
1115 memcpy(_th, th, sizeof(*th));
1118 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1119 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1120 sizeof(*bp) + sizeof(*th));
1121 return crypto_ahash_update(hp->md5_req);
1124 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1125 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1127 struct tcp_md5sig_pool *hp;
1128 struct ahash_request *req;
1130 hp = tcp_get_md5sig_pool();
1132 goto clear_hash_noput;
1135 if (crypto_ahash_init(req))
1137 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1139 if (tcp_md5_hash_key(hp, key))
1141 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1142 if (crypto_ahash_final(req))
1145 tcp_put_md5sig_pool();
1149 tcp_put_md5sig_pool();
1151 memset(md5_hash, 0, 16);
1155 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1156 const struct sock *sk,
1157 const struct sk_buff *skb)
1159 struct tcp_md5sig_pool *hp;
1160 struct ahash_request *req;
1161 const struct tcphdr *th = tcp_hdr(skb);
1162 __be32 saddr, daddr;
1164 if (sk) { /* valid for establish/request sockets */
1165 saddr = sk->sk_rcv_saddr;
1166 daddr = sk->sk_daddr;
1168 const struct iphdr *iph = ip_hdr(skb);
1173 hp = tcp_get_md5sig_pool();
1175 goto clear_hash_noput;
1178 if (crypto_ahash_init(req))
1181 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1183 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1185 if (tcp_md5_hash_key(hp, key))
1187 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1188 if (crypto_ahash_final(req))
1191 tcp_put_md5sig_pool();
1195 tcp_put_md5sig_pool();
1197 memset(md5_hash, 0, 16);
1200 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1204 /* Called with rcu_read_lock() */
1205 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1206 const struct sk_buff *skb)
1208 #ifdef CONFIG_TCP_MD5SIG
1210 * This gets called for each TCP segment that arrives
1211 * so we want to be efficient.
1212 * We have 3 drop cases:
1213 * o No MD5 hash and one expected.
1214 * o MD5 hash and we're not expecting one.
1215 * o MD5 hash and its wrong.
1217 const __u8 *hash_location = NULL;
1218 struct tcp_md5sig_key *hash_expected;
1219 const struct iphdr *iph = ip_hdr(skb);
1220 const struct tcphdr *th = tcp_hdr(skb);
1222 unsigned char newhash[16];
1224 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1226 hash_location = tcp_parse_md5sig_option(th);
1228 /* We've parsed the options - do we have a hash? */
1229 if (!hash_expected && !hash_location)
1232 if (hash_expected && !hash_location) {
1233 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1237 if (!hash_expected && hash_location) {
1238 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1242 /* Okay, so this is hash_expected and hash_location -
1243 * so we need to calculate the checksum.
1245 genhash = tcp_v4_md5_hash_skb(newhash,
1249 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1250 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1251 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1252 &iph->saddr, ntohs(th->source),
1253 &iph->daddr, ntohs(th->dest),
1254 genhash ? " tcp_v4_calc_md5_hash failed"
1263 static void tcp_v4_init_req(struct request_sock *req,
1264 const struct sock *sk_listener,
1265 struct sk_buff *skb)
1267 struct inet_request_sock *ireq = inet_rsk(req);
1268 struct net *net = sock_net(sk_listener);
1270 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1271 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1272 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1275 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1277 const struct request_sock *req)
1279 return inet_csk_route_req(sk, &fl->u.ip4, req);
1282 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1284 .obj_size = sizeof(struct tcp_request_sock),
1285 .rtx_syn_ack = tcp_rtx_synack,
1286 .send_ack = tcp_v4_reqsk_send_ack,
1287 .destructor = tcp_v4_reqsk_destructor,
1288 .send_reset = tcp_v4_send_reset,
1289 .syn_ack_timeout = tcp_syn_ack_timeout,
1292 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1293 .mss_clamp = TCP_MSS_DEFAULT,
1294 #ifdef CONFIG_TCP_MD5SIG
1295 .req_md5_lookup = tcp_v4_md5_lookup,
1296 .calc_md5_hash = tcp_v4_md5_hash_skb,
1298 .init_req = tcp_v4_init_req,
1299 #ifdef CONFIG_SYN_COOKIES
1300 .cookie_init_seq = cookie_v4_init_sequence,
1302 .route_req = tcp_v4_route_req,
1303 .init_seq = tcp_v4_init_seq,
1304 .init_ts_off = tcp_v4_init_ts_off,
1305 .send_synack = tcp_v4_send_synack,
1308 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1310 /* Never answer to SYNs send to broadcast or multicast */
1311 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1314 return tcp_conn_request(&tcp_request_sock_ops,
1315 &tcp_request_sock_ipv4_ops, sk, skb);
1321 EXPORT_SYMBOL(tcp_v4_conn_request);
1325 * The three way handshake has completed - we got a valid synack -
1326 * now create the new socket.
1328 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1329 struct request_sock *req,
1330 struct dst_entry *dst,
1331 struct request_sock *req_unhash,
1334 struct inet_request_sock *ireq;
1335 struct inet_sock *newinet;
1336 struct tcp_sock *newtp;
1338 #ifdef CONFIG_TCP_MD5SIG
1339 struct tcp_md5sig_key *key;
1341 struct ip_options_rcu *inet_opt;
1343 if (sk_acceptq_is_full(sk))
1346 newsk = tcp_create_openreq_child(sk, req, skb);
1350 newsk->sk_gso_type = SKB_GSO_TCPV4;
1351 inet_sk_rx_dst_set(newsk, skb);
1353 newtp = tcp_sk(newsk);
1354 newinet = inet_sk(newsk);
1355 ireq = inet_rsk(req);
1356 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1357 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1358 newsk->sk_bound_dev_if = ireq->ir_iif;
1359 newinet->inet_saddr = ireq->ir_loc_addr;
1360 inet_opt = rcu_dereference(ireq->ireq_opt);
1361 RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1362 newinet->mc_index = inet_iif(skb);
1363 newinet->mc_ttl = ip_hdr(skb)->ttl;
1364 newinet->rcv_tos = ip_hdr(skb)->tos;
1365 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1367 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1368 newinet->inet_id = newtp->write_seq ^ jiffies;
1371 dst = inet_csk_route_child_sock(sk, newsk, req);
1375 /* syncookie case : see end of cookie_v4_check() */
1377 sk_setup_caps(newsk, dst);
1379 tcp_ca_openreq_child(newsk, dst);
1381 tcp_sync_mss(newsk, dst_mtu(dst));
1382 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1384 tcp_initialize_rcv_mss(newsk);
1386 #ifdef CONFIG_TCP_MD5SIG
1387 /* Copy over the MD5 key from the original socket */
1388 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1392 * We're using one, so create a matching key
1393 * on the newsk structure. If we fail to get
1394 * memory, then we end up not copying the key
1397 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1398 AF_INET, 32, key->key, key->keylen, GFP_ATOMIC);
1399 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1403 if (__inet_inherit_port(sk, newsk) < 0)
1405 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1406 if (likely(*own_req)) {
1407 tcp_move_syn(newtp, req);
1408 ireq->ireq_opt = NULL;
1410 newinet->inet_opt = NULL;
1415 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1422 newinet->inet_opt = NULL;
1423 inet_csk_prepare_forced_close(newsk);
1427 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1429 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1431 #ifdef CONFIG_SYN_COOKIES
1432 const struct tcphdr *th = tcp_hdr(skb);
1435 sk = cookie_v4_check(sk, skb);
1440 /* The socket must have it's spinlock held when we get
1441 * here, unless it is a TCP_LISTEN socket.
1443 * We have a potential double-lock case here, so even when
1444 * doing backlog processing we use the BH locking scheme.
1445 * This is because we cannot sleep with the original spinlock
1448 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1452 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1453 struct dst_entry *dst = sk->sk_rx_dst;
1455 sock_rps_save_rxhash(sk, skb);
1456 sk_mark_napi_id(sk, skb);
1458 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1459 !dst->ops->check(dst, 0)) {
1461 sk->sk_rx_dst = NULL;
1464 tcp_rcv_established(sk, skb, tcp_hdr(skb));
1468 if (tcp_checksum_complete(skb))
1471 if (sk->sk_state == TCP_LISTEN) {
1472 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1477 if (tcp_child_process(sk, nsk, skb)) {
1484 sock_rps_save_rxhash(sk, skb);
1486 if (tcp_rcv_state_process(sk, skb)) {
1493 tcp_v4_send_reset(rsk, skb);
1496 /* Be careful here. If this function gets more complicated and
1497 * gcc suffers from register pressure on the x86, sk (in %ebx)
1498 * might be destroyed here. This current version compiles correctly,
1499 * but you have been warned.
1504 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1505 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1508 EXPORT_SYMBOL(tcp_v4_do_rcv);
1510 int tcp_v4_early_demux(struct sk_buff *skb)
1512 const struct iphdr *iph;
1513 const struct tcphdr *th;
1516 if (skb->pkt_type != PACKET_HOST)
1519 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1525 if (th->doff < sizeof(struct tcphdr) / 4)
1528 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1529 iph->saddr, th->source,
1530 iph->daddr, ntohs(th->dest),
1531 skb->skb_iif, inet_sdif(skb));
1534 skb->destructor = sock_edemux;
1535 if (sk_fullsock(sk)) {
1536 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1539 dst = dst_check(dst, 0);
1541 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1542 skb_dst_set_noref(skb, dst);
1548 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1550 u32 limit = sk->sk_rcvbuf + sk->sk_sndbuf;
1552 /* Only socket owner can try to collapse/prune rx queues
1553 * to reduce memory overhead, so add a little headroom here.
1554 * Few sockets backlog are possibly concurrently non empty.
1558 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1559 * we can fix skb->truesize to its real value to avoid future drops.
1560 * This is valid because skb is not yet charged to the socket.
1561 * It has been noticed pure SACK packets were sometimes dropped
1562 * (if cooked by drivers without copybreak feature).
1566 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1568 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1573 EXPORT_SYMBOL(tcp_add_backlog);
1575 int tcp_filter(struct sock *sk, struct sk_buff *skb)
1577 struct tcphdr *th = (struct tcphdr *)skb->data;
1578 unsigned int eaten = skb->len;
1581 err = sk_filter_trim_cap(sk, skb, th->doff * 4);
1584 TCP_SKB_CB(skb)->end_seq -= eaten;
1588 EXPORT_SYMBOL(tcp_filter);
1594 int tcp_v4_rcv(struct sk_buff *skb)
1596 struct net *net = dev_net(skb->dev);
1597 int sdif = inet_sdif(skb);
1598 const struct iphdr *iph;
1599 const struct tcphdr *th;
1604 if (skb->pkt_type != PACKET_HOST)
1607 /* Count it even if it's bad */
1608 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1610 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1613 th = (const struct tcphdr *)skb->data;
1615 if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1617 if (!pskb_may_pull(skb, th->doff * 4))
1620 /* An explanation is required here, I think.
1621 * Packet length and doff are validated by header prediction,
1622 * provided case of th->doff==0 is eliminated.
1623 * So, we defer the checks. */
1625 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1628 th = (const struct tcphdr *)skb->data;
1630 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1631 * barrier() makes sure compiler wont play fool^Waliasing games.
1633 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1634 sizeof(struct inet_skb_parm));
1637 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1638 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1639 skb->len - th->doff * 4);
1640 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1641 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1642 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1643 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1644 TCP_SKB_CB(skb)->sacked = 0;
1645 TCP_SKB_CB(skb)->has_rxtstamp =
1646 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1649 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1650 th->dest, sdif, &refcounted);
1655 if (sk->sk_state == TCP_TIME_WAIT)
1658 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1659 struct request_sock *req = inet_reqsk(sk);
1662 sk = req->rsk_listener;
1663 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1664 sk_drops_add(sk, skb);
1668 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1669 inet_csk_reqsk_queue_drop_and_put(sk, req);
1672 /* We own a reference on the listener, increase it again
1673 * as we might lose it too soon.
1678 if (!tcp_filter(sk, skb))
1679 nsk = tcp_check_req(sk, skb, req, false);
1682 goto discard_and_relse;
1686 } else if (tcp_child_process(sk, nsk, skb)) {
1687 tcp_v4_send_reset(nsk, skb);
1688 goto discard_and_relse;
1694 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1695 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1696 goto discard_and_relse;
1699 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1700 goto discard_and_relse;
1702 if (tcp_v4_inbound_md5_hash(sk, skb))
1703 goto discard_and_relse;
1707 if (tcp_filter(sk, skb))
1708 goto discard_and_relse;
1709 th = (const struct tcphdr *)skb->data;
1714 if (sk->sk_state == TCP_LISTEN) {
1715 ret = tcp_v4_do_rcv(sk, skb);
1716 goto put_and_return;
1719 sk_incoming_cpu_update(sk);
1721 bh_lock_sock_nested(sk);
1722 tcp_segs_in(tcp_sk(sk), skb);
1724 if (!sock_owned_by_user(sk)) {
1725 ret = tcp_v4_do_rcv(sk, skb);
1726 } else if (tcp_add_backlog(sk, skb)) {
1727 goto discard_and_relse;
1738 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1741 if (tcp_checksum_complete(skb)) {
1743 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1745 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1747 tcp_v4_send_reset(NULL, skb);
1751 /* Discard frame. */
1756 sk_drops_add(sk, skb);
1762 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1763 inet_twsk_put(inet_twsk(sk));
1767 if (tcp_checksum_complete(skb)) {
1768 inet_twsk_put(inet_twsk(sk));
1771 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1773 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1776 iph->saddr, th->source,
1777 iph->daddr, th->dest,
1781 inet_twsk_deschedule_put(inet_twsk(sk));
1790 tcp_v4_timewait_ack(sk, skb);
1793 tcp_v4_send_reset(sk, skb);
1794 inet_twsk_deschedule_put(inet_twsk(sk));
1796 case TCP_TW_SUCCESS:;
1801 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1802 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1803 .twsk_unique = tcp_twsk_unique,
1804 .twsk_destructor= tcp_twsk_destructor,
1807 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1809 struct dst_entry *dst = skb_dst(skb);
1811 if (dst && dst_hold_safe(dst)) {
1812 sk->sk_rx_dst = dst;
1813 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1816 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1818 const struct inet_connection_sock_af_ops ipv4_specific = {
1819 .queue_xmit = ip_queue_xmit,
1820 .send_check = tcp_v4_send_check,
1821 .rebuild_header = inet_sk_rebuild_header,
1822 .sk_rx_dst_set = inet_sk_rx_dst_set,
1823 .conn_request = tcp_v4_conn_request,
1824 .syn_recv_sock = tcp_v4_syn_recv_sock,
1825 .net_header_len = sizeof(struct iphdr),
1826 .setsockopt = ip_setsockopt,
1827 .getsockopt = ip_getsockopt,
1828 .addr2sockaddr = inet_csk_addr2sockaddr,
1829 .sockaddr_len = sizeof(struct sockaddr_in),
1830 #ifdef CONFIG_COMPAT
1831 .compat_setsockopt = compat_ip_setsockopt,
1832 .compat_getsockopt = compat_ip_getsockopt,
1834 .mtu_reduced = tcp_v4_mtu_reduced,
1836 EXPORT_SYMBOL(ipv4_specific);
1838 #ifdef CONFIG_TCP_MD5SIG
1839 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1840 .md5_lookup = tcp_v4_md5_lookup,
1841 .calc_md5_hash = tcp_v4_md5_hash_skb,
1842 .md5_parse = tcp_v4_parse_md5_keys,
1846 /* NOTE: A lot of things set to zero explicitly by call to
1847 * sk_alloc() so need not be done here.
1849 static int tcp_v4_init_sock(struct sock *sk)
1851 struct inet_connection_sock *icsk = inet_csk(sk);
1855 icsk->icsk_af_ops = &ipv4_specific;
1857 #ifdef CONFIG_TCP_MD5SIG
1858 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1864 void tcp_v4_destroy_sock(struct sock *sk)
1866 struct tcp_sock *tp = tcp_sk(sk);
1868 tcp_clear_xmit_timers(sk);
1870 tcp_cleanup_congestion_control(sk);
1872 tcp_cleanup_ulp(sk);
1874 /* Cleanup up the write buffer. */
1875 tcp_write_queue_purge(sk);
1877 /* Check if we want to disable active TFO */
1878 tcp_fastopen_active_disable_ofo_check(sk);
1880 /* Cleans up our, hopefully empty, out_of_order_queue. */
1881 skb_rbtree_purge(&tp->out_of_order_queue);
1883 #ifdef CONFIG_TCP_MD5SIG
1884 /* Clean up the MD5 key list, if any */
1885 if (tp->md5sig_info) {
1886 tcp_clear_md5_list(sk);
1887 kfree_rcu(tp->md5sig_info, rcu);
1888 tp->md5sig_info = NULL;
1892 /* Clean up a referenced TCP bind bucket. */
1893 if (inet_csk(sk)->icsk_bind_hash)
1896 BUG_ON(tp->fastopen_rsk);
1898 /* If socket is aborted during connect operation */
1899 tcp_free_fastopen_req(tp);
1900 tcp_fastopen_destroy_cipher(sk);
1901 tcp_saved_syn_free(tp);
1903 sk_sockets_allocated_dec(sk);
1905 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1907 #ifdef CONFIG_PROC_FS
1908 /* Proc filesystem TCP sock list dumping. */
1911 * Get next listener socket follow cur. If cur is NULL, get first socket
1912 * starting from bucket given in st->bucket; when st->bucket is zero the
1913 * very first socket in the hash table is returned.
1915 static void *listening_get_next(struct seq_file *seq, void *cur)
1917 struct tcp_iter_state *st = seq->private;
1918 struct net *net = seq_file_net(seq);
1919 struct inet_listen_hashbucket *ilb;
1920 struct sock *sk = cur;
1924 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1925 spin_lock(&ilb->lock);
1926 sk = sk_head(&ilb->head);
1930 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1936 sk_for_each_from(sk) {
1937 if (!net_eq(sock_net(sk), net))
1939 if (sk->sk_family == st->family)
1942 spin_unlock(&ilb->lock);
1944 if (++st->bucket < INET_LHTABLE_SIZE)
1949 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1951 struct tcp_iter_state *st = seq->private;
1956 rc = listening_get_next(seq, NULL);
1958 while (rc && *pos) {
1959 rc = listening_get_next(seq, rc);
1965 static inline bool empty_bucket(const struct tcp_iter_state *st)
1967 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1971 * Get first established socket starting from bucket given in st->bucket.
1972 * If st->bucket is zero, the very first socket in the hash is returned.
1974 static void *established_get_first(struct seq_file *seq)
1976 struct tcp_iter_state *st = seq->private;
1977 struct net *net = seq_file_net(seq);
1981 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1983 struct hlist_nulls_node *node;
1984 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1986 /* Lockless fast path for the common case of empty buckets */
1987 if (empty_bucket(st))
1991 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1992 if (sk->sk_family != st->family ||
1993 !net_eq(sock_net(sk), net)) {
1999 spin_unlock_bh(lock);
2005 static void *established_get_next(struct seq_file *seq, void *cur)
2007 struct sock *sk = cur;
2008 struct hlist_nulls_node *node;
2009 struct tcp_iter_state *st = seq->private;
2010 struct net *net = seq_file_net(seq);
2015 sk = sk_nulls_next(sk);
2017 sk_nulls_for_each_from(sk, node) {
2018 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2022 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2024 return established_get_first(seq);
2027 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2029 struct tcp_iter_state *st = seq->private;
2033 rc = established_get_first(seq);
2036 rc = established_get_next(seq, rc);
2042 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2045 struct tcp_iter_state *st = seq->private;
2047 st->state = TCP_SEQ_STATE_LISTENING;
2048 rc = listening_get_idx(seq, &pos);
2051 st->state = TCP_SEQ_STATE_ESTABLISHED;
2052 rc = established_get_idx(seq, pos);
2058 static void *tcp_seek_last_pos(struct seq_file *seq)
2060 struct tcp_iter_state *st = seq->private;
2061 int offset = st->offset;
2062 int orig_num = st->num;
2065 switch (st->state) {
2066 case TCP_SEQ_STATE_LISTENING:
2067 if (st->bucket >= INET_LHTABLE_SIZE)
2069 st->state = TCP_SEQ_STATE_LISTENING;
2070 rc = listening_get_next(seq, NULL);
2071 while (offset-- && rc)
2072 rc = listening_get_next(seq, rc);
2076 st->state = TCP_SEQ_STATE_ESTABLISHED;
2078 case TCP_SEQ_STATE_ESTABLISHED:
2079 if (st->bucket > tcp_hashinfo.ehash_mask)
2081 rc = established_get_first(seq);
2082 while (offset-- && rc)
2083 rc = established_get_next(seq, rc);
2091 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2093 struct tcp_iter_state *st = seq->private;
2096 if (*pos && *pos == st->last_pos) {
2097 rc = tcp_seek_last_pos(seq);
2102 st->state = TCP_SEQ_STATE_LISTENING;
2106 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2109 st->last_pos = *pos;
2113 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2115 struct tcp_iter_state *st = seq->private;
2118 if (v == SEQ_START_TOKEN) {
2119 rc = tcp_get_idx(seq, 0);
2123 switch (st->state) {
2124 case TCP_SEQ_STATE_LISTENING:
2125 rc = listening_get_next(seq, v);
2127 st->state = TCP_SEQ_STATE_ESTABLISHED;
2130 rc = established_get_first(seq);
2133 case TCP_SEQ_STATE_ESTABLISHED:
2134 rc = established_get_next(seq, v);
2139 st->last_pos = *pos;
2143 static void tcp_seq_stop(struct seq_file *seq, void *v)
2145 struct tcp_iter_state *st = seq->private;
2147 switch (st->state) {
2148 case TCP_SEQ_STATE_LISTENING:
2149 if (v != SEQ_START_TOKEN)
2150 spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2152 case TCP_SEQ_STATE_ESTABLISHED:
2154 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2159 int tcp_seq_open(struct inode *inode, struct file *file)
2161 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2162 struct tcp_iter_state *s;
2165 err = seq_open_net(inode, file, &afinfo->seq_ops,
2166 sizeof(struct tcp_iter_state));
2170 s = ((struct seq_file *)file->private_data)->private;
2171 s->family = afinfo->family;
2175 EXPORT_SYMBOL(tcp_seq_open);
2177 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2180 struct proc_dir_entry *p;
2182 afinfo->seq_ops.start = tcp_seq_start;
2183 afinfo->seq_ops.next = tcp_seq_next;
2184 afinfo->seq_ops.stop = tcp_seq_stop;
2186 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2187 afinfo->seq_fops, afinfo);
2192 EXPORT_SYMBOL(tcp_proc_register);
2194 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2196 remove_proc_entry(afinfo->name, net->proc_net);
2198 EXPORT_SYMBOL(tcp_proc_unregister);
2200 static void get_openreq4(const struct request_sock *req,
2201 struct seq_file *f, int i)
2203 const struct inet_request_sock *ireq = inet_rsk(req);
2204 long delta = req->rsk_timer.expires - jiffies;
2206 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2207 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2212 ntohs(ireq->ir_rmt_port),
2214 0, 0, /* could print option size, but that is af dependent. */
2215 1, /* timers active (only the expire timer) */
2216 jiffies_delta_to_clock_t(delta),
2218 from_kuid_munged(seq_user_ns(f),
2219 sock_i_uid(req->rsk_listener)),
2220 0, /* non standard timer */
2221 0, /* open_requests have no inode */
2226 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2229 unsigned long timer_expires;
2230 const struct tcp_sock *tp = tcp_sk(sk);
2231 const struct inet_connection_sock *icsk = inet_csk(sk);
2232 const struct inet_sock *inet = inet_sk(sk);
2233 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2234 __be32 dest = inet->inet_daddr;
2235 __be32 src = inet->inet_rcv_saddr;
2236 __u16 destp = ntohs(inet->inet_dport);
2237 __u16 srcp = ntohs(inet->inet_sport);
2241 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2242 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2243 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2245 timer_expires = icsk->icsk_timeout;
2246 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2248 timer_expires = icsk->icsk_timeout;
2249 } else if (timer_pending(&sk->sk_timer)) {
2251 timer_expires = sk->sk_timer.expires;
2254 timer_expires = jiffies;
2257 state = sk_state_load(sk);
2258 if (state == TCP_LISTEN)
2259 rx_queue = sk->sk_ack_backlog;
2261 /* Because we don't lock the socket,
2262 * we might find a transient negative value.
2264 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2266 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2267 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2268 i, src, srcp, dest, destp, state,
2269 tp->write_seq - tp->snd_una,
2272 jiffies_delta_to_clock_t(timer_expires - jiffies),
2273 icsk->icsk_retransmits,
2274 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2275 icsk->icsk_probes_out,
2277 refcount_read(&sk->sk_refcnt), sk,
2278 jiffies_to_clock_t(icsk->icsk_rto),
2279 jiffies_to_clock_t(icsk->icsk_ack.ato),
2280 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2282 state == TCP_LISTEN ?
2283 fastopenq->max_qlen :
2284 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2287 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2288 struct seq_file *f, int i)
2290 long delta = tw->tw_timer.expires - jiffies;
2294 dest = tw->tw_daddr;
2295 src = tw->tw_rcv_saddr;
2296 destp = ntohs(tw->tw_dport);
2297 srcp = ntohs(tw->tw_sport);
2299 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2300 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2301 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2302 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2303 refcount_read(&tw->tw_refcnt), tw);
2308 static int tcp4_seq_show(struct seq_file *seq, void *v)
2310 struct tcp_iter_state *st;
2311 struct sock *sk = v;
2313 seq_setwidth(seq, TMPSZ - 1);
2314 if (v == SEQ_START_TOKEN) {
2315 seq_puts(seq, " sl local_address rem_address st tx_queue "
2316 "rx_queue tr tm->when retrnsmt uid timeout "
2322 if (sk->sk_state == TCP_TIME_WAIT)
2323 get_timewait4_sock(v, seq, st->num);
2324 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2325 get_openreq4(v, seq, st->num);
2327 get_tcp4_sock(v, seq, st->num);
2333 static const struct file_operations tcp_afinfo_seq_fops = {
2334 .owner = THIS_MODULE,
2335 .open = tcp_seq_open,
2337 .llseek = seq_lseek,
2338 .release = seq_release_net
2341 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2344 .seq_fops = &tcp_afinfo_seq_fops,
2346 .show = tcp4_seq_show,
2350 static int __net_init tcp4_proc_init_net(struct net *net)
2352 return tcp_proc_register(net, &tcp4_seq_afinfo);
2355 static void __net_exit tcp4_proc_exit_net(struct net *net)
2357 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2360 static struct pernet_operations tcp4_net_ops = {
2361 .init = tcp4_proc_init_net,
2362 .exit = tcp4_proc_exit_net,
2365 int __init tcp4_proc_init(void)
2367 return register_pernet_subsys(&tcp4_net_ops);
2370 void tcp4_proc_exit(void)
2372 unregister_pernet_subsys(&tcp4_net_ops);
2374 #endif /* CONFIG_PROC_FS */
2376 struct proto tcp_prot = {
2378 .owner = THIS_MODULE,
2380 .connect = tcp_v4_connect,
2381 .disconnect = tcp_disconnect,
2382 .accept = inet_csk_accept,
2384 .init = tcp_v4_init_sock,
2385 .destroy = tcp_v4_destroy_sock,
2386 .shutdown = tcp_shutdown,
2387 .setsockopt = tcp_setsockopt,
2388 .getsockopt = tcp_getsockopt,
2389 .keepalive = tcp_set_keepalive,
2390 .recvmsg = tcp_recvmsg,
2391 .sendmsg = tcp_sendmsg,
2392 .sendpage = tcp_sendpage,
2393 .backlog_rcv = tcp_v4_do_rcv,
2394 .release_cb = tcp_release_cb,
2396 .unhash = inet_unhash,
2397 .get_port = inet_csk_get_port,
2398 .enter_memory_pressure = tcp_enter_memory_pressure,
2399 .leave_memory_pressure = tcp_leave_memory_pressure,
2400 .stream_memory_free = tcp_stream_memory_free,
2401 .sockets_allocated = &tcp_sockets_allocated,
2402 .orphan_count = &tcp_orphan_count,
2403 .memory_allocated = &tcp_memory_allocated,
2404 .memory_pressure = &tcp_memory_pressure,
2405 .sysctl_mem = sysctl_tcp_mem,
2406 .sysctl_wmem = sysctl_tcp_wmem,
2407 .sysctl_rmem = sysctl_tcp_rmem,
2408 .max_header = MAX_TCP_HEADER,
2409 .obj_size = sizeof(struct tcp_sock),
2410 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2411 .twsk_prot = &tcp_timewait_sock_ops,
2412 .rsk_prot = &tcp_request_sock_ops,
2413 .h.hashinfo = &tcp_hashinfo,
2414 .no_autobind = true,
2415 #ifdef CONFIG_COMPAT
2416 .compat_setsockopt = compat_tcp_setsockopt,
2417 .compat_getsockopt = compat_tcp_getsockopt,
2419 .diag_destroy = tcp_abort,
2421 EXPORT_SYMBOL(tcp_prot);
2423 static void __net_exit tcp_sk_exit(struct net *net)
2427 for_each_possible_cpu(cpu)
2428 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2429 free_percpu(net->ipv4.tcp_sk);
2432 static int __net_init tcp_sk_init(struct net *net)
2436 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2437 if (!net->ipv4.tcp_sk)
2440 for_each_possible_cpu(cpu) {
2443 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2447 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2448 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2451 net->ipv4.sysctl_tcp_ecn = 2;
2452 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2454 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2455 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2456 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2458 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2459 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2460 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2462 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2463 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2464 net->ipv4.sysctl_tcp_syncookies = 1;
2465 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2466 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2467 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2468 net->ipv4.sysctl_tcp_orphan_retries = 0;
2469 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2470 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2471 net->ipv4.sysctl_tcp_tw_reuse = 0;
2473 cnt = tcp_hashinfo.ehash_mask + 1;
2474 net->ipv4.tcp_death_row.sysctl_max_tw_buckets = (cnt + 1) / 2;
2475 net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo;
2477 net->ipv4.sysctl_max_syn_backlog = max(128, cnt / 256);
2478 net->ipv4.sysctl_tcp_sack = 1;
2479 net->ipv4.sysctl_tcp_window_scaling = 1;
2480 net->ipv4.sysctl_tcp_timestamps = 1;
2482 net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
2483 spin_lock_init(&net->ipv4.tcp_fastopen_ctx_lock);
2484 net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 60 * 60;
2485 atomic_set(&net->ipv4.tfo_active_disable_times, 0);
2494 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2498 inet_twsk_purge(&tcp_hashinfo, AF_INET);
2500 list_for_each_entry(net, net_exit_list, exit_list)
2501 tcp_fastopen_ctx_destroy(net);
2504 static struct pernet_operations __net_initdata tcp_sk_ops = {
2505 .init = tcp_sk_init,
2506 .exit = tcp_sk_exit,
2507 .exit_batch = tcp_sk_exit_batch,
2510 void __init tcp_v4_init(void)
2512 if (register_pernet_subsys(&tcp_sk_ops))
2513 panic("Failed to create the TCP control socket.\n");