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, 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 struct tcp_sock *tp = tcp_sk(sk);
902 struct tcp_md5sig_key *key;
903 unsigned int size = sizeof(struct in_addr);
904 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 struct sock *addr_sk)
929 union tcp_md5_addr *addr;
931 addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
932 return tcp_md5_do_lookup(sk, addr, AF_INET);
934 EXPORT_SYMBOL(tcp_v4_md5_lookup);
936 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
937 struct request_sock *req)
939 union tcp_md5_addr *addr;
941 addr = (union tcp_md5_addr *)&inet_rsk(req)->ir_rmt_addr;
942 return tcp_md5_do_lookup(sk, addr, AF_INET);
945 /* This can be called on a newly created socket, from other files */
946 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
947 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
949 /* Add Key to the list */
950 struct tcp_md5sig_key *key;
951 struct tcp_sock *tp = tcp_sk(sk);
952 struct tcp_md5sig_info *md5sig;
954 key = tcp_md5_do_lookup(sk, addr, family);
956 /* Pre-existing entry - just update that one. */
957 memcpy(key->key, newkey, newkeylen);
958 key->keylen = newkeylen;
962 md5sig = rcu_dereference_protected(tp->md5sig_info,
963 sock_owned_by_user(sk));
965 md5sig = kmalloc(sizeof(*md5sig), gfp);
969 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
970 INIT_HLIST_HEAD(&md5sig->head);
971 rcu_assign_pointer(tp->md5sig_info, md5sig);
974 key = sock_kmalloc(sk, sizeof(*key), gfp);
977 if (!tcp_alloc_md5sig_pool()) {
978 sock_kfree_s(sk, key, sizeof(*key));
982 memcpy(key->key, newkey, newkeylen);
983 key->keylen = newkeylen;
984 key->family = family;
985 memcpy(&key->addr, addr,
986 (family == AF_INET6) ? sizeof(struct in6_addr) :
987 sizeof(struct in_addr));
988 hlist_add_head_rcu(&key->node, &md5sig->head);
991 EXPORT_SYMBOL(tcp_md5_do_add);
993 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
995 struct tcp_md5sig_key *key;
997 key = tcp_md5_do_lookup(sk, addr, family);
1000 hlist_del_rcu(&key->node);
1001 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1002 kfree_rcu(key, rcu);
1005 EXPORT_SYMBOL(tcp_md5_do_del);
1007 static void tcp_clear_md5_list(struct sock *sk)
1009 struct tcp_sock *tp = tcp_sk(sk);
1010 struct tcp_md5sig_key *key;
1011 struct hlist_node *n;
1012 struct tcp_md5sig_info *md5sig;
1014 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1016 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1017 hlist_del_rcu(&key->node);
1018 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1019 kfree_rcu(key, rcu);
1023 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1026 struct tcp_md5sig cmd;
1027 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1029 if (optlen < sizeof(cmd))
1032 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1035 if (sin->sin_family != AF_INET)
1038 if (!cmd.tcpm_keylen)
1039 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1042 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1045 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1046 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1050 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1051 __be32 daddr, __be32 saddr, int nbytes)
1053 struct tcp4_pseudohdr *bp;
1054 struct scatterlist sg;
1056 bp = &hp->md5_blk.ip4;
1059 * 1. the TCP pseudo-header (in the order: source IP address,
1060 * destination IP address, zero-padded protocol number, and
1066 bp->protocol = IPPROTO_TCP;
1067 bp->len = cpu_to_be16(nbytes);
1069 sg_init_one(&sg, bp, sizeof(*bp));
1070 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1073 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1074 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1076 struct tcp_md5sig_pool *hp;
1077 struct hash_desc *desc;
1079 hp = tcp_get_md5sig_pool();
1081 goto clear_hash_noput;
1082 desc = &hp->md5_desc;
1084 if (crypto_hash_init(desc))
1086 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1088 if (tcp_md5_hash_header(hp, th))
1090 if (tcp_md5_hash_key(hp, key))
1092 if (crypto_hash_final(desc, md5_hash))
1095 tcp_put_md5sig_pool();
1099 tcp_put_md5sig_pool();
1101 memset(md5_hash, 0, 16);
1105 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1106 const struct sock *sk, const struct request_sock *req,
1107 const struct sk_buff *skb)
1109 struct tcp_md5sig_pool *hp;
1110 struct hash_desc *desc;
1111 const struct tcphdr *th = tcp_hdr(skb);
1112 __be32 saddr, daddr;
1115 saddr = inet_sk(sk)->inet_saddr;
1116 daddr = inet_sk(sk)->inet_daddr;
1118 saddr = inet_rsk(req)->ir_loc_addr;
1119 daddr = inet_rsk(req)->ir_rmt_addr;
1121 const struct iphdr *iph = ip_hdr(skb);
1126 hp = tcp_get_md5sig_pool();
1128 goto clear_hash_noput;
1129 desc = &hp->md5_desc;
1131 if (crypto_hash_init(desc))
1134 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1136 if (tcp_md5_hash_header(hp, th))
1138 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1140 if (tcp_md5_hash_key(hp, key))
1142 if (crypto_hash_final(desc, md5_hash))
1145 tcp_put_md5sig_pool();
1149 tcp_put_md5sig_pool();
1151 memset(md5_hash, 0, 16);
1154 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1156 /* Called with rcu_read_lock() */
1157 static bool tcp_v4_inbound_md5_hash(struct sock *sk,
1158 const struct sk_buff *skb)
1161 * This gets called for each TCP segment that arrives
1162 * so we want to be efficient.
1163 * We have 3 drop cases:
1164 * o No MD5 hash and one expected.
1165 * o MD5 hash and we're not expecting one.
1166 * o MD5 hash and its wrong.
1168 const __u8 *hash_location = NULL;
1169 struct tcp_md5sig_key *hash_expected;
1170 const struct iphdr *iph = ip_hdr(skb);
1171 const struct tcphdr *th = tcp_hdr(skb);
1173 unsigned char newhash[16];
1175 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1177 hash_location = tcp_parse_md5sig_option(th);
1179 /* We've parsed the options - do we have a hash? */
1180 if (!hash_expected && !hash_location)
1183 if (hash_expected && !hash_location) {
1184 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1188 if (!hash_expected && hash_location) {
1189 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1193 /* Okay, so this is hash_expected and hash_location -
1194 * so we need to calculate the checksum.
1196 genhash = tcp_v4_md5_hash_skb(newhash,
1200 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1201 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1202 &iph->saddr, ntohs(th->source),
1203 &iph->daddr, ntohs(th->dest),
1204 genhash ? " tcp_v4_calc_md5_hash failed"
1212 static void tcp_v4_init_req(struct request_sock *req, struct sock *sk_listener,
1213 struct sk_buff *skb)
1215 struct inet_request_sock *ireq = inet_rsk(req);
1217 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1218 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1219 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
1220 ireq->opt = tcp_v4_save_options(skb);
1221 ireq->ireq_family = AF_INET;
1224 static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl,
1225 const struct request_sock *req,
1228 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1231 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1240 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1242 .obj_size = sizeof(struct tcp_request_sock),
1243 .rtx_syn_ack = tcp_rtx_synack,
1244 .send_ack = tcp_v4_reqsk_send_ack,
1245 .destructor = tcp_v4_reqsk_destructor,
1246 .send_reset = tcp_v4_send_reset,
1247 .syn_ack_timeout = tcp_syn_ack_timeout,
1250 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1251 .mss_clamp = TCP_MSS_DEFAULT,
1252 #ifdef CONFIG_TCP_MD5SIG
1253 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1254 .calc_md5_hash = tcp_v4_md5_hash_skb,
1256 .init_req = tcp_v4_init_req,
1257 #ifdef CONFIG_SYN_COOKIES
1258 .cookie_init_seq = cookie_v4_init_sequence,
1260 .route_req = tcp_v4_route_req,
1261 .init_seq = tcp_v4_init_sequence,
1262 .send_synack = tcp_v4_send_synack,
1263 .queue_hash_add = inet_csk_reqsk_queue_hash_add,
1266 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1268 /* Never answer to SYNs send to broadcast or multicast */
1269 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1272 return tcp_conn_request(&tcp_request_sock_ops,
1273 &tcp_request_sock_ipv4_ops, sk, skb);
1276 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1279 EXPORT_SYMBOL(tcp_v4_conn_request);
1283 * The three way handshake has completed - we got a valid synack -
1284 * now create the new socket.
1286 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1287 struct request_sock *req,
1288 struct dst_entry *dst)
1290 struct inet_request_sock *ireq;
1291 struct inet_sock *newinet;
1292 struct tcp_sock *newtp;
1294 #ifdef CONFIG_TCP_MD5SIG
1295 struct tcp_md5sig_key *key;
1297 struct ip_options_rcu *inet_opt;
1299 if (sk_acceptq_is_full(sk))
1302 newsk = tcp_create_openreq_child(sk, req, skb);
1306 newsk->sk_gso_type = SKB_GSO_TCPV4;
1307 inet_sk_rx_dst_set(newsk, skb);
1309 newtp = tcp_sk(newsk);
1310 newinet = inet_sk(newsk);
1311 ireq = inet_rsk(req);
1312 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1313 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1314 newinet->inet_saddr = ireq->ir_loc_addr;
1315 inet_opt = ireq->opt;
1316 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1318 newinet->mc_index = inet_iif(skb);
1319 newinet->mc_ttl = ip_hdr(skb)->ttl;
1320 newinet->rcv_tos = ip_hdr(skb)->tos;
1321 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1322 inet_set_txhash(newsk);
1324 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1325 newinet->inet_id = newtp->write_seq ^ jiffies;
1328 dst = inet_csk_route_child_sock(sk, newsk, req);
1332 /* syncookie case : see end of cookie_v4_check() */
1334 sk_setup_caps(newsk, dst);
1336 tcp_ca_openreq_child(newsk, dst);
1338 tcp_sync_mss(newsk, dst_mtu(dst));
1339 newtp->advmss = dst_metric_advmss(dst);
1340 if (tcp_sk(sk)->rx_opt.user_mss &&
1341 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1342 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1344 tcp_initialize_rcv_mss(newsk);
1346 #ifdef CONFIG_TCP_MD5SIG
1347 /* Copy over the MD5 key from the original socket */
1348 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1352 * We're using one, so create a matching key
1353 * on the newsk structure. If we fail to get
1354 * memory, then we end up not copying the key
1357 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1358 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1359 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1363 if (__inet_inherit_port(sk, newsk) < 0)
1365 __inet_hash_nolisten(newsk, NULL);
1370 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1374 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1377 inet_csk_prepare_forced_close(newsk);
1381 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1383 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1385 const struct tcphdr *th = tcp_hdr(skb);
1386 const struct iphdr *iph = ip_hdr(skb);
1387 struct request_sock *req;
1390 req = inet_csk_search_req(sk, th->source, iph->saddr, iph->daddr);
1392 nsk = tcp_check_req(sk, skb, req, false);
1397 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1398 th->source, iph->daddr, th->dest, inet_iif(skb));
1401 if (nsk->sk_state != TCP_TIME_WAIT) {
1405 inet_twsk_put(inet_twsk(nsk));
1409 #ifdef CONFIG_SYN_COOKIES
1411 sk = cookie_v4_check(sk, skb);
1416 /* The socket must have it's spinlock held when we get
1419 * We have a potential double-lock case here, so even when
1420 * doing backlog processing we use the BH locking scheme.
1421 * This is because we cannot sleep with the original spinlock
1424 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1428 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1429 struct dst_entry *dst = sk->sk_rx_dst;
1431 sock_rps_save_rxhash(sk, skb);
1432 sk_mark_napi_id(sk, skb);
1434 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1435 dst->ops->check(dst, 0) == NULL) {
1437 sk->sk_rx_dst = NULL;
1440 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1444 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1447 if (sk->sk_state == TCP_LISTEN) {
1448 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1453 sock_rps_save_rxhash(nsk, skb);
1454 sk_mark_napi_id(sk, skb);
1455 if (tcp_child_process(sk, nsk, skb)) {
1462 sock_rps_save_rxhash(sk, skb);
1464 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1471 tcp_v4_send_reset(rsk, skb);
1474 /* Be careful here. If this function gets more complicated and
1475 * gcc suffers from register pressure on the x86, sk (in %ebx)
1476 * might be destroyed here. This current version compiles correctly,
1477 * but you have been warned.
1482 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1483 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1486 EXPORT_SYMBOL(tcp_v4_do_rcv);
1488 void tcp_v4_early_demux(struct sk_buff *skb)
1490 const struct iphdr *iph;
1491 const struct tcphdr *th;
1494 if (skb->pkt_type != PACKET_HOST)
1497 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1503 if (th->doff < sizeof(struct tcphdr) / 4)
1506 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1507 iph->saddr, th->source,
1508 iph->daddr, ntohs(th->dest),
1512 skb->destructor = sock_edemux;
1513 if (sk_fullsock(sk)) {
1514 struct dst_entry *dst = sk->sk_rx_dst;
1517 dst = dst_check(dst, 0);
1519 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1520 skb_dst_set_noref(skb, dst);
1525 /* Packet is added to VJ-style prequeue for processing in process
1526 * context, if a reader task is waiting. Apparently, this exciting
1527 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1528 * failed somewhere. Latency? Burstiness? Well, at least now we will
1529 * see, why it failed. 8)8) --ANK
1532 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1534 struct tcp_sock *tp = tcp_sk(sk);
1536 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1539 if (skb->len <= tcp_hdrlen(skb) &&
1540 skb_queue_len(&tp->ucopy.prequeue) == 0)
1543 /* Before escaping RCU protected region, we need to take care of skb
1544 * dst. Prequeue is only enabled for established sockets.
1545 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1546 * Instead of doing full sk_rx_dst validity here, let's perform
1547 * an optimistic check.
1549 if (likely(sk->sk_rx_dst))
1554 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1555 tp->ucopy.memory += skb->truesize;
1556 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1557 struct sk_buff *skb1;
1559 BUG_ON(sock_owned_by_user(sk));
1561 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1562 sk_backlog_rcv(sk, skb1);
1563 NET_INC_STATS_BH(sock_net(sk),
1564 LINUX_MIB_TCPPREQUEUEDROPPED);
1567 tp->ucopy.memory = 0;
1568 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1569 wake_up_interruptible_sync_poll(sk_sleep(sk),
1570 POLLIN | POLLRDNORM | POLLRDBAND);
1571 if (!inet_csk_ack_scheduled(sk))
1572 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1573 (3 * tcp_rto_min(sk)) / 4,
1578 EXPORT_SYMBOL(tcp_prequeue);
1584 int tcp_v4_rcv(struct sk_buff *skb)
1586 const struct iphdr *iph;
1587 const struct tcphdr *th;
1590 struct net *net = dev_net(skb->dev);
1592 if (skb->pkt_type != PACKET_HOST)
1595 /* Count it even if it's bad */
1596 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1598 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1603 if (th->doff < sizeof(struct tcphdr) / 4)
1605 if (!pskb_may_pull(skb, th->doff * 4))
1608 /* An explanation is required here, I think.
1609 * Packet length and doff are validated by header prediction,
1610 * provided case of th->doff==0 is eliminated.
1611 * So, we defer the checks. */
1613 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1618 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1619 * barrier() makes sure compiler wont play fool^Waliasing games.
1621 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1622 sizeof(struct inet_skb_parm));
1625 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1626 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1627 skb->len - th->doff * 4);
1628 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1629 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1630 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1631 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1632 TCP_SKB_CB(skb)->sacked = 0;
1634 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1639 if (sk->sk_state == TCP_TIME_WAIT)
1642 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1643 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1644 goto discard_and_relse;
1647 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1648 goto discard_and_relse;
1650 #ifdef CONFIG_TCP_MD5SIG
1652 * We really want to reject the packet as early as possible
1654 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1655 * o There is an MD5 option and we're not expecting one
1657 if (tcp_v4_inbound_md5_hash(sk, skb))
1658 goto discard_and_relse;
1663 if (sk_filter(sk, skb))
1664 goto discard_and_relse;
1666 sk_incoming_cpu_update(sk);
1669 bh_lock_sock_nested(sk);
1671 if (!sock_owned_by_user(sk)) {
1672 if (!tcp_prequeue(sk, skb))
1673 ret = tcp_v4_do_rcv(sk, skb);
1674 } else if (unlikely(sk_add_backlog(sk, skb,
1675 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1677 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1678 goto discard_and_relse;
1687 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1690 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1692 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1694 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1696 tcp_v4_send_reset(NULL, skb);
1700 /* Discard frame. */
1709 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1710 inet_twsk_put(inet_twsk(sk));
1714 if (skb->len < (th->doff << 2)) {
1715 inet_twsk_put(inet_twsk(sk));
1718 if (tcp_checksum_complete(skb)) {
1719 inet_twsk_put(inet_twsk(sk));
1722 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1724 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1726 iph->saddr, th->source,
1727 iph->daddr, th->dest,
1730 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1731 inet_twsk_put(inet_twsk(sk));
1735 /* Fall through to ACK */
1738 tcp_v4_timewait_ack(sk, skb);
1742 case TCP_TW_SUCCESS:;
1747 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1748 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1749 .twsk_unique = tcp_twsk_unique,
1750 .twsk_destructor= tcp_twsk_destructor,
1753 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1755 struct dst_entry *dst = skb_dst(skb);
1759 sk->sk_rx_dst = dst;
1760 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1763 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1765 const struct inet_connection_sock_af_ops ipv4_specific = {
1766 .queue_xmit = ip_queue_xmit,
1767 .send_check = tcp_v4_send_check,
1768 .rebuild_header = inet_sk_rebuild_header,
1769 .sk_rx_dst_set = inet_sk_rx_dst_set,
1770 .conn_request = tcp_v4_conn_request,
1771 .syn_recv_sock = tcp_v4_syn_recv_sock,
1772 .net_header_len = sizeof(struct iphdr),
1773 .setsockopt = ip_setsockopt,
1774 .getsockopt = ip_getsockopt,
1775 .addr2sockaddr = inet_csk_addr2sockaddr,
1776 .sockaddr_len = sizeof(struct sockaddr_in),
1777 .bind_conflict = inet_csk_bind_conflict,
1778 #ifdef CONFIG_COMPAT
1779 .compat_setsockopt = compat_ip_setsockopt,
1780 .compat_getsockopt = compat_ip_getsockopt,
1782 .mtu_reduced = tcp_v4_mtu_reduced,
1784 EXPORT_SYMBOL(ipv4_specific);
1786 #ifdef CONFIG_TCP_MD5SIG
1787 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1788 .md5_lookup = tcp_v4_md5_lookup,
1789 .calc_md5_hash = tcp_v4_md5_hash_skb,
1790 .md5_parse = tcp_v4_parse_md5_keys,
1794 /* NOTE: A lot of things set to zero explicitly by call to
1795 * sk_alloc() so need not be done here.
1797 static int tcp_v4_init_sock(struct sock *sk)
1799 struct inet_connection_sock *icsk = inet_csk(sk);
1803 icsk->icsk_af_ops = &ipv4_specific;
1805 #ifdef CONFIG_TCP_MD5SIG
1806 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1812 void tcp_v4_destroy_sock(struct sock *sk)
1814 struct tcp_sock *tp = tcp_sk(sk);
1816 tcp_clear_xmit_timers(sk);
1818 tcp_cleanup_congestion_control(sk);
1820 /* Cleanup up the write buffer. */
1821 tcp_write_queue_purge(sk);
1823 /* Cleans up our, hopefully empty, out_of_order_queue. */
1824 __skb_queue_purge(&tp->out_of_order_queue);
1826 #ifdef CONFIG_TCP_MD5SIG
1827 /* Clean up the MD5 key list, if any */
1828 if (tp->md5sig_info) {
1829 tcp_clear_md5_list(sk);
1830 kfree_rcu(tp->md5sig_info, rcu);
1831 tp->md5sig_info = NULL;
1835 /* Clean prequeue, it must be empty really */
1836 __skb_queue_purge(&tp->ucopy.prequeue);
1838 /* Clean up a referenced TCP bind bucket. */
1839 if (inet_csk(sk)->icsk_bind_hash)
1842 BUG_ON(tp->fastopen_rsk != NULL);
1844 /* If socket is aborted during connect operation */
1845 tcp_free_fastopen_req(tp);
1847 sk_sockets_allocated_dec(sk);
1848 sock_release_memcg(sk);
1850 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1852 #ifdef CONFIG_PROC_FS
1853 /* Proc filesystem TCP sock list dumping. */
1856 * Get next listener socket follow cur. If cur is NULL, get first socket
1857 * starting from bucket given in st->bucket; when st->bucket is zero the
1858 * very first socket in the hash table is returned.
1860 static void *listening_get_next(struct seq_file *seq, void *cur)
1862 struct inet_connection_sock *icsk;
1863 struct hlist_nulls_node *node;
1864 struct sock *sk = cur;
1865 struct inet_listen_hashbucket *ilb;
1866 struct tcp_iter_state *st = seq->private;
1867 struct net *net = seq_file_net(seq);
1870 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1871 spin_lock_bh(&ilb->lock);
1872 sk = sk_nulls_head(&ilb->head);
1876 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1880 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1881 struct request_sock *req = cur;
1883 icsk = inet_csk(st->syn_wait_sk);
1887 if (req->rsk_ops->family == st->family) {
1893 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1896 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1898 sk = sk_nulls_next(st->syn_wait_sk);
1899 st->state = TCP_SEQ_STATE_LISTENING;
1900 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1902 icsk = inet_csk(sk);
1903 spin_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1904 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1906 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1907 sk = sk_nulls_next(sk);
1910 sk_nulls_for_each_from(sk, node) {
1911 if (!net_eq(sock_net(sk), net))
1913 if (sk->sk_family == st->family) {
1917 icsk = inet_csk(sk);
1918 spin_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1919 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1921 st->uid = sock_i_uid(sk);
1922 st->syn_wait_sk = sk;
1923 st->state = TCP_SEQ_STATE_OPENREQ;
1927 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1929 spin_unlock_bh(&ilb->lock);
1931 if (++st->bucket < INET_LHTABLE_SIZE) {
1932 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1933 spin_lock_bh(&ilb->lock);
1934 sk = sk_nulls_head(&ilb->head);
1942 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1944 struct tcp_iter_state *st = seq->private;
1949 rc = listening_get_next(seq, NULL);
1951 while (rc && *pos) {
1952 rc = listening_get_next(seq, rc);
1958 static inline bool empty_bucket(const struct tcp_iter_state *st)
1960 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1964 * Get first established socket starting from bucket given in st->bucket.
1965 * If st->bucket is zero, the very first socket in the hash is returned.
1967 static void *established_get_first(struct seq_file *seq)
1969 struct tcp_iter_state *st = seq->private;
1970 struct net *net = seq_file_net(seq);
1974 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1976 struct hlist_nulls_node *node;
1977 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1979 /* Lockless fast path for the common case of empty buckets */
1980 if (empty_bucket(st))
1984 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1985 if (sk->sk_family != st->family ||
1986 !net_eq(sock_net(sk), net)) {
1992 spin_unlock_bh(lock);
1998 static void *established_get_next(struct seq_file *seq, void *cur)
2000 struct sock *sk = cur;
2001 struct hlist_nulls_node *node;
2002 struct tcp_iter_state *st = seq->private;
2003 struct net *net = seq_file_net(seq);
2008 sk = sk_nulls_next(sk);
2010 sk_nulls_for_each_from(sk, node) {
2011 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2015 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2017 return established_get_first(seq);
2020 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2022 struct tcp_iter_state *st = seq->private;
2026 rc = established_get_first(seq);
2029 rc = established_get_next(seq, rc);
2035 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2038 struct tcp_iter_state *st = seq->private;
2040 st->state = TCP_SEQ_STATE_LISTENING;
2041 rc = listening_get_idx(seq, &pos);
2044 st->state = TCP_SEQ_STATE_ESTABLISHED;
2045 rc = established_get_idx(seq, pos);
2051 static void *tcp_seek_last_pos(struct seq_file *seq)
2053 struct tcp_iter_state *st = seq->private;
2054 int offset = st->offset;
2055 int orig_num = st->num;
2058 switch (st->state) {
2059 case TCP_SEQ_STATE_OPENREQ:
2060 case TCP_SEQ_STATE_LISTENING:
2061 if (st->bucket >= INET_LHTABLE_SIZE)
2063 st->state = TCP_SEQ_STATE_LISTENING;
2064 rc = listening_get_next(seq, NULL);
2065 while (offset-- && rc)
2066 rc = listening_get_next(seq, rc);
2070 st->state = TCP_SEQ_STATE_ESTABLISHED;
2072 case TCP_SEQ_STATE_ESTABLISHED:
2073 if (st->bucket > tcp_hashinfo.ehash_mask)
2075 rc = established_get_first(seq);
2076 while (offset-- && rc)
2077 rc = established_get_next(seq, rc);
2085 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2087 struct tcp_iter_state *st = seq->private;
2090 if (*pos && *pos == st->last_pos) {
2091 rc = tcp_seek_last_pos(seq);
2096 st->state = TCP_SEQ_STATE_LISTENING;
2100 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2103 st->last_pos = *pos;
2107 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2109 struct tcp_iter_state *st = seq->private;
2112 if (v == SEQ_START_TOKEN) {
2113 rc = tcp_get_idx(seq, 0);
2117 switch (st->state) {
2118 case TCP_SEQ_STATE_OPENREQ:
2119 case TCP_SEQ_STATE_LISTENING:
2120 rc = listening_get_next(seq, v);
2122 st->state = TCP_SEQ_STATE_ESTABLISHED;
2125 rc = established_get_first(seq);
2128 case TCP_SEQ_STATE_ESTABLISHED:
2129 rc = established_get_next(seq, v);
2134 st->last_pos = *pos;
2138 static void tcp_seq_stop(struct seq_file *seq, void *v)
2140 struct tcp_iter_state *st = seq->private;
2142 switch (st->state) {
2143 case TCP_SEQ_STATE_OPENREQ:
2145 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2146 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2148 case TCP_SEQ_STATE_LISTENING:
2149 if (v != SEQ_START_TOKEN)
2150 spin_unlock_bh(&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, kuid_t uid)
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), uid),
2219 0, /* non standard timer */
2220 0, /* open_requests have no inode */
2225 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2228 unsigned long timer_expires;
2229 const struct tcp_sock *tp = tcp_sk(sk);
2230 const struct inet_connection_sock *icsk = inet_csk(sk);
2231 const struct inet_sock *inet = inet_sk(sk);
2232 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2233 __be32 dest = inet->inet_daddr;
2234 __be32 src = inet->inet_rcv_saddr;
2235 __u16 destp = ntohs(inet->inet_dport);
2236 __u16 srcp = ntohs(inet->inet_sport);
2239 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2240 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2241 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2243 timer_expires = icsk->icsk_timeout;
2244 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2246 timer_expires = icsk->icsk_timeout;
2247 } else if (timer_pending(&sk->sk_timer)) {
2249 timer_expires = sk->sk_timer.expires;
2252 timer_expires = jiffies;
2255 if (sk->sk_state == TCP_LISTEN)
2256 rx_queue = sk->sk_ack_backlog;
2259 * because we dont lock socket, we might find a transient negative value
2261 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2263 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2264 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2265 i, src, srcp, dest, destp, sk->sk_state,
2266 tp->write_seq - tp->snd_una,
2269 jiffies_delta_to_clock_t(timer_expires - jiffies),
2270 icsk->icsk_retransmits,
2271 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2272 icsk->icsk_probes_out,
2274 atomic_read(&sk->sk_refcnt), sk,
2275 jiffies_to_clock_t(icsk->icsk_rto),
2276 jiffies_to_clock_t(icsk->icsk_ack.ato),
2277 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2279 sk->sk_state == TCP_LISTEN ?
2280 (fastopenq ? fastopenq->max_qlen : 0) :
2281 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2284 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2285 struct seq_file *f, int i)
2289 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
2291 dest = tw->tw_daddr;
2292 src = tw->tw_rcv_saddr;
2293 destp = ntohs(tw->tw_dport);
2294 srcp = ntohs(tw->tw_sport);
2296 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2297 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2298 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2299 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2300 atomic_read(&tw->tw_refcnt), tw);
2305 static int tcp4_seq_show(struct seq_file *seq, void *v)
2307 struct tcp_iter_state *st;
2308 struct sock *sk = v;
2310 seq_setwidth(seq, TMPSZ - 1);
2311 if (v == SEQ_START_TOKEN) {
2312 seq_puts(seq, " sl local_address rem_address st tx_queue "
2313 "rx_queue tr tm->when retrnsmt uid timeout "
2319 switch (st->state) {
2320 case TCP_SEQ_STATE_LISTENING:
2321 case TCP_SEQ_STATE_ESTABLISHED:
2322 if (sk->sk_state == TCP_TIME_WAIT)
2323 get_timewait4_sock(v, seq, st->num);
2325 get_tcp4_sock(v, seq, st->num);
2327 case TCP_SEQ_STATE_OPENREQ:
2328 get_openreq4(v, seq, st->num, st->uid);
2336 static const struct file_operations tcp_afinfo_seq_fops = {
2337 .owner = THIS_MODULE,
2338 .open = tcp_seq_open,
2340 .llseek = seq_lseek,
2341 .release = seq_release_net
2344 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2347 .seq_fops = &tcp_afinfo_seq_fops,
2349 .show = tcp4_seq_show,
2353 static int __net_init tcp4_proc_init_net(struct net *net)
2355 return tcp_proc_register(net, &tcp4_seq_afinfo);
2358 static void __net_exit tcp4_proc_exit_net(struct net *net)
2360 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2363 static struct pernet_operations tcp4_net_ops = {
2364 .init = tcp4_proc_init_net,
2365 .exit = tcp4_proc_exit_net,
2368 int __init tcp4_proc_init(void)
2370 return register_pernet_subsys(&tcp4_net_ops);
2373 void tcp4_proc_exit(void)
2375 unregister_pernet_subsys(&tcp4_net_ops);
2377 #endif /* CONFIG_PROC_FS */
2379 struct proto tcp_prot = {
2381 .owner = THIS_MODULE,
2383 .connect = tcp_v4_connect,
2384 .disconnect = tcp_disconnect,
2385 .accept = inet_csk_accept,
2387 .init = tcp_v4_init_sock,
2388 .destroy = tcp_v4_destroy_sock,
2389 .shutdown = tcp_shutdown,
2390 .setsockopt = tcp_setsockopt,
2391 .getsockopt = tcp_getsockopt,
2392 .recvmsg = tcp_recvmsg,
2393 .sendmsg = tcp_sendmsg,
2394 .sendpage = tcp_sendpage,
2395 .backlog_rcv = tcp_v4_do_rcv,
2396 .release_cb = tcp_release_cb,
2398 .unhash = inet_unhash,
2399 .get_port = inet_csk_get_port,
2400 .enter_memory_pressure = tcp_enter_memory_pressure,
2401 .stream_memory_free = tcp_stream_memory_free,
2402 .sockets_allocated = &tcp_sockets_allocated,
2403 .orphan_count = &tcp_orphan_count,
2404 .memory_allocated = &tcp_memory_allocated,
2405 .memory_pressure = &tcp_memory_pressure,
2406 .sysctl_mem = sysctl_tcp_mem,
2407 .sysctl_wmem = sysctl_tcp_wmem,
2408 .sysctl_rmem = sysctl_tcp_rmem,
2409 .max_header = MAX_TCP_HEADER,
2410 .obj_size = sizeof(struct tcp_sock),
2411 .slab_flags = SLAB_DESTROY_BY_RCU,
2412 .twsk_prot = &tcp_timewait_sock_ops,
2413 .rsk_prot = &tcp_request_sock_ops,
2414 .h.hashinfo = &tcp_hashinfo,
2415 .no_autobind = true,
2416 #ifdef CONFIG_COMPAT
2417 .compat_setsockopt = compat_tcp_setsockopt,
2418 .compat_getsockopt = compat_tcp_getsockopt,
2420 #ifdef CONFIG_MEMCG_KMEM
2421 .init_cgroup = tcp_init_cgroup,
2422 .destroy_cgroup = tcp_destroy_cgroup,
2423 .proto_cgroup = tcp_proto_cgroup,
2426 EXPORT_SYMBOL(tcp_prot);
2428 static void __net_exit tcp_sk_exit(struct net *net)
2432 for_each_possible_cpu(cpu)
2433 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2434 free_percpu(net->ipv4.tcp_sk);
2437 static int __net_init tcp_sk_init(struct net *net)
2441 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2442 if (!net->ipv4.tcp_sk)
2445 for_each_possible_cpu(cpu) {
2448 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2452 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2454 net->ipv4.sysctl_tcp_ecn = 2;
2455 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2456 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2457 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2466 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2468 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2471 static struct pernet_operations __net_initdata tcp_sk_ops = {
2472 .init = tcp_sk_init,
2473 .exit = tcp_sk_exit,
2474 .exit_batch = tcp_sk_exit_batch,
2477 void __init tcp_v4_init(void)
2479 inet_hashinfo_init(&tcp_hashinfo);
2480 if (register_pernet_subsys(&tcp_sk_ops))
2481 panic("Failed to create the TCP control socket.\n");