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/netdma.h>
76 #include <net/secure_seq.h>
77 #include <net/tcp_memcontrol.h>
78 #include <net/busy_poll.h>
80 #include <linux/inet.h>
81 #include <linux/ipv6.h>
82 #include <linux/stddef.h>
83 #include <linux/proc_fs.h>
84 #include <linux/seq_file.h>
86 #include <linux/crypto.h>
87 #include <linux/scatterlist.h>
89 int sysctl_tcp_tw_reuse __read_mostly;
90 int sysctl_tcp_low_latency __read_mostly;
91 EXPORT_SYMBOL(sysctl_tcp_low_latency);
94 #ifdef CONFIG_TCP_MD5SIG
95 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
96 __be32 daddr, __be32 saddr, const struct tcphdr *th);
99 struct inet_hashinfo tcp_hashinfo;
100 EXPORT_SYMBOL(tcp_hashinfo);
102 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
104 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
107 tcp_hdr(skb)->source);
110 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
112 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
113 struct tcp_sock *tp = tcp_sk(sk);
115 /* With PAWS, it is safe from the viewpoint
116 of data integrity. Even without PAWS it is safe provided sequence
117 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
119 Actually, the idea is close to VJ's one, only timestamp cache is
120 held not per host, but per port pair and TW bucket is used as state
123 If TW bucket has been already destroyed we fall back to VJ's scheme
124 and use initial timestamp retrieved from peer table.
126 if (tcptw->tw_ts_recent_stamp &&
127 (twp == NULL || (sysctl_tcp_tw_reuse &&
128 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
129 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
130 if (tp->write_seq == 0)
132 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
133 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
140 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
142 /* This will initiate an outgoing connection. */
143 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
145 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
146 struct inet_sock *inet = inet_sk(sk);
147 struct tcp_sock *tp = tcp_sk(sk);
148 __be16 orig_sport, orig_dport;
149 __be32 daddr, nexthop;
153 struct ip_options_rcu *inet_opt;
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 sock_owned_by_user(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 inet->inet_rcv_saddr = 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 if (tcp_death_row.sysctl_tw_recycle &&
205 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
206 tcp_fetch_timewait_stamp(sk, &rt->dst);
208 inet->inet_dport = usin->sin_port;
209 inet->inet_daddr = daddr;
213 inet_csk(sk)->icsk_ext_hdr_len = 0;
215 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
217 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
219 /* Socket identity is still unknown (sport may be zero).
220 * However we set state to SYN-SENT and not releasing socket
221 * lock select source port, enter ourselves into the hash tables and
222 * complete initialization after this.
224 tcp_set_state(sk, TCP_SYN_SENT);
225 err = inet_hash_connect(&tcp_death_row, sk);
229 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
230 inet->inet_sport, inet->inet_dport, sk);
236 /* OK, now commit destination to socket. */
237 sk->sk_gso_type = SKB_GSO_TCPV4;
238 sk_setup_caps(sk, &rt->dst);
240 if (!tp->write_seq && likely(!tp->repair))
241 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
246 inet->inet_id = tp->write_seq ^ jiffies;
248 err = tcp_connect(sk);
258 * This unhashes the socket and releases the local port,
261 tcp_set_state(sk, TCP_CLOSE);
263 sk->sk_route_caps = 0;
264 inet->inet_dport = 0;
267 EXPORT_SYMBOL(tcp_v4_connect);
270 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
271 * It can be called through tcp_release_cb() if socket was owned by user
272 * at the time tcp_v4_err() was called to handle ICMP message.
274 void tcp_v4_mtu_reduced(struct sock *sk)
276 struct dst_entry *dst;
277 struct inet_sock *inet = inet_sk(sk);
278 u32 mtu = tcp_sk(sk)->mtu_info;
280 dst = inet_csk_update_pmtu(sk, mtu);
284 /* Something is about to be wrong... Remember soft error
285 * for the case, if this connection will not able to recover.
287 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
288 sk->sk_err_soft = EMSGSIZE;
292 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
293 ip_sk_accept_pmtu(sk) &&
294 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
295 tcp_sync_mss(sk, mtu);
297 /* Resend the TCP packet because it's
298 * clear that the old packet has been
299 * dropped. This is the new "fast" path mtu
302 tcp_simple_retransmit(sk);
303 } /* else let the usual retransmit timer handle it */
305 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
307 static void do_redirect(struct sk_buff *skb, struct sock *sk)
309 struct dst_entry *dst = __sk_dst_check(sk, 0);
312 dst->ops->redirect(dst, sk, skb);
316 * This routine is called by the ICMP module when it gets some
317 * sort of error condition. If err < 0 then the socket should
318 * be closed and the error returned to the user. If err > 0
319 * it's just the icmp type << 8 | icmp code. After adjustment
320 * header points to the first 8 bytes of the tcp header. We need
321 * to find the appropriate port.
323 * The locking strategy used here is very "optimistic". When
324 * someone else accesses the socket the ICMP is just dropped
325 * and for some paths there is no check at all.
326 * A more general error queue to queue errors for later handling
327 * is probably better.
331 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
333 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
334 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
335 struct inet_connection_sock *icsk;
337 struct inet_sock *inet;
338 const int type = icmp_hdr(icmp_skb)->type;
339 const int code = icmp_hdr(icmp_skb)->code;
342 struct request_sock *fastopen;
346 struct net *net = dev_net(icmp_skb->dev);
348 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
349 iph->saddr, th->source, inet_iif(icmp_skb));
351 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
354 if (sk->sk_state == TCP_TIME_WAIT) {
355 inet_twsk_put(inet_twsk(sk));
360 /* If too many ICMPs get dropped on busy
361 * servers this needs to be solved differently.
362 * We do take care of PMTU discovery (RFC1191) special case :
363 * we can receive locally generated ICMP messages while socket is held.
365 if (sock_owned_by_user(sk)) {
366 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
367 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
369 if (sk->sk_state == TCP_CLOSE)
372 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
373 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
379 seq = ntohl(th->seq);
380 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
381 fastopen = tp->fastopen_rsk;
382 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
383 if (sk->sk_state != TCP_LISTEN &&
384 !between(seq, snd_una, tp->snd_nxt)) {
385 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
391 do_redirect(icmp_skb, sk);
393 case ICMP_SOURCE_QUENCH:
394 /* Just silently ignore these. */
396 case ICMP_PARAMETERPROB:
399 case ICMP_DEST_UNREACH:
400 if (code > NR_ICMP_UNREACH)
403 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
404 /* We are not interested in TCP_LISTEN and open_requests
405 * (SYN-ACKs send out by Linux are always <576bytes so
406 * they should go through unfragmented).
408 if (sk->sk_state == TCP_LISTEN)
412 if (!sock_owned_by_user(sk)) {
413 tcp_v4_mtu_reduced(sk);
415 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
421 err = icmp_err_convert[code].errno;
422 /* check if icmp_skb allows revert of backoff
423 * (see draft-zimmermann-tcp-lcd) */
424 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
426 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
427 !icsk->icsk_backoff || fastopen)
430 if (sock_owned_by_user(sk))
433 icsk->icsk_backoff--;
434 inet_csk(sk)->icsk_rto = (tp->srtt_us ? __tcp_set_rto(tp) :
435 TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
438 skb = tcp_write_queue_head(sk);
441 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
442 tcp_time_stamp - TCP_SKB_CB(skb)->when);
445 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
446 remaining, TCP_RTO_MAX);
448 /* RTO revert clocked out retransmission.
449 * Will retransmit now */
450 tcp_retransmit_timer(sk);
454 case ICMP_TIME_EXCEEDED:
461 switch (sk->sk_state) {
462 struct request_sock *req, **prev;
464 if (sock_owned_by_user(sk))
467 req = inet_csk_search_req(sk, &prev, th->dest,
468 iph->daddr, iph->saddr);
472 /* ICMPs are not backlogged, hence we cannot get
473 an established socket here.
477 if (seq != tcp_rsk(req)->snt_isn) {
478 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
483 * Still in SYN_RECV, just remove it silently.
484 * There is no good way to pass the error to the newly
485 * created socket, and POSIX does not want network
486 * errors returned from accept().
488 inet_csk_reqsk_queue_drop(sk, req, prev);
489 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
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 (skb_rtable(skb)->rt_type != RTN_LOCAL)
605 /* Swap the send and the receive. */
606 memset(&rep, 0, sizeof(rep));
607 rep.th.dest = th->source;
608 rep.th.source = th->dest;
609 rep.th.doff = sizeof(struct tcphdr) / 4;
613 rep.th.seq = th->ack_seq;
616 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
617 skb->len - (th->doff << 2));
620 memset(&arg, 0, sizeof(arg));
621 arg.iov[0].iov_base = (unsigned char *)&rep;
622 arg.iov[0].iov_len = sizeof(rep.th);
624 #ifdef CONFIG_TCP_MD5SIG
625 hash_location = tcp_parse_md5sig_option(th);
626 if (!sk && hash_location) {
628 * active side is lost. Try to find listening socket through
629 * source port, and then find md5 key through listening socket.
630 * we are not loose security here:
631 * Incoming packet is checked with md5 hash with finding key,
632 * no RST generated if md5 hash doesn't match.
634 sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev),
635 &tcp_hashinfo, ip_hdr(skb)->saddr,
636 th->source, ip_hdr(skb)->daddr,
637 ntohs(th->source), inet_iif(skb));
638 /* don't send rst if it can't find key */
642 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
643 &ip_hdr(skb)->saddr, AF_INET);
647 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb);
648 if (genhash || memcmp(hash_location, newhash, 16) != 0)
651 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
657 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
659 (TCPOPT_MD5SIG << 8) |
661 /* Update length and the length the header thinks exists */
662 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
663 rep.th.doff = arg.iov[0].iov_len / 4;
665 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
666 key, ip_hdr(skb)->saddr,
667 ip_hdr(skb)->daddr, &rep.th);
670 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
671 ip_hdr(skb)->saddr, /* XXX */
672 arg.iov[0].iov_len, IPPROTO_TCP, 0);
673 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
674 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
675 /* When socket is gone, all binding information is lost.
676 * routing might fail in this case. No choice here, if we choose to force
677 * input interface, we will misroute in case of asymmetric route.
680 arg.bound_dev_if = sk->sk_bound_dev_if;
682 net = dev_net(skb_dst(skb)->dev);
683 arg.tos = ip_hdr(skb)->tos;
684 ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
685 ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
687 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
688 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
690 #ifdef CONFIG_TCP_MD5SIG
699 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
700 outside socket context is ugly, certainly. What can I do?
703 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
704 u32 win, u32 tsval, u32 tsecr, int oif,
705 struct tcp_md5sig_key *key,
706 int reply_flags, u8 tos)
708 const struct tcphdr *th = tcp_hdr(skb);
711 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
712 #ifdef CONFIG_TCP_MD5SIG
713 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
717 struct ip_reply_arg arg;
718 struct net *net = dev_net(skb_dst(skb)->dev);
720 memset(&rep.th, 0, sizeof(struct tcphdr));
721 memset(&arg, 0, sizeof(arg));
723 arg.iov[0].iov_base = (unsigned char *)&rep;
724 arg.iov[0].iov_len = sizeof(rep.th);
726 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
727 (TCPOPT_TIMESTAMP << 8) |
729 rep.opt[1] = htonl(tsval);
730 rep.opt[2] = htonl(tsecr);
731 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
734 /* Swap the send and the receive. */
735 rep.th.dest = th->source;
736 rep.th.source = th->dest;
737 rep.th.doff = arg.iov[0].iov_len / 4;
738 rep.th.seq = htonl(seq);
739 rep.th.ack_seq = htonl(ack);
741 rep.th.window = htons(win);
743 #ifdef CONFIG_TCP_MD5SIG
745 int offset = (tsecr) ? 3 : 0;
747 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
749 (TCPOPT_MD5SIG << 8) |
751 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
752 rep.th.doff = arg.iov[0].iov_len/4;
754 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
755 key, ip_hdr(skb)->saddr,
756 ip_hdr(skb)->daddr, &rep.th);
759 arg.flags = reply_flags;
760 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
761 ip_hdr(skb)->saddr, /* XXX */
762 arg.iov[0].iov_len, IPPROTO_TCP, 0);
763 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
765 arg.bound_dev_if = oif;
767 ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
768 ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
770 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
773 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
775 struct inet_timewait_sock *tw = inet_twsk(sk);
776 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
778 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
779 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
780 tcp_time_stamp + tcptw->tw_ts_offset,
783 tcp_twsk_md5_key(tcptw),
784 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
791 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
792 struct request_sock *req)
794 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
795 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
797 tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
798 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
799 tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
803 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
805 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
810 * Send a SYN-ACK after having received a SYN.
811 * This still operates on a request_sock only, not on a big
814 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
816 struct request_sock *req,
818 struct tcp_fastopen_cookie *foc)
820 const struct inet_request_sock *ireq = inet_rsk(req);
825 /* First, grab a route. */
826 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
829 skb = tcp_make_synack(sk, dst, req, foc);
832 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
834 skb_set_queue_mapping(skb, queue_mapping);
835 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
838 err = net_xmit_eval(err);
845 * IPv4 request_sock destructor.
847 static void tcp_v4_reqsk_destructor(struct request_sock *req)
849 kfree(inet_rsk(req)->opt);
853 * Return true if a syncookie should be sent
855 bool tcp_syn_flood_action(struct sock *sk,
856 const struct sk_buff *skb,
859 const char *msg = "Dropping request";
860 bool want_cookie = false;
861 struct listen_sock *lopt;
863 #ifdef CONFIG_SYN_COOKIES
864 if (sysctl_tcp_syncookies) {
865 msg = "Sending cookies";
867 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
870 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
872 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
873 if (!lopt->synflood_warned && sysctl_tcp_syncookies != 2) {
874 lopt->synflood_warned = 1;
875 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
876 proto, ntohs(tcp_hdr(skb)->dest), msg);
880 EXPORT_SYMBOL(tcp_syn_flood_action);
883 * Save and compile IPv4 options into the request_sock if needed.
885 static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
887 const struct ip_options *opt = &(IPCB(skb)->opt);
888 struct ip_options_rcu *dopt = NULL;
890 if (opt && opt->optlen) {
891 int opt_size = sizeof(*dopt) + opt->optlen;
893 dopt = kmalloc(opt_size, GFP_ATOMIC);
895 if (ip_options_echo(&dopt->opt, skb)) {
904 #ifdef CONFIG_TCP_MD5SIG
906 * RFC2385 MD5 checksumming requires a mapping of
907 * IP address->MD5 Key.
908 * We need to maintain these in the sk structure.
911 /* Find the Key structure for an address. */
912 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
913 const union tcp_md5_addr *addr,
916 struct tcp_sock *tp = tcp_sk(sk);
917 struct tcp_md5sig_key *key;
918 unsigned int size = sizeof(struct in_addr);
919 struct tcp_md5sig_info *md5sig;
921 /* caller either holds rcu_read_lock() or socket lock */
922 md5sig = rcu_dereference_check(tp->md5sig_info,
923 sock_owned_by_user(sk) ||
924 lockdep_is_held(&sk->sk_lock.slock));
927 #if IS_ENABLED(CONFIG_IPV6)
928 if (family == AF_INET6)
929 size = sizeof(struct in6_addr);
931 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
932 if (key->family != family)
934 if (!memcmp(&key->addr, addr, size))
939 EXPORT_SYMBOL(tcp_md5_do_lookup);
941 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
942 struct sock *addr_sk)
944 union tcp_md5_addr *addr;
946 addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
947 return tcp_md5_do_lookup(sk, addr, AF_INET);
949 EXPORT_SYMBOL(tcp_v4_md5_lookup);
951 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
952 struct request_sock *req)
954 union tcp_md5_addr *addr;
956 addr = (union tcp_md5_addr *)&inet_rsk(req)->ir_rmt_addr;
957 return tcp_md5_do_lookup(sk, addr, AF_INET);
960 /* This can be called on a newly created socket, from other files */
961 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
962 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
964 /* Add Key to the list */
965 struct tcp_md5sig_key *key;
966 struct tcp_sock *tp = tcp_sk(sk);
967 struct tcp_md5sig_info *md5sig;
969 key = tcp_md5_do_lookup(sk, addr, family);
971 /* Pre-existing entry - just update that one. */
972 memcpy(key->key, newkey, newkeylen);
973 key->keylen = newkeylen;
977 md5sig = rcu_dereference_protected(tp->md5sig_info,
978 sock_owned_by_user(sk));
980 md5sig = kmalloc(sizeof(*md5sig), gfp);
984 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
985 INIT_HLIST_HEAD(&md5sig->head);
986 rcu_assign_pointer(tp->md5sig_info, md5sig);
989 key = sock_kmalloc(sk, sizeof(*key), gfp);
992 if (!tcp_alloc_md5sig_pool()) {
993 sock_kfree_s(sk, key, sizeof(*key));
997 memcpy(key->key, newkey, newkeylen);
998 key->keylen = newkeylen;
999 key->family = family;
1000 memcpy(&key->addr, addr,
1001 (family == AF_INET6) ? sizeof(struct in6_addr) :
1002 sizeof(struct in_addr));
1003 hlist_add_head_rcu(&key->node, &md5sig->head);
1006 EXPORT_SYMBOL(tcp_md5_do_add);
1008 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
1010 struct tcp_md5sig_key *key;
1012 key = tcp_md5_do_lookup(sk, addr, family);
1015 hlist_del_rcu(&key->node);
1016 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1017 kfree_rcu(key, rcu);
1020 EXPORT_SYMBOL(tcp_md5_do_del);
1022 static void tcp_clear_md5_list(struct sock *sk)
1024 struct tcp_sock *tp = tcp_sk(sk);
1025 struct tcp_md5sig_key *key;
1026 struct hlist_node *n;
1027 struct tcp_md5sig_info *md5sig;
1029 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1031 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1032 hlist_del_rcu(&key->node);
1033 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1034 kfree_rcu(key, rcu);
1038 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1041 struct tcp_md5sig cmd;
1042 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1044 if (optlen < sizeof(cmd))
1047 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1050 if (sin->sin_family != AF_INET)
1053 if (!cmd.tcpm_keylen)
1054 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1057 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1060 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1061 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1065 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1066 __be32 daddr, __be32 saddr, int nbytes)
1068 struct tcp4_pseudohdr *bp;
1069 struct scatterlist sg;
1071 bp = &hp->md5_blk.ip4;
1074 * 1. the TCP pseudo-header (in the order: source IP address,
1075 * destination IP address, zero-padded protocol number, and
1081 bp->protocol = IPPROTO_TCP;
1082 bp->len = cpu_to_be16(nbytes);
1084 sg_init_one(&sg, bp, sizeof(*bp));
1085 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1088 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1089 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1091 struct tcp_md5sig_pool *hp;
1092 struct hash_desc *desc;
1094 hp = tcp_get_md5sig_pool();
1096 goto clear_hash_noput;
1097 desc = &hp->md5_desc;
1099 if (crypto_hash_init(desc))
1101 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1103 if (tcp_md5_hash_header(hp, th))
1105 if (tcp_md5_hash_key(hp, key))
1107 if (crypto_hash_final(desc, md5_hash))
1110 tcp_put_md5sig_pool();
1114 tcp_put_md5sig_pool();
1116 memset(md5_hash, 0, 16);
1120 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1121 const struct sock *sk, const struct request_sock *req,
1122 const struct sk_buff *skb)
1124 struct tcp_md5sig_pool *hp;
1125 struct hash_desc *desc;
1126 const struct tcphdr *th = tcp_hdr(skb);
1127 __be32 saddr, daddr;
1130 saddr = inet_sk(sk)->inet_saddr;
1131 daddr = inet_sk(sk)->inet_daddr;
1133 saddr = inet_rsk(req)->ir_loc_addr;
1134 daddr = inet_rsk(req)->ir_rmt_addr;
1136 const struct iphdr *iph = ip_hdr(skb);
1141 hp = tcp_get_md5sig_pool();
1143 goto clear_hash_noput;
1144 desc = &hp->md5_desc;
1146 if (crypto_hash_init(desc))
1149 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1151 if (tcp_md5_hash_header(hp, th))
1153 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1155 if (tcp_md5_hash_key(hp, key))
1157 if (crypto_hash_final(desc, md5_hash))
1160 tcp_put_md5sig_pool();
1164 tcp_put_md5sig_pool();
1166 memset(md5_hash, 0, 16);
1169 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1171 static bool __tcp_v4_inbound_md5_hash(struct sock *sk,
1172 const struct sk_buff *skb)
1175 * This gets called for each TCP segment that arrives
1176 * so we want to be efficient.
1177 * We have 3 drop cases:
1178 * o No MD5 hash and one expected.
1179 * o MD5 hash and we're not expecting one.
1180 * o MD5 hash and its wrong.
1182 const __u8 *hash_location = NULL;
1183 struct tcp_md5sig_key *hash_expected;
1184 const struct iphdr *iph = ip_hdr(skb);
1185 const struct tcphdr *th = tcp_hdr(skb);
1187 unsigned char newhash[16];
1189 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1191 hash_location = tcp_parse_md5sig_option(th);
1193 /* We've parsed the options - do we have a hash? */
1194 if (!hash_expected && !hash_location)
1197 if (hash_expected && !hash_location) {
1198 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1202 if (!hash_expected && hash_location) {
1203 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1207 /* Okay, so this is hash_expected and hash_location -
1208 * so we need to calculate the checksum.
1210 genhash = tcp_v4_md5_hash_skb(newhash,
1214 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1215 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1216 &iph->saddr, ntohs(th->source),
1217 &iph->daddr, ntohs(th->dest),
1218 genhash ? " tcp_v4_calc_md5_hash failed"
1225 static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1230 ret = __tcp_v4_inbound_md5_hash(sk, skb);
1238 static void tcp_v4_init_req(struct request_sock *req, struct sock *sk,
1239 struct sk_buff *skb)
1241 struct inet_request_sock *ireq = inet_rsk(req);
1243 ireq->ir_loc_addr = ip_hdr(skb)->daddr;
1244 ireq->ir_rmt_addr = ip_hdr(skb)->saddr;
1245 ireq->no_srccheck = inet_sk(sk)->transparent;
1246 ireq->opt = tcp_v4_save_options(skb);
1249 static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl,
1250 const struct request_sock *req,
1253 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1256 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1265 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1267 .obj_size = sizeof(struct tcp_request_sock),
1268 .rtx_syn_ack = tcp_rtx_synack,
1269 .send_ack = tcp_v4_reqsk_send_ack,
1270 .destructor = tcp_v4_reqsk_destructor,
1271 .send_reset = tcp_v4_send_reset,
1272 .syn_ack_timeout = tcp_syn_ack_timeout,
1275 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1276 .mss_clamp = TCP_MSS_DEFAULT,
1277 #ifdef CONFIG_TCP_MD5SIG
1278 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1279 .calc_md5_hash = tcp_v4_md5_hash_skb,
1281 .init_req = tcp_v4_init_req,
1282 #ifdef CONFIG_SYN_COOKIES
1283 .cookie_init_seq = cookie_v4_init_sequence,
1285 .route_req = tcp_v4_route_req,
1286 .init_seq = tcp_v4_init_sequence,
1287 .send_synack = tcp_v4_send_synack,
1288 .queue_hash_add = inet_csk_reqsk_queue_hash_add,
1291 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1293 /* Never answer to SYNs send to broadcast or multicast */
1294 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1297 return tcp_conn_request(&tcp_request_sock_ops,
1298 &tcp_request_sock_ipv4_ops, sk, skb);
1301 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1304 EXPORT_SYMBOL(tcp_v4_conn_request);
1308 * The three way handshake has completed - we got a valid synack -
1309 * now create the new socket.
1311 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1312 struct request_sock *req,
1313 struct dst_entry *dst)
1315 struct inet_request_sock *ireq;
1316 struct inet_sock *newinet;
1317 struct tcp_sock *newtp;
1319 #ifdef CONFIG_TCP_MD5SIG
1320 struct tcp_md5sig_key *key;
1322 struct ip_options_rcu *inet_opt;
1324 if (sk_acceptq_is_full(sk))
1327 newsk = tcp_create_openreq_child(sk, req, skb);
1331 newsk->sk_gso_type = SKB_GSO_TCPV4;
1332 inet_sk_rx_dst_set(newsk, skb);
1334 newtp = tcp_sk(newsk);
1335 newinet = inet_sk(newsk);
1336 ireq = inet_rsk(req);
1337 newinet->inet_daddr = ireq->ir_rmt_addr;
1338 newinet->inet_rcv_saddr = ireq->ir_loc_addr;
1339 newinet->inet_saddr = ireq->ir_loc_addr;
1340 inet_opt = ireq->opt;
1341 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1343 newinet->mc_index = inet_iif(skb);
1344 newinet->mc_ttl = ip_hdr(skb)->ttl;
1345 newinet->rcv_tos = ip_hdr(skb)->tos;
1346 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1347 inet_set_txhash(newsk);
1349 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1350 newinet->inet_id = newtp->write_seq ^ jiffies;
1353 dst = inet_csk_route_child_sock(sk, newsk, req);
1357 /* syncookie case : see end of cookie_v4_check() */
1359 sk_setup_caps(newsk, dst);
1361 tcp_sync_mss(newsk, dst_mtu(dst));
1362 newtp->advmss = dst_metric_advmss(dst);
1363 if (tcp_sk(sk)->rx_opt.user_mss &&
1364 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1365 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1367 tcp_initialize_rcv_mss(newsk);
1369 #ifdef CONFIG_TCP_MD5SIG
1370 /* Copy over the MD5 key from the original socket */
1371 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1375 * We're using one, so create a matching key
1376 * on the newsk structure. If we fail to get
1377 * memory, then we end up not copying the key
1380 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1381 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1382 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1386 if (__inet_inherit_port(sk, newsk) < 0)
1388 __inet_hash_nolisten(newsk, NULL);
1393 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1397 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1400 inet_csk_prepare_forced_close(newsk);
1404 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1406 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1408 struct tcphdr *th = tcp_hdr(skb);
1409 const struct iphdr *iph = ip_hdr(skb);
1411 struct request_sock **prev;
1412 /* Find possible connection requests. */
1413 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1414 iph->saddr, iph->daddr);
1416 return tcp_check_req(sk, skb, req, prev, false);
1418 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1419 th->source, iph->daddr, th->dest, inet_iif(skb));
1422 if (nsk->sk_state != TCP_TIME_WAIT) {
1426 inet_twsk_put(inet_twsk(nsk));
1430 #ifdef CONFIG_SYN_COOKIES
1432 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1437 /* The socket must have it's spinlock held when we get
1440 * We have a potential double-lock case here, so even when
1441 * doing backlog processing we use the BH locking scheme.
1442 * This is because we cannot sleep with the original spinlock
1445 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1449 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1450 struct dst_entry *dst = sk->sk_rx_dst;
1452 sock_rps_save_rxhash(sk, skb);
1454 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1455 dst->ops->check(dst, 0) == NULL) {
1457 sk->sk_rx_dst = NULL;
1460 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1464 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1467 if (sk->sk_state == TCP_LISTEN) {
1468 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1473 sock_rps_save_rxhash(nsk, skb);
1474 if (tcp_child_process(sk, nsk, skb)) {
1481 sock_rps_save_rxhash(sk, skb);
1483 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1490 tcp_v4_send_reset(rsk, skb);
1493 /* Be careful here. If this function gets more complicated and
1494 * gcc suffers from register pressure on the x86, sk (in %ebx)
1495 * might be destroyed here. This current version compiles correctly,
1496 * but you have been warned.
1501 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1502 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1505 EXPORT_SYMBOL(tcp_v4_do_rcv);
1507 void tcp_v4_early_demux(struct sk_buff *skb)
1509 const struct iphdr *iph;
1510 const struct tcphdr *th;
1513 if (skb->pkt_type != PACKET_HOST)
1516 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1522 if (th->doff < sizeof(struct tcphdr) / 4)
1525 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1526 iph->saddr, th->source,
1527 iph->daddr, ntohs(th->dest),
1531 skb->destructor = sock_edemux;
1532 if (sk->sk_state != TCP_TIME_WAIT) {
1533 struct dst_entry *dst = sk->sk_rx_dst;
1536 dst = dst_check(dst, 0);
1538 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1539 skb_dst_set_noref(skb, dst);
1544 /* Packet is added to VJ-style prequeue for processing in process
1545 * context, if a reader task is waiting. Apparently, this exciting
1546 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1547 * failed somewhere. Latency? Burstiness? Well, at least now we will
1548 * see, why it failed. 8)8) --ANK
1551 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1553 struct tcp_sock *tp = tcp_sk(sk);
1555 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1558 if (skb->len <= tcp_hdrlen(skb) &&
1559 skb_queue_len(&tp->ucopy.prequeue) == 0)
1563 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1564 tp->ucopy.memory += skb->truesize;
1565 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1566 struct sk_buff *skb1;
1568 BUG_ON(sock_owned_by_user(sk));
1570 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1571 sk_backlog_rcv(sk, skb1);
1572 NET_INC_STATS_BH(sock_net(sk),
1573 LINUX_MIB_TCPPREQUEUEDROPPED);
1576 tp->ucopy.memory = 0;
1577 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1578 wake_up_interruptible_sync_poll(sk_sleep(sk),
1579 POLLIN | POLLRDNORM | POLLRDBAND);
1580 if (!inet_csk_ack_scheduled(sk))
1581 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1582 (3 * tcp_rto_min(sk)) / 4,
1587 EXPORT_SYMBOL(tcp_prequeue);
1593 int tcp_v4_rcv(struct sk_buff *skb)
1595 const struct iphdr *iph;
1596 const struct tcphdr *th;
1599 struct net *net = dev_net(skb->dev);
1601 if (skb->pkt_type != PACKET_HOST)
1604 /* Count it even if it's bad */
1605 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1607 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1612 if (th->doff < sizeof(struct tcphdr) / 4)
1614 if (!pskb_may_pull(skb, th->doff * 4))
1617 /* An explanation is required here, I think.
1618 * Packet length and doff are validated by header prediction,
1619 * provided case of th->doff==0 is eliminated.
1620 * So, we defer the checks. */
1622 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1627 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1628 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1629 skb->len - th->doff * 4);
1630 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1631 TCP_SKB_CB(skb)->when = 0;
1632 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1633 TCP_SKB_CB(skb)->sacked = 0;
1635 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1640 if (sk->sk_state == TCP_TIME_WAIT)
1643 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1644 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1645 goto discard_and_relse;
1648 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1649 goto discard_and_relse;
1651 #ifdef CONFIG_TCP_MD5SIG
1653 * We really want to reject the packet as early as possible
1655 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1656 * o There is an MD5 option and we're not expecting one
1658 if (tcp_v4_inbound_md5_hash(sk, skb))
1659 goto discard_and_relse;
1664 if (sk_filter(sk, skb))
1665 goto discard_and_relse;
1667 sk_mark_napi_id(sk, skb);
1670 bh_lock_sock_nested(sk);
1672 if (!sock_owned_by_user(sk)) {
1673 #ifdef CONFIG_NET_DMA
1674 struct tcp_sock *tp = tcp_sk(sk);
1675 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1676 tp->ucopy.dma_chan = net_dma_find_channel();
1677 if (tp->ucopy.dma_chan)
1678 ret = tcp_v4_do_rcv(sk, skb);
1682 if (!tcp_prequeue(sk, skb))
1683 ret = tcp_v4_do_rcv(sk, skb);
1685 } else if (unlikely(sk_add_backlog(sk, skb,
1686 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1688 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1689 goto discard_and_relse;
1698 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1701 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1703 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1705 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1707 tcp_v4_send_reset(NULL, skb);
1711 /* Discard frame. */
1720 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1721 inet_twsk_put(inet_twsk(sk));
1725 if (skb->len < (th->doff << 2)) {
1726 inet_twsk_put(inet_twsk(sk));
1729 if (tcp_checksum_complete(skb)) {
1730 inet_twsk_put(inet_twsk(sk));
1733 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1735 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1737 iph->saddr, th->source,
1738 iph->daddr, th->dest,
1741 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1742 inet_twsk_put(inet_twsk(sk));
1746 /* Fall through to ACK */
1749 tcp_v4_timewait_ack(sk, skb);
1753 case TCP_TW_SUCCESS:;
1758 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1759 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1760 .twsk_unique = tcp_twsk_unique,
1761 .twsk_destructor= tcp_twsk_destructor,
1764 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1766 struct dst_entry *dst = skb_dst(skb);
1769 sk->sk_rx_dst = dst;
1770 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1772 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1774 const struct inet_connection_sock_af_ops ipv4_specific = {
1775 .queue_xmit = ip_queue_xmit,
1776 .send_check = tcp_v4_send_check,
1777 .rebuild_header = inet_sk_rebuild_header,
1778 .sk_rx_dst_set = inet_sk_rx_dst_set,
1779 .conn_request = tcp_v4_conn_request,
1780 .syn_recv_sock = tcp_v4_syn_recv_sock,
1781 .net_header_len = sizeof(struct iphdr),
1782 .setsockopt = ip_setsockopt,
1783 .getsockopt = ip_getsockopt,
1784 .addr2sockaddr = inet_csk_addr2sockaddr,
1785 .sockaddr_len = sizeof(struct sockaddr_in),
1786 .bind_conflict = inet_csk_bind_conflict,
1787 #ifdef CONFIG_COMPAT
1788 .compat_setsockopt = compat_ip_setsockopt,
1789 .compat_getsockopt = compat_ip_getsockopt,
1791 .mtu_reduced = tcp_v4_mtu_reduced,
1793 EXPORT_SYMBOL(ipv4_specific);
1795 #ifdef CONFIG_TCP_MD5SIG
1796 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1797 .md5_lookup = tcp_v4_md5_lookup,
1798 .calc_md5_hash = tcp_v4_md5_hash_skb,
1799 .md5_parse = tcp_v4_parse_md5_keys,
1803 /* NOTE: A lot of things set to zero explicitly by call to
1804 * sk_alloc() so need not be done here.
1806 static int tcp_v4_init_sock(struct sock *sk)
1808 struct inet_connection_sock *icsk = inet_csk(sk);
1812 icsk->icsk_af_ops = &ipv4_specific;
1814 #ifdef CONFIG_TCP_MD5SIG
1815 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1821 void tcp_v4_destroy_sock(struct sock *sk)
1823 struct tcp_sock *tp = tcp_sk(sk);
1825 tcp_clear_xmit_timers(sk);
1827 tcp_cleanup_congestion_control(sk);
1829 /* Cleanup up the write buffer. */
1830 tcp_write_queue_purge(sk);
1832 /* Cleans up our, hopefully empty, out_of_order_queue. */
1833 __skb_queue_purge(&tp->out_of_order_queue);
1835 #ifdef CONFIG_TCP_MD5SIG
1836 /* Clean up the MD5 key list, if any */
1837 if (tp->md5sig_info) {
1838 tcp_clear_md5_list(sk);
1839 kfree_rcu(tp->md5sig_info, rcu);
1840 tp->md5sig_info = NULL;
1844 #ifdef CONFIG_NET_DMA
1845 /* Cleans up our sk_async_wait_queue */
1846 __skb_queue_purge(&sk->sk_async_wait_queue);
1849 /* Clean prequeue, it must be empty really */
1850 __skb_queue_purge(&tp->ucopy.prequeue);
1852 /* Clean up a referenced TCP bind bucket. */
1853 if (inet_csk(sk)->icsk_bind_hash)
1856 BUG_ON(tp->fastopen_rsk != NULL);
1858 /* If socket is aborted during connect operation */
1859 tcp_free_fastopen_req(tp);
1861 sk_sockets_allocated_dec(sk);
1862 sock_release_memcg(sk);
1864 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1866 #ifdef CONFIG_PROC_FS
1867 /* Proc filesystem TCP sock list dumping. */
1870 * Get next listener socket follow cur. If cur is NULL, get first socket
1871 * starting from bucket given in st->bucket; when st->bucket is zero the
1872 * very first socket in the hash table is returned.
1874 static void *listening_get_next(struct seq_file *seq, void *cur)
1876 struct inet_connection_sock *icsk;
1877 struct hlist_nulls_node *node;
1878 struct sock *sk = cur;
1879 struct inet_listen_hashbucket *ilb;
1880 struct tcp_iter_state *st = seq->private;
1881 struct net *net = seq_file_net(seq);
1884 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1885 spin_lock_bh(&ilb->lock);
1886 sk = sk_nulls_head(&ilb->head);
1890 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1894 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1895 struct request_sock *req = cur;
1897 icsk = inet_csk(st->syn_wait_sk);
1901 if (req->rsk_ops->family == st->family) {
1907 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1910 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1912 sk = sk_nulls_next(st->syn_wait_sk);
1913 st->state = TCP_SEQ_STATE_LISTENING;
1914 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1916 icsk = inet_csk(sk);
1917 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1918 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1920 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1921 sk = sk_nulls_next(sk);
1924 sk_nulls_for_each_from(sk, node) {
1925 if (!net_eq(sock_net(sk), net))
1927 if (sk->sk_family == st->family) {
1931 icsk = inet_csk(sk);
1932 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1933 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1935 st->uid = sock_i_uid(sk);
1936 st->syn_wait_sk = sk;
1937 st->state = TCP_SEQ_STATE_OPENREQ;
1941 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1943 spin_unlock_bh(&ilb->lock);
1945 if (++st->bucket < INET_LHTABLE_SIZE) {
1946 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1947 spin_lock_bh(&ilb->lock);
1948 sk = sk_nulls_head(&ilb->head);
1956 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1958 struct tcp_iter_state *st = seq->private;
1963 rc = listening_get_next(seq, NULL);
1965 while (rc && *pos) {
1966 rc = listening_get_next(seq, rc);
1972 static inline bool empty_bucket(const struct tcp_iter_state *st)
1974 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1978 * Get first established socket starting from bucket given in st->bucket.
1979 * If st->bucket is zero, the very first socket in the hash is returned.
1981 static void *established_get_first(struct seq_file *seq)
1983 struct tcp_iter_state *st = seq->private;
1984 struct net *net = seq_file_net(seq);
1988 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1990 struct hlist_nulls_node *node;
1991 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1993 /* Lockless fast path for the common case of empty buckets */
1994 if (empty_bucket(st))
1998 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1999 if (sk->sk_family != st->family ||
2000 !net_eq(sock_net(sk), net)) {
2006 spin_unlock_bh(lock);
2012 static void *established_get_next(struct seq_file *seq, void *cur)
2014 struct sock *sk = cur;
2015 struct hlist_nulls_node *node;
2016 struct tcp_iter_state *st = seq->private;
2017 struct net *net = seq_file_net(seq);
2022 sk = sk_nulls_next(sk);
2024 sk_nulls_for_each_from(sk, node) {
2025 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2029 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2031 return established_get_first(seq);
2034 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2036 struct tcp_iter_state *st = seq->private;
2040 rc = established_get_first(seq);
2043 rc = established_get_next(seq, rc);
2049 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2052 struct tcp_iter_state *st = seq->private;
2054 st->state = TCP_SEQ_STATE_LISTENING;
2055 rc = listening_get_idx(seq, &pos);
2058 st->state = TCP_SEQ_STATE_ESTABLISHED;
2059 rc = established_get_idx(seq, pos);
2065 static void *tcp_seek_last_pos(struct seq_file *seq)
2067 struct tcp_iter_state *st = seq->private;
2068 int offset = st->offset;
2069 int orig_num = st->num;
2072 switch (st->state) {
2073 case TCP_SEQ_STATE_OPENREQ:
2074 case TCP_SEQ_STATE_LISTENING:
2075 if (st->bucket >= INET_LHTABLE_SIZE)
2077 st->state = TCP_SEQ_STATE_LISTENING;
2078 rc = listening_get_next(seq, NULL);
2079 while (offset-- && rc)
2080 rc = listening_get_next(seq, rc);
2084 st->state = TCP_SEQ_STATE_ESTABLISHED;
2086 case TCP_SEQ_STATE_ESTABLISHED:
2087 if (st->bucket > tcp_hashinfo.ehash_mask)
2089 rc = established_get_first(seq);
2090 while (offset-- && rc)
2091 rc = established_get_next(seq, rc);
2099 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2101 struct tcp_iter_state *st = seq->private;
2104 if (*pos && *pos == st->last_pos) {
2105 rc = tcp_seek_last_pos(seq);
2110 st->state = TCP_SEQ_STATE_LISTENING;
2114 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2117 st->last_pos = *pos;
2121 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2123 struct tcp_iter_state *st = seq->private;
2126 if (v == SEQ_START_TOKEN) {
2127 rc = tcp_get_idx(seq, 0);
2131 switch (st->state) {
2132 case TCP_SEQ_STATE_OPENREQ:
2133 case TCP_SEQ_STATE_LISTENING:
2134 rc = listening_get_next(seq, v);
2136 st->state = TCP_SEQ_STATE_ESTABLISHED;
2139 rc = established_get_first(seq);
2142 case TCP_SEQ_STATE_ESTABLISHED:
2143 rc = established_get_next(seq, v);
2148 st->last_pos = *pos;
2152 static void tcp_seq_stop(struct seq_file *seq, void *v)
2154 struct tcp_iter_state *st = seq->private;
2156 switch (st->state) {
2157 case TCP_SEQ_STATE_OPENREQ:
2159 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2160 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2162 case TCP_SEQ_STATE_LISTENING:
2163 if (v != SEQ_START_TOKEN)
2164 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2166 case TCP_SEQ_STATE_ESTABLISHED:
2168 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2173 int tcp_seq_open(struct inode *inode, struct file *file)
2175 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2176 struct tcp_iter_state *s;
2179 err = seq_open_net(inode, file, &afinfo->seq_ops,
2180 sizeof(struct tcp_iter_state));
2184 s = ((struct seq_file *)file->private_data)->private;
2185 s->family = afinfo->family;
2189 EXPORT_SYMBOL(tcp_seq_open);
2191 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2194 struct proc_dir_entry *p;
2196 afinfo->seq_ops.start = tcp_seq_start;
2197 afinfo->seq_ops.next = tcp_seq_next;
2198 afinfo->seq_ops.stop = tcp_seq_stop;
2200 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2201 afinfo->seq_fops, afinfo);
2206 EXPORT_SYMBOL(tcp_proc_register);
2208 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2210 remove_proc_entry(afinfo->name, net->proc_net);
2212 EXPORT_SYMBOL(tcp_proc_unregister);
2214 static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2215 struct seq_file *f, int i, kuid_t uid)
2217 const struct inet_request_sock *ireq = inet_rsk(req);
2218 long delta = req->expires - jiffies;
2220 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2221 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2224 ntohs(inet_sk(sk)->inet_sport),
2226 ntohs(ireq->ir_rmt_port),
2228 0, 0, /* could print option size, but that is af dependent. */
2229 1, /* timers active (only the expire timer) */
2230 jiffies_delta_to_clock_t(delta),
2232 from_kuid_munged(seq_user_ns(f), uid),
2233 0, /* non standard timer */
2234 0, /* open_requests have no inode */
2235 atomic_read(&sk->sk_refcnt),
2239 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2242 unsigned long timer_expires;
2243 const struct tcp_sock *tp = tcp_sk(sk);
2244 const struct inet_connection_sock *icsk = inet_csk(sk);
2245 const struct inet_sock *inet = inet_sk(sk);
2246 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2247 __be32 dest = inet->inet_daddr;
2248 __be32 src = inet->inet_rcv_saddr;
2249 __u16 destp = ntohs(inet->inet_dport);
2250 __u16 srcp = ntohs(inet->inet_sport);
2253 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2254 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2255 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2257 timer_expires = icsk->icsk_timeout;
2258 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2260 timer_expires = icsk->icsk_timeout;
2261 } else if (timer_pending(&sk->sk_timer)) {
2263 timer_expires = sk->sk_timer.expires;
2266 timer_expires = jiffies;
2269 if (sk->sk_state == TCP_LISTEN)
2270 rx_queue = sk->sk_ack_backlog;
2273 * because we dont lock socket, we might find a transient negative value
2275 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2277 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2278 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2279 i, src, srcp, dest, destp, sk->sk_state,
2280 tp->write_seq - tp->snd_una,
2283 jiffies_delta_to_clock_t(timer_expires - jiffies),
2284 icsk->icsk_retransmits,
2285 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2286 icsk->icsk_probes_out,
2288 atomic_read(&sk->sk_refcnt), sk,
2289 jiffies_to_clock_t(icsk->icsk_rto),
2290 jiffies_to_clock_t(icsk->icsk_ack.ato),
2291 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2293 sk->sk_state == TCP_LISTEN ?
2294 (fastopenq ? fastopenq->max_qlen : 0) :
2295 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2298 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2299 struct seq_file *f, int i)
2303 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
2305 dest = tw->tw_daddr;
2306 src = tw->tw_rcv_saddr;
2307 destp = ntohs(tw->tw_dport);
2308 srcp = ntohs(tw->tw_sport);
2310 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2311 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2312 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2313 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2314 atomic_read(&tw->tw_refcnt), tw);
2319 static int tcp4_seq_show(struct seq_file *seq, void *v)
2321 struct tcp_iter_state *st;
2322 struct sock *sk = v;
2324 seq_setwidth(seq, TMPSZ - 1);
2325 if (v == SEQ_START_TOKEN) {
2326 seq_puts(seq, " sl local_address rem_address st tx_queue "
2327 "rx_queue tr tm->when retrnsmt uid timeout "
2333 switch (st->state) {
2334 case TCP_SEQ_STATE_LISTENING:
2335 case TCP_SEQ_STATE_ESTABLISHED:
2336 if (sk->sk_state == TCP_TIME_WAIT)
2337 get_timewait4_sock(v, seq, st->num);
2339 get_tcp4_sock(v, seq, st->num);
2341 case TCP_SEQ_STATE_OPENREQ:
2342 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid);
2350 static const struct file_operations tcp_afinfo_seq_fops = {
2351 .owner = THIS_MODULE,
2352 .open = tcp_seq_open,
2354 .llseek = seq_lseek,
2355 .release = seq_release_net
2358 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2361 .seq_fops = &tcp_afinfo_seq_fops,
2363 .show = tcp4_seq_show,
2367 static int __net_init tcp4_proc_init_net(struct net *net)
2369 return tcp_proc_register(net, &tcp4_seq_afinfo);
2372 static void __net_exit tcp4_proc_exit_net(struct net *net)
2374 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2377 static struct pernet_operations tcp4_net_ops = {
2378 .init = tcp4_proc_init_net,
2379 .exit = tcp4_proc_exit_net,
2382 int __init tcp4_proc_init(void)
2384 return register_pernet_subsys(&tcp4_net_ops);
2387 void tcp4_proc_exit(void)
2389 unregister_pernet_subsys(&tcp4_net_ops);
2391 #endif /* CONFIG_PROC_FS */
2393 struct proto tcp_prot = {
2395 .owner = THIS_MODULE,
2397 .connect = tcp_v4_connect,
2398 .disconnect = tcp_disconnect,
2399 .accept = inet_csk_accept,
2401 .init = tcp_v4_init_sock,
2402 .destroy = tcp_v4_destroy_sock,
2403 .shutdown = tcp_shutdown,
2404 .setsockopt = tcp_setsockopt,
2405 .getsockopt = tcp_getsockopt,
2406 .recvmsg = tcp_recvmsg,
2407 .sendmsg = tcp_sendmsg,
2408 .sendpage = tcp_sendpage,
2409 .backlog_rcv = tcp_v4_do_rcv,
2410 .release_cb = tcp_release_cb,
2412 .unhash = inet_unhash,
2413 .get_port = inet_csk_get_port,
2414 .enter_memory_pressure = tcp_enter_memory_pressure,
2415 .stream_memory_free = tcp_stream_memory_free,
2416 .sockets_allocated = &tcp_sockets_allocated,
2417 .orphan_count = &tcp_orphan_count,
2418 .memory_allocated = &tcp_memory_allocated,
2419 .memory_pressure = &tcp_memory_pressure,
2420 .sysctl_mem = sysctl_tcp_mem,
2421 .sysctl_wmem = sysctl_tcp_wmem,
2422 .sysctl_rmem = sysctl_tcp_rmem,
2423 .max_header = MAX_TCP_HEADER,
2424 .obj_size = sizeof(struct tcp_sock),
2425 .slab_flags = SLAB_DESTROY_BY_RCU,
2426 .twsk_prot = &tcp_timewait_sock_ops,
2427 .rsk_prot = &tcp_request_sock_ops,
2428 .h.hashinfo = &tcp_hashinfo,
2429 .no_autobind = true,
2430 #ifdef CONFIG_COMPAT
2431 .compat_setsockopt = compat_tcp_setsockopt,
2432 .compat_getsockopt = compat_tcp_getsockopt,
2434 #ifdef CONFIG_MEMCG_KMEM
2435 .init_cgroup = tcp_init_cgroup,
2436 .destroy_cgroup = tcp_destroy_cgroup,
2437 .proto_cgroup = tcp_proto_cgroup,
2440 EXPORT_SYMBOL(tcp_prot);
2442 static int __net_init tcp_sk_init(struct net *net)
2444 net->ipv4.sysctl_tcp_ecn = 2;
2448 static void __net_exit tcp_sk_exit(struct net *net)
2452 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2454 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2457 static struct pernet_operations __net_initdata tcp_sk_ops = {
2458 .init = tcp_sk_init,
2459 .exit = tcp_sk_exit,
2460 .exit_batch = tcp_sk_exit_batch,
2463 void __init tcp_v4_init(void)
2465 inet_hashinfo_init(&tcp_hashinfo);
2466 if (register_pernet_subsys(&tcp_sk_ops))
2467 panic("Failed to create the TCP control socket.\n");