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>
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);
93 #ifdef CONFIG_TCP_MD5SIG
94 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
95 __be32 daddr, __be32 saddr, const struct tcphdr *th);
98 struct inet_hashinfo tcp_hashinfo;
99 EXPORT_SYMBOL(tcp_hashinfo);
101 static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
103 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
106 tcp_hdr(skb)->source);
109 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
111 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
112 struct tcp_sock *tp = tcp_sk(sk);
114 /* With PAWS, it is safe from the viewpoint
115 of data integrity. Even without PAWS it is safe provided sequence
116 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
118 Actually, the idea is close to VJ's one, only timestamp cache is
119 held not per host, but per port pair and TW bucket is used as state
122 If TW bucket has been already destroyed we fall back to VJ's scheme
123 and use initial timestamp retrieved from peer table.
125 if (tcptw->tw_ts_recent_stamp &&
126 (twp == NULL || (sysctl_tcp_tw_reuse &&
127 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
128 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
129 if (tp->write_seq == 0)
131 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
132 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
139 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
141 /* This will initiate an outgoing connection. */
142 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
144 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
145 struct inet_sock *inet = inet_sk(sk);
146 struct tcp_sock *tp = tcp_sk(sk);
147 __be16 orig_sport, orig_dport;
148 __be32 daddr, nexthop;
152 struct ip_options_rcu *inet_opt;
154 if (addr_len < sizeof(struct sockaddr_in))
157 if (usin->sin_family != AF_INET)
158 return -EAFNOSUPPORT;
160 nexthop = daddr = usin->sin_addr.s_addr;
161 inet_opt = rcu_dereference_protected(inet->inet_opt,
162 sock_owned_by_user(sk));
163 if (inet_opt && inet_opt->opt.srr) {
166 nexthop = inet_opt->opt.faddr;
169 orig_sport = inet->inet_sport;
170 orig_dport = usin->sin_port;
171 fl4 = &inet->cork.fl.u.ip4;
172 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
173 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
175 orig_sport, orig_dport, sk, true);
178 if (err == -ENETUNREACH)
179 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
183 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
188 if (!inet_opt || !inet_opt->opt.srr)
191 if (!inet->inet_saddr)
192 inet->inet_saddr = fl4->saddr;
193 inet->inet_rcv_saddr = inet->inet_saddr;
195 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
196 /* Reset inherited state */
197 tp->rx_opt.ts_recent = 0;
198 tp->rx_opt.ts_recent_stamp = 0;
199 if (likely(!tp->repair))
203 if (tcp_death_row.sysctl_tw_recycle &&
204 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
205 tcp_fetch_timewait_stamp(sk, &rt->dst);
207 inet->inet_dport = usin->sin_port;
208 inet->inet_daddr = daddr;
210 inet_csk(sk)->icsk_ext_hdr_len = 0;
212 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
214 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
216 /* Socket identity is still unknown (sport may be zero).
217 * However we set state to SYN-SENT and not releasing socket
218 * lock select source port, enter ourselves into the hash tables and
219 * complete initialization after this.
221 tcp_set_state(sk, TCP_SYN_SENT);
222 err = inet_hash_connect(&tcp_death_row, sk);
226 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
227 inet->inet_sport, inet->inet_dport, sk);
233 /* OK, now commit destination to socket. */
234 sk->sk_gso_type = SKB_GSO_TCPV4;
235 sk_setup_caps(sk, &rt->dst);
237 if (!tp->write_seq && likely(!tp->repair))
238 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
243 inet->inet_id = tp->write_seq ^ jiffies;
245 err = tcp_connect(sk);
255 * This unhashes the socket and releases the local port,
258 tcp_set_state(sk, TCP_CLOSE);
260 sk->sk_route_caps = 0;
261 inet->inet_dport = 0;
264 EXPORT_SYMBOL(tcp_v4_connect);
267 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
268 * It can be called through tcp_release_cb() if socket was owned by user
269 * at the time tcp_v4_err() was called to handle ICMP message.
271 static void tcp_v4_mtu_reduced(struct sock *sk)
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
275 u32 mtu = tcp_sk(sk)->mtu_info;
277 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
278 * send out by Linux are always <576bytes so they should go through
281 if (sk->sk_state == TCP_LISTEN)
284 dst = inet_csk_update_pmtu(sk, mtu);
288 /* Something is about to be wrong... Remember soft error
289 * for the case, if this connection will not able to recover.
291 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
292 sk->sk_err_soft = EMSGSIZE;
296 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
297 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
298 tcp_sync_mss(sk, mtu);
300 /* Resend the TCP packet because it's
301 * clear that the old packet has been
302 * dropped. This is the new "fast" path mtu
305 tcp_simple_retransmit(sk);
306 } /* else let the usual retransmit timer handle it */
309 static void do_redirect(struct sk_buff *skb, struct sock *sk)
311 struct dst_entry *dst = __sk_dst_check(sk, 0);
314 dst->ops->redirect(dst, sk, skb);
318 * This routine is called by the ICMP module when it gets some
319 * sort of error condition. If err < 0 then the socket should
320 * be closed and the error returned to the user. If err > 0
321 * it's just the icmp type << 8 | icmp code. After adjustment
322 * header points to the first 8 bytes of the tcp header. We need
323 * to find the appropriate port.
325 * The locking strategy used here is very "optimistic". When
326 * someone else accesses the socket the ICMP is just dropped
327 * and for some paths there is no check at all.
328 * A more general error queue to queue errors for later handling
329 * is probably better.
333 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
335 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
336 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
337 struct inet_connection_sock *icsk;
339 struct inet_sock *inet;
340 const int type = icmp_hdr(icmp_skb)->type;
341 const int code = icmp_hdr(icmp_skb)->code;
344 struct request_sock *req;
348 struct net *net = dev_net(icmp_skb->dev);
350 if (icmp_skb->len < (iph->ihl << 2) + 8) {
351 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
355 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
356 iph->saddr, th->source, inet_iif(icmp_skb));
358 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
361 if (sk->sk_state == TCP_TIME_WAIT) {
362 inet_twsk_put(inet_twsk(sk));
367 /* If too many ICMPs get dropped on busy
368 * servers this needs to be solved differently.
369 * We do take care of PMTU discovery (RFC1191) special case :
370 * we can receive locally generated ICMP messages while socket is held.
372 if (sock_owned_by_user(sk)) {
373 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
374 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
376 if (sk->sk_state == TCP_CLOSE)
379 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
380 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
386 req = tp->fastopen_rsk;
387 seq = ntohl(th->seq);
388 if (sk->sk_state != TCP_LISTEN &&
389 !between(seq, tp->snd_una, tp->snd_nxt) &&
390 (req == NULL || seq != tcp_rsk(req)->snt_isn)) {
391 /* For a Fast Open socket, allow seq to be snt_isn. */
392 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
398 do_redirect(icmp_skb, sk);
400 case ICMP_SOURCE_QUENCH:
401 /* Just silently ignore these. */
403 case ICMP_PARAMETERPROB:
406 case ICMP_DEST_UNREACH:
407 if (code > NR_ICMP_UNREACH)
410 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
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 ||
430 /* XXX (TFO) - revisit the following logic for TFO */
432 if (sock_owned_by_user(sk))
435 icsk->icsk_backoff--;
436 inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
437 TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
440 skb = tcp_write_queue_head(sk);
443 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
444 tcp_time_stamp - TCP_SKB_CB(skb)->when);
447 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
448 remaining, TCP_RTO_MAX);
450 /* RTO revert clocked out retransmission.
451 * Will retransmit now */
452 tcp_retransmit_timer(sk);
456 case ICMP_TIME_EXCEEDED:
463 /* XXX (TFO) - if it's a TFO socket and has been accepted, rather
464 * than following the TCP_SYN_RECV case and closing the socket,
465 * we ignore the ICMP error and keep trying like a fully established
466 * socket. Is this the right thing to do?
468 if (req && req->sk == NULL)
471 switch (sk->sk_state) {
472 struct request_sock *req, **prev;
474 if (sock_owned_by_user(sk))
477 req = inet_csk_search_req(sk, &prev, th->dest,
478 iph->daddr, iph->saddr);
482 /* ICMPs are not backlogged, hence we cannot get
483 an established socket here.
487 if (seq != tcp_rsk(req)->snt_isn) {
488 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
493 * Still in SYN_RECV, just remove it silently.
494 * There is no good way to pass the error to the newly
495 * created socket, and POSIX does not want network
496 * errors returned from accept().
498 inet_csk_reqsk_queue_drop(sk, req, prev);
499 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
503 case TCP_SYN_RECV: /* Cannot happen.
504 It can f.e. if SYNs crossed,
507 if (!sock_owned_by_user(sk)) {
510 sk->sk_error_report(sk);
514 sk->sk_err_soft = err;
519 /* If we've already connected we will keep trying
520 * until we time out, or the user gives up.
522 * rfc1122 4.2.3.9 allows to consider as hard errors
523 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
524 * but it is obsoleted by pmtu discovery).
526 * Note, that in modern internet, where routing is unreliable
527 * and in each dark corner broken firewalls sit, sending random
528 * errors ordered by their masters even this two messages finally lose
529 * their original sense (even Linux sends invalid PORT_UNREACHs)
531 * Now we are in compliance with RFCs.
536 if (!sock_owned_by_user(sk) && inet->recverr) {
538 sk->sk_error_report(sk);
539 } else { /* Only an error on timeout */
540 sk->sk_err_soft = err;
548 static void __tcp_v4_send_check(struct sk_buff *skb,
549 __be32 saddr, __be32 daddr)
551 struct tcphdr *th = tcp_hdr(skb);
553 if (skb->ip_summed == CHECKSUM_PARTIAL) {
554 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
555 skb->csum_start = skb_transport_header(skb) - skb->head;
556 skb->csum_offset = offsetof(struct tcphdr, check);
558 th->check = tcp_v4_check(skb->len, saddr, daddr,
565 /* This routine computes an IPv4 TCP checksum. */
566 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
568 const struct inet_sock *inet = inet_sk(sk);
570 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
572 EXPORT_SYMBOL(tcp_v4_send_check);
574 int tcp_v4_gso_send_check(struct sk_buff *skb)
576 const struct iphdr *iph;
579 if (!pskb_may_pull(skb, sizeof(*th)))
586 skb->ip_summed = CHECKSUM_PARTIAL;
587 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
592 * This routine will send an RST to the other tcp.
594 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
596 * Answer: if a packet caused RST, it is not for a socket
597 * existing in our system, if it is matched to a socket,
598 * it is just duplicate segment or bug in other side's TCP.
599 * So that we build reply only basing on parameters
600 * arrived with segment.
601 * Exception: precedence violation. We do not implement it in any case.
604 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
606 const struct tcphdr *th = tcp_hdr(skb);
609 #ifdef CONFIG_TCP_MD5SIG
610 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
613 struct ip_reply_arg arg;
614 #ifdef CONFIG_TCP_MD5SIG
615 struct tcp_md5sig_key *key;
616 const __u8 *hash_location = NULL;
617 unsigned char newhash[16];
619 struct sock *sk1 = NULL;
623 /* Never send a reset in response to a reset. */
627 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
630 /* Swap the send and the receive. */
631 memset(&rep, 0, sizeof(rep));
632 rep.th.dest = th->source;
633 rep.th.source = th->dest;
634 rep.th.doff = sizeof(struct tcphdr) / 4;
638 rep.th.seq = th->ack_seq;
641 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
642 skb->len - (th->doff << 2));
645 memset(&arg, 0, sizeof(arg));
646 arg.iov[0].iov_base = (unsigned char *)&rep;
647 arg.iov[0].iov_len = sizeof(rep.th);
649 #ifdef CONFIG_TCP_MD5SIG
650 hash_location = tcp_parse_md5sig_option(th);
651 if (!sk && hash_location) {
653 * active side is lost. Try to find listening socket through
654 * source port, and then find md5 key through listening socket.
655 * we are not loose security here:
656 * Incoming packet is checked with md5 hash with finding key,
657 * no RST generated if md5 hash doesn't match.
659 sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev),
660 &tcp_hashinfo, ip_hdr(skb)->saddr,
661 th->source, ip_hdr(skb)->daddr,
662 ntohs(th->source), inet_iif(skb));
663 /* don't send rst if it can't find key */
667 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
668 &ip_hdr(skb)->saddr, AF_INET);
672 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb);
673 if (genhash || memcmp(hash_location, newhash, 16) != 0)
676 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
682 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
684 (TCPOPT_MD5SIG << 8) |
686 /* Update length and the length the header thinks exists */
687 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
688 rep.th.doff = arg.iov[0].iov_len / 4;
690 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
691 key, ip_hdr(skb)->saddr,
692 ip_hdr(skb)->daddr, &rep.th);
695 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
696 ip_hdr(skb)->saddr, /* XXX */
697 arg.iov[0].iov_len, IPPROTO_TCP, 0);
698 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
699 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
700 /* When socket is gone, all binding information is lost.
701 * routing might fail in this case. No choice here, if we choose to force
702 * input interface, we will misroute in case of asymmetric route.
705 arg.bound_dev_if = sk->sk_bound_dev_if;
707 net = dev_net(skb_dst(skb)->dev);
708 arg.tos = ip_hdr(skb)->tos;
709 ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
710 ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
712 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
713 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
715 #ifdef CONFIG_TCP_MD5SIG
724 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
725 outside socket context is ugly, certainly. What can I do?
728 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
729 u32 win, u32 tsval, u32 tsecr, int oif,
730 struct tcp_md5sig_key *key,
731 int reply_flags, u8 tos)
733 const struct tcphdr *th = tcp_hdr(skb);
736 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
737 #ifdef CONFIG_TCP_MD5SIG
738 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
742 struct ip_reply_arg arg;
743 struct net *net = dev_net(skb_dst(skb)->dev);
745 memset(&rep.th, 0, sizeof(struct tcphdr));
746 memset(&arg, 0, sizeof(arg));
748 arg.iov[0].iov_base = (unsigned char *)&rep;
749 arg.iov[0].iov_len = sizeof(rep.th);
751 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
752 (TCPOPT_TIMESTAMP << 8) |
754 rep.opt[1] = htonl(tsval);
755 rep.opt[2] = htonl(tsecr);
756 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
759 /* Swap the send and the receive. */
760 rep.th.dest = th->source;
761 rep.th.source = th->dest;
762 rep.th.doff = arg.iov[0].iov_len / 4;
763 rep.th.seq = htonl(seq);
764 rep.th.ack_seq = htonl(ack);
766 rep.th.window = htons(win);
768 #ifdef CONFIG_TCP_MD5SIG
770 int offset = (tsecr) ? 3 : 0;
772 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
774 (TCPOPT_MD5SIG << 8) |
776 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
777 rep.th.doff = arg.iov[0].iov_len/4;
779 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
780 key, ip_hdr(skb)->saddr,
781 ip_hdr(skb)->daddr, &rep.th);
784 arg.flags = reply_flags;
785 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
786 ip_hdr(skb)->saddr, /* XXX */
787 arg.iov[0].iov_len, IPPROTO_TCP, 0);
788 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
790 arg.bound_dev_if = oif;
792 ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
793 ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
795 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
798 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
800 struct inet_timewait_sock *tw = inet_twsk(sk);
801 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
803 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
804 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
805 tcp_time_stamp + tcptw->tw_ts_offset,
808 tcp_twsk_md5_key(tcptw),
809 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
816 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
817 struct request_sock *req)
819 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
820 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
822 tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
823 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
824 tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
828 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
830 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
835 * Send a SYN-ACK after having received a SYN.
836 * This still operates on a request_sock only, not on a big
839 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
840 struct request_sock *req,
844 const struct inet_request_sock *ireq = inet_rsk(req);
847 struct sk_buff * skb;
849 /* First, grab a route. */
850 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
853 skb = tcp_make_synack(sk, dst, req, NULL);
856 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
858 skb_set_queue_mapping(skb, queue_mapping);
859 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
862 err = net_xmit_eval(err);
863 if (!tcp_rsk(req)->snt_synack && !err)
864 tcp_rsk(req)->snt_synack = tcp_time_stamp;
870 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req)
872 int res = tcp_v4_send_synack(sk, NULL, req, 0, false);
875 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
880 * IPv4 request_sock destructor.
882 static void tcp_v4_reqsk_destructor(struct request_sock *req)
884 kfree(inet_rsk(req)->opt);
888 * Return true if a syncookie should be sent
890 bool tcp_syn_flood_action(struct sock *sk,
891 const struct sk_buff *skb,
894 const char *msg = "Dropping request";
895 bool want_cookie = false;
896 struct listen_sock *lopt;
900 #ifdef CONFIG_SYN_COOKIES
901 if (sysctl_tcp_syncookies) {
902 msg = "Sending cookies";
904 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
907 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
909 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
910 if (!lopt->synflood_warned) {
911 lopt->synflood_warned = 1;
912 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
913 proto, ntohs(tcp_hdr(skb)->dest), msg);
917 EXPORT_SYMBOL(tcp_syn_flood_action);
920 * Save and compile IPv4 options into the request_sock if needed.
922 static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
924 const struct ip_options *opt = &(IPCB(skb)->opt);
925 struct ip_options_rcu *dopt = NULL;
927 if (opt && opt->optlen) {
928 int opt_size = sizeof(*dopt) + opt->optlen;
930 dopt = kmalloc(opt_size, GFP_ATOMIC);
932 if (ip_options_echo(&dopt->opt, skb)) {
941 #ifdef CONFIG_TCP_MD5SIG
943 * RFC2385 MD5 checksumming requires a mapping of
944 * IP address->MD5 Key.
945 * We need to maintain these in the sk structure.
948 /* Find the Key structure for an address. */
949 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
950 const union tcp_md5_addr *addr,
953 struct tcp_sock *tp = tcp_sk(sk);
954 struct tcp_md5sig_key *key;
955 unsigned int size = sizeof(struct in_addr);
956 struct tcp_md5sig_info *md5sig;
958 /* caller either holds rcu_read_lock() or socket lock */
959 md5sig = rcu_dereference_check(tp->md5sig_info,
960 sock_owned_by_user(sk) ||
961 lockdep_is_held(&sk->sk_lock.slock));
964 #if IS_ENABLED(CONFIG_IPV6)
965 if (family == AF_INET6)
966 size = sizeof(struct in6_addr);
968 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
969 if (key->family != family)
971 if (!memcmp(&key->addr, addr, size))
976 EXPORT_SYMBOL(tcp_md5_do_lookup);
978 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
979 struct sock *addr_sk)
981 union tcp_md5_addr *addr;
983 addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
984 return tcp_md5_do_lookup(sk, addr, AF_INET);
986 EXPORT_SYMBOL(tcp_v4_md5_lookup);
988 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
989 struct request_sock *req)
991 union tcp_md5_addr *addr;
993 addr = (union tcp_md5_addr *)&inet_rsk(req)->rmt_addr;
994 return tcp_md5_do_lookup(sk, addr, AF_INET);
997 /* This can be called on a newly created socket, from other files */
998 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
999 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
1001 /* Add Key to the list */
1002 struct tcp_md5sig_key *key;
1003 struct tcp_sock *tp = tcp_sk(sk);
1004 struct tcp_md5sig_info *md5sig;
1006 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
1008 /* Pre-existing entry - just update that one. */
1009 memcpy(key->key, newkey, newkeylen);
1010 key->keylen = newkeylen;
1014 md5sig = rcu_dereference_protected(tp->md5sig_info,
1015 sock_owned_by_user(sk));
1017 md5sig = kmalloc(sizeof(*md5sig), gfp);
1021 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1022 INIT_HLIST_HEAD(&md5sig->head);
1023 rcu_assign_pointer(tp->md5sig_info, md5sig);
1026 key = sock_kmalloc(sk, sizeof(*key), gfp);
1029 if (hlist_empty(&md5sig->head) && !tcp_alloc_md5sig_pool(sk)) {
1030 sock_kfree_s(sk, key, sizeof(*key));
1034 memcpy(key->key, newkey, newkeylen);
1035 key->keylen = newkeylen;
1036 key->family = family;
1037 memcpy(&key->addr, addr,
1038 (family == AF_INET6) ? sizeof(struct in6_addr) :
1039 sizeof(struct in_addr));
1040 hlist_add_head_rcu(&key->node, &md5sig->head);
1043 EXPORT_SYMBOL(tcp_md5_do_add);
1045 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
1047 struct tcp_sock *tp = tcp_sk(sk);
1048 struct tcp_md5sig_key *key;
1049 struct tcp_md5sig_info *md5sig;
1051 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
1054 hlist_del_rcu(&key->node);
1055 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1056 kfree_rcu(key, rcu);
1057 md5sig = rcu_dereference_protected(tp->md5sig_info,
1058 sock_owned_by_user(sk));
1059 if (hlist_empty(&md5sig->head))
1060 tcp_free_md5sig_pool();
1063 EXPORT_SYMBOL(tcp_md5_do_del);
1065 static void tcp_clear_md5_list(struct sock *sk)
1067 struct tcp_sock *tp = tcp_sk(sk);
1068 struct tcp_md5sig_key *key;
1069 struct hlist_node *n;
1070 struct tcp_md5sig_info *md5sig;
1072 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1074 if (!hlist_empty(&md5sig->head))
1075 tcp_free_md5sig_pool();
1076 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1077 hlist_del_rcu(&key->node);
1078 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1079 kfree_rcu(key, rcu);
1083 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1086 struct tcp_md5sig cmd;
1087 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1089 if (optlen < sizeof(cmd))
1092 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1095 if (sin->sin_family != AF_INET)
1098 if (!cmd.tcpm_key || !cmd.tcpm_keylen)
1099 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1102 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1105 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1106 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1110 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1111 __be32 daddr, __be32 saddr, int nbytes)
1113 struct tcp4_pseudohdr *bp;
1114 struct scatterlist sg;
1116 bp = &hp->md5_blk.ip4;
1119 * 1. the TCP pseudo-header (in the order: source IP address,
1120 * destination IP address, zero-padded protocol number, and
1126 bp->protocol = IPPROTO_TCP;
1127 bp->len = cpu_to_be16(nbytes);
1129 sg_init_one(&sg, bp, sizeof(*bp));
1130 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1133 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1134 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1136 struct tcp_md5sig_pool *hp;
1137 struct hash_desc *desc;
1139 hp = tcp_get_md5sig_pool();
1141 goto clear_hash_noput;
1142 desc = &hp->md5_desc;
1144 if (crypto_hash_init(desc))
1146 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1148 if (tcp_md5_hash_header(hp, th))
1150 if (tcp_md5_hash_key(hp, key))
1152 if (crypto_hash_final(desc, md5_hash))
1155 tcp_put_md5sig_pool();
1159 tcp_put_md5sig_pool();
1161 memset(md5_hash, 0, 16);
1165 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1166 const struct sock *sk, const struct request_sock *req,
1167 const struct sk_buff *skb)
1169 struct tcp_md5sig_pool *hp;
1170 struct hash_desc *desc;
1171 const struct tcphdr *th = tcp_hdr(skb);
1172 __be32 saddr, daddr;
1175 saddr = inet_sk(sk)->inet_saddr;
1176 daddr = inet_sk(sk)->inet_daddr;
1178 saddr = inet_rsk(req)->loc_addr;
1179 daddr = inet_rsk(req)->rmt_addr;
1181 const struct iphdr *iph = ip_hdr(skb);
1186 hp = tcp_get_md5sig_pool();
1188 goto clear_hash_noput;
1189 desc = &hp->md5_desc;
1191 if (crypto_hash_init(desc))
1194 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1196 if (tcp_md5_hash_header(hp, th))
1198 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1200 if (tcp_md5_hash_key(hp, key))
1202 if (crypto_hash_final(desc, md5_hash))
1205 tcp_put_md5sig_pool();
1209 tcp_put_md5sig_pool();
1211 memset(md5_hash, 0, 16);
1214 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1216 static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1219 * This gets called for each TCP segment that arrives
1220 * so we want to be efficient.
1221 * We have 3 drop cases:
1222 * o No MD5 hash and one expected.
1223 * o MD5 hash and we're not expecting one.
1224 * o MD5 hash and its wrong.
1226 const __u8 *hash_location = NULL;
1227 struct tcp_md5sig_key *hash_expected;
1228 const struct iphdr *iph = ip_hdr(skb);
1229 const struct tcphdr *th = tcp_hdr(skb);
1231 unsigned char newhash[16];
1233 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1235 hash_location = tcp_parse_md5sig_option(th);
1237 /* We've parsed the options - do we have a hash? */
1238 if (!hash_expected && !hash_location)
1241 if (hash_expected && !hash_location) {
1242 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1246 if (!hash_expected && hash_location) {
1247 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1251 /* Okay, so this is hash_expected and hash_location -
1252 * so we need to calculate the checksum.
1254 genhash = tcp_v4_md5_hash_skb(newhash,
1258 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1259 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1260 &iph->saddr, ntohs(th->source),
1261 &iph->daddr, ntohs(th->dest),
1262 genhash ? " tcp_v4_calc_md5_hash failed"
1271 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1273 .obj_size = sizeof(struct tcp_request_sock),
1274 .rtx_syn_ack = tcp_v4_rtx_synack,
1275 .send_ack = tcp_v4_reqsk_send_ack,
1276 .destructor = tcp_v4_reqsk_destructor,
1277 .send_reset = tcp_v4_send_reset,
1278 .syn_ack_timeout = tcp_syn_ack_timeout,
1281 #ifdef CONFIG_TCP_MD5SIG
1282 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1283 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1284 .calc_md5_hash = tcp_v4_md5_hash_skb,
1288 static bool tcp_fastopen_check(struct sock *sk, struct sk_buff *skb,
1289 struct request_sock *req,
1290 struct tcp_fastopen_cookie *foc,
1291 struct tcp_fastopen_cookie *valid_foc)
1293 bool skip_cookie = false;
1294 struct fastopen_queue *fastopenq;
1296 if (likely(!fastopen_cookie_present(foc))) {
1297 /* See include/net/tcp.h for the meaning of these knobs */
1298 if ((sysctl_tcp_fastopen & TFO_SERVER_ALWAYS) ||
1299 ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_REQD) &&
1300 (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1)))
1301 skip_cookie = true; /* no cookie to validate */
1305 fastopenq = inet_csk(sk)->icsk_accept_queue.fastopenq;
1306 /* A FO option is present; bump the counter. */
1307 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVE);
1309 /* Make sure the listener has enabled fastopen, and we don't
1310 * exceed the max # of pending TFO requests allowed before trying
1311 * to validating the cookie in order to avoid burning CPU cycles
1314 * XXX (TFO) - The implication of checking the max_qlen before
1315 * processing a cookie request is that clients can't differentiate
1316 * between qlen overflow causing Fast Open to be disabled
1317 * temporarily vs a server not supporting Fast Open at all.
1319 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) == 0 ||
1320 fastopenq == NULL || fastopenq->max_qlen == 0)
1323 if (fastopenq->qlen >= fastopenq->max_qlen) {
1324 struct request_sock *req1;
1325 spin_lock(&fastopenq->lock);
1326 req1 = fastopenq->rskq_rst_head;
1327 if ((req1 == NULL) || time_after(req1->expires, jiffies)) {
1328 spin_unlock(&fastopenq->lock);
1329 NET_INC_STATS_BH(sock_net(sk),
1330 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW);
1331 /* Avoid bumping LINUX_MIB_TCPFASTOPENPASSIVEFAIL*/
1335 fastopenq->rskq_rst_head = req1->dl_next;
1337 spin_unlock(&fastopenq->lock);
1341 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1344 if (foc->len == TCP_FASTOPEN_COOKIE_SIZE) {
1345 if ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_CHKED) == 0) {
1346 tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc);
1347 if ((valid_foc->len != TCP_FASTOPEN_COOKIE_SIZE) ||
1348 memcmp(&foc->val[0], &valid_foc->val[0],
1349 TCP_FASTOPEN_COOKIE_SIZE) != 0)
1351 valid_foc->len = -1;
1353 /* Acknowledge the data received from the peer. */
1354 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1356 } else if (foc->len == 0) { /* Client requesting a cookie */
1357 tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc);
1358 NET_INC_STATS_BH(sock_net(sk),
1359 LINUX_MIB_TCPFASTOPENCOOKIEREQD);
1361 /* Client sent a cookie with wrong size. Treat it
1362 * the same as invalid and return a valid one.
1364 tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc);
1369 static int tcp_v4_conn_req_fastopen(struct sock *sk,
1370 struct sk_buff *skb,
1371 struct sk_buff *skb_synack,
1372 struct request_sock *req)
1374 struct tcp_sock *tp = tcp_sk(sk);
1375 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1376 const struct inet_request_sock *ireq = inet_rsk(req);
1380 req->num_retrans = 0;
1381 req->num_timeout = 0;
1384 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
1385 if (child == NULL) {
1386 NET_INC_STATS_BH(sock_net(sk),
1387 LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
1388 kfree_skb(skb_synack);
1391 err = ip_build_and_send_pkt(skb_synack, sk, ireq->loc_addr,
1392 ireq->rmt_addr, ireq->opt);
1393 err = net_xmit_eval(err);
1395 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1396 /* XXX (TFO) - is it ok to ignore error and continue? */
1398 spin_lock(&queue->fastopenq->lock);
1399 queue->fastopenq->qlen++;
1400 spin_unlock(&queue->fastopenq->lock);
1402 /* Initialize the child socket. Have to fix some values to take
1403 * into account the child is a Fast Open socket and is created
1404 * only out of the bits carried in the SYN packet.
1408 tp->fastopen_rsk = req;
1409 /* Do a hold on the listner sk so that if the listener is being
1410 * closed, the child that has been accepted can live on and still
1411 * access listen_lock.
1414 tcp_rsk(req)->listener = sk;
1416 /* RFC1323: The window in SYN & SYN/ACK segments is never
1417 * scaled. So correct it appropriately.
1419 tp->snd_wnd = ntohs(tcp_hdr(skb)->window);
1421 /* Activate the retrans timer so that SYNACK can be retransmitted.
1422 * The request socket is not added to the SYN table of the parent
1423 * because it's been added to the accept queue directly.
1425 inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
1426 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
1428 /* Add the child socket directly into the accept queue */
1429 inet_csk_reqsk_queue_add(sk, req, child);
1431 /* Now finish processing the fastopen child socket. */
1432 inet_csk(child)->icsk_af_ops->rebuild_header(child);
1433 tcp_init_congestion_control(child);
1434 tcp_mtup_init(child);
1435 tcp_init_buffer_space(child);
1436 tcp_init_metrics(child);
1438 /* Queue the data carried in the SYN packet. We need to first
1439 * bump skb's refcnt because the caller will attempt to free it.
1441 * XXX (TFO) - we honor a zero-payload TFO request for now.
1442 * (Any reason not to?)
1444 if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq + 1) {
1445 /* Don't queue the skb if there is no payload in SYN.
1446 * XXX (TFO) - How about SYN+FIN?
1448 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1452 __skb_pull(skb, tcp_hdr(skb)->doff * 4);
1453 skb_set_owner_r(skb, child);
1454 __skb_queue_tail(&child->sk_receive_queue, skb);
1455 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1456 tp->syn_data_acked = 1;
1458 sk->sk_data_ready(sk, 0);
1459 bh_unlock_sock(child);
1461 WARN_ON(req->sk == NULL);
1465 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1467 struct tcp_options_received tmp_opt;
1468 struct request_sock *req;
1469 struct inet_request_sock *ireq;
1470 struct tcp_sock *tp = tcp_sk(sk);
1471 struct dst_entry *dst = NULL;
1472 __be32 saddr = ip_hdr(skb)->saddr;
1473 __be32 daddr = ip_hdr(skb)->daddr;
1474 __u32 isn = TCP_SKB_CB(skb)->when;
1475 bool want_cookie = false;
1477 struct tcp_fastopen_cookie foc = { .len = -1 };
1478 struct tcp_fastopen_cookie valid_foc = { .len = -1 };
1479 struct sk_buff *skb_synack;
1482 /* Never answer to SYNs send to broadcast or multicast */
1483 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1486 /* TW buckets are converted to open requests without
1487 * limitations, they conserve resources and peer is
1488 * evidently real one.
1490 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1491 want_cookie = tcp_syn_flood_action(sk, skb, "TCP");
1496 /* Accept backlog is full. If we have already queued enough
1497 * of warm entries in syn queue, drop request. It is better than
1498 * clogging syn queue with openreqs with exponentially increasing
1501 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
1502 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1506 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1510 #ifdef CONFIG_TCP_MD5SIG
1511 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1514 tcp_clear_options(&tmp_opt);
1515 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1516 tmp_opt.user_mss = tp->rx_opt.user_mss;
1517 tcp_parse_options(skb, &tmp_opt, 0, want_cookie ? NULL : &foc);
1519 if (want_cookie && !tmp_opt.saw_tstamp)
1520 tcp_clear_options(&tmp_opt);
1522 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1523 tcp_openreq_init(req, &tmp_opt, skb);
1525 ireq = inet_rsk(req);
1526 ireq->loc_addr = daddr;
1527 ireq->rmt_addr = saddr;
1528 ireq->no_srccheck = inet_sk(sk)->transparent;
1529 ireq->opt = tcp_v4_save_options(skb);
1531 if (security_inet_conn_request(sk, skb, req))
1534 if (!want_cookie || tmp_opt.tstamp_ok)
1535 TCP_ECN_create_request(req, skb, sock_net(sk));
1538 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1539 req->cookie_ts = tmp_opt.tstamp_ok;
1541 /* VJ's idea. We save last timestamp seen
1542 * from the destination in peer table, when entering
1543 * state TIME-WAIT, and check against it before
1544 * accepting new connection request.
1546 * If "isn" is not zero, this request hit alive
1547 * timewait bucket, so that all the necessary checks
1548 * are made in the function processing timewait state.
1550 if (tmp_opt.saw_tstamp &&
1551 tcp_death_row.sysctl_tw_recycle &&
1552 (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1553 fl4.daddr == saddr) {
1554 if (!tcp_peer_is_proven(req, dst, true)) {
1555 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1556 goto drop_and_release;
1559 /* Kill the following clause, if you dislike this way. */
1560 else if (!sysctl_tcp_syncookies &&
1561 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1562 (sysctl_max_syn_backlog >> 2)) &&
1563 !tcp_peer_is_proven(req, dst, false)) {
1564 /* Without syncookies last quarter of
1565 * backlog is filled with destinations,
1566 * proven to be alive.
1567 * It means that we continue to communicate
1568 * to destinations, already remembered
1569 * to the moment of synflood.
1571 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"),
1572 &saddr, ntohs(tcp_hdr(skb)->source));
1573 goto drop_and_release;
1576 isn = tcp_v4_init_sequence(skb);
1578 tcp_rsk(req)->snt_isn = isn;
1581 dst = inet_csk_route_req(sk, &fl4, req);
1585 do_fastopen = tcp_fastopen_check(sk, skb, req, &foc, &valid_foc);
1587 /* We don't call tcp_v4_send_synack() directly because we need
1588 * to make sure a child socket can be created successfully before
1589 * sending back synack!
1591 * XXX (TFO) - Ideally one would simply call tcp_v4_send_synack()
1592 * (or better yet, call tcp_send_synack() in the child context
1593 * directly, but will have to fix bunch of other code first)
1594 * after syn_recv_sock() except one will need to first fix the
1595 * latter to remove its dependency on the current implementation
1596 * of tcp_v4_send_synack()->tcp_select_initial_window().
1598 skb_synack = tcp_make_synack(sk, dst, req,
1599 fastopen_cookie_present(&valid_foc) ? &valid_foc : NULL);
1602 __tcp_v4_send_check(skb_synack, ireq->loc_addr, ireq->rmt_addr);
1603 skb_set_queue_mapping(skb_synack, skb_get_queue_mapping(skb));
1607 if (likely(!do_fastopen)) {
1609 err = ip_build_and_send_pkt(skb_synack, sk, ireq->loc_addr,
1610 ireq->rmt_addr, ireq->opt);
1611 err = net_xmit_eval(err);
1612 if (err || want_cookie)
1615 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1616 tcp_rsk(req)->listener = NULL;
1617 /* Add the request_sock to the SYN table */
1618 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1619 if (fastopen_cookie_present(&foc) && foc.len != 0)
1620 NET_INC_STATS_BH(sock_net(sk),
1621 LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
1622 } else if (tcp_v4_conn_req_fastopen(sk, skb, skb_synack, req))
1632 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1635 EXPORT_SYMBOL(tcp_v4_conn_request);
1639 * The three way handshake has completed - we got a valid synack -
1640 * now create the new socket.
1642 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1643 struct request_sock *req,
1644 struct dst_entry *dst)
1646 struct inet_request_sock *ireq;
1647 struct inet_sock *newinet;
1648 struct tcp_sock *newtp;
1650 #ifdef CONFIG_TCP_MD5SIG
1651 struct tcp_md5sig_key *key;
1653 struct ip_options_rcu *inet_opt;
1655 if (sk_acceptq_is_full(sk))
1658 newsk = tcp_create_openreq_child(sk, req, skb);
1662 newsk->sk_gso_type = SKB_GSO_TCPV4;
1663 inet_sk_rx_dst_set(newsk, skb);
1665 newtp = tcp_sk(newsk);
1666 newinet = inet_sk(newsk);
1667 ireq = inet_rsk(req);
1668 newinet->inet_daddr = ireq->rmt_addr;
1669 newinet->inet_rcv_saddr = ireq->loc_addr;
1670 newinet->inet_saddr = ireq->loc_addr;
1671 inet_opt = ireq->opt;
1672 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1674 newinet->mc_index = inet_iif(skb);
1675 newinet->mc_ttl = ip_hdr(skb)->ttl;
1676 newinet->rcv_tos = ip_hdr(skb)->tos;
1677 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1679 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1680 newinet->inet_id = newtp->write_seq ^ jiffies;
1683 dst = inet_csk_route_child_sock(sk, newsk, req);
1687 /* syncookie case : see end of cookie_v4_check() */
1689 sk_setup_caps(newsk, dst);
1691 tcp_mtup_init(newsk);
1692 tcp_sync_mss(newsk, dst_mtu(dst));
1693 newtp->advmss = dst_metric_advmss(dst);
1694 if (tcp_sk(sk)->rx_opt.user_mss &&
1695 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1696 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1698 tcp_initialize_rcv_mss(newsk);
1699 tcp_synack_rtt_meas(newsk, req);
1700 newtp->total_retrans = req->num_retrans;
1702 #ifdef CONFIG_TCP_MD5SIG
1703 /* Copy over the MD5 key from the original socket */
1704 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1708 * We're using one, so create a matching key
1709 * on the newsk structure. If we fail to get
1710 * memory, then we end up not copying the key
1713 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1714 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1715 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1719 if (__inet_inherit_port(sk, newsk) < 0)
1721 __inet_hash_nolisten(newsk, NULL);
1726 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1730 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1733 inet_csk_prepare_forced_close(newsk);
1737 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1739 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1741 struct tcphdr *th = tcp_hdr(skb);
1742 const struct iphdr *iph = ip_hdr(skb);
1744 struct request_sock **prev;
1745 /* Find possible connection requests. */
1746 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1747 iph->saddr, iph->daddr);
1749 return tcp_check_req(sk, skb, req, prev, false);
1751 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1752 th->source, iph->daddr, th->dest, inet_iif(skb));
1755 if (nsk->sk_state != TCP_TIME_WAIT) {
1759 inet_twsk_put(inet_twsk(nsk));
1763 #ifdef CONFIG_SYN_COOKIES
1765 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1770 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1772 const struct iphdr *iph = ip_hdr(skb);
1774 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1775 if (!tcp_v4_check(skb->len, iph->saddr,
1776 iph->daddr, skb->csum)) {
1777 skb->ip_summed = CHECKSUM_UNNECESSARY;
1782 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1783 skb->len, IPPROTO_TCP, 0);
1785 if (skb->len <= 76) {
1786 return __skb_checksum_complete(skb);
1792 /* The socket must have it's spinlock held when we get
1795 * We have a potential double-lock case here, so even when
1796 * doing backlog processing we use the BH locking scheme.
1797 * This is because we cannot sleep with the original spinlock
1800 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1803 #ifdef CONFIG_TCP_MD5SIG
1805 * We really want to reject the packet as early as possible
1807 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1808 * o There is an MD5 option and we're not expecting one
1810 if (tcp_v4_inbound_md5_hash(sk, skb))
1814 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1815 struct dst_entry *dst = sk->sk_rx_dst;
1817 sock_rps_save_rxhash(sk, skb);
1819 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1820 dst->ops->check(dst, 0) == NULL) {
1822 sk->sk_rx_dst = NULL;
1825 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1832 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1835 if (sk->sk_state == TCP_LISTEN) {
1836 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1841 sock_rps_save_rxhash(nsk, skb);
1842 if (tcp_child_process(sk, nsk, skb)) {
1849 sock_rps_save_rxhash(sk, skb);
1851 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1858 tcp_v4_send_reset(rsk, skb);
1861 /* Be careful here. If this function gets more complicated and
1862 * gcc suffers from register pressure on the x86, sk (in %ebx)
1863 * might be destroyed here. This current version compiles correctly,
1864 * but you have been warned.
1869 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1872 EXPORT_SYMBOL(tcp_v4_do_rcv);
1874 void tcp_v4_early_demux(struct sk_buff *skb)
1876 const struct iphdr *iph;
1877 const struct tcphdr *th;
1880 if (skb->pkt_type != PACKET_HOST)
1883 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1889 if (th->doff < sizeof(struct tcphdr) / 4)
1892 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1893 iph->saddr, th->source,
1894 iph->daddr, ntohs(th->dest),
1898 skb->destructor = sock_edemux;
1899 if (sk->sk_state != TCP_TIME_WAIT) {
1900 struct dst_entry *dst = sk->sk_rx_dst;
1903 dst = dst_check(dst, 0);
1905 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1906 skb_dst_set_noref(skb, dst);
1911 /* Packet is added to VJ-style prequeue for processing in process
1912 * context, if a reader task is waiting. Apparently, this exciting
1913 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1914 * failed somewhere. Latency? Burstiness? Well, at least now we will
1915 * see, why it failed. 8)8) --ANK
1918 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1920 struct tcp_sock *tp = tcp_sk(sk);
1922 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1925 if (skb->len <= tcp_hdrlen(skb) &&
1926 skb_queue_len(&tp->ucopy.prequeue) == 0)
1929 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1930 tp->ucopy.memory += skb->truesize;
1931 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1932 struct sk_buff *skb1;
1934 BUG_ON(sock_owned_by_user(sk));
1936 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1937 sk_backlog_rcv(sk, skb1);
1938 NET_INC_STATS_BH(sock_net(sk),
1939 LINUX_MIB_TCPPREQUEUEDROPPED);
1942 tp->ucopy.memory = 0;
1943 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1944 wake_up_interruptible_sync_poll(sk_sleep(sk),
1945 POLLIN | POLLRDNORM | POLLRDBAND);
1946 if (!inet_csk_ack_scheduled(sk))
1947 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1948 (3 * tcp_rto_min(sk)) / 4,
1953 EXPORT_SYMBOL(tcp_prequeue);
1959 int tcp_v4_rcv(struct sk_buff *skb)
1961 const struct iphdr *iph;
1962 const struct tcphdr *th;
1965 struct net *net = dev_net(skb->dev);
1967 if (skb->pkt_type != PACKET_HOST)
1970 /* Count it even if it's bad */
1971 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1973 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1978 if (th->doff < sizeof(struct tcphdr) / 4)
1980 if (!pskb_may_pull(skb, th->doff * 4))
1983 /* An explanation is required here, I think.
1984 * Packet length and doff are validated by header prediction,
1985 * provided case of th->doff==0 is eliminated.
1986 * So, we defer the checks. */
1987 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1992 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1993 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1994 skb->len - th->doff * 4);
1995 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1996 TCP_SKB_CB(skb)->when = 0;
1997 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1998 TCP_SKB_CB(skb)->sacked = 0;
2000 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
2005 if (sk->sk_state == TCP_TIME_WAIT)
2008 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
2009 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
2010 goto discard_and_relse;
2013 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2014 goto discard_and_relse;
2017 if (sk_filter(sk, skb))
2018 goto discard_and_relse;
2022 bh_lock_sock_nested(sk);
2024 if (!sock_owned_by_user(sk)) {
2025 #ifdef CONFIG_NET_DMA
2026 struct tcp_sock *tp = tcp_sk(sk);
2027 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
2028 tp->ucopy.dma_chan = net_dma_find_channel();
2029 if (tp->ucopy.dma_chan)
2030 ret = tcp_v4_do_rcv(sk, skb);
2034 if (!tcp_prequeue(sk, skb))
2035 ret = tcp_v4_do_rcv(sk, skb);
2037 } else if (unlikely(sk_add_backlog(sk, skb,
2038 sk->sk_rcvbuf + sk->sk_sndbuf))) {
2040 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
2041 goto discard_and_relse;
2050 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2053 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
2055 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
2057 tcp_v4_send_reset(NULL, skb);
2061 /* Discard frame. */
2070 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2071 inet_twsk_put(inet_twsk(sk));
2075 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
2076 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
2077 inet_twsk_put(inet_twsk(sk));
2080 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2082 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
2084 iph->saddr, th->source,
2085 iph->daddr, th->dest,
2088 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
2089 inet_twsk_put(inet_twsk(sk));
2093 /* Fall through to ACK */
2096 tcp_v4_timewait_ack(sk, skb);
2100 case TCP_TW_SUCCESS:;
2105 static struct timewait_sock_ops tcp_timewait_sock_ops = {
2106 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
2107 .twsk_unique = tcp_twsk_unique,
2108 .twsk_destructor= tcp_twsk_destructor,
2111 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2113 struct dst_entry *dst = skb_dst(skb);
2116 sk->sk_rx_dst = dst;
2117 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
2119 EXPORT_SYMBOL(inet_sk_rx_dst_set);
2121 const struct inet_connection_sock_af_ops ipv4_specific = {
2122 .queue_xmit = ip_queue_xmit,
2123 .send_check = tcp_v4_send_check,
2124 .rebuild_header = inet_sk_rebuild_header,
2125 .sk_rx_dst_set = inet_sk_rx_dst_set,
2126 .conn_request = tcp_v4_conn_request,
2127 .syn_recv_sock = tcp_v4_syn_recv_sock,
2128 .net_header_len = sizeof(struct iphdr),
2129 .setsockopt = ip_setsockopt,
2130 .getsockopt = ip_getsockopt,
2131 .addr2sockaddr = inet_csk_addr2sockaddr,
2132 .sockaddr_len = sizeof(struct sockaddr_in),
2133 .bind_conflict = inet_csk_bind_conflict,
2134 #ifdef CONFIG_COMPAT
2135 .compat_setsockopt = compat_ip_setsockopt,
2136 .compat_getsockopt = compat_ip_getsockopt,
2139 EXPORT_SYMBOL(ipv4_specific);
2141 #ifdef CONFIG_TCP_MD5SIG
2142 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2143 .md5_lookup = tcp_v4_md5_lookup,
2144 .calc_md5_hash = tcp_v4_md5_hash_skb,
2145 .md5_parse = tcp_v4_parse_md5_keys,
2149 /* NOTE: A lot of things set to zero explicitly by call to
2150 * sk_alloc() so need not be done here.
2152 static int tcp_v4_init_sock(struct sock *sk)
2154 struct inet_connection_sock *icsk = inet_csk(sk);
2158 icsk->icsk_af_ops = &ipv4_specific;
2160 #ifdef CONFIG_TCP_MD5SIG
2161 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2167 void tcp_v4_destroy_sock(struct sock *sk)
2169 struct tcp_sock *tp = tcp_sk(sk);
2171 tcp_clear_xmit_timers(sk);
2173 tcp_cleanup_congestion_control(sk);
2175 /* Cleanup up the write buffer. */
2176 tcp_write_queue_purge(sk);
2178 /* Cleans up our, hopefully empty, out_of_order_queue. */
2179 __skb_queue_purge(&tp->out_of_order_queue);
2181 #ifdef CONFIG_TCP_MD5SIG
2182 /* Clean up the MD5 key list, if any */
2183 if (tp->md5sig_info) {
2184 tcp_clear_md5_list(sk);
2185 kfree_rcu(tp->md5sig_info, rcu);
2186 tp->md5sig_info = NULL;
2190 #ifdef CONFIG_NET_DMA
2191 /* Cleans up our sk_async_wait_queue */
2192 __skb_queue_purge(&sk->sk_async_wait_queue);
2195 /* Clean prequeue, it must be empty really */
2196 __skb_queue_purge(&tp->ucopy.prequeue);
2198 /* Clean up a referenced TCP bind bucket. */
2199 if (inet_csk(sk)->icsk_bind_hash)
2202 BUG_ON(tp->fastopen_rsk != NULL);
2204 /* If socket is aborted during connect operation */
2205 tcp_free_fastopen_req(tp);
2207 sk_sockets_allocated_dec(sk);
2208 sock_release_memcg(sk);
2210 EXPORT_SYMBOL(tcp_v4_destroy_sock);
2212 #ifdef CONFIG_PROC_FS
2213 /* Proc filesystem TCP sock list dumping. */
2215 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
2217 return hlist_nulls_empty(head) ? NULL :
2218 list_entry(head->first, struct inet_timewait_sock, tw_node);
2221 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
2223 return !is_a_nulls(tw->tw_node.next) ?
2224 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
2228 * Get next listener socket follow cur. If cur is NULL, get first socket
2229 * starting from bucket given in st->bucket; when st->bucket is zero the
2230 * very first socket in the hash table is returned.
2232 static void *listening_get_next(struct seq_file *seq, void *cur)
2234 struct inet_connection_sock *icsk;
2235 struct hlist_nulls_node *node;
2236 struct sock *sk = cur;
2237 struct inet_listen_hashbucket *ilb;
2238 struct tcp_iter_state *st = seq->private;
2239 struct net *net = seq_file_net(seq);
2242 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2243 spin_lock_bh(&ilb->lock);
2244 sk = sk_nulls_head(&ilb->head);
2248 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2252 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2253 struct request_sock *req = cur;
2255 icsk = inet_csk(st->syn_wait_sk);
2259 if (req->rsk_ops->family == st->family) {
2265 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2268 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2270 sk = sk_nulls_next(st->syn_wait_sk);
2271 st->state = TCP_SEQ_STATE_LISTENING;
2272 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2274 icsk = inet_csk(sk);
2275 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2276 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2278 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2279 sk = sk_nulls_next(sk);
2282 sk_nulls_for_each_from(sk, node) {
2283 if (!net_eq(sock_net(sk), net))
2285 if (sk->sk_family == st->family) {
2289 icsk = inet_csk(sk);
2290 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2291 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2293 st->uid = sock_i_uid(sk);
2294 st->syn_wait_sk = sk;
2295 st->state = TCP_SEQ_STATE_OPENREQ;
2299 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2301 spin_unlock_bh(&ilb->lock);
2303 if (++st->bucket < INET_LHTABLE_SIZE) {
2304 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2305 spin_lock_bh(&ilb->lock);
2306 sk = sk_nulls_head(&ilb->head);
2314 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2316 struct tcp_iter_state *st = seq->private;
2321 rc = listening_get_next(seq, NULL);
2323 while (rc && *pos) {
2324 rc = listening_get_next(seq, rc);
2330 static inline bool empty_bucket(struct tcp_iter_state *st)
2332 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2333 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2337 * Get first established socket starting from bucket given in st->bucket.
2338 * If st->bucket is zero, the very first socket in the hash is returned.
2340 static void *established_get_first(struct seq_file *seq)
2342 struct tcp_iter_state *st = seq->private;
2343 struct net *net = seq_file_net(seq);
2347 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2349 struct hlist_nulls_node *node;
2350 struct inet_timewait_sock *tw;
2351 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2353 /* Lockless fast path for the common case of empty buckets */
2354 if (empty_bucket(st))
2358 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2359 if (sk->sk_family != st->family ||
2360 !net_eq(sock_net(sk), net)) {
2366 st->state = TCP_SEQ_STATE_TIME_WAIT;
2367 inet_twsk_for_each(tw, node,
2368 &tcp_hashinfo.ehash[st->bucket].twchain) {
2369 if (tw->tw_family != st->family ||
2370 !net_eq(twsk_net(tw), net)) {
2376 spin_unlock_bh(lock);
2377 st->state = TCP_SEQ_STATE_ESTABLISHED;
2383 static void *established_get_next(struct seq_file *seq, void *cur)
2385 struct sock *sk = cur;
2386 struct inet_timewait_sock *tw;
2387 struct hlist_nulls_node *node;
2388 struct tcp_iter_state *st = seq->private;
2389 struct net *net = seq_file_net(seq);
2394 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2398 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2405 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2406 st->state = TCP_SEQ_STATE_ESTABLISHED;
2408 /* Look for next non empty bucket */
2410 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2413 if (st->bucket > tcp_hashinfo.ehash_mask)
2416 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2417 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2419 sk = sk_nulls_next(sk);
2421 sk_nulls_for_each_from(sk, node) {
2422 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2426 st->state = TCP_SEQ_STATE_TIME_WAIT;
2427 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2435 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2437 struct tcp_iter_state *st = seq->private;
2441 rc = established_get_first(seq);
2444 rc = established_get_next(seq, rc);
2450 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2453 struct tcp_iter_state *st = seq->private;
2455 st->state = TCP_SEQ_STATE_LISTENING;
2456 rc = listening_get_idx(seq, &pos);
2459 st->state = TCP_SEQ_STATE_ESTABLISHED;
2460 rc = established_get_idx(seq, pos);
2466 static void *tcp_seek_last_pos(struct seq_file *seq)
2468 struct tcp_iter_state *st = seq->private;
2469 int offset = st->offset;
2470 int orig_num = st->num;
2473 switch (st->state) {
2474 case TCP_SEQ_STATE_OPENREQ:
2475 case TCP_SEQ_STATE_LISTENING:
2476 if (st->bucket >= INET_LHTABLE_SIZE)
2478 st->state = TCP_SEQ_STATE_LISTENING;
2479 rc = listening_get_next(seq, NULL);
2480 while (offset-- && rc)
2481 rc = listening_get_next(seq, rc);
2486 case TCP_SEQ_STATE_ESTABLISHED:
2487 case TCP_SEQ_STATE_TIME_WAIT:
2488 st->state = TCP_SEQ_STATE_ESTABLISHED;
2489 if (st->bucket > tcp_hashinfo.ehash_mask)
2491 rc = established_get_first(seq);
2492 while (offset-- && rc)
2493 rc = established_get_next(seq, rc);
2501 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2503 struct tcp_iter_state *st = seq->private;
2506 if (*pos && *pos == st->last_pos) {
2507 rc = tcp_seek_last_pos(seq);
2512 st->state = TCP_SEQ_STATE_LISTENING;
2516 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2519 st->last_pos = *pos;
2523 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2525 struct tcp_iter_state *st = seq->private;
2528 if (v == SEQ_START_TOKEN) {
2529 rc = tcp_get_idx(seq, 0);
2533 switch (st->state) {
2534 case TCP_SEQ_STATE_OPENREQ:
2535 case TCP_SEQ_STATE_LISTENING:
2536 rc = listening_get_next(seq, v);
2538 st->state = TCP_SEQ_STATE_ESTABLISHED;
2541 rc = established_get_first(seq);
2544 case TCP_SEQ_STATE_ESTABLISHED:
2545 case TCP_SEQ_STATE_TIME_WAIT:
2546 rc = established_get_next(seq, v);
2551 st->last_pos = *pos;
2555 static void tcp_seq_stop(struct seq_file *seq, void *v)
2557 struct tcp_iter_state *st = seq->private;
2559 switch (st->state) {
2560 case TCP_SEQ_STATE_OPENREQ:
2562 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2563 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2565 case TCP_SEQ_STATE_LISTENING:
2566 if (v != SEQ_START_TOKEN)
2567 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2569 case TCP_SEQ_STATE_TIME_WAIT:
2570 case TCP_SEQ_STATE_ESTABLISHED:
2572 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2577 int tcp_seq_open(struct inode *inode, struct file *file)
2579 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2580 struct tcp_iter_state *s;
2583 err = seq_open_net(inode, file, &afinfo->seq_ops,
2584 sizeof(struct tcp_iter_state));
2588 s = ((struct seq_file *)file->private_data)->private;
2589 s->family = afinfo->family;
2593 EXPORT_SYMBOL(tcp_seq_open);
2595 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2598 struct proc_dir_entry *p;
2600 afinfo->seq_ops.start = tcp_seq_start;
2601 afinfo->seq_ops.next = tcp_seq_next;
2602 afinfo->seq_ops.stop = tcp_seq_stop;
2604 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2605 afinfo->seq_fops, afinfo);
2610 EXPORT_SYMBOL(tcp_proc_register);
2612 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2614 remove_proc_entry(afinfo->name, net->proc_net);
2616 EXPORT_SYMBOL(tcp_proc_unregister);
2618 static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2619 struct seq_file *f, int i, kuid_t uid, int *len)
2621 const struct inet_request_sock *ireq = inet_rsk(req);
2622 long delta = req->expires - jiffies;
2624 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2625 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2628 ntohs(inet_sk(sk)->inet_sport),
2630 ntohs(ireq->rmt_port),
2632 0, 0, /* could print option size, but that is af dependent. */
2633 1, /* timers active (only the expire timer) */
2634 jiffies_delta_to_clock_t(delta),
2636 from_kuid_munged(seq_user_ns(f), uid),
2637 0, /* non standard timer */
2638 0, /* open_requests have no inode */
2639 atomic_read(&sk->sk_refcnt),
2644 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2647 unsigned long timer_expires;
2648 const struct tcp_sock *tp = tcp_sk(sk);
2649 const struct inet_connection_sock *icsk = inet_csk(sk);
2650 const struct inet_sock *inet = inet_sk(sk);
2651 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2652 __be32 dest = inet->inet_daddr;
2653 __be32 src = inet->inet_rcv_saddr;
2654 __u16 destp = ntohs(inet->inet_dport);
2655 __u16 srcp = ntohs(inet->inet_sport);
2658 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2659 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2660 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2662 timer_expires = icsk->icsk_timeout;
2663 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2665 timer_expires = icsk->icsk_timeout;
2666 } else if (timer_pending(&sk->sk_timer)) {
2668 timer_expires = sk->sk_timer.expires;
2671 timer_expires = jiffies;
2674 if (sk->sk_state == TCP_LISTEN)
2675 rx_queue = sk->sk_ack_backlog;
2678 * because we dont lock socket, we might find a transient negative value
2680 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2682 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2683 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2684 i, src, srcp, dest, destp, sk->sk_state,
2685 tp->write_seq - tp->snd_una,
2688 jiffies_delta_to_clock_t(timer_expires - jiffies),
2689 icsk->icsk_retransmits,
2690 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2691 icsk->icsk_probes_out,
2693 atomic_read(&sk->sk_refcnt), sk,
2694 jiffies_to_clock_t(icsk->icsk_rto),
2695 jiffies_to_clock_t(icsk->icsk_ack.ato),
2696 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2698 sk->sk_state == TCP_LISTEN ?
2699 (fastopenq ? fastopenq->max_qlen : 0) :
2700 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh),
2704 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2705 struct seq_file *f, int i, int *len)
2709 long delta = tw->tw_ttd - jiffies;
2711 dest = tw->tw_daddr;
2712 src = tw->tw_rcv_saddr;
2713 destp = ntohs(tw->tw_dport);
2714 srcp = ntohs(tw->tw_sport);
2716 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2717 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2718 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2719 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2720 atomic_read(&tw->tw_refcnt), tw, len);
2725 static int tcp4_seq_show(struct seq_file *seq, void *v)
2727 struct tcp_iter_state *st;
2730 if (v == SEQ_START_TOKEN) {
2731 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2732 " sl local_address rem_address st tx_queue "
2733 "rx_queue tr tm->when retrnsmt uid timeout "
2739 switch (st->state) {
2740 case TCP_SEQ_STATE_LISTENING:
2741 case TCP_SEQ_STATE_ESTABLISHED:
2742 get_tcp4_sock(v, seq, st->num, &len);
2744 case TCP_SEQ_STATE_OPENREQ:
2745 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2747 case TCP_SEQ_STATE_TIME_WAIT:
2748 get_timewait4_sock(v, seq, st->num, &len);
2751 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2756 static const struct file_operations tcp_afinfo_seq_fops = {
2757 .owner = THIS_MODULE,
2758 .open = tcp_seq_open,
2760 .llseek = seq_lseek,
2761 .release = seq_release_net
2764 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2767 .seq_fops = &tcp_afinfo_seq_fops,
2769 .show = tcp4_seq_show,
2773 static int __net_init tcp4_proc_init_net(struct net *net)
2775 return tcp_proc_register(net, &tcp4_seq_afinfo);
2778 static void __net_exit tcp4_proc_exit_net(struct net *net)
2780 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2783 static struct pernet_operations tcp4_net_ops = {
2784 .init = tcp4_proc_init_net,
2785 .exit = tcp4_proc_exit_net,
2788 int __init tcp4_proc_init(void)
2790 return register_pernet_subsys(&tcp4_net_ops);
2793 void tcp4_proc_exit(void)
2795 unregister_pernet_subsys(&tcp4_net_ops);
2797 #endif /* CONFIG_PROC_FS */
2799 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2801 const struct iphdr *iph = skb_gro_network_header(skb);
2805 switch (skb->ip_summed) {
2806 case CHECKSUM_COMPLETE:
2807 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2809 skb->ip_summed = CHECKSUM_UNNECESSARY;
2813 NAPI_GRO_CB(skb)->flush = 1;
2817 wsum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
2818 skb_gro_len(skb), IPPROTO_TCP, 0);
2819 sum = csum_fold(skb_checksum(skb,
2820 skb_gro_offset(skb),
2826 skb->ip_summed = CHECKSUM_UNNECESSARY;
2830 return tcp_gro_receive(head, skb);
2833 int tcp4_gro_complete(struct sk_buff *skb)
2835 const struct iphdr *iph = ip_hdr(skb);
2836 struct tcphdr *th = tcp_hdr(skb);
2838 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2839 iph->saddr, iph->daddr, 0);
2840 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2842 return tcp_gro_complete(skb);
2845 struct proto tcp_prot = {
2847 .owner = THIS_MODULE,
2849 .connect = tcp_v4_connect,
2850 .disconnect = tcp_disconnect,
2851 .accept = inet_csk_accept,
2853 .init = tcp_v4_init_sock,
2854 .destroy = tcp_v4_destroy_sock,
2855 .shutdown = tcp_shutdown,
2856 .setsockopt = tcp_setsockopt,
2857 .getsockopt = tcp_getsockopt,
2858 .recvmsg = tcp_recvmsg,
2859 .sendmsg = tcp_sendmsg,
2860 .sendpage = tcp_sendpage,
2861 .backlog_rcv = tcp_v4_do_rcv,
2862 .release_cb = tcp_release_cb,
2863 .mtu_reduced = tcp_v4_mtu_reduced,
2865 .unhash = inet_unhash,
2866 .get_port = inet_csk_get_port,
2867 .enter_memory_pressure = tcp_enter_memory_pressure,
2868 .sockets_allocated = &tcp_sockets_allocated,
2869 .orphan_count = &tcp_orphan_count,
2870 .memory_allocated = &tcp_memory_allocated,
2871 .memory_pressure = &tcp_memory_pressure,
2872 .sysctl_wmem = sysctl_tcp_wmem,
2873 .sysctl_rmem = sysctl_tcp_rmem,
2874 .max_header = MAX_TCP_HEADER,
2875 .obj_size = sizeof(struct tcp_sock),
2876 .slab_flags = SLAB_DESTROY_BY_RCU,
2877 .twsk_prot = &tcp_timewait_sock_ops,
2878 .rsk_prot = &tcp_request_sock_ops,
2879 .h.hashinfo = &tcp_hashinfo,
2880 .no_autobind = true,
2881 #ifdef CONFIG_COMPAT
2882 .compat_setsockopt = compat_tcp_setsockopt,
2883 .compat_getsockopt = compat_tcp_getsockopt,
2885 #ifdef CONFIG_MEMCG_KMEM
2886 .init_cgroup = tcp_init_cgroup,
2887 .destroy_cgroup = tcp_destroy_cgroup,
2888 .proto_cgroup = tcp_proto_cgroup,
2891 EXPORT_SYMBOL(tcp_prot);
2893 static int __net_init tcp_sk_init(struct net *net)
2895 net->ipv4.sysctl_tcp_ecn = 2;
2899 static void __net_exit tcp_sk_exit(struct net *net)
2903 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2905 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2908 static struct pernet_operations __net_initdata tcp_sk_ops = {
2909 .init = tcp_sk_init,
2910 .exit = tcp_sk_exit,
2911 .exit_batch = tcp_sk_exit_batch,
2914 void __init tcp_v4_init(void)
2916 inet_hashinfo_init(&tcp_hashinfo);
2917 if (register_pernet_subsys(&tcp_sk_ops))
2918 panic("Failed to create the TCP control socket.\n");