inet: add RCU protection to inet->opt
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / ipv4 / tcp_ipv4.c
blobd60732fe5f21ed9a41b5ac8f9675c06a7aaab556
1 /*
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
11 * code split from:
12 * linux/ipv4/tcp.c
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.
25 * Changes:
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
34 * ACK bit.
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
45 * coma.
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.
54 #include <linux/bottom_half.h>
55 #include <linux/types.h>
56 #include <linux/fcntl.h>
57 #include <linux/module.h>
58 #include <linux/random.h>
59 #include <linux/cache.h>
60 #include <linux/jhash.h>
61 #include <linux/init.h>
62 #include <linux/times.h>
63 #include <linux/slab.h>
65 #include <net/net_namespace.h>
66 #include <net/icmp.h>
67 #include <net/inet_hashtables.h>
68 #include <net/tcp.h>
69 #include <net/transp_v6.h>
70 #include <net/ipv6.h>
71 #include <net/inet_common.h>
72 #include <net/timewait_sock.h>
73 #include <net/xfrm.h>
74 #include <net/netdma.h>
76 #include <linux/inet.h>
77 #include <linux/ipv6.h>
78 #include <linux/stddef.h>
79 #include <linux/proc_fs.h>
80 #include <linux/seq_file.h>
82 #include <linux/crypto.h>
83 #include <linux/scatterlist.h>
85 int sysctl_tcp_tw_reuse __read_mostly;
86 int sysctl_tcp_low_latency __read_mostly;
87 EXPORT_SYMBOL(sysctl_tcp_low_latency);
90 #ifdef CONFIG_TCP_MD5SIG
91 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
92 __be32 addr);
93 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
94 __be32 daddr, __be32 saddr, struct tcphdr *th);
95 #else
96 static inline
97 struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
99 return NULL;
101 #endif
103 struct inet_hashinfo tcp_hashinfo;
104 EXPORT_SYMBOL(tcp_hashinfo);
106 static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
108 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
109 ip_hdr(skb)->saddr,
110 tcp_hdr(skb)->dest,
111 tcp_hdr(skb)->source);
114 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
116 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
117 struct tcp_sock *tp = tcp_sk(sk);
119 /* With PAWS, it is safe from the viewpoint
120 of data integrity. Even without PAWS it is safe provided sequence
121 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
123 Actually, the idea is close to VJ's one, only timestamp cache is
124 held not per host, but per port pair and TW bucket is used as state
125 holder.
127 If TW bucket has been already destroyed we fall back to VJ's scheme
128 and use initial timestamp retrieved from peer table.
130 if (tcptw->tw_ts_recent_stamp &&
131 (twp == NULL || (sysctl_tcp_tw_reuse &&
132 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
133 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
134 if (tp->write_seq == 0)
135 tp->write_seq = 1;
136 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
137 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
138 sock_hold(sktw);
139 return 1;
142 return 0;
144 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
146 /* This will initiate an outgoing connection. */
147 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
149 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
150 struct inet_sock *inet = inet_sk(sk);
151 struct tcp_sock *tp = tcp_sk(sk);
152 __be16 orig_sport, orig_dport;
153 __be32 daddr, nexthop;
154 struct flowi4 fl4;
155 struct rtable *rt;
156 int err;
157 struct ip_options_rcu *inet_opt;
159 if (addr_len < sizeof(struct sockaddr_in))
160 return -EINVAL;
162 if (usin->sin_family != AF_INET)
163 return -EAFNOSUPPORT;
165 nexthop = daddr = usin->sin_addr.s_addr;
166 inet_opt = rcu_dereference_protected(inet->inet_opt,
167 sock_owned_by_user(sk));
168 if (inet_opt && inet_opt->opt.srr) {
169 if (!daddr)
170 return -EINVAL;
171 nexthop = inet_opt->opt.faddr;
174 orig_sport = inet->inet_sport;
175 orig_dport = usin->sin_port;
176 rt = ip_route_connect(&fl4, nexthop, inet->inet_saddr,
177 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
178 IPPROTO_TCP,
179 orig_sport, orig_dport, sk, true);
180 if (IS_ERR(rt)) {
181 err = PTR_ERR(rt);
182 if (err == -ENETUNREACH)
183 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
184 return err;
187 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
188 ip_rt_put(rt);
189 return -ENETUNREACH;
192 if (!inet_opt || !inet_opt->opt.srr)
193 daddr = rt->rt_dst;
195 if (!inet->inet_saddr)
196 inet->inet_saddr = rt->rt_src;
197 inet->inet_rcv_saddr = inet->inet_saddr;
199 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
200 /* Reset inherited state */
201 tp->rx_opt.ts_recent = 0;
202 tp->rx_opt.ts_recent_stamp = 0;
203 tp->write_seq = 0;
206 if (tcp_death_row.sysctl_tw_recycle &&
207 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
208 struct inet_peer *peer = rt_get_peer(rt);
210 * VJ's idea. We save last timestamp seen from
211 * the destination in peer table, when entering state
212 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
213 * when trying new connection.
215 if (peer) {
216 inet_peer_refcheck(peer);
217 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
218 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
219 tp->rx_opt.ts_recent = peer->tcp_ts;
224 inet->inet_dport = usin->sin_port;
225 inet->inet_daddr = daddr;
227 inet_csk(sk)->icsk_ext_hdr_len = 0;
228 if (inet_opt)
229 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
231 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
233 /* Socket identity is still unknown (sport may be zero).
234 * However we set state to SYN-SENT and not releasing socket
235 * lock select source port, enter ourselves into the hash tables and
236 * complete initialization after this.
238 tcp_set_state(sk, TCP_SYN_SENT);
239 err = inet_hash_connect(&tcp_death_row, sk);
240 if (err)
241 goto failure;
243 rt = ip_route_newports(&fl4, rt, orig_sport, orig_dport,
244 inet->inet_sport, inet->inet_dport, sk);
245 if (IS_ERR(rt)) {
246 err = PTR_ERR(rt);
247 rt = NULL;
248 goto failure;
250 /* OK, now commit destination to socket. */
251 sk->sk_gso_type = SKB_GSO_TCPV4;
252 sk_setup_caps(sk, &rt->dst);
254 if (!tp->write_seq)
255 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
256 inet->inet_daddr,
257 inet->inet_sport,
258 usin->sin_port);
260 inet->inet_id = tp->write_seq ^ jiffies;
262 err = tcp_connect(sk);
263 rt = NULL;
264 if (err)
265 goto failure;
267 return 0;
269 failure:
271 * This unhashes the socket and releases the local port,
272 * if necessary.
274 tcp_set_state(sk, TCP_CLOSE);
275 ip_rt_put(rt);
276 sk->sk_route_caps = 0;
277 inet->inet_dport = 0;
278 return err;
280 EXPORT_SYMBOL(tcp_v4_connect);
283 * This routine does path mtu discovery as defined in RFC1191.
285 static void do_pmtu_discovery(struct sock *sk, const struct iphdr *iph, u32 mtu)
287 struct dst_entry *dst;
288 struct inet_sock *inet = inet_sk(sk);
290 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
291 * send out by Linux are always <576bytes so they should go through
292 * unfragmented).
294 if (sk->sk_state == TCP_LISTEN)
295 return;
297 /* We don't check in the destentry if pmtu discovery is forbidden
298 * on this route. We just assume that no packet_to_big packets
299 * are send back when pmtu discovery is not active.
300 * There is a small race when the user changes this flag in the
301 * route, but I think that's acceptable.
303 if ((dst = __sk_dst_check(sk, 0)) == NULL)
304 return;
306 dst->ops->update_pmtu(dst, mtu);
308 /* Something is about to be wrong... Remember soft error
309 * for the case, if this connection will not able to recover.
311 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
312 sk->sk_err_soft = EMSGSIZE;
314 mtu = dst_mtu(dst);
316 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
317 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
318 tcp_sync_mss(sk, mtu);
320 /* Resend the TCP packet because it's
321 * clear that the old packet has been
322 * dropped. This is the new "fast" path mtu
323 * discovery.
325 tcp_simple_retransmit(sk);
326 } /* else let the usual retransmit timer handle it */
330 * This routine is called by the ICMP module when it gets some
331 * sort of error condition. If err < 0 then the socket should
332 * be closed and the error returned to the user. If err > 0
333 * it's just the icmp type << 8 | icmp code. After adjustment
334 * header points to the first 8 bytes of the tcp header. We need
335 * to find the appropriate port.
337 * The locking strategy used here is very "optimistic". When
338 * someone else accesses the socket the ICMP is just dropped
339 * and for some paths there is no check at all.
340 * A more general error queue to queue errors for later handling
341 * is probably better.
345 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
347 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
348 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
349 struct inet_connection_sock *icsk;
350 struct tcp_sock *tp;
351 struct inet_sock *inet;
352 const int type = icmp_hdr(icmp_skb)->type;
353 const int code = icmp_hdr(icmp_skb)->code;
354 struct sock *sk;
355 struct sk_buff *skb;
356 __u32 seq;
357 __u32 remaining;
358 int err;
359 struct net *net = dev_net(icmp_skb->dev);
361 if (icmp_skb->len < (iph->ihl << 2) + 8) {
362 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
363 return;
366 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
367 iph->saddr, th->source, inet_iif(icmp_skb));
368 if (!sk) {
369 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
370 return;
372 if (sk->sk_state == TCP_TIME_WAIT) {
373 inet_twsk_put(inet_twsk(sk));
374 return;
377 bh_lock_sock(sk);
378 /* If too many ICMPs get dropped on busy
379 * servers this needs to be solved differently.
381 if (sock_owned_by_user(sk))
382 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
384 if (sk->sk_state == TCP_CLOSE)
385 goto out;
387 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
388 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
389 goto out;
392 icsk = inet_csk(sk);
393 tp = tcp_sk(sk);
394 seq = ntohl(th->seq);
395 if (sk->sk_state != TCP_LISTEN &&
396 !between(seq, tp->snd_una, tp->snd_nxt)) {
397 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
398 goto out;
401 switch (type) {
402 case ICMP_SOURCE_QUENCH:
403 /* Just silently ignore these. */
404 goto out;
405 case ICMP_PARAMETERPROB:
406 err = EPROTO;
407 break;
408 case ICMP_DEST_UNREACH:
409 if (code > NR_ICMP_UNREACH)
410 goto out;
412 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
413 if (!sock_owned_by_user(sk))
414 do_pmtu_discovery(sk, iph, info);
415 goto out;
418 err = icmp_err_convert[code].errno;
419 /* check if icmp_skb allows revert of backoff
420 * (see draft-zimmermann-tcp-lcd) */
421 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
422 break;
423 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
424 !icsk->icsk_backoff)
425 break;
427 if (sock_owned_by_user(sk))
428 break;
430 icsk->icsk_backoff--;
431 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) <<
432 icsk->icsk_backoff;
433 tcp_bound_rto(sk);
435 skb = tcp_write_queue_head(sk);
436 BUG_ON(!skb);
438 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
439 tcp_time_stamp - TCP_SKB_CB(skb)->when);
441 if (remaining) {
442 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
443 remaining, TCP_RTO_MAX);
444 } else {
445 /* RTO revert clocked out retransmission.
446 * Will retransmit now */
447 tcp_retransmit_timer(sk);
450 break;
451 case ICMP_TIME_EXCEEDED:
452 err = EHOSTUNREACH;
453 break;
454 default:
455 goto out;
458 switch (sk->sk_state) {
459 struct request_sock *req, **prev;
460 case TCP_LISTEN:
461 if (sock_owned_by_user(sk))
462 goto out;
464 req = inet_csk_search_req(sk, &prev, th->dest,
465 iph->daddr, iph->saddr);
466 if (!req)
467 goto out;
469 /* ICMPs are not backlogged, hence we cannot get
470 an established socket here.
472 WARN_ON(req->sk);
474 if (seq != tcp_rsk(req)->snt_isn) {
475 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
476 goto out;
480 * Still in SYN_RECV, just remove it silently.
481 * There is no good way to pass the error to the newly
482 * created socket, and POSIX does not want network
483 * errors returned from accept().
485 inet_csk_reqsk_queue_drop(sk, req, prev);
486 goto out;
488 case TCP_SYN_SENT:
489 case TCP_SYN_RECV: /* Cannot happen.
490 It can f.e. if SYNs crossed.
492 if (!sock_owned_by_user(sk)) {
493 sk->sk_err = err;
495 sk->sk_error_report(sk);
497 tcp_done(sk);
498 } else {
499 sk->sk_err_soft = err;
501 goto out;
504 /* If we've already connected we will keep trying
505 * until we time out, or the user gives up.
507 * rfc1122 4.2.3.9 allows to consider as hard errors
508 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
509 * but it is obsoleted by pmtu discovery).
511 * Note, that in modern internet, where routing is unreliable
512 * and in each dark corner broken firewalls sit, sending random
513 * errors ordered by their masters even this two messages finally lose
514 * their original sense (even Linux sends invalid PORT_UNREACHs)
516 * Now we are in compliance with RFCs.
517 * --ANK (980905)
520 inet = inet_sk(sk);
521 if (!sock_owned_by_user(sk) && inet->recverr) {
522 sk->sk_err = err;
523 sk->sk_error_report(sk);
524 } else { /* Only an error on timeout */
525 sk->sk_err_soft = err;
528 out:
529 bh_unlock_sock(sk);
530 sock_put(sk);
533 static void __tcp_v4_send_check(struct sk_buff *skb,
534 __be32 saddr, __be32 daddr)
536 struct tcphdr *th = tcp_hdr(skb);
538 if (skb->ip_summed == CHECKSUM_PARTIAL) {
539 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
540 skb->csum_start = skb_transport_header(skb) - skb->head;
541 skb->csum_offset = offsetof(struct tcphdr, check);
542 } else {
543 th->check = tcp_v4_check(skb->len, saddr, daddr,
544 csum_partial(th,
545 th->doff << 2,
546 skb->csum));
550 /* This routine computes an IPv4 TCP checksum. */
551 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
553 struct inet_sock *inet = inet_sk(sk);
555 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
557 EXPORT_SYMBOL(tcp_v4_send_check);
559 int tcp_v4_gso_send_check(struct sk_buff *skb)
561 const struct iphdr *iph;
562 struct tcphdr *th;
564 if (!pskb_may_pull(skb, sizeof(*th)))
565 return -EINVAL;
567 iph = ip_hdr(skb);
568 th = tcp_hdr(skb);
570 th->check = 0;
571 skb->ip_summed = CHECKSUM_PARTIAL;
572 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
573 return 0;
577 * This routine will send an RST to the other tcp.
579 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
580 * for reset.
581 * Answer: if a packet caused RST, it is not for a socket
582 * existing in our system, if it is matched to a socket,
583 * it is just duplicate segment or bug in other side's TCP.
584 * So that we build reply only basing on parameters
585 * arrived with segment.
586 * Exception: precedence violation. We do not implement it in any case.
589 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
591 struct tcphdr *th = tcp_hdr(skb);
592 struct {
593 struct tcphdr th;
594 #ifdef CONFIG_TCP_MD5SIG
595 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
596 #endif
597 } rep;
598 struct ip_reply_arg arg;
599 #ifdef CONFIG_TCP_MD5SIG
600 struct tcp_md5sig_key *key;
601 #endif
602 struct net *net;
604 /* Never send a reset in response to a reset. */
605 if (th->rst)
606 return;
608 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
609 return;
611 /* Swap the send and the receive. */
612 memset(&rep, 0, sizeof(rep));
613 rep.th.dest = th->source;
614 rep.th.source = th->dest;
615 rep.th.doff = sizeof(struct tcphdr) / 4;
616 rep.th.rst = 1;
618 if (th->ack) {
619 rep.th.seq = th->ack_seq;
620 } else {
621 rep.th.ack = 1;
622 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
623 skb->len - (th->doff << 2));
626 memset(&arg, 0, sizeof(arg));
627 arg.iov[0].iov_base = (unsigned char *)&rep;
628 arg.iov[0].iov_len = sizeof(rep.th);
630 #ifdef CONFIG_TCP_MD5SIG
631 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
632 if (key) {
633 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
634 (TCPOPT_NOP << 16) |
635 (TCPOPT_MD5SIG << 8) |
636 TCPOLEN_MD5SIG);
637 /* Update length and the length the header thinks exists */
638 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
639 rep.th.doff = arg.iov[0].iov_len / 4;
641 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
642 key, ip_hdr(skb)->saddr,
643 ip_hdr(skb)->daddr, &rep.th);
645 #endif
646 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
647 ip_hdr(skb)->saddr, /* XXX */
648 arg.iov[0].iov_len, IPPROTO_TCP, 0);
649 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
650 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
652 net = dev_net(skb_dst(skb)->dev);
653 ip_send_reply(net->ipv4.tcp_sock, skb,
654 &arg, arg.iov[0].iov_len);
656 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
657 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
660 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
661 outside socket context is ugly, certainly. What can I do?
664 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
665 u32 win, u32 ts, int oif,
666 struct tcp_md5sig_key *key,
667 int reply_flags)
669 struct tcphdr *th = tcp_hdr(skb);
670 struct {
671 struct tcphdr th;
672 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
673 #ifdef CONFIG_TCP_MD5SIG
674 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
675 #endif
677 } rep;
678 struct ip_reply_arg arg;
679 struct net *net = dev_net(skb_dst(skb)->dev);
681 memset(&rep.th, 0, sizeof(struct tcphdr));
682 memset(&arg, 0, sizeof(arg));
684 arg.iov[0].iov_base = (unsigned char *)&rep;
685 arg.iov[0].iov_len = sizeof(rep.th);
686 if (ts) {
687 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
688 (TCPOPT_TIMESTAMP << 8) |
689 TCPOLEN_TIMESTAMP);
690 rep.opt[1] = htonl(tcp_time_stamp);
691 rep.opt[2] = htonl(ts);
692 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
695 /* Swap the send and the receive. */
696 rep.th.dest = th->source;
697 rep.th.source = th->dest;
698 rep.th.doff = arg.iov[0].iov_len / 4;
699 rep.th.seq = htonl(seq);
700 rep.th.ack_seq = htonl(ack);
701 rep.th.ack = 1;
702 rep.th.window = htons(win);
704 #ifdef CONFIG_TCP_MD5SIG
705 if (key) {
706 int offset = (ts) ? 3 : 0;
708 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
709 (TCPOPT_NOP << 16) |
710 (TCPOPT_MD5SIG << 8) |
711 TCPOLEN_MD5SIG);
712 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
713 rep.th.doff = arg.iov[0].iov_len/4;
715 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
716 key, ip_hdr(skb)->saddr,
717 ip_hdr(skb)->daddr, &rep.th);
719 #endif
720 arg.flags = reply_flags;
721 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
722 ip_hdr(skb)->saddr, /* XXX */
723 arg.iov[0].iov_len, IPPROTO_TCP, 0);
724 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
725 if (oif)
726 arg.bound_dev_if = oif;
728 ip_send_reply(net->ipv4.tcp_sock, skb,
729 &arg, arg.iov[0].iov_len);
731 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
734 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
736 struct inet_timewait_sock *tw = inet_twsk(sk);
737 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
739 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
740 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
741 tcptw->tw_ts_recent,
742 tw->tw_bound_dev_if,
743 tcp_twsk_md5_key(tcptw),
744 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
747 inet_twsk_put(tw);
750 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
751 struct request_sock *req)
753 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
754 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
755 req->ts_recent,
757 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
758 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
762 * Send a SYN-ACK after having received a SYN.
763 * This still operates on a request_sock only, not on a big
764 * socket.
766 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
767 struct request_sock *req,
768 struct request_values *rvp)
770 const struct inet_request_sock *ireq = inet_rsk(req);
771 int err = -1;
772 struct sk_buff * skb;
774 /* First, grab a route. */
775 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
776 return -1;
778 skb = tcp_make_synack(sk, dst, req, rvp);
780 if (skb) {
781 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
783 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
784 ireq->rmt_addr,
785 ireq->opt);
786 err = net_xmit_eval(err);
789 dst_release(dst);
790 return err;
793 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
794 struct request_values *rvp)
796 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
797 return tcp_v4_send_synack(sk, NULL, req, rvp);
801 * IPv4 request_sock destructor.
803 static void tcp_v4_reqsk_destructor(struct request_sock *req)
805 kfree(inet_rsk(req)->opt);
808 static void syn_flood_warning(const struct sk_buff *skb)
810 const char *msg;
812 #ifdef CONFIG_SYN_COOKIES
813 if (sysctl_tcp_syncookies)
814 msg = "Sending cookies";
815 else
816 #endif
817 msg = "Dropping request";
819 pr_info("TCP: Possible SYN flooding on port %d. %s.\n",
820 ntohs(tcp_hdr(skb)->dest), msg);
824 * Save and compile IPv4 options into the request_sock if needed.
826 static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk,
827 struct sk_buff *skb)
829 const struct ip_options *opt = &(IPCB(skb)->opt);
830 struct ip_options_rcu *dopt = NULL;
832 if (opt && opt->optlen) {
833 int opt_size = sizeof(*dopt) + opt->optlen;
835 dopt = kmalloc(opt_size, GFP_ATOMIC);
836 if (dopt) {
837 if (ip_options_echo(&dopt->opt, skb)) {
838 kfree(dopt);
839 dopt = NULL;
843 return dopt;
846 #ifdef CONFIG_TCP_MD5SIG
848 * RFC2385 MD5 checksumming requires a mapping of
849 * IP address->MD5 Key.
850 * We need to maintain these in the sk structure.
853 /* Find the Key structure for an address. */
854 static struct tcp_md5sig_key *
855 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
857 struct tcp_sock *tp = tcp_sk(sk);
858 int i;
860 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
861 return NULL;
862 for (i = 0; i < tp->md5sig_info->entries4; i++) {
863 if (tp->md5sig_info->keys4[i].addr == addr)
864 return &tp->md5sig_info->keys4[i].base;
866 return NULL;
869 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
870 struct sock *addr_sk)
872 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
874 EXPORT_SYMBOL(tcp_v4_md5_lookup);
876 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
877 struct request_sock *req)
879 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
882 /* This can be called on a newly created socket, from other files */
883 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
884 u8 *newkey, u8 newkeylen)
886 /* Add Key to the list */
887 struct tcp_md5sig_key *key;
888 struct tcp_sock *tp = tcp_sk(sk);
889 struct tcp4_md5sig_key *keys;
891 key = tcp_v4_md5_do_lookup(sk, addr);
892 if (key) {
893 /* Pre-existing entry - just update that one. */
894 kfree(key->key);
895 key->key = newkey;
896 key->keylen = newkeylen;
897 } else {
898 struct tcp_md5sig_info *md5sig;
900 if (!tp->md5sig_info) {
901 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
902 GFP_ATOMIC);
903 if (!tp->md5sig_info) {
904 kfree(newkey);
905 return -ENOMEM;
907 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
909 if (tcp_alloc_md5sig_pool(sk) == NULL) {
910 kfree(newkey);
911 return -ENOMEM;
913 md5sig = tp->md5sig_info;
915 if (md5sig->alloced4 == md5sig->entries4) {
916 keys = kmalloc((sizeof(*keys) *
917 (md5sig->entries4 + 1)), GFP_ATOMIC);
918 if (!keys) {
919 kfree(newkey);
920 tcp_free_md5sig_pool();
921 return -ENOMEM;
924 if (md5sig->entries4)
925 memcpy(keys, md5sig->keys4,
926 sizeof(*keys) * md5sig->entries4);
928 /* Free old key list, and reference new one */
929 kfree(md5sig->keys4);
930 md5sig->keys4 = keys;
931 md5sig->alloced4++;
933 md5sig->entries4++;
934 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
935 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
936 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
938 return 0;
940 EXPORT_SYMBOL(tcp_v4_md5_do_add);
942 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
943 u8 *newkey, u8 newkeylen)
945 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
946 newkey, newkeylen);
949 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
951 struct tcp_sock *tp = tcp_sk(sk);
952 int i;
954 for (i = 0; i < tp->md5sig_info->entries4; i++) {
955 if (tp->md5sig_info->keys4[i].addr == addr) {
956 /* Free the key */
957 kfree(tp->md5sig_info->keys4[i].base.key);
958 tp->md5sig_info->entries4--;
960 if (tp->md5sig_info->entries4 == 0) {
961 kfree(tp->md5sig_info->keys4);
962 tp->md5sig_info->keys4 = NULL;
963 tp->md5sig_info->alloced4 = 0;
964 } else if (tp->md5sig_info->entries4 != i) {
965 /* Need to do some manipulation */
966 memmove(&tp->md5sig_info->keys4[i],
967 &tp->md5sig_info->keys4[i+1],
968 (tp->md5sig_info->entries4 - i) *
969 sizeof(struct tcp4_md5sig_key));
971 tcp_free_md5sig_pool();
972 return 0;
975 return -ENOENT;
977 EXPORT_SYMBOL(tcp_v4_md5_do_del);
979 static void tcp_v4_clear_md5_list(struct sock *sk)
981 struct tcp_sock *tp = tcp_sk(sk);
983 /* Free each key, then the set of key keys,
984 * the crypto element, and then decrement our
985 * hold on the last resort crypto.
987 if (tp->md5sig_info->entries4) {
988 int i;
989 for (i = 0; i < tp->md5sig_info->entries4; i++)
990 kfree(tp->md5sig_info->keys4[i].base.key);
991 tp->md5sig_info->entries4 = 0;
992 tcp_free_md5sig_pool();
994 if (tp->md5sig_info->keys4) {
995 kfree(tp->md5sig_info->keys4);
996 tp->md5sig_info->keys4 = NULL;
997 tp->md5sig_info->alloced4 = 0;
1001 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1002 int optlen)
1004 struct tcp_md5sig cmd;
1005 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1006 u8 *newkey;
1008 if (optlen < sizeof(cmd))
1009 return -EINVAL;
1011 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1012 return -EFAULT;
1014 if (sin->sin_family != AF_INET)
1015 return -EINVAL;
1017 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1018 if (!tcp_sk(sk)->md5sig_info)
1019 return -ENOENT;
1020 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1023 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1024 return -EINVAL;
1026 if (!tcp_sk(sk)->md5sig_info) {
1027 struct tcp_sock *tp = tcp_sk(sk);
1028 struct tcp_md5sig_info *p;
1030 p = kzalloc(sizeof(*p), sk->sk_allocation);
1031 if (!p)
1032 return -EINVAL;
1034 tp->md5sig_info = p;
1035 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1038 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1039 if (!newkey)
1040 return -ENOMEM;
1041 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1042 newkey, cmd.tcpm_keylen);
1045 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1046 __be32 daddr, __be32 saddr, int nbytes)
1048 struct tcp4_pseudohdr *bp;
1049 struct scatterlist sg;
1051 bp = &hp->md5_blk.ip4;
1054 * 1. the TCP pseudo-header (in the order: source IP address,
1055 * destination IP address, zero-padded protocol number, and
1056 * segment length)
1058 bp->saddr = saddr;
1059 bp->daddr = daddr;
1060 bp->pad = 0;
1061 bp->protocol = IPPROTO_TCP;
1062 bp->len = cpu_to_be16(nbytes);
1064 sg_init_one(&sg, bp, sizeof(*bp));
1065 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1068 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1069 __be32 daddr, __be32 saddr, struct tcphdr *th)
1071 struct tcp_md5sig_pool *hp;
1072 struct hash_desc *desc;
1074 hp = tcp_get_md5sig_pool();
1075 if (!hp)
1076 goto clear_hash_noput;
1077 desc = &hp->md5_desc;
1079 if (crypto_hash_init(desc))
1080 goto clear_hash;
1081 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1082 goto clear_hash;
1083 if (tcp_md5_hash_header(hp, th))
1084 goto clear_hash;
1085 if (tcp_md5_hash_key(hp, key))
1086 goto clear_hash;
1087 if (crypto_hash_final(desc, md5_hash))
1088 goto clear_hash;
1090 tcp_put_md5sig_pool();
1091 return 0;
1093 clear_hash:
1094 tcp_put_md5sig_pool();
1095 clear_hash_noput:
1096 memset(md5_hash, 0, 16);
1097 return 1;
1100 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1101 struct sock *sk, struct request_sock *req,
1102 struct sk_buff *skb)
1104 struct tcp_md5sig_pool *hp;
1105 struct hash_desc *desc;
1106 struct tcphdr *th = tcp_hdr(skb);
1107 __be32 saddr, daddr;
1109 if (sk) {
1110 saddr = inet_sk(sk)->inet_saddr;
1111 daddr = inet_sk(sk)->inet_daddr;
1112 } else if (req) {
1113 saddr = inet_rsk(req)->loc_addr;
1114 daddr = inet_rsk(req)->rmt_addr;
1115 } else {
1116 const struct iphdr *iph = ip_hdr(skb);
1117 saddr = iph->saddr;
1118 daddr = iph->daddr;
1121 hp = tcp_get_md5sig_pool();
1122 if (!hp)
1123 goto clear_hash_noput;
1124 desc = &hp->md5_desc;
1126 if (crypto_hash_init(desc))
1127 goto clear_hash;
1129 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1130 goto clear_hash;
1131 if (tcp_md5_hash_header(hp, th))
1132 goto clear_hash;
1133 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1134 goto clear_hash;
1135 if (tcp_md5_hash_key(hp, key))
1136 goto clear_hash;
1137 if (crypto_hash_final(desc, md5_hash))
1138 goto clear_hash;
1140 tcp_put_md5sig_pool();
1141 return 0;
1143 clear_hash:
1144 tcp_put_md5sig_pool();
1145 clear_hash_noput:
1146 memset(md5_hash, 0, 16);
1147 return 1;
1149 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1151 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1154 * This gets called for each TCP segment that arrives
1155 * so we want to be efficient.
1156 * We have 3 drop cases:
1157 * o No MD5 hash and one expected.
1158 * o MD5 hash and we're not expecting one.
1159 * o MD5 hash and its wrong.
1161 __u8 *hash_location = NULL;
1162 struct tcp_md5sig_key *hash_expected;
1163 const struct iphdr *iph = ip_hdr(skb);
1164 struct tcphdr *th = tcp_hdr(skb);
1165 int genhash;
1166 unsigned char newhash[16];
1168 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1169 hash_location = tcp_parse_md5sig_option(th);
1171 /* We've parsed the options - do we have a hash? */
1172 if (!hash_expected && !hash_location)
1173 return 0;
1175 if (hash_expected && !hash_location) {
1176 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1177 return 1;
1180 if (!hash_expected && hash_location) {
1181 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1182 return 1;
1185 /* Okay, so this is hash_expected and hash_location -
1186 * so we need to calculate the checksum.
1188 genhash = tcp_v4_md5_hash_skb(newhash,
1189 hash_expected,
1190 NULL, NULL, skb);
1192 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1193 if (net_ratelimit()) {
1194 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1195 &iph->saddr, ntohs(th->source),
1196 &iph->daddr, ntohs(th->dest),
1197 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1199 return 1;
1201 return 0;
1204 #endif
1206 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1207 .family = PF_INET,
1208 .obj_size = sizeof(struct tcp_request_sock),
1209 .rtx_syn_ack = tcp_v4_rtx_synack,
1210 .send_ack = tcp_v4_reqsk_send_ack,
1211 .destructor = tcp_v4_reqsk_destructor,
1212 .send_reset = tcp_v4_send_reset,
1213 .syn_ack_timeout = tcp_syn_ack_timeout,
1216 #ifdef CONFIG_TCP_MD5SIG
1217 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1218 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1219 .calc_md5_hash = tcp_v4_md5_hash_skb,
1221 #endif
1223 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1225 struct tcp_extend_values tmp_ext;
1226 struct tcp_options_received tmp_opt;
1227 u8 *hash_location;
1228 struct request_sock *req;
1229 struct inet_request_sock *ireq;
1230 struct tcp_sock *tp = tcp_sk(sk);
1231 struct dst_entry *dst = NULL;
1232 __be32 saddr = ip_hdr(skb)->saddr;
1233 __be32 daddr = ip_hdr(skb)->daddr;
1234 __u32 isn = TCP_SKB_CB(skb)->when;
1235 #ifdef CONFIG_SYN_COOKIES
1236 int want_cookie = 0;
1237 #else
1238 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1239 #endif
1241 /* Never answer to SYNs send to broadcast or multicast */
1242 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1243 goto drop;
1245 /* TW buckets are converted to open requests without
1246 * limitations, they conserve resources and peer is
1247 * evidently real one.
1249 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1250 if (net_ratelimit())
1251 syn_flood_warning(skb);
1252 #ifdef CONFIG_SYN_COOKIES
1253 if (sysctl_tcp_syncookies) {
1254 want_cookie = 1;
1255 } else
1256 #endif
1257 goto drop;
1260 /* Accept backlog is full. If we have already queued enough
1261 * of warm entries in syn queue, drop request. It is better than
1262 * clogging syn queue with openreqs with exponentially increasing
1263 * timeout.
1265 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1266 goto drop;
1268 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1269 if (!req)
1270 goto drop;
1272 #ifdef CONFIG_TCP_MD5SIG
1273 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1274 #endif
1276 tcp_clear_options(&tmp_opt);
1277 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1278 tmp_opt.user_mss = tp->rx_opt.user_mss;
1279 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1281 if (tmp_opt.cookie_plus > 0 &&
1282 tmp_opt.saw_tstamp &&
1283 !tp->rx_opt.cookie_out_never &&
1284 (sysctl_tcp_cookie_size > 0 ||
1285 (tp->cookie_values != NULL &&
1286 tp->cookie_values->cookie_desired > 0))) {
1287 u8 *c;
1288 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1289 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1291 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1292 goto drop_and_release;
1294 /* Secret recipe starts with IP addresses */
1295 *mess++ ^= (__force u32)daddr;
1296 *mess++ ^= (__force u32)saddr;
1298 /* plus variable length Initiator Cookie */
1299 c = (u8 *)mess;
1300 while (l-- > 0)
1301 *c++ ^= *hash_location++;
1303 #ifdef CONFIG_SYN_COOKIES
1304 want_cookie = 0; /* not our kind of cookie */
1305 #endif
1306 tmp_ext.cookie_out_never = 0; /* false */
1307 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1308 } else if (!tp->rx_opt.cookie_in_always) {
1309 /* redundant indications, but ensure initialization. */
1310 tmp_ext.cookie_out_never = 1; /* true */
1311 tmp_ext.cookie_plus = 0;
1312 } else {
1313 goto drop_and_release;
1315 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1317 if (want_cookie && !tmp_opt.saw_tstamp)
1318 tcp_clear_options(&tmp_opt);
1320 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1321 tcp_openreq_init(req, &tmp_opt, skb);
1323 ireq = inet_rsk(req);
1324 ireq->loc_addr = daddr;
1325 ireq->rmt_addr = saddr;
1326 ireq->no_srccheck = inet_sk(sk)->transparent;
1327 ireq->opt = tcp_v4_save_options(sk, skb);
1329 if (security_inet_conn_request(sk, skb, req))
1330 goto drop_and_free;
1332 if (!want_cookie || tmp_opt.tstamp_ok)
1333 TCP_ECN_create_request(req, tcp_hdr(skb));
1335 if (want_cookie) {
1336 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1337 req->cookie_ts = tmp_opt.tstamp_ok;
1338 } else if (!isn) {
1339 struct inet_peer *peer = NULL;
1341 /* VJ's idea. We save last timestamp seen
1342 * from the destination in peer table, when entering
1343 * state TIME-WAIT, and check against it before
1344 * accepting new connection request.
1346 * If "isn" is not zero, this request hit alive
1347 * timewait bucket, so that all the necessary checks
1348 * are made in the function processing timewait state.
1350 if (tmp_opt.saw_tstamp &&
1351 tcp_death_row.sysctl_tw_recycle &&
1352 (dst = inet_csk_route_req(sk, req)) != NULL &&
1353 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1354 peer->daddr.addr.a4 == saddr) {
1355 inet_peer_refcheck(peer);
1356 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1357 (s32)(peer->tcp_ts - req->ts_recent) >
1358 TCP_PAWS_WINDOW) {
1359 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1360 goto drop_and_release;
1363 /* Kill the following clause, if you dislike this way. */
1364 else if (!sysctl_tcp_syncookies &&
1365 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1366 (sysctl_max_syn_backlog >> 2)) &&
1367 (!peer || !peer->tcp_ts_stamp) &&
1368 (!dst || !dst_metric(dst, RTAX_RTT))) {
1369 /* Without syncookies last quarter of
1370 * backlog is filled with destinations,
1371 * proven to be alive.
1372 * It means that we continue to communicate
1373 * to destinations, already remembered
1374 * to the moment of synflood.
1376 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1377 &saddr, ntohs(tcp_hdr(skb)->source));
1378 goto drop_and_release;
1381 isn = tcp_v4_init_sequence(skb);
1383 tcp_rsk(req)->snt_isn = isn;
1385 if (tcp_v4_send_synack(sk, dst, req,
1386 (struct request_values *)&tmp_ext) ||
1387 want_cookie)
1388 goto drop_and_free;
1390 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1391 return 0;
1393 drop_and_release:
1394 dst_release(dst);
1395 drop_and_free:
1396 reqsk_free(req);
1397 drop:
1398 return 0;
1400 EXPORT_SYMBOL(tcp_v4_conn_request);
1404 * The three way handshake has completed - we got a valid synack -
1405 * now create the new socket.
1407 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1408 struct request_sock *req,
1409 struct dst_entry *dst)
1411 struct inet_request_sock *ireq;
1412 struct inet_sock *newinet;
1413 struct tcp_sock *newtp;
1414 struct sock *newsk;
1415 #ifdef CONFIG_TCP_MD5SIG
1416 struct tcp_md5sig_key *key;
1417 #endif
1418 struct ip_options_rcu *inet_opt;
1420 if (sk_acceptq_is_full(sk))
1421 goto exit_overflow;
1423 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1424 goto exit;
1426 newsk = tcp_create_openreq_child(sk, req, skb);
1427 if (!newsk)
1428 goto exit_nonewsk;
1430 newsk->sk_gso_type = SKB_GSO_TCPV4;
1431 sk_setup_caps(newsk, dst);
1433 newtp = tcp_sk(newsk);
1434 newinet = inet_sk(newsk);
1435 ireq = inet_rsk(req);
1436 newinet->inet_daddr = ireq->rmt_addr;
1437 newinet->inet_rcv_saddr = ireq->loc_addr;
1438 newinet->inet_saddr = ireq->loc_addr;
1439 inet_opt = ireq->opt;
1440 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1441 ireq->opt = NULL;
1442 newinet->mc_index = inet_iif(skb);
1443 newinet->mc_ttl = ip_hdr(skb)->ttl;
1444 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1445 if (inet_opt)
1446 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1447 newinet->inet_id = newtp->write_seq ^ jiffies;
1449 tcp_mtup_init(newsk);
1450 tcp_sync_mss(newsk, dst_mtu(dst));
1451 newtp->advmss = dst_metric_advmss(dst);
1452 if (tcp_sk(sk)->rx_opt.user_mss &&
1453 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1454 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1456 tcp_initialize_rcv_mss(newsk);
1458 #ifdef CONFIG_TCP_MD5SIG
1459 /* Copy over the MD5 key from the original socket */
1460 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1461 if (key != NULL) {
1463 * We're using one, so create a matching key
1464 * on the newsk structure. If we fail to get
1465 * memory, then we end up not copying the key
1466 * across. Shucks.
1468 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1469 if (newkey != NULL)
1470 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1471 newkey, key->keylen);
1472 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1474 #endif
1476 if (__inet_inherit_port(sk, newsk) < 0) {
1477 sock_put(newsk);
1478 goto exit;
1480 __inet_hash_nolisten(newsk, NULL);
1482 return newsk;
1484 exit_overflow:
1485 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1486 exit_nonewsk:
1487 dst_release(dst);
1488 exit:
1489 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1490 return NULL;
1492 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1494 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1496 struct tcphdr *th = tcp_hdr(skb);
1497 const struct iphdr *iph = ip_hdr(skb);
1498 struct sock *nsk;
1499 struct request_sock **prev;
1500 /* Find possible connection requests. */
1501 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1502 iph->saddr, iph->daddr);
1503 if (req)
1504 return tcp_check_req(sk, skb, req, prev);
1506 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1507 th->source, iph->daddr, th->dest, inet_iif(skb));
1509 if (nsk) {
1510 if (nsk->sk_state != TCP_TIME_WAIT) {
1511 bh_lock_sock(nsk);
1512 return nsk;
1514 inet_twsk_put(inet_twsk(nsk));
1515 return NULL;
1518 #ifdef CONFIG_SYN_COOKIES
1519 if (!th->syn)
1520 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1521 #endif
1522 return sk;
1525 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1527 const struct iphdr *iph = ip_hdr(skb);
1529 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1530 if (!tcp_v4_check(skb->len, iph->saddr,
1531 iph->daddr, skb->csum)) {
1532 skb->ip_summed = CHECKSUM_UNNECESSARY;
1533 return 0;
1537 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1538 skb->len, IPPROTO_TCP, 0);
1540 if (skb->len <= 76) {
1541 return __skb_checksum_complete(skb);
1543 return 0;
1547 /* The socket must have it's spinlock held when we get
1548 * here.
1550 * We have a potential double-lock case here, so even when
1551 * doing backlog processing we use the BH locking scheme.
1552 * This is because we cannot sleep with the original spinlock
1553 * held.
1555 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1557 struct sock *rsk;
1558 #ifdef CONFIG_TCP_MD5SIG
1560 * We really want to reject the packet as early as possible
1561 * if:
1562 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1563 * o There is an MD5 option and we're not expecting one
1565 if (tcp_v4_inbound_md5_hash(sk, skb))
1566 goto discard;
1567 #endif
1569 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1570 sock_rps_save_rxhash(sk, skb->rxhash);
1571 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1572 rsk = sk;
1573 goto reset;
1575 return 0;
1578 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1579 goto csum_err;
1581 if (sk->sk_state == TCP_LISTEN) {
1582 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1583 if (!nsk)
1584 goto discard;
1586 if (nsk != sk) {
1587 if (tcp_child_process(sk, nsk, skb)) {
1588 rsk = nsk;
1589 goto reset;
1591 return 0;
1593 } else
1594 sock_rps_save_rxhash(sk, skb->rxhash);
1596 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1597 rsk = sk;
1598 goto reset;
1600 return 0;
1602 reset:
1603 tcp_v4_send_reset(rsk, skb);
1604 discard:
1605 kfree_skb(skb);
1606 /* Be careful here. If this function gets more complicated and
1607 * gcc suffers from register pressure on the x86, sk (in %ebx)
1608 * might be destroyed here. This current version compiles correctly,
1609 * but you have been warned.
1611 return 0;
1613 csum_err:
1614 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1615 goto discard;
1617 EXPORT_SYMBOL(tcp_v4_do_rcv);
1620 * From tcp_input.c
1623 int tcp_v4_rcv(struct sk_buff *skb)
1625 const struct iphdr *iph;
1626 struct tcphdr *th;
1627 struct sock *sk;
1628 int ret;
1629 struct net *net = dev_net(skb->dev);
1631 if (skb->pkt_type != PACKET_HOST)
1632 goto discard_it;
1634 /* Count it even if it's bad */
1635 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1637 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1638 goto discard_it;
1640 th = tcp_hdr(skb);
1642 if (th->doff < sizeof(struct tcphdr) / 4)
1643 goto bad_packet;
1644 if (!pskb_may_pull(skb, th->doff * 4))
1645 goto discard_it;
1647 /* An explanation is required here, I think.
1648 * Packet length and doff are validated by header prediction,
1649 * provided case of th->doff==0 is eliminated.
1650 * So, we defer the checks. */
1651 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1652 goto bad_packet;
1654 th = tcp_hdr(skb);
1655 iph = ip_hdr(skb);
1656 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1657 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1658 skb->len - th->doff * 4);
1659 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1660 TCP_SKB_CB(skb)->when = 0;
1661 TCP_SKB_CB(skb)->flags = iph->tos;
1662 TCP_SKB_CB(skb)->sacked = 0;
1664 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1665 if (!sk)
1666 goto no_tcp_socket;
1668 process:
1669 if (sk->sk_state == TCP_TIME_WAIT)
1670 goto do_time_wait;
1672 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1673 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1674 goto discard_and_relse;
1677 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1678 goto discard_and_relse;
1679 nf_reset(skb);
1681 if (sk_filter(sk, skb))
1682 goto discard_and_relse;
1684 skb->dev = NULL;
1686 bh_lock_sock_nested(sk);
1687 ret = 0;
1688 if (!sock_owned_by_user(sk)) {
1689 #ifdef CONFIG_NET_DMA
1690 struct tcp_sock *tp = tcp_sk(sk);
1691 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1692 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1693 if (tp->ucopy.dma_chan)
1694 ret = tcp_v4_do_rcv(sk, skb);
1695 else
1696 #endif
1698 if (!tcp_prequeue(sk, skb))
1699 ret = tcp_v4_do_rcv(sk, skb);
1701 } else if (unlikely(sk_add_backlog(sk, skb))) {
1702 bh_unlock_sock(sk);
1703 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1704 goto discard_and_relse;
1706 bh_unlock_sock(sk);
1708 sock_put(sk);
1710 return ret;
1712 no_tcp_socket:
1713 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1714 goto discard_it;
1716 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1717 bad_packet:
1718 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1719 } else {
1720 tcp_v4_send_reset(NULL, skb);
1723 discard_it:
1724 /* Discard frame. */
1725 kfree_skb(skb);
1726 return 0;
1728 discard_and_relse:
1729 sock_put(sk);
1730 goto discard_it;
1732 do_time_wait:
1733 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1734 inet_twsk_put(inet_twsk(sk));
1735 goto discard_it;
1738 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1739 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1740 inet_twsk_put(inet_twsk(sk));
1741 goto discard_it;
1743 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1744 case TCP_TW_SYN: {
1745 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1746 &tcp_hashinfo,
1747 iph->daddr, th->dest,
1748 inet_iif(skb));
1749 if (sk2) {
1750 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1751 inet_twsk_put(inet_twsk(sk));
1752 sk = sk2;
1753 goto process;
1755 /* Fall through to ACK */
1757 case TCP_TW_ACK:
1758 tcp_v4_timewait_ack(sk, skb);
1759 break;
1760 case TCP_TW_RST:
1761 goto no_tcp_socket;
1762 case TCP_TW_SUCCESS:;
1764 goto discard_it;
1767 struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it)
1769 struct rtable *rt = (struct rtable *) __sk_dst_get(sk);
1770 struct inet_sock *inet = inet_sk(sk);
1771 struct inet_peer *peer;
1773 if (!rt || rt->rt_dst != inet->inet_daddr) {
1774 peer = inet_getpeer_v4(inet->inet_daddr, 1);
1775 *release_it = true;
1776 } else {
1777 if (!rt->peer)
1778 rt_bind_peer(rt, 1);
1779 peer = rt->peer;
1780 *release_it = false;
1783 return peer;
1785 EXPORT_SYMBOL(tcp_v4_get_peer);
1787 void *tcp_v4_tw_get_peer(struct sock *sk)
1789 struct inet_timewait_sock *tw = inet_twsk(sk);
1791 return inet_getpeer_v4(tw->tw_daddr, 1);
1793 EXPORT_SYMBOL(tcp_v4_tw_get_peer);
1795 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1796 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1797 .twsk_unique = tcp_twsk_unique,
1798 .twsk_destructor= tcp_twsk_destructor,
1799 .twsk_getpeer = tcp_v4_tw_get_peer,
1802 const struct inet_connection_sock_af_ops ipv4_specific = {
1803 .queue_xmit = ip_queue_xmit,
1804 .send_check = tcp_v4_send_check,
1805 .rebuild_header = inet_sk_rebuild_header,
1806 .conn_request = tcp_v4_conn_request,
1807 .syn_recv_sock = tcp_v4_syn_recv_sock,
1808 .get_peer = tcp_v4_get_peer,
1809 .net_header_len = sizeof(struct iphdr),
1810 .setsockopt = ip_setsockopt,
1811 .getsockopt = ip_getsockopt,
1812 .addr2sockaddr = inet_csk_addr2sockaddr,
1813 .sockaddr_len = sizeof(struct sockaddr_in),
1814 .bind_conflict = inet_csk_bind_conflict,
1815 #ifdef CONFIG_COMPAT
1816 .compat_setsockopt = compat_ip_setsockopt,
1817 .compat_getsockopt = compat_ip_getsockopt,
1818 #endif
1820 EXPORT_SYMBOL(ipv4_specific);
1822 #ifdef CONFIG_TCP_MD5SIG
1823 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1824 .md5_lookup = tcp_v4_md5_lookup,
1825 .calc_md5_hash = tcp_v4_md5_hash_skb,
1826 .md5_add = tcp_v4_md5_add_func,
1827 .md5_parse = tcp_v4_parse_md5_keys,
1829 #endif
1831 /* NOTE: A lot of things set to zero explicitly by call to
1832 * sk_alloc() so need not be done here.
1834 static int tcp_v4_init_sock(struct sock *sk)
1836 struct inet_connection_sock *icsk = inet_csk(sk);
1837 struct tcp_sock *tp = tcp_sk(sk);
1839 skb_queue_head_init(&tp->out_of_order_queue);
1840 tcp_init_xmit_timers(sk);
1841 tcp_prequeue_init(tp);
1843 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1844 tp->mdev = TCP_TIMEOUT_INIT;
1846 /* So many TCP implementations out there (incorrectly) count the
1847 * initial SYN frame in their delayed-ACK and congestion control
1848 * algorithms that we must have the following bandaid to talk
1849 * efficiently to them. -DaveM
1851 tp->snd_cwnd = 2;
1853 /* See draft-stevens-tcpca-spec-01 for discussion of the
1854 * initialization of these values.
1856 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1857 tp->snd_cwnd_clamp = ~0;
1858 tp->mss_cache = TCP_MSS_DEFAULT;
1860 tp->reordering = sysctl_tcp_reordering;
1861 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1863 sk->sk_state = TCP_CLOSE;
1865 sk->sk_write_space = sk_stream_write_space;
1866 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1868 icsk->icsk_af_ops = &ipv4_specific;
1869 icsk->icsk_sync_mss = tcp_sync_mss;
1870 #ifdef CONFIG_TCP_MD5SIG
1871 tp->af_specific = &tcp_sock_ipv4_specific;
1872 #endif
1874 /* TCP Cookie Transactions */
1875 if (sysctl_tcp_cookie_size > 0) {
1876 /* Default, cookies without s_data_payload. */
1877 tp->cookie_values =
1878 kzalloc(sizeof(*tp->cookie_values),
1879 sk->sk_allocation);
1880 if (tp->cookie_values != NULL)
1881 kref_init(&tp->cookie_values->kref);
1883 /* Presumed zeroed, in order of appearance:
1884 * cookie_in_always, cookie_out_never,
1885 * s_data_constant, s_data_in, s_data_out
1887 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1888 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1890 local_bh_disable();
1891 percpu_counter_inc(&tcp_sockets_allocated);
1892 local_bh_enable();
1894 return 0;
1897 void tcp_v4_destroy_sock(struct sock *sk)
1899 struct tcp_sock *tp = tcp_sk(sk);
1901 tcp_clear_xmit_timers(sk);
1903 tcp_cleanup_congestion_control(sk);
1905 /* Cleanup up the write buffer. */
1906 tcp_write_queue_purge(sk);
1908 /* Cleans up our, hopefully empty, out_of_order_queue. */
1909 __skb_queue_purge(&tp->out_of_order_queue);
1911 #ifdef CONFIG_TCP_MD5SIG
1912 /* Clean up the MD5 key list, if any */
1913 if (tp->md5sig_info) {
1914 tcp_v4_clear_md5_list(sk);
1915 kfree(tp->md5sig_info);
1916 tp->md5sig_info = NULL;
1918 #endif
1920 #ifdef CONFIG_NET_DMA
1921 /* Cleans up our sk_async_wait_queue */
1922 __skb_queue_purge(&sk->sk_async_wait_queue);
1923 #endif
1925 /* Clean prequeue, it must be empty really */
1926 __skb_queue_purge(&tp->ucopy.prequeue);
1928 /* Clean up a referenced TCP bind bucket. */
1929 if (inet_csk(sk)->icsk_bind_hash)
1930 inet_put_port(sk);
1933 * If sendmsg cached page exists, toss it.
1935 if (sk->sk_sndmsg_page) {
1936 __free_page(sk->sk_sndmsg_page);
1937 sk->sk_sndmsg_page = NULL;
1940 /* TCP Cookie Transactions */
1941 if (tp->cookie_values != NULL) {
1942 kref_put(&tp->cookie_values->kref,
1943 tcp_cookie_values_release);
1944 tp->cookie_values = NULL;
1947 percpu_counter_dec(&tcp_sockets_allocated);
1949 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1951 #ifdef CONFIG_PROC_FS
1952 /* Proc filesystem TCP sock list dumping. */
1954 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1956 return hlist_nulls_empty(head) ? NULL :
1957 list_entry(head->first, struct inet_timewait_sock, tw_node);
1960 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1962 return !is_a_nulls(tw->tw_node.next) ?
1963 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1967 * Get next listener socket follow cur. If cur is NULL, get first socket
1968 * starting from bucket given in st->bucket; when st->bucket is zero the
1969 * very first socket in the hash table is returned.
1971 static void *listening_get_next(struct seq_file *seq, void *cur)
1973 struct inet_connection_sock *icsk;
1974 struct hlist_nulls_node *node;
1975 struct sock *sk = cur;
1976 struct inet_listen_hashbucket *ilb;
1977 struct tcp_iter_state *st = seq->private;
1978 struct net *net = seq_file_net(seq);
1980 if (!sk) {
1981 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1982 spin_lock_bh(&ilb->lock);
1983 sk = sk_nulls_head(&ilb->head);
1984 st->offset = 0;
1985 goto get_sk;
1987 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1988 ++st->num;
1989 ++st->offset;
1991 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1992 struct request_sock *req = cur;
1994 icsk = inet_csk(st->syn_wait_sk);
1995 req = req->dl_next;
1996 while (1) {
1997 while (req) {
1998 if (req->rsk_ops->family == st->family) {
1999 cur = req;
2000 goto out;
2002 req = req->dl_next;
2004 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2005 break;
2006 get_req:
2007 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2009 sk = sk_nulls_next(st->syn_wait_sk);
2010 st->state = TCP_SEQ_STATE_LISTENING;
2011 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2012 } else {
2013 icsk = inet_csk(sk);
2014 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2015 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2016 goto start_req;
2017 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2018 sk = sk_nulls_next(sk);
2020 get_sk:
2021 sk_nulls_for_each_from(sk, node) {
2022 if (!net_eq(sock_net(sk), net))
2023 continue;
2024 if (sk->sk_family == st->family) {
2025 cur = sk;
2026 goto out;
2028 icsk = inet_csk(sk);
2029 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2030 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2031 start_req:
2032 st->uid = sock_i_uid(sk);
2033 st->syn_wait_sk = sk;
2034 st->state = TCP_SEQ_STATE_OPENREQ;
2035 st->sbucket = 0;
2036 goto get_req;
2038 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2040 spin_unlock_bh(&ilb->lock);
2041 st->offset = 0;
2042 if (++st->bucket < INET_LHTABLE_SIZE) {
2043 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2044 spin_lock_bh(&ilb->lock);
2045 sk = sk_nulls_head(&ilb->head);
2046 goto get_sk;
2048 cur = NULL;
2049 out:
2050 return cur;
2053 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2055 struct tcp_iter_state *st = seq->private;
2056 void *rc;
2058 st->bucket = 0;
2059 st->offset = 0;
2060 rc = listening_get_next(seq, NULL);
2062 while (rc && *pos) {
2063 rc = listening_get_next(seq, rc);
2064 --*pos;
2066 return rc;
2069 static inline int empty_bucket(struct tcp_iter_state *st)
2071 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2072 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2076 * Get first established socket starting from bucket given in st->bucket.
2077 * If st->bucket is zero, the very first socket in the hash is returned.
2079 static void *established_get_first(struct seq_file *seq)
2081 struct tcp_iter_state *st = seq->private;
2082 struct net *net = seq_file_net(seq);
2083 void *rc = NULL;
2085 st->offset = 0;
2086 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2087 struct sock *sk;
2088 struct hlist_nulls_node *node;
2089 struct inet_timewait_sock *tw;
2090 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2092 /* Lockless fast path for the common case of empty buckets */
2093 if (empty_bucket(st))
2094 continue;
2096 spin_lock_bh(lock);
2097 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2098 if (sk->sk_family != st->family ||
2099 !net_eq(sock_net(sk), net)) {
2100 continue;
2102 rc = sk;
2103 goto out;
2105 st->state = TCP_SEQ_STATE_TIME_WAIT;
2106 inet_twsk_for_each(tw, node,
2107 &tcp_hashinfo.ehash[st->bucket].twchain) {
2108 if (tw->tw_family != st->family ||
2109 !net_eq(twsk_net(tw), net)) {
2110 continue;
2112 rc = tw;
2113 goto out;
2115 spin_unlock_bh(lock);
2116 st->state = TCP_SEQ_STATE_ESTABLISHED;
2118 out:
2119 return rc;
2122 static void *established_get_next(struct seq_file *seq, void *cur)
2124 struct sock *sk = cur;
2125 struct inet_timewait_sock *tw;
2126 struct hlist_nulls_node *node;
2127 struct tcp_iter_state *st = seq->private;
2128 struct net *net = seq_file_net(seq);
2130 ++st->num;
2131 ++st->offset;
2133 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2134 tw = cur;
2135 tw = tw_next(tw);
2136 get_tw:
2137 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2138 tw = tw_next(tw);
2140 if (tw) {
2141 cur = tw;
2142 goto out;
2144 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2145 st->state = TCP_SEQ_STATE_ESTABLISHED;
2147 /* Look for next non empty bucket */
2148 st->offset = 0;
2149 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2150 empty_bucket(st))
2152 if (st->bucket > tcp_hashinfo.ehash_mask)
2153 return NULL;
2155 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2156 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2157 } else
2158 sk = sk_nulls_next(sk);
2160 sk_nulls_for_each_from(sk, node) {
2161 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2162 goto found;
2165 st->state = TCP_SEQ_STATE_TIME_WAIT;
2166 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2167 goto get_tw;
2168 found:
2169 cur = sk;
2170 out:
2171 return cur;
2174 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2176 struct tcp_iter_state *st = seq->private;
2177 void *rc;
2179 st->bucket = 0;
2180 rc = established_get_first(seq);
2182 while (rc && pos) {
2183 rc = established_get_next(seq, rc);
2184 --pos;
2186 return rc;
2189 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2191 void *rc;
2192 struct tcp_iter_state *st = seq->private;
2194 st->state = TCP_SEQ_STATE_LISTENING;
2195 rc = listening_get_idx(seq, &pos);
2197 if (!rc) {
2198 st->state = TCP_SEQ_STATE_ESTABLISHED;
2199 rc = established_get_idx(seq, pos);
2202 return rc;
2205 static void *tcp_seek_last_pos(struct seq_file *seq)
2207 struct tcp_iter_state *st = seq->private;
2208 int offset = st->offset;
2209 int orig_num = st->num;
2210 void *rc = NULL;
2212 switch (st->state) {
2213 case TCP_SEQ_STATE_OPENREQ:
2214 case TCP_SEQ_STATE_LISTENING:
2215 if (st->bucket >= INET_LHTABLE_SIZE)
2216 break;
2217 st->state = TCP_SEQ_STATE_LISTENING;
2218 rc = listening_get_next(seq, NULL);
2219 while (offset-- && rc)
2220 rc = listening_get_next(seq, rc);
2221 if (rc)
2222 break;
2223 st->bucket = 0;
2224 /* Fallthrough */
2225 case TCP_SEQ_STATE_ESTABLISHED:
2226 case TCP_SEQ_STATE_TIME_WAIT:
2227 st->state = TCP_SEQ_STATE_ESTABLISHED;
2228 if (st->bucket > tcp_hashinfo.ehash_mask)
2229 break;
2230 rc = established_get_first(seq);
2231 while (offset-- && rc)
2232 rc = established_get_next(seq, rc);
2235 st->num = orig_num;
2237 return rc;
2240 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2242 struct tcp_iter_state *st = seq->private;
2243 void *rc;
2245 if (*pos && *pos == st->last_pos) {
2246 rc = tcp_seek_last_pos(seq);
2247 if (rc)
2248 goto out;
2251 st->state = TCP_SEQ_STATE_LISTENING;
2252 st->num = 0;
2253 st->bucket = 0;
2254 st->offset = 0;
2255 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2257 out:
2258 st->last_pos = *pos;
2259 return rc;
2262 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2264 struct tcp_iter_state *st = seq->private;
2265 void *rc = NULL;
2267 if (v == SEQ_START_TOKEN) {
2268 rc = tcp_get_idx(seq, 0);
2269 goto out;
2272 switch (st->state) {
2273 case TCP_SEQ_STATE_OPENREQ:
2274 case TCP_SEQ_STATE_LISTENING:
2275 rc = listening_get_next(seq, v);
2276 if (!rc) {
2277 st->state = TCP_SEQ_STATE_ESTABLISHED;
2278 st->bucket = 0;
2279 st->offset = 0;
2280 rc = established_get_first(seq);
2282 break;
2283 case TCP_SEQ_STATE_ESTABLISHED:
2284 case TCP_SEQ_STATE_TIME_WAIT:
2285 rc = established_get_next(seq, v);
2286 break;
2288 out:
2289 ++*pos;
2290 st->last_pos = *pos;
2291 return rc;
2294 static void tcp_seq_stop(struct seq_file *seq, void *v)
2296 struct tcp_iter_state *st = seq->private;
2298 switch (st->state) {
2299 case TCP_SEQ_STATE_OPENREQ:
2300 if (v) {
2301 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2302 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2304 case TCP_SEQ_STATE_LISTENING:
2305 if (v != SEQ_START_TOKEN)
2306 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2307 break;
2308 case TCP_SEQ_STATE_TIME_WAIT:
2309 case TCP_SEQ_STATE_ESTABLISHED:
2310 if (v)
2311 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2312 break;
2316 static int tcp_seq_open(struct inode *inode, struct file *file)
2318 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2319 struct tcp_iter_state *s;
2320 int err;
2322 err = seq_open_net(inode, file, &afinfo->seq_ops,
2323 sizeof(struct tcp_iter_state));
2324 if (err < 0)
2325 return err;
2327 s = ((struct seq_file *)file->private_data)->private;
2328 s->family = afinfo->family;
2329 s->last_pos = 0;
2330 return 0;
2333 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2335 int rc = 0;
2336 struct proc_dir_entry *p;
2338 afinfo->seq_fops.open = tcp_seq_open;
2339 afinfo->seq_fops.read = seq_read;
2340 afinfo->seq_fops.llseek = seq_lseek;
2341 afinfo->seq_fops.release = seq_release_net;
2343 afinfo->seq_ops.start = tcp_seq_start;
2344 afinfo->seq_ops.next = tcp_seq_next;
2345 afinfo->seq_ops.stop = tcp_seq_stop;
2347 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2348 &afinfo->seq_fops, afinfo);
2349 if (!p)
2350 rc = -ENOMEM;
2351 return rc;
2353 EXPORT_SYMBOL(tcp_proc_register);
2355 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2357 proc_net_remove(net, afinfo->name);
2359 EXPORT_SYMBOL(tcp_proc_unregister);
2361 static void get_openreq4(struct sock *sk, struct request_sock *req,
2362 struct seq_file *f, int i, int uid, int *len)
2364 const struct inet_request_sock *ireq = inet_rsk(req);
2365 int ttd = req->expires - jiffies;
2367 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2368 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
2370 ireq->loc_addr,
2371 ntohs(inet_sk(sk)->inet_sport),
2372 ireq->rmt_addr,
2373 ntohs(ireq->rmt_port),
2374 TCP_SYN_RECV,
2375 0, 0, /* could print option size, but that is af dependent. */
2376 1, /* timers active (only the expire timer) */
2377 jiffies_to_clock_t(ttd),
2378 req->retrans,
2379 uid,
2380 0, /* non standard timer */
2381 0, /* open_requests have no inode */
2382 atomic_read(&sk->sk_refcnt),
2383 req,
2384 len);
2387 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2389 int timer_active;
2390 unsigned long timer_expires;
2391 struct tcp_sock *tp = tcp_sk(sk);
2392 const struct inet_connection_sock *icsk = inet_csk(sk);
2393 struct inet_sock *inet = inet_sk(sk);
2394 __be32 dest = inet->inet_daddr;
2395 __be32 src = inet->inet_rcv_saddr;
2396 __u16 destp = ntohs(inet->inet_dport);
2397 __u16 srcp = ntohs(inet->inet_sport);
2398 int rx_queue;
2400 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2401 timer_active = 1;
2402 timer_expires = icsk->icsk_timeout;
2403 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2404 timer_active = 4;
2405 timer_expires = icsk->icsk_timeout;
2406 } else if (timer_pending(&sk->sk_timer)) {
2407 timer_active = 2;
2408 timer_expires = sk->sk_timer.expires;
2409 } else {
2410 timer_active = 0;
2411 timer_expires = jiffies;
2414 if (sk->sk_state == TCP_LISTEN)
2415 rx_queue = sk->sk_ack_backlog;
2416 else
2418 * because we dont lock socket, we might find a transient negative value
2420 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2422 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2423 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
2424 i, src, srcp, dest, destp, sk->sk_state,
2425 tp->write_seq - tp->snd_una,
2426 rx_queue,
2427 timer_active,
2428 jiffies_to_clock_t(timer_expires - jiffies),
2429 icsk->icsk_retransmits,
2430 sock_i_uid(sk),
2431 icsk->icsk_probes_out,
2432 sock_i_ino(sk),
2433 atomic_read(&sk->sk_refcnt), sk,
2434 jiffies_to_clock_t(icsk->icsk_rto),
2435 jiffies_to_clock_t(icsk->icsk_ack.ato),
2436 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2437 tp->snd_cwnd,
2438 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2439 len);
2442 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2443 struct seq_file *f, int i, int *len)
2445 __be32 dest, src;
2446 __u16 destp, srcp;
2447 int ttd = tw->tw_ttd - jiffies;
2449 if (ttd < 0)
2450 ttd = 0;
2452 dest = tw->tw_daddr;
2453 src = tw->tw_rcv_saddr;
2454 destp = ntohs(tw->tw_dport);
2455 srcp = ntohs(tw->tw_sport);
2457 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2458 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
2459 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2460 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2461 atomic_read(&tw->tw_refcnt), tw, len);
2464 #define TMPSZ 150
2466 static int tcp4_seq_show(struct seq_file *seq, void *v)
2468 struct tcp_iter_state *st;
2469 int len;
2471 if (v == SEQ_START_TOKEN) {
2472 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2473 " sl local_address rem_address st tx_queue "
2474 "rx_queue tr tm->when retrnsmt uid timeout "
2475 "inode");
2476 goto out;
2478 st = seq->private;
2480 switch (st->state) {
2481 case TCP_SEQ_STATE_LISTENING:
2482 case TCP_SEQ_STATE_ESTABLISHED:
2483 get_tcp4_sock(v, seq, st->num, &len);
2484 break;
2485 case TCP_SEQ_STATE_OPENREQ:
2486 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2487 break;
2488 case TCP_SEQ_STATE_TIME_WAIT:
2489 get_timewait4_sock(v, seq, st->num, &len);
2490 break;
2492 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2493 out:
2494 return 0;
2497 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2498 .name = "tcp",
2499 .family = AF_INET,
2500 .seq_fops = {
2501 .owner = THIS_MODULE,
2503 .seq_ops = {
2504 .show = tcp4_seq_show,
2508 static int __net_init tcp4_proc_init_net(struct net *net)
2510 return tcp_proc_register(net, &tcp4_seq_afinfo);
2513 static void __net_exit tcp4_proc_exit_net(struct net *net)
2515 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2518 static struct pernet_operations tcp4_net_ops = {
2519 .init = tcp4_proc_init_net,
2520 .exit = tcp4_proc_exit_net,
2523 int __init tcp4_proc_init(void)
2525 return register_pernet_subsys(&tcp4_net_ops);
2528 void tcp4_proc_exit(void)
2530 unregister_pernet_subsys(&tcp4_net_ops);
2532 #endif /* CONFIG_PROC_FS */
2534 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2536 const struct iphdr *iph = skb_gro_network_header(skb);
2538 switch (skb->ip_summed) {
2539 case CHECKSUM_COMPLETE:
2540 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2541 skb->csum)) {
2542 skb->ip_summed = CHECKSUM_UNNECESSARY;
2543 break;
2546 /* fall through */
2547 case CHECKSUM_NONE:
2548 NAPI_GRO_CB(skb)->flush = 1;
2549 return NULL;
2552 return tcp_gro_receive(head, skb);
2555 int tcp4_gro_complete(struct sk_buff *skb)
2557 const struct iphdr *iph = ip_hdr(skb);
2558 struct tcphdr *th = tcp_hdr(skb);
2560 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2561 iph->saddr, iph->daddr, 0);
2562 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2564 return tcp_gro_complete(skb);
2567 struct proto tcp_prot = {
2568 .name = "TCP",
2569 .owner = THIS_MODULE,
2570 .close = tcp_close,
2571 .connect = tcp_v4_connect,
2572 .disconnect = tcp_disconnect,
2573 .accept = inet_csk_accept,
2574 .ioctl = tcp_ioctl,
2575 .init = tcp_v4_init_sock,
2576 .destroy = tcp_v4_destroy_sock,
2577 .shutdown = tcp_shutdown,
2578 .setsockopt = tcp_setsockopt,
2579 .getsockopt = tcp_getsockopt,
2580 .recvmsg = tcp_recvmsg,
2581 .sendmsg = tcp_sendmsg,
2582 .sendpage = tcp_sendpage,
2583 .backlog_rcv = tcp_v4_do_rcv,
2584 .hash = inet_hash,
2585 .unhash = inet_unhash,
2586 .get_port = inet_csk_get_port,
2587 .enter_memory_pressure = tcp_enter_memory_pressure,
2588 .sockets_allocated = &tcp_sockets_allocated,
2589 .orphan_count = &tcp_orphan_count,
2590 .memory_allocated = &tcp_memory_allocated,
2591 .memory_pressure = &tcp_memory_pressure,
2592 .sysctl_mem = sysctl_tcp_mem,
2593 .sysctl_wmem = sysctl_tcp_wmem,
2594 .sysctl_rmem = sysctl_tcp_rmem,
2595 .max_header = MAX_TCP_HEADER,
2596 .obj_size = sizeof(struct tcp_sock),
2597 .slab_flags = SLAB_DESTROY_BY_RCU,
2598 .twsk_prot = &tcp_timewait_sock_ops,
2599 .rsk_prot = &tcp_request_sock_ops,
2600 .h.hashinfo = &tcp_hashinfo,
2601 .no_autobind = true,
2602 #ifdef CONFIG_COMPAT
2603 .compat_setsockopt = compat_tcp_setsockopt,
2604 .compat_getsockopt = compat_tcp_getsockopt,
2605 #endif
2607 EXPORT_SYMBOL(tcp_prot);
2610 static int __net_init tcp_sk_init(struct net *net)
2612 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2613 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2616 static void __net_exit tcp_sk_exit(struct net *net)
2618 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2621 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2623 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2626 static struct pernet_operations __net_initdata tcp_sk_ops = {
2627 .init = tcp_sk_init,
2628 .exit = tcp_sk_exit,
2629 .exit_batch = tcp_sk_exit_batch,
2632 void __init tcp_v4_init(void)
2634 inet_hashinfo_init(&tcp_hashinfo);
2635 if (register_pernet_subsys(&tcp_sk_ops))
2636 panic("Failed to create the TCP control socket.\n");