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.
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>
67 #include <net/inet_hashtables.h>
69 #include <net/transp_v6.h>
71 #include <net/inet_common.h>
72 #include <net/timewait_sock.h>
74 #include <net/netdma.h>
75 #include <net/secure_seq.h>
77 #include <linux/inet.h>
78 #include <linux/ipv6.h>
79 #include <linux/stddef.h>
80 #include <linux/proc_fs.h>
81 #include <linux/seq_file.h>
83 #include <linux/crypto.h>
84 #include <linux/scatterlist.h>
86 int sysctl_tcp_tw_reuse __read_mostly
;
87 int sysctl_tcp_low_latency __read_mostly
;
88 EXPORT_SYMBOL(sysctl_tcp_low_latency
);
91 #ifdef CONFIG_TCP_MD5SIG
92 static struct tcp_md5sig_key
*tcp_v4_md5_do_lookup(struct sock
*sk
,
94 static int tcp_v4_md5_hash_hdr(char *md5_hash
, struct tcp_md5sig_key
*key
,
95 __be32 daddr
, __be32 saddr
, struct tcphdr
*th
);
98 struct tcp_md5sig_key
*tcp_v4_md5_do_lookup(struct sock
*sk
, __be32 addr
)
104 struct inet_hashinfo tcp_hashinfo
;
105 EXPORT_SYMBOL(tcp_hashinfo
);
107 static inline __u32
tcp_v4_init_sequence(struct sk_buff
*skb
)
109 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
112 tcp_hdr(skb
)->source
);
115 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
)
117 const struct tcp_timewait_sock
*tcptw
= tcp_twsk(sktw
);
118 struct tcp_sock
*tp
= tcp_sk(sk
);
120 /* With PAWS, it is safe from the viewpoint
121 of data integrity. Even without PAWS it is safe provided sequence
122 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
124 Actually, the idea is close to VJ's one, only timestamp cache is
125 held not per host, but per port pair and TW bucket is used as state
128 If TW bucket has been already destroyed we fall back to VJ's scheme
129 and use initial timestamp retrieved from peer table.
131 if (tcptw
->tw_ts_recent_stamp
&&
132 (twp
== NULL
|| (sysctl_tcp_tw_reuse
&&
133 get_seconds() - tcptw
->tw_ts_recent_stamp
> 1))) {
134 tp
->write_seq
= tcptw
->tw_snd_nxt
+ 65535 + 2;
135 if (tp
->write_seq
== 0)
137 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
138 tp
->rx_opt
.ts_recent_stamp
= tcptw
->tw_ts_recent_stamp
;
145 EXPORT_SYMBOL_GPL(tcp_twsk_unique
);
147 /* This will initiate an outgoing connection. */
148 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
150 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
151 struct inet_sock
*inet
= inet_sk(sk
);
152 struct tcp_sock
*tp
= tcp_sk(sk
);
153 __be16 orig_sport
, orig_dport
;
154 __be32 daddr
, nexthop
;
158 struct ip_options_rcu
*inet_opt
;
160 if (addr_len
< sizeof(struct sockaddr_in
))
163 if (usin
->sin_family
!= AF_INET
)
164 return -EAFNOSUPPORT
;
166 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
167 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
168 sock_owned_by_user(sk
));
169 if (inet_opt
&& inet_opt
->opt
.srr
) {
172 nexthop
= inet_opt
->opt
.faddr
;
175 orig_sport
= inet
->inet_sport
;
176 orig_dport
= usin
->sin_port
;
177 fl4
= &inet
->cork
.fl
.u
.ip4
;
178 rt
= ip_route_connect(fl4
, nexthop
, inet
->inet_saddr
,
179 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
181 orig_sport
, orig_dport
, sk
, true);
184 if (err
== -ENETUNREACH
)
185 IP_INC_STATS_BH(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
189 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
194 if (!inet_opt
|| !inet_opt
->opt
.srr
)
197 if (!inet
->inet_saddr
)
198 inet
->inet_saddr
= fl4
->saddr
;
199 inet
->inet_rcv_saddr
= inet
->inet_saddr
;
201 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->inet_daddr
!= daddr
) {
202 /* Reset inherited state */
203 tp
->rx_opt
.ts_recent
= 0;
204 tp
->rx_opt
.ts_recent_stamp
= 0;
208 if (tcp_death_row
.sysctl_tw_recycle
&&
209 !tp
->rx_opt
.ts_recent_stamp
&& fl4
->daddr
== daddr
) {
210 struct inet_peer
*peer
= rt_get_peer(rt
, fl4
->daddr
);
212 * VJ's idea. We save last timestamp seen from
213 * the destination in peer table, when entering state
214 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
215 * when trying new connection.
218 inet_peer_refcheck(peer
);
219 if ((u32
)get_seconds() - peer
->tcp_ts_stamp
<= TCP_PAWS_MSL
) {
220 tp
->rx_opt
.ts_recent_stamp
= peer
->tcp_ts_stamp
;
221 tp
->rx_opt
.ts_recent
= peer
->tcp_ts
;
226 inet
->inet_dport
= usin
->sin_port
;
227 inet
->inet_daddr
= daddr
;
229 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
231 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
233 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
235 /* Socket identity is still unknown (sport may be zero).
236 * However we set state to SYN-SENT and not releasing socket
237 * lock select source port, enter ourselves into the hash tables and
238 * complete initialization after this.
240 tcp_set_state(sk
, TCP_SYN_SENT
);
241 err
= inet_hash_connect(&tcp_death_row
, sk
);
245 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
246 inet
->inet_sport
, inet
->inet_dport
, sk
);
252 /* OK, now commit destination to socket. */
253 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
254 sk_setup_caps(sk
, &rt
->dst
);
257 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
262 inet
->inet_id
= tp
->write_seq
^ jiffies
;
264 err
= tcp_connect(sk
);
273 * This unhashes the socket and releases the local port,
276 tcp_set_state(sk
, TCP_CLOSE
);
278 sk
->sk_route_caps
= 0;
279 inet
->inet_dport
= 0;
282 EXPORT_SYMBOL(tcp_v4_connect
);
285 * This routine does path mtu discovery as defined in RFC1191.
287 static void do_pmtu_discovery(struct sock
*sk
, const struct iphdr
*iph
, u32 mtu
)
289 struct dst_entry
*dst
;
290 struct inet_sock
*inet
= inet_sk(sk
);
292 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
293 * send out by Linux are always <576bytes so they should go through
296 if (sk
->sk_state
== TCP_LISTEN
)
299 /* We don't check in the destentry if pmtu discovery is forbidden
300 * on this route. We just assume that no packet_to_big packets
301 * are send back when pmtu discovery is not active.
302 * There is a small race when the user changes this flag in the
303 * route, but I think that's acceptable.
305 if ((dst
= __sk_dst_check(sk
, 0)) == NULL
)
308 dst
->ops
->update_pmtu(dst
, mtu
);
310 /* Something is about to be wrong... Remember soft error
311 * for the case, if this connection will not able to recover.
313 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
314 sk
->sk_err_soft
= EMSGSIZE
;
318 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
319 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
320 tcp_sync_mss(sk
, mtu
);
322 /* Resend the TCP packet because it's
323 * clear that the old packet has been
324 * dropped. This is the new "fast" path mtu
327 tcp_simple_retransmit(sk
);
328 } /* else let the usual retransmit timer handle it */
332 * This routine is called by the ICMP module when it gets some
333 * sort of error condition. If err < 0 then the socket should
334 * be closed and the error returned to the user. If err > 0
335 * it's just the icmp type << 8 | icmp code. After adjustment
336 * header points to the first 8 bytes of the tcp header. We need
337 * to find the appropriate port.
339 * The locking strategy used here is very "optimistic". When
340 * someone else accesses the socket the ICMP is just dropped
341 * and for some paths there is no check at all.
342 * A more general error queue to queue errors for later handling
343 * is probably better.
347 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
349 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
350 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
351 struct inet_connection_sock
*icsk
;
353 struct inet_sock
*inet
;
354 const int type
= icmp_hdr(icmp_skb
)->type
;
355 const int code
= icmp_hdr(icmp_skb
)->code
;
361 struct net
*net
= dev_net(icmp_skb
->dev
);
363 if (icmp_skb
->len
< (iph
->ihl
<< 2) + 8) {
364 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
368 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
369 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
371 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
374 if (sk
->sk_state
== TCP_TIME_WAIT
) {
375 inet_twsk_put(inet_twsk(sk
));
380 /* If too many ICMPs get dropped on busy
381 * servers this needs to be solved differently.
383 if (sock_owned_by_user(sk
))
384 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
386 if (sk
->sk_state
== TCP_CLOSE
)
389 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
390 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
396 seq
= ntohl(th
->seq
);
397 if (sk
->sk_state
!= TCP_LISTEN
&&
398 !between(seq
, tp
->snd_una
, tp
->snd_nxt
)) {
399 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
404 case ICMP_SOURCE_QUENCH
:
405 /* Just silently ignore these. */
407 case ICMP_PARAMETERPROB
:
410 case ICMP_DEST_UNREACH
:
411 if (code
> NR_ICMP_UNREACH
)
414 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
415 if (!sock_owned_by_user(sk
))
416 do_pmtu_discovery(sk
, iph
, info
);
420 err
= icmp_err_convert
[code
].errno
;
421 /* check if icmp_skb allows revert of backoff
422 * (see draft-zimmermann-tcp-lcd) */
423 if (code
!= ICMP_NET_UNREACH
&& code
!= ICMP_HOST_UNREACH
)
425 if (seq
!= tp
->snd_una
|| !icsk
->icsk_retransmits
||
429 if (sock_owned_by_user(sk
))
432 icsk
->icsk_backoff
--;
433 inet_csk(sk
)->icsk_rto
= (tp
->srtt
? __tcp_set_rto(tp
) :
434 TCP_TIMEOUT_INIT
) << icsk
->icsk_backoff
;
437 skb
= tcp_write_queue_head(sk
);
440 remaining
= icsk
->icsk_rto
- min(icsk
->icsk_rto
,
441 tcp_time_stamp
- TCP_SKB_CB(skb
)->when
);
444 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
445 remaining
, TCP_RTO_MAX
);
447 /* RTO revert clocked out retransmission.
448 * Will retransmit now */
449 tcp_retransmit_timer(sk
);
453 case ICMP_TIME_EXCEEDED
:
460 switch (sk
->sk_state
) {
461 struct request_sock
*req
, **prev
;
463 if (sock_owned_by_user(sk
))
466 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
467 iph
->daddr
, iph
->saddr
);
471 /* ICMPs are not backlogged, hence we cannot get
472 an established socket here.
476 if (seq
!= tcp_rsk(req
)->snt_isn
) {
477 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
482 * Still in SYN_RECV, just remove it silently.
483 * There is no good way to pass the error to the newly
484 * created socket, and POSIX does not want network
485 * errors returned from accept().
487 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
491 case TCP_SYN_RECV
: /* Cannot happen.
492 It can f.e. if SYNs crossed.
494 if (!sock_owned_by_user(sk
)) {
497 sk
->sk_error_report(sk
);
501 sk
->sk_err_soft
= err
;
506 /* If we've already connected we will keep trying
507 * until we time out, or the user gives up.
509 * rfc1122 4.2.3.9 allows to consider as hard errors
510 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
511 * but it is obsoleted by pmtu discovery).
513 * Note, that in modern internet, where routing is unreliable
514 * and in each dark corner broken firewalls sit, sending random
515 * errors ordered by their masters even this two messages finally lose
516 * their original sense (even Linux sends invalid PORT_UNREACHs)
518 * Now we are in compliance with RFCs.
523 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
525 sk
->sk_error_report(sk
);
526 } else { /* Only an error on timeout */
527 sk
->sk_err_soft
= err
;
535 static void __tcp_v4_send_check(struct sk_buff
*skb
,
536 __be32 saddr
, __be32 daddr
)
538 struct tcphdr
*th
= tcp_hdr(skb
);
540 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
541 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
542 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
543 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
545 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
552 /* This routine computes an IPv4 TCP checksum. */
553 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
555 struct inet_sock
*inet
= inet_sk(sk
);
557 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
559 EXPORT_SYMBOL(tcp_v4_send_check
);
561 int tcp_v4_gso_send_check(struct sk_buff
*skb
)
563 const struct iphdr
*iph
;
566 if (!pskb_may_pull(skb
, sizeof(*th
)))
573 skb
->ip_summed
= CHECKSUM_PARTIAL
;
574 __tcp_v4_send_check(skb
, iph
->saddr
, iph
->daddr
);
579 * This routine will send an RST to the other tcp.
581 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
583 * Answer: if a packet caused RST, it is not for a socket
584 * existing in our system, if it is matched to a socket,
585 * it is just duplicate segment or bug in other side's TCP.
586 * So that we build reply only basing on parameters
587 * arrived with segment.
588 * Exception: precedence violation. We do not implement it in any case.
591 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
593 struct tcphdr
*th
= tcp_hdr(skb
);
596 #ifdef CONFIG_TCP_MD5SIG
597 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
600 struct ip_reply_arg arg
;
601 #ifdef CONFIG_TCP_MD5SIG
602 struct tcp_md5sig_key
*key
;
606 /* Never send a reset in response to a reset. */
610 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
613 /* Swap the send and the receive. */
614 memset(&rep
, 0, sizeof(rep
));
615 rep
.th
.dest
= th
->source
;
616 rep
.th
.source
= th
->dest
;
617 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
621 rep
.th
.seq
= th
->ack_seq
;
624 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
625 skb
->len
- (th
->doff
<< 2));
628 memset(&arg
, 0, sizeof(arg
));
629 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
630 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
632 #ifdef CONFIG_TCP_MD5SIG
633 key
= sk
? tcp_v4_md5_do_lookup(sk
, ip_hdr(skb
)->daddr
) : NULL
;
635 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
637 (TCPOPT_MD5SIG
<< 8) |
639 /* Update length and the length the header thinks exists */
640 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
641 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
643 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
644 key
, ip_hdr(skb
)->saddr
,
645 ip_hdr(skb
)->daddr
, &rep
.th
);
648 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
649 ip_hdr(skb
)->saddr
, /* XXX */
650 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
651 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
652 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
654 net
= dev_net(skb_dst(skb
)->dev
);
655 ip_send_reply(net
->ipv4
.tcp_sock
, skb
, ip_hdr(skb
)->saddr
,
656 &arg
, arg
.iov
[0].iov_len
);
658 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
659 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
662 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
663 outside socket context is ugly, certainly. What can I do?
666 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
667 u32 win
, u32 ts
, int oif
,
668 struct tcp_md5sig_key
*key
,
671 struct tcphdr
*th
= tcp_hdr(skb
);
674 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
675 #ifdef CONFIG_TCP_MD5SIG
676 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
680 struct ip_reply_arg arg
;
681 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
683 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
684 memset(&arg
, 0, sizeof(arg
));
686 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
687 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
689 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
690 (TCPOPT_TIMESTAMP
<< 8) |
692 rep
.opt
[1] = htonl(tcp_time_stamp
);
693 rep
.opt
[2] = htonl(ts
);
694 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
697 /* Swap the send and the receive. */
698 rep
.th
.dest
= th
->source
;
699 rep
.th
.source
= th
->dest
;
700 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
701 rep
.th
.seq
= htonl(seq
);
702 rep
.th
.ack_seq
= htonl(ack
);
704 rep
.th
.window
= htons(win
);
706 #ifdef CONFIG_TCP_MD5SIG
708 int offset
= (ts
) ? 3 : 0;
710 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
712 (TCPOPT_MD5SIG
<< 8) |
714 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
715 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
717 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
718 key
, ip_hdr(skb
)->saddr
,
719 ip_hdr(skb
)->daddr
, &rep
.th
);
722 arg
.flags
= reply_flags
;
723 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
724 ip_hdr(skb
)->saddr
, /* XXX */
725 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
726 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
728 arg
.bound_dev_if
= oif
;
730 ip_send_reply(net
->ipv4
.tcp_sock
, skb
, ip_hdr(skb
)->saddr
,
731 &arg
, arg
.iov
[0].iov_len
);
733 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
736 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
738 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
739 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
741 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
742 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
745 tcp_twsk_md5_key(tcptw
),
746 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0
752 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
753 struct request_sock
*req
)
755 tcp_v4_send_ack(skb
, tcp_rsk(req
)->snt_isn
+ 1,
756 tcp_rsk(req
)->rcv_isn
+ 1, req
->rcv_wnd
,
759 tcp_v4_md5_do_lookup(sk
, ip_hdr(skb
)->daddr
),
760 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0);
764 * Send a SYN-ACK after having received a SYN.
765 * This still operates on a request_sock only, not on a big
768 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
769 struct request_sock
*req
,
770 struct request_values
*rvp
)
772 const struct inet_request_sock
*ireq
= inet_rsk(req
);
775 struct sk_buff
* skb
;
777 /* First, grab a route. */
778 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
781 skb
= tcp_make_synack(sk
, dst
, req
, rvp
);
784 __tcp_v4_send_check(skb
, ireq
->loc_addr
, ireq
->rmt_addr
);
786 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
789 err
= net_xmit_eval(err
);
796 static int tcp_v4_rtx_synack(struct sock
*sk
, struct request_sock
*req
,
797 struct request_values
*rvp
)
799 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
800 return tcp_v4_send_synack(sk
, NULL
, req
, rvp
);
804 * IPv4 request_sock destructor.
806 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
808 kfree(inet_rsk(req
)->opt
);
812 * Return 1 if a syncookie should be sent
814 int tcp_syn_flood_action(struct sock
*sk
,
815 const struct sk_buff
*skb
,
818 const char *msg
= "Dropping request";
820 struct listen_sock
*lopt
;
824 #ifdef CONFIG_SYN_COOKIES
825 if (sysctl_tcp_syncookies
) {
826 msg
= "Sending cookies";
828 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
831 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
833 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
834 if (!lopt
->synflood_warned
) {
835 lopt
->synflood_warned
= 1;
836 pr_info("%s: Possible SYN flooding on port %d. %s. "
837 " Check SNMP counters.\n",
838 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
842 EXPORT_SYMBOL(tcp_syn_flood_action
);
845 * Save and compile IPv4 options into the request_sock if needed.
847 static struct ip_options_rcu
*tcp_v4_save_options(struct sock
*sk
,
850 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
851 struct ip_options_rcu
*dopt
= NULL
;
853 if (opt
&& opt
->optlen
) {
854 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
856 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
858 if (ip_options_echo(&dopt
->opt
, skb
)) {
867 #ifdef CONFIG_TCP_MD5SIG
869 * RFC2385 MD5 checksumming requires a mapping of
870 * IP address->MD5 Key.
871 * We need to maintain these in the sk structure.
874 /* Find the Key structure for an address. */
875 static struct tcp_md5sig_key
*
876 tcp_v4_md5_do_lookup(struct sock
*sk
, __be32 addr
)
878 struct tcp_sock
*tp
= tcp_sk(sk
);
881 if (!tp
->md5sig_info
|| !tp
->md5sig_info
->entries4
)
883 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++) {
884 if (tp
->md5sig_info
->keys4
[i
].addr
== addr
)
885 return &tp
->md5sig_info
->keys4
[i
].base
;
890 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
891 struct sock
*addr_sk
)
893 return tcp_v4_md5_do_lookup(sk
, inet_sk(addr_sk
)->inet_daddr
);
895 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
897 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
898 struct request_sock
*req
)
900 return tcp_v4_md5_do_lookup(sk
, inet_rsk(req
)->rmt_addr
);
903 /* This can be called on a newly created socket, from other files */
904 int tcp_v4_md5_do_add(struct sock
*sk
, __be32 addr
,
905 u8
*newkey
, u8 newkeylen
)
907 /* Add Key to the list */
908 struct tcp_md5sig_key
*key
;
909 struct tcp_sock
*tp
= tcp_sk(sk
);
910 struct tcp4_md5sig_key
*keys
;
912 key
= tcp_v4_md5_do_lookup(sk
, addr
);
914 /* Pre-existing entry - just update that one. */
917 key
->keylen
= newkeylen
;
919 struct tcp_md5sig_info
*md5sig
;
921 if (!tp
->md5sig_info
) {
922 tp
->md5sig_info
= kzalloc(sizeof(*tp
->md5sig_info
),
924 if (!tp
->md5sig_info
) {
928 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
930 if (tcp_alloc_md5sig_pool(sk
) == NULL
) {
934 md5sig
= tp
->md5sig_info
;
936 if (md5sig
->alloced4
== md5sig
->entries4
) {
937 keys
= kmalloc((sizeof(*keys
) *
938 (md5sig
->entries4
+ 1)), GFP_ATOMIC
);
941 tcp_free_md5sig_pool();
945 if (md5sig
->entries4
)
946 memcpy(keys
, md5sig
->keys4
,
947 sizeof(*keys
) * md5sig
->entries4
);
949 /* Free old key list, and reference new one */
950 kfree(md5sig
->keys4
);
951 md5sig
->keys4
= keys
;
955 md5sig
->keys4
[md5sig
->entries4
- 1].addr
= addr
;
956 md5sig
->keys4
[md5sig
->entries4
- 1].base
.key
= newkey
;
957 md5sig
->keys4
[md5sig
->entries4
- 1].base
.keylen
= newkeylen
;
961 EXPORT_SYMBOL(tcp_v4_md5_do_add
);
963 static int tcp_v4_md5_add_func(struct sock
*sk
, struct sock
*addr_sk
,
964 u8
*newkey
, u8 newkeylen
)
966 return tcp_v4_md5_do_add(sk
, inet_sk(addr_sk
)->inet_daddr
,
970 int tcp_v4_md5_do_del(struct sock
*sk
, __be32 addr
)
972 struct tcp_sock
*tp
= tcp_sk(sk
);
975 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++) {
976 if (tp
->md5sig_info
->keys4
[i
].addr
== addr
) {
978 kfree(tp
->md5sig_info
->keys4
[i
].base
.key
);
979 tp
->md5sig_info
->entries4
--;
981 if (tp
->md5sig_info
->entries4
== 0) {
982 kfree(tp
->md5sig_info
->keys4
);
983 tp
->md5sig_info
->keys4
= NULL
;
984 tp
->md5sig_info
->alloced4
= 0;
985 } else if (tp
->md5sig_info
->entries4
!= i
) {
986 /* Need to do some manipulation */
987 memmove(&tp
->md5sig_info
->keys4
[i
],
988 &tp
->md5sig_info
->keys4
[i
+1],
989 (tp
->md5sig_info
->entries4
- i
) *
990 sizeof(struct tcp4_md5sig_key
));
992 tcp_free_md5sig_pool();
998 EXPORT_SYMBOL(tcp_v4_md5_do_del
);
1000 static void tcp_v4_clear_md5_list(struct sock
*sk
)
1002 struct tcp_sock
*tp
= tcp_sk(sk
);
1004 /* Free each key, then the set of key keys,
1005 * the crypto element, and then decrement our
1006 * hold on the last resort crypto.
1008 if (tp
->md5sig_info
->entries4
) {
1010 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++)
1011 kfree(tp
->md5sig_info
->keys4
[i
].base
.key
);
1012 tp
->md5sig_info
->entries4
= 0;
1013 tcp_free_md5sig_pool();
1015 if (tp
->md5sig_info
->keys4
) {
1016 kfree(tp
->md5sig_info
->keys4
);
1017 tp
->md5sig_info
->keys4
= NULL
;
1018 tp
->md5sig_info
->alloced4
= 0;
1022 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1025 struct tcp_md5sig cmd
;
1026 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1029 if (optlen
< sizeof(cmd
))
1032 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1035 if (sin
->sin_family
!= AF_INET
)
1038 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
) {
1039 if (!tcp_sk(sk
)->md5sig_info
)
1041 return tcp_v4_md5_do_del(sk
, sin
->sin_addr
.s_addr
);
1044 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1047 if (!tcp_sk(sk
)->md5sig_info
) {
1048 struct tcp_sock
*tp
= tcp_sk(sk
);
1049 struct tcp_md5sig_info
*p
;
1051 p
= kzalloc(sizeof(*p
), sk
->sk_allocation
);
1055 tp
->md5sig_info
= p
;
1056 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
1059 newkey
= kmemdup(cmd
.tcpm_key
, cmd
.tcpm_keylen
, sk
->sk_allocation
);
1062 return tcp_v4_md5_do_add(sk
, sin
->sin_addr
.s_addr
,
1063 newkey
, cmd
.tcpm_keylen
);
1066 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1067 __be32 daddr
, __be32 saddr
, int nbytes
)
1069 struct tcp4_pseudohdr
*bp
;
1070 struct scatterlist sg
;
1072 bp
= &hp
->md5_blk
.ip4
;
1075 * 1. the TCP pseudo-header (in the order: source IP address,
1076 * destination IP address, zero-padded protocol number, and
1082 bp
->protocol
= IPPROTO_TCP
;
1083 bp
->len
= cpu_to_be16(nbytes
);
1085 sg_init_one(&sg
, bp
, sizeof(*bp
));
1086 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1089 static int tcp_v4_md5_hash_hdr(char *md5_hash
, struct tcp_md5sig_key
*key
,
1090 __be32 daddr
, __be32 saddr
, struct tcphdr
*th
)
1092 struct tcp_md5sig_pool
*hp
;
1093 struct hash_desc
*desc
;
1095 hp
= tcp_get_md5sig_pool();
1097 goto clear_hash_noput
;
1098 desc
= &hp
->md5_desc
;
1100 if (crypto_hash_init(desc
))
1102 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1104 if (tcp_md5_hash_header(hp
, th
))
1106 if (tcp_md5_hash_key(hp
, key
))
1108 if (crypto_hash_final(desc
, md5_hash
))
1111 tcp_put_md5sig_pool();
1115 tcp_put_md5sig_pool();
1117 memset(md5_hash
, 0, 16);
1121 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1122 struct sock
*sk
, struct request_sock
*req
,
1123 struct sk_buff
*skb
)
1125 struct tcp_md5sig_pool
*hp
;
1126 struct hash_desc
*desc
;
1127 struct tcphdr
*th
= tcp_hdr(skb
);
1128 __be32 saddr
, daddr
;
1131 saddr
= inet_sk(sk
)->inet_saddr
;
1132 daddr
= inet_sk(sk
)->inet_daddr
;
1134 saddr
= inet_rsk(req
)->loc_addr
;
1135 daddr
= inet_rsk(req
)->rmt_addr
;
1137 const struct iphdr
*iph
= ip_hdr(skb
);
1142 hp
= tcp_get_md5sig_pool();
1144 goto clear_hash_noput
;
1145 desc
= &hp
->md5_desc
;
1147 if (crypto_hash_init(desc
))
1150 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1152 if (tcp_md5_hash_header(hp
, th
))
1154 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1156 if (tcp_md5_hash_key(hp
, key
))
1158 if (crypto_hash_final(desc
, md5_hash
))
1161 tcp_put_md5sig_pool();
1165 tcp_put_md5sig_pool();
1167 memset(md5_hash
, 0, 16);
1170 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1172 static int tcp_v4_inbound_md5_hash(struct sock
*sk
, struct sk_buff
*skb
)
1175 * This gets called for each TCP segment that arrives
1176 * so we want to be efficient.
1177 * We have 3 drop cases:
1178 * o No MD5 hash and one expected.
1179 * o MD5 hash and we're not expecting one.
1180 * o MD5 hash and its wrong.
1182 __u8
*hash_location
= NULL
;
1183 struct tcp_md5sig_key
*hash_expected
;
1184 const struct iphdr
*iph
= ip_hdr(skb
);
1185 struct tcphdr
*th
= tcp_hdr(skb
);
1187 unsigned char newhash
[16];
1189 hash_expected
= tcp_v4_md5_do_lookup(sk
, iph
->saddr
);
1190 hash_location
= tcp_parse_md5sig_option(th
);
1192 /* We've parsed the options - do we have a hash? */
1193 if (!hash_expected
&& !hash_location
)
1196 if (hash_expected
&& !hash_location
) {
1197 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1201 if (!hash_expected
&& hash_location
) {
1202 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1206 /* Okay, so this is hash_expected and hash_location -
1207 * so we need to calculate the checksum.
1209 genhash
= tcp_v4_md5_hash_skb(newhash
,
1213 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1214 if (net_ratelimit()) {
1215 printk(KERN_INFO
"MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1216 &iph
->saddr
, ntohs(th
->source
),
1217 &iph
->daddr
, ntohs(th
->dest
),
1218 genhash
? " tcp_v4_calc_md5_hash failed" : "");
1227 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1229 .obj_size
= sizeof(struct tcp_request_sock
),
1230 .rtx_syn_ack
= tcp_v4_rtx_synack
,
1231 .send_ack
= tcp_v4_reqsk_send_ack
,
1232 .destructor
= tcp_v4_reqsk_destructor
,
1233 .send_reset
= tcp_v4_send_reset
,
1234 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1237 #ifdef CONFIG_TCP_MD5SIG
1238 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1239 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1240 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1244 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1246 struct tcp_extend_values tmp_ext
;
1247 struct tcp_options_received tmp_opt
;
1249 struct request_sock
*req
;
1250 struct inet_request_sock
*ireq
;
1251 struct tcp_sock
*tp
= tcp_sk(sk
);
1252 struct dst_entry
*dst
= NULL
;
1253 __be32 saddr
= ip_hdr(skb
)->saddr
;
1254 __be32 daddr
= ip_hdr(skb
)->daddr
;
1255 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1256 int want_cookie
= 0;
1258 /* Never answer to SYNs send to broadcast or multicast */
1259 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1262 /* TW buckets are converted to open requests without
1263 * limitations, they conserve resources and peer is
1264 * evidently real one.
1266 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
1267 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCP");
1272 /* Accept backlog is full. If we have already queued enough
1273 * of warm entries in syn queue, drop request. It is better than
1274 * clogging syn queue with openreqs with exponentially increasing
1277 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1)
1280 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1284 #ifdef CONFIG_TCP_MD5SIG
1285 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1288 tcp_clear_options(&tmp_opt
);
1289 tmp_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
1290 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
1291 tcp_parse_options(skb
, &tmp_opt
, &hash_location
, 0);
1293 if (tmp_opt
.cookie_plus
> 0 &&
1294 tmp_opt
.saw_tstamp
&&
1295 !tp
->rx_opt
.cookie_out_never
&&
1296 (sysctl_tcp_cookie_size
> 0 ||
1297 (tp
->cookie_values
!= NULL
&&
1298 tp
->cookie_values
->cookie_desired
> 0))) {
1300 u32
*mess
= &tmp_ext
.cookie_bakery
[COOKIE_DIGEST_WORDS
];
1301 int l
= tmp_opt
.cookie_plus
- TCPOLEN_COOKIE_BASE
;
1303 if (tcp_cookie_generator(&tmp_ext
.cookie_bakery
[0]) != 0)
1304 goto drop_and_release
;
1306 /* Secret recipe starts with IP addresses */
1307 *mess
++ ^= (__force u32
)daddr
;
1308 *mess
++ ^= (__force u32
)saddr
;
1310 /* plus variable length Initiator Cookie */
1313 *c
++ ^= *hash_location
++;
1315 want_cookie
= 0; /* not our kind of cookie */
1316 tmp_ext
.cookie_out_never
= 0; /* false */
1317 tmp_ext
.cookie_plus
= tmp_opt
.cookie_plus
;
1318 } else if (!tp
->rx_opt
.cookie_in_always
) {
1319 /* redundant indications, but ensure initialization. */
1320 tmp_ext
.cookie_out_never
= 1; /* true */
1321 tmp_ext
.cookie_plus
= 0;
1323 goto drop_and_release
;
1325 tmp_ext
.cookie_in_always
= tp
->rx_opt
.cookie_in_always
;
1327 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1328 tcp_clear_options(&tmp_opt
);
1330 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1331 tcp_openreq_init(req
, &tmp_opt
, skb
);
1333 ireq
= inet_rsk(req
);
1334 ireq
->loc_addr
= daddr
;
1335 ireq
->rmt_addr
= saddr
;
1336 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1337 ireq
->opt
= tcp_v4_save_options(sk
, skb
);
1339 if (security_inet_conn_request(sk
, skb
, req
))
1342 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1343 TCP_ECN_create_request(req
, tcp_hdr(skb
));
1346 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1347 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1349 struct inet_peer
*peer
= NULL
;
1352 /* VJ's idea. We save last timestamp seen
1353 * from the destination in peer table, when entering
1354 * state TIME-WAIT, and check against it before
1355 * accepting new connection request.
1357 * If "isn" is not zero, this request hit alive
1358 * timewait bucket, so that all the necessary checks
1359 * are made in the function processing timewait state.
1361 if (tmp_opt
.saw_tstamp
&&
1362 tcp_death_row
.sysctl_tw_recycle
&&
1363 (dst
= inet_csk_route_req(sk
, &fl4
, req
)) != NULL
&&
1364 fl4
.daddr
== saddr
&&
1365 (peer
= rt_get_peer((struct rtable
*)dst
, fl4
.daddr
)) != NULL
) {
1366 inet_peer_refcheck(peer
);
1367 if ((u32
)get_seconds() - peer
->tcp_ts_stamp
< TCP_PAWS_MSL
&&
1368 (s32
)(peer
->tcp_ts
- req
->ts_recent
) >
1370 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1371 goto drop_and_release
;
1374 /* Kill the following clause, if you dislike this way. */
1375 else if (!sysctl_tcp_syncookies
&&
1376 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1377 (sysctl_max_syn_backlog
>> 2)) &&
1378 (!peer
|| !peer
->tcp_ts_stamp
) &&
1379 (!dst
|| !dst_metric(dst
, RTAX_RTT
))) {
1380 /* Without syncookies last quarter of
1381 * backlog is filled with destinations,
1382 * proven to be alive.
1383 * It means that we continue to communicate
1384 * to destinations, already remembered
1385 * to the moment of synflood.
1387 LIMIT_NETDEBUG(KERN_DEBUG
"TCP: drop open request from %pI4/%u\n",
1388 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1389 goto drop_and_release
;
1392 isn
= tcp_v4_init_sequence(skb
);
1394 tcp_rsk(req
)->snt_isn
= isn
;
1395 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1397 if (tcp_v4_send_synack(sk
, dst
, req
,
1398 (struct request_values
*)&tmp_ext
) ||
1402 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1412 EXPORT_SYMBOL(tcp_v4_conn_request
);
1416 * The three way handshake has completed - we got a valid synack -
1417 * now create the new socket.
1419 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1420 struct request_sock
*req
,
1421 struct dst_entry
*dst
)
1423 struct inet_request_sock
*ireq
;
1424 struct inet_sock
*newinet
;
1425 struct tcp_sock
*newtp
;
1427 #ifdef CONFIG_TCP_MD5SIG
1428 struct tcp_md5sig_key
*key
;
1430 struct ip_options_rcu
*inet_opt
;
1432 if (sk_acceptq_is_full(sk
))
1435 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1439 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1441 newtp
= tcp_sk(newsk
);
1442 newinet
= inet_sk(newsk
);
1443 ireq
= inet_rsk(req
);
1444 newinet
->inet_daddr
= ireq
->rmt_addr
;
1445 newinet
->inet_rcv_saddr
= ireq
->loc_addr
;
1446 newinet
->inet_saddr
= ireq
->loc_addr
;
1447 inet_opt
= ireq
->opt
;
1448 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1450 newinet
->mc_index
= inet_iif(skb
);
1451 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1452 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1454 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1455 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1457 if (!dst
&& (dst
= inet_csk_route_child_sock(sk
, newsk
, req
)) == NULL
)
1460 sk_setup_caps(newsk
, dst
);
1462 tcp_mtup_init(newsk
);
1463 tcp_sync_mss(newsk
, dst_mtu(dst
));
1464 newtp
->advmss
= dst_metric_advmss(dst
);
1465 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1466 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1467 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1469 tcp_initialize_rcv_mss(newsk
);
1470 if (tcp_rsk(req
)->snt_synack
)
1471 tcp_valid_rtt_meas(newsk
,
1472 tcp_time_stamp
- tcp_rsk(req
)->snt_synack
);
1473 newtp
->total_retrans
= req
->retrans
;
1475 #ifdef CONFIG_TCP_MD5SIG
1476 /* Copy over the MD5 key from the original socket */
1477 key
= tcp_v4_md5_do_lookup(sk
, newinet
->inet_daddr
);
1480 * We're using one, so create a matching key
1481 * on the newsk structure. If we fail to get
1482 * memory, then we end up not copying the key
1485 char *newkey
= kmemdup(key
->key
, key
->keylen
, GFP_ATOMIC
);
1487 tcp_v4_md5_do_add(newsk
, newinet
->inet_daddr
,
1488 newkey
, key
->keylen
);
1489 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1493 if (__inet_inherit_port(sk
, newsk
) < 0)
1495 __inet_hash_nolisten(newsk
, NULL
);
1500 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1504 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1510 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1512 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1514 struct tcphdr
*th
= tcp_hdr(skb
);
1515 const struct iphdr
*iph
= ip_hdr(skb
);
1517 struct request_sock
**prev
;
1518 /* Find possible connection requests. */
1519 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1520 iph
->saddr
, iph
->daddr
);
1522 return tcp_check_req(sk
, skb
, req
, prev
);
1524 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1525 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1528 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1532 inet_twsk_put(inet_twsk(nsk
));
1536 #ifdef CONFIG_SYN_COOKIES
1538 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1543 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1545 const struct iphdr
*iph
= ip_hdr(skb
);
1547 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1548 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1549 iph
->daddr
, skb
->csum
)) {
1550 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1555 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1556 skb
->len
, IPPROTO_TCP
, 0);
1558 if (skb
->len
<= 76) {
1559 return __skb_checksum_complete(skb
);
1565 /* The socket must have it's spinlock held when we get
1568 * We have a potential double-lock case here, so even when
1569 * doing backlog processing we use the BH locking scheme.
1570 * This is because we cannot sleep with the original spinlock
1573 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1576 #ifdef CONFIG_TCP_MD5SIG
1578 * We really want to reject the packet as early as possible
1580 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1581 * o There is an MD5 option and we're not expecting one
1583 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1587 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1588 sock_rps_save_rxhash(sk
, skb
->rxhash
);
1589 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1596 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1599 if (sk
->sk_state
== TCP_LISTEN
) {
1600 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1605 sock_rps_save_rxhash(nsk
, skb
->rxhash
);
1606 if (tcp_child_process(sk
, nsk
, skb
)) {
1613 sock_rps_save_rxhash(sk
, skb
->rxhash
);
1615 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1622 tcp_v4_send_reset(rsk
, skb
);
1625 /* Be careful here. If this function gets more complicated and
1626 * gcc suffers from register pressure on the x86, sk (in %ebx)
1627 * might be destroyed here. This current version compiles correctly,
1628 * but you have been warned.
1633 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1636 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1642 int tcp_v4_rcv(struct sk_buff
*skb
)
1644 const struct iphdr
*iph
;
1648 struct net
*net
= dev_net(skb
->dev
);
1650 if (skb
->pkt_type
!= PACKET_HOST
)
1653 /* Count it even if it's bad */
1654 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1656 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1661 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1663 if (!pskb_may_pull(skb
, th
->doff
* 4))
1666 /* An explanation is required here, I think.
1667 * Packet length and doff are validated by header prediction,
1668 * provided case of th->doff==0 is eliminated.
1669 * So, we defer the checks. */
1670 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
1675 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1676 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1677 skb
->len
- th
->doff
* 4);
1678 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1679 TCP_SKB_CB(skb
)->when
= 0;
1680 TCP_SKB_CB(skb
)->flags
= iph
->tos
;
1681 TCP_SKB_CB(skb
)->sacked
= 0;
1683 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1688 if (sk
->sk_state
== TCP_TIME_WAIT
)
1691 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1692 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1693 goto discard_and_relse
;
1696 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1697 goto discard_and_relse
;
1700 if (sk_filter(sk
, skb
))
1701 goto discard_and_relse
;
1705 bh_lock_sock_nested(sk
);
1707 if (!sock_owned_by_user(sk
)) {
1708 #ifdef CONFIG_NET_DMA
1709 struct tcp_sock
*tp
= tcp_sk(sk
);
1710 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1711 tp
->ucopy
.dma_chan
= dma_find_channel(DMA_MEMCPY
);
1712 if (tp
->ucopy
.dma_chan
)
1713 ret
= tcp_v4_do_rcv(sk
, skb
);
1717 if (!tcp_prequeue(sk
, skb
))
1718 ret
= tcp_v4_do_rcv(sk
, skb
);
1720 } else if (unlikely(sk_add_backlog(sk
, skb
))) {
1722 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1723 goto discard_and_relse
;
1732 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1735 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1737 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1739 tcp_v4_send_reset(NULL
, skb
);
1743 /* Discard frame. */
1752 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1753 inet_twsk_put(inet_twsk(sk
));
1757 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1758 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1759 inet_twsk_put(inet_twsk(sk
));
1762 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1764 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1766 iph
->daddr
, th
->dest
,
1769 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1770 inet_twsk_put(inet_twsk(sk
));
1774 /* Fall through to ACK */
1777 tcp_v4_timewait_ack(sk
, skb
);
1781 case TCP_TW_SUCCESS
:;
1786 struct inet_peer
*tcp_v4_get_peer(struct sock
*sk
, bool *release_it
)
1788 struct rtable
*rt
= (struct rtable
*) __sk_dst_get(sk
);
1789 struct inet_sock
*inet
= inet_sk(sk
);
1790 struct inet_peer
*peer
;
1793 inet
->cork
.fl
.u
.ip4
.daddr
!= inet
->inet_daddr
) {
1794 peer
= inet_getpeer_v4(inet
->inet_daddr
, 1);
1798 rt_bind_peer(rt
, inet
->inet_daddr
, 1);
1800 *release_it
= false;
1805 EXPORT_SYMBOL(tcp_v4_get_peer
);
1807 void *tcp_v4_tw_get_peer(struct sock
*sk
)
1809 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
1811 return inet_getpeer_v4(tw
->tw_daddr
, 1);
1813 EXPORT_SYMBOL(tcp_v4_tw_get_peer
);
1815 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1816 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1817 .twsk_unique
= tcp_twsk_unique
,
1818 .twsk_destructor
= tcp_twsk_destructor
,
1819 .twsk_getpeer
= tcp_v4_tw_get_peer
,
1822 const struct inet_connection_sock_af_ops ipv4_specific
= {
1823 .queue_xmit
= ip_queue_xmit
,
1824 .send_check
= tcp_v4_send_check
,
1825 .rebuild_header
= inet_sk_rebuild_header
,
1826 .conn_request
= tcp_v4_conn_request
,
1827 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1828 .get_peer
= tcp_v4_get_peer
,
1829 .net_header_len
= sizeof(struct iphdr
),
1830 .setsockopt
= ip_setsockopt
,
1831 .getsockopt
= ip_getsockopt
,
1832 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1833 .sockaddr_len
= sizeof(struct sockaddr_in
),
1834 .bind_conflict
= inet_csk_bind_conflict
,
1835 #ifdef CONFIG_COMPAT
1836 .compat_setsockopt
= compat_ip_setsockopt
,
1837 .compat_getsockopt
= compat_ip_getsockopt
,
1840 EXPORT_SYMBOL(ipv4_specific
);
1842 #ifdef CONFIG_TCP_MD5SIG
1843 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1844 .md5_lookup
= tcp_v4_md5_lookup
,
1845 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1846 .md5_add
= tcp_v4_md5_add_func
,
1847 .md5_parse
= tcp_v4_parse_md5_keys
,
1851 /* NOTE: A lot of things set to zero explicitly by call to
1852 * sk_alloc() so need not be done here.
1854 static int tcp_v4_init_sock(struct sock
*sk
)
1856 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1857 struct tcp_sock
*tp
= tcp_sk(sk
);
1859 skb_queue_head_init(&tp
->out_of_order_queue
);
1860 tcp_init_xmit_timers(sk
);
1861 tcp_prequeue_init(tp
);
1863 icsk
->icsk_rto
= TCP_TIMEOUT_INIT
;
1864 tp
->mdev
= TCP_TIMEOUT_INIT
;
1866 /* So many TCP implementations out there (incorrectly) count the
1867 * initial SYN frame in their delayed-ACK and congestion control
1868 * algorithms that we must have the following bandaid to talk
1869 * efficiently to them. -DaveM
1871 tp
->snd_cwnd
= TCP_INIT_CWND
;
1873 /* See draft-stevens-tcpca-spec-01 for discussion of the
1874 * initialization of these values.
1876 tp
->snd_ssthresh
= TCP_INFINITE_SSTHRESH
;
1877 tp
->snd_cwnd_clamp
= ~0;
1878 tp
->mss_cache
= TCP_MSS_DEFAULT
;
1880 tp
->reordering
= sysctl_tcp_reordering
;
1881 icsk
->icsk_ca_ops
= &tcp_init_congestion_ops
;
1883 sk
->sk_state
= TCP_CLOSE
;
1885 sk
->sk_write_space
= sk_stream_write_space
;
1886 sock_set_flag(sk
, SOCK_USE_WRITE_QUEUE
);
1888 icsk
->icsk_af_ops
= &ipv4_specific
;
1889 icsk
->icsk_sync_mss
= tcp_sync_mss
;
1890 #ifdef CONFIG_TCP_MD5SIG
1891 tp
->af_specific
= &tcp_sock_ipv4_specific
;
1894 /* TCP Cookie Transactions */
1895 if (sysctl_tcp_cookie_size
> 0) {
1896 /* Default, cookies without s_data_payload. */
1898 kzalloc(sizeof(*tp
->cookie_values
),
1900 if (tp
->cookie_values
!= NULL
)
1901 kref_init(&tp
->cookie_values
->kref
);
1903 /* Presumed zeroed, in order of appearance:
1904 * cookie_in_always, cookie_out_never,
1905 * s_data_constant, s_data_in, s_data_out
1907 sk
->sk_sndbuf
= sysctl_tcp_wmem
[1];
1908 sk
->sk_rcvbuf
= sysctl_tcp_rmem
[1];
1911 percpu_counter_inc(&tcp_sockets_allocated
);
1917 void tcp_v4_destroy_sock(struct sock
*sk
)
1919 struct tcp_sock
*tp
= tcp_sk(sk
);
1921 tcp_clear_xmit_timers(sk
);
1923 tcp_cleanup_congestion_control(sk
);
1925 /* Cleanup up the write buffer. */
1926 tcp_write_queue_purge(sk
);
1928 /* Cleans up our, hopefully empty, out_of_order_queue. */
1929 __skb_queue_purge(&tp
->out_of_order_queue
);
1931 #ifdef CONFIG_TCP_MD5SIG
1932 /* Clean up the MD5 key list, if any */
1933 if (tp
->md5sig_info
) {
1934 tcp_v4_clear_md5_list(sk
);
1935 kfree(tp
->md5sig_info
);
1936 tp
->md5sig_info
= NULL
;
1940 #ifdef CONFIG_NET_DMA
1941 /* Cleans up our sk_async_wait_queue */
1942 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1945 /* Clean prequeue, it must be empty really */
1946 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1948 /* Clean up a referenced TCP bind bucket. */
1949 if (inet_csk(sk
)->icsk_bind_hash
)
1953 * If sendmsg cached page exists, toss it.
1955 if (sk
->sk_sndmsg_page
) {
1956 __free_page(sk
->sk_sndmsg_page
);
1957 sk
->sk_sndmsg_page
= NULL
;
1960 /* TCP Cookie Transactions */
1961 if (tp
->cookie_values
!= NULL
) {
1962 kref_put(&tp
->cookie_values
->kref
,
1963 tcp_cookie_values_release
);
1964 tp
->cookie_values
= NULL
;
1967 percpu_counter_dec(&tcp_sockets_allocated
);
1969 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1971 #ifdef CONFIG_PROC_FS
1972 /* Proc filesystem TCP sock list dumping. */
1974 static inline struct inet_timewait_sock
*tw_head(struct hlist_nulls_head
*head
)
1976 return hlist_nulls_empty(head
) ? NULL
:
1977 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
1980 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
1982 return !is_a_nulls(tw
->tw_node
.next
) ?
1983 hlist_nulls_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
1987 * Get next listener socket follow cur. If cur is NULL, get first socket
1988 * starting from bucket given in st->bucket; when st->bucket is zero the
1989 * very first socket in the hash table is returned.
1991 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
1993 struct inet_connection_sock
*icsk
;
1994 struct hlist_nulls_node
*node
;
1995 struct sock
*sk
= cur
;
1996 struct inet_listen_hashbucket
*ilb
;
1997 struct tcp_iter_state
*st
= seq
->private;
1998 struct net
*net
= seq_file_net(seq
);
2001 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2002 spin_lock_bh(&ilb
->lock
);
2003 sk
= sk_nulls_head(&ilb
->head
);
2007 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2011 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
2012 struct request_sock
*req
= cur
;
2014 icsk
= inet_csk(st
->syn_wait_sk
);
2018 if (req
->rsk_ops
->family
== st
->family
) {
2024 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
2027 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
2029 sk
= sk_nulls_next(st
->syn_wait_sk
);
2030 st
->state
= TCP_SEQ_STATE_LISTENING
;
2031 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2033 icsk
= inet_csk(sk
);
2034 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2035 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
2037 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2038 sk
= sk_nulls_next(sk
);
2041 sk_nulls_for_each_from(sk
, node
) {
2042 if (!net_eq(sock_net(sk
), net
))
2044 if (sk
->sk_family
== st
->family
) {
2048 icsk
= inet_csk(sk
);
2049 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2050 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2052 st
->uid
= sock_i_uid(sk
);
2053 st
->syn_wait_sk
= sk
;
2054 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2058 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2060 spin_unlock_bh(&ilb
->lock
);
2062 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2063 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2064 spin_lock_bh(&ilb
->lock
);
2065 sk
= sk_nulls_head(&ilb
->head
);
2073 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2075 struct tcp_iter_state
*st
= seq
->private;
2080 rc
= listening_get_next(seq
, NULL
);
2082 while (rc
&& *pos
) {
2083 rc
= listening_get_next(seq
, rc
);
2089 static inline int empty_bucket(struct tcp_iter_state
*st
)
2091 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
) &&
2092 hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2096 * Get first established socket starting from bucket given in st->bucket.
2097 * If st->bucket is zero, the very first socket in the hash is returned.
2099 static void *established_get_first(struct seq_file
*seq
)
2101 struct tcp_iter_state
*st
= seq
->private;
2102 struct net
*net
= seq_file_net(seq
);
2106 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2108 struct hlist_nulls_node
*node
;
2109 struct inet_timewait_sock
*tw
;
2110 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2112 /* Lockless fast path for the common case of empty buckets */
2113 if (empty_bucket(st
))
2117 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2118 if (sk
->sk_family
!= st
->family
||
2119 !net_eq(sock_net(sk
), net
)) {
2125 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2126 inet_twsk_for_each(tw
, node
,
2127 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
2128 if (tw
->tw_family
!= st
->family
||
2129 !net_eq(twsk_net(tw
), net
)) {
2135 spin_unlock_bh(lock
);
2136 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2142 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2144 struct sock
*sk
= cur
;
2145 struct inet_timewait_sock
*tw
;
2146 struct hlist_nulls_node
*node
;
2147 struct tcp_iter_state
*st
= seq
->private;
2148 struct net
*net
= seq_file_net(seq
);
2153 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2157 while (tw
&& (tw
->tw_family
!= st
->family
|| !net_eq(twsk_net(tw
), net
))) {
2164 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2165 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2167 /* Look for next non empty bucket */
2169 while (++st
->bucket
<= tcp_hashinfo
.ehash_mask
&&
2172 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2175 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2176 sk
= sk_nulls_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2178 sk
= sk_nulls_next(sk
);
2180 sk_nulls_for_each_from(sk
, node
) {
2181 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2185 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2186 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2194 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2196 struct tcp_iter_state
*st
= seq
->private;
2200 rc
= established_get_first(seq
);
2203 rc
= established_get_next(seq
, rc
);
2209 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2212 struct tcp_iter_state
*st
= seq
->private;
2214 st
->state
= TCP_SEQ_STATE_LISTENING
;
2215 rc
= listening_get_idx(seq
, &pos
);
2218 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2219 rc
= established_get_idx(seq
, pos
);
2225 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2227 struct tcp_iter_state
*st
= seq
->private;
2228 int offset
= st
->offset
;
2229 int orig_num
= st
->num
;
2232 switch (st
->state
) {
2233 case TCP_SEQ_STATE_OPENREQ
:
2234 case TCP_SEQ_STATE_LISTENING
:
2235 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2237 st
->state
= TCP_SEQ_STATE_LISTENING
;
2238 rc
= listening_get_next(seq
, NULL
);
2239 while (offset
-- && rc
)
2240 rc
= listening_get_next(seq
, rc
);
2245 case TCP_SEQ_STATE_ESTABLISHED
:
2246 case TCP_SEQ_STATE_TIME_WAIT
:
2247 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2248 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2250 rc
= established_get_first(seq
);
2251 while (offset
-- && rc
)
2252 rc
= established_get_next(seq
, rc
);
2260 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2262 struct tcp_iter_state
*st
= seq
->private;
2265 if (*pos
&& *pos
== st
->last_pos
) {
2266 rc
= tcp_seek_last_pos(seq
);
2271 st
->state
= TCP_SEQ_STATE_LISTENING
;
2275 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2278 st
->last_pos
= *pos
;
2282 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2284 struct tcp_iter_state
*st
= seq
->private;
2287 if (v
== SEQ_START_TOKEN
) {
2288 rc
= tcp_get_idx(seq
, 0);
2292 switch (st
->state
) {
2293 case TCP_SEQ_STATE_OPENREQ
:
2294 case TCP_SEQ_STATE_LISTENING
:
2295 rc
= listening_get_next(seq
, v
);
2297 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2300 rc
= established_get_first(seq
);
2303 case TCP_SEQ_STATE_ESTABLISHED
:
2304 case TCP_SEQ_STATE_TIME_WAIT
:
2305 rc
= established_get_next(seq
, v
);
2310 st
->last_pos
= *pos
;
2314 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2316 struct tcp_iter_state
*st
= seq
->private;
2318 switch (st
->state
) {
2319 case TCP_SEQ_STATE_OPENREQ
:
2321 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2322 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2324 case TCP_SEQ_STATE_LISTENING
:
2325 if (v
!= SEQ_START_TOKEN
)
2326 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2328 case TCP_SEQ_STATE_TIME_WAIT
:
2329 case TCP_SEQ_STATE_ESTABLISHED
:
2331 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2336 static int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2338 struct tcp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
2339 struct tcp_iter_state
*s
;
2342 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2343 sizeof(struct tcp_iter_state
));
2347 s
= ((struct seq_file
*)file
->private_data
)->private;
2348 s
->family
= afinfo
->family
;
2353 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2356 struct proc_dir_entry
*p
;
2358 afinfo
->seq_fops
.open
= tcp_seq_open
;
2359 afinfo
->seq_fops
.read
= seq_read
;
2360 afinfo
->seq_fops
.llseek
= seq_lseek
;
2361 afinfo
->seq_fops
.release
= seq_release_net
;
2363 afinfo
->seq_ops
.start
= tcp_seq_start
;
2364 afinfo
->seq_ops
.next
= tcp_seq_next
;
2365 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2367 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2368 &afinfo
->seq_fops
, afinfo
);
2373 EXPORT_SYMBOL(tcp_proc_register
);
2375 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2377 proc_net_remove(net
, afinfo
->name
);
2379 EXPORT_SYMBOL(tcp_proc_unregister
);
2381 static void get_openreq4(struct sock
*sk
, struct request_sock
*req
,
2382 struct seq_file
*f
, int i
, int uid
, int *len
)
2384 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2385 int ttd
= req
->expires
- jiffies
;
2387 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2388 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2391 ntohs(inet_sk(sk
)->inet_sport
),
2393 ntohs(ireq
->rmt_port
),
2395 0, 0, /* could print option size, but that is af dependent. */
2396 1, /* timers active (only the expire timer) */
2397 jiffies_to_clock_t(ttd
),
2400 0, /* non standard timer */
2401 0, /* open_requests have no inode */
2402 atomic_read(&sk
->sk_refcnt
),
2407 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
, int *len
)
2410 unsigned long timer_expires
;
2411 struct tcp_sock
*tp
= tcp_sk(sk
);
2412 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2413 struct inet_sock
*inet
= inet_sk(sk
);
2414 __be32 dest
= inet
->inet_daddr
;
2415 __be32 src
= inet
->inet_rcv_saddr
;
2416 __u16 destp
= ntohs(inet
->inet_dport
);
2417 __u16 srcp
= ntohs(inet
->inet_sport
);
2420 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
) {
2422 timer_expires
= icsk
->icsk_timeout
;
2423 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2425 timer_expires
= icsk
->icsk_timeout
;
2426 } else if (timer_pending(&sk
->sk_timer
)) {
2428 timer_expires
= sk
->sk_timer
.expires
;
2431 timer_expires
= jiffies
;
2434 if (sk
->sk_state
== TCP_LISTEN
)
2435 rx_queue
= sk
->sk_ack_backlog
;
2438 * because we dont lock socket, we might find a transient negative value
2440 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2442 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2443 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2444 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2445 tp
->write_seq
- tp
->snd_una
,
2448 jiffies_to_clock_t(timer_expires
- jiffies
),
2449 icsk
->icsk_retransmits
,
2451 icsk
->icsk_probes_out
,
2453 atomic_read(&sk
->sk_refcnt
), sk
,
2454 jiffies_to_clock_t(icsk
->icsk_rto
),
2455 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2456 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2458 tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
,
2462 static void get_timewait4_sock(struct inet_timewait_sock
*tw
,
2463 struct seq_file
*f
, int i
, int *len
)
2467 int ttd
= tw
->tw_ttd
- jiffies
;
2472 dest
= tw
->tw_daddr
;
2473 src
= tw
->tw_rcv_saddr
;
2474 destp
= ntohs(tw
->tw_dport
);
2475 srcp
= ntohs(tw
->tw_sport
);
2477 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2478 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2479 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2480 3, jiffies_to_clock_t(ttd
), 0, 0, 0, 0,
2481 atomic_read(&tw
->tw_refcnt
), tw
, len
);
2486 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2488 struct tcp_iter_state
*st
;
2491 if (v
== SEQ_START_TOKEN
) {
2492 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2493 " sl local_address rem_address st tx_queue "
2494 "rx_queue tr tm->when retrnsmt uid timeout "
2500 switch (st
->state
) {
2501 case TCP_SEQ_STATE_LISTENING
:
2502 case TCP_SEQ_STATE_ESTABLISHED
:
2503 get_tcp4_sock(v
, seq
, st
->num
, &len
);
2505 case TCP_SEQ_STATE_OPENREQ
:
2506 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
, &len
);
2508 case TCP_SEQ_STATE_TIME_WAIT
:
2509 get_timewait4_sock(v
, seq
, st
->num
, &len
);
2512 seq_printf(seq
, "%*s\n", TMPSZ
- 1 - len
, "");
2517 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2521 .owner
= THIS_MODULE
,
2524 .show
= tcp4_seq_show
,
2528 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2530 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2533 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2535 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2538 static struct pernet_operations tcp4_net_ops
= {
2539 .init
= tcp4_proc_init_net
,
2540 .exit
= tcp4_proc_exit_net
,
2543 int __init
tcp4_proc_init(void)
2545 return register_pernet_subsys(&tcp4_net_ops
);
2548 void tcp4_proc_exit(void)
2550 unregister_pernet_subsys(&tcp4_net_ops
);
2552 #endif /* CONFIG_PROC_FS */
2554 struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2556 const struct iphdr
*iph
= skb_gro_network_header(skb
);
2558 switch (skb
->ip_summed
) {
2559 case CHECKSUM_COMPLETE
:
2560 if (!tcp_v4_check(skb_gro_len(skb
), iph
->saddr
, iph
->daddr
,
2562 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2568 NAPI_GRO_CB(skb
)->flush
= 1;
2572 return tcp_gro_receive(head
, skb
);
2575 int tcp4_gro_complete(struct sk_buff
*skb
)
2577 const struct iphdr
*iph
= ip_hdr(skb
);
2578 struct tcphdr
*th
= tcp_hdr(skb
);
2580 th
->check
= ~tcp_v4_check(skb
->len
- skb_transport_offset(skb
),
2581 iph
->saddr
, iph
->daddr
, 0);
2582 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
2584 return tcp_gro_complete(skb
);
2587 struct proto tcp_prot
= {
2589 .owner
= THIS_MODULE
,
2591 .connect
= tcp_v4_connect
,
2592 .disconnect
= tcp_disconnect
,
2593 .accept
= inet_csk_accept
,
2595 .init
= tcp_v4_init_sock
,
2596 .destroy
= tcp_v4_destroy_sock
,
2597 .shutdown
= tcp_shutdown
,
2598 .setsockopt
= tcp_setsockopt
,
2599 .getsockopt
= tcp_getsockopt
,
2600 .recvmsg
= tcp_recvmsg
,
2601 .sendmsg
= tcp_sendmsg
,
2602 .sendpage
= tcp_sendpage
,
2603 .backlog_rcv
= tcp_v4_do_rcv
,
2605 .unhash
= inet_unhash
,
2606 .get_port
= inet_csk_get_port
,
2607 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2608 .sockets_allocated
= &tcp_sockets_allocated
,
2609 .orphan_count
= &tcp_orphan_count
,
2610 .memory_allocated
= &tcp_memory_allocated
,
2611 .memory_pressure
= &tcp_memory_pressure
,
2612 .sysctl_mem
= sysctl_tcp_mem
,
2613 .sysctl_wmem
= sysctl_tcp_wmem
,
2614 .sysctl_rmem
= sysctl_tcp_rmem
,
2615 .max_header
= MAX_TCP_HEADER
,
2616 .obj_size
= sizeof(struct tcp_sock
),
2617 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2618 .twsk_prot
= &tcp_timewait_sock_ops
,
2619 .rsk_prot
= &tcp_request_sock_ops
,
2620 .h
.hashinfo
= &tcp_hashinfo
,
2621 .no_autobind
= true,
2622 #ifdef CONFIG_COMPAT
2623 .compat_setsockopt
= compat_tcp_setsockopt
,
2624 .compat_getsockopt
= compat_tcp_getsockopt
,
2627 EXPORT_SYMBOL(tcp_prot
);
2630 static int __net_init
tcp_sk_init(struct net
*net
)
2632 return inet_ctl_sock_create(&net
->ipv4
.tcp_sock
,
2633 PF_INET
, SOCK_RAW
, IPPROTO_TCP
, net
);
2636 static void __net_exit
tcp_sk_exit(struct net
*net
)
2638 inet_ctl_sock_destroy(net
->ipv4
.tcp_sock
);
2641 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2643 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2646 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2647 .init
= tcp_sk_init
,
2648 .exit
= tcp_sk_exit
,
2649 .exit_batch
= tcp_sk_exit_batch
,
2652 void __init
tcp_v4_init(void)
2654 inet_hashinfo_init(&tcp_hashinfo
);
2655 if (register_pernet_subsys(&tcp_sk_ops
))
2656 panic("Failed to create the TCP control socket.\n");