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/types.h>
55 #include <linux/fcntl.h>
56 #include <linux/module.h>
57 #include <linux/random.h>
58 #include <linux/cache.h>
59 #include <linux/jhash.h>
60 #include <linux/init.h>
61 #include <linux/times.h>
63 #include <net/net_namespace.h>
65 #include <net/inet_hashtables.h>
67 #include <net/transp_v6.h>
69 #include <net/inet_common.h>
70 #include <net/timewait_sock.h>
72 #include <net/netdma.h>
74 #include <linux/inet.h>
75 #include <linux/ipv6.h>
76 #include <linux/stddef.h>
77 #include <linux/proc_fs.h>
78 #include <linux/seq_file.h>
80 #include <linux/crypto.h>
81 #include <linux/scatterlist.h>
83 int sysctl_tcp_tw_reuse __read_mostly
;
84 int sysctl_tcp_low_latency __read_mostly
;
87 #ifdef CONFIG_TCP_MD5SIG
88 static struct tcp_md5sig_key
*tcp_v4_md5_do_lookup(struct sock
*sk
,
90 static int tcp_v4_md5_hash_hdr(char *md5_hash
, struct tcp_md5sig_key
*key
,
91 __be32 daddr
, __be32 saddr
, struct tcphdr
*th
);
94 struct tcp_md5sig_key
*tcp_v4_md5_do_lookup(struct sock
*sk
, __be32 addr
)
100 struct inet_hashinfo __cacheline_aligned tcp_hashinfo
= {
101 .lhash_lock
= __RW_LOCK_UNLOCKED(tcp_hashinfo
.lhash_lock
),
102 .lhash_users
= ATOMIC_INIT(0),
103 .lhash_wait
= __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo
.lhash_wait
),
106 static inline __u32
tcp_v4_init_sequence(struct sk_buff
*skb
)
108 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
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
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)
136 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
137 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 inet_sock
*inet
= inet_sk(sk
);
151 struct tcp_sock
*tp
= tcp_sk(sk
);
152 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
154 __be32 daddr
, nexthop
;
158 if (addr_len
< sizeof(struct sockaddr_in
))
161 if (usin
->sin_family
!= AF_INET
)
162 return -EAFNOSUPPORT
;
164 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
165 if (inet
->opt
&& inet
->opt
->srr
) {
168 nexthop
= inet
->opt
->faddr
;
171 tmp
= ip_route_connect(&rt
, nexthop
, inet
->saddr
,
172 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
174 inet
->sport
, usin
->sin_port
, sk
, 1);
176 if (tmp
== -ENETUNREACH
)
177 IP_INC_STATS_BH(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
181 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
186 if (!inet
->opt
|| !inet
->opt
->srr
)
190 inet
->saddr
= rt
->rt_src
;
191 inet
->rcv_saddr
= inet
->saddr
;
193 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->daddr
!= daddr
) {
194 /* Reset inherited state */
195 tp
->rx_opt
.ts_recent
= 0;
196 tp
->rx_opt
.ts_recent_stamp
= 0;
200 if (tcp_death_row
.sysctl_tw_recycle
&&
201 !tp
->rx_opt
.ts_recent_stamp
&& rt
->rt_dst
== daddr
) {
202 struct inet_peer
*peer
= rt_get_peer(rt
);
204 * VJ's idea. We save last timestamp seen from
205 * the destination in peer table, when entering state
206 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
207 * when trying new connection.
210 peer
->tcp_ts_stamp
+ TCP_PAWS_MSL
>= get_seconds()) {
211 tp
->rx_opt
.ts_recent_stamp
= peer
->tcp_ts_stamp
;
212 tp
->rx_opt
.ts_recent
= peer
->tcp_ts
;
216 inet
->dport
= usin
->sin_port
;
219 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
221 inet_csk(sk
)->icsk_ext_hdr_len
= inet
->opt
->optlen
;
223 tp
->rx_opt
.mss_clamp
= 536;
225 /* Socket identity is still unknown (sport may be zero).
226 * However we set state to SYN-SENT and not releasing socket
227 * lock select source port, enter ourselves into the hash tables and
228 * complete initialization after this.
230 tcp_set_state(sk
, TCP_SYN_SENT
);
231 err
= inet_hash_connect(&tcp_death_row
, sk
);
235 err
= ip_route_newports(&rt
, IPPROTO_TCP
,
236 inet
->sport
, inet
->dport
, sk
);
240 /* OK, now commit destination to socket. */
241 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
242 sk_setup_caps(sk
, &rt
->u
.dst
);
245 tp
->write_seq
= secure_tcp_sequence_number(inet
->saddr
,
250 inet
->id
= tp
->write_seq
^ jiffies
;
252 err
= tcp_connect(sk
);
261 * This unhashes the socket and releases the local port,
264 tcp_set_state(sk
, TCP_CLOSE
);
266 sk
->sk_route_caps
= 0;
272 * This routine does path mtu discovery as defined in RFC1191.
274 static void do_pmtu_discovery(struct sock
*sk
, struct iphdr
*iph
, u32 mtu
)
276 struct dst_entry
*dst
;
277 struct inet_sock
*inet
= inet_sk(sk
);
279 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
280 * send out by Linux are always <576bytes so they should go through
283 if (sk
->sk_state
== TCP_LISTEN
)
286 /* We don't check in the destentry if pmtu discovery is forbidden
287 * on this route. We just assume that no packet_to_big packets
288 * are send back when pmtu discovery is not active.
289 * There is a small race when the user changes this flag in the
290 * route, but I think that's acceptable.
292 if ((dst
= __sk_dst_check(sk
, 0)) == NULL
)
295 dst
->ops
->update_pmtu(dst
, mtu
);
297 /* Something is about to be wrong... Remember soft error
298 * for the case, if this connection will not able to recover.
300 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
301 sk
->sk_err_soft
= EMSGSIZE
;
305 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
306 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
307 tcp_sync_mss(sk
, mtu
);
309 /* Resend the TCP packet because it's
310 * clear that the old packet has been
311 * dropped. This is the new "fast" path mtu
314 tcp_simple_retransmit(sk
);
315 } /* else let the usual retransmit timer handle it */
319 * This routine is called by the ICMP module when it gets some
320 * sort of error condition. If err < 0 then the socket should
321 * be closed and the error returned to the user. If err > 0
322 * it's just the icmp type << 8 | icmp code. After adjustment
323 * header points to the first 8 bytes of the tcp header. We need
324 * to find the appropriate port.
326 * The locking strategy used here is very "optimistic". When
327 * someone else accesses the socket the ICMP is just dropped
328 * and for some paths there is no check at all.
329 * A more general error queue to queue errors for later handling
330 * is probably better.
334 void tcp_v4_err(struct sk_buff
*skb
, u32 info
)
336 struct iphdr
*iph
= (struct iphdr
*)skb
->data
;
337 struct tcphdr
*th
= (struct tcphdr
*)(skb
->data
+ (iph
->ihl
<< 2));
339 struct inet_sock
*inet
;
340 const int type
= icmp_hdr(skb
)->type
;
341 const int code
= icmp_hdr(skb
)->code
;
345 struct net
*net
= dev_net(skb
->dev
);
347 if (skb
->len
< (iph
->ihl
<< 2) + 8) {
348 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
352 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
353 iph
->saddr
, th
->source
, inet_iif(skb
));
355 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
358 if (sk
->sk_state
== TCP_TIME_WAIT
) {
359 inet_twsk_put(inet_twsk(sk
));
364 /* If too many ICMPs get dropped on busy
365 * servers this needs to be solved differently.
367 if (sock_owned_by_user(sk
))
368 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
370 if (sk
->sk_state
== TCP_CLOSE
)
374 seq
= ntohl(th
->seq
);
375 if (sk
->sk_state
!= TCP_LISTEN
&&
376 !between(seq
, tp
->snd_una
, tp
->snd_nxt
)) {
377 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
382 case ICMP_SOURCE_QUENCH
:
383 /* Just silently ignore these. */
385 case ICMP_PARAMETERPROB
:
388 case ICMP_DEST_UNREACH
:
389 if (code
> NR_ICMP_UNREACH
)
392 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
393 if (!sock_owned_by_user(sk
))
394 do_pmtu_discovery(sk
, iph
, info
);
398 err
= icmp_err_convert
[code
].errno
;
400 case ICMP_TIME_EXCEEDED
:
407 switch (sk
->sk_state
) {
408 struct request_sock
*req
, **prev
;
410 if (sock_owned_by_user(sk
))
413 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
414 iph
->daddr
, iph
->saddr
);
418 /* ICMPs are not backlogged, hence we cannot get
419 an established socket here.
423 if (seq
!= tcp_rsk(req
)->snt_isn
) {
424 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
429 * Still in SYN_RECV, just remove it silently.
430 * There is no good way to pass the error to the newly
431 * created socket, and POSIX does not want network
432 * errors returned from accept().
434 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
438 case TCP_SYN_RECV
: /* Cannot happen.
439 It can f.e. if SYNs crossed.
441 if (!sock_owned_by_user(sk
)) {
444 sk
->sk_error_report(sk
);
448 sk
->sk_err_soft
= err
;
453 /* If we've already connected we will keep trying
454 * until we time out, or the user gives up.
456 * rfc1122 4.2.3.9 allows to consider as hard errors
457 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
458 * but it is obsoleted by pmtu discovery).
460 * Note, that in modern internet, where routing is unreliable
461 * and in each dark corner broken firewalls sit, sending random
462 * errors ordered by their masters even this two messages finally lose
463 * their original sense (even Linux sends invalid PORT_UNREACHs)
465 * Now we are in compliance with RFCs.
470 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
472 sk
->sk_error_report(sk
);
473 } else { /* Only an error on timeout */
474 sk
->sk_err_soft
= err
;
482 /* This routine computes an IPv4 TCP checksum. */
483 void tcp_v4_send_check(struct sock
*sk
, int len
, struct sk_buff
*skb
)
485 struct inet_sock
*inet
= inet_sk(sk
);
486 struct tcphdr
*th
= tcp_hdr(skb
);
488 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
489 th
->check
= ~tcp_v4_check(len
, inet
->saddr
,
491 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
492 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
494 th
->check
= tcp_v4_check(len
, inet
->saddr
, inet
->daddr
,
495 csum_partial((char *)th
,
501 int tcp_v4_gso_send_check(struct sk_buff
*skb
)
503 const struct iphdr
*iph
;
506 if (!pskb_may_pull(skb
, sizeof(*th
)))
513 th
->check
= ~tcp_v4_check(skb
->len
, iph
->saddr
, iph
->daddr
, 0);
514 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
515 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
516 skb
->ip_summed
= CHECKSUM_PARTIAL
;
521 * This routine will send an RST to the other tcp.
523 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
525 * Answer: if a packet caused RST, it is not for a socket
526 * existing in our system, if it is matched to a socket,
527 * it is just duplicate segment or bug in other side's TCP.
528 * So that we build reply only basing on parameters
529 * arrived with segment.
530 * Exception: precedence violation. We do not implement it in any case.
533 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
535 struct tcphdr
*th
= tcp_hdr(skb
);
538 #ifdef CONFIG_TCP_MD5SIG
539 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
542 struct ip_reply_arg arg
;
543 #ifdef CONFIG_TCP_MD5SIG
544 struct tcp_md5sig_key
*key
;
548 /* Never send a reset in response to a reset. */
552 if (skb
->rtable
->rt_type
!= RTN_LOCAL
)
555 /* Swap the send and the receive. */
556 memset(&rep
, 0, sizeof(rep
));
557 rep
.th
.dest
= th
->source
;
558 rep
.th
.source
= th
->dest
;
559 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
563 rep
.th
.seq
= th
->ack_seq
;
566 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
567 skb
->len
- (th
->doff
<< 2));
570 memset(&arg
, 0, sizeof(arg
));
571 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
572 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
574 #ifdef CONFIG_TCP_MD5SIG
575 key
= sk
? tcp_v4_md5_do_lookup(sk
, ip_hdr(skb
)->daddr
) : NULL
;
577 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
579 (TCPOPT_MD5SIG
<< 8) |
581 /* Update length and the length the header thinks exists */
582 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
583 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
585 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
586 key
, ip_hdr(skb
)->saddr
,
587 ip_hdr(skb
)->daddr
, &rep
.th
);
590 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
591 ip_hdr(skb
)->saddr
, /* XXX */
592 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
593 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
594 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
596 net
= dev_net(skb
->dst
->dev
);
597 ip_send_reply(net
->ipv4
.tcp_sock
, skb
,
598 &arg
, arg
.iov
[0].iov_len
);
600 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
601 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
604 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
605 outside socket context is ugly, certainly. What can I do?
608 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
609 u32 win
, u32 ts
, int oif
,
610 struct tcp_md5sig_key
*key
,
613 struct tcphdr
*th
= tcp_hdr(skb
);
616 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
617 #ifdef CONFIG_TCP_MD5SIG
618 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
622 struct ip_reply_arg arg
;
623 struct net
*net
= dev_net(skb
->dst
->dev
);
625 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
626 memset(&arg
, 0, sizeof(arg
));
628 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
629 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
631 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
632 (TCPOPT_TIMESTAMP
<< 8) |
634 rep
.opt
[1] = htonl(tcp_time_stamp
);
635 rep
.opt
[2] = htonl(ts
);
636 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
639 /* Swap the send and the receive. */
640 rep
.th
.dest
= th
->source
;
641 rep
.th
.source
= th
->dest
;
642 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
643 rep
.th
.seq
= htonl(seq
);
644 rep
.th
.ack_seq
= htonl(ack
);
646 rep
.th
.window
= htons(win
);
648 #ifdef CONFIG_TCP_MD5SIG
650 int offset
= (ts
) ? 3 : 0;
652 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
654 (TCPOPT_MD5SIG
<< 8) |
656 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
657 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
659 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
660 key
, ip_hdr(skb
)->saddr
,
661 ip_hdr(skb
)->daddr
, &rep
.th
);
664 arg
.flags
= reply_flags
;
665 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
666 ip_hdr(skb
)->saddr
, /* XXX */
667 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
668 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
670 arg
.bound_dev_if
= oif
;
672 ip_send_reply(net
->ipv4
.tcp_sock
, skb
,
673 &arg
, arg
.iov
[0].iov_len
);
675 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
678 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
680 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
681 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
683 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
684 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
687 tcp_twsk_md5_key(tcptw
),
688 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0
694 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
695 struct request_sock
*req
)
697 tcp_v4_send_ack(skb
, tcp_rsk(req
)->snt_isn
+ 1,
698 tcp_rsk(req
)->rcv_isn
+ 1, req
->rcv_wnd
,
701 tcp_v4_md5_do_lookup(sk
, ip_hdr(skb
)->daddr
),
702 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0);
706 * Send a SYN-ACK after having received a SYN.
707 * This still operates on a request_sock only, not on a big
710 static int __tcp_v4_send_synack(struct sock
*sk
, struct request_sock
*req
,
711 struct dst_entry
*dst
)
713 const struct inet_request_sock
*ireq
= inet_rsk(req
);
715 struct sk_buff
* skb
;
717 /* First, grab a route. */
718 if (!dst
&& (dst
= inet_csk_route_req(sk
, req
)) == NULL
)
721 skb
= tcp_make_synack(sk
, dst
, req
);
724 struct tcphdr
*th
= tcp_hdr(skb
);
726 th
->check
= tcp_v4_check(skb
->len
,
729 csum_partial((char *)th
, skb
->len
,
732 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
735 err
= net_xmit_eval(err
);
742 static int tcp_v4_send_synack(struct sock
*sk
, struct request_sock
*req
)
744 return __tcp_v4_send_synack(sk
, req
, NULL
);
748 * IPv4 request_sock destructor.
750 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
752 kfree(inet_rsk(req
)->opt
);
755 #ifdef CONFIG_SYN_COOKIES
756 static void syn_flood_warning(struct sk_buff
*skb
)
758 static unsigned long warntime
;
760 if (time_after(jiffies
, (warntime
+ HZ
* 60))) {
763 "possible SYN flooding on port %d. Sending cookies.\n",
764 ntohs(tcp_hdr(skb
)->dest
));
770 * Save and compile IPv4 options into the request_sock if needed.
772 static struct ip_options
*tcp_v4_save_options(struct sock
*sk
,
775 struct ip_options
*opt
= &(IPCB(skb
)->opt
);
776 struct ip_options
*dopt
= NULL
;
778 if (opt
&& opt
->optlen
) {
779 int opt_size
= optlength(opt
);
780 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
782 if (ip_options_echo(dopt
, skb
)) {
791 #ifdef CONFIG_TCP_MD5SIG
793 * RFC2385 MD5 checksumming requires a mapping of
794 * IP address->MD5 Key.
795 * We need to maintain these in the sk structure.
798 /* Find the Key structure for an address. */
799 static struct tcp_md5sig_key
*
800 tcp_v4_md5_do_lookup(struct sock
*sk
, __be32 addr
)
802 struct tcp_sock
*tp
= tcp_sk(sk
);
805 if (!tp
->md5sig_info
|| !tp
->md5sig_info
->entries4
)
807 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++) {
808 if (tp
->md5sig_info
->keys4
[i
].addr
== addr
)
809 return &tp
->md5sig_info
->keys4
[i
].base
;
814 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
815 struct sock
*addr_sk
)
817 return tcp_v4_md5_do_lookup(sk
, inet_sk(addr_sk
)->daddr
);
820 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
822 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
823 struct request_sock
*req
)
825 return tcp_v4_md5_do_lookup(sk
, inet_rsk(req
)->rmt_addr
);
828 /* This can be called on a newly created socket, from other files */
829 int tcp_v4_md5_do_add(struct sock
*sk
, __be32 addr
,
830 u8
*newkey
, u8 newkeylen
)
832 /* Add Key to the list */
833 struct tcp_md5sig_key
*key
;
834 struct tcp_sock
*tp
= tcp_sk(sk
);
835 struct tcp4_md5sig_key
*keys
;
837 key
= tcp_v4_md5_do_lookup(sk
, addr
);
839 /* Pre-existing entry - just update that one. */
842 key
->keylen
= newkeylen
;
844 struct tcp_md5sig_info
*md5sig
;
846 if (!tp
->md5sig_info
) {
847 tp
->md5sig_info
= kzalloc(sizeof(*tp
->md5sig_info
),
849 if (!tp
->md5sig_info
) {
853 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
855 if (tcp_alloc_md5sig_pool() == NULL
) {
859 md5sig
= tp
->md5sig_info
;
861 if (md5sig
->alloced4
== md5sig
->entries4
) {
862 keys
= kmalloc((sizeof(*keys
) *
863 (md5sig
->entries4
+ 1)), GFP_ATOMIC
);
866 tcp_free_md5sig_pool();
870 if (md5sig
->entries4
)
871 memcpy(keys
, md5sig
->keys4
,
872 sizeof(*keys
) * md5sig
->entries4
);
874 /* Free old key list, and reference new one */
875 kfree(md5sig
->keys4
);
876 md5sig
->keys4
= keys
;
880 md5sig
->keys4
[md5sig
->entries4
- 1].addr
= addr
;
881 md5sig
->keys4
[md5sig
->entries4
- 1].base
.key
= newkey
;
882 md5sig
->keys4
[md5sig
->entries4
- 1].base
.keylen
= newkeylen
;
887 EXPORT_SYMBOL(tcp_v4_md5_do_add
);
889 static int tcp_v4_md5_add_func(struct sock
*sk
, struct sock
*addr_sk
,
890 u8
*newkey
, u8 newkeylen
)
892 return tcp_v4_md5_do_add(sk
, inet_sk(addr_sk
)->daddr
,
896 int tcp_v4_md5_do_del(struct sock
*sk
, __be32 addr
)
898 struct tcp_sock
*tp
= tcp_sk(sk
);
901 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++) {
902 if (tp
->md5sig_info
->keys4
[i
].addr
== addr
) {
904 kfree(tp
->md5sig_info
->keys4
[i
].base
.key
);
905 tp
->md5sig_info
->entries4
--;
907 if (tp
->md5sig_info
->entries4
== 0) {
908 kfree(tp
->md5sig_info
->keys4
);
909 tp
->md5sig_info
->keys4
= NULL
;
910 tp
->md5sig_info
->alloced4
= 0;
911 } else if (tp
->md5sig_info
->entries4
!= i
) {
912 /* Need to do some manipulation */
913 memmove(&tp
->md5sig_info
->keys4
[i
],
914 &tp
->md5sig_info
->keys4
[i
+1],
915 (tp
->md5sig_info
->entries4
- i
) *
916 sizeof(struct tcp4_md5sig_key
));
918 tcp_free_md5sig_pool();
925 EXPORT_SYMBOL(tcp_v4_md5_do_del
);
927 static void tcp_v4_clear_md5_list(struct sock
*sk
)
929 struct tcp_sock
*tp
= tcp_sk(sk
);
931 /* Free each key, then the set of key keys,
932 * the crypto element, and then decrement our
933 * hold on the last resort crypto.
935 if (tp
->md5sig_info
->entries4
) {
937 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++)
938 kfree(tp
->md5sig_info
->keys4
[i
].base
.key
);
939 tp
->md5sig_info
->entries4
= 0;
940 tcp_free_md5sig_pool();
942 if (tp
->md5sig_info
->keys4
) {
943 kfree(tp
->md5sig_info
->keys4
);
944 tp
->md5sig_info
->keys4
= NULL
;
945 tp
->md5sig_info
->alloced4
= 0;
949 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
952 struct tcp_md5sig cmd
;
953 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
956 if (optlen
< sizeof(cmd
))
959 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
962 if (sin
->sin_family
!= AF_INET
)
965 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
) {
966 if (!tcp_sk(sk
)->md5sig_info
)
968 return tcp_v4_md5_do_del(sk
, sin
->sin_addr
.s_addr
);
971 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
974 if (!tcp_sk(sk
)->md5sig_info
) {
975 struct tcp_sock
*tp
= tcp_sk(sk
);
976 struct tcp_md5sig_info
*p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
982 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
985 newkey
= kmemdup(cmd
.tcpm_key
, cmd
.tcpm_keylen
, GFP_KERNEL
);
988 return tcp_v4_md5_do_add(sk
, sin
->sin_addr
.s_addr
,
989 newkey
, cmd
.tcpm_keylen
);
992 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
993 __be32 daddr
, __be32 saddr
, int nbytes
)
995 struct tcp4_pseudohdr
*bp
;
996 struct scatterlist sg
;
998 bp
= &hp
->md5_blk
.ip4
;
1001 * 1. the TCP pseudo-header (in the order: source IP address,
1002 * destination IP address, zero-padded protocol number, and
1008 bp
->protocol
= IPPROTO_TCP
;
1009 bp
->len
= cpu_to_be16(nbytes
);
1011 sg_init_one(&sg
, bp
, sizeof(*bp
));
1012 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1015 static int tcp_v4_md5_hash_hdr(char *md5_hash
, struct tcp_md5sig_key
*key
,
1016 __be32 daddr
, __be32 saddr
, struct tcphdr
*th
)
1018 struct tcp_md5sig_pool
*hp
;
1019 struct hash_desc
*desc
;
1021 hp
= tcp_get_md5sig_pool();
1023 goto clear_hash_noput
;
1024 desc
= &hp
->md5_desc
;
1026 if (crypto_hash_init(desc
))
1028 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1030 if (tcp_md5_hash_header(hp
, th
))
1032 if (tcp_md5_hash_key(hp
, key
))
1034 if (crypto_hash_final(desc
, md5_hash
))
1037 tcp_put_md5sig_pool();
1041 tcp_put_md5sig_pool();
1043 memset(md5_hash
, 0, 16);
1047 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1048 struct sock
*sk
, struct request_sock
*req
,
1049 struct sk_buff
*skb
)
1051 struct tcp_md5sig_pool
*hp
;
1052 struct hash_desc
*desc
;
1053 struct tcphdr
*th
= tcp_hdr(skb
);
1054 __be32 saddr
, daddr
;
1057 saddr
= inet_sk(sk
)->saddr
;
1058 daddr
= inet_sk(sk
)->daddr
;
1060 saddr
= inet_rsk(req
)->loc_addr
;
1061 daddr
= inet_rsk(req
)->rmt_addr
;
1063 const struct iphdr
*iph
= ip_hdr(skb
);
1068 hp
= tcp_get_md5sig_pool();
1070 goto clear_hash_noput
;
1071 desc
= &hp
->md5_desc
;
1073 if (crypto_hash_init(desc
))
1076 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1078 if (tcp_md5_hash_header(hp
, th
))
1080 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1082 if (tcp_md5_hash_key(hp
, key
))
1084 if (crypto_hash_final(desc
, md5_hash
))
1087 tcp_put_md5sig_pool();
1091 tcp_put_md5sig_pool();
1093 memset(md5_hash
, 0, 16);
1097 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1099 static int tcp_v4_inbound_md5_hash(struct sock
*sk
, struct sk_buff
*skb
)
1102 * This gets called for each TCP segment that arrives
1103 * so we want to be efficient.
1104 * We have 3 drop cases:
1105 * o No MD5 hash and one expected.
1106 * o MD5 hash and we're not expecting one.
1107 * o MD5 hash and its wrong.
1109 __u8
*hash_location
= NULL
;
1110 struct tcp_md5sig_key
*hash_expected
;
1111 const struct iphdr
*iph
= ip_hdr(skb
);
1112 struct tcphdr
*th
= tcp_hdr(skb
);
1114 unsigned char newhash
[16];
1116 hash_expected
= tcp_v4_md5_do_lookup(sk
, iph
->saddr
);
1117 hash_location
= tcp_parse_md5sig_option(th
);
1119 /* We've parsed the options - do we have a hash? */
1120 if (!hash_expected
&& !hash_location
)
1123 if (hash_expected
&& !hash_location
) {
1124 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1128 if (!hash_expected
&& hash_location
) {
1129 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1133 /* Okay, so this is hash_expected and hash_location -
1134 * so we need to calculate the checksum.
1136 genhash
= tcp_v4_md5_hash_skb(newhash
,
1140 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1141 if (net_ratelimit()) {
1142 printk(KERN_INFO
"MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1143 &iph
->saddr
, ntohs(th
->source
),
1144 &iph
->daddr
, ntohs(th
->dest
),
1145 genhash
? " tcp_v4_calc_md5_hash failed" : "");
1154 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1156 .obj_size
= sizeof(struct tcp_request_sock
),
1157 .rtx_syn_ack
= tcp_v4_send_synack
,
1158 .send_ack
= tcp_v4_reqsk_send_ack
,
1159 .destructor
= tcp_v4_reqsk_destructor
,
1160 .send_reset
= tcp_v4_send_reset
,
1163 #ifdef CONFIG_TCP_MD5SIG
1164 static struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1165 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1169 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1170 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1171 .twsk_unique
= tcp_twsk_unique
,
1172 .twsk_destructor
= tcp_twsk_destructor
,
1175 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1177 struct inet_request_sock
*ireq
;
1178 struct tcp_options_received tmp_opt
;
1179 struct request_sock
*req
;
1180 __be32 saddr
= ip_hdr(skb
)->saddr
;
1181 __be32 daddr
= ip_hdr(skb
)->daddr
;
1182 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1183 struct dst_entry
*dst
= NULL
;
1184 #ifdef CONFIG_SYN_COOKIES
1185 int want_cookie
= 0;
1187 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1190 /* Never answer to SYNs send to broadcast or multicast */
1191 if (skb
->rtable
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1194 /* TW buckets are converted to open requests without
1195 * limitations, they conserve resources and peer is
1196 * evidently real one.
1198 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
1199 #ifdef CONFIG_SYN_COOKIES
1200 if (sysctl_tcp_syncookies
) {
1207 /* Accept backlog is full. If we have already queued enough
1208 * of warm entries in syn queue, drop request. It is better than
1209 * clogging syn queue with openreqs with exponentially increasing
1212 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1)
1215 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1219 #ifdef CONFIG_TCP_MD5SIG
1220 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1223 tcp_clear_options(&tmp_opt
);
1224 tmp_opt
.mss_clamp
= 536;
1225 tmp_opt
.user_mss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1227 tcp_parse_options(skb
, &tmp_opt
, 0);
1229 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1230 tcp_clear_options(&tmp_opt
);
1232 if (tmp_opt
.saw_tstamp
&& !tmp_opt
.rcv_tsval
) {
1233 /* Some OSes (unknown ones, but I see them on web server, which
1234 * contains information interesting only for windows'
1235 * users) do not send their stamp in SYN. It is easy case.
1236 * We simply do not advertise TS support.
1238 tmp_opt
.saw_tstamp
= 0;
1239 tmp_opt
.tstamp_ok
= 0;
1241 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1243 tcp_openreq_init(req
, &tmp_opt
, skb
);
1245 if (security_inet_conn_request(sk
, skb
, req
))
1248 ireq
= inet_rsk(req
);
1249 ireq
->loc_addr
= daddr
;
1250 ireq
->rmt_addr
= saddr
;
1251 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1252 ireq
->opt
= tcp_v4_save_options(sk
, skb
);
1254 TCP_ECN_create_request(req
, tcp_hdr(skb
));
1257 #ifdef CONFIG_SYN_COOKIES
1258 syn_flood_warning(skb
);
1259 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1261 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1263 struct inet_peer
*peer
= NULL
;
1265 /* VJ's idea. We save last timestamp seen
1266 * from the destination in peer table, when entering
1267 * state TIME-WAIT, and check against it before
1268 * accepting new connection request.
1270 * If "isn" is not zero, this request hit alive
1271 * timewait bucket, so that all the necessary checks
1272 * are made in the function processing timewait state.
1274 if (tmp_opt
.saw_tstamp
&&
1275 tcp_death_row
.sysctl_tw_recycle
&&
1276 (dst
= inet_csk_route_req(sk
, req
)) != NULL
&&
1277 (peer
= rt_get_peer((struct rtable
*)dst
)) != NULL
&&
1278 peer
->v4daddr
== saddr
) {
1279 if (get_seconds() < peer
->tcp_ts_stamp
+ TCP_PAWS_MSL
&&
1280 (s32
)(peer
->tcp_ts
- req
->ts_recent
) >
1282 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1283 goto drop_and_release
;
1286 /* Kill the following clause, if you dislike this way. */
1287 else if (!sysctl_tcp_syncookies
&&
1288 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1289 (sysctl_max_syn_backlog
>> 2)) &&
1290 (!peer
|| !peer
->tcp_ts_stamp
) &&
1291 (!dst
|| !dst_metric(dst
, RTAX_RTT
))) {
1292 /* Without syncookies last quarter of
1293 * backlog is filled with destinations,
1294 * proven to be alive.
1295 * It means that we continue to communicate
1296 * to destinations, already remembered
1297 * to the moment of synflood.
1299 LIMIT_NETDEBUG(KERN_DEBUG
"TCP: drop open request from %pI4/%u\n",
1300 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1301 goto drop_and_release
;
1304 isn
= tcp_v4_init_sequence(skb
);
1306 tcp_rsk(req
)->snt_isn
= isn
;
1308 if (__tcp_v4_send_synack(sk
, req
, dst
) || want_cookie
)
1311 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1324 * The three way handshake has completed - we got a valid synack -
1325 * now create the new socket.
1327 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1328 struct request_sock
*req
,
1329 struct dst_entry
*dst
)
1331 struct inet_request_sock
*ireq
;
1332 struct inet_sock
*newinet
;
1333 struct tcp_sock
*newtp
;
1335 #ifdef CONFIG_TCP_MD5SIG
1336 struct tcp_md5sig_key
*key
;
1339 if (sk_acceptq_is_full(sk
))
1342 if (!dst
&& (dst
= inet_csk_route_req(sk
, req
)) == NULL
)
1345 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1349 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1350 sk_setup_caps(newsk
, dst
);
1352 newtp
= tcp_sk(newsk
);
1353 newinet
= inet_sk(newsk
);
1354 ireq
= inet_rsk(req
);
1355 newinet
->daddr
= ireq
->rmt_addr
;
1356 newinet
->rcv_saddr
= ireq
->loc_addr
;
1357 newinet
->saddr
= ireq
->loc_addr
;
1358 newinet
->opt
= ireq
->opt
;
1360 newinet
->mc_index
= inet_iif(skb
);
1361 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1362 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1364 inet_csk(newsk
)->icsk_ext_hdr_len
= newinet
->opt
->optlen
;
1365 newinet
->id
= newtp
->write_seq
^ jiffies
;
1367 tcp_mtup_init(newsk
);
1368 tcp_sync_mss(newsk
, dst_mtu(dst
));
1369 newtp
->advmss
= dst_metric(dst
, RTAX_ADVMSS
);
1370 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1371 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1372 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1374 tcp_initialize_rcv_mss(newsk
);
1376 #ifdef CONFIG_TCP_MD5SIG
1377 /* Copy over the MD5 key from the original socket */
1378 if ((key
= tcp_v4_md5_do_lookup(sk
, newinet
->daddr
)) != NULL
) {
1380 * We're using one, so create a matching key
1381 * on the newsk structure. If we fail to get
1382 * memory, then we end up not copying the key
1385 char *newkey
= kmemdup(key
->key
, key
->keylen
, GFP_ATOMIC
);
1387 tcp_v4_md5_do_add(newsk
, inet_sk(sk
)->daddr
,
1388 newkey
, key
->keylen
);
1389 newsk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1393 __inet_hash_nolisten(newsk
);
1394 __inet_inherit_port(sk
, newsk
);
1399 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1401 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1406 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1408 struct tcphdr
*th
= tcp_hdr(skb
);
1409 const struct iphdr
*iph
= ip_hdr(skb
);
1411 struct request_sock
**prev
;
1412 /* Find possible connection requests. */
1413 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1414 iph
->saddr
, iph
->daddr
);
1416 return tcp_check_req(sk
, skb
, req
, prev
);
1418 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1419 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1422 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1426 inet_twsk_put(inet_twsk(nsk
));
1430 #ifdef CONFIG_SYN_COOKIES
1431 if (!th
->rst
&& !th
->syn
&& th
->ack
)
1432 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1437 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1439 const struct iphdr
*iph
= ip_hdr(skb
);
1441 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1442 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1443 iph
->daddr
, skb
->csum
)) {
1444 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1449 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1450 skb
->len
, IPPROTO_TCP
, 0);
1452 if (skb
->len
<= 76) {
1453 return __skb_checksum_complete(skb
);
1459 /* The socket must have it's spinlock held when we get
1462 * We have a potential double-lock case here, so even when
1463 * doing backlog processing we use the BH locking scheme.
1464 * This is because we cannot sleep with the original spinlock
1467 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1470 #ifdef CONFIG_TCP_MD5SIG
1472 * We really want to reject the packet as early as possible
1474 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1475 * o There is an MD5 option and we're not expecting one
1477 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1481 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1482 TCP_CHECK_TIMER(sk
);
1483 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1487 TCP_CHECK_TIMER(sk
);
1491 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1494 if (sk
->sk_state
== TCP_LISTEN
) {
1495 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1500 if (tcp_child_process(sk
, nsk
, skb
)) {
1508 TCP_CHECK_TIMER(sk
);
1509 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1513 TCP_CHECK_TIMER(sk
);
1517 tcp_v4_send_reset(rsk
, skb
);
1520 /* Be careful here. If this function gets more complicated and
1521 * gcc suffers from register pressure on the x86, sk (in %ebx)
1522 * might be destroyed here. This current version compiles correctly,
1523 * but you have been warned.
1528 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1536 int tcp_v4_rcv(struct sk_buff
*skb
)
1538 const struct iphdr
*iph
;
1542 struct net
*net
= dev_net(skb
->dev
);
1544 if (skb
->pkt_type
!= PACKET_HOST
)
1547 /* Count it even if it's bad */
1548 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1550 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1555 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1557 if (!pskb_may_pull(skb
, th
->doff
* 4))
1560 /* An explanation is required here, I think.
1561 * Packet length and doff are validated by header prediction,
1562 * provided case of th->doff==0 is eliminated.
1563 * So, we defer the checks. */
1564 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
1569 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1570 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1571 skb
->len
- th
->doff
* 4);
1572 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1573 TCP_SKB_CB(skb
)->when
= 0;
1574 TCP_SKB_CB(skb
)->flags
= iph
->tos
;
1575 TCP_SKB_CB(skb
)->sacked
= 0;
1577 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1582 if (sk
->sk_state
== TCP_TIME_WAIT
)
1585 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1586 goto discard_and_relse
;
1589 if (sk_filter(sk
, skb
))
1590 goto discard_and_relse
;
1594 bh_lock_sock_nested(sk
);
1596 if (!sock_owned_by_user(sk
)) {
1597 #ifdef CONFIG_NET_DMA
1598 struct tcp_sock
*tp
= tcp_sk(sk
);
1599 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1600 tp
->ucopy
.dma_chan
= get_softnet_dma();
1601 if (tp
->ucopy
.dma_chan
)
1602 ret
= tcp_v4_do_rcv(sk
, skb
);
1606 if (!tcp_prequeue(sk
, skb
))
1607 ret
= tcp_v4_do_rcv(sk
, skb
);
1610 sk_add_backlog(sk
, skb
);
1618 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1621 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1623 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1625 tcp_v4_send_reset(NULL
, skb
);
1629 /* Discard frame. */
1638 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1639 inet_twsk_put(inet_twsk(sk
));
1643 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1644 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1645 inet_twsk_put(inet_twsk(sk
));
1648 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1650 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1652 iph
->daddr
, th
->dest
,
1655 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1656 inet_twsk_put(inet_twsk(sk
));
1660 /* Fall through to ACK */
1663 tcp_v4_timewait_ack(sk
, skb
);
1667 case TCP_TW_SUCCESS
:;
1672 /* VJ's idea. Save last timestamp seen from this destination
1673 * and hold it at least for normal timewait interval to use for duplicate
1674 * segment detection in subsequent connections, before they enter synchronized
1678 int tcp_v4_remember_stamp(struct sock
*sk
)
1680 struct inet_sock
*inet
= inet_sk(sk
);
1681 struct tcp_sock
*tp
= tcp_sk(sk
);
1682 struct rtable
*rt
= (struct rtable
*)__sk_dst_get(sk
);
1683 struct inet_peer
*peer
= NULL
;
1686 if (!rt
|| rt
->rt_dst
!= inet
->daddr
) {
1687 peer
= inet_getpeer(inet
->daddr
, 1);
1691 rt_bind_peer(rt
, 1);
1696 if ((s32
)(peer
->tcp_ts
- tp
->rx_opt
.ts_recent
) <= 0 ||
1697 (peer
->tcp_ts_stamp
+ TCP_PAWS_MSL
< get_seconds() &&
1698 peer
->tcp_ts_stamp
<= tp
->rx_opt
.ts_recent_stamp
)) {
1699 peer
->tcp_ts_stamp
= tp
->rx_opt
.ts_recent_stamp
;
1700 peer
->tcp_ts
= tp
->rx_opt
.ts_recent
;
1710 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
)
1712 struct inet_peer
*peer
= inet_getpeer(tw
->tw_daddr
, 1);
1715 const struct tcp_timewait_sock
*tcptw
= tcp_twsk((struct sock
*)tw
);
1717 if ((s32
)(peer
->tcp_ts
- tcptw
->tw_ts_recent
) <= 0 ||
1718 (peer
->tcp_ts_stamp
+ TCP_PAWS_MSL
< get_seconds() &&
1719 peer
->tcp_ts_stamp
<= tcptw
->tw_ts_recent_stamp
)) {
1720 peer
->tcp_ts_stamp
= tcptw
->tw_ts_recent_stamp
;
1721 peer
->tcp_ts
= tcptw
->tw_ts_recent
;
1730 struct inet_connection_sock_af_ops ipv4_specific
= {
1731 .queue_xmit
= ip_queue_xmit
,
1732 .send_check
= tcp_v4_send_check
,
1733 .rebuild_header
= inet_sk_rebuild_header
,
1734 .conn_request
= tcp_v4_conn_request
,
1735 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1736 .remember_stamp
= tcp_v4_remember_stamp
,
1737 .net_header_len
= sizeof(struct iphdr
),
1738 .setsockopt
= ip_setsockopt
,
1739 .getsockopt
= ip_getsockopt
,
1740 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1741 .sockaddr_len
= sizeof(struct sockaddr_in
),
1742 .bind_conflict
= inet_csk_bind_conflict
,
1743 #ifdef CONFIG_COMPAT
1744 .compat_setsockopt
= compat_ip_setsockopt
,
1745 .compat_getsockopt
= compat_ip_getsockopt
,
1749 #ifdef CONFIG_TCP_MD5SIG
1750 static struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1751 .md5_lookup
= tcp_v4_md5_lookup
,
1752 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1753 .md5_add
= tcp_v4_md5_add_func
,
1754 .md5_parse
= tcp_v4_parse_md5_keys
,
1758 /* NOTE: A lot of things set to zero explicitly by call to
1759 * sk_alloc() so need not be done here.
1761 static int tcp_v4_init_sock(struct sock
*sk
)
1763 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1764 struct tcp_sock
*tp
= tcp_sk(sk
);
1766 skb_queue_head_init(&tp
->out_of_order_queue
);
1767 tcp_init_xmit_timers(sk
);
1768 tcp_prequeue_init(tp
);
1770 icsk
->icsk_rto
= TCP_TIMEOUT_INIT
;
1771 tp
->mdev
= TCP_TIMEOUT_INIT
;
1773 /* So many TCP implementations out there (incorrectly) count the
1774 * initial SYN frame in their delayed-ACK and congestion control
1775 * algorithms that we must have the following bandaid to talk
1776 * efficiently to them. -DaveM
1780 /* See draft-stevens-tcpca-spec-01 for discussion of the
1781 * initialization of these values.
1783 tp
->snd_ssthresh
= 0x7fffffff; /* Infinity */
1784 tp
->snd_cwnd_clamp
= ~0;
1785 tp
->mss_cache
= 536;
1787 tp
->reordering
= sysctl_tcp_reordering
;
1788 icsk
->icsk_ca_ops
= &tcp_init_congestion_ops
;
1790 sk
->sk_state
= TCP_CLOSE
;
1792 sk
->sk_write_space
= sk_stream_write_space
;
1793 sock_set_flag(sk
, SOCK_USE_WRITE_QUEUE
);
1795 icsk
->icsk_af_ops
= &ipv4_specific
;
1796 icsk
->icsk_sync_mss
= tcp_sync_mss
;
1797 #ifdef CONFIG_TCP_MD5SIG
1798 tp
->af_specific
= &tcp_sock_ipv4_specific
;
1801 sk
->sk_sndbuf
= sysctl_tcp_wmem
[1];
1802 sk
->sk_rcvbuf
= sysctl_tcp_rmem
[1];
1804 atomic_inc(&tcp_sockets_allocated
);
1809 void tcp_v4_destroy_sock(struct sock
*sk
)
1811 struct tcp_sock
*tp
= tcp_sk(sk
);
1813 tcp_clear_xmit_timers(sk
);
1815 tcp_cleanup_congestion_control(sk
);
1817 /* Cleanup up the write buffer. */
1818 tcp_write_queue_purge(sk
);
1820 /* Cleans up our, hopefully empty, out_of_order_queue. */
1821 __skb_queue_purge(&tp
->out_of_order_queue
);
1823 #ifdef CONFIG_TCP_MD5SIG
1824 /* Clean up the MD5 key list, if any */
1825 if (tp
->md5sig_info
) {
1826 tcp_v4_clear_md5_list(sk
);
1827 kfree(tp
->md5sig_info
);
1828 tp
->md5sig_info
= NULL
;
1832 #ifdef CONFIG_NET_DMA
1833 /* Cleans up our sk_async_wait_queue */
1834 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1837 /* Clean prequeue, it must be empty really */
1838 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1840 /* Clean up a referenced TCP bind bucket. */
1841 if (inet_csk(sk
)->icsk_bind_hash
)
1845 * If sendmsg cached page exists, toss it.
1847 if (sk
->sk_sndmsg_page
) {
1848 __free_page(sk
->sk_sndmsg_page
);
1849 sk
->sk_sndmsg_page
= NULL
;
1852 atomic_dec(&tcp_sockets_allocated
);
1855 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1857 #ifdef CONFIG_PROC_FS
1858 /* Proc filesystem TCP sock list dumping. */
1860 static inline struct inet_timewait_sock
*tw_head(struct hlist_head
*head
)
1862 return hlist_empty(head
) ? NULL
:
1863 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
1866 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
1868 return tw
->tw_node
.next
?
1869 hlist_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
1872 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
1874 struct inet_connection_sock
*icsk
;
1875 struct hlist_node
*node
;
1876 struct sock
*sk
= cur
;
1877 struct tcp_iter_state
* st
= seq
->private;
1878 struct net
*net
= seq_file_net(seq
);
1882 sk
= sk_head(&tcp_hashinfo
.listening_hash
[0]);
1888 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
1889 struct request_sock
*req
= cur
;
1891 icsk
= inet_csk(st
->syn_wait_sk
);
1895 if (req
->rsk_ops
->family
== st
->family
) {
1901 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
1904 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
1906 sk
= sk_next(st
->syn_wait_sk
);
1907 st
->state
= TCP_SEQ_STATE_LISTENING
;
1908 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1910 icsk
= inet_csk(sk
);
1911 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1912 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
1914 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1918 sk_for_each_from(sk
, node
) {
1919 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
)) {
1923 icsk
= inet_csk(sk
);
1924 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1925 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
1927 st
->uid
= sock_i_uid(sk
);
1928 st
->syn_wait_sk
= sk
;
1929 st
->state
= TCP_SEQ_STATE_OPENREQ
;
1933 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1935 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
1936 sk
= sk_head(&tcp_hashinfo
.listening_hash
[st
->bucket
]);
1944 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
1946 void *rc
= listening_get_next(seq
, NULL
);
1948 while (rc
&& *pos
) {
1949 rc
= listening_get_next(seq
, rc
);
1955 static inline int empty_bucket(struct tcp_iter_state
*st
)
1957 return hlist_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
) &&
1958 hlist_empty(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
1961 static void *established_get_first(struct seq_file
*seq
)
1963 struct tcp_iter_state
* st
= seq
->private;
1964 struct net
*net
= seq_file_net(seq
);
1967 for (st
->bucket
= 0; st
->bucket
< tcp_hashinfo
.ehash_size
; ++st
->bucket
) {
1969 struct hlist_node
*node
;
1970 struct inet_timewait_sock
*tw
;
1971 rwlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
1973 /* Lockless fast path for the common case of empty buckets */
1974 if (empty_bucket(st
))
1978 sk_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
1979 if (sk
->sk_family
!= st
->family
||
1980 !net_eq(sock_net(sk
), net
)) {
1986 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
1987 inet_twsk_for_each(tw
, node
,
1988 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
1989 if (tw
->tw_family
!= st
->family
||
1990 !net_eq(twsk_net(tw
), net
)) {
1996 read_unlock_bh(lock
);
1997 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2003 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2005 struct sock
*sk
= cur
;
2006 struct inet_timewait_sock
*tw
;
2007 struct hlist_node
*node
;
2008 struct tcp_iter_state
* st
= seq
->private;
2009 struct net
*net
= seq_file_net(seq
);
2013 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2017 while (tw
&& (tw
->tw_family
!= st
->family
|| !net_eq(twsk_net(tw
), net
))) {
2024 read_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2025 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2027 /* Look for next non empty bucket */
2028 while (++st
->bucket
< tcp_hashinfo
.ehash_size
&&
2031 if (st
->bucket
>= tcp_hashinfo
.ehash_size
)
2034 read_lock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2035 sk
= sk_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2039 sk_for_each_from(sk
, node
) {
2040 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2044 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2045 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2053 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2055 void *rc
= established_get_first(seq
);
2058 rc
= established_get_next(seq
, rc
);
2064 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2067 struct tcp_iter_state
* st
= seq
->private;
2069 inet_listen_lock(&tcp_hashinfo
);
2070 st
->state
= TCP_SEQ_STATE_LISTENING
;
2071 rc
= listening_get_idx(seq
, &pos
);
2074 inet_listen_unlock(&tcp_hashinfo
);
2075 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2076 rc
= established_get_idx(seq
, pos
);
2082 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2084 struct tcp_iter_state
* st
= seq
->private;
2085 st
->state
= TCP_SEQ_STATE_LISTENING
;
2087 return *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2090 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2093 struct tcp_iter_state
* st
;
2095 if (v
== SEQ_START_TOKEN
) {
2096 rc
= tcp_get_idx(seq
, 0);
2101 switch (st
->state
) {
2102 case TCP_SEQ_STATE_OPENREQ
:
2103 case TCP_SEQ_STATE_LISTENING
:
2104 rc
= listening_get_next(seq
, v
);
2106 inet_listen_unlock(&tcp_hashinfo
);
2107 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2108 rc
= established_get_first(seq
);
2111 case TCP_SEQ_STATE_ESTABLISHED
:
2112 case TCP_SEQ_STATE_TIME_WAIT
:
2113 rc
= established_get_next(seq
, v
);
2121 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2123 struct tcp_iter_state
* st
= seq
->private;
2125 switch (st
->state
) {
2126 case TCP_SEQ_STATE_OPENREQ
:
2128 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2129 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2131 case TCP_SEQ_STATE_LISTENING
:
2132 if (v
!= SEQ_START_TOKEN
)
2133 inet_listen_unlock(&tcp_hashinfo
);
2135 case TCP_SEQ_STATE_TIME_WAIT
:
2136 case TCP_SEQ_STATE_ESTABLISHED
:
2138 read_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2143 static int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2145 struct tcp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
2146 struct tcp_iter_state
*s
;
2149 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2150 sizeof(struct tcp_iter_state
));
2154 s
= ((struct seq_file
*)file
->private_data
)->private;
2155 s
->family
= afinfo
->family
;
2159 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2162 struct proc_dir_entry
*p
;
2164 afinfo
->seq_fops
.open
= tcp_seq_open
;
2165 afinfo
->seq_fops
.read
= seq_read
;
2166 afinfo
->seq_fops
.llseek
= seq_lseek
;
2167 afinfo
->seq_fops
.release
= seq_release_net
;
2169 afinfo
->seq_ops
.start
= tcp_seq_start
;
2170 afinfo
->seq_ops
.next
= tcp_seq_next
;
2171 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2173 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2174 &afinfo
->seq_fops
, afinfo
);
2180 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2182 proc_net_remove(net
, afinfo
->name
);
2185 static void get_openreq4(struct sock
*sk
, struct request_sock
*req
,
2186 struct seq_file
*f
, int i
, int uid
, int *len
)
2188 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2189 int ttd
= req
->expires
- jiffies
;
2191 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2192 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
2195 ntohs(inet_sk(sk
)->sport
),
2197 ntohs(ireq
->rmt_port
),
2199 0, 0, /* could print option size, but that is af dependent. */
2200 1, /* timers active (only the expire timer) */
2201 jiffies_to_clock_t(ttd
),
2204 0, /* non standard timer */
2205 0, /* open_requests have no inode */
2206 atomic_read(&sk
->sk_refcnt
),
2211 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
, int *len
)
2214 unsigned long timer_expires
;
2215 struct tcp_sock
*tp
= tcp_sk(sk
);
2216 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2217 struct inet_sock
*inet
= inet_sk(sk
);
2218 __be32 dest
= inet
->daddr
;
2219 __be32 src
= inet
->rcv_saddr
;
2220 __u16 destp
= ntohs(inet
->dport
);
2221 __u16 srcp
= ntohs(inet
->sport
);
2223 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
) {
2225 timer_expires
= icsk
->icsk_timeout
;
2226 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2228 timer_expires
= icsk
->icsk_timeout
;
2229 } else if (timer_pending(&sk
->sk_timer
)) {
2231 timer_expires
= sk
->sk_timer
.expires
;
2234 timer_expires
= jiffies
;
2237 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2238 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
2239 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2240 tp
->write_seq
- tp
->snd_una
,
2241 sk
->sk_state
== TCP_LISTEN
? sk
->sk_ack_backlog
:
2242 (tp
->rcv_nxt
- tp
->copied_seq
),
2244 jiffies_to_clock_t(timer_expires
- jiffies
),
2245 icsk
->icsk_retransmits
,
2247 icsk
->icsk_probes_out
,
2249 atomic_read(&sk
->sk_refcnt
), sk
,
2250 jiffies_to_clock_t(icsk
->icsk_rto
),
2251 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2252 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2254 tp
->snd_ssthresh
>= 0xFFFF ? -1 : tp
->snd_ssthresh
,
2258 static void get_timewait4_sock(struct inet_timewait_sock
*tw
,
2259 struct seq_file
*f
, int i
, int *len
)
2263 int ttd
= tw
->tw_ttd
- jiffies
;
2268 dest
= tw
->tw_daddr
;
2269 src
= tw
->tw_rcv_saddr
;
2270 destp
= ntohs(tw
->tw_dport
);
2271 srcp
= ntohs(tw
->tw_sport
);
2273 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2274 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
2275 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2276 3, jiffies_to_clock_t(ttd
), 0, 0, 0, 0,
2277 atomic_read(&tw
->tw_refcnt
), tw
, len
);
2282 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2284 struct tcp_iter_state
* st
;
2287 if (v
== SEQ_START_TOKEN
) {
2288 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2289 " sl local_address rem_address st tx_queue "
2290 "rx_queue tr tm->when retrnsmt uid timeout "
2296 switch (st
->state
) {
2297 case TCP_SEQ_STATE_LISTENING
:
2298 case TCP_SEQ_STATE_ESTABLISHED
:
2299 get_tcp4_sock(v
, seq
, st
->num
, &len
);
2301 case TCP_SEQ_STATE_OPENREQ
:
2302 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
, &len
);
2304 case TCP_SEQ_STATE_TIME_WAIT
:
2305 get_timewait4_sock(v
, seq
, st
->num
, &len
);
2308 seq_printf(seq
, "%*s\n", TMPSZ
- 1 - len
, "");
2313 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2317 .owner
= THIS_MODULE
,
2320 .show
= tcp4_seq_show
,
2324 static int tcp4_proc_init_net(struct net
*net
)
2326 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2329 static void tcp4_proc_exit_net(struct net
*net
)
2331 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2334 static struct pernet_operations tcp4_net_ops
= {
2335 .init
= tcp4_proc_init_net
,
2336 .exit
= tcp4_proc_exit_net
,
2339 int __init
tcp4_proc_init(void)
2341 return register_pernet_subsys(&tcp4_net_ops
);
2344 void tcp4_proc_exit(void)
2346 unregister_pernet_subsys(&tcp4_net_ops
);
2348 #endif /* CONFIG_PROC_FS */
2350 struct proto tcp_prot
= {
2352 .owner
= THIS_MODULE
,
2354 .connect
= tcp_v4_connect
,
2355 .disconnect
= tcp_disconnect
,
2356 .accept
= inet_csk_accept
,
2358 .init
= tcp_v4_init_sock
,
2359 .destroy
= tcp_v4_destroy_sock
,
2360 .shutdown
= tcp_shutdown
,
2361 .setsockopt
= tcp_setsockopt
,
2362 .getsockopt
= tcp_getsockopt
,
2363 .recvmsg
= tcp_recvmsg
,
2364 .backlog_rcv
= tcp_v4_do_rcv
,
2366 .unhash
= inet_unhash
,
2367 .get_port
= inet_csk_get_port
,
2368 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2369 .sockets_allocated
= &tcp_sockets_allocated
,
2370 .orphan_count
= &tcp_orphan_count
,
2371 .memory_allocated
= &tcp_memory_allocated
,
2372 .memory_pressure
= &tcp_memory_pressure
,
2373 .sysctl_mem
= sysctl_tcp_mem
,
2374 .sysctl_wmem
= sysctl_tcp_wmem
,
2375 .sysctl_rmem
= sysctl_tcp_rmem
,
2376 .max_header
= MAX_TCP_HEADER
,
2377 .obj_size
= sizeof(struct tcp_sock
),
2378 .twsk_prot
= &tcp_timewait_sock_ops
,
2379 .rsk_prot
= &tcp_request_sock_ops
,
2380 .h
.hashinfo
= &tcp_hashinfo
,
2381 #ifdef CONFIG_COMPAT
2382 .compat_setsockopt
= compat_tcp_setsockopt
,
2383 .compat_getsockopt
= compat_tcp_getsockopt
,
2388 static int __net_init
tcp_sk_init(struct net
*net
)
2390 return inet_ctl_sock_create(&net
->ipv4
.tcp_sock
,
2391 PF_INET
, SOCK_RAW
, IPPROTO_TCP
, net
);
2394 static void __net_exit
tcp_sk_exit(struct net
*net
)
2396 inet_ctl_sock_destroy(net
->ipv4
.tcp_sock
);
2397 inet_twsk_purge(net
, &tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2400 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2401 .init
= tcp_sk_init
,
2402 .exit
= tcp_sk_exit
,
2405 void __init
tcp_v4_init(void)
2407 if (register_pernet_device(&tcp_sk_ops
))
2408 panic("Failed to create the TCP control socket.\n");
2411 EXPORT_SYMBOL(ipv4_specific
);
2412 EXPORT_SYMBOL(tcp_hashinfo
);
2413 EXPORT_SYMBOL(tcp_prot
);
2414 EXPORT_SYMBOL(tcp_v4_conn_request
);
2415 EXPORT_SYMBOL(tcp_v4_connect
);
2416 EXPORT_SYMBOL(tcp_v4_do_rcv
);
2417 EXPORT_SYMBOL(tcp_v4_remember_stamp
);
2418 EXPORT_SYMBOL(tcp_v4_send_check
);
2419 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
2421 #ifdef CONFIG_PROC_FS
2422 EXPORT_SYMBOL(tcp_proc_register
);
2423 EXPORT_SYMBOL(tcp_proc_unregister
);
2425 EXPORT_SYMBOL(sysctl_tcp_low_latency
);