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 * Version: $Id: tcp_ipv4.c,v 1.240 2002/02/01 22:01:04 davem Exp $
10 * IPv4 specific functions
15 * linux/ipv4/tcp_input.c
16 * linux/ipv4/tcp_output.c
18 * See tcp.c for author information
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
28 * David S. Miller : New socket lookup architecture.
29 * This code is dedicated to John Dyson.
30 * David S. Miller : Change semantics of established hash,
31 * half is devoted to TIME_WAIT sockets
32 * and the rest go in the other half.
33 * Andi Kleen : Add support for syncookies and fixed
34 * some bugs: ip options weren't passed to
35 * the TCP layer, missed a check for an
37 * Andi Kleen : Implemented fast path mtu discovery.
38 * Fixed many serious bugs in the
39 * request_sock handling and moved
40 * most of it into the af independent code.
41 * Added tail drop and some other bugfixes.
42 * Added new listen semantics.
43 * Mike McLagan : Routing by source
44 * Juan Jose Ciarlante: ip_dynaddr bits
45 * Andi Kleen: various fixes.
46 * Vitaly E. Lavrov : Transparent proxy revived after year
48 * Andi Kleen : Fix new listen.
49 * Andi Kleen : Fix accept error reporting.
50 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
51 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
52 * a single port at the same time.
56 #include <linux/types.h>
57 #include <linux/fcntl.h>
58 #include <linux/module.h>
59 #include <linux/random.h>
60 #include <linux/cache.h>
61 #include <linux/jhash.h>
62 #include <linux/init.h>
63 #include <linux/times.h>
66 #include <net/inet_hashtables.h>
68 #include <net/transp_v6.h>
70 #include <net/inet_common.h>
71 #include <net/timewait_sock.h>
73 #include <net/netdma.h>
75 #include <linux/inet.h>
76 #include <linux/ipv6.h>
77 #include <linux/stddef.h>
78 #include <linux/proc_fs.h>
79 #include <linux/seq_file.h>
81 #include <linux/crypto.h>
82 #include <linux/scatterlist.h>
84 int sysctl_tcp_tw_reuse __read_mostly
;
85 int sysctl_tcp_low_latency __read_mostly
;
87 /* Check TCP sequence numbers in ICMP packets. */
88 #define ICMP_MIN_LENGTH 8
90 /* Socket used for sending RSTs */
91 static struct socket
*tcp_socket __read_mostly
;
93 void tcp_v4_send_check(struct sock
*sk
, int len
, struct sk_buff
*skb
);
95 #ifdef CONFIG_TCP_MD5SIG
96 static struct tcp_md5sig_key
*tcp_v4_md5_do_lookup(struct sock
*sk
,
98 static int tcp_v4_do_calc_md5_hash(char *md5_hash
, struct tcp_md5sig_key
*key
,
99 __be32 saddr
, __be32 daddr
,
100 struct tcphdr
*th
, int protocol
,
104 struct inet_hashinfo __cacheline_aligned tcp_hashinfo
= {
105 .lhash_lock
= __RW_LOCK_UNLOCKED(tcp_hashinfo
.lhash_lock
),
106 .lhash_users
= ATOMIC_INIT(0),
107 .lhash_wait
= __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo
.lhash_wait
),
110 static int tcp_v4_get_port(struct sock
*sk
, unsigned short snum
)
112 return inet_csk_get_port(&tcp_hashinfo
, sk
, snum
,
113 inet_csk_bind_conflict
);
116 static void tcp_v4_hash(struct sock
*sk
)
118 inet_hash(&tcp_hashinfo
, sk
);
121 void tcp_unhash(struct sock
*sk
)
123 inet_unhash(&tcp_hashinfo
, sk
);
126 static inline __u32
tcp_v4_init_sequence(struct sk_buff
*skb
)
128 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
131 tcp_hdr(skb
)->source
);
134 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
)
136 const struct tcp_timewait_sock
*tcptw
= tcp_twsk(sktw
);
137 struct tcp_sock
*tp
= tcp_sk(sk
);
139 /* With PAWS, it is safe from the viewpoint
140 of data integrity. Even without PAWS it is safe provided sequence
141 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
143 Actually, the idea is close to VJ's one, only timestamp cache is
144 held not per host, but per port pair and TW bucket is used as state
147 If TW bucket has been already destroyed we fall back to VJ's scheme
148 and use initial timestamp retrieved from peer table.
150 if (tcptw
->tw_ts_recent_stamp
&&
151 (twp
== NULL
|| (sysctl_tcp_tw_reuse
&&
152 get_seconds() - tcptw
->tw_ts_recent_stamp
> 1))) {
153 tp
->write_seq
= tcptw
->tw_snd_nxt
+ 65535 + 2;
154 if (tp
->write_seq
== 0)
156 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
157 tp
->rx_opt
.ts_recent_stamp
= tcptw
->tw_ts_recent_stamp
;
165 EXPORT_SYMBOL_GPL(tcp_twsk_unique
);
167 /* This will initiate an outgoing connection. */
168 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
170 struct inet_sock
*inet
= inet_sk(sk
);
171 struct tcp_sock
*tp
= tcp_sk(sk
);
172 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
174 __be32 daddr
, nexthop
;
178 if (addr_len
< sizeof(struct sockaddr_in
))
181 if (usin
->sin_family
!= AF_INET
)
182 return -EAFNOSUPPORT
;
184 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
185 if (inet
->opt
&& inet
->opt
->srr
) {
188 nexthop
= inet
->opt
->faddr
;
191 tmp
= ip_route_connect(&rt
, nexthop
, inet
->saddr
,
192 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
194 inet
->sport
, usin
->sin_port
, sk
, 1);
196 if (tmp
== -ENETUNREACH
)
197 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES
);
201 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
206 if (!inet
->opt
|| !inet
->opt
->srr
)
210 inet
->saddr
= rt
->rt_src
;
211 inet
->rcv_saddr
= inet
->saddr
;
213 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->daddr
!= daddr
) {
214 /* Reset inherited state */
215 tp
->rx_opt
.ts_recent
= 0;
216 tp
->rx_opt
.ts_recent_stamp
= 0;
220 if (tcp_death_row
.sysctl_tw_recycle
&&
221 !tp
->rx_opt
.ts_recent_stamp
&& rt
->rt_dst
== daddr
) {
222 struct inet_peer
*peer
= rt_get_peer(rt
);
224 * VJ's idea. We save last timestamp seen from
225 * the destination in peer table, when entering state
226 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
227 * when trying new connection.
230 peer
->tcp_ts_stamp
+ TCP_PAWS_MSL
>= get_seconds()) {
231 tp
->rx_opt
.ts_recent_stamp
= peer
->tcp_ts_stamp
;
232 tp
->rx_opt
.ts_recent
= peer
->tcp_ts
;
236 inet
->dport
= usin
->sin_port
;
239 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
241 inet_csk(sk
)->icsk_ext_hdr_len
= inet
->opt
->optlen
;
243 tp
->rx_opt
.mss_clamp
= 536;
245 /* Socket identity is still unknown (sport may be zero).
246 * However we set state to SYN-SENT and not releasing socket
247 * lock select source port, enter ourselves into the hash tables and
248 * complete initialization after this.
250 tcp_set_state(sk
, TCP_SYN_SENT
);
251 err
= inet_hash_connect(&tcp_death_row
, sk
);
255 err
= ip_route_newports(&rt
, IPPROTO_TCP
,
256 inet
->sport
, inet
->dport
, sk
);
260 /* OK, now commit destination to socket. */
261 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
262 sk_setup_caps(sk
, &rt
->u
.dst
);
265 tp
->write_seq
= secure_tcp_sequence_number(inet
->saddr
,
270 inet
->id
= tp
->write_seq
^ jiffies
;
272 err
= tcp_connect(sk
);
281 * This unhashes the socket and releases the local port,
284 tcp_set_state(sk
, TCP_CLOSE
);
286 sk
->sk_route_caps
= 0;
292 * This routine does path mtu discovery as defined in RFC1191.
294 static void do_pmtu_discovery(struct sock
*sk
, struct iphdr
*iph
, u32 mtu
)
296 struct dst_entry
*dst
;
297 struct inet_sock
*inet
= inet_sk(sk
);
299 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
300 * send out by Linux are always <576bytes so they should go through
303 if (sk
->sk_state
== TCP_LISTEN
)
306 /* We don't check in the destentry if pmtu discovery is forbidden
307 * on this route. We just assume that no packet_to_big packets
308 * are send back when pmtu discovery is not active.
309 * There is a small race when the user changes this flag in the
310 * route, but I think that's acceptable.
312 if ((dst
= __sk_dst_check(sk
, 0)) == NULL
)
315 dst
->ops
->update_pmtu(dst
, mtu
);
317 /* Something is about to be wrong... Remember soft error
318 * for the case, if this connection will not able to recover.
320 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
321 sk
->sk_err_soft
= EMSGSIZE
;
325 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
326 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
327 tcp_sync_mss(sk
, mtu
);
329 /* Resend the TCP packet because it's
330 * clear that the old packet has been
331 * dropped. This is the new "fast" path mtu
334 tcp_simple_retransmit(sk
);
335 } /* else let the usual retransmit timer handle it */
339 * This routine is called by the ICMP module when it gets some
340 * sort of error condition. If err < 0 then the socket should
341 * be closed and the error returned to the user. If err > 0
342 * it's just the icmp type << 8 | icmp code. After adjustment
343 * header points to the first 8 bytes of the tcp header. We need
344 * to find the appropriate port.
346 * The locking strategy used here is very "optimistic". When
347 * someone else accesses the socket the ICMP is just dropped
348 * and for some paths there is no check at all.
349 * A more general error queue to queue errors for later handling
350 * is probably better.
354 void tcp_v4_err(struct sk_buff
*skb
, u32 info
)
356 struct iphdr
*iph
= (struct iphdr
*)skb
->data
;
357 struct tcphdr
*th
= (struct tcphdr
*)(skb
->data
+ (iph
->ihl
<< 2));
359 struct inet_sock
*inet
;
360 const int type
= icmp_hdr(skb
)->type
;
361 const int code
= icmp_hdr(skb
)->code
;
366 if (skb
->len
< (iph
->ihl
<< 2) + 8) {
367 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS
);
371 sk
= inet_lookup(&tcp_hashinfo
, iph
->daddr
, th
->dest
, iph
->saddr
,
372 th
->source
, inet_iif(skb
));
374 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS
);
377 if (sk
->sk_state
== TCP_TIME_WAIT
) {
378 inet_twsk_put(inet_twsk(sk
));
383 /* If too many ICMPs get dropped on busy
384 * servers this needs to be solved differently.
386 if (sock_owned_by_user(sk
))
387 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS
);
389 if (sk
->sk_state
== TCP_CLOSE
)
393 seq
= ntohl(th
->seq
);
394 if (sk
->sk_state
!= TCP_LISTEN
&&
395 !between(seq
, tp
->snd_una
, tp
->snd_nxt
)) {
396 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS
);
401 case ICMP_SOURCE_QUENCH
:
402 /* Just silently ignore these. */
404 case ICMP_PARAMETERPROB
:
407 case ICMP_DEST_UNREACH
:
408 if (code
> NR_ICMP_UNREACH
)
411 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
412 if (!sock_owned_by_user(sk
))
413 do_pmtu_discovery(sk
, iph
, info
);
417 err
= icmp_err_convert
[code
].errno
;
419 case ICMP_TIME_EXCEEDED
:
426 switch (sk
->sk_state
) {
427 struct request_sock
*req
, **prev
;
429 if (sock_owned_by_user(sk
))
432 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
433 iph
->daddr
, iph
->saddr
);
437 /* ICMPs are not backlogged, hence we cannot get
438 an established socket here.
442 if (seq
!= tcp_rsk(req
)->snt_isn
) {
443 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS
);
448 * Still in SYN_RECV, just remove it silently.
449 * There is no good way to pass the error to the newly
450 * created socket, and POSIX does not want network
451 * errors returned from accept().
453 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
457 case TCP_SYN_RECV
: /* Cannot happen.
458 It can f.e. if SYNs crossed.
460 if (!sock_owned_by_user(sk
)) {
463 sk
->sk_error_report(sk
);
467 sk
->sk_err_soft
= err
;
472 /* If we've already connected we will keep trying
473 * until we time out, or the user gives up.
475 * rfc1122 4.2.3.9 allows to consider as hard errors
476 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
477 * but it is obsoleted by pmtu discovery).
479 * Note, that in modern internet, where routing is unreliable
480 * and in each dark corner broken firewalls sit, sending random
481 * errors ordered by their masters even this two messages finally lose
482 * their original sense (even Linux sends invalid PORT_UNREACHs)
484 * Now we are in compliance with RFCs.
489 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
491 sk
->sk_error_report(sk
);
492 } else { /* Only an error on timeout */
493 sk
->sk_err_soft
= err
;
501 /* This routine computes an IPv4 TCP checksum. */
502 void tcp_v4_send_check(struct sock
*sk
, int len
, struct sk_buff
*skb
)
504 struct inet_sock
*inet
= inet_sk(sk
);
505 struct tcphdr
*th
= tcp_hdr(skb
);
507 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
508 th
->check
= ~tcp_v4_check(len
, inet
->saddr
,
510 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
511 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
513 th
->check
= tcp_v4_check(len
, inet
->saddr
, inet
->daddr
,
514 csum_partial((char *)th
,
520 int tcp_v4_gso_send_check(struct sk_buff
*skb
)
522 const struct iphdr
*iph
;
525 if (!pskb_may_pull(skb
, sizeof(*th
)))
532 th
->check
= ~tcp_v4_check(skb
->len
, iph
->saddr
, iph
->daddr
, 0);
533 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
534 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
535 skb
->ip_summed
= CHECKSUM_PARTIAL
;
540 * This routine will send an RST to the other tcp.
542 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
544 * Answer: if a packet caused RST, it is not for a socket
545 * existing in our system, if it is matched to a socket,
546 * it is just duplicate segment or bug in other side's TCP.
547 * So that we build reply only basing on parameters
548 * arrived with segment.
549 * Exception: precedence violation. We do not implement it in any case.
552 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
554 struct tcphdr
*th
= tcp_hdr(skb
);
557 #ifdef CONFIG_TCP_MD5SIG
558 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
561 struct ip_reply_arg arg
;
562 #ifdef CONFIG_TCP_MD5SIG
563 struct tcp_md5sig_key
*key
;
566 /* Never send a reset in response to a reset. */
570 if (((struct rtable
*)skb
->dst
)->rt_type
!= RTN_LOCAL
)
573 /* Swap the send and the receive. */
574 memset(&rep
, 0, sizeof(rep
));
575 rep
.th
.dest
= th
->source
;
576 rep
.th
.source
= th
->dest
;
577 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
581 rep
.th
.seq
= th
->ack_seq
;
584 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
585 skb
->len
- (th
->doff
<< 2));
588 memset(&arg
, 0, sizeof(arg
));
589 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
590 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
592 #ifdef CONFIG_TCP_MD5SIG
593 key
= sk
? tcp_v4_md5_do_lookup(sk
, ip_hdr(skb
)->daddr
) : NULL
;
595 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
597 (TCPOPT_MD5SIG
<< 8) |
599 /* Update length and the length the header thinks exists */
600 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
601 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
603 tcp_v4_do_calc_md5_hash((__u8
*)&rep
.opt
[1],
607 &rep
.th
, IPPROTO_TCP
,
611 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
612 ip_hdr(skb
)->saddr
, /* XXX */
613 sizeof(struct tcphdr
), IPPROTO_TCP
, 0);
614 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
616 ip_send_reply(tcp_socket
->sk
, skb
, &arg
, arg
.iov
[0].iov_len
);
618 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS
);
619 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS
);
622 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
623 outside socket context is ugly, certainly. What can I do?
626 static void tcp_v4_send_ack(struct tcp_timewait_sock
*twsk
,
627 struct sk_buff
*skb
, u32 seq
, u32 ack
,
630 struct tcphdr
*th
= tcp_hdr(skb
);
633 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
634 #ifdef CONFIG_TCP_MD5SIG
635 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
639 struct ip_reply_arg arg
;
640 #ifdef CONFIG_TCP_MD5SIG
641 struct tcp_md5sig_key
*key
;
642 struct tcp_md5sig_key tw_key
;
645 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
646 memset(&arg
, 0, sizeof(arg
));
648 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
649 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
651 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
652 (TCPOPT_TIMESTAMP
<< 8) |
654 rep
.opt
[1] = htonl(tcp_time_stamp
);
655 rep
.opt
[2] = htonl(ts
);
656 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
659 /* Swap the send and the receive. */
660 rep
.th
.dest
= th
->source
;
661 rep
.th
.source
= th
->dest
;
662 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
663 rep
.th
.seq
= htonl(seq
);
664 rep
.th
.ack_seq
= htonl(ack
);
666 rep
.th
.window
= htons(win
);
668 #ifdef CONFIG_TCP_MD5SIG
670 * The SKB holds an imcoming packet, but may not have a valid ->sk
671 * pointer. This is especially the case when we're dealing with a
672 * TIME_WAIT ack, because the sk structure is long gone, and only
673 * the tcp_timewait_sock remains. So the md5 key is stashed in that
674 * structure, and we use it in preference. I believe that (twsk ||
675 * skb->sk) holds true, but we program defensively.
677 if (!twsk
&& skb
->sk
) {
678 key
= tcp_v4_md5_do_lookup(skb
->sk
, ip_hdr(skb
)->daddr
);
679 } else if (twsk
&& twsk
->tw_md5_keylen
) {
680 tw_key
.key
= twsk
->tw_md5_key
;
681 tw_key
.keylen
= twsk
->tw_md5_keylen
;
687 int offset
= (ts
) ? 3 : 0;
689 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
691 (TCPOPT_MD5SIG
<< 8) |
693 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
694 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
696 tcp_v4_do_calc_md5_hash((__u8
*)&rep
.opt
[offset
],
700 &rep
.th
, IPPROTO_TCP
,
704 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
705 ip_hdr(skb
)->saddr
, /* XXX */
706 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
707 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
709 arg
.bound_dev_if
= twsk
->tw_sk
.tw_bound_dev_if
;
711 ip_send_reply(tcp_socket
->sk
, skb
, &arg
, arg
.iov
[0].iov_len
);
713 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS
);
716 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
718 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
719 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
721 tcp_v4_send_ack(tcptw
, skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
722 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
723 tcptw
->tw_ts_recent
);
728 static void tcp_v4_reqsk_send_ack(struct sk_buff
*skb
,
729 struct request_sock
*req
)
731 tcp_v4_send_ack(NULL
, skb
, tcp_rsk(req
)->snt_isn
+ 1,
732 tcp_rsk(req
)->rcv_isn
+ 1, req
->rcv_wnd
,
737 * Send a SYN-ACK after having received an ACK.
738 * This still operates on a request_sock only, not on a big
741 static int tcp_v4_send_synack(struct sock
*sk
, struct request_sock
*req
,
742 struct dst_entry
*dst
)
744 const struct inet_request_sock
*ireq
= inet_rsk(req
);
746 struct sk_buff
* skb
;
748 /* First, grab a route. */
749 if (!dst
&& (dst
= inet_csk_route_req(sk
, req
)) == NULL
)
752 skb
= tcp_make_synack(sk
, dst
, req
);
755 struct tcphdr
*th
= tcp_hdr(skb
);
757 th
->check
= tcp_v4_check(skb
->len
,
760 csum_partial((char *)th
, skb
->len
,
763 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
766 err
= net_xmit_eval(err
);
775 * IPv4 request_sock destructor.
777 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
779 kfree(inet_rsk(req
)->opt
);
782 #ifdef CONFIG_SYN_COOKIES
783 static void syn_flood_warning(struct sk_buff
*skb
)
785 static unsigned long warntime
;
787 if (time_after(jiffies
, (warntime
+ HZ
* 60))) {
790 "possible SYN flooding on port %d. Sending cookies.\n",
791 ntohs(tcp_hdr(skb
)->dest
));
797 * Save and compile IPv4 options into the request_sock if needed.
799 static struct ip_options
*tcp_v4_save_options(struct sock
*sk
,
802 struct ip_options
*opt
= &(IPCB(skb
)->opt
);
803 struct ip_options
*dopt
= NULL
;
805 if (opt
&& opt
->optlen
) {
806 int opt_size
= optlength(opt
);
807 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
809 if (ip_options_echo(dopt
, skb
)) {
818 #ifdef CONFIG_TCP_MD5SIG
820 * RFC2385 MD5 checksumming requires a mapping of
821 * IP address->MD5 Key.
822 * We need to maintain these in the sk structure.
825 /* Find the Key structure for an address. */
826 static struct tcp_md5sig_key
*
827 tcp_v4_md5_do_lookup(struct sock
*sk
, __be32 addr
)
829 struct tcp_sock
*tp
= tcp_sk(sk
);
832 if (!tp
->md5sig_info
|| !tp
->md5sig_info
->entries4
)
834 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++) {
835 if (tp
->md5sig_info
->keys4
[i
].addr
== addr
)
836 return (struct tcp_md5sig_key
*)
837 &tp
->md5sig_info
->keys4
[i
];
842 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
843 struct sock
*addr_sk
)
845 return tcp_v4_md5_do_lookup(sk
, inet_sk(addr_sk
)->daddr
);
848 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
850 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
851 struct request_sock
*req
)
853 return tcp_v4_md5_do_lookup(sk
, inet_rsk(req
)->rmt_addr
);
856 /* This can be called on a newly created socket, from other files */
857 int tcp_v4_md5_do_add(struct sock
*sk
, __be32 addr
,
858 u8
*newkey
, u8 newkeylen
)
860 /* Add Key to the list */
861 struct tcp4_md5sig_key
*key
;
862 struct tcp_sock
*tp
= tcp_sk(sk
);
863 struct tcp4_md5sig_key
*keys
;
865 key
= (struct tcp4_md5sig_key
*)tcp_v4_md5_do_lookup(sk
, addr
);
867 /* Pre-existing entry - just update that one. */
870 key
->keylen
= newkeylen
;
872 struct tcp_md5sig_info
*md5sig
;
874 if (!tp
->md5sig_info
) {
875 tp
->md5sig_info
= kzalloc(sizeof(*tp
->md5sig_info
),
877 if (!tp
->md5sig_info
) {
881 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
883 if (tcp_alloc_md5sig_pool() == NULL
) {
887 md5sig
= tp
->md5sig_info
;
889 if (md5sig
->alloced4
== md5sig
->entries4
) {
890 keys
= kmalloc((sizeof(*keys
) *
891 (md5sig
->entries4
+ 1)), GFP_ATOMIC
);
894 tcp_free_md5sig_pool();
898 if (md5sig
->entries4
)
899 memcpy(keys
, md5sig
->keys4
,
900 sizeof(*keys
) * md5sig
->entries4
);
902 /* Free old key list, and reference new one */
904 kfree(md5sig
->keys4
);
905 md5sig
->keys4
= keys
;
909 md5sig
->keys4
[md5sig
->entries4
- 1].addr
= addr
;
910 md5sig
->keys4
[md5sig
->entries4
- 1].key
= newkey
;
911 md5sig
->keys4
[md5sig
->entries4
- 1].keylen
= newkeylen
;
916 EXPORT_SYMBOL(tcp_v4_md5_do_add
);
918 static int tcp_v4_md5_add_func(struct sock
*sk
, struct sock
*addr_sk
,
919 u8
*newkey
, u8 newkeylen
)
921 return tcp_v4_md5_do_add(sk
, inet_sk(addr_sk
)->daddr
,
925 int tcp_v4_md5_do_del(struct sock
*sk
, __be32 addr
)
927 struct tcp_sock
*tp
= tcp_sk(sk
);
930 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++) {
931 if (tp
->md5sig_info
->keys4
[i
].addr
== addr
) {
933 kfree(tp
->md5sig_info
->keys4
[i
].key
);
934 tp
->md5sig_info
->entries4
--;
936 if (tp
->md5sig_info
->entries4
== 0) {
937 kfree(tp
->md5sig_info
->keys4
);
938 tp
->md5sig_info
->keys4
= NULL
;
939 tp
->md5sig_info
->alloced4
= 0;
940 } else if (tp
->md5sig_info
->entries4
!= i
) {
941 /* Need to do some manipulation */
942 memcpy(&tp
->md5sig_info
->keys4
[i
],
943 &tp
->md5sig_info
->keys4
[i
+1],
944 (tp
->md5sig_info
->entries4
- i
) *
945 sizeof(struct tcp4_md5sig_key
));
947 tcp_free_md5sig_pool();
954 EXPORT_SYMBOL(tcp_v4_md5_do_del
);
956 static void tcp_v4_clear_md5_list(struct sock
*sk
)
958 struct tcp_sock
*tp
= tcp_sk(sk
);
960 /* Free each key, then the set of key keys,
961 * the crypto element, and then decrement our
962 * hold on the last resort crypto.
964 if (tp
->md5sig_info
->entries4
) {
966 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++)
967 kfree(tp
->md5sig_info
->keys4
[i
].key
);
968 tp
->md5sig_info
->entries4
= 0;
969 tcp_free_md5sig_pool();
971 if (tp
->md5sig_info
->keys4
) {
972 kfree(tp
->md5sig_info
->keys4
);
973 tp
->md5sig_info
->keys4
= NULL
;
974 tp
->md5sig_info
->alloced4
= 0;
978 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
981 struct tcp_md5sig cmd
;
982 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
985 if (optlen
< sizeof(cmd
))
988 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
991 if (sin
->sin_family
!= AF_INET
)
994 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
) {
995 if (!tcp_sk(sk
)->md5sig_info
)
997 return tcp_v4_md5_do_del(sk
, sin
->sin_addr
.s_addr
);
1000 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1003 if (!tcp_sk(sk
)->md5sig_info
) {
1004 struct tcp_sock
*tp
= tcp_sk(sk
);
1005 struct tcp_md5sig_info
*p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
1010 tp
->md5sig_info
= p
;
1011 sk
->sk_route_caps
&= ~NETIF_F_GSO_MASK
;
1014 newkey
= kmemdup(cmd
.tcpm_key
, cmd
.tcpm_keylen
, GFP_KERNEL
);
1017 return tcp_v4_md5_do_add(sk
, sin
->sin_addr
.s_addr
,
1018 newkey
, cmd
.tcpm_keylen
);
1021 static int tcp_v4_do_calc_md5_hash(char *md5_hash
, struct tcp_md5sig_key
*key
,
1022 __be32 saddr
, __be32 daddr
,
1023 struct tcphdr
*th
, int protocol
,
1026 struct scatterlist sg
[4];
1029 __sum16 old_checksum
;
1030 struct tcp_md5sig_pool
*hp
;
1031 struct tcp4_pseudohdr
*bp
;
1032 struct hash_desc
*desc
;
1034 unsigned int nbytes
= 0;
1037 * Okay, so RFC2385 is turned on for this connection,
1038 * so we need to generate the MD5 hash for the packet now.
1041 hp
= tcp_get_md5sig_pool();
1043 goto clear_hash_noput
;
1045 bp
= &hp
->md5_blk
.ip4
;
1046 desc
= &hp
->md5_desc
;
1049 * 1. the TCP pseudo-header (in the order: source IP address,
1050 * destination IP address, zero-padded protocol number, and
1056 bp
->protocol
= protocol
;
1057 bp
->len
= htons(tcplen
);
1058 sg_set_buf(&sg
[block
++], bp
, sizeof(*bp
));
1059 nbytes
+= sizeof(*bp
);
1061 /* 2. the TCP header, excluding options, and assuming a
1064 old_checksum
= th
->check
;
1066 sg_set_buf(&sg
[block
++], th
, sizeof(struct tcphdr
));
1067 nbytes
+= sizeof(struct tcphdr
);
1069 /* 3. the TCP segment data (if any) */
1070 data_len
= tcplen
- (th
->doff
<< 2);
1072 unsigned char *data
= (unsigned char *)th
+ (th
->doff
<< 2);
1073 sg_set_buf(&sg
[block
++], data
, data_len
);
1077 /* 4. an independently-specified key or password, known to both
1078 * TCPs and presumably connection-specific
1080 sg_set_buf(&sg
[block
++], key
->key
, key
->keylen
);
1081 nbytes
+= key
->keylen
;
1083 /* Now store the Hash into the packet */
1084 err
= crypto_hash_init(desc
);
1087 err
= crypto_hash_update(desc
, sg
, nbytes
);
1090 err
= crypto_hash_final(desc
, md5_hash
);
1094 /* Reset header, and free up the crypto */
1095 tcp_put_md5sig_pool();
1096 th
->check
= old_checksum
;
1101 tcp_put_md5sig_pool();
1103 memset(md5_hash
, 0, 16);
1107 int tcp_v4_calc_md5_hash(char *md5_hash
, struct tcp_md5sig_key
*key
,
1109 struct dst_entry
*dst
,
1110 struct request_sock
*req
,
1111 struct tcphdr
*th
, int protocol
,
1114 __be32 saddr
, daddr
;
1117 saddr
= inet_sk(sk
)->saddr
;
1118 daddr
= inet_sk(sk
)->daddr
;
1120 struct rtable
*rt
= (struct rtable
*)dst
;
1125 return tcp_v4_do_calc_md5_hash(md5_hash
, key
,
1127 th
, protocol
, tcplen
);
1130 EXPORT_SYMBOL(tcp_v4_calc_md5_hash
);
1132 static int tcp_v4_inbound_md5_hash(struct sock
*sk
, struct sk_buff
*skb
)
1135 * This gets called for each TCP segment that arrives
1136 * so we want to be efficient.
1137 * We have 3 drop cases:
1138 * o No MD5 hash and one expected.
1139 * o MD5 hash and we're not expecting one.
1140 * o MD5 hash and its wrong.
1142 __u8
*hash_location
= NULL
;
1143 struct tcp_md5sig_key
*hash_expected
;
1144 const struct iphdr
*iph
= ip_hdr(skb
);
1145 struct tcphdr
*th
= tcp_hdr(skb
);
1146 int length
= (th
->doff
<< 2) - sizeof(struct tcphdr
);
1149 unsigned char newhash
[16];
1151 hash_expected
= tcp_v4_md5_do_lookup(sk
, iph
->saddr
);
1154 * If the TCP option length is less than the TCP_MD5SIG
1155 * option length, then we can shortcut
1157 if (length
< TCPOLEN_MD5SIG
) {
1164 /* Okay, we can't shortcut - we have to grub through the options */
1165 ptr
= (unsigned char *)(th
+ 1);
1166 while (length
> 0) {
1167 int opcode
= *ptr
++;
1180 if (opsize
> length
)
1183 if (opcode
== TCPOPT_MD5SIG
) {
1184 hash_location
= ptr
;
1192 /* We've parsed the options - do we have a hash? */
1193 if (!hash_expected
&& !hash_location
)
1196 if (hash_expected
&& !hash_location
) {
1197 LIMIT_NETDEBUG(KERN_INFO
"MD5 Hash expected but NOT found "
1198 "(" NIPQUAD_FMT
", %d)->(" NIPQUAD_FMT
", %d)\n",
1199 NIPQUAD(iph
->saddr
), ntohs(th
->source
),
1200 NIPQUAD(iph
->daddr
), ntohs(th
->dest
));
1204 if (!hash_expected
&& hash_location
) {
1205 LIMIT_NETDEBUG(KERN_INFO
"MD5 Hash NOT expected but found "
1206 "(" NIPQUAD_FMT
", %d)->(" NIPQUAD_FMT
", %d)\n",
1207 NIPQUAD(iph
->saddr
), ntohs(th
->source
),
1208 NIPQUAD(iph
->daddr
), ntohs(th
->dest
));
1212 /* Okay, so this is hash_expected and hash_location -
1213 * so we need to calculate the checksum.
1215 genhash
= tcp_v4_do_calc_md5_hash(newhash
,
1217 iph
->saddr
, iph
->daddr
,
1218 th
, sk
->sk_protocol
,
1221 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1222 if (net_ratelimit()) {
1223 printk(KERN_INFO
"MD5 Hash failed for "
1224 "(" NIPQUAD_FMT
", %d)->(" NIPQUAD_FMT
", %d)%s\n",
1225 NIPQUAD(iph
->saddr
), ntohs(th
->source
),
1226 NIPQUAD(iph
->daddr
), ntohs(th
->dest
),
1227 genhash
? " tcp_v4_calc_md5_hash failed" : "");
1236 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1238 .obj_size
= sizeof(struct tcp_request_sock
),
1239 .rtx_syn_ack
= tcp_v4_send_synack
,
1240 .send_ack
= tcp_v4_reqsk_send_ack
,
1241 .destructor
= tcp_v4_reqsk_destructor
,
1242 .send_reset
= tcp_v4_send_reset
,
1245 #ifdef CONFIG_TCP_MD5SIG
1246 static struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1247 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1251 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1252 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1253 .twsk_unique
= tcp_twsk_unique
,
1254 .twsk_destructor
= tcp_twsk_destructor
,
1257 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1259 struct inet_request_sock
*ireq
;
1260 struct tcp_options_received tmp_opt
;
1261 struct request_sock
*req
;
1262 __be32 saddr
= ip_hdr(skb
)->saddr
;
1263 __be32 daddr
= ip_hdr(skb
)->daddr
;
1264 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1265 struct dst_entry
*dst
= NULL
;
1266 #ifdef CONFIG_SYN_COOKIES
1267 int want_cookie
= 0;
1269 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1272 /* Never answer to SYNs send to broadcast or multicast */
1273 if (((struct rtable
*)skb
->dst
)->rt_flags
&
1274 (RTCF_BROADCAST
| RTCF_MULTICAST
))
1277 /* TW buckets are converted to open requests without
1278 * limitations, they conserve resources and peer is
1279 * evidently real one.
1281 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
1282 #ifdef CONFIG_SYN_COOKIES
1283 if (sysctl_tcp_syncookies
) {
1290 /* Accept backlog is full. If we have already queued enough
1291 * of warm entries in syn queue, drop request. It is better than
1292 * clogging syn queue with openreqs with exponentially increasing
1295 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1)
1298 req
= reqsk_alloc(&tcp_request_sock_ops
);
1302 #ifdef CONFIG_TCP_MD5SIG
1303 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1306 tcp_clear_options(&tmp_opt
);
1307 tmp_opt
.mss_clamp
= 536;
1308 tmp_opt
.user_mss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1310 tcp_parse_options(skb
, &tmp_opt
, 0);
1313 tcp_clear_options(&tmp_opt
);
1314 tmp_opt
.saw_tstamp
= 0;
1317 if (tmp_opt
.saw_tstamp
&& !tmp_opt
.rcv_tsval
) {
1318 /* Some OSes (unknown ones, but I see them on web server, which
1319 * contains information interesting only for windows'
1320 * users) do not send their stamp in SYN. It is easy case.
1321 * We simply do not advertise TS support.
1323 tmp_opt
.saw_tstamp
= 0;
1324 tmp_opt
.tstamp_ok
= 0;
1326 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1328 tcp_openreq_init(req
, &tmp_opt
, skb
);
1330 if (security_inet_conn_request(sk
, skb
, req
))
1333 ireq
= inet_rsk(req
);
1334 ireq
->loc_addr
= daddr
;
1335 ireq
->rmt_addr
= saddr
;
1336 ireq
->opt
= tcp_v4_save_options(sk
, skb
);
1338 TCP_ECN_create_request(req
, tcp_hdr(skb
));
1341 #ifdef CONFIG_SYN_COOKIES
1342 syn_flood_warning(skb
);
1344 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1346 struct inet_peer
*peer
= NULL
;
1348 /* VJ's idea. We save last timestamp seen
1349 * from the destination in peer table, when entering
1350 * state TIME-WAIT, and check against it before
1351 * accepting new connection request.
1353 * If "isn" is not zero, this request hit alive
1354 * timewait bucket, so that all the necessary checks
1355 * are made in the function processing timewait state.
1357 if (tmp_opt
.saw_tstamp
&&
1358 tcp_death_row
.sysctl_tw_recycle
&&
1359 (dst
= inet_csk_route_req(sk
, req
)) != NULL
&&
1360 (peer
= rt_get_peer((struct rtable
*)dst
)) != NULL
&&
1361 peer
->v4daddr
== saddr
) {
1362 if (get_seconds() < peer
->tcp_ts_stamp
+ TCP_PAWS_MSL
&&
1363 (s32
)(peer
->tcp_ts
- req
->ts_recent
) >
1365 NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED
);
1370 /* Kill the following clause, if you dislike this way. */
1371 else if (!sysctl_tcp_syncookies
&&
1372 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1373 (sysctl_max_syn_backlog
>> 2)) &&
1374 (!peer
|| !peer
->tcp_ts_stamp
) &&
1375 (!dst
|| !dst_metric(dst
, RTAX_RTT
))) {
1376 /* Without syncookies last quarter of
1377 * backlog is filled with destinations,
1378 * proven to be alive.
1379 * It means that we continue to communicate
1380 * to destinations, already remembered
1381 * to the moment of synflood.
1383 LIMIT_NETDEBUG(KERN_DEBUG
"TCP: drop open "
1384 "request from %u.%u.%u.%u/%u\n",
1386 ntohs(tcp_hdr(skb
)->source
));
1391 isn
= tcp_v4_init_sequence(skb
);
1393 tcp_rsk(req
)->snt_isn
= isn
;
1395 if (tcp_v4_send_synack(sk
, req
, dst
))
1401 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1413 * The three way handshake has completed - we got a valid synack -
1414 * now create the new socket.
1416 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1417 struct request_sock
*req
,
1418 struct dst_entry
*dst
)
1420 struct inet_request_sock
*ireq
;
1421 struct inet_sock
*newinet
;
1422 struct tcp_sock
*newtp
;
1424 #ifdef CONFIG_TCP_MD5SIG
1425 struct tcp_md5sig_key
*key
;
1428 if (sk_acceptq_is_full(sk
))
1431 if (!dst
&& (dst
= inet_csk_route_req(sk
, req
)) == NULL
)
1434 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1438 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1439 sk_setup_caps(newsk
, dst
);
1441 newtp
= tcp_sk(newsk
);
1442 newinet
= inet_sk(newsk
);
1443 ireq
= inet_rsk(req
);
1444 newinet
->daddr
= ireq
->rmt_addr
;
1445 newinet
->rcv_saddr
= ireq
->loc_addr
;
1446 newinet
->saddr
= ireq
->loc_addr
;
1447 newinet
->opt
= ireq
->opt
;
1449 newinet
->mc_index
= inet_iif(skb
);
1450 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1451 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1453 inet_csk(newsk
)->icsk_ext_hdr_len
= newinet
->opt
->optlen
;
1454 newinet
->id
= newtp
->write_seq
^ jiffies
;
1456 tcp_mtup_init(newsk
);
1457 tcp_sync_mss(newsk
, dst_mtu(dst
));
1458 newtp
->advmss
= dst_metric(dst
, RTAX_ADVMSS
);
1459 tcp_initialize_rcv_mss(newsk
);
1461 #ifdef CONFIG_TCP_MD5SIG
1462 /* Copy over the MD5 key from the original socket */
1463 if ((key
= tcp_v4_md5_do_lookup(sk
, newinet
->daddr
)) != NULL
) {
1465 * We're using one, so create a matching key
1466 * on the newsk structure. If we fail to get
1467 * memory, then we end up not copying the key
1470 char *newkey
= kmemdup(key
->key
, key
->keylen
, GFP_ATOMIC
);
1472 tcp_v4_md5_do_add(newsk
, inet_sk(sk
)->daddr
,
1473 newkey
, key
->keylen
);
1477 __inet_hash(&tcp_hashinfo
, newsk
, 0);
1478 __inet_inherit_port(&tcp_hashinfo
, sk
, newsk
);
1483 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS
);
1485 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS
);
1490 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1492 struct tcphdr
*th
= tcp_hdr(skb
);
1493 const struct iphdr
*iph
= ip_hdr(skb
);
1495 struct request_sock
**prev
;
1496 /* Find possible connection requests. */
1497 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1498 iph
->saddr
, iph
->daddr
);
1500 return tcp_check_req(sk
, skb
, req
, prev
);
1502 nsk
= inet_lookup_established(&tcp_hashinfo
, iph
->saddr
, th
->source
,
1503 iph
->daddr
, th
->dest
, inet_iif(skb
));
1506 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1510 inet_twsk_put(inet_twsk(nsk
));
1514 #ifdef CONFIG_SYN_COOKIES
1515 if (!th
->rst
&& !th
->syn
&& th
->ack
)
1516 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1521 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1523 const struct iphdr
*iph
= ip_hdr(skb
);
1525 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1526 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1527 iph
->daddr
, skb
->csum
)) {
1528 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1533 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1534 skb
->len
, IPPROTO_TCP
, 0);
1536 if (skb
->len
<= 76) {
1537 return __skb_checksum_complete(skb
);
1543 /* The socket must have it's spinlock held when we get
1546 * We have a potential double-lock case here, so even when
1547 * doing backlog processing we use the BH locking scheme.
1548 * This is because we cannot sleep with the original spinlock
1551 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1554 #ifdef CONFIG_TCP_MD5SIG
1556 * We really want to reject the packet as early as possible
1558 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1559 * o There is an MD5 option and we're not expecting one
1561 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1565 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1566 TCP_CHECK_TIMER(sk
);
1567 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1571 TCP_CHECK_TIMER(sk
);
1575 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1578 if (sk
->sk_state
== TCP_LISTEN
) {
1579 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1584 if (tcp_child_process(sk
, nsk
, skb
)) {
1592 TCP_CHECK_TIMER(sk
);
1593 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1597 TCP_CHECK_TIMER(sk
);
1601 tcp_v4_send_reset(rsk
, skb
);
1604 /* Be careful here. If this function gets more complicated and
1605 * gcc suffers from register pressure on the x86, sk (in %ebx)
1606 * might be destroyed here. This current version compiles correctly,
1607 * but you have been warned.
1612 TCP_INC_STATS_BH(TCP_MIB_INERRS
);
1620 int tcp_v4_rcv(struct sk_buff
*skb
)
1622 const struct iphdr
*iph
;
1627 if (skb
->pkt_type
!= PACKET_HOST
)
1630 /* Count it even if it's bad */
1631 TCP_INC_STATS_BH(TCP_MIB_INSEGS
);
1633 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1638 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1640 if (!pskb_may_pull(skb
, th
->doff
* 4))
1643 /* An explanation is required here, I think.
1644 * Packet length and doff are validated by header prediction,
1645 * provided case of th->doff==0 is eliminated.
1646 * So, we defer the checks. */
1647 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
1652 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1653 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1654 skb
->len
- th
->doff
* 4);
1655 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1656 TCP_SKB_CB(skb
)->when
= 0;
1657 TCP_SKB_CB(skb
)->flags
= iph
->tos
;
1658 TCP_SKB_CB(skb
)->sacked
= 0;
1660 sk
= __inet_lookup(&tcp_hashinfo
, iph
->saddr
, th
->source
,
1661 iph
->daddr
, th
->dest
, inet_iif(skb
));
1666 if (sk
->sk_state
== TCP_TIME_WAIT
)
1669 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1670 goto discard_and_relse
;
1673 if (sk_filter(sk
, skb
))
1674 goto discard_and_relse
;
1678 bh_lock_sock_nested(sk
);
1680 if (!sock_owned_by_user(sk
)) {
1681 #ifdef CONFIG_NET_DMA
1682 struct tcp_sock
*tp
= tcp_sk(sk
);
1683 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1684 tp
->ucopy
.dma_chan
= get_softnet_dma();
1685 if (tp
->ucopy
.dma_chan
)
1686 ret
= tcp_v4_do_rcv(sk
, skb
);
1690 if (!tcp_prequeue(sk
, skb
))
1691 ret
= tcp_v4_do_rcv(sk
, skb
);
1694 sk_add_backlog(sk
, skb
);
1702 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1705 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1707 TCP_INC_STATS_BH(TCP_MIB_INERRS
);
1709 tcp_v4_send_reset(NULL
, skb
);
1713 /* Discard frame. */
1722 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1723 inet_twsk_put(inet_twsk(sk
));
1727 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1728 TCP_INC_STATS_BH(TCP_MIB_INERRS
);
1729 inet_twsk_put(inet_twsk(sk
));
1732 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1734 struct sock
*sk2
= inet_lookup_listener(&tcp_hashinfo
,
1735 iph
->daddr
, th
->dest
,
1738 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1739 inet_twsk_put(inet_twsk(sk
));
1743 /* Fall through to ACK */
1746 tcp_v4_timewait_ack(sk
, skb
);
1750 case TCP_TW_SUCCESS
:;
1755 /* VJ's idea. Save last timestamp seen from this destination
1756 * and hold it at least for normal timewait interval to use for duplicate
1757 * segment detection in subsequent connections, before they enter synchronized
1761 int tcp_v4_remember_stamp(struct sock
*sk
)
1763 struct inet_sock
*inet
= inet_sk(sk
);
1764 struct tcp_sock
*tp
= tcp_sk(sk
);
1765 struct rtable
*rt
= (struct rtable
*)__sk_dst_get(sk
);
1766 struct inet_peer
*peer
= NULL
;
1769 if (!rt
|| rt
->rt_dst
!= inet
->daddr
) {
1770 peer
= inet_getpeer(inet
->daddr
, 1);
1774 rt_bind_peer(rt
, 1);
1779 if ((s32
)(peer
->tcp_ts
- tp
->rx_opt
.ts_recent
) <= 0 ||
1780 (peer
->tcp_ts_stamp
+ TCP_PAWS_MSL
< get_seconds() &&
1781 peer
->tcp_ts_stamp
<= tp
->rx_opt
.ts_recent_stamp
)) {
1782 peer
->tcp_ts_stamp
= tp
->rx_opt
.ts_recent_stamp
;
1783 peer
->tcp_ts
= tp
->rx_opt
.ts_recent
;
1793 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock
*tw
)
1795 struct inet_peer
*peer
= inet_getpeer(tw
->tw_daddr
, 1);
1798 const struct tcp_timewait_sock
*tcptw
= tcp_twsk((struct sock
*)tw
);
1800 if ((s32
)(peer
->tcp_ts
- tcptw
->tw_ts_recent
) <= 0 ||
1801 (peer
->tcp_ts_stamp
+ TCP_PAWS_MSL
< get_seconds() &&
1802 peer
->tcp_ts_stamp
<= tcptw
->tw_ts_recent_stamp
)) {
1803 peer
->tcp_ts_stamp
= tcptw
->tw_ts_recent_stamp
;
1804 peer
->tcp_ts
= tcptw
->tw_ts_recent
;
1813 struct inet_connection_sock_af_ops ipv4_specific
= {
1814 .queue_xmit
= ip_queue_xmit
,
1815 .send_check
= tcp_v4_send_check
,
1816 .rebuild_header
= inet_sk_rebuild_header
,
1817 .conn_request
= tcp_v4_conn_request
,
1818 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1819 .remember_stamp
= tcp_v4_remember_stamp
,
1820 .net_header_len
= sizeof(struct iphdr
),
1821 .setsockopt
= ip_setsockopt
,
1822 .getsockopt
= ip_getsockopt
,
1823 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1824 .sockaddr_len
= sizeof(struct sockaddr_in
),
1825 #ifdef CONFIG_COMPAT
1826 .compat_setsockopt
= compat_ip_setsockopt
,
1827 .compat_getsockopt
= compat_ip_getsockopt
,
1831 #ifdef CONFIG_TCP_MD5SIG
1832 static struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1833 .md5_lookup
= tcp_v4_md5_lookup
,
1834 .calc_md5_hash
= tcp_v4_calc_md5_hash
,
1835 .md5_add
= tcp_v4_md5_add_func
,
1836 .md5_parse
= tcp_v4_parse_md5_keys
,
1840 /* NOTE: A lot of things set to zero explicitly by call to
1841 * sk_alloc() so need not be done here.
1843 static int tcp_v4_init_sock(struct sock
*sk
)
1845 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1846 struct tcp_sock
*tp
= tcp_sk(sk
);
1848 skb_queue_head_init(&tp
->out_of_order_queue
);
1849 tcp_init_xmit_timers(sk
);
1850 tcp_prequeue_init(tp
);
1852 icsk
->icsk_rto
= TCP_TIMEOUT_INIT
;
1853 tp
->mdev
= TCP_TIMEOUT_INIT
;
1855 /* So many TCP implementations out there (incorrectly) count the
1856 * initial SYN frame in their delayed-ACK and congestion control
1857 * algorithms that we must have the following bandaid to talk
1858 * efficiently to them. -DaveM
1862 /* See draft-stevens-tcpca-spec-01 for discussion of the
1863 * initialization of these values.
1865 tp
->snd_ssthresh
= 0x7fffffff; /* Infinity */
1866 tp
->snd_cwnd_clamp
= ~0;
1867 tp
->mss_cache
= 536;
1869 tp
->reordering
= sysctl_tcp_reordering
;
1870 icsk
->icsk_ca_ops
= &tcp_init_congestion_ops
;
1872 sk
->sk_state
= TCP_CLOSE
;
1874 sk
->sk_write_space
= sk_stream_write_space
;
1875 sock_set_flag(sk
, SOCK_USE_WRITE_QUEUE
);
1877 icsk
->icsk_af_ops
= &ipv4_specific
;
1878 icsk
->icsk_sync_mss
= tcp_sync_mss
;
1879 #ifdef CONFIG_TCP_MD5SIG
1880 tp
->af_specific
= &tcp_sock_ipv4_specific
;
1883 sk
->sk_sndbuf
= sysctl_tcp_wmem
[1];
1884 sk
->sk_rcvbuf
= sysctl_tcp_rmem
[1];
1886 atomic_inc(&tcp_sockets_allocated
);
1891 int tcp_v4_destroy_sock(struct sock
*sk
)
1893 struct tcp_sock
*tp
= tcp_sk(sk
);
1895 tcp_clear_xmit_timers(sk
);
1897 tcp_cleanup_congestion_control(sk
);
1899 /* Cleanup up the write buffer. */
1900 tcp_write_queue_purge(sk
);
1902 /* Cleans up our, hopefully empty, out_of_order_queue. */
1903 __skb_queue_purge(&tp
->out_of_order_queue
);
1905 #ifdef CONFIG_TCP_MD5SIG
1906 /* Clean up the MD5 key list, if any */
1907 if (tp
->md5sig_info
) {
1908 tcp_v4_clear_md5_list(sk
);
1909 kfree(tp
->md5sig_info
);
1910 tp
->md5sig_info
= NULL
;
1914 #ifdef CONFIG_NET_DMA
1915 /* Cleans up our sk_async_wait_queue */
1916 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1919 /* Clean prequeue, it must be empty really */
1920 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1922 /* Clean up a referenced TCP bind bucket. */
1923 if (inet_csk(sk
)->icsk_bind_hash
)
1924 inet_put_port(&tcp_hashinfo
, sk
);
1927 * If sendmsg cached page exists, toss it.
1929 if (sk
->sk_sndmsg_page
) {
1930 __free_page(sk
->sk_sndmsg_page
);
1931 sk
->sk_sndmsg_page
= NULL
;
1934 atomic_dec(&tcp_sockets_allocated
);
1939 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1941 #ifdef CONFIG_PROC_FS
1942 /* Proc filesystem TCP sock list dumping. */
1944 static inline struct inet_timewait_sock
*tw_head(struct hlist_head
*head
)
1946 return hlist_empty(head
) ? NULL
:
1947 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
1950 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
1952 return tw
->tw_node
.next
?
1953 hlist_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
1956 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
1958 struct inet_connection_sock
*icsk
;
1959 struct hlist_node
*node
;
1960 struct sock
*sk
= cur
;
1961 struct tcp_iter_state
* st
= seq
->private;
1965 sk
= sk_head(&tcp_hashinfo
.listening_hash
[0]);
1971 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
1972 struct request_sock
*req
= cur
;
1974 icsk
= inet_csk(st
->syn_wait_sk
);
1978 if (req
->rsk_ops
->family
== st
->family
) {
1984 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
1987 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
1989 sk
= sk_next(st
->syn_wait_sk
);
1990 st
->state
= TCP_SEQ_STATE_LISTENING
;
1991 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1993 icsk
= inet_csk(sk
);
1994 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1995 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
1997 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2001 sk_for_each_from(sk
, node
) {
2002 if (sk
->sk_family
== st
->family
) {
2006 icsk
= inet_csk(sk
);
2007 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2008 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2010 st
->uid
= sock_i_uid(sk
);
2011 st
->syn_wait_sk
= sk
;
2012 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2016 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2018 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2019 sk
= sk_head(&tcp_hashinfo
.listening_hash
[st
->bucket
]);
2027 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2029 void *rc
= listening_get_next(seq
, NULL
);
2031 while (rc
&& *pos
) {
2032 rc
= listening_get_next(seq
, rc
);
2038 static void *established_get_first(struct seq_file
*seq
)
2040 struct tcp_iter_state
* st
= seq
->private;
2043 for (st
->bucket
= 0; st
->bucket
< tcp_hashinfo
.ehash_size
; ++st
->bucket
) {
2045 struct hlist_node
*node
;
2046 struct inet_timewait_sock
*tw
;
2048 read_lock_bh(&tcp_hashinfo
.ehash
[st
->bucket
].lock
);
2049 sk_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2050 if (sk
->sk_family
!= st
->family
) {
2056 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2057 inet_twsk_for_each(tw
, node
,
2058 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
2059 if (tw
->tw_family
!= st
->family
) {
2065 read_unlock_bh(&tcp_hashinfo
.ehash
[st
->bucket
].lock
);
2066 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2072 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2074 struct sock
*sk
= cur
;
2075 struct inet_timewait_sock
*tw
;
2076 struct hlist_node
*node
;
2077 struct tcp_iter_state
* st
= seq
->private;
2081 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2085 while (tw
&& tw
->tw_family
!= st
->family
) {
2092 read_unlock_bh(&tcp_hashinfo
.ehash
[st
->bucket
].lock
);
2093 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2095 if (++st
->bucket
< tcp_hashinfo
.ehash_size
) {
2096 read_lock_bh(&tcp_hashinfo
.ehash
[st
->bucket
].lock
);
2097 sk
= sk_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2105 sk_for_each_from(sk
, node
) {
2106 if (sk
->sk_family
== st
->family
)
2110 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2111 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2119 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2121 void *rc
= established_get_first(seq
);
2124 rc
= established_get_next(seq
, rc
);
2130 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2133 struct tcp_iter_state
* st
= seq
->private;
2135 inet_listen_lock(&tcp_hashinfo
);
2136 st
->state
= TCP_SEQ_STATE_LISTENING
;
2137 rc
= listening_get_idx(seq
, &pos
);
2140 inet_listen_unlock(&tcp_hashinfo
);
2141 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2142 rc
= established_get_idx(seq
, pos
);
2148 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2150 struct tcp_iter_state
* st
= seq
->private;
2151 st
->state
= TCP_SEQ_STATE_LISTENING
;
2153 return *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2156 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2159 struct tcp_iter_state
* st
;
2161 if (v
== SEQ_START_TOKEN
) {
2162 rc
= tcp_get_idx(seq
, 0);
2167 switch (st
->state
) {
2168 case TCP_SEQ_STATE_OPENREQ
:
2169 case TCP_SEQ_STATE_LISTENING
:
2170 rc
= listening_get_next(seq
, v
);
2172 inet_listen_unlock(&tcp_hashinfo
);
2173 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2174 rc
= established_get_first(seq
);
2177 case TCP_SEQ_STATE_ESTABLISHED
:
2178 case TCP_SEQ_STATE_TIME_WAIT
:
2179 rc
= established_get_next(seq
, v
);
2187 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2189 struct tcp_iter_state
* st
= seq
->private;
2191 switch (st
->state
) {
2192 case TCP_SEQ_STATE_OPENREQ
:
2194 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2195 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2197 case TCP_SEQ_STATE_LISTENING
:
2198 if (v
!= SEQ_START_TOKEN
)
2199 inet_listen_unlock(&tcp_hashinfo
);
2201 case TCP_SEQ_STATE_TIME_WAIT
:
2202 case TCP_SEQ_STATE_ESTABLISHED
:
2204 read_unlock_bh(&tcp_hashinfo
.ehash
[st
->bucket
].lock
);
2209 static int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2211 struct tcp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
2212 struct seq_file
*seq
;
2213 struct tcp_iter_state
*s
;
2216 if (unlikely(afinfo
== NULL
))
2219 s
= kzalloc(sizeof(*s
), GFP_KERNEL
);
2222 s
->family
= afinfo
->family
;
2223 s
->seq_ops
.start
= tcp_seq_start
;
2224 s
->seq_ops
.next
= tcp_seq_next
;
2225 s
->seq_ops
.show
= afinfo
->seq_show
;
2226 s
->seq_ops
.stop
= tcp_seq_stop
;
2228 rc
= seq_open(file
, &s
->seq_ops
);
2231 seq
= file
->private_data
;
2240 int tcp_proc_register(struct tcp_seq_afinfo
*afinfo
)
2243 struct proc_dir_entry
*p
;
2247 afinfo
->seq_fops
->owner
= afinfo
->owner
;
2248 afinfo
->seq_fops
->open
= tcp_seq_open
;
2249 afinfo
->seq_fops
->read
= seq_read
;
2250 afinfo
->seq_fops
->llseek
= seq_lseek
;
2251 afinfo
->seq_fops
->release
= seq_release_private
;
2253 p
= proc_net_fops_create(afinfo
->name
, S_IRUGO
, afinfo
->seq_fops
);
2261 void tcp_proc_unregister(struct tcp_seq_afinfo
*afinfo
)
2265 proc_net_remove(afinfo
->name
);
2266 memset(afinfo
->seq_fops
, 0, sizeof(*afinfo
->seq_fops
));
2269 static void get_openreq4(struct sock
*sk
, struct request_sock
*req
,
2270 char *tmpbuf
, int i
, int uid
)
2272 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2273 int ttd
= req
->expires
- jiffies
;
2275 sprintf(tmpbuf
, "%4d: %08X:%04X %08X:%04X"
2276 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p",
2279 ntohs(inet_sk(sk
)->sport
),
2281 ntohs(ireq
->rmt_port
),
2283 0, 0, /* could print option size, but that is af dependent. */
2284 1, /* timers active (only the expire timer) */
2285 jiffies_to_clock_t(ttd
),
2288 0, /* non standard timer */
2289 0, /* open_requests have no inode */
2290 atomic_read(&sk
->sk_refcnt
),
2294 static void get_tcp4_sock(struct sock
*sk
, char *tmpbuf
, int i
)
2297 unsigned long timer_expires
;
2298 struct tcp_sock
*tp
= tcp_sk(sk
);
2299 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2300 struct inet_sock
*inet
= inet_sk(sk
);
2301 __be32 dest
= inet
->daddr
;
2302 __be32 src
= inet
->rcv_saddr
;
2303 __u16 destp
= ntohs(inet
->dport
);
2304 __u16 srcp
= ntohs(inet
->sport
);
2306 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
) {
2308 timer_expires
= icsk
->icsk_timeout
;
2309 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2311 timer_expires
= icsk
->icsk_timeout
;
2312 } else if (timer_pending(&sk
->sk_timer
)) {
2314 timer_expires
= sk
->sk_timer
.expires
;
2317 timer_expires
= jiffies
;
2320 sprintf(tmpbuf
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2321 "%08X %5d %8d %lu %d %p %u %u %u %u %d",
2322 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2323 tp
->write_seq
- tp
->snd_una
,
2324 sk
->sk_state
== TCP_LISTEN
? sk
->sk_ack_backlog
:
2325 (tp
->rcv_nxt
- tp
->copied_seq
),
2327 jiffies_to_clock_t(timer_expires
- jiffies
),
2328 icsk
->icsk_retransmits
,
2330 icsk
->icsk_probes_out
,
2332 atomic_read(&sk
->sk_refcnt
), sk
,
2335 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2337 tp
->snd_ssthresh
>= 0xFFFF ? -1 : tp
->snd_ssthresh
);
2340 static void get_timewait4_sock(struct inet_timewait_sock
*tw
,
2341 char *tmpbuf
, int i
)
2345 int ttd
= tw
->tw_ttd
- jiffies
;
2350 dest
= tw
->tw_daddr
;
2351 src
= tw
->tw_rcv_saddr
;
2352 destp
= ntohs(tw
->tw_dport
);
2353 srcp
= ntohs(tw
->tw_sport
);
2355 sprintf(tmpbuf
, "%4d: %08X:%04X %08X:%04X"
2356 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p",
2357 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2358 3, jiffies_to_clock_t(ttd
), 0, 0, 0, 0,
2359 atomic_read(&tw
->tw_refcnt
), tw
);
2364 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2366 struct tcp_iter_state
* st
;
2367 char tmpbuf
[TMPSZ
+ 1];
2369 if (v
== SEQ_START_TOKEN
) {
2370 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2371 " sl local_address rem_address st tx_queue "
2372 "rx_queue tr tm->when retrnsmt uid timeout "
2378 switch (st
->state
) {
2379 case TCP_SEQ_STATE_LISTENING
:
2380 case TCP_SEQ_STATE_ESTABLISHED
:
2381 get_tcp4_sock(v
, tmpbuf
, st
->num
);
2383 case TCP_SEQ_STATE_OPENREQ
:
2384 get_openreq4(st
->syn_wait_sk
, v
, tmpbuf
, st
->num
, st
->uid
);
2386 case TCP_SEQ_STATE_TIME_WAIT
:
2387 get_timewait4_sock(v
, tmpbuf
, st
->num
);
2390 seq_printf(seq
, "%-*s\n", TMPSZ
- 1, tmpbuf
);
2395 static struct file_operations tcp4_seq_fops
;
2396 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2397 .owner
= THIS_MODULE
,
2400 .seq_show
= tcp4_seq_show
,
2401 .seq_fops
= &tcp4_seq_fops
,
2404 int __init
tcp4_proc_init(void)
2406 return tcp_proc_register(&tcp4_seq_afinfo
);
2409 void tcp4_proc_exit(void)
2411 tcp_proc_unregister(&tcp4_seq_afinfo
);
2413 #endif /* CONFIG_PROC_FS */
2415 struct proto tcp_prot
= {
2417 .owner
= THIS_MODULE
,
2419 .connect
= tcp_v4_connect
,
2420 .disconnect
= tcp_disconnect
,
2421 .accept
= inet_csk_accept
,
2423 .init
= tcp_v4_init_sock
,
2424 .destroy
= tcp_v4_destroy_sock
,
2425 .shutdown
= tcp_shutdown
,
2426 .setsockopt
= tcp_setsockopt
,
2427 .getsockopt
= tcp_getsockopt
,
2428 .recvmsg
= tcp_recvmsg
,
2429 .backlog_rcv
= tcp_v4_do_rcv
,
2430 .hash
= tcp_v4_hash
,
2431 .unhash
= tcp_unhash
,
2432 .get_port
= tcp_v4_get_port
,
2433 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2434 .sockets_allocated
= &tcp_sockets_allocated
,
2435 .orphan_count
= &tcp_orphan_count
,
2436 .memory_allocated
= &tcp_memory_allocated
,
2437 .memory_pressure
= &tcp_memory_pressure
,
2438 .sysctl_mem
= sysctl_tcp_mem
,
2439 .sysctl_wmem
= sysctl_tcp_wmem
,
2440 .sysctl_rmem
= sysctl_tcp_rmem
,
2441 .max_header
= MAX_TCP_HEADER
,
2442 .obj_size
= sizeof(struct tcp_sock
),
2443 .twsk_prot
= &tcp_timewait_sock_ops
,
2444 .rsk_prot
= &tcp_request_sock_ops
,
2445 #ifdef CONFIG_COMPAT
2446 .compat_setsockopt
= compat_tcp_setsockopt
,
2447 .compat_getsockopt
= compat_tcp_getsockopt
,
2451 void __init
tcp_v4_init(struct net_proto_family
*ops
)
2453 if (inet_csk_ctl_sock_create(&tcp_socket
, PF_INET
, SOCK_RAW
,
2455 panic("Failed to create the TCP control socket.\n");
2458 EXPORT_SYMBOL(ipv4_specific
);
2459 EXPORT_SYMBOL(tcp_hashinfo
);
2460 EXPORT_SYMBOL(tcp_prot
);
2461 EXPORT_SYMBOL(tcp_unhash
);
2462 EXPORT_SYMBOL(tcp_v4_conn_request
);
2463 EXPORT_SYMBOL(tcp_v4_connect
);
2464 EXPORT_SYMBOL(tcp_v4_do_rcv
);
2465 EXPORT_SYMBOL(tcp_v4_remember_stamp
);
2466 EXPORT_SYMBOL(tcp_v4_send_check
);
2467 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
2469 #ifdef CONFIG_PROC_FS
2470 EXPORT_SYMBOL(tcp_proc_register
);
2471 EXPORT_SYMBOL(tcp_proc_unregister
);
2473 EXPORT_SYMBOL(sysctl_local_port_range
);
2474 EXPORT_SYMBOL(sysctl_tcp_low_latency
);