2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Implementation of the Transmission Control Protocol(TCP).
8 * IPv4 specific functions
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
16 * See tcp.c for author information
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
54 #include <linux/bottom_half.h>
55 #include <linux/types.h>
56 #include <linux/fcntl.h>
57 #include <linux/module.h>
58 #include <linux/random.h>
59 #include <linux/cache.h>
60 #include <linux/jhash.h>
61 #include <linux/init.h>
62 #include <linux/times.h>
63 #include <linux/slab.h>
65 #include <net/net_namespace.h>
67 #include <net/inet_hashtables.h>
69 #include <net/transp_v6.h>
71 #include <net/inet_common.h>
72 #include <net/timewait_sock.h>
74 #include <net/netdma.h>
75 #include <net/secure_seq.h>
77 #include <linux/inet.h>
78 #include <linux/ipv6.h>
79 #include <linux/stddef.h>
80 #include <linux/proc_fs.h>
81 #include <linux/seq_file.h>
83 #include <linux/crypto.h>
84 #include <linux/scatterlist.h>
86 int sysctl_tcp_tw_reuse __read_mostly
;
87 int sysctl_tcp_low_latency __read_mostly
;
88 EXPORT_SYMBOL(sysctl_tcp_low_latency
);
91 #ifdef CONFIG_TCP_MD5SIG
92 static struct tcp_md5sig_key
*tcp_v4_md5_do_lookup(struct sock
*sk
,
94 static int tcp_v4_md5_hash_hdr(char *md5_hash
, struct tcp_md5sig_key
*key
,
95 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
);
98 struct tcp_md5sig_key
*tcp_v4_md5_do_lookup(struct sock
*sk
, __be32 addr
)
104 struct inet_hashinfo tcp_hashinfo
;
105 EXPORT_SYMBOL(tcp_hashinfo
);
107 static inline __u32
tcp_v4_init_sequence(const struct sk_buff
*skb
)
109 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
112 tcp_hdr(skb
)->source
);
115 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
)
117 const struct tcp_timewait_sock
*tcptw
= tcp_twsk(sktw
);
118 struct tcp_sock
*tp
= tcp_sk(sk
);
120 /* With PAWS, it is safe from the viewpoint
121 of data integrity. Even without PAWS it is safe provided sequence
122 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
124 Actually, the idea is close to VJ's one, only timestamp cache is
125 held not per host, but per port pair and TW bucket is used as state
128 If TW bucket has been already destroyed we fall back to VJ's scheme
129 and use initial timestamp retrieved from peer table.
131 if (tcptw
->tw_ts_recent_stamp
&&
132 (twp
== NULL
|| (sysctl_tcp_tw_reuse
&&
133 get_seconds() - tcptw
->tw_ts_recent_stamp
> 1))) {
134 tp
->write_seq
= tcptw
->tw_snd_nxt
+ 65535 + 2;
135 if (tp
->write_seq
== 0)
137 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
138 tp
->rx_opt
.ts_recent_stamp
= tcptw
->tw_ts_recent_stamp
;
145 EXPORT_SYMBOL_GPL(tcp_twsk_unique
);
147 /* This will initiate an outgoing connection. */
148 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
150 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
151 struct inet_sock
*inet
= inet_sk(sk
);
152 struct tcp_sock
*tp
= tcp_sk(sk
);
153 __be16 orig_sport
, orig_dport
;
154 __be32 daddr
, nexthop
;
158 struct ip_options_rcu
*inet_opt
;
160 if (addr_len
< sizeof(struct sockaddr_in
))
163 if (usin
->sin_family
!= AF_INET
)
164 return -EAFNOSUPPORT
;
166 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
167 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
168 sock_owned_by_user(sk
));
169 if (inet_opt
&& inet_opt
->opt
.srr
) {
172 nexthop
= inet_opt
->opt
.faddr
;
175 orig_sport
= inet
->inet_sport
;
176 orig_dport
= usin
->sin_port
;
177 fl4
= &inet
->cork
.fl
.u
.ip4
;
178 rt
= ip_route_connect(fl4
, nexthop
, inet
->inet_saddr
,
179 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
181 orig_sport
, orig_dport
, sk
, true);
184 if (err
== -ENETUNREACH
)
185 IP_INC_STATS_BH(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
189 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
194 if (!inet_opt
|| !inet_opt
->opt
.srr
)
197 if (!inet
->inet_saddr
)
198 inet
->inet_saddr
= fl4
->saddr
;
199 inet
->inet_rcv_saddr
= inet
->inet_saddr
;
201 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->inet_daddr
!= daddr
) {
202 /* Reset inherited state */
203 tp
->rx_opt
.ts_recent
= 0;
204 tp
->rx_opt
.ts_recent_stamp
= 0;
208 if (tcp_death_row
.sysctl_tw_recycle
&&
209 !tp
->rx_opt
.ts_recent_stamp
&& fl4
->daddr
== daddr
) {
210 struct inet_peer
*peer
= rt_get_peer(rt
, fl4
->daddr
);
212 * VJ's idea. We save last timestamp seen from
213 * the destination in peer table, when entering state
214 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
215 * when trying new connection.
218 inet_peer_refcheck(peer
);
219 if ((u32
)get_seconds() - peer
->tcp_ts_stamp
<= TCP_PAWS_MSL
) {
220 tp
->rx_opt
.ts_recent_stamp
= peer
->tcp_ts_stamp
;
221 tp
->rx_opt
.ts_recent
= peer
->tcp_ts
;
226 inet
->inet_dport
= usin
->sin_port
;
227 inet
->inet_daddr
= daddr
;
229 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
231 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
233 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
235 /* Socket identity is still unknown (sport may be zero).
236 * However we set state to SYN-SENT and not releasing socket
237 * lock select source port, enter ourselves into the hash tables and
238 * complete initialization after this.
240 tcp_set_state(sk
, TCP_SYN_SENT
);
241 err
= inet_hash_connect(&tcp_death_row
, sk
);
245 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
246 inet
->inet_sport
, inet
->inet_dport
, sk
);
252 /* OK, now commit destination to socket. */
253 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
254 sk_setup_caps(sk
, &rt
->dst
);
257 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
262 inet
->inet_id
= tp
->write_seq
^ jiffies
;
264 err
= tcp_connect(sk
);
273 * This unhashes the socket and releases the local port,
276 tcp_set_state(sk
, TCP_CLOSE
);
278 sk
->sk_route_caps
= 0;
279 inet
->inet_dport
= 0;
282 EXPORT_SYMBOL(tcp_v4_connect
);
285 * This routine does path mtu discovery as defined in RFC1191.
287 static void do_pmtu_discovery(struct sock
*sk
, const struct iphdr
*iph
, u32 mtu
)
289 struct dst_entry
*dst
;
290 struct inet_sock
*inet
= inet_sk(sk
);
292 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
293 * send out by Linux are always <576bytes so they should go through
296 if (sk
->sk_state
== TCP_LISTEN
)
299 /* We don't check in the destentry if pmtu discovery is forbidden
300 * on this route. We just assume that no packet_to_big packets
301 * are send back when pmtu discovery is not active.
302 * There is a small race when the user changes this flag in the
303 * route, but I think that's acceptable.
305 if ((dst
= __sk_dst_check(sk
, 0)) == NULL
)
308 dst
->ops
->update_pmtu(dst
, mtu
);
310 /* Something is about to be wrong... Remember soft error
311 * for the case, if this connection will not able to recover.
313 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
314 sk
->sk_err_soft
= EMSGSIZE
;
318 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
319 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
320 tcp_sync_mss(sk
, mtu
);
322 /* Resend the TCP packet because it's
323 * clear that the old packet has been
324 * dropped. This is the new "fast" path mtu
327 tcp_simple_retransmit(sk
);
328 } /* else let the usual retransmit timer handle it */
332 * This routine is called by the ICMP module when it gets some
333 * sort of error condition. If err < 0 then the socket should
334 * be closed and the error returned to the user. If err > 0
335 * it's just the icmp type << 8 | icmp code. After adjustment
336 * header points to the first 8 bytes of the tcp header. We need
337 * to find the appropriate port.
339 * The locking strategy used here is very "optimistic". When
340 * someone else accesses the socket the ICMP is just dropped
341 * and for some paths there is no check at all.
342 * A more general error queue to queue errors for later handling
343 * is probably better.
347 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
349 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
350 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
351 struct inet_connection_sock
*icsk
;
353 struct inet_sock
*inet
;
354 const int type
= icmp_hdr(icmp_skb
)->type
;
355 const int code
= icmp_hdr(icmp_skb
)->code
;
361 struct net
*net
= dev_net(icmp_skb
->dev
);
363 if (icmp_skb
->len
< (iph
->ihl
<< 2) + 8) {
364 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
368 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
369 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
371 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
374 if (sk
->sk_state
== TCP_TIME_WAIT
) {
375 inet_twsk_put(inet_twsk(sk
));
380 /* If too many ICMPs get dropped on busy
381 * servers this needs to be solved differently.
383 if (sock_owned_by_user(sk
))
384 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
386 if (sk
->sk_state
== TCP_CLOSE
)
389 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
390 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
396 seq
= ntohl(th
->seq
);
397 if (sk
->sk_state
!= TCP_LISTEN
&&
398 !between(seq
, tp
->snd_una
, tp
->snd_nxt
)) {
399 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
404 case ICMP_SOURCE_QUENCH
:
405 /* Just silently ignore these. */
407 case ICMP_PARAMETERPROB
:
410 case ICMP_DEST_UNREACH
:
411 if (code
> NR_ICMP_UNREACH
)
414 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
415 if (!sock_owned_by_user(sk
))
416 do_pmtu_discovery(sk
, iph
, info
);
420 err
= icmp_err_convert
[code
].errno
;
421 /* check if icmp_skb allows revert of backoff
422 * (see draft-zimmermann-tcp-lcd) */
423 if (code
!= ICMP_NET_UNREACH
&& code
!= ICMP_HOST_UNREACH
)
425 if (seq
!= tp
->snd_una
|| !icsk
->icsk_retransmits
||
429 if (sock_owned_by_user(sk
))
432 icsk
->icsk_backoff
--;
433 inet_csk(sk
)->icsk_rto
= (tp
->srtt
? __tcp_set_rto(tp
) :
434 TCP_TIMEOUT_INIT
) << icsk
->icsk_backoff
;
437 skb
= tcp_write_queue_head(sk
);
440 remaining
= icsk
->icsk_rto
- min(icsk
->icsk_rto
,
441 tcp_time_stamp
- TCP_SKB_CB(skb
)->when
);
444 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
445 remaining
, TCP_RTO_MAX
);
447 /* RTO revert clocked out retransmission.
448 * Will retransmit now */
449 tcp_retransmit_timer(sk
);
453 case ICMP_TIME_EXCEEDED
:
460 switch (sk
->sk_state
) {
461 struct request_sock
*req
, **prev
;
463 if (sock_owned_by_user(sk
))
466 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
467 iph
->daddr
, iph
->saddr
);
471 /* ICMPs are not backlogged, hence we cannot get
472 an established socket here.
476 if (seq
!= tcp_rsk(req
)->snt_isn
) {
477 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
482 * Still in SYN_RECV, just remove it silently.
483 * There is no good way to pass the error to the newly
484 * created socket, and POSIX does not want network
485 * errors returned from accept().
487 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
491 case TCP_SYN_RECV
: /* Cannot happen.
492 It can f.e. if SYNs crossed.
494 if (!sock_owned_by_user(sk
)) {
497 sk
->sk_error_report(sk
);
501 sk
->sk_err_soft
= err
;
506 /* If we've already connected we will keep trying
507 * until we time out, or the user gives up.
509 * rfc1122 4.2.3.9 allows to consider as hard errors
510 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
511 * but it is obsoleted by pmtu discovery).
513 * Note, that in modern internet, where routing is unreliable
514 * and in each dark corner broken firewalls sit, sending random
515 * errors ordered by their masters even this two messages finally lose
516 * their original sense (even Linux sends invalid PORT_UNREACHs)
518 * Now we are in compliance with RFCs.
523 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
525 sk
->sk_error_report(sk
);
526 } else { /* Only an error on timeout */
527 sk
->sk_err_soft
= err
;
535 static void __tcp_v4_send_check(struct sk_buff
*skb
,
536 __be32 saddr
, __be32 daddr
)
538 struct tcphdr
*th
= tcp_hdr(skb
);
540 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
541 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
542 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
543 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
545 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
552 /* This routine computes an IPv4 TCP checksum. */
553 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
555 const struct inet_sock
*inet
= inet_sk(sk
);
557 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
559 EXPORT_SYMBOL(tcp_v4_send_check
);
561 int tcp_v4_gso_send_check(struct sk_buff
*skb
)
563 const struct iphdr
*iph
;
566 if (!pskb_may_pull(skb
, sizeof(*th
)))
573 skb
->ip_summed
= CHECKSUM_PARTIAL
;
574 __tcp_v4_send_check(skb
, iph
->saddr
, iph
->daddr
);
579 * This routine will send an RST to the other tcp.
581 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
583 * Answer: if a packet caused RST, it is not for a socket
584 * existing in our system, if it is matched to a socket,
585 * it is just duplicate segment or bug in other side's TCP.
586 * So that we build reply only basing on parameters
587 * arrived with segment.
588 * Exception: precedence violation. We do not implement it in any case.
591 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
593 const struct tcphdr
*th
= tcp_hdr(skb
);
596 #ifdef CONFIG_TCP_MD5SIG
597 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
600 struct ip_reply_arg arg
;
601 #ifdef CONFIG_TCP_MD5SIG
602 struct tcp_md5sig_key
*key
;
606 /* Never send a reset in response to a reset. */
610 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
613 /* Swap the send and the receive. */
614 memset(&rep
, 0, sizeof(rep
));
615 rep
.th
.dest
= th
->source
;
616 rep
.th
.source
= th
->dest
;
617 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
621 rep
.th
.seq
= th
->ack_seq
;
624 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
625 skb
->len
- (th
->doff
<< 2));
628 memset(&arg
, 0, sizeof(arg
));
629 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
630 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
632 #ifdef CONFIG_TCP_MD5SIG
633 key
= sk
? tcp_v4_md5_do_lookup(sk
, ip_hdr(skb
)->daddr
) : NULL
;
635 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
637 (TCPOPT_MD5SIG
<< 8) |
639 /* Update length and the length the header thinks exists */
640 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
641 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
643 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
644 key
, ip_hdr(skb
)->saddr
,
645 ip_hdr(skb
)->daddr
, &rep
.th
);
648 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
649 ip_hdr(skb
)->saddr
, /* XXX */
650 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
651 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
652 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
654 net
= dev_net(skb_dst(skb
)->dev
);
655 arg
.tos
= ip_hdr(skb
)->tos
;
656 ip_send_reply(net
->ipv4
.tcp_sock
, skb
, ip_hdr(skb
)->saddr
,
657 &arg
, arg
.iov
[0].iov_len
);
659 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
660 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
663 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
664 outside socket context is ugly, certainly. What can I do?
667 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
668 u32 win
, u32 ts
, int oif
,
669 struct tcp_md5sig_key
*key
,
670 int reply_flags
, u8 tos
)
672 const struct tcphdr
*th
= tcp_hdr(skb
);
675 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
676 #ifdef CONFIG_TCP_MD5SIG
677 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
681 struct ip_reply_arg arg
;
682 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
684 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
685 memset(&arg
, 0, sizeof(arg
));
687 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
688 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
690 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
691 (TCPOPT_TIMESTAMP
<< 8) |
693 rep
.opt
[1] = htonl(tcp_time_stamp
);
694 rep
.opt
[2] = htonl(ts
);
695 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
698 /* Swap the send and the receive. */
699 rep
.th
.dest
= th
->source
;
700 rep
.th
.source
= th
->dest
;
701 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
702 rep
.th
.seq
= htonl(seq
);
703 rep
.th
.ack_seq
= htonl(ack
);
705 rep
.th
.window
= htons(win
);
707 #ifdef CONFIG_TCP_MD5SIG
709 int offset
= (ts
) ? 3 : 0;
711 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
713 (TCPOPT_MD5SIG
<< 8) |
715 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
716 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
718 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
719 key
, ip_hdr(skb
)->saddr
,
720 ip_hdr(skb
)->daddr
, &rep
.th
);
723 arg
.flags
= reply_flags
;
724 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
725 ip_hdr(skb
)->saddr
, /* XXX */
726 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
727 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
729 arg
.bound_dev_if
= oif
;
731 ip_send_reply(net
->ipv4
.tcp_sock
, skb
, ip_hdr(skb
)->saddr
,
732 &arg
, arg
.iov
[0].iov_len
);
734 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
737 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
739 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
740 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
742 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
743 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
746 tcp_twsk_md5_key(tcptw
),
747 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
754 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
755 struct request_sock
*req
)
757 tcp_v4_send_ack(skb
, tcp_rsk(req
)->snt_isn
+ 1,
758 tcp_rsk(req
)->rcv_isn
+ 1, req
->rcv_wnd
,
761 tcp_v4_md5_do_lookup(sk
, ip_hdr(skb
)->daddr
),
762 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
767 * Send a SYN-ACK after having received a SYN.
768 * This still operates on a request_sock only, not on a big
771 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
772 struct request_sock
*req
,
773 struct request_values
*rvp
)
775 const struct inet_request_sock
*ireq
= inet_rsk(req
);
778 struct sk_buff
* skb
;
780 /* First, grab a route. */
781 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
784 skb
= tcp_make_synack(sk
, dst
, req
, rvp
);
787 __tcp_v4_send_check(skb
, ireq
->loc_addr
, ireq
->rmt_addr
);
789 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
792 err
= net_xmit_eval(err
);
799 static int tcp_v4_rtx_synack(struct sock
*sk
, struct request_sock
*req
,
800 struct request_values
*rvp
)
802 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
803 return tcp_v4_send_synack(sk
, NULL
, req
, rvp
);
807 * IPv4 request_sock destructor.
809 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
811 kfree(inet_rsk(req
)->opt
);
815 * Return 1 if a syncookie should be sent
817 int tcp_syn_flood_action(struct sock
*sk
,
818 const struct sk_buff
*skb
,
821 const char *msg
= "Dropping request";
823 struct listen_sock
*lopt
;
827 #ifdef CONFIG_SYN_COOKIES
828 if (sysctl_tcp_syncookies
) {
829 msg
= "Sending cookies";
831 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
834 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
836 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
837 if (!lopt
->synflood_warned
) {
838 lopt
->synflood_warned
= 1;
839 pr_info("%s: Possible SYN flooding on port %d. %s. "
840 " Check SNMP counters.\n",
841 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
845 EXPORT_SYMBOL(tcp_syn_flood_action
);
848 * Save and compile IPv4 options into the request_sock if needed.
850 static struct ip_options_rcu
*tcp_v4_save_options(struct sock
*sk
,
853 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
854 struct ip_options_rcu
*dopt
= NULL
;
856 if (opt
&& opt
->optlen
) {
857 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
859 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
861 if (ip_options_echo(&dopt
->opt
, skb
)) {
870 #ifdef CONFIG_TCP_MD5SIG
872 * RFC2385 MD5 checksumming requires a mapping of
873 * IP address->MD5 Key.
874 * We need to maintain these in the sk structure.
877 /* Find the Key structure for an address. */
878 static struct tcp_md5sig_key
*
879 tcp_v4_md5_do_lookup(struct sock
*sk
, __be32 addr
)
881 struct tcp_sock
*tp
= tcp_sk(sk
);
884 if (!tp
->md5sig_info
|| !tp
->md5sig_info
->entries4
)
886 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++) {
887 if (tp
->md5sig_info
->keys4
[i
].addr
== addr
)
888 return &tp
->md5sig_info
->keys4
[i
].base
;
893 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
894 struct sock
*addr_sk
)
896 return tcp_v4_md5_do_lookup(sk
, inet_sk(addr_sk
)->inet_daddr
);
898 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
900 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
901 struct request_sock
*req
)
903 return tcp_v4_md5_do_lookup(sk
, inet_rsk(req
)->rmt_addr
);
906 /* This can be called on a newly created socket, from other files */
907 int tcp_v4_md5_do_add(struct sock
*sk
, __be32 addr
,
908 u8
*newkey
, u8 newkeylen
)
910 /* Add Key to the list */
911 struct tcp_md5sig_key
*key
;
912 struct tcp_sock
*tp
= tcp_sk(sk
);
913 struct tcp4_md5sig_key
*keys
;
915 key
= tcp_v4_md5_do_lookup(sk
, addr
);
917 /* Pre-existing entry - just update that one. */
920 key
->keylen
= newkeylen
;
922 struct tcp_md5sig_info
*md5sig
;
924 if (!tp
->md5sig_info
) {
925 tp
->md5sig_info
= kzalloc(sizeof(*tp
->md5sig_info
),
927 if (!tp
->md5sig_info
) {
931 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
934 md5sig
= tp
->md5sig_info
;
935 if (md5sig
->entries4
== 0 &&
936 tcp_alloc_md5sig_pool(sk
) == NULL
) {
941 if (md5sig
->alloced4
== md5sig
->entries4
) {
942 keys
= kmalloc((sizeof(*keys
) *
943 (md5sig
->entries4
+ 1)), GFP_ATOMIC
);
946 if (md5sig
->entries4
== 0)
947 tcp_free_md5sig_pool();
951 if (md5sig
->entries4
)
952 memcpy(keys
, md5sig
->keys4
,
953 sizeof(*keys
) * md5sig
->entries4
);
955 /* Free old key list, and reference new one */
956 kfree(md5sig
->keys4
);
957 md5sig
->keys4
= keys
;
961 md5sig
->keys4
[md5sig
->entries4
- 1].addr
= addr
;
962 md5sig
->keys4
[md5sig
->entries4
- 1].base
.key
= newkey
;
963 md5sig
->keys4
[md5sig
->entries4
- 1].base
.keylen
= newkeylen
;
967 EXPORT_SYMBOL(tcp_v4_md5_do_add
);
969 static int tcp_v4_md5_add_func(struct sock
*sk
, struct sock
*addr_sk
,
970 u8
*newkey
, u8 newkeylen
)
972 return tcp_v4_md5_do_add(sk
, inet_sk(addr_sk
)->inet_daddr
,
976 int tcp_v4_md5_do_del(struct sock
*sk
, __be32 addr
)
978 struct tcp_sock
*tp
= tcp_sk(sk
);
981 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++) {
982 if (tp
->md5sig_info
->keys4
[i
].addr
== addr
) {
984 kfree(tp
->md5sig_info
->keys4
[i
].base
.key
);
985 tp
->md5sig_info
->entries4
--;
987 if (tp
->md5sig_info
->entries4
== 0) {
988 kfree(tp
->md5sig_info
->keys4
);
989 tp
->md5sig_info
->keys4
= NULL
;
990 tp
->md5sig_info
->alloced4
= 0;
991 tcp_free_md5sig_pool();
992 } else if (tp
->md5sig_info
->entries4
!= i
) {
993 /* Need to do some manipulation */
994 memmove(&tp
->md5sig_info
->keys4
[i
],
995 &tp
->md5sig_info
->keys4
[i
+1],
996 (tp
->md5sig_info
->entries4
- i
) *
997 sizeof(struct tcp4_md5sig_key
));
1004 EXPORT_SYMBOL(tcp_v4_md5_do_del
);
1006 static void tcp_v4_clear_md5_list(struct sock
*sk
)
1008 struct tcp_sock
*tp
= tcp_sk(sk
);
1010 /* Free each key, then the set of key keys,
1011 * the crypto element, and then decrement our
1012 * hold on the last resort crypto.
1014 if (tp
->md5sig_info
->entries4
) {
1016 for (i
= 0; i
< tp
->md5sig_info
->entries4
; i
++)
1017 kfree(tp
->md5sig_info
->keys4
[i
].base
.key
);
1018 tp
->md5sig_info
->entries4
= 0;
1019 tcp_free_md5sig_pool();
1021 if (tp
->md5sig_info
->keys4
) {
1022 kfree(tp
->md5sig_info
->keys4
);
1023 tp
->md5sig_info
->keys4
= NULL
;
1024 tp
->md5sig_info
->alloced4
= 0;
1028 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1031 struct tcp_md5sig cmd
;
1032 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1035 if (optlen
< sizeof(cmd
))
1038 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1041 if (sin
->sin_family
!= AF_INET
)
1044 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
) {
1045 if (!tcp_sk(sk
)->md5sig_info
)
1047 return tcp_v4_md5_do_del(sk
, sin
->sin_addr
.s_addr
);
1050 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1053 if (!tcp_sk(sk
)->md5sig_info
) {
1054 struct tcp_sock
*tp
= tcp_sk(sk
);
1055 struct tcp_md5sig_info
*p
;
1057 p
= kzalloc(sizeof(*p
), sk
->sk_allocation
);
1061 tp
->md5sig_info
= p
;
1062 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
1065 newkey
= kmemdup(cmd
.tcpm_key
, cmd
.tcpm_keylen
, sk
->sk_allocation
);
1068 return tcp_v4_md5_do_add(sk
, sin
->sin_addr
.s_addr
,
1069 newkey
, cmd
.tcpm_keylen
);
1072 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1073 __be32 daddr
, __be32 saddr
, int nbytes
)
1075 struct tcp4_pseudohdr
*bp
;
1076 struct scatterlist sg
;
1078 bp
= &hp
->md5_blk
.ip4
;
1081 * 1. the TCP pseudo-header (in the order: source IP address,
1082 * destination IP address, zero-padded protocol number, and
1088 bp
->protocol
= IPPROTO_TCP
;
1089 bp
->len
= cpu_to_be16(nbytes
);
1091 sg_init_one(&sg
, bp
, sizeof(*bp
));
1092 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1095 static int tcp_v4_md5_hash_hdr(char *md5_hash
, struct tcp_md5sig_key
*key
,
1096 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1098 struct tcp_md5sig_pool
*hp
;
1099 struct hash_desc
*desc
;
1101 hp
= tcp_get_md5sig_pool();
1103 goto clear_hash_noput
;
1104 desc
= &hp
->md5_desc
;
1106 if (crypto_hash_init(desc
))
1108 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1110 if (tcp_md5_hash_header(hp
, th
))
1112 if (tcp_md5_hash_key(hp
, key
))
1114 if (crypto_hash_final(desc
, md5_hash
))
1117 tcp_put_md5sig_pool();
1121 tcp_put_md5sig_pool();
1123 memset(md5_hash
, 0, 16);
1127 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1128 const struct sock
*sk
, const struct request_sock
*req
,
1129 const struct sk_buff
*skb
)
1131 struct tcp_md5sig_pool
*hp
;
1132 struct hash_desc
*desc
;
1133 const struct tcphdr
*th
= tcp_hdr(skb
);
1134 __be32 saddr
, daddr
;
1137 saddr
= inet_sk(sk
)->inet_saddr
;
1138 daddr
= inet_sk(sk
)->inet_daddr
;
1140 saddr
= inet_rsk(req
)->loc_addr
;
1141 daddr
= inet_rsk(req
)->rmt_addr
;
1143 const struct iphdr
*iph
= ip_hdr(skb
);
1148 hp
= tcp_get_md5sig_pool();
1150 goto clear_hash_noput
;
1151 desc
= &hp
->md5_desc
;
1153 if (crypto_hash_init(desc
))
1156 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1158 if (tcp_md5_hash_header(hp
, th
))
1160 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1162 if (tcp_md5_hash_key(hp
, key
))
1164 if (crypto_hash_final(desc
, md5_hash
))
1167 tcp_put_md5sig_pool();
1171 tcp_put_md5sig_pool();
1173 memset(md5_hash
, 0, 16);
1176 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1178 static int tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1181 * This gets called for each TCP segment that arrives
1182 * so we want to be efficient.
1183 * We have 3 drop cases:
1184 * o No MD5 hash and one expected.
1185 * o MD5 hash and we're not expecting one.
1186 * o MD5 hash and its wrong.
1188 const __u8
*hash_location
= NULL
;
1189 struct tcp_md5sig_key
*hash_expected
;
1190 const struct iphdr
*iph
= ip_hdr(skb
);
1191 const struct tcphdr
*th
= tcp_hdr(skb
);
1193 unsigned char newhash
[16];
1195 hash_expected
= tcp_v4_md5_do_lookup(sk
, iph
->saddr
);
1196 hash_location
= tcp_parse_md5sig_option(th
);
1198 /* We've parsed the options - do we have a hash? */
1199 if (!hash_expected
&& !hash_location
)
1202 if (hash_expected
&& !hash_location
) {
1203 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1207 if (!hash_expected
&& hash_location
) {
1208 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1212 /* Okay, so this is hash_expected and hash_location -
1213 * so we need to calculate the checksum.
1215 genhash
= tcp_v4_md5_hash_skb(newhash
,
1219 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1220 if (net_ratelimit()) {
1221 printk(KERN_INFO
"MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1222 &iph
->saddr
, ntohs(th
->source
),
1223 &iph
->daddr
, ntohs(th
->dest
),
1224 genhash
? " tcp_v4_calc_md5_hash failed" : "");
1233 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1235 .obj_size
= sizeof(struct tcp_request_sock
),
1236 .rtx_syn_ack
= tcp_v4_rtx_synack
,
1237 .send_ack
= tcp_v4_reqsk_send_ack
,
1238 .destructor
= tcp_v4_reqsk_destructor
,
1239 .send_reset
= tcp_v4_send_reset
,
1240 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1243 #ifdef CONFIG_TCP_MD5SIG
1244 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1245 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1246 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1250 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1252 struct tcp_extend_values tmp_ext
;
1253 struct tcp_options_received tmp_opt
;
1254 const u8
*hash_location
;
1255 struct request_sock
*req
;
1256 struct inet_request_sock
*ireq
;
1257 struct tcp_sock
*tp
= tcp_sk(sk
);
1258 struct dst_entry
*dst
= NULL
;
1259 __be32 saddr
= ip_hdr(skb
)->saddr
;
1260 __be32 daddr
= ip_hdr(skb
)->daddr
;
1261 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1262 int want_cookie
= 0;
1264 /* Never answer to SYNs send to broadcast or multicast */
1265 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1268 /* TW buckets are converted to open requests without
1269 * limitations, they conserve resources and peer is
1270 * evidently real one.
1272 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
1273 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCP");
1278 /* Accept backlog is full. If we have already queued enough
1279 * of warm entries in syn queue, drop request. It is better than
1280 * clogging syn queue with openreqs with exponentially increasing
1283 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1)
1286 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1290 #ifdef CONFIG_TCP_MD5SIG
1291 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1294 tcp_clear_options(&tmp_opt
);
1295 tmp_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
1296 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
1297 tcp_parse_options(skb
, &tmp_opt
, &hash_location
, 0);
1299 if (tmp_opt
.cookie_plus
> 0 &&
1300 tmp_opt
.saw_tstamp
&&
1301 !tp
->rx_opt
.cookie_out_never
&&
1302 (sysctl_tcp_cookie_size
> 0 ||
1303 (tp
->cookie_values
!= NULL
&&
1304 tp
->cookie_values
->cookie_desired
> 0))) {
1306 u32
*mess
= &tmp_ext
.cookie_bakery
[COOKIE_DIGEST_WORDS
];
1307 int l
= tmp_opt
.cookie_plus
- TCPOLEN_COOKIE_BASE
;
1309 if (tcp_cookie_generator(&tmp_ext
.cookie_bakery
[0]) != 0)
1310 goto drop_and_release
;
1312 /* Secret recipe starts with IP addresses */
1313 *mess
++ ^= (__force u32
)daddr
;
1314 *mess
++ ^= (__force u32
)saddr
;
1316 /* plus variable length Initiator Cookie */
1319 *c
++ ^= *hash_location
++;
1321 want_cookie
= 0; /* not our kind of cookie */
1322 tmp_ext
.cookie_out_never
= 0; /* false */
1323 tmp_ext
.cookie_plus
= tmp_opt
.cookie_plus
;
1324 } else if (!tp
->rx_opt
.cookie_in_always
) {
1325 /* redundant indications, but ensure initialization. */
1326 tmp_ext
.cookie_out_never
= 1; /* true */
1327 tmp_ext
.cookie_plus
= 0;
1329 goto drop_and_release
;
1331 tmp_ext
.cookie_in_always
= tp
->rx_opt
.cookie_in_always
;
1333 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1334 tcp_clear_options(&tmp_opt
);
1336 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1337 tcp_openreq_init(req
, &tmp_opt
, skb
);
1339 ireq
= inet_rsk(req
);
1340 ireq
->loc_addr
= daddr
;
1341 ireq
->rmt_addr
= saddr
;
1342 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1343 ireq
->opt
= tcp_v4_save_options(sk
, skb
);
1345 if (security_inet_conn_request(sk
, skb
, req
))
1348 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1349 TCP_ECN_create_request(req
, tcp_hdr(skb
));
1352 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1353 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1355 struct inet_peer
*peer
= NULL
;
1358 /* VJ's idea. We save last timestamp seen
1359 * from the destination in peer table, when entering
1360 * state TIME-WAIT, and check against it before
1361 * accepting new connection request.
1363 * If "isn" is not zero, this request hit alive
1364 * timewait bucket, so that all the necessary checks
1365 * are made in the function processing timewait state.
1367 if (tmp_opt
.saw_tstamp
&&
1368 tcp_death_row
.sysctl_tw_recycle
&&
1369 (dst
= inet_csk_route_req(sk
, &fl4
, req
)) != NULL
&&
1370 fl4
.daddr
== saddr
&&
1371 (peer
= rt_get_peer((struct rtable
*)dst
, fl4
.daddr
)) != NULL
) {
1372 inet_peer_refcheck(peer
);
1373 if ((u32
)get_seconds() - peer
->tcp_ts_stamp
< TCP_PAWS_MSL
&&
1374 (s32
)(peer
->tcp_ts
- req
->ts_recent
) >
1376 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1377 goto drop_and_release
;
1380 /* Kill the following clause, if you dislike this way. */
1381 else if (!sysctl_tcp_syncookies
&&
1382 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1383 (sysctl_max_syn_backlog
>> 2)) &&
1384 (!peer
|| !peer
->tcp_ts_stamp
) &&
1385 (!dst
|| !dst_metric(dst
, RTAX_RTT
))) {
1386 /* Without syncookies last quarter of
1387 * backlog is filled with destinations,
1388 * proven to be alive.
1389 * It means that we continue to communicate
1390 * to destinations, already remembered
1391 * to the moment of synflood.
1393 LIMIT_NETDEBUG(KERN_DEBUG
"TCP: drop open request from %pI4/%u\n",
1394 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1395 goto drop_and_release
;
1398 isn
= tcp_v4_init_sequence(skb
);
1400 tcp_rsk(req
)->snt_isn
= isn
;
1401 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1403 if (tcp_v4_send_synack(sk
, dst
, req
,
1404 (struct request_values
*)&tmp_ext
) ||
1408 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1418 EXPORT_SYMBOL(tcp_v4_conn_request
);
1422 * The three way handshake has completed - we got a valid synack -
1423 * now create the new socket.
1425 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1426 struct request_sock
*req
,
1427 struct dst_entry
*dst
)
1429 struct inet_request_sock
*ireq
;
1430 struct inet_sock
*newinet
;
1431 struct tcp_sock
*newtp
;
1433 #ifdef CONFIG_TCP_MD5SIG
1434 struct tcp_md5sig_key
*key
;
1436 struct ip_options_rcu
*inet_opt
;
1438 if (sk_acceptq_is_full(sk
))
1441 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1445 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1447 newtp
= tcp_sk(newsk
);
1448 newinet
= inet_sk(newsk
);
1449 ireq
= inet_rsk(req
);
1450 newinet
->inet_daddr
= ireq
->rmt_addr
;
1451 newinet
->inet_rcv_saddr
= ireq
->loc_addr
;
1452 newinet
->inet_saddr
= ireq
->loc_addr
;
1453 inet_opt
= ireq
->opt
;
1454 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1456 newinet
->mc_index
= inet_iif(skb
);
1457 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1458 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1460 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1461 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1463 if (!dst
&& (dst
= inet_csk_route_child_sock(sk
, newsk
, req
)) == NULL
)
1466 sk_setup_caps(newsk
, dst
);
1468 tcp_mtup_init(newsk
);
1469 tcp_sync_mss(newsk
, dst_mtu(dst
));
1470 newtp
->advmss
= dst_metric_advmss(dst
);
1471 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1472 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1473 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1475 tcp_initialize_rcv_mss(newsk
);
1476 if (tcp_rsk(req
)->snt_synack
)
1477 tcp_valid_rtt_meas(newsk
,
1478 tcp_time_stamp
- tcp_rsk(req
)->snt_synack
);
1479 newtp
->total_retrans
= req
->retrans
;
1481 #ifdef CONFIG_TCP_MD5SIG
1482 /* Copy over the MD5 key from the original socket */
1483 key
= tcp_v4_md5_do_lookup(sk
, newinet
->inet_daddr
);
1486 * We're using one, so create a matching key
1487 * on the newsk structure. If we fail to get
1488 * memory, then we end up not copying the key
1491 char *newkey
= kmemdup(key
->key
, key
->keylen
, GFP_ATOMIC
);
1493 tcp_v4_md5_do_add(newsk
, newinet
->inet_daddr
,
1494 newkey
, key
->keylen
);
1495 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1499 if (__inet_inherit_port(sk
, newsk
) < 0)
1501 __inet_hash_nolisten(newsk
, NULL
);
1506 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1510 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1513 tcp_clear_xmit_timers(newsk
);
1514 bh_unlock_sock(newsk
);
1518 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1520 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1522 struct tcphdr
*th
= tcp_hdr(skb
);
1523 const struct iphdr
*iph
= ip_hdr(skb
);
1525 struct request_sock
**prev
;
1526 /* Find possible connection requests. */
1527 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1528 iph
->saddr
, iph
->daddr
);
1530 return tcp_check_req(sk
, skb
, req
, prev
);
1532 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1533 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1536 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1540 inet_twsk_put(inet_twsk(nsk
));
1544 #ifdef CONFIG_SYN_COOKIES
1546 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1551 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1553 const struct iphdr
*iph
= ip_hdr(skb
);
1555 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1556 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1557 iph
->daddr
, skb
->csum
)) {
1558 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1563 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1564 skb
->len
, IPPROTO_TCP
, 0);
1566 if (skb
->len
<= 76) {
1567 return __skb_checksum_complete(skb
);
1573 /* The socket must have it's spinlock held when we get
1576 * We have a potential double-lock case here, so even when
1577 * doing backlog processing we use the BH locking scheme.
1578 * This is because we cannot sleep with the original spinlock
1581 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1584 #ifdef CONFIG_TCP_MD5SIG
1586 * We really want to reject the packet as early as possible
1588 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1589 * o There is an MD5 option and we're not expecting one
1591 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1595 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1596 sock_rps_save_rxhash(sk
, skb
);
1597 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1604 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1607 if (sk
->sk_state
== TCP_LISTEN
) {
1608 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1613 sock_rps_save_rxhash(nsk
, skb
);
1614 if (tcp_child_process(sk
, nsk
, skb
)) {
1621 sock_rps_save_rxhash(sk
, skb
);
1623 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1630 tcp_v4_send_reset(rsk
, skb
);
1633 /* Be careful here. If this function gets more complicated and
1634 * gcc suffers from register pressure on the x86, sk (in %ebx)
1635 * might be destroyed here. This current version compiles correctly,
1636 * but you have been warned.
1641 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1644 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1650 int tcp_v4_rcv(struct sk_buff
*skb
)
1652 const struct iphdr
*iph
;
1653 const struct tcphdr
*th
;
1656 struct net
*net
= dev_net(skb
->dev
);
1658 if (skb
->pkt_type
!= PACKET_HOST
)
1661 /* Count it even if it's bad */
1662 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1664 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1669 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1671 if (!pskb_may_pull(skb
, th
->doff
* 4))
1674 /* An explanation is required here, I think.
1675 * Packet length and doff are validated by header prediction,
1676 * provided case of th->doff==0 is eliminated.
1677 * So, we defer the checks. */
1678 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
1683 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1684 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1685 skb
->len
- th
->doff
* 4);
1686 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1687 TCP_SKB_CB(skb
)->when
= 0;
1688 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
1689 TCP_SKB_CB(skb
)->sacked
= 0;
1691 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1696 if (sk
->sk_state
== TCP_TIME_WAIT
)
1699 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1700 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1701 goto discard_and_relse
;
1704 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1705 goto discard_and_relse
;
1708 if (sk_filter(sk
, skb
))
1709 goto discard_and_relse
;
1713 bh_lock_sock_nested(sk
);
1715 if (!sock_owned_by_user(sk
)) {
1716 #ifdef CONFIG_NET_DMA
1717 struct tcp_sock
*tp
= tcp_sk(sk
);
1718 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1719 tp
->ucopy
.dma_chan
= dma_find_channel(DMA_MEMCPY
);
1720 if (tp
->ucopy
.dma_chan
)
1721 ret
= tcp_v4_do_rcv(sk
, skb
);
1725 if (!tcp_prequeue(sk
, skb
))
1726 ret
= tcp_v4_do_rcv(sk
, skb
);
1728 } else if (unlikely(sk_add_backlog(sk
, skb
))) {
1730 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1731 goto discard_and_relse
;
1740 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1743 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1745 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1747 tcp_v4_send_reset(NULL
, skb
);
1751 /* Discard frame. */
1760 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1761 inet_twsk_put(inet_twsk(sk
));
1765 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1766 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1767 inet_twsk_put(inet_twsk(sk
));
1770 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1772 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1774 iph
->daddr
, th
->dest
,
1777 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1778 inet_twsk_put(inet_twsk(sk
));
1782 /* Fall through to ACK */
1785 tcp_v4_timewait_ack(sk
, skb
);
1789 case TCP_TW_SUCCESS
:;
1794 struct inet_peer
*tcp_v4_get_peer(struct sock
*sk
, bool *release_it
)
1796 struct rtable
*rt
= (struct rtable
*) __sk_dst_get(sk
);
1797 struct inet_sock
*inet
= inet_sk(sk
);
1798 struct inet_peer
*peer
;
1801 inet
->cork
.fl
.u
.ip4
.daddr
!= inet
->inet_daddr
) {
1802 peer
= inet_getpeer_v4(inet
->inet_daddr
, 1);
1806 rt_bind_peer(rt
, inet
->inet_daddr
, 1);
1808 *release_it
= false;
1813 EXPORT_SYMBOL(tcp_v4_get_peer
);
1815 void *tcp_v4_tw_get_peer(struct sock
*sk
)
1817 const struct inet_timewait_sock
*tw
= inet_twsk(sk
);
1819 return inet_getpeer_v4(tw
->tw_daddr
, 1);
1821 EXPORT_SYMBOL(tcp_v4_tw_get_peer
);
1823 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1824 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1825 .twsk_unique
= tcp_twsk_unique
,
1826 .twsk_destructor
= tcp_twsk_destructor
,
1827 .twsk_getpeer
= tcp_v4_tw_get_peer
,
1830 const struct inet_connection_sock_af_ops ipv4_specific
= {
1831 .queue_xmit
= ip_queue_xmit
,
1832 .send_check
= tcp_v4_send_check
,
1833 .rebuild_header
= inet_sk_rebuild_header
,
1834 .conn_request
= tcp_v4_conn_request
,
1835 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1836 .get_peer
= tcp_v4_get_peer
,
1837 .net_header_len
= sizeof(struct iphdr
),
1838 .setsockopt
= ip_setsockopt
,
1839 .getsockopt
= ip_getsockopt
,
1840 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1841 .sockaddr_len
= sizeof(struct sockaddr_in
),
1842 .bind_conflict
= inet_csk_bind_conflict
,
1843 #ifdef CONFIG_COMPAT
1844 .compat_setsockopt
= compat_ip_setsockopt
,
1845 .compat_getsockopt
= compat_ip_getsockopt
,
1848 EXPORT_SYMBOL(ipv4_specific
);
1850 #ifdef CONFIG_TCP_MD5SIG
1851 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1852 .md5_lookup
= tcp_v4_md5_lookup
,
1853 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1854 .md5_add
= tcp_v4_md5_add_func
,
1855 .md5_parse
= tcp_v4_parse_md5_keys
,
1859 /* NOTE: A lot of things set to zero explicitly by call to
1860 * sk_alloc() so need not be done here.
1862 static int tcp_v4_init_sock(struct sock
*sk
)
1864 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1865 struct tcp_sock
*tp
= tcp_sk(sk
);
1867 skb_queue_head_init(&tp
->out_of_order_queue
);
1868 tcp_init_xmit_timers(sk
);
1869 tcp_prequeue_init(tp
);
1871 icsk
->icsk_rto
= TCP_TIMEOUT_INIT
;
1872 tp
->mdev
= TCP_TIMEOUT_INIT
;
1874 /* So many TCP implementations out there (incorrectly) count the
1875 * initial SYN frame in their delayed-ACK and congestion control
1876 * algorithms that we must have the following bandaid to talk
1877 * efficiently to them. -DaveM
1879 tp
->snd_cwnd
= TCP_INIT_CWND
;
1881 /* See draft-stevens-tcpca-spec-01 for discussion of the
1882 * initialization of these values.
1884 tp
->snd_ssthresh
= TCP_INFINITE_SSTHRESH
;
1885 tp
->snd_cwnd_clamp
= ~0;
1886 tp
->mss_cache
= TCP_MSS_DEFAULT
;
1888 tp
->reordering
= sysctl_tcp_reordering
;
1889 icsk
->icsk_ca_ops
= &tcp_init_congestion_ops
;
1891 sk
->sk_state
= TCP_CLOSE
;
1893 sk
->sk_write_space
= sk_stream_write_space
;
1894 sock_set_flag(sk
, SOCK_USE_WRITE_QUEUE
);
1896 icsk
->icsk_af_ops
= &ipv4_specific
;
1897 icsk
->icsk_sync_mss
= tcp_sync_mss
;
1898 #ifdef CONFIG_TCP_MD5SIG
1899 tp
->af_specific
= &tcp_sock_ipv4_specific
;
1902 /* TCP Cookie Transactions */
1903 if (sysctl_tcp_cookie_size
> 0) {
1904 /* Default, cookies without s_data_payload. */
1906 kzalloc(sizeof(*tp
->cookie_values
),
1908 if (tp
->cookie_values
!= NULL
)
1909 kref_init(&tp
->cookie_values
->kref
);
1911 /* Presumed zeroed, in order of appearance:
1912 * cookie_in_always, cookie_out_never,
1913 * s_data_constant, s_data_in, s_data_out
1915 sk
->sk_sndbuf
= sysctl_tcp_wmem
[1];
1916 sk
->sk_rcvbuf
= sysctl_tcp_rmem
[1];
1919 percpu_counter_inc(&tcp_sockets_allocated
);
1925 void tcp_v4_destroy_sock(struct sock
*sk
)
1927 struct tcp_sock
*tp
= tcp_sk(sk
);
1929 tcp_clear_xmit_timers(sk
);
1931 tcp_cleanup_congestion_control(sk
);
1933 /* Cleanup up the write buffer. */
1934 tcp_write_queue_purge(sk
);
1936 /* Cleans up our, hopefully empty, out_of_order_queue. */
1937 __skb_queue_purge(&tp
->out_of_order_queue
);
1939 #ifdef CONFIG_TCP_MD5SIG
1940 /* Clean up the MD5 key list, if any */
1941 if (tp
->md5sig_info
) {
1942 tcp_v4_clear_md5_list(sk
);
1943 kfree(tp
->md5sig_info
);
1944 tp
->md5sig_info
= NULL
;
1948 #ifdef CONFIG_NET_DMA
1949 /* Cleans up our sk_async_wait_queue */
1950 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1953 /* Clean prequeue, it must be empty really */
1954 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1956 /* Clean up a referenced TCP bind bucket. */
1957 if (inet_csk(sk
)->icsk_bind_hash
)
1961 * If sendmsg cached page exists, toss it.
1963 if (sk
->sk_sndmsg_page
) {
1964 __free_page(sk
->sk_sndmsg_page
);
1965 sk
->sk_sndmsg_page
= NULL
;
1968 /* TCP Cookie Transactions */
1969 if (tp
->cookie_values
!= NULL
) {
1970 kref_put(&tp
->cookie_values
->kref
,
1971 tcp_cookie_values_release
);
1972 tp
->cookie_values
= NULL
;
1975 percpu_counter_dec(&tcp_sockets_allocated
);
1977 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1979 #ifdef CONFIG_PROC_FS
1980 /* Proc filesystem TCP sock list dumping. */
1982 static inline struct inet_timewait_sock
*tw_head(struct hlist_nulls_head
*head
)
1984 return hlist_nulls_empty(head
) ? NULL
:
1985 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
1988 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
1990 return !is_a_nulls(tw
->tw_node
.next
) ?
1991 hlist_nulls_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
1995 * Get next listener socket follow cur. If cur is NULL, get first socket
1996 * starting from bucket given in st->bucket; when st->bucket is zero the
1997 * very first socket in the hash table is returned.
1999 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
2001 struct inet_connection_sock
*icsk
;
2002 struct hlist_nulls_node
*node
;
2003 struct sock
*sk
= cur
;
2004 struct inet_listen_hashbucket
*ilb
;
2005 struct tcp_iter_state
*st
= seq
->private;
2006 struct net
*net
= seq_file_net(seq
);
2009 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2010 spin_lock_bh(&ilb
->lock
);
2011 sk
= sk_nulls_head(&ilb
->head
);
2015 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2019 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
2020 struct request_sock
*req
= cur
;
2022 icsk
= inet_csk(st
->syn_wait_sk
);
2026 if (req
->rsk_ops
->family
== st
->family
) {
2032 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
2035 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
2037 sk
= sk_nulls_next(st
->syn_wait_sk
);
2038 st
->state
= TCP_SEQ_STATE_LISTENING
;
2039 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2041 icsk
= inet_csk(sk
);
2042 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2043 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
2045 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2046 sk
= sk_nulls_next(sk
);
2049 sk_nulls_for_each_from(sk
, node
) {
2050 if (!net_eq(sock_net(sk
), net
))
2052 if (sk
->sk_family
== st
->family
) {
2056 icsk
= inet_csk(sk
);
2057 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2058 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2060 st
->uid
= sock_i_uid(sk
);
2061 st
->syn_wait_sk
= sk
;
2062 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2066 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2068 spin_unlock_bh(&ilb
->lock
);
2070 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2071 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2072 spin_lock_bh(&ilb
->lock
);
2073 sk
= sk_nulls_head(&ilb
->head
);
2081 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2083 struct tcp_iter_state
*st
= seq
->private;
2088 rc
= listening_get_next(seq
, NULL
);
2090 while (rc
&& *pos
) {
2091 rc
= listening_get_next(seq
, rc
);
2097 static inline int empty_bucket(struct tcp_iter_state
*st
)
2099 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
) &&
2100 hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2104 * Get first established socket starting from bucket given in st->bucket.
2105 * If st->bucket is zero, the very first socket in the hash is returned.
2107 static void *established_get_first(struct seq_file
*seq
)
2109 struct tcp_iter_state
*st
= seq
->private;
2110 struct net
*net
= seq_file_net(seq
);
2114 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2116 struct hlist_nulls_node
*node
;
2117 struct inet_timewait_sock
*tw
;
2118 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2120 /* Lockless fast path for the common case of empty buckets */
2121 if (empty_bucket(st
))
2125 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2126 if (sk
->sk_family
!= st
->family
||
2127 !net_eq(sock_net(sk
), net
)) {
2133 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2134 inet_twsk_for_each(tw
, node
,
2135 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
2136 if (tw
->tw_family
!= st
->family
||
2137 !net_eq(twsk_net(tw
), net
)) {
2143 spin_unlock_bh(lock
);
2144 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2150 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2152 struct sock
*sk
= cur
;
2153 struct inet_timewait_sock
*tw
;
2154 struct hlist_nulls_node
*node
;
2155 struct tcp_iter_state
*st
= seq
->private;
2156 struct net
*net
= seq_file_net(seq
);
2161 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2165 while (tw
&& (tw
->tw_family
!= st
->family
|| !net_eq(twsk_net(tw
), net
))) {
2172 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2173 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2175 /* Look for next non empty bucket */
2177 while (++st
->bucket
<= tcp_hashinfo
.ehash_mask
&&
2180 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2183 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2184 sk
= sk_nulls_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2186 sk
= sk_nulls_next(sk
);
2188 sk_nulls_for_each_from(sk
, node
) {
2189 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2193 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2194 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2202 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2204 struct tcp_iter_state
*st
= seq
->private;
2208 rc
= established_get_first(seq
);
2211 rc
= established_get_next(seq
, rc
);
2217 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2220 struct tcp_iter_state
*st
= seq
->private;
2222 st
->state
= TCP_SEQ_STATE_LISTENING
;
2223 rc
= listening_get_idx(seq
, &pos
);
2226 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2227 rc
= established_get_idx(seq
, pos
);
2233 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2235 struct tcp_iter_state
*st
= seq
->private;
2236 int offset
= st
->offset
;
2237 int orig_num
= st
->num
;
2240 switch (st
->state
) {
2241 case TCP_SEQ_STATE_OPENREQ
:
2242 case TCP_SEQ_STATE_LISTENING
:
2243 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2245 st
->state
= TCP_SEQ_STATE_LISTENING
;
2246 rc
= listening_get_next(seq
, NULL
);
2247 while (offset
-- && rc
)
2248 rc
= listening_get_next(seq
, rc
);
2253 case TCP_SEQ_STATE_ESTABLISHED
:
2254 case TCP_SEQ_STATE_TIME_WAIT
:
2255 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2256 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2258 rc
= established_get_first(seq
);
2259 while (offset
-- && rc
)
2260 rc
= established_get_next(seq
, rc
);
2268 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2270 struct tcp_iter_state
*st
= seq
->private;
2273 if (*pos
&& *pos
== st
->last_pos
) {
2274 rc
= tcp_seek_last_pos(seq
);
2279 st
->state
= TCP_SEQ_STATE_LISTENING
;
2283 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2286 st
->last_pos
= *pos
;
2290 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2292 struct tcp_iter_state
*st
= seq
->private;
2295 if (v
== SEQ_START_TOKEN
) {
2296 rc
= tcp_get_idx(seq
, 0);
2300 switch (st
->state
) {
2301 case TCP_SEQ_STATE_OPENREQ
:
2302 case TCP_SEQ_STATE_LISTENING
:
2303 rc
= listening_get_next(seq
, v
);
2305 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2308 rc
= established_get_first(seq
);
2311 case TCP_SEQ_STATE_ESTABLISHED
:
2312 case TCP_SEQ_STATE_TIME_WAIT
:
2313 rc
= established_get_next(seq
, v
);
2318 st
->last_pos
= *pos
;
2322 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2324 struct tcp_iter_state
*st
= seq
->private;
2326 switch (st
->state
) {
2327 case TCP_SEQ_STATE_OPENREQ
:
2329 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2330 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2332 case TCP_SEQ_STATE_LISTENING
:
2333 if (v
!= SEQ_START_TOKEN
)
2334 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2336 case TCP_SEQ_STATE_TIME_WAIT
:
2337 case TCP_SEQ_STATE_ESTABLISHED
:
2339 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2344 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2346 struct tcp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
2347 struct tcp_iter_state
*s
;
2350 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2351 sizeof(struct tcp_iter_state
));
2355 s
= ((struct seq_file
*)file
->private_data
)->private;
2356 s
->family
= afinfo
->family
;
2360 EXPORT_SYMBOL(tcp_seq_open
);
2362 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2365 struct proc_dir_entry
*p
;
2367 afinfo
->seq_ops
.start
= tcp_seq_start
;
2368 afinfo
->seq_ops
.next
= tcp_seq_next
;
2369 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2371 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2372 afinfo
->seq_fops
, afinfo
);
2377 EXPORT_SYMBOL(tcp_proc_register
);
2379 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2381 proc_net_remove(net
, afinfo
->name
);
2383 EXPORT_SYMBOL(tcp_proc_unregister
);
2385 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2386 struct seq_file
*f
, int i
, int uid
, int *len
)
2388 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2389 int ttd
= req
->expires
- jiffies
;
2391 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2392 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2395 ntohs(inet_sk(sk
)->inet_sport
),
2397 ntohs(ireq
->rmt_port
),
2399 0, 0, /* could print option size, but that is af dependent. */
2400 1, /* timers active (only the expire timer) */
2401 jiffies_to_clock_t(ttd
),
2404 0, /* non standard timer */
2405 0, /* open_requests have no inode */
2406 atomic_read(&sk
->sk_refcnt
),
2411 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
, int *len
)
2414 unsigned long timer_expires
;
2415 const struct tcp_sock
*tp
= tcp_sk(sk
);
2416 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2417 const struct inet_sock
*inet
= inet_sk(sk
);
2418 __be32 dest
= inet
->inet_daddr
;
2419 __be32 src
= inet
->inet_rcv_saddr
;
2420 __u16 destp
= ntohs(inet
->inet_dport
);
2421 __u16 srcp
= ntohs(inet
->inet_sport
);
2424 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
) {
2426 timer_expires
= icsk
->icsk_timeout
;
2427 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2429 timer_expires
= icsk
->icsk_timeout
;
2430 } else if (timer_pending(&sk
->sk_timer
)) {
2432 timer_expires
= sk
->sk_timer
.expires
;
2435 timer_expires
= jiffies
;
2438 if (sk
->sk_state
== TCP_LISTEN
)
2439 rx_queue
= sk
->sk_ack_backlog
;
2442 * because we dont lock socket, we might find a transient negative value
2444 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2446 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2447 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2448 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2449 tp
->write_seq
- tp
->snd_una
,
2452 jiffies_to_clock_t(timer_expires
- jiffies
),
2453 icsk
->icsk_retransmits
,
2455 icsk
->icsk_probes_out
,
2457 atomic_read(&sk
->sk_refcnt
), sk
,
2458 jiffies_to_clock_t(icsk
->icsk_rto
),
2459 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2460 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2462 tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
,
2466 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2467 struct seq_file
*f
, int i
, int *len
)
2471 int ttd
= tw
->tw_ttd
- jiffies
;
2476 dest
= tw
->tw_daddr
;
2477 src
= tw
->tw_rcv_saddr
;
2478 destp
= ntohs(tw
->tw_dport
);
2479 srcp
= ntohs(tw
->tw_sport
);
2481 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2482 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2483 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2484 3, jiffies_to_clock_t(ttd
), 0, 0, 0, 0,
2485 atomic_read(&tw
->tw_refcnt
), tw
, len
);
2490 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2492 struct tcp_iter_state
*st
;
2495 if (v
== SEQ_START_TOKEN
) {
2496 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2497 " sl local_address rem_address st tx_queue "
2498 "rx_queue tr tm->when retrnsmt uid timeout "
2504 switch (st
->state
) {
2505 case TCP_SEQ_STATE_LISTENING
:
2506 case TCP_SEQ_STATE_ESTABLISHED
:
2507 get_tcp4_sock(v
, seq
, st
->num
, &len
);
2509 case TCP_SEQ_STATE_OPENREQ
:
2510 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
, &len
);
2512 case TCP_SEQ_STATE_TIME_WAIT
:
2513 get_timewait4_sock(v
, seq
, st
->num
, &len
);
2516 seq_printf(seq
, "%*s\n", TMPSZ
- 1 - len
, "");
2521 static const struct file_operations tcp_afinfo_seq_fops
= {
2522 .owner
= THIS_MODULE
,
2523 .open
= tcp_seq_open
,
2525 .llseek
= seq_lseek
,
2526 .release
= seq_release_net
2529 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2532 .seq_fops
= &tcp_afinfo_seq_fops
,
2534 .show
= tcp4_seq_show
,
2538 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2540 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2543 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2545 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2548 static struct pernet_operations tcp4_net_ops
= {
2549 .init
= tcp4_proc_init_net
,
2550 .exit
= tcp4_proc_exit_net
,
2553 int __init
tcp4_proc_init(void)
2555 return register_pernet_subsys(&tcp4_net_ops
);
2558 void tcp4_proc_exit(void)
2560 unregister_pernet_subsys(&tcp4_net_ops
);
2562 #endif /* CONFIG_PROC_FS */
2564 struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2566 const struct iphdr
*iph
= skb_gro_network_header(skb
);
2568 switch (skb
->ip_summed
) {
2569 case CHECKSUM_COMPLETE
:
2570 if (!tcp_v4_check(skb_gro_len(skb
), iph
->saddr
, iph
->daddr
,
2572 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2578 NAPI_GRO_CB(skb
)->flush
= 1;
2582 return tcp_gro_receive(head
, skb
);
2585 int tcp4_gro_complete(struct sk_buff
*skb
)
2587 const struct iphdr
*iph
= ip_hdr(skb
);
2588 struct tcphdr
*th
= tcp_hdr(skb
);
2590 th
->check
= ~tcp_v4_check(skb
->len
- skb_transport_offset(skb
),
2591 iph
->saddr
, iph
->daddr
, 0);
2592 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
2594 return tcp_gro_complete(skb
);
2597 struct proto tcp_prot
= {
2599 .owner
= THIS_MODULE
,
2601 .connect
= tcp_v4_connect
,
2602 .disconnect
= tcp_disconnect
,
2603 .accept
= inet_csk_accept
,
2605 .init
= tcp_v4_init_sock
,
2606 .destroy
= tcp_v4_destroy_sock
,
2607 .shutdown
= tcp_shutdown
,
2608 .setsockopt
= tcp_setsockopt
,
2609 .getsockopt
= tcp_getsockopt
,
2610 .recvmsg
= tcp_recvmsg
,
2611 .sendmsg
= tcp_sendmsg
,
2612 .sendpage
= tcp_sendpage
,
2613 .backlog_rcv
= tcp_v4_do_rcv
,
2615 .unhash
= inet_unhash
,
2616 .get_port
= inet_csk_get_port
,
2617 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2618 .sockets_allocated
= &tcp_sockets_allocated
,
2619 .orphan_count
= &tcp_orphan_count
,
2620 .memory_allocated
= &tcp_memory_allocated
,
2621 .memory_pressure
= &tcp_memory_pressure
,
2622 .sysctl_mem
= sysctl_tcp_mem
,
2623 .sysctl_wmem
= sysctl_tcp_wmem
,
2624 .sysctl_rmem
= sysctl_tcp_rmem
,
2625 .max_header
= MAX_TCP_HEADER
,
2626 .obj_size
= sizeof(struct tcp_sock
),
2627 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2628 .twsk_prot
= &tcp_timewait_sock_ops
,
2629 .rsk_prot
= &tcp_request_sock_ops
,
2630 .h
.hashinfo
= &tcp_hashinfo
,
2631 .no_autobind
= true,
2632 #ifdef CONFIG_COMPAT
2633 .compat_setsockopt
= compat_tcp_setsockopt
,
2634 .compat_getsockopt
= compat_tcp_getsockopt
,
2637 EXPORT_SYMBOL(tcp_prot
);
2640 static int __net_init
tcp_sk_init(struct net
*net
)
2642 return inet_ctl_sock_create(&net
->ipv4
.tcp_sock
,
2643 PF_INET
, SOCK_RAW
, IPPROTO_TCP
, net
);
2646 static void __net_exit
tcp_sk_exit(struct net
*net
)
2648 inet_ctl_sock_destroy(net
->ipv4
.tcp_sock
);
2651 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2653 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2656 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2657 .init
= tcp_sk_init
,
2658 .exit
= tcp_sk_exit
,
2659 .exit_batch
= tcp_sk_exit_batch
,
2662 void __init
tcp_v4_init(void)
2664 inet_hashinfo_init(&tcp_hashinfo
);
2665 if (register_pernet_subsys(&tcp_sk_ops
))
2666 panic("Failed to create the TCP control socket.\n");