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>
76 #include <net/tcp_memcontrol.h>
78 #include <linux/inet.h>
79 #include <linux/ipv6.h>
80 #include <linux/stddef.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
84 #include <linux/crypto.h>
85 #include <linux/scatterlist.h>
87 int sysctl_tcp_tw_reuse __read_mostly
;
88 int sysctl_tcp_low_latency __read_mostly
;
89 EXPORT_SYMBOL(sysctl_tcp_low_latency
);
92 #ifdef CONFIG_TCP_MD5SIG
93 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
94 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
);
97 struct inet_hashinfo tcp_hashinfo
;
98 EXPORT_SYMBOL(tcp_hashinfo
);
100 static inline __u32
tcp_v4_init_sequence(const struct sk_buff
*skb
)
102 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
105 tcp_hdr(skb
)->source
);
108 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
)
110 const struct tcp_timewait_sock
*tcptw
= tcp_twsk(sktw
);
111 struct tcp_sock
*tp
= tcp_sk(sk
);
113 /* With PAWS, it is safe from the viewpoint
114 of data integrity. Even without PAWS it is safe provided sequence
115 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
117 Actually, the idea is close to VJ's one, only timestamp cache is
118 held not per host, but per port pair and TW bucket is used as state
121 If TW bucket has been already destroyed we fall back to VJ's scheme
122 and use initial timestamp retrieved from peer table.
124 if (tcptw
->tw_ts_recent_stamp
&&
125 (twp
== NULL
|| (sysctl_tcp_tw_reuse
&&
126 get_seconds() - tcptw
->tw_ts_recent_stamp
> 1))) {
127 tp
->write_seq
= tcptw
->tw_snd_nxt
+ 65535 + 2;
128 if (tp
->write_seq
== 0)
130 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
131 tp
->rx_opt
.ts_recent_stamp
= tcptw
->tw_ts_recent_stamp
;
138 EXPORT_SYMBOL_GPL(tcp_twsk_unique
);
140 /* This will initiate an outgoing connection. */
141 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
143 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
144 struct inet_sock
*inet
= inet_sk(sk
);
145 struct tcp_sock
*tp
= tcp_sk(sk
);
146 __be16 orig_sport
, orig_dport
;
147 __be32 daddr
, nexthop
;
151 struct ip_options_rcu
*inet_opt
;
153 if (addr_len
< sizeof(struct sockaddr_in
))
156 if (usin
->sin_family
!= AF_INET
)
157 return -EAFNOSUPPORT
;
159 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
160 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
161 sock_owned_by_user(sk
));
162 if (inet_opt
&& inet_opt
->opt
.srr
) {
165 nexthop
= inet_opt
->opt
.faddr
;
168 orig_sport
= inet
->inet_sport
;
169 orig_dport
= usin
->sin_port
;
170 fl4
= &inet
->cork
.fl
.u
.ip4
;
171 rt
= ip_route_connect(fl4
, nexthop
, inet
->inet_saddr
,
172 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
174 orig_sport
, orig_dport
, sk
, true);
177 if (err
== -ENETUNREACH
)
178 IP_INC_STATS_BH(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
182 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
187 if (!inet_opt
|| !inet_opt
->opt
.srr
)
190 if (!inet
->inet_saddr
)
191 inet
->inet_saddr
= fl4
->saddr
;
192 inet
->inet_rcv_saddr
= inet
->inet_saddr
;
194 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->inet_daddr
!= daddr
) {
195 /* Reset inherited state */
196 tp
->rx_opt
.ts_recent
= 0;
197 tp
->rx_opt
.ts_recent_stamp
= 0;
201 if (tcp_death_row
.sysctl_tw_recycle
&&
202 !tp
->rx_opt
.ts_recent_stamp
&& fl4
->daddr
== daddr
) {
203 struct inet_peer
*peer
= rt_get_peer(rt
, fl4
->daddr
);
205 * VJ's idea. We save last timestamp seen from
206 * the destination in peer table, when entering state
207 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
208 * when trying new connection.
211 inet_peer_refcheck(peer
);
212 if ((u32
)get_seconds() - peer
->tcp_ts_stamp
<= TCP_PAWS_MSL
) {
213 tp
->rx_opt
.ts_recent_stamp
= peer
->tcp_ts_stamp
;
214 tp
->rx_opt
.ts_recent
= peer
->tcp_ts
;
219 inet
->inet_dport
= usin
->sin_port
;
220 inet
->inet_daddr
= daddr
;
222 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
224 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
226 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
228 /* Socket identity is still unknown (sport may be zero).
229 * However we set state to SYN-SENT and not releasing socket
230 * lock select source port, enter ourselves into the hash tables and
231 * complete initialization after this.
233 tcp_set_state(sk
, TCP_SYN_SENT
);
234 err
= inet_hash_connect(&tcp_death_row
, sk
);
238 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
239 inet
->inet_sport
, inet
->inet_dport
, sk
);
245 /* OK, now commit destination to socket. */
246 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
247 sk_setup_caps(sk
, &rt
->dst
);
250 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
255 inet
->inet_id
= tp
->write_seq
^ jiffies
;
257 err
= tcp_connect(sk
);
266 * This unhashes the socket and releases the local port,
269 tcp_set_state(sk
, TCP_CLOSE
);
271 sk
->sk_route_caps
= 0;
272 inet
->inet_dport
= 0;
275 EXPORT_SYMBOL(tcp_v4_connect
);
278 * This routine does path mtu discovery as defined in RFC1191.
280 static void do_pmtu_discovery(struct sock
*sk
, const struct iphdr
*iph
, u32 mtu
)
282 struct dst_entry
*dst
;
283 struct inet_sock
*inet
= inet_sk(sk
);
285 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
286 * send out by Linux are always <576bytes so they should go through
289 if (sk
->sk_state
== TCP_LISTEN
)
292 /* We don't check in the destentry if pmtu discovery is forbidden
293 * on this route. We just assume that no packet_to_big packets
294 * are send back when pmtu discovery is not active.
295 * There is a small race when the user changes this flag in the
296 * route, but I think that's acceptable.
298 if ((dst
= __sk_dst_check(sk
, 0)) == NULL
)
301 dst
->ops
->update_pmtu(dst
, mtu
);
303 /* Something is about to be wrong... Remember soft error
304 * for the case, if this connection will not able to recover.
306 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
307 sk
->sk_err_soft
= EMSGSIZE
;
311 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
312 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
313 tcp_sync_mss(sk
, mtu
);
315 /* Resend the TCP packet because it's
316 * clear that the old packet has been
317 * dropped. This is the new "fast" path mtu
320 tcp_simple_retransmit(sk
);
321 } /* else let the usual retransmit timer handle it */
325 * This routine is called by the ICMP module when it gets some
326 * sort of error condition. If err < 0 then the socket should
327 * be closed and the error returned to the user. If err > 0
328 * it's just the icmp type << 8 | icmp code. After adjustment
329 * header points to the first 8 bytes of the tcp header. We need
330 * to find the appropriate port.
332 * The locking strategy used here is very "optimistic". When
333 * someone else accesses the socket the ICMP is just dropped
334 * and for some paths there is no check at all.
335 * A more general error queue to queue errors for later handling
336 * is probably better.
340 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
342 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
343 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
344 struct inet_connection_sock
*icsk
;
346 struct inet_sock
*inet
;
347 const int type
= icmp_hdr(icmp_skb
)->type
;
348 const int code
= icmp_hdr(icmp_skb
)->code
;
354 struct net
*net
= dev_net(icmp_skb
->dev
);
356 if (icmp_skb
->len
< (iph
->ihl
<< 2) + 8) {
357 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
361 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
362 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
364 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
367 if (sk
->sk_state
== TCP_TIME_WAIT
) {
368 inet_twsk_put(inet_twsk(sk
));
373 /* If too many ICMPs get dropped on busy
374 * servers this needs to be solved differently.
376 if (sock_owned_by_user(sk
))
377 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
379 if (sk
->sk_state
== TCP_CLOSE
)
382 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
383 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
389 seq
= ntohl(th
->seq
);
390 if (sk
->sk_state
!= TCP_LISTEN
&&
391 !between(seq
, tp
->snd_una
, tp
->snd_nxt
)) {
392 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
397 case ICMP_SOURCE_QUENCH
:
398 /* Just silently ignore these. */
400 case ICMP_PARAMETERPROB
:
403 case ICMP_DEST_UNREACH
:
404 if (code
> NR_ICMP_UNREACH
)
407 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
408 if (!sock_owned_by_user(sk
))
409 do_pmtu_discovery(sk
, iph
, info
);
413 err
= icmp_err_convert
[code
].errno
;
414 /* check if icmp_skb allows revert of backoff
415 * (see draft-zimmermann-tcp-lcd) */
416 if (code
!= ICMP_NET_UNREACH
&& code
!= ICMP_HOST_UNREACH
)
418 if (seq
!= tp
->snd_una
|| !icsk
->icsk_retransmits
||
422 if (sock_owned_by_user(sk
))
425 icsk
->icsk_backoff
--;
426 inet_csk(sk
)->icsk_rto
= (tp
->srtt
? __tcp_set_rto(tp
) :
427 TCP_TIMEOUT_INIT
) << icsk
->icsk_backoff
;
430 skb
= tcp_write_queue_head(sk
);
433 remaining
= icsk
->icsk_rto
- min(icsk
->icsk_rto
,
434 tcp_time_stamp
- TCP_SKB_CB(skb
)->when
);
437 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
438 remaining
, TCP_RTO_MAX
);
440 /* RTO revert clocked out retransmission.
441 * Will retransmit now */
442 tcp_retransmit_timer(sk
);
446 case ICMP_TIME_EXCEEDED
:
453 switch (sk
->sk_state
) {
454 struct request_sock
*req
, **prev
;
456 if (sock_owned_by_user(sk
))
459 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
460 iph
->daddr
, iph
->saddr
);
464 /* ICMPs are not backlogged, hence we cannot get
465 an established socket here.
469 if (seq
!= tcp_rsk(req
)->snt_isn
) {
470 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
475 * Still in SYN_RECV, just remove it silently.
476 * There is no good way to pass the error to the newly
477 * created socket, and POSIX does not want network
478 * errors returned from accept().
480 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
484 case TCP_SYN_RECV
: /* Cannot happen.
485 It can f.e. if SYNs crossed.
487 if (!sock_owned_by_user(sk
)) {
490 sk
->sk_error_report(sk
);
494 sk
->sk_err_soft
= err
;
499 /* If we've already connected we will keep trying
500 * until we time out, or the user gives up.
502 * rfc1122 4.2.3.9 allows to consider as hard errors
503 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
504 * but it is obsoleted by pmtu discovery).
506 * Note, that in modern internet, where routing is unreliable
507 * and in each dark corner broken firewalls sit, sending random
508 * errors ordered by their masters even this two messages finally lose
509 * their original sense (even Linux sends invalid PORT_UNREACHs)
511 * Now we are in compliance with RFCs.
516 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
518 sk
->sk_error_report(sk
);
519 } else { /* Only an error on timeout */
520 sk
->sk_err_soft
= err
;
528 static void __tcp_v4_send_check(struct sk_buff
*skb
,
529 __be32 saddr
, __be32 daddr
)
531 struct tcphdr
*th
= tcp_hdr(skb
);
533 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
534 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
535 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
536 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
538 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
545 /* This routine computes an IPv4 TCP checksum. */
546 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
548 const struct inet_sock
*inet
= inet_sk(sk
);
550 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
552 EXPORT_SYMBOL(tcp_v4_send_check
);
554 int tcp_v4_gso_send_check(struct sk_buff
*skb
)
556 const struct iphdr
*iph
;
559 if (!pskb_may_pull(skb
, sizeof(*th
)))
566 skb
->ip_summed
= CHECKSUM_PARTIAL
;
567 __tcp_v4_send_check(skb
, iph
->saddr
, iph
->daddr
);
572 * This routine will send an RST to the other tcp.
574 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
576 * Answer: if a packet caused RST, it is not for a socket
577 * existing in our system, if it is matched to a socket,
578 * it is just duplicate segment or bug in other side's TCP.
579 * So that we build reply only basing on parameters
580 * arrived with segment.
581 * Exception: precedence violation. We do not implement it in any case.
584 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
586 const struct tcphdr
*th
= tcp_hdr(skb
);
589 #ifdef CONFIG_TCP_MD5SIG
590 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
593 struct ip_reply_arg arg
;
594 #ifdef CONFIG_TCP_MD5SIG
595 struct tcp_md5sig_key
*key
;
596 const __u8
*hash_location
= NULL
;
597 unsigned char newhash
[16];
599 struct sock
*sk1
= NULL
;
603 /* Never send a reset in response to a reset. */
607 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
610 /* Swap the send and the receive. */
611 memset(&rep
, 0, sizeof(rep
));
612 rep
.th
.dest
= th
->source
;
613 rep
.th
.source
= th
->dest
;
614 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
618 rep
.th
.seq
= th
->ack_seq
;
621 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
622 skb
->len
- (th
->doff
<< 2));
625 memset(&arg
, 0, sizeof(arg
));
626 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
627 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
629 #ifdef CONFIG_TCP_MD5SIG
630 hash_location
= tcp_parse_md5sig_option(th
);
631 if (!sk
&& hash_location
) {
633 * active side is lost. Try to find listening socket through
634 * source port, and then find md5 key through listening socket.
635 * we are not loose security here:
636 * Incoming packet is checked with md5 hash with finding key,
637 * no RST generated if md5 hash doesn't match.
639 sk1
= __inet_lookup_listener(dev_net(skb_dst(skb
)->dev
),
640 &tcp_hashinfo
, ip_hdr(skb
)->daddr
,
641 ntohs(th
->source
), inet_iif(skb
));
642 /* don't send rst if it can't find key */
646 key
= tcp_md5_do_lookup(sk1
, (union tcp_md5_addr
*)
647 &ip_hdr(skb
)->saddr
, AF_INET
);
651 genhash
= tcp_v4_md5_hash_skb(newhash
, key
, NULL
, NULL
, skb
);
652 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
655 key
= sk
? tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)
661 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
663 (TCPOPT_MD5SIG
<< 8) |
665 /* Update length and the length the header thinks exists */
666 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
667 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
669 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
670 key
, ip_hdr(skb
)->saddr
,
671 ip_hdr(skb
)->daddr
, &rep
.th
);
674 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
675 ip_hdr(skb
)->saddr
, /* XXX */
676 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
677 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
678 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
679 /* When socket is gone, all binding information is lost.
680 * routing might fail in this case. using iif for oif to
681 * make sure we can deliver it
683 arg
.bound_dev_if
= sk
? sk
->sk_bound_dev_if
: inet_iif(skb
);
685 net
= dev_net(skb_dst(skb
)->dev
);
686 arg
.tos
= ip_hdr(skb
)->tos
;
687 ip_send_reply(net
->ipv4
.tcp_sock
, skb
, ip_hdr(skb
)->saddr
,
688 &arg
, arg
.iov
[0].iov_len
);
690 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
691 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
693 #ifdef CONFIG_TCP_MD5SIG
702 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
703 outside socket context is ugly, certainly. What can I do?
706 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
707 u32 win
, u32 ts
, int oif
,
708 struct tcp_md5sig_key
*key
,
709 int reply_flags
, u8 tos
)
711 const struct tcphdr
*th
= tcp_hdr(skb
);
714 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
715 #ifdef CONFIG_TCP_MD5SIG
716 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
720 struct ip_reply_arg arg
;
721 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
723 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
724 memset(&arg
, 0, sizeof(arg
));
726 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
727 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
729 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
730 (TCPOPT_TIMESTAMP
<< 8) |
732 rep
.opt
[1] = htonl(tcp_time_stamp
);
733 rep
.opt
[2] = htonl(ts
);
734 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
737 /* Swap the send and the receive. */
738 rep
.th
.dest
= th
->source
;
739 rep
.th
.source
= th
->dest
;
740 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
741 rep
.th
.seq
= htonl(seq
);
742 rep
.th
.ack_seq
= htonl(ack
);
744 rep
.th
.window
= htons(win
);
746 #ifdef CONFIG_TCP_MD5SIG
748 int offset
= (ts
) ? 3 : 0;
750 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
752 (TCPOPT_MD5SIG
<< 8) |
754 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
755 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
757 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
758 key
, ip_hdr(skb
)->saddr
,
759 ip_hdr(skb
)->daddr
, &rep
.th
);
762 arg
.flags
= reply_flags
;
763 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
764 ip_hdr(skb
)->saddr
, /* XXX */
765 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
766 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
768 arg
.bound_dev_if
= oif
;
770 ip_send_reply(net
->ipv4
.tcp_sock
, skb
, ip_hdr(skb
)->saddr
,
771 &arg
, arg
.iov
[0].iov_len
);
773 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
776 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
778 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
779 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
781 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
782 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
785 tcp_twsk_md5_key(tcptw
),
786 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
793 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
794 struct request_sock
*req
)
796 tcp_v4_send_ack(skb
, tcp_rsk(req
)->snt_isn
+ 1,
797 tcp_rsk(req
)->rcv_isn
+ 1, req
->rcv_wnd
,
800 tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&ip_hdr(skb
)->daddr
,
802 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
807 * Send a SYN-ACK after having received a SYN.
808 * This still operates on a request_sock only, not on a big
811 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
812 struct request_sock
*req
,
813 struct request_values
*rvp
)
815 const struct inet_request_sock
*ireq
= inet_rsk(req
);
818 struct sk_buff
* skb
;
820 /* First, grab a route. */
821 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
824 skb
= tcp_make_synack(sk
, dst
, req
, rvp
);
827 __tcp_v4_send_check(skb
, ireq
->loc_addr
, ireq
->rmt_addr
);
829 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
832 err
= net_xmit_eval(err
);
839 static int tcp_v4_rtx_synack(struct sock
*sk
, struct request_sock
*req
,
840 struct request_values
*rvp
)
842 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
843 return tcp_v4_send_synack(sk
, NULL
, req
, rvp
);
847 * IPv4 request_sock destructor.
849 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
851 kfree(inet_rsk(req
)->opt
);
855 * Return 1 if a syncookie should be sent
857 int tcp_syn_flood_action(struct sock
*sk
,
858 const struct sk_buff
*skb
,
861 const char *msg
= "Dropping request";
863 struct listen_sock
*lopt
;
867 #ifdef CONFIG_SYN_COOKIES
868 if (sysctl_tcp_syncookies
) {
869 msg
= "Sending cookies";
871 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
874 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
876 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
877 if (!lopt
->synflood_warned
) {
878 lopt
->synflood_warned
= 1;
879 pr_info("%s: Possible SYN flooding on port %d. %s. "
880 " Check SNMP counters.\n",
881 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
885 EXPORT_SYMBOL(tcp_syn_flood_action
);
888 * Save and compile IPv4 options into the request_sock if needed.
890 static struct ip_options_rcu
*tcp_v4_save_options(struct sock
*sk
,
893 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
894 struct ip_options_rcu
*dopt
= NULL
;
896 if (opt
&& opt
->optlen
) {
897 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
899 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
901 if (ip_options_echo(&dopt
->opt
, skb
)) {
910 #ifdef CONFIG_TCP_MD5SIG
912 * RFC2385 MD5 checksumming requires a mapping of
913 * IP address->MD5 Key.
914 * We need to maintain these in the sk structure.
917 /* Find the Key structure for an address. */
918 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
919 const union tcp_md5_addr
*addr
,
922 struct tcp_sock
*tp
= tcp_sk(sk
);
923 struct tcp_md5sig_key
*key
;
924 struct hlist_node
*pos
;
925 unsigned int size
= sizeof(struct in_addr
);
926 struct tcp_md5sig_info
*md5sig
;
928 /* caller either holds rcu_read_lock() or socket lock */
929 md5sig
= rcu_dereference_check(tp
->md5sig_info
,
930 sock_owned_by_user(sk
) ||
931 lockdep_is_held(&sk
->sk_lock
.slock
));
934 #if IS_ENABLED(CONFIG_IPV6)
935 if (family
== AF_INET6
)
936 size
= sizeof(struct in6_addr
);
938 hlist_for_each_entry_rcu(key
, pos
, &md5sig
->head
, node
) {
939 if (key
->family
!= family
)
941 if (!memcmp(&key
->addr
, addr
, size
))
946 EXPORT_SYMBOL(tcp_md5_do_lookup
);
948 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
949 struct sock
*addr_sk
)
951 union tcp_md5_addr
*addr
;
953 addr
= (union tcp_md5_addr
*)&inet_sk(addr_sk
)->inet_daddr
;
954 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
956 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
958 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
959 struct request_sock
*req
)
961 union tcp_md5_addr
*addr
;
963 addr
= (union tcp_md5_addr
*)&inet_rsk(req
)->rmt_addr
;
964 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
967 /* This can be called on a newly created socket, from other files */
968 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
969 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
)
971 /* Add Key to the list */
972 struct tcp_md5sig_key
*key
;
973 struct tcp_sock
*tp
= tcp_sk(sk
);
974 struct tcp_md5sig_info
*md5sig
;
976 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&addr
, AF_INET
);
978 /* Pre-existing entry - just update that one. */
979 memcpy(key
->key
, newkey
, newkeylen
);
980 key
->keylen
= newkeylen
;
984 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
985 sock_owned_by_user(sk
));
987 md5sig
= kmalloc(sizeof(*md5sig
), gfp
);
991 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
992 INIT_HLIST_HEAD(&md5sig
->head
);
993 rcu_assign_pointer(tp
->md5sig_info
, md5sig
);
996 key
= sock_kmalloc(sk
, sizeof(*key
), gfp
);
999 if (hlist_empty(&md5sig
->head
) && !tcp_alloc_md5sig_pool(sk
)) {
1000 sock_kfree_s(sk
, key
, sizeof(*key
));
1004 memcpy(key
->key
, newkey
, newkeylen
);
1005 key
->keylen
= newkeylen
;
1006 key
->family
= family
;
1007 memcpy(&key
->addr
, addr
,
1008 (family
== AF_INET6
) ? sizeof(struct in6_addr
) :
1009 sizeof(struct in_addr
));
1010 hlist_add_head_rcu(&key
->node
, &md5sig
->head
);
1013 EXPORT_SYMBOL(tcp_md5_do_add
);
1015 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
, int family
)
1017 struct tcp_sock
*tp
= tcp_sk(sk
);
1018 struct tcp_md5sig_key
*key
;
1019 struct tcp_md5sig_info
*md5sig
;
1021 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&addr
, AF_INET
);
1024 hlist_del_rcu(&key
->node
);
1025 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1026 kfree_rcu(key
, rcu
);
1027 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
1028 sock_owned_by_user(sk
));
1029 if (hlist_empty(&md5sig
->head
))
1030 tcp_free_md5sig_pool();
1033 EXPORT_SYMBOL(tcp_md5_do_del
);
1035 void tcp_clear_md5_list(struct sock
*sk
)
1037 struct tcp_sock
*tp
= tcp_sk(sk
);
1038 struct tcp_md5sig_key
*key
;
1039 struct hlist_node
*pos
, *n
;
1040 struct tcp_md5sig_info
*md5sig
;
1042 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
1044 if (!hlist_empty(&md5sig
->head
))
1045 tcp_free_md5sig_pool();
1046 hlist_for_each_entry_safe(key
, pos
, n
, &md5sig
->head
, node
) {
1047 hlist_del_rcu(&key
->node
);
1048 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1049 kfree_rcu(key
, rcu
);
1053 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1056 struct tcp_md5sig cmd
;
1057 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1059 if (optlen
< sizeof(cmd
))
1062 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1065 if (sin
->sin_family
!= AF_INET
)
1068 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
)
1069 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1072 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1075 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1076 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1080 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1081 __be32 daddr
, __be32 saddr
, int nbytes
)
1083 struct tcp4_pseudohdr
*bp
;
1084 struct scatterlist sg
;
1086 bp
= &hp
->md5_blk
.ip4
;
1089 * 1. the TCP pseudo-header (in the order: source IP address,
1090 * destination IP address, zero-padded protocol number, and
1096 bp
->protocol
= IPPROTO_TCP
;
1097 bp
->len
= cpu_to_be16(nbytes
);
1099 sg_init_one(&sg
, bp
, sizeof(*bp
));
1100 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1103 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1104 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1106 struct tcp_md5sig_pool
*hp
;
1107 struct hash_desc
*desc
;
1109 hp
= tcp_get_md5sig_pool();
1111 goto clear_hash_noput
;
1112 desc
= &hp
->md5_desc
;
1114 if (crypto_hash_init(desc
))
1116 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1118 if (tcp_md5_hash_header(hp
, th
))
1120 if (tcp_md5_hash_key(hp
, key
))
1122 if (crypto_hash_final(desc
, md5_hash
))
1125 tcp_put_md5sig_pool();
1129 tcp_put_md5sig_pool();
1131 memset(md5_hash
, 0, 16);
1135 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1136 const struct sock
*sk
, const struct request_sock
*req
,
1137 const struct sk_buff
*skb
)
1139 struct tcp_md5sig_pool
*hp
;
1140 struct hash_desc
*desc
;
1141 const struct tcphdr
*th
= tcp_hdr(skb
);
1142 __be32 saddr
, daddr
;
1145 saddr
= inet_sk(sk
)->inet_saddr
;
1146 daddr
= inet_sk(sk
)->inet_daddr
;
1148 saddr
= inet_rsk(req
)->loc_addr
;
1149 daddr
= inet_rsk(req
)->rmt_addr
;
1151 const struct iphdr
*iph
= ip_hdr(skb
);
1156 hp
= tcp_get_md5sig_pool();
1158 goto clear_hash_noput
;
1159 desc
= &hp
->md5_desc
;
1161 if (crypto_hash_init(desc
))
1164 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1166 if (tcp_md5_hash_header(hp
, th
))
1168 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1170 if (tcp_md5_hash_key(hp
, key
))
1172 if (crypto_hash_final(desc
, md5_hash
))
1175 tcp_put_md5sig_pool();
1179 tcp_put_md5sig_pool();
1181 memset(md5_hash
, 0, 16);
1184 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1186 static int tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1189 * This gets called for each TCP segment that arrives
1190 * so we want to be efficient.
1191 * We have 3 drop cases:
1192 * o No MD5 hash and one expected.
1193 * o MD5 hash and we're not expecting one.
1194 * o MD5 hash and its wrong.
1196 const __u8
*hash_location
= NULL
;
1197 struct tcp_md5sig_key
*hash_expected
;
1198 const struct iphdr
*iph
= ip_hdr(skb
);
1199 const struct tcphdr
*th
= tcp_hdr(skb
);
1201 unsigned char newhash
[16];
1203 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1205 hash_location
= tcp_parse_md5sig_option(th
);
1207 /* We've parsed the options - do we have a hash? */
1208 if (!hash_expected
&& !hash_location
)
1211 if (hash_expected
&& !hash_location
) {
1212 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1216 if (!hash_expected
&& hash_location
) {
1217 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1221 /* Okay, so this is hash_expected and hash_location -
1222 * so we need to calculate the checksum.
1224 genhash
= tcp_v4_md5_hash_skb(newhash
,
1228 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1229 if (net_ratelimit()) {
1230 pr_info("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1231 &iph
->saddr
, ntohs(th
->source
),
1232 &iph
->daddr
, ntohs(th
->dest
),
1233 genhash
? " tcp_v4_calc_md5_hash failed" : "");
1242 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1244 .obj_size
= sizeof(struct tcp_request_sock
),
1245 .rtx_syn_ack
= tcp_v4_rtx_synack
,
1246 .send_ack
= tcp_v4_reqsk_send_ack
,
1247 .destructor
= tcp_v4_reqsk_destructor
,
1248 .send_reset
= tcp_v4_send_reset
,
1249 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1252 #ifdef CONFIG_TCP_MD5SIG
1253 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1254 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1255 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1259 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1261 struct tcp_extend_values tmp_ext
;
1262 struct tcp_options_received tmp_opt
;
1263 const u8
*hash_location
;
1264 struct request_sock
*req
;
1265 struct inet_request_sock
*ireq
;
1266 struct tcp_sock
*tp
= tcp_sk(sk
);
1267 struct dst_entry
*dst
= NULL
;
1268 __be32 saddr
= ip_hdr(skb
)->saddr
;
1269 __be32 daddr
= ip_hdr(skb
)->daddr
;
1270 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1271 int want_cookie
= 0;
1273 /* Never answer to SYNs send to broadcast or multicast */
1274 if (skb_rtable(skb
)->rt_flags
& (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 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCP");
1287 /* Accept backlog is full. If we have already queued enough
1288 * of warm entries in syn queue, drop request. It is better than
1289 * clogging syn queue with openreqs with exponentially increasing
1292 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1)
1295 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1299 #ifdef CONFIG_TCP_MD5SIG
1300 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1303 tcp_clear_options(&tmp_opt
);
1304 tmp_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
1305 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
1306 tcp_parse_options(skb
, &tmp_opt
, &hash_location
, 0);
1308 if (tmp_opt
.cookie_plus
> 0 &&
1309 tmp_opt
.saw_tstamp
&&
1310 !tp
->rx_opt
.cookie_out_never
&&
1311 (sysctl_tcp_cookie_size
> 0 ||
1312 (tp
->cookie_values
!= NULL
&&
1313 tp
->cookie_values
->cookie_desired
> 0))) {
1315 u32
*mess
= &tmp_ext
.cookie_bakery
[COOKIE_DIGEST_WORDS
];
1316 int l
= tmp_opt
.cookie_plus
- TCPOLEN_COOKIE_BASE
;
1318 if (tcp_cookie_generator(&tmp_ext
.cookie_bakery
[0]) != 0)
1319 goto drop_and_release
;
1321 /* Secret recipe starts with IP addresses */
1322 *mess
++ ^= (__force u32
)daddr
;
1323 *mess
++ ^= (__force u32
)saddr
;
1325 /* plus variable length Initiator Cookie */
1328 *c
++ ^= *hash_location
++;
1330 want_cookie
= 0; /* not our kind of cookie */
1331 tmp_ext
.cookie_out_never
= 0; /* false */
1332 tmp_ext
.cookie_plus
= tmp_opt
.cookie_plus
;
1333 } else if (!tp
->rx_opt
.cookie_in_always
) {
1334 /* redundant indications, but ensure initialization. */
1335 tmp_ext
.cookie_out_never
= 1; /* true */
1336 tmp_ext
.cookie_plus
= 0;
1338 goto drop_and_release
;
1340 tmp_ext
.cookie_in_always
= tp
->rx_opt
.cookie_in_always
;
1342 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1343 tcp_clear_options(&tmp_opt
);
1345 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1346 tcp_openreq_init(req
, &tmp_opt
, skb
);
1348 ireq
= inet_rsk(req
);
1349 ireq
->loc_addr
= daddr
;
1350 ireq
->rmt_addr
= saddr
;
1351 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1352 ireq
->opt
= tcp_v4_save_options(sk
, skb
);
1354 if (security_inet_conn_request(sk
, skb
, req
))
1357 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1358 TCP_ECN_create_request(req
, tcp_hdr(skb
));
1361 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1362 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1364 struct inet_peer
*peer
= NULL
;
1367 /* VJ's idea. We save last timestamp seen
1368 * from the destination in peer table, when entering
1369 * state TIME-WAIT, and check against it before
1370 * accepting new connection request.
1372 * If "isn" is not zero, this request hit alive
1373 * timewait bucket, so that all the necessary checks
1374 * are made in the function processing timewait state.
1376 if (tmp_opt
.saw_tstamp
&&
1377 tcp_death_row
.sysctl_tw_recycle
&&
1378 (dst
= inet_csk_route_req(sk
, &fl4
, req
)) != NULL
&&
1379 fl4
.daddr
== saddr
&&
1380 (peer
= rt_get_peer((struct rtable
*)dst
, fl4
.daddr
)) != NULL
) {
1381 inet_peer_refcheck(peer
);
1382 if ((u32
)get_seconds() - peer
->tcp_ts_stamp
< TCP_PAWS_MSL
&&
1383 (s32
)(peer
->tcp_ts
- req
->ts_recent
) >
1385 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1386 goto drop_and_release
;
1389 /* Kill the following clause, if you dislike this way. */
1390 else if (!sysctl_tcp_syncookies
&&
1391 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1392 (sysctl_max_syn_backlog
>> 2)) &&
1393 (!peer
|| !peer
->tcp_ts_stamp
) &&
1394 (!dst
|| !dst_metric(dst
, RTAX_RTT
))) {
1395 /* Without syncookies last quarter of
1396 * backlog is filled with destinations,
1397 * proven to be alive.
1398 * It means that we continue to communicate
1399 * to destinations, already remembered
1400 * to the moment of synflood.
1402 LIMIT_NETDEBUG(KERN_DEBUG
"TCP: drop open request from %pI4/%u\n",
1403 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1404 goto drop_and_release
;
1407 isn
= tcp_v4_init_sequence(skb
);
1409 tcp_rsk(req
)->snt_isn
= isn
;
1410 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1412 if (tcp_v4_send_synack(sk
, dst
, req
,
1413 (struct request_values
*)&tmp_ext
) ||
1417 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1427 EXPORT_SYMBOL(tcp_v4_conn_request
);
1431 * The three way handshake has completed - we got a valid synack -
1432 * now create the new socket.
1434 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1435 struct request_sock
*req
,
1436 struct dst_entry
*dst
)
1438 struct inet_request_sock
*ireq
;
1439 struct inet_sock
*newinet
;
1440 struct tcp_sock
*newtp
;
1442 #ifdef CONFIG_TCP_MD5SIG
1443 struct tcp_md5sig_key
*key
;
1445 struct ip_options_rcu
*inet_opt
;
1447 if (sk_acceptq_is_full(sk
))
1450 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1454 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1456 newtp
= tcp_sk(newsk
);
1457 newinet
= inet_sk(newsk
);
1458 ireq
= inet_rsk(req
);
1459 newinet
->inet_daddr
= ireq
->rmt_addr
;
1460 newinet
->inet_rcv_saddr
= ireq
->loc_addr
;
1461 newinet
->inet_saddr
= ireq
->loc_addr
;
1462 inet_opt
= ireq
->opt
;
1463 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1465 newinet
->mc_index
= inet_iif(skb
);
1466 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1467 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1468 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1470 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1471 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1473 if (!dst
&& (dst
= inet_csk_route_child_sock(sk
, newsk
, req
)) == NULL
)
1476 sk_setup_caps(newsk
, dst
);
1478 tcp_mtup_init(newsk
);
1479 tcp_sync_mss(newsk
, dst_mtu(dst
));
1480 newtp
->advmss
= dst_metric_advmss(dst
);
1481 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1482 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1483 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1485 tcp_initialize_rcv_mss(newsk
);
1486 if (tcp_rsk(req
)->snt_synack
)
1487 tcp_valid_rtt_meas(newsk
,
1488 tcp_time_stamp
- tcp_rsk(req
)->snt_synack
);
1489 newtp
->total_retrans
= req
->retrans
;
1491 #ifdef CONFIG_TCP_MD5SIG
1492 /* Copy over the MD5 key from the original socket */
1493 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1497 * We're using one, so create a matching key
1498 * on the newsk structure. If we fail to get
1499 * memory, then we end up not copying the key
1502 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1503 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1504 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1508 if (__inet_inherit_port(sk
, newsk
) < 0)
1510 __inet_hash_nolisten(newsk
, NULL
);
1515 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1519 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1522 tcp_clear_xmit_timers(newsk
);
1523 tcp_cleanup_congestion_control(newsk
);
1524 bh_unlock_sock(newsk
);
1528 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1530 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1532 struct tcphdr
*th
= tcp_hdr(skb
);
1533 const struct iphdr
*iph
= ip_hdr(skb
);
1535 struct request_sock
**prev
;
1536 /* Find possible connection requests. */
1537 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1538 iph
->saddr
, iph
->daddr
);
1540 return tcp_check_req(sk
, skb
, req
, prev
);
1542 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1543 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1546 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1550 inet_twsk_put(inet_twsk(nsk
));
1554 #ifdef CONFIG_SYN_COOKIES
1556 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1561 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1563 const struct iphdr
*iph
= ip_hdr(skb
);
1565 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1566 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1567 iph
->daddr
, skb
->csum
)) {
1568 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1573 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1574 skb
->len
, IPPROTO_TCP
, 0);
1576 if (skb
->len
<= 76) {
1577 return __skb_checksum_complete(skb
);
1583 /* The socket must have it's spinlock held when we get
1586 * We have a potential double-lock case here, so even when
1587 * doing backlog processing we use the BH locking scheme.
1588 * This is because we cannot sleep with the original spinlock
1591 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1594 #ifdef CONFIG_TCP_MD5SIG
1596 * We really want to reject the packet as early as possible
1598 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1599 * o There is an MD5 option and we're not expecting one
1601 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1605 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1606 sock_rps_save_rxhash(sk
, skb
);
1607 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1614 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1617 if (sk
->sk_state
== TCP_LISTEN
) {
1618 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1623 sock_rps_save_rxhash(nsk
, skb
);
1624 if (tcp_child_process(sk
, nsk
, skb
)) {
1631 sock_rps_save_rxhash(sk
, skb
);
1633 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1640 tcp_v4_send_reset(rsk
, skb
);
1643 /* Be careful here. If this function gets more complicated and
1644 * gcc suffers from register pressure on the x86, sk (in %ebx)
1645 * might be destroyed here. This current version compiles correctly,
1646 * but you have been warned.
1651 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1654 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1660 int tcp_v4_rcv(struct sk_buff
*skb
)
1662 const struct iphdr
*iph
;
1663 const struct tcphdr
*th
;
1666 struct net
*net
= dev_net(skb
->dev
);
1668 if (skb
->pkt_type
!= PACKET_HOST
)
1671 /* Count it even if it's bad */
1672 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1674 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1679 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1681 if (!pskb_may_pull(skb
, th
->doff
* 4))
1684 /* An explanation is required here, I think.
1685 * Packet length and doff are validated by header prediction,
1686 * provided case of th->doff==0 is eliminated.
1687 * So, we defer the checks. */
1688 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
1693 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1694 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1695 skb
->len
- th
->doff
* 4);
1696 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1697 TCP_SKB_CB(skb
)->when
= 0;
1698 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
1699 TCP_SKB_CB(skb
)->sacked
= 0;
1701 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1706 if (sk
->sk_state
== TCP_TIME_WAIT
)
1709 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1710 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1711 goto discard_and_relse
;
1714 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1715 goto discard_and_relse
;
1718 if (sk_filter(sk
, skb
))
1719 goto discard_and_relse
;
1723 bh_lock_sock_nested(sk
);
1725 if (!sock_owned_by_user(sk
)) {
1726 #ifdef CONFIG_NET_DMA
1727 struct tcp_sock
*tp
= tcp_sk(sk
);
1728 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1729 tp
->ucopy
.dma_chan
= dma_find_channel(DMA_MEMCPY
);
1730 if (tp
->ucopy
.dma_chan
)
1731 ret
= tcp_v4_do_rcv(sk
, skb
);
1735 if (!tcp_prequeue(sk
, skb
))
1736 ret
= tcp_v4_do_rcv(sk
, skb
);
1738 } else if (unlikely(sk_add_backlog(sk
, skb
))) {
1740 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1741 goto discard_and_relse
;
1750 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1753 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1755 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1757 tcp_v4_send_reset(NULL
, skb
);
1761 /* Discard frame. */
1770 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1771 inet_twsk_put(inet_twsk(sk
));
1775 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1776 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1777 inet_twsk_put(inet_twsk(sk
));
1780 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1782 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1784 iph
->daddr
, th
->dest
,
1787 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1788 inet_twsk_put(inet_twsk(sk
));
1792 /* Fall through to ACK */
1795 tcp_v4_timewait_ack(sk
, skb
);
1799 case TCP_TW_SUCCESS
:;
1804 struct inet_peer
*tcp_v4_get_peer(struct sock
*sk
, bool *release_it
)
1806 struct rtable
*rt
= (struct rtable
*) __sk_dst_get(sk
);
1807 struct inet_sock
*inet
= inet_sk(sk
);
1808 struct inet_peer
*peer
;
1811 inet
->cork
.fl
.u
.ip4
.daddr
!= inet
->inet_daddr
) {
1812 peer
= inet_getpeer_v4(inet
->inet_daddr
, 1);
1816 rt_bind_peer(rt
, inet
->inet_daddr
, 1);
1818 *release_it
= false;
1823 EXPORT_SYMBOL(tcp_v4_get_peer
);
1825 void *tcp_v4_tw_get_peer(struct sock
*sk
)
1827 const struct inet_timewait_sock
*tw
= inet_twsk(sk
);
1829 return inet_getpeer_v4(tw
->tw_daddr
, 1);
1831 EXPORT_SYMBOL(tcp_v4_tw_get_peer
);
1833 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1834 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1835 .twsk_unique
= tcp_twsk_unique
,
1836 .twsk_destructor
= tcp_twsk_destructor
,
1837 .twsk_getpeer
= tcp_v4_tw_get_peer
,
1840 const struct inet_connection_sock_af_ops ipv4_specific
= {
1841 .queue_xmit
= ip_queue_xmit
,
1842 .send_check
= tcp_v4_send_check
,
1843 .rebuild_header
= inet_sk_rebuild_header
,
1844 .conn_request
= tcp_v4_conn_request
,
1845 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1846 .get_peer
= tcp_v4_get_peer
,
1847 .net_header_len
= sizeof(struct iphdr
),
1848 .setsockopt
= ip_setsockopt
,
1849 .getsockopt
= ip_getsockopt
,
1850 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1851 .sockaddr_len
= sizeof(struct sockaddr_in
),
1852 .bind_conflict
= inet_csk_bind_conflict
,
1853 #ifdef CONFIG_COMPAT
1854 .compat_setsockopt
= compat_ip_setsockopt
,
1855 .compat_getsockopt
= compat_ip_getsockopt
,
1858 EXPORT_SYMBOL(ipv4_specific
);
1860 #ifdef CONFIG_TCP_MD5SIG
1861 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1862 .md5_lookup
= tcp_v4_md5_lookup
,
1863 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1864 .md5_parse
= tcp_v4_parse_md5_keys
,
1868 /* NOTE: A lot of things set to zero explicitly by call to
1869 * sk_alloc() so need not be done here.
1871 static int tcp_v4_init_sock(struct sock
*sk
)
1873 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1874 struct tcp_sock
*tp
= tcp_sk(sk
);
1876 skb_queue_head_init(&tp
->out_of_order_queue
);
1877 tcp_init_xmit_timers(sk
);
1878 tcp_prequeue_init(tp
);
1880 icsk
->icsk_rto
= TCP_TIMEOUT_INIT
;
1881 tp
->mdev
= TCP_TIMEOUT_INIT
;
1883 /* So many TCP implementations out there (incorrectly) count the
1884 * initial SYN frame in their delayed-ACK and congestion control
1885 * algorithms that we must have the following bandaid to talk
1886 * efficiently to them. -DaveM
1888 tp
->snd_cwnd
= TCP_INIT_CWND
;
1890 /* See draft-stevens-tcpca-spec-01 for discussion of the
1891 * initialization of these values.
1893 tp
->snd_ssthresh
= TCP_INFINITE_SSTHRESH
;
1894 tp
->snd_cwnd_clamp
= ~0;
1895 tp
->mss_cache
= TCP_MSS_DEFAULT
;
1897 tp
->reordering
= sysctl_tcp_reordering
;
1898 icsk
->icsk_ca_ops
= &tcp_init_congestion_ops
;
1900 sk
->sk_state
= TCP_CLOSE
;
1902 sk
->sk_write_space
= sk_stream_write_space
;
1903 sock_set_flag(sk
, SOCK_USE_WRITE_QUEUE
);
1905 icsk
->icsk_af_ops
= &ipv4_specific
;
1906 icsk
->icsk_sync_mss
= tcp_sync_mss
;
1907 #ifdef CONFIG_TCP_MD5SIG
1908 tp
->af_specific
= &tcp_sock_ipv4_specific
;
1911 /* TCP Cookie Transactions */
1912 if (sysctl_tcp_cookie_size
> 0) {
1913 /* Default, cookies without s_data_payload. */
1915 kzalloc(sizeof(*tp
->cookie_values
),
1917 if (tp
->cookie_values
!= NULL
)
1918 kref_init(&tp
->cookie_values
->kref
);
1920 /* Presumed zeroed, in order of appearance:
1921 * cookie_in_always, cookie_out_never,
1922 * s_data_constant, s_data_in, s_data_out
1924 sk
->sk_sndbuf
= sysctl_tcp_wmem
[1];
1925 sk
->sk_rcvbuf
= sysctl_tcp_rmem
[1];
1928 sock_update_memcg(sk
);
1929 sk_sockets_allocated_inc(sk
);
1935 void tcp_v4_destroy_sock(struct sock
*sk
)
1937 struct tcp_sock
*tp
= tcp_sk(sk
);
1939 tcp_clear_xmit_timers(sk
);
1941 tcp_cleanup_congestion_control(sk
);
1943 /* Cleanup up the write buffer. */
1944 tcp_write_queue_purge(sk
);
1946 /* Cleans up our, hopefully empty, out_of_order_queue. */
1947 __skb_queue_purge(&tp
->out_of_order_queue
);
1949 #ifdef CONFIG_TCP_MD5SIG
1950 /* Clean up the MD5 key list, if any */
1951 if (tp
->md5sig_info
) {
1952 tcp_clear_md5_list(sk
);
1953 kfree_rcu(tp
->md5sig_info
, rcu
);
1954 tp
->md5sig_info
= NULL
;
1958 #ifdef CONFIG_NET_DMA
1959 /* Cleans up our sk_async_wait_queue */
1960 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1963 /* Clean prequeue, it must be empty really */
1964 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1966 /* Clean up a referenced TCP bind bucket. */
1967 if (inet_csk(sk
)->icsk_bind_hash
)
1971 * If sendmsg cached page exists, toss it.
1973 if (sk
->sk_sndmsg_page
) {
1974 __free_page(sk
->sk_sndmsg_page
);
1975 sk
->sk_sndmsg_page
= NULL
;
1978 /* TCP Cookie Transactions */
1979 if (tp
->cookie_values
!= NULL
) {
1980 kref_put(&tp
->cookie_values
->kref
,
1981 tcp_cookie_values_release
);
1982 tp
->cookie_values
= NULL
;
1985 sk_sockets_allocated_dec(sk
);
1986 sock_release_memcg(sk
);
1988 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1990 #ifdef CONFIG_PROC_FS
1991 /* Proc filesystem TCP sock list dumping. */
1993 static inline struct inet_timewait_sock
*tw_head(struct hlist_nulls_head
*head
)
1995 return hlist_nulls_empty(head
) ? NULL
:
1996 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
1999 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
2001 return !is_a_nulls(tw
->tw_node
.next
) ?
2002 hlist_nulls_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
2006 * Get next listener socket follow cur. If cur is NULL, get first socket
2007 * starting from bucket given in st->bucket; when st->bucket is zero the
2008 * very first socket in the hash table is returned.
2010 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
2012 struct inet_connection_sock
*icsk
;
2013 struct hlist_nulls_node
*node
;
2014 struct sock
*sk
= cur
;
2015 struct inet_listen_hashbucket
*ilb
;
2016 struct tcp_iter_state
*st
= seq
->private;
2017 struct net
*net
= seq_file_net(seq
);
2020 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2021 spin_lock_bh(&ilb
->lock
);
2022 sk
= sk_nulls_head(&ilb
->head
);
2026 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2030 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
2031 struct request_sock
*req
= cur
;
2033 icsk
= inet_csk(st
->syn_wait_sk
);
2037 if (req
->rsk_ops
->family
== st
->family
) {
2043 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
2046 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
2048 sk
= sk_nulls_next(st
->syn_wait_sk
);
2049 st
->state
= TCP_SEQ_STATE_LISTENING
;
2050 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2052 icsk
= inet_csk(sk
);
2053 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2054 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
2056 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2057 sk
= sk_nulls_next(sk
);
2060 sk_nulls_for_each_from(sk
, node
) {
2061 if (!net_eq(sock_net(sk
), net
))
2063 if (sk
->sk_family
== st
->family
) {
2067 icsk
= inet_csk(sk
);
2068 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2069 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2071 st
->uid
= sock_i_uid(sk
);
2072 st
->syn_wait_sk
= sk
;
2073 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2077 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2079 spin_unlock_bh(&ilb
->lock
);
2081 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2082 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2083 spin_lock_bh(&ilb
->lock
);
2084 sk
= sk_nulls_head(&ilb
->head
);
2092 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2094 struct tcp_iter_state
*st
= seq
->private;
2099 rc
= listening_get_next(seq
, NULL
);
2101 while (rc
&& *pos
) {
2102 rc
= listening_get_next(seq
, rc
);
2108 static inline int empty_bucket(struct tcp_iter_state
*st
)
2110 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
) &&
2111 hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2115 * Get first established socket starting from bucket given in st->bucket.
2116 * If st->bucket is zero, the very first socket in the hash is returned.
2118 static void *established_get_first(struct seq_file
*seq
)
2120 struct tcp_iter_state
*st
= seq
->private;
2121 struct net
*net
= seq_file_net(seq
);
2125 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2127 struct hlist_nulls_node
*node
;
2128 struct inet_timewait_sock
*tw
;
2129 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2131 /* Lockless fast path for the common case of empty buckets */
2132 if (empty_bucket(st
))
2136 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2137 if (sk
->sk_family
!= st
->family
||
2138 !net_eq(sock_net(sk
), net
)) {
2144 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2145 inet_twsk_for_each(tw
, node
,
2146 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
2147 if (tw
->tw_family
!= st
->family
||
2148 !net_eq(twsk_net(tw
), net
)) {
2154 spin_unlock_bh(lock
);
2155 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2161 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2163 struct sock
*sk
= cur
;
2164 struct inet_timewait_sock
*tw
;
2165 struct hlist_nulls_node
*node
;
2166 struct tcp_iter_state
*st
= seq
->private;
2167 struct net
*net
= seq_file_net(seq
);
2172 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2176 while (tw
&& (tw
->tw_family
!= st
->family
|| !net_eq(twsk_net(tw
), net
))) {
2183 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2184 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2186 /* Look for next non empty bucket */
2188 while (++st
->bucket
<= tcp_hashinfo
.ehash_mask
&&
2191 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2194 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2195 sk
= sk_nulls_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2197 sk
= sk_nulls_next(sk
);
2199 sk_nulls_for_each_from(sk
, node
) {
2200 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2204 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2205 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2213 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2215 struct tcp_iter_state
*st
= seq
->private;
2219 rc
= established_get_first(seq
);
2222 rc
= established_get_next(seq
, rc
);
2228 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2231 struct tcp_iter_state
*st
= seq
->private;
2233 st
->state
= TCP_SEQ_STATE_LISTENING
;
2234 rc
= listening_get_idx(seq
, &pos
);
2237 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2238 rc
= established_get_idx(seq
, pos
);
2244 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2246 struct tcp_iter_state
*st
= seq
->private;
2247 int offset
= st
->offset
;
2248 int orig_num
= st
->num
;
2251 switch (st
->state
) {
2252 case TCP_SEQ_STATE_OPENREQ
:
2253 case TCP_SEQ_STATE_LISTENING
:
2254 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2256 st
->state
= TCP_SEQ_STATE_LISTENING
;
2257 rc
= listening_get_next(seq
, NULL
);
2258 while (offset
-- && rc
)
2259 rc
= listening_get_next(seq
, rc
);
2264 case TCP_SEQ_STATE_ESTABLISHED
:
2265 case TCP_SEQ_STATE_TIME_WAIT
:
2266 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2267 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2269 rc
= established_get_first(seq
);
2270 while (offset
-- && rc
)
2271 rc
= established_get_next(seq
, rc
);
2279 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2281 struct tcp_iter_state
*st
= seq
->private;
2284 if (*pos
&& *pos
== st
->last_pos
) {
2285 rc
= tcp_seek_last_pos(seq
);
2290 st
->state
= TCP_SEQ_STATE_LISTENING
;
2294 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2297 st
->last_pos
= *pos
;
2301 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2303 struct tcp_iter_state
*st
= seq
->private;
2306 if (v
== SEQ_START_TOKEN
) {
2307 rc
= tcp_get_idx(seq
, 0);
2311 switch (st
->state
) {
2312 case TCP_SEQ_STATE_OPENREQ
:
2313 case TCP_SEQ_STATE_LISTENING
:
2314 rc
= listening_get_next(seq
, v
);
2316 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2319 rc
= established_get_first(seq
);
2322 case TCP_SEQ_STATE_ESTABLISHED
:
2323 case TCP_SEQ_STATE_TIME_WAIT
:
2324 rc
= established_get_next(seq
, v
);
2329 st
->last_pos
= *pos
;
2333 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2335 struct tcp_iter_state
*st
= seq
->private;
2337 switch (st
->state
) {
2338 case TCP_SEQ_STATE_OPENREQ
:
2340 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2341 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2343 case TCP_SEQ_STATE_LISTENING
:
2344 if (v
!= SEQ_START_TOKEN
)
2345 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2347 case TCP_SEQ_STATE_TIME_WAIT
:
2348 case TCP_SEQ_STATE_ESTABLISHED
:
2350 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2355 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2357 struct tcp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
2358 struct tcp_iter_state
*s
;
2361 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2362 sizeof(struct tcp_iter_state
));
2366 s
= ((struct seq_file
*)file
->private_data
)->private;
2367 s
->family
= afinfo
->family
;
2371 EXPORT_SYMBOL(tcp_seq_open
);
2373 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2376 struct proc_dir_entry
*p
;
2378 afinfo
->seq_ops
.start
= tcp_seq_start
;
2379 afinfo
->seq_ops
.next
= tcp_seq_next
;
2380 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2382 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2383 afinfo
->seq_fops
, afinfo
);
2388 EXPORT_SYMBOL(tcp_proc_register
);
2390 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2392 proc_net_remove(net
, afinfo
->name
);
2394 EXPORT_SYMBOL(tcp_proc_unregister
);
2396 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2397 struct seq_file
*f
, int i
, int uid
, int *len
)
2399 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2400 int ttd
= req
->expires
- jiffies
;
2402 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2403 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2406 ntohs(inet_sk(sk
)->inet_sport
),
2408 ntohs(ireq
->rmt_port
),
2410 0, 0, /* could print option size, but that is af dependent. */
2411 1, /* timers active (only the expire timer) */
2412 jiffies_to_clock_t(ttd
),
2415 0, /* non standard timer */
2416 0, /* open_requests have no inode */
2417 atomic_read(&sk
->sk_refcnt
),
2422 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
, int *len
)
2425 unsigned long timer_expires
;
2426 const struct tcp_sock
*tp
= tcp_sk(sk
);
2427 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2428 const struct inet_sock
*inet
= inet_sk(sk
);
2429 __be32 dest
= inet
->inet_daddr
;
2430 __be32 src
= inet
->inet_rcv_saddr
;
2431 __u16 destp
= ntohs(inet
->inet_dport
);
2432 __u16 srcp
= ntohs(inet
->inet_sport
);
2435 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
) {
2437 timer_expires
= icsk
->icsk_timeout
;
2438 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2440 timer_expires
= icsk
->icsk_timeout
;
2441 } else if (timer_pending(&sk
->sk_timer
)) {
2443 timer_expires
= sk
->sk_timer
.expires
;
2446 timer_expires
= jiffies
;
2449 if (sk
->sk_state
== TCP_LISTEN
)
2450 rx_queue
= sk
->sk_ack_backlog
;
2453 * because we dont lock socket, we might find a transient negative value
2455 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2457 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2458 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2459 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2460 tp
->write_seq
- tp
->snd_una
,
2463 jiffies_to_clock_t(timer_expires
- jiffies
),
2464 icsk
->icsk_retransmits
,
2466 icsk
->icsk_probes_out
,
2468 atomic_read(&sk
->sk_refcnt
), sk
,
2469 jiffies_to_clock_t(icsk
->icsk_rto
),
2470 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2471 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2473 tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
,
2477 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2478 struct seq_file
*f
, int i
, int *len
)
2482 int ttd
= tw
->tw_ttd
- jiffies
;
2487 dest
= tw
->tw_daddr
;
2488 src
= tw
->tw_rcv_saddr
;
2489 destp
= ntohs(tw
->tw_dport
);
2490 srcp
= ntohs(tw
->tw_sport
);
2492 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2493 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2494 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2495 3, jiffies_to_clock_t(ttd
), 0, 0, 0, 0,
2496 atomic_read(&tw
->tw_refcnt
), tw
, len
);
2501 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2503 struct tcp_iter_state
*st
;
2506 if (v
== SEQ_START_TOKEN
) {
2507 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2508 " sl local_address rem_address st tx_queue "
2509 "rx_queue tr tm->when retrnsmt uid timeout "
2515 switch (st
->state
) {
2516 case TCP_SEQ_STATE_LISTENING
:
2517 case TCP_SEQ_STATE_ESTABLISHED
:
2518 get_tcp4_sock(v
, seq
, st
->num
, &len
);
2520 case TCP_SEQ_STATE_OPENREQ
:
2521 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
, &len
);
2523 case TCP_SEQ_STATE_TIME_WAIT
:
2524 get_timewait4_sock(v
, seq
, st
->num
, &len
);
2527 seq_printf(seq
, "%*s\n", TMPSZ
- 1 - len
, "");
2532 static const struct file_operations tcp_afinfo_seq_fops
= {
2533 .owner
= THIS_MODULE
,
2534 .open
= tcp_seq_open
,
2536 .llseek
= seq_lseek
,
2537 .release
= seq_release_net
2540 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2543 .seq_fops
= &tcp_afinfo_seq_fops
,
2545 .show
= tcp4_seq_show
,
2549 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2551 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2554 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2556 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2559 static struct pernet_operations tcp4_net_ops
= {
2560 .init
= tcp4_proc_init_net
,
2561 .exit
= tcp4_proc_exit_net
,
2564 int __init
tcp4_proc_init(void)
2566 return register_pernet_subsys(&tcp4_net_ops
);
2569 void tcp4_proc_exit(void)
2571 unregister_pernet_subsys(&tcp4_net_ops
);
2573 #endif /* CONFIG_PROC_FS */
2575 struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2577 const struct iphdr
*iph
= skb_gro_network_header(skb
);
2579 switch (skb
->ip_summed
) {
2580 case CHECKSUM_COMPLETE
:
2581 if (!tcp_v4_check(skb_gro_len(skb
), iph
->saddr
, iph
->daddr
,
2583 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2589 NAPI_GRO_CB(skb
)->flush
= 1;
2593 return tcp_gro_receive(head
, skb
);
2596 int tcp4_gro_complete(struct sk_buff
*skb
)
2598 const struct iphdr
*iph
= ip_hdr(skb
);
2599 struct tcphdr
*th
= tcp_hdr(skb
);
2601 th
->check
= ~tcp_v4_check(skb
->len
- skb_transport_offset(skb
),
2602 iph
->saddr
, iph
->daddr
, 0);
2603 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
2605 return tcp_gro_complete(skb
);
2608 struct proto tcp_prot
= {
2610 .owner
= THIS_MODULE
,
2612 .connect
= tcp_v4_connect
,
2613 .disconnect
= tcp_disconnect
,
2614 .accept
= inet_csk_accept
,
2616 .init
= tcp_v4_init_sock
,
2617 .destroy
= tcp_v4_destroy_sock
,
2618 .shutdown
= tcp_shutdown
,
2619 .setsockopt
= tcp_setsockopt
,
2620 .getsockopt
= tcp_getsockopt
,
2621 .recvmsg
= tcp_recvmsg
,
2622 .sendmsg
= tcp_sendmsg
,
2623 .sendpage
= tcp_sendpage
,
2624 .backlog_rcv
= tcp_v4_do_rcv
,
2626 .unhash
= inet_unhash
,
2627 .get_port
= inet_csk_get_port
,
2628 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2629 .sockets_allocated
= &tcp_sockets_allocated
,
2630 .orphan_count
= &tcp_orphan_count
,
2631 .memory_allocated
= &tcp_memory_allocated
,
2632 .memory_pressure
= &tcp_memory_pressure
,
2633 .sysctl_wmem
= sysctl_tcp_wmem
,
2634 .sysctl_rmem
= sysctl_tcp_rmem
,
2635 .max_header
= MAX_TCP_HEADER
,
2636 .obj_size
= sizeof(struct tcp_sock
),
2637 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2638 .twsk_prot
= &tcp_timewait_sock_ops
,
2639 .rsk_prot
= &tcp_request_sock_ops
,
2640 .h
.hashinfo
= &tcp_hashinfo
,
2641 .no_autobind
= true,
2642 #ifdef CONFIG_COMPAT
2643 .compat_setsockopt
= compat_tcp_setsockopt
,
2644 .compat_getsockopt
= compat_tcp_getsockopt
,
2646 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2647 .init_cgroup
= tcp_init_cgroup
,
2648 .destroy_cgroup
= tcp_destroy_cgroup
,
2649 .proto_cgroup
= tcp_proto_cgroup
,
2652 EXPORT_SYMBOL(tcp_prot
);
2654 static int __net_init
tcp_sk_init(struct net
*net
)
2656 return inet_ctl_sock_create(&net
->ipv4
.tcp_sock
,
2657 PF_INET
, SOCK_RAW
, IPPROTO_TCP
, net
);
2660 static void __net_exit
tcp_sk_exit(struct net
*net
)
2662 inet_ctl_sock_destroy(net
->ipv4
.tcp_sock
);
2665 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2667 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2670 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2671 .init
= tcp_sk_init
,
2672 .exit
= tcp_sk_exit
,
2673 .exit_batch
= tcp_sk_exit_batch
,
2676 void __init
tcp_v4_init(void)
2678 inet_hashinfo_init(&tcp_hashinfo
);
2679 if (register_pernet_subsys(&tcp_sk_ops
))
2680 panic("Failed to create the TCP control socket.\n");