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.
53 #define pr_fmt(fmt) "TCP: " fmt
55 #include <linux/bottom_half.h>
56 #include <linux/types.h>
57 #include <linux/fcntl.h>
58 #include <linux/module.h>
59 #include <linux/random.h>
60 #include <linux/cache.h>
61 #include <linux/jhash.h>
62 #include <linux/init.h>
63 #include <linux/times.h>
64 #include <linux/slab.h>
66 #include <net/net_namespace.h>
68 #include <net/inet_hashtables.h>
70 #include <net/transp_v6.h>
72 #include <net/inet_common.h>
73 #include <net/timewait_sock.h>
75 #include <net/netdma.h>
76 #include <net/secure_seq.h>
77 #include <net/tcp_memcontrol.h>
78 #include <net/busy_poll.h>
80 #include <linux/inet.h>
81 #include <linux/ipv6.h>
82 #include <linux/stddef.h>
83 #include <linux/proc_fs.h>
84 #include <linux/seq_file.h>
86 #include <linux/crypto.h>
87 #include <linux/scatterlist.h>
89 int sysctl_tcp_tw_reuse __read_mostly
;
90 int sysctl_tcp_low_latency __read_mostly
;
91 EXPORT_SYMBOL(sysctl_tcp_low_latency
);
93 #ifdef CONFIG_TCP_MD5SIG
94 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
95 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
);
98 struct inet_hashinfo tcp_hashinfo
;
99 EXPORT_SYMBOL(tcp_hashinfo
);
101 static __u32
tcp_v4_init_sequence(const struct sk_buff
*skb
)
103 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
106 tcp_hdr(skb
)->source
);
109 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
)
111 const struct tcp_timewait_sock
*tcptw
= tcp_twsk(sktw
);
112 struct tcp_sock
*tp
= tcp_sk(sk
);
114 /* With PAWS, it is safe from the viewpoint
115 of data integrity. Even without PAWS it is safe provided sequence
116 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
118 Actually, the idea is close to VJ's one, only timestamp cache is
119 held not per host, but per port pair and TW bucket is used as state
122 If TW bucket has been already destroyed we fall back to VJ's scheme
123 and use initial timestamp retrieved from peer table.
125 if (tcptw
->tw_ts_recent_stamp
&&
126 (twp
== NULL
|| (sysctl_tcp_tw_reuse
&&
127 get_seconds() - tcptw
->tw_ts_recent_stamp
> 1))) {
128 tp
->write_seq
= tcptw
->tw_snd_nxt
+ 65535 + 2;
129 if (tp
->write_seq
== 0)
131 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
132 tp
->rx_opt
.ts_recent_stamp
= tcptw
->tw_ts_recent_stamp
;
139 EXPORT_SYMBOL_GPL(tcp_twsk_unique
);
141 /* This will initiate an outgoing connection. */
142 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
144 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
145 struct inet_sock
*inet
= inet_sk(sk
);
146 struct tcp_sock
*tp
= tcp_sk(sk
);
147 __be16 orig_sport
, orig_dport
;
148 __be32 daddr
, nexthop
;
152 struct ip_options_rcu
*inet_opt
;
154 if (addr_len
< sizeof(struct sockaddr_in
))
157 if (usin
->sin_family
!= AF_INET
)
158 return -EAFNOSUPPORT
;
160 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
161 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
162 sock_owned_by_user(sk
));
163 if (inet_opt
&& inet_opt
->opt
.srr
) {
166 nexthop
= inet_opt
->opt
.faddr
;
169 orig_sport
= inet
->inet_sport
;
170 orig_dport
= usin
->sin_port
;
171 fl4
= &inet
->cork
.fl
.u
.ip4
;
172 rt
= ip_route_connect(fl4
, nexthop
, inet
->inet_saddr
,
173 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
175 orig_sport
, orig_dport
, sk
);
178 if (err
== -ENETUNREACH
)
179 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
183 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
188 if (!inet_opt
|| !inet_opt
->opt
.srr
)
191 if (!inet
->inet_saddr
)
192 inet
->inet_saddr
= fl4
->saddr
;
193 inet
->inet_rcv_saddr
= inet
->inet_saddr
;
195 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->inet_daddr
!= daddr
) {
196 /* Reset inherited state */
197 tp
->rx_opt
.ts_recent
= 0;
198 tp
->rx_opt
.ts_recent_stamp
= 0;
199 if (likely(!tp
->repair
))
203 if (tcp_death_row
.sysctl_tw_recycle
&&
204 !tp
->rx_opt
.ts_recent_stamp
&& fl4
->daddr
== daddr
)
205 tcp_fetch_timewait_stamp(sk
, &rt
->dst
);
207 inet
->inet_dport
= usin
->sin_port
;
208 inet
->inet_daddr
= daddr
;
212 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
214 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
216 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
218 /* Socket identity is still unknown (sport may be zero).
219 * However we set state to SYN-SENT and not releasing socket
220 * lock select source port, enter ourselves into the hash tables and
221 * complete initialization after this.
223 tcp_set_state(sk
, TCP_SYN_SENT
);
224 err
= inet_hash_connect(&tcp_death_row
, sk
);
228 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
229 inet
->inet_sport
, inet
->inet_dport
, sk
);
235 /* OK, now commit destination to socket. */
236 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
237 sk_setup_caps(sk
, &rt
->dst
);
239 if (!tp
->write_seq
&& likely(!tp
->repair
))
240 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
245 inet
->inet_id
= tp
->write_seq
^ jiffies
;
247 err
= tcp_connect(sk
);
257 * This unhashes the socket and releases the local port,
260 tcp_set_state(sk
, TCP_CLOSE
);
262 sk
->sk_route_caps
= 0;
263 inet
->inet_dport
= 0;
266 EXPORT_SYMBOL(tcp_v4_connect
);
269 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
270 * It can be called through tcp_release_cb() if socket was owned by user
271 * at the time tcp_v4_err() was called to handle ICMP message.
273 void tcp_v4_mtu_reduced(struct sock
*sk
)
275 struct dst_entry
*dst
;
276 struct inet_sock
*inet
= inet_sk(sk
);
277 u32 mtu
= tcp_sk(sk
)->mtu_info
;
279 dst
= inet_csk_update_pmtu(sk
, mtu
);
283 /* Something is about to be wrong... Remember soft error
284 * for the case, if this connection will not able to recover.
286 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
287 sk
->sk_err_soft
= EMSGSIZE
;
291 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
292 ip_sk_accept_pmtu(sk
) &&
293 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
294 tcp_sync_mss(sk
, mtu
);
296 /* Resend the TCP packet because it's
297 * clear that the old packet has been
298 * dropped. This is the new "fast" path mtu
301 tcp_simple_retransmit(sk
);
302 } /* else let the usual retransmit timer handle it */
304 EXPORT_SYMBOL(tcp_v4_mtu_reduced
);
306 static void do_redirect(struct sk_buff
*skb
, struct sock
*sk
)
308 struct dst_entry
*dst
= __sk_dst_check(sk
, 0);
311 dst
->ops
->redirect(dst
, sk
, skb
);
315 * This routine is called by the ICMP module when it gets some
316 * sort of error condition. If err < 0 then the socket should
317 * be closed and the error returned to the user. If err > 0
318 * it's just the icmp type << 8 | icmp code. After adjustment
319 * header points to the first 8 bytes of the tcp header. We need
320 * to find the appropriate port.
322 * The locking strategy used here is very "optimistic". When
323 * someone else accesses the socket the ICMP is just dropped
324 * and for some paths there is no check at all.
325 * A more general error queue to queue errors for later handling
326 * is probably better.
330 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
332 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
333 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
334 struct inet_connection_sock
*icsk
;
336 struct inet_sock
*inet
;
337 const int type
= icmp_hdr(icmp_skb
)->type
;
338 const int code
= icmp_hdr(icmp_skb
)->code
;
341 struct request_sock
*fastopen
;
345 struct net
*net
= dev_net(icmp_skb
->dev
);
347 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
348 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
350 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
353 if (sk
->sk_state
== TCP_TIME_WAIT
) {
354 inet_twsk_put(inet_twsk(sk
));
359 /* If too many ICMPs get dropped on busy
360 * servers this needs to be solved differently.
361 * We do take care of PMTU discovery (RFC1191) special case :
362 * we can receive locally generated ICMP messages while socket is held.
364 if (sock_owned_by_user(sk
)) {
365 if (!(type
== ICMP_DEST_UNREACH
&& code
== ICMP_FRAG_NEEDED
))
366 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
368 if (sk
->sk_state
== TCP_CLOSE
)
371 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
372 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
378 seq
= ntohl(th
->seq
);
379 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
380 fastopen
= tp
->fastopen_rsk
;
381 snd_una
= fastopen
? tcp_rsk(fastopen
)->snt_isn
: tp
->snd_una
;
382 if (sk
->sk_state
!= TCP_LISTEN
&&
383 !between(seq
, snd_una
, tp
->snd_nxt
)) {
384 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
390 do_redirect(icmp_skb
, sk
);
392 case ICMP_SOURCE_QUENCH
:
393 /* Just silently ignore these. */
395 case ICMP_PARAMETERPROB
:
398 case ICMP_DEST_UNREACH
:
399 if (code
> NR_ICMP_UNREACH
)
402 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
403 /* We are not interested in TCP_LISTEN and open_requests
404 * (SYN-ACKs send out by Linux are always <576bytes so
405 * they should go through unfragmented).
407 if (sk
->sk_state
== TCP_LISTEN
)
411 if (!sock_owned_by_user(sk
)) {
412 tcp_v4_mtu_reduced(sk
);
414 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED
, &tp
->tsq_flags
))
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
||
426 !icsk
->icsk_backoff
|| fastopen
)
429 if (sock_owned_by_user(sk
))
432 icsk
->icsk_backoff
--;
433 icsk
->icsk_rto
= tp
->srtt_us
? __tcp_set_rto(tp
) :
435 icsk
->icsk_rto
= inet_csk_rto_backoff(icsk
, TCP_RTO_MAX
);
437 skb
= tcp_write_queue_head(sk
);
440 remaining
= icsk
->icsk_rto
-
442 tcp_time_stamp
- tcp_skb_timestamp(skb
));
445 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
446 remaining
, TCP_RTO_MAX
);
448 /* RTO revert clocked out retransmission.
449 * Will retransmit now */
450 tcp_retransmit_timer(sk
);
454 case ICMP_TIME_EXCEEDED
:
461 switch (sk
->sk_state
) {
462 struct request_sock
*req
, **prev
;
464 if (sock_owned_by_user(sk
))
467 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
468 iph
->daddr
, iph
->saddr
);
472 /* ICMPs are not backlogged, hence we cannot get
473 an established socket here.
477 if (seq
!= tcp_rsk(req
)->snt_isn
) {
478 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
483 * Still in SYN_RECV, just remove it silently.
484 * There is no good way to pass the error to the newly
485 * created socket, and POSIX does not want network
486 * errors returned from accept().
488 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
489 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
494 /* Only in fast or simultaneous open. If a fast open socket is
495 * is already accepted it is treated as a connected one below.
497 if (fastopen
&& fastopen
->sk
== NULL
)
500 if (!sock_owned_by_user(sk
)) {
503 sk
->sk_error_report(sk
);
507 sk
->sk_err_soft
= err
;
512 /* If we've already connected we will keep trying
513 * until we time out, or the user gives up.
515 * rfc1122 4.2.3.9 allows to consider as hard errors
516 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
517 * but it is obsoleted by pmtu discovery).
519 * Note, that in modern internet, where routing is unreliable
520 * and in each dark corner broken firewalls sit, sending random
521 * errors ordered by their masters even this two messages finally lose
522 * their original sense (even Linux sends invalid PORT_UNREACHs)
524 * Now we are in compliance with RFCs.
529 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
531 sk
->sk_error_report(sk
);
532 } else { /* Only an error on timeout */
533 sk
->sk_err_soft
= err
;
541 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
)
543 struct tcphdr
*th
= tcp_hdr(skb
);
545 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
546 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
547 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
548 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
550 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
557 /* This routine computes an IPv4 TCP checksum. */
558 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
560 const struct inet_sock
*inet
= inet_sk(sk
);
562 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
564 EXPORT_SYMBOL(tcp_v4_send_check
);
567 * This routine will send an RST to the other tcp.
569 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
571 * Answer: if a packet caused RST, it is not for a socket
572 * existing in our system, if it is matched to a socket,
573 * it is just duplicate segment or bug in other side's TCP.
574 * So that we build reply only basing on parameters
575 * arrived with segment.
576 * Exception: precedence violation. We do not implement it in any case.
579 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
581 const struct tcphdr
*th
= tcp_hdr(skb
);
584 #ifdef CONFIG_TCP_MD5SIG
585 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
588 struct ip_reply_arg arg
;
589 #ifdef CONFIG_TCP_MD5SIG
590 struct tcp_md5sig_key
*key
;
591 const __u8
*hash_location
= NULL
;
592 unsigned char newhash
[16];
594 struct sock
*sk1
= NULL
;
598 /* Never send a reset in response to a reset. */
602 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
605 /* Swap the send and the receive. */
606 memset(&rep
, 0, sizeof(rep
));
607 rep
.th
.dest
= th
->source
;
608 rep
.th
.source
= th
->dest
;
609 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
613 rep
.th
.seq
= th
->ack_seq
;
616 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
617 skb
->len
- (th
->doff
<< 2));
620 memset(&arg
, 0, sizeof(arg
));
621 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
622 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
624 #ifdef CONFIG_TCP_MD5SIG
625 hash_location
= tcp_parse_md5sig_option(th
);
626 if (!sk
&& hash_location
) {
628 * active side is lost. Try to find listening socket through
629 * source port, and then find md5 key through listening socket.
630 * we are not loose security here:
631 * Incoming packet is checked with md5 hash with finding key,
632 * no RST generated if md5 hash doesn't match.
634 sk1
= __inet_lookup_listener(dev_net(skb_dst(skb
)->dev
),
635 &tcp_hashinfo
, ip_hdr(skb
)->saddr
,
636 th
->source
, ip_hdr(skb
)->daddr
,
637 ntohs(th
->source
), inet_iif(skb
));
638 /* don't send rst if it can't find key */
642 key
= tcp_md5_do_lookup(sk1
, (union tcp_md5_addr
*)
643 &ip_hdr(skb
)->saddr
, AF_INET
);
647 genhash
= tcp_v4_md5_hash_skb(newhash
, key
, NULL
, NULL
, skb
);
648 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
651 key
= sk
? tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)
657 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
659 (TCPOPT_MD5SIG
<< 8) |
661 /* Update length and the length the header thinks exists */
662 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
663 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
665 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
666 key
, ip_hdr(skb
)->saddr
,
667 ip_hdr(skb
)->daddr
, &rep
.th
);
670 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
671 ip_hdr(skb
)->saddr
, /* XXX */
672 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
673 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
674 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
675 /* When socket is gone, all binding information is lost.
676 * routing might fail in this case. No choice here, if we choose to force
677 * input interface, we will misroute in case of asymmetric route.
680 arg
.bound_dev_if
= sk
->sk_bound_dev_if
;
682 net
= dev_net(skb_dst(skb
)->dev
);
683 arg
.tos
= ip_hdr(skb
)->tos
;
684 ip_send_unicast_reply(net
, skb
, ip_hdr(skb
)->saddr
,
685 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
687 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
688 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
690 #ifdef CONFIG_TCP_MD5SIG
699 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
700 outside socket context is ugly, certainly. What can I do?
703 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
704 u32 win
, u32 tsval
, u32 tsecr
, int oif
,
705 struct tcp_md5sig_key
*key
,
706 int reply_flags
, u8 tos
)
708 const struct tcphdr
*th
= tcp_hdr(skb
);
711 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
712 #ifdef CONFIG_TCP_MD5SIG
713 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
717 struct ip_reply_arg arg
;
718 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
720 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
721 memset(&arg
, 0, sizeof(arg
));
723 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
724 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
726 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
727 (TCPOPT_TIMESTAMP
<< 8) |
729 rep
.opt
[1] = htonl(tsval
);
730 rep
.opt
[2] = htonl(tsecr
);
731 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
734 /* Swap the send and the receive. */
735 rep
.th
.dest
= th
->source
;
736 rep
.th
.source
= th
->dest
;
737 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
738 rep
.th
.seq
= htonl(seq
);
739 rep
.th
.ack_seq
= htonl(ack
);
741 rep
.th
.window
= htons(win
);
743 #ifdef CONFIG_TCP_MD5SIG
745 int offset
= (tsecr
) ? 3 : 0;
747 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
749 (TCPOPT_MD5SIG
<< 8) |
751 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
752 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
754 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
755 key
, ip_hdr(skb
)->saddr
,
756 ip_hdr(skb
)->daddr
, &rep
.th
);
759 arg
.flags
= reply_flags
;
760 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
761 ip_hdr(skb
)->saddr
, /* XXX */
762 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
763 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
765 arg
.bound_dev_if
= oif
;
767 ip_send_unicast_reply(net
, skb
, ip_hdr(skb
)->saddr
,
768 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
770 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
773 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
775 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
776 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
778 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
779 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
780 tcp_time_stamp
+ tcptw
->tw_ts_offset
,
783 tcp_twsk_md5_key(tcptw
),
784 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
791 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
792 struct request_sock
*req
)
794 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
795 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
797 tcp_v4_send_ack(skb
, (sk
->sk_state
== TCP_LISTEN
) ?
798 tcp_rsk(req
)->snt_isn
+ 1 : tcp_sk(sk
)->snd_nxt
,
799 tcp_rsk(req
)->rcv_nxt
, req
->rcv_wnd
,
803 tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&ip_hdr(skb
)->daddr
,
805 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
810 * Send a SYN-ACK after having received a SYN.
811 * This still operates on a request_sock only, not on a big
814 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
816 struct request_sock
*req
,
818 struct tcp_fastopen_cookie
*foc
)
820 const struct inet_request_sock
*ireq
= inet_rsk(req
);
825 /* First, grab a route. */
826 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
829 skb
= tcp_make_synack(sk
, dst
, req
, foc
);
832 __tcp_v4_send_check(skb
, ireq
->ir_loc_addr
, ireq
->ir_rmt_addr
);
834 skb_set_queue_mapping(skb
, queue_mapping
);
835 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->ir_loc_addr
,
838 err
= net_xmit_eval(err
);
845 * IPv4 request_sock destructor.
847 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
849 kfree(inet_rsk(req
)->opt
);
853 * Return true if a syncookie should be sent
855 bool tcp_syn_flood_action(struct sock
*sk
,
856 const struct sk_buff
*skb
,
859 const char *msg
= "Dropping request";
860 bool want_cookie
= false;
861 struct listen_sock
*lopt
;
863 #ifdef CONFIG_SYN_COOKIES
864 if (sysctl_tcp_syncookies
) {
865 msg
= "Sending cookies";
867 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
870 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
872 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
873 if (!lopt
->synflood_warned
&& sysctl_tcp_syncookies
!= 2) {
874 lopt
->synflood_warned
= 1;
875 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
876 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
880 EXPORT_SYMBOL(tcp_syn_flood_action
);
883 * Save and compile IPv4 options into the request_sock if needed.
885 static struct ip_options_rcu
*tcp_v4_save_options(struct sk_buff
*skb
)
887 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
888 struct ip_options_rcu
*dopt
= NULL
;
890 if (opt
&& opt
->optlen
) {
891 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
893 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
895 if (ip_options_echo(&dopt
->opt
, skb
)) {
904 #ifdef CONFIG_TCP_MD5SIG
906 * RFC2385 MD5 checksumming requires a mapping of
907 * IP address->MD5 Key.
908 * We need to maintain these in the sk structure.
911 /* Find the Key structure for an address. */
912 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
913 const union tcp_md5_addr
*addr
,
916 struct tcp_sock
*tp
= tcp_sk(sk
);
917 struct tcp_md5sig_key
*key
;
918 unsigned int size
= sizeof(struct in_addr
);
919 struct tcp_md5sig_info
*md5sig
;
921 /* caller either holds rcu_read_lock() or socket lock */
922 md5sig
= rcu_dereference_check(tp
->md5sig_info
,
923 sock_owned_by_user(sk
) ||
924 lockdep_is_held(&sk
->sk_lock
.slock
));
927 #if IS_ENABLED(CONFIG_IPV6)
928 if (family
== AF_INET6
)
929 size
= sizeof(struct in6_addr
);
931 hlist_for_each_entry_rcu(key
, &md5sig
->head
, node
) {
932 if (key
->family
!= family
)
934 if (!memcmp(&key
->addr
, addr
, size
))
939 EXPORT_SYMBOL(tcp_md5_do_lookup
);
941 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
942 struct sock
*addr_sk
)
944 union tcp_md5_addr
*addr
;
946 addr
= (union tcp_md5_addr
*)&inet_sk(addr_sk
)->inet_daddr
;
947 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
949 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
951 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
952 struct request_sock
*req
)
954 union tcp_md5_addr
*addr
;
956 addr
= (union tcp_md5_addr
*)&inet_rsk(req
)->ir_rmt_addr
;
957 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
960 /* This can be called on a newly created socket, from other files */
961 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
962 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
)
964 /* Add Key to the list */
965 struct tcp_md5sig_key
*key
;
966 struct tcp_sock
*tp
= tcp_sk(sk
);
967 struct tcp_md5sig_info
*md5sig
;
969 key
= tcp_md5_do_lookup(sk
, addr
, family
);
971 /* Pre-existing entry - just update that one. */
972 memcpy(key
->key
, newkey
, newkeylen
);
973 key
->keylen
= newkeylen
;
977 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
978 sock_owned_by_user(sk
));
980 md5sig
= kmalloc(sizeof(*md5sig
), gfp
);
984 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
985 INIT_HLIST_HEAD(&md5sig
->head
);
986 rcu_assign_pointer(tp
->md5sig_info
, md5sig
);
989 key
= sock_kmalloc(sk
, sizeof(*key
), gfp
);
992 if (!tcp_alloc_md5sig_pool()) {
993 sock_kfree_s(sk
, key
, sizeof(*key
));
997 memcpy(key
->key
, newkey
, newkeylen
);
998 key
->keylen
= newkeylen
;
999 key
->family
= family
;
1000 memcpy(&key
->addr
, addr
,
1001 (family
== AF_INET6
) ? sizeof(struct in6_addr
) :
1002 sizeof(struct in_addr
));
1003 hlist_add_head_rcu(&key
->node
, &md5sig
->head
);
1006 EXPORT_SYMBOL(tcp_md5_do_add
);
1008 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
, int family
)
1010 struct tcp_md5sig_key
*key
;
1012 key
= tcp_md5_do_lookup(sk
, addr
, family
);
1015 hlist_del_rcu(&key
->node
);
1016 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1017 kfree_rcu(key
, rcu
);
1020 EXPORT_SYMBOL(tcp_md5_do_del
);
1022 static void tcp_clear_md5_list(struct sock
*sk
)
1024 struct tcp_sock
*tp
= tcp_sk(sk
);
1025 struct tcp_md5sig_key
*key
;
1026 struct hlist_node
*n
;
1027 struct tcp_md5sig_info
*md5sig
;
1029 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
1031 hlist_for_each_entry_safe(key
, n
, &md5sig
->head
, node
) {
1032 hlist_del_rcu(&key
->node
);
1033 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1034 kfree_rcu(key
, rcu
);
1038 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1041 struct tcp_md5sig cmd
;
1042 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1044 if (optlen
< sizeof(cmd
))
1047 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1050 if (sin
->sin_family
!= AF_INET
)
1053 if (!cmd
.tcpm_keylen
)
1054 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1057 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1060 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1061 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1065 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1066 __be32 daddr
, __be32 saddr
, int nbytes
)
1068 struct tcp4_pseudohdr
*bp
;
1069 struct scatterlist sg
;
1071 bp
= &hp
->md5_blk
.ip4
;
1074 * 1. the TCP pseudo-header (in the order: source IP address,
1075 * destination IP address, zero-padded protocol number, and
1081 bp
->protocol
= IPPROTO_TCP
;
1082 bp
->len
= cpu_to_be16(nbytes
);
1084 sg_init_one(&sg
, bp
, sizeof(*bp
));
1085 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1088 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1089 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1091 struct tcp_md5sig_pool
*hp
;
1092 struct hash_desc
*desc
;
1094 hp
= tcp_get_md5sig_pool();
1096 goto clear_hash_noput
;
1097 desc
= &hp
->md5_desc
;
1099 if (crypto_hash_init(desc
))
1101 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1103 if (tcp_md5_hash_header(hp
, th
))
1105 if (tcp_md5_hash_key(hp
, key
))
1107 if (crypto_hash_final(desc
, md5_hash
))
1110 tcp_put_md5sig_pool();
1114 tcp_put_md5sig_pool();
1116 memset(md5_hash
, 0, 16);
1120 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1121 const struct sock
*sk
, const struct request_sock
*req
,
1122 const struct sk_buff
*skb
)
1124 struct tcp_md5sig_pool
*hp
;
1125 struct hash_desc
*desc
;
1126 const struct tcphdr
*th
= tcp_hdr(skb
);
1127 __be32 saddr
, daddr
;
1130 saddr
= inet_sk(sk
)->inet_saddr
;
1131 daddr
= inet_sk(sk
)->inet_daddr
;
1133 saddr
= inet_rsk(req
)->ir_loc_addr
;
1134 daddr
= inet_rsk(req
)->ir_rmt_addr
;
1136 const struct iphdr
*iph
= ip_hdr(skb
);
1141 hp
= tcp_get_md5sig_pool();
1143 goto clear_hash_noput
;
1144 desc
= &hp
->md5_desc
;
1146 if (crypto_hash_init(desc
))
1149 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1151 if (tcp_md5_hash_header(hp
, th
))
1153 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1155 if (tcp_md5_hash_key(hp
, key
))
1157 if (crypto_hash_final(desc
, md5_hash
))
1160 tcp_put_md5sig_pool();
1164 tcp_put_md5sig_pool();
1166 memset(md5_hash
, 0, 16);
1169 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1171 static bool __tcp_v4_inbound_md5_hash(struct sock
*sk
,
1172 const struct sk_buff
*skb
)
1175 * This gets called for each TCP segment that arrives
1176 * so we want to be efficient.
1177 * We have 3 drop cases:
1178 * o No MD5 hash and one expected.
1179 * o MD5 hash and we're not expecting one.
1180 * o MD5 hash and its wrong.
1182 const __u8
*hash_location
= NULL
;
1183 struct tcp_md5sig_key
*hash_expected
;
1184 const struct iphdr
*iph
= ip_hdr(skb
);
1185 const struct tcphdr
*th
= tcp_hdr(skb
);
1187 unsigned char newhash
[16];
1189 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1191 hash_location
= tcp_parse_md5sig_option(th
);
1193 /* We've parsed the options - do we have a hash? */
1194 if (!hash_expected
&& !hash_location
)
1197 if (hash_expected
&& !hash_location
) {
1198 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1202 if (!hash_expected
&& hash_location
) {
1203 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1207 /* Okay, so this is hash_expected and hash_location -
1208 * so we need to calculate the checksum.
1210 genhash
= tcp_v4_md5_hash_skb(newhash
,
1214 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1215 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1216 &iph
->saddr
, ntohs(th
->source
),
1217 &iph
->daddr
, ntohs(th
->dest
),
1218 genhash
? " tcp_v4_calc_md5_hash failed"
1225 static bool tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1230 ret
= __tcp_v4_inbound_md5_hash(sk
, skb
);
1238 static void tcp_v4_init_req(struct request_sock
*req
, struct sock
*sk
,
1239 struct sk_buff
*skb
)
1241 struct inet_request_sock
*ireq
= inet_rsk(req
);
1243 ireq
->ir_loc_addr
= ip_hdr(skb
)->daddr
;
1244 ireq
->ir_rmt_addr
= ip_hdr(skb
)->saddr
;
1245 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1246 ireq
->opt
= tcp_v4_save_options(skb
);
1249 static struct dst_entry
*tcp_v4_route_req(struct sock
*sk
, struct flowi
*fl
,
1250 const struct request_sock
*req
,
1253 struct dst_entry
*dst
= inet_csk_route_req(sk
, &fl
->u
.ip4
, req
);
1256 if (fl
->u
.ip4
.daddr
== inet_rsk(req
)->ir_rmt_addr
)
1265 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1267 .obj_size
= sizeof(struct tcp_request_sock
),
1268 .rtx_syn_ack
= tcp_rtx_synack
,
1269 .send_ack
= tcp_v4_reqsk_send_ack
,
1270 .destructor
= tcp_v4_reqsk_destructor
,
1271 .send_reset
= tcp_v4_send_reset
,
1272 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1275 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1276 .mss_clamp
= TCP_MSS_DEFAULT
,
1277 #ifdef CONFIG_TCP_MD5SIG
1278 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1279 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1281 .init_req
= tcp_v4_init_req
,
1282 #ifdef CONFIG_SYN_COOKIES
1283 .cookie_init_seq
= cookie_v4_init_sequence
,
1285 .route_req
= tcp_v4_route_req
,
1286 .init_seq
= tcp_v4_init_sequence
,
1287 .send_synack
= tcp_v4_send_synack
,
1288 .queue_hash_add
= inet_csk_reqsk_queue_hash_add
,
1291 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1293 /* Never answer to SYNs send to broadcast or multicast */
1294 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1297 return tcp_conn_request(&tcp_request_sock_ops
,
1298 &tcp_request_sock_ipv4_ops
, sk
, skb
);
1301 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1304 EXPORT_SYMBOL(tcp_v4_conn_request
);
1308 * The three way handshake has completed - we got a valid synack -
1309 * now create the new socket.
1311 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1312 struct request_sock
*req
,
1313 struct dst_entry
*dst
)
1315 struct inet_request_sock
*ireq
;
1316 struct inet_sock
*newinet
;
1317 struct tcp_sock
*newtp
;
1319 #ifdef CONFIG_TCP_MD5SIG
1320 struct tcp_md5sig_key
*key
;
1322 struct ip_options_rcu
*inet_opt
;
1324 if (sk_acceptq_is_full(sk
))
1327 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1331 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1332 inet_sk_rx_dst_set(newsk
, skb
);
1334 newtp
= tcp_sk(newsk
);
1335 newinet
= inet_sk(newsk
);
1336 ireq
= inet_rsk(req
);
1337 newinet
->inet_daddr
= ireq
->ir_rmt_addr
;
1338 newinet
->inet_rcv_saddr
= ireq
->ir_loc_addr
;
1339 newinet
->inet_saddr
= ireq
->ir_loc_addr
;
1340 inet_opt
= ireq
->opt
;
1341 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1343 newinet
->mc_index
= inet_iif(skb
);
1344 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1345 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1346 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1347 inet_set_txhash(newsk
);
1349 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1350 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1353 dst
= inet_csk_route_child_sock(sk
, newsk
, req
);
1357 /* syncookie case : see end of cookie_v4_check() */
1359 sk_setup_caps(newsk
, dst
);
1361 tcp_sync_mss(newsk
, dst_mtu(dst
));
1362 newtp
->advmss
= dst_metric_advmss(dst
);
1363 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1364 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1365 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1367 tcp_initialize_rcv_mss(newsk
);
1369 #ifdef CONFIG_TCP_MD5SIG
1370 /* Copy over the MD5 key from the original socket */
1371 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1375 * We're using one, so create a matching key
1376 * on the newsk structure. If we fail to get
1377 * memory, then we end up not copying the key
1380 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1381 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1382 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1386 if (__inet_inherit_port(sk
, newsk
) < 0)
1388 __inet_hash_nolisten(newsk
, NULL
);
1393 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1397 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1400 inet_csk_prepare_forced_close(newsk
);
1404 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1406 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1408 struct tcphdr
*th
= tcp_hdr(skb
);
1409 const struct iphdr
*iph
= ip_hdr(skb
);
1411 struct request_sock
**prev
;
1412 /* Find possible connection requests. */
1413 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1414 iph
->saddr
, iph
->daddr
);
1416 return tcp_check_req(sk
, skb
, req
, prev
, false);
1418 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1419 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1422 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1426 inet_twsk_put(inet_twsk(nsk
));
1430 #ifdef CONFIG_SYN_COOKIES
1432 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1437 /* The socket must have it's spinlock held when we get
1440 * We have a potential double-lock case here, so even when
1441 * doing backlog processing we use the BH locking scheme.
1442 * This is because we cannot sleep with the original spinlock
1445 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1449 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1450 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1452 sock_rps_save_rxhash(sk
, skb
);
1454 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1455 dst
->ops
->check(dst
, 0) == NULL
) {
1457 sk
->sk_rx_dst
= NULL
;
1460 tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
);
1464 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1467 if (sk
->sk_state
== TCP_LISTEN
) {
1468 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1473 sock_rps_save_rxhash(nsk
, skb
);
1474 if (tcp_child_process(sk
, nsk
, skb
)) {
1481 sock_rps_save_rxhash(sk
, skb
);
1483 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1490 tcp_v4_send_reset(rsk
, skb
);
1493 /* Be careful here. If this function gets more complicated and
1494 * gcc suffers from register pressure on the x86, sk (in %ebx)
1495 * might be destroyed here. This current version compiles correctly,
1496 * but you have been warned.
1501 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1502 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1505 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1507 void tcp_v4_early_demux(struct sk_buff
*skb
)
1509 const struct iphdr
*iph
;
1510 const struct tcphdr
*th
;
1513 if (skb
->pkt_type
!= PACKET_HOST
)
1516 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1522 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1525 sk
= __inet_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1526 iph
->saddr
, th
->source
,
1527 iph
->daddr
, ntohs(th
->dest
),
1531 skb
->destructor
= sock_edemux
;
1532 if (sk
->sk_state
!= TCP_TIME_WAIT
) {
1533 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1536 dst
= dst_check(dst
, 0);
1538 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1539 skb_dst_set_noref(skb
, dst
);
1544 /* Packet is added to VJ-style prequeue for processing in process
1545 * context, if a reader task is waiting. Apparently, this exciting
1546 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1547 * failed somewhere. Latency? Burstiness? Well, at least now we will
1548 * see, why it failed. 8)8) --ANK
1551 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1553 struct tcp_sock
*tp
= tcp_sk(sk
);
1555 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
1558 if (skb
->len
<= tcp_hdrlen(skb
) &&
1559 skb_queue_len(&tp
->ucopy
.prequeue
) == 0)
1562 /* Before escaping RCU protected region, we need to take care of skb
1563 * dst. Prequeue is only enabled for established sockets.
1564 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1565 * Instead of doing full sk_rx_dst validity here, let's perform
1566 * an optimistic check.
1568 if (likely(sk
->sk_rx_dst
))
1573 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1574 tp
->ucopy
.memory
+= skb
->truesize
;
1575 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1576 struct sk_buff
*skb1
;
1578 BUG_ON(sock_owned_by_user(sk
));
1580 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1581 sk_backlog_rcv(sk
, skb1
);
1582 NET_INC_STATS_BH(sock_net(sk
),
1583 LINUX_MIB_TCPPREQUEUEDROPPED
);
1586 tp
->ucopy
.memory
= 0;
1587 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1588 wake_up_interruptible_sync_poll(sk_sleep(sk
),
1589 POLLIN
| POLLRDNORM
| POLLRDBAND
);
1590 if (!inet_csk_ack_scheduled(sk
))
1591 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
1592 (3 * tcp_rto_min(sk
)) / 4,
1597 EXPORT_SYMBOL(tcp_prequeue
);
1603 int tcp_v4_rcv(struct sk_buff
*skb
)
1605 const struct iphdr
*iph
;
1606 const struct tcphdr
*th
;
1609 struct net
*net
= dev_net(skb
->dev
);
1611 if (skb
->pkt_type
!= PACKET_HOST
)
1614 /* Count it even if it's bad */
1615 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1617 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1622 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1624 if (!pskb_may_pull(skb
, th
->doff
* 4))
1627 /* An explanation is required here, I think.
1628 * Packet length and doff are validated by header prediction,
1629 * provided case of th->doff==0 is eliminated.
1630 * So, we defer the checks. */
1632 if (skb_checksum_init(skb
, IPPROTO_TCP
, inet_compute_pseudo
))
1637 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1638 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1639 skb
->len
- th
->doff
* 4);
1640 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1641 TCP_SKB_CB(skb
)->tcp_flags
= tcp_flag_byte(th
);
1642 TCP_SKB_CB(skb
)->tcp_tw_isn
= 0;
1643 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
1644 TCP_SKB_CB(skb
)->sacked
= 0;
1646 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1651 if (sk
->sk_state
== TCP_TIME_WAIT
)
1654 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1655 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1656 goto discard_and_relse
;
1659 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1660 goto discard_and_relse
;
1662 #ifdef CONFIG_TCP_MD5SIG
1664 * We really want to reject the packet as early as possible
1666 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1667 * o There is an MD5 option and we're not expecting one
1669 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1670 goto discard_and_relse
;
1675 if (sk_filter(sk
, skb
))
1676 goto discard_and_relse
;
1678 sk_mark_napi_id(sk
, skb
);
1681 bh_lock_sock_nested(sk
);
1683 if (!sock_owned_by_user(sk
)) {
1684 #ifdef CONFIG_NET_DMA
1685 struct tcp_sock
*tp
= tcp_sk(sk
);
1686 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1687 tp
->ucopy
.dma_chan
= net_dma_find_channel();
1688 if (tp
->ucopy
.dma_chan
)
1689 ret
= tcp_v4_do_rcv(sk
, skb
);
1693 if (!tcp_prequeue(sk
, skb
))
1694 ret
= tcp_v4_do_rcv(sk
, skb
);
1696 } else if (unlikely(sk_add_backlog(sk
, skb
,
1697 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
1699 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1700 goto discard_and_relse
;
1709 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1712 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1714 TCP_INC_STATS_BH(net
, TCP_MIB_CSUMERRORS
);
1716 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1718 tcp_v4_send_reset(NULL
, skb
);
1722 /* Discard frame. */
1731 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1732 inet_twsk_put(inet_twsk(sk
));
1736 if (skb
->len
< (th
->doff
<< 2)) {
1737 inet_twsk_put(inet_twsk(sk
));
1740 if (tcp_checksum_complete(skb
)) {
1741 inet_twsk_put(inet_twsk(sk
));
1744 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1746 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1748 iph
->saddr
, th
->source
,
1749 iph
->daddr
, th
->dest
,
1752 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1753 inet_twsk_put(inet_twsk(sk
));
1757 /* Fall through to ACK */
1760 tcp_v4_timewait_ack(sk
, skb
);
1764 case TCP_TW_SUCCESS
:;
1769 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1770 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1771 .twsk_unique
= tcp_twsk_unique
,
1772 .twsk_destructor
= tcp_twsk_destructor
,
1775 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
1777 struct dst_entry
*dst
= skb_dst(skb
);
1781 sk
->sk_rx_dst
= dst
;
1782 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
1785 EXPORT_SYMBOL(inet_sk_rx_dst_set
);
1787 const struct inet_connection_sock_af_ops ipv4_specific
= {
1788 .queue_xmit
= ip_queue_xmit
,
1789 .send_check
= tcp_v4_send_check
,
1790 .rebuild_header
= inet_sk_rebuild_header
,
1791 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
1792 .conn_request
= tcp_v4_conn_request
,
1793 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1794 .net_header_len
= sizeof(struct iphdr
),
1795 .setsockopt
= ip_setsockopt
,
1796 .getsockopt
= ip_getsockopt
,
1797 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1798 .sockaddr_len
= sizeof(struct sockaddr_in
),
1799 .bind_conflict
= inet_csk_bind_conflict
,
1800 #ifdef CONFIG_COMPAT
1801 .compat_setsockopt
= compat_ip_setsockopt
,
1802 .compat_getsockopt
= compat_ip_getsockopt
,
1804 .mtu_reduced
= tcp_v4_mtu_reduced
,
1806 EXPORT_SYMBOL(ipv4_specific
);
1808 #ifdef CONFIG_TCP_MD5SIG
1809 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1810 .md5_lookup
= tcp_v4_md5_lookup
,
1811 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1812 .md5_parse
= tcp_v4_parse_md5_keys
,
1816 /* NOTE: A lot of things set to zero explicitly by call to
1817 * sk_alloc() so need not be done here.
1819 static int tcp_v4_init_sock(struct sock
*sk
)
1821 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1825 icsk
->icsk_af_ops
= &ipv4_specific
;
1827 #ifdef CONFIG_TCP_MD5SIG
1828 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv4_specific
;
1834 void tcp_v4_destroy_sock(struct sock
*sk
)
1836 struct tcp_sock
*tp
= tcp_sk(sk
);
1838 tcp_clear_xmit_timers(sk
);
1840 tcp_cleanup_congestion_control(sk
);
1842 /* Cleanup up the write buffer. */
1843 tcp_write_queue_purge(sk
);
1845 /* Cleans up our, hopefully empty, out_of_order_queue. */
1846 __skb_queue_purge(&tp
->out_of_order_queue
);
1848 #ifdef CONFIG_TCP_MD5SIG
1849 /* Clean up the MD5 key list, if any */
1850 if (tp
->md5sig_info
) {
1851 tcp_clear_md5_list(sk
);
1852 kfree_rcu(tp
->md5sig_info
, rcu
);
1853 tp
->md5sig_info
= NULL
;
1857 #ifdef CONFIG_NET_DMA
1858 /* Cleans up our sk_async_wait_queue */
1859 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1862 /* Clean prequeue, it must be empty really */
1863 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1865 /* Clean up a referenced TCP bind bucket. */
1866 if (inet_csk(sk
)->icsk_bind_hash
)
1869 BUG_ON(tp
->fastopen_rsk
!= NULL
);
1871 /* If socket is aborted during connect operation */
1872 tcp_free_fastopen_req(tp
);
1874 sk_sockets_allocated_dec(sk
);
1875 sock_release_memcg(sk
);
1877 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1879 #ifdef CONFIG_PROC_FS
1880 /* Proc filesystem TCP sock list dumping. */
1883 * Get next listener socket follow cur. If cur is NULL, get first socket
1884 * starting from bucket given in st->bucket; when st->bucket is zero the
1885 * very first socket in the hash table is returned.
1887 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
1889 struct inet_connection_sock
*icsk
;
1890 struct hlist_nulls_node
*node
;
1891 struct sock
*sk
= cur
;
1892 struct inet_listen_hashbucket
*ilb
;
1893 struct tcp_iter_state
*st
= seq
->private;
1894 struct net
*net
= seq_file_net(seq
);
1897 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1898 spin_lock_bh(&ilb
->lock
);
1899 sk
= sk_nulls_head(&ilb
->head
);
1903 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1907 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
1908 struct request_sock
*req
= cur
;
1910 icsk
= inet_csk(st
->syn_wait_sk
);
1914 if (req
->rsk_ops
->family
== st
->family
) {
1920 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
1923 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
1925 sk
= sk_nulls_next(st
->syn_wait_sk
);
1926 st
->state
= TCP_SEQ_STATE_LISTENING
;
1927 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1929 icsk
= inet_csk(sk
);
1930 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1931 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
1933 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1934 sk
= sk_nulls_next(sk
);
1937 sk_nulls_for_each_from(sk
, node
) {
1938 if (!net_eq(sock_net(sk
), net
))
1940 if (sk
->sk_family
== st
->family
) {
1944 icsk
= inet_csk(sk
);
1945 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1946 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
1948 st
->uid
= sock_i_uid(sk
);
1949 st
->syn_wait_sk
= sk
;
1950 st
->state
= TCP_SEQ_STATE_OPENREQ
;
1954 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1956 spin_unlock_bh(&ilb
->lock
);
1958 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
1959 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1960 spin_lock_bh(&ilb
->lock
);
1961 sk
= sk_nulls_head(&ilb
->head
);
1969 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
1971 struct tcp_iter_state
*st
= seq
->private;
1976 rc
= listening_get_next(seq
, NULL
);
1978 while (rc
&& *pos
) {
1979 rc
= listening_get_next(seq
, rc
);
1985 static inline bool empty_bucket(const struct tcp_iter_state
*st
)
1987 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
1991 * Get first established socket starting from bucket given in st->bucket.
1992 * If st->bucket is zero, the very first socket in the hash is returned.
1994 static void *established_get_first(struct seq_file
*seq
)
1996 struct tcp_iter_state
*st
= seq
->private;
1997 struct net
*net
= seq_file_net(seq
);
2001 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2003 struct hlist_nulls_node
*node
;
2004 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2006 /* Lockless fast path for the common case of empty buckets */
2007 if (empty_bucket(st
))
2011 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2012 if (sk
->sk_family
!= st
->family
||
2013 !net_eq(sock_net(sk
), net
)) {
2019 spin_unlock_bh(lock
);
2025 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2027 struct sock
*sk
= cur
;
2028 struct hlist_nulls_node
*node
;
2029 struct tcp_iter_state
*st
= seq
->private;
2030 struct net
*net
= seq_file_net(seq
);
2035 sk
= sk_nulls_next(sk
);
2037 sk_nulls_for_each_from(sk
, node
) {
2038 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2042 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2044 return established_get_first(seq
);
2047 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2049 struct tcp_iter_state
*st
= seq
->private;
2053 rc
= established_get_first(seq
);
2056 rc
= established_get_next(seq
, rc
);
2062 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2065 struct tcp_iter_state
*st
= seq
->private;
2067 st
->state
= TCP_SEQ_STATE_LISTENING
;
2068 rc
= listening_get_idx(seq
, &pos
);
2071 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2072 rc
= established_get_idx(seq
, pos
);
2078 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2080 struct tcp_iter_state
*st
= seq
->private;
2081 int offset
= st
->offset
;
2082 int orig_num
= st
->num
;
2085 switch (st
->state
) {
2086 case TCP_SEQ_STATE_OPENREQ
:
2087 case TCP_SEQ_STATE_LISTENING
:
2088 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2090 st
->state
= TCP_SEQ_STATE_LISTENING
;
2091 rc
= listening_get_next(seq
, NULL
);
2092 while (offset
-- && rc
)
2093 rc
= listening_get_next(seq
, rc
);
2097 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2099 case TCP_SEQ_STATE_ESTABLISHED
:
2100 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2102 rc
= established_get_first(seq
);
2103 while (offset
-- && rc
)
2104 rc
= established_get_next(seq
, rc
);
2112 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2114 struct tcp_iter_state
*st
= seq
->private;
2117 if (*pos
&& *pos
== st
->last_pos
) {
2118 rc
= tcp_seek_last_pos(seq
);
2123 st
->state
= TCP_SEQ_STATE_LISTENING
;
2127 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2130 st
->last_pos
= *pos
;
2134 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2136 struct tcp_iter_state
*st
= seq
->private;
2139 if (v
== SEQ_START_TOKEN
) {
2140 rc
= tcp_get_idx(seq
, 0);
2144 switch (st
->state
) {
2145 case TCP_SEQ_STATE_OPENREQ
:
2146 case TCP_SEQ_STATE_LISTENING
:
2147 rc
= listening_get_next(seq
, v
);
2149 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2152 rc
= established_get_first(seq
);
2155 case TCP_SEQ_STATE_ESTABLISHED
:
2156 rc
= established_get_next(seq
, v
);
2161 st
->last_pos
= *pos
;
2165 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2167 struct tcp_iter_state
*st
= seq
->private;
2169 switch (st
->state
) {
2170 case TCP_SEQ_STATE_OPENREQ
:
2172 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2173 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2175 case TCP_SEQ_STATE_LISTENING
:
2176 if (v
!= SEQ_START_TOKEN
)
2177 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2179 case TCP_SEQ_STATE_ESTABLISHED
:
2181 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2186 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2188 struct tcp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2189 struct tcp_iter_state
*s
;
2192 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2193 sizeof(struct tcp_iter_state
));
2197 s
= ((struct seq_file
*)file
->private_data
)->private;
2198 s
->family
= afinfo
->family
;
2202 EXPORT_SYMBOL(tcp_seq_open
);
2204 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2207 struct proc_dir_entry
*p
;
2209 afinfo
->seq_ops
.start
= tcp_seq_start
;
2210 afinfo
->seq_ops
.next
= tcp_seq_next
;
2211 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2213 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2214 afinfo
->seq_fops
, afinfo
);
2219 EXPORT_SYMBOL(tcp_proc_register
);
2221 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2223 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2225 EXPORT_SYMBOL(tcp_proc_unregister
);
2227 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2228 struct seq_file
*f
, int i
, kuid_t uid
)
2230 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2231 long delta
= req
->expires
- jiffies
;
2233 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2234 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2237 ntohs(inet_sk(sk
)->inet_sport
),
2239 ntohs(ireq
->ir_rmt_port
),
2241 0, 0, /* could print option size, but that is af dependent. */
2242 1, /* timers active (only the expire timer) */
2243 jiffies_delta_to_clock_t(delta
),
2245 from_kuid_munged(seq_user_ns(f
), uid
),
2246 0, /* non standard timer */
2247 0, /* open_requests have no inode */
2248 atomic_read(&sk
->sk_refcnt
),
2252 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
)
2255 unsigned long timer_expires
;
2256 const struct tcp_sock
*tp
= tcp_sk(sk
);
2257 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2258 const struct inet_sock
*inet
= inet_sk(sk
);
2259 struct fastopen_queue
*fastopenq
= icsk
->icsk_accept_queue
.fastopenq
;
2260 __be32 dest
= inet
->inet_daddr
;
2261 __be32 src
= inet
->inet_rcv_saddr
;
2262 __u16 destp
= ntohs(inet
->inet_dport
);
2263 __u16 srcp
= ntohs(inet
->inet_sport
);
2266 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
2267 icsk
->icsk_pending
== ICSK_TIME_EARLY_RETRANS
||
2268 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
2270 timer_expires
= icsk
->icsk_timeout
;
2271 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2273 timer_expires
= icsk
->icsk_timeout
;
2274 } else if (timer_pending(&sk
->sk_timer
)) {
2276 timer_expires
= sk
->sk_timer
.expires
;
2279 timer_expires
= jiffies
;
2282 if (sk
->sk_state
== TCP_LISTEN
)
2283 rx_queue
= sk
->sk_ack_backlog
;
2286 * because we dont lock socket, we might find a transient negative value
2288 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2290 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2291 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2292 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2293 tp
->write_seq
- tp
->snd_una
,
2296 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
2297 icsk
->icsk_retransmits
,
2298 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sk
)),
2299 icsk
->icsk_probes_out
,
2301 atomic_read(&sk
->sk_refcnt
), sk
,
2302 jiffies_to_clock_t(icsk
->icsk_rto
),
2303 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2304 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2306 sk
->sk_state
== TCP_LISTEN
?
2307 (fastopenq
? fastopenq
->max_qlen
: 0) :
2308 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
));
2311 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2312 struct seq_file
*f
, int i
)
2316 s32 delta
= tw
->tw_ttd
- inet_tw_time_stamp();
2318 dest
= tw
->tw_daddr
;
2319 src
= tw
->tw_rcv_saddr
;
2320 destp
= ntohs(tw
->tw_dport
);
2321 srcp
= ntohs(tw
->tw_sport
);
2323 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2324 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2325 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2326 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
2327 atomic_read(&tw
->tw_refcnt
), tw
);
2332 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2334 struct tcp_iter_state
*st
;
2335 struct sock
*sk
= v
;
2337 seq_setwidth(seq
, TMPSZ
- 1);
2338 if (v
== SEQ_START_TOKEN
) {
2339 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2340 "rx_queue tr tm->when retrnsmt uid timeout "
2346 switch (st
->state
) {
2347 case TCP_SEQ_STATE_LISTENING
:
2348 case TCP_SEQ_STATE_ESTABLISHED
:
2349 if (sk
->sk_state
== TCP_TIME_WAIT
)
2350 get_timewait4_sock(v
, seq
, st
->num
);
2352 get_tcp4_sock(v
, seq
, st
->num
);
2354 case TCP_SEQ_STATE_OPENREQ
:
2355 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
);
2363 static const struct file_operations tcp_afinfo_seq_fops
= {
2364 .owner
= THIS_MODULE
,
2365 .open
= tcp_seq_open
,
2367 .llseek
= seq_lseek
,
2368 .release
= seq_release_net
2371 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2374 .seq_fops
= &tcp_afinfo_seq_fops
,
2376 .show
= tcp4_seq_show
,
2380 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2382 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2385 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2387 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2390 static struct pernet_operations tcp4_net_ops
= {
2391 .init
= tcp4_proc_init_net
,
2392 .exit
= tcp4_proc_exit_net
,
2395 int __init
tcp4_proc_init(void)
2397 return register_pernet_subsys(&tcp4_net_ops
);
2400 void tcp4_proc_exit(void)
2402 unregister_pernet_subsys(&tcp4_net_ops
);
2404 #endif /* CONFIG_PROC_FS */
2406 struct proto tcp_prot
= {
2408 .owner
= THIS_MODULE
,
2410 .connect
= tcp_v4_connect
,
2411 .disconnect
= tcp_disconnect
,
2412 .accept
= inet_csk_accept
,
2414 .init
= tcp_v4_init_sock
,
2415 .destroy
= tcp_v4_destroy_sock
,
2416 .shutdown
= tcp_shutdown
,
2417 .setsockopt
= tcp_setsockopt
,
2418 .getsockopt
= tcp_getsockopt
,
2419 .recvmsg
= tcp_recvmsg
,
2420 .sendmsg
= tcp_sendmsg
,
2421 .sendpage
= tcp_sendpage
,
2422 .backlog_rcv
= tcp_v4_do_rcv
,
2423 .release_cb
= tcp_release_cb
,
2425 .unhash
= inet_unhash
,
2426 .get_port
= inet_csk_get_port
,
2427 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2428 .stream_memory_free
= tcp_stream_memory_free
,
2429 .sockets_allocated
= &tcp_sockets_allocated
,
2430 .orphan_count
= &tcp_orphan_count
,
2431 .memory_allocated
= &tcp_memory_allocated
,
2432 .memory_pressure
= &tcp_memory_pressure
,
2433 .sysctl_mem
= sysctl_tcp_mem
,
2434 .sysctl_wmem
= sysctl_tcp_wmem
,
2435 .sysctl_rmem
= sysctl_tcp_rmem
,
2436 .max_header
= MAX_TCP_HEADER
,
2437 .obj_size
= sizeof(struct tcp_sock
),
2438 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2439 .twsk_prot
= &tcp_timewait_sock_ops
,
2440 .rsk_prot
= &tcp_request_sock_ops
,
2441 .h
.hashinfo
= &tcp_hashinfo
,
2442 .no_autobind
= true,
2443 #ifdef CONFIG_COMPAT
2444 .compat_setsockopt
= compat_tcp_setsockopt
,
2445 .compat_getsockopt
= compat_tcp_getsockopt
,
2447 #ifdef CONFIG_MEMCG_KMEM
2448 .init_cgroup
= tcp_init_cgroup
,
2449 .destroy_cgroup
= tcp_destroy_cgroup
,
2450 .proto_cgroup
= tcp_proto_cgroup
,
2453 EXPORT_SYMBOL(tcp_prot
);
2455 static int __net_init
tcp_sk_init(struct net
*net
)
2457 net
->ipv4
.sysctl_tcp_ecn
= 2;
2461 static void __net_exit
tcp_sk_exit(struct net
*net
)
2465 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2467 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2470 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2471 .init
= tcp_sk_init
,
2472 .exit
= tcp_sk_exit
,
2473 .exit_batch
= tcp_sk_exit_batch
,
2476 void __init
tcp_v4_init(void)
2478 inet_hashinfo_init(&tcp_hashinfo
);
2479 if (register_pernet_subsys(&tcp_sk_ops
))
2480 panic("Failed to create the TCP control socket.\n");