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
);
94 #ifdef CONFIG_TCP_MD5SIG
95 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
96 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
);
99 struct inet_hashinfo tcp_hashinfo
;
100 EXPORT_SYMBOL(tcp_hashinfo
);
102 static __u32
tcp_v4_init_sequence(const struct sk_buff
*skb
)
104 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
107 tcp_hdr(skb
)->source
);
110 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
)
112 const struct tcp_timewait_sock
*tcptw
= tcp_twsk(sktw
);
113 struct tcp_sock
*tp
= tcp_sk(sk
);
115 /* With PAWS, it is safe from the viewpoint
116 of data integrity. Even without PAWS it is safe provided sequence
117 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
119 Actually, the idea is close to VJ's one, only timestamp cache is
120 held not per host, but per port pair and TW bucket is used as state
123 If TW bucket has been already destroyed we fall back to VJ's scheme
124 and use initial timestamp retrieved from peer table.
126 if (tcptw
->tw_ts_recent_stamp
&&
127 (twp
== NULL
|| (sysctl_tcp_tw_reuse
&&
128 get_seconds() - tcptw
->tw_ts_recent_stamp
> 1))) {
129 tp
->write_seq
= tcptw
->tw_snd_nxt
+ 65535 + 2;
130 if (tp
->write_seq
== 0)
132 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
133 tp
->rx_opt
.ts_recent_stamp
= tcptw
->tw_ts_recent_stamp
;
140 EXPORT_SYMBOL_GPL(tcp_twsk_unique
);
142 /* This will initiate an outgoing connection. */
143 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
145 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
146 struct inet_sock
*inet
= inet_sk(sk
);
147 struct tcp_sock
*tp
= tcp_sk(sk
);
148 __be16 orig_sport
, orig_dport
;
149 __be32 daddr
, nexthop
;
153 struct ip_options_rcu
*inet_opt
;
155 if (addr_len
< sizeof(struct sockaddr_in
))
158 if (usin
->sin_family
!= AF_INET
)
159 return -EAFNOSUPPORT
;
161 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
162 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
163 sock_owned_by_user(sk
));
164 if (inet_opt
&& inet_opt
->opt
.srr
) {
167 nexthop
= inet_opt
->opt
.faddr
;
170 orig_sport
= inet
->inet_sport
;
171 orig_dport
= usin
->sin_port
;
172 fl4
= &inet
->cork
.fl
.u
.ip4
;
173 rt
= ip_route_connect(fl4
, nexthop
, inet
->inet_saddr
,
174 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
176 orig_sport
, orig_dport
, sk
);
179 if (err
== -ENETUNREACH
)
180 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
184 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
189 if (!inet_opt
|| !inet_opt
->opt
.srr
)
192 if (!inet
->inet_saddr
)
193 inet
->inet_saddr
= fl4
->saddr
;
194 inet
->inet_rcv_saddr
= inet
->inet_saddr
;
196 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->inet_daddr
!= daddr
) {
197 /* Reset inherited state */
198 tp
->rx_opt
.ts_recent
= 0;
199 tp
->rx_opt
.ts_recent_stamp
= 0;
200 if (likely(!tp
->repair
))
204 if (tcp_death_row
.sysctl_tw_recycle
&&
205 !tp
->rx_opt
.ts_recent_stamp
&& fl4
->daddr
== daddr
)
206 tcp_fetch_timewait_stamp(sk
, &rt
->dst
);
208 inet
->inet_dport
= usin
->sin_port
;
209 inet
->inet_daddr
= daddr
;
213 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
215 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
217 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
219 /* Socket identity is still unknown (sport may be zero).
220 * However we set state to SYN-SENT and not releasing socket
221 * lock select source port, enter ourselves into the hash tables and
222 * complete initialization after this.
224 tcp_set_state(sk
, TCP_SYN_SENT
);
225 err
= inet_hash_connect(&tcp_death_row
, sk
);
229 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
230 inet
->inet_sport
, inet
->inet_dport
, sk
);
236 /* OK, now commit destination to socket. */
237 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
238 sk_setup_caps(sk
, &rt
->dst
);
240 if (!tp
->write_seq
&& likely(!tp
->repair
))
241 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
246 inet
->inet_id
= tp
->write_seq
^ jiffies
;
248 err
= tcp_connect(sk
);
258 * This unhashes the socket and releases the local port,
261 tcp_set_state(sk
, TCP_CLOSE
);
263 sk
->sk_route_caps
= 0;
264 inet
->inet_dport
= 0;
267 EXPORT_SYMBOL(tcp_v4_connect
);
270 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
271 * It can be called through tcp_release_cb() if socket was owned by user
272 * at the time tcp_v4_err() was called to handle ICMP message.
274 static void tcp_v4_mtu_reduced(struct sock
*sk
)
276 struct dst_entry
*dst
;
277 struct inet_sock
*inet
= inet_sk(sk
);
278 u32 mtu
= tcp_sk(sk
)->mtu_info
;
280 dst
= inet_csk_update_pmtu(sk
, mtu
);
284 /* Something is about to be wrong... Remember soft error
285 * for the case, if this connection will not able to recover.
287 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
288 sk
->sk_err_soft
= EMSGSIZE
;
292 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
293 ip_sk_accept_pmtu(sk
) &&
294 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
295 tcp_sync_mss(sk
, mtu
);
297 /* Resend the TCP packet because it's
298 * clear that the old packet has been
299 * dropped. This is the new "fast" path mtu
302 tcp_simple_retransmit(sk
);
303 } /* else let the usual retransmit timer handle it */
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 inet_csk(sk
)->icsk_rto
= (tp
->srtt_us
? __tcp_set_rto(tp
) :
434 TCP_TIMEOUT_INIT
) << icsk
->icsk_backoff
;
437 skb
= tcp_write_queue_head(sk
);
440 remaining
= icsk
->icsk_rto
- min(icsk
->icsk_rto
,
441 tcp_time_stamp
- TCP_SKB_CB(skb
)->when
);
444 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
445 remaining
, TCP_RTO_MAX
);
447 /* RTO revert clocked out retransmission.
448 * Will retransmit now */
449 tcp_retransmit_timer(sk
);
453 case ICMP_TIME_EXCEEDED
:
460 switch (sk
->sk_state
) {
461 struct request_sock
*req
, **prev
;
463 if (sock_owned_by_user(sk
))
466 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
467 iph
->daddr
, iph
->saddr
);
471 /* ICMPs are not backlogged, hence we cannot get
472 an established socket here.
476 if (seq
!= tcp_rsk(req
)->snt_isn
) {
477 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
482 * Still in SYN_RECV, just remove it silently.
483 * There is no good way to pass the error to the newly
484 * created socket, and POSIX does not want network
485 * errors returned from accept().
487 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
488 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
493 /* Only in fast or simultaneous open. If a fast open socket is
494 * is already accepted it is treated as a connected one below.
496 if (fastopen
&& fastopen
->sk
== NULL
)
499 if (!sock_owned_by_user(sk
)) {
502 sk
->sk_error_report(sk
);
506 sk
->sk_err_soft
= err
;
511 /* If we've already connected we will keep trying
512 * until we time out, or the user gives up.
514 * rfc1122 4.2.3.9 allows to consider as hard errors
515 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
516 * but it is obsoleted by pmtu discovery).
518 * Note, that in modern internet, where routing is unreliable
519 * and in each dark corner broken firewalls sit, sending random
520 * errors ordered by their masters even this two messages finally lose
521 * their original sense (even Linux sends invalid PORT_UNREACHs)
523 * Now we are in compliance with RFCs.
528 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
530 sk
->sk_error_report(sk
);
531 } else { /* Only an error on timeout */
532 sk
->sk_err_soft
= err
;
540 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
)
542 struct tcphdr
*th
= tcp_hdr(skb
);
544 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
545 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
546 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
547 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
549 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
556 /* This routine computes an IPv4 TCP checksum. */
557 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
559 const struct inet_sock
*inet
= inet_sk(sk
);
561 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
563 EXPORT_SYMBOL(tcp_v4_send_check
);
566 * This routine will send an RST to the other tcp.
568 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
570 * Answer: if a packet caused RST, it is not for a socket
571 * existing in our system, if it is matched to a socket,
572 * it is just duplicate segment or bug in other side's TCP.
573 * So that we build reply only basing on parameters
574 * arrived with segment.
575 * Exception: precedence violation. We do not implement it in any case.
578 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
580 const struct tcphdr
*th
= tcp_hdr(skb
);
583 #ifdef CONFIG_TCP_MD5SIG
584 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
587 struct ip_reply_arg arg
;
588 #ifdef CONFIG_TCP_MD5SIG
589 struct tcp_md5sig_key
*key
;
590 const __u8
*hash_location
= NULL
;
591 unsigned char newhash
[16];
593 struct sock
*sk1
= NULL
;
597 /* Never send a reset in response to a reset. */
601 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
604 /* Swap the send and the receive. */
605 memset(&rep
, 0, sizeof(rep
));
606 rep
.th
.dest
= th
->source
;
607 rep
.th
.source
= th
->dest
;
608 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
612 rep
.th
.seq
= th
->ack_seq
;
615 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
616 skb
->len
- (th
->doff
<< 2));
619 memset(&arg
, 0, sizeof(arg
));
620 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
621 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
623 #ifdef CONFIG_TCP_MD5SIG
624 hash_location
= tcp_parse_md5sig_option(th
);
625 if (!sk
&& hash_location
) {
627 * active side is lost. Try to find listening socket through
628 * source port, and then find md5 key through listening socket.
629 * we are not loose security here:
630 * Incoming packet is checked with md5 hash with finding key,
631 * no RST generated if md5 hash doesn't match.
633 sk1
= __inet_lookup_listener(dev_net(skb_dst(skb
)->dev
),
634 &tcp_hashinfo
, ip_hdr(skb
)->saddr
,
635 th
->source
, ip_hdr(skb
)->daddr
,
636 ntohs(th
->source
), inet_iif(skb
));
637 /* don't send rst if it can't find key */
641 key
= tcp_md5_do_lookup(sk1
, (union tcp_md5_addr
*)
642 &ip_hdr(skb
)->saddr
, AF_INET
);
646 genhash
= tcp_v4_md5_hash_skb(newhash
, key
, NULL
, NULL
, skb
);
647 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
650 key
= sk
? tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)
656 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
658 (TCPOPT_MD5SIG
<< 8) |
660 /* Update length and the length the header thinks exists */
661 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
662 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
664 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
665 key
, ip_hdr(skb
)->saddr
,
666 ip_hdr(skb
)->daddr
, &rep
.th
);
669 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
670 ip_hdr(skb
)->saddr
, /* XXX */
671 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
672 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
673 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
674 /* When socket is gone, all binding information is lost.
675 * routing might fail in this case. No choice here, if we choose to force
676 * input interface, we will misroute in case of asymmetric route.
679 arg
.bound_dev_if
= sk
->sk_bound_dev_if
;
681 net
= dev_net(skb_dst(skb
)->dev
);
682 arg
.tos
= ip_hdr(skb
)->tos
;
683 ip_send_unicast_reply(net
, skb
, ip_hdr(skb
)->saddr
,
684 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
686 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
687 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
689 #ifdef CONFIG_TCP_MD5SIG
698 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
699 outside socket context is ugly, certainly. What can I do?
702 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
703 u32 win
, u32 tsval
, u32 tsecr
, int oif
,
704 struct tcp_md5sig_key
*key
,
705 int reply_flags
, u8 tos
)
707 const struct tcphdr
*th
= tcp_hdr(skb
);
710 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
711 #ifdef CONFIG_TCP_MD5SIG
712 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
716 struct ip_reply_arg arg
;
717 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
719 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
720 memset(&arg
, 0, sizeof(arg
));
722 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
723 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
725 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
726 (TCPOPT_TIMESTAMP
<< 8) |
728 rep
.opt
[1] = htonl(tsval
);
729 rep
.opt
[2] = htonl(tsecr
);
730 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
733 /* Swap the send and the receive. */
734 rep
.th
.dest
= th
->source
;
735 rep
.th
.source
= th
->dest
;
736 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
737 rep
.th
.seq
= htonl(seq
);
738 rep
.th
.ack_seq
= htonl(ack
);
740 rep
.th
.window
= htons(win
);
742 #ifdef CONFIG_TCP_MD5SIG
744 int offset
= (tsecr
) ? 3 : 0;
746 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
748 (TCPOPT_MD5SIG
<< 8) |
750 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
751 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
753 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
754 key
, ip_hdr(skb
)->saddr
,
755 ip_hdr(skb
)->daddr
, &rep
.th
);
758 arg
.flags
= reply_flags
;
759 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
760 ip_hdr(skb
)->saddr
, /* XXX */
761 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
762 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
764 arg
.bound_dev_if
= oif
;
766 ip_send_unicast_reply(net
, skb
, ip_hdr(skb
)->saddr
,
767 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
769 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
772 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
774 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
775 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
777 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
778 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
779 tcp_time_stamp
+ tcptw
->tw_ts_offset
,
782 tcp_twsk_md5_key(tcptw
),
783 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
790 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
791 struct request_sock
*req
)
793 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
794 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
796 tcp_v4_send_ack(skb
, (sk
->sk_state
== TCP_LISTEN
) ?
797 tcp_rsk(req
)->snt_isn
+ 1 : tcp_sk(sk
)->snd_nxt
,
798 tcp_rsk(req
)->rcv_nxt
, req
->rcv_wnd
,
802 tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&ip_hdr(skb
)->daddr
,
804 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
809 * Send a SYN-ACK after having received a SYN.
810 * This still operates on a request_sock only, not on a big
813 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
815 struct request_sock
*req
,
817 struct tcp_fastopen_cookie
*foc
)
819 const struct inet_request_sock
*ireq
= inet_rsk(req
);
824 /* First, grab a route. */
825 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
828 skb
= tcp_make_synack(sk
, dst
, req
, foc
);
831 __tcp_v4_send_check(skb
, ireq
->ir_loc_addr
, ireq
->ir_rmt_addr
);
833 skb_set_queue_mapping(skb
, queue_mapping
);
834 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->ir_loc_addr
,
837 err
= net_xmit_eval(err
);
844 * IPv4 request_sock destructor.
846 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
848 kfree(inet_rsk(req
)->opt
);
852 * Return true if a syncookie should be sent
854 bool tcp_syn_flood_action(struct sock
*sk
,
855 const struct sk_buff
*skb
,
858 const char *msg
= "Dropping request";
859 bool want_cookie
= false;
860 struct listen_sock
*lopt
;
862 #ifdef CONFIG_SYN_COOKIES
863 if (sysctl_tcp_syncookies
) {
864 msg
= "Sending cookies";
866 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
869 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
871 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
872 if (!lopt
->synflood_warned
&& sysctl_tcp_syncookies
!= 2) {
873 lopt
->synflood_warned
= 1;
874 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
875 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
879 EXPORT_SYMBOL(tcp_syn_flood_action
);
882 * Save and compile IPv4 options into the request_sock if needed.
884 static struct ip_options_rcu
*tcp_v4_save_options(struct sk_buff
*skb
)
886 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
887 struct ip_options_rcu
*dopt
= NULL
;
889 if (opt
&& opt
->optlen
) {
890 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
892 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
894 if (ip_options_echo(&dopt
->opt
, skb
)) {
903 #ifdef CONFIG_TCP_MD5SIG
905 * RFC2385 MD5 checksumming requires a mapping of
906 * IP address->MD5 Key.
907 * We need to maintain these in the sk structure.
910 /* Find the Key structure for an address. */
911 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
912 const union tcp_md5_addr
*addr
,
915 struct tcp_sock
*tp
= tcp_sk(sk
);
916 struct tcp_md5sig_key
*key
;
917 unsigned int size
= sizeof(struct in_addr
);
918 struct tcp_md5sig_info
*md5sig
;
920 /* caller either holds rcu_read_lock() or socket lock */
921 md5sig
= rcu_dereference_check(tp
->md5sig_info
,
922 sock_owned_by_user(sk
) ||
923 lockdep_is_held(&sk
->sk_lock
.slock
));
926 #if IS_ENABLED(CONFIG_IPV6)
927 if (family
== AF_INET6
)
928 size
= sizeof(struct in6_addr
);
930 hlist_for_each_entry_rcu(key
, &md5sig
->head
, node
) {
931 if (key
->family
!= family
)
933 if (!memcmp(&key
->addr
, addr
, size
))
938 EXPORT_SYMBOL(tcp_md5_do_lookup
);
940 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
941 struct sock
*addr_sk
)
943 union tcp_md5_addr
*addr
;
945 addr
= (union tcp_md5_addr
*)&inet_sk(addr_sk
)->inet_daddr
;
946 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
948 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
950 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
951 struct request_sock
*req
)
953 union tcp_md5_addr
*addr
;
955 addr
= (union tcp_md5_addr
*)&inet_rsk(req
)->ir_rmt_addr
;
956 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
959 /* This can be called on a newly created socket, from other files */
960 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
961 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
)
963 /* Add Key to the list */
964 struct tcp_md5sig_key
*key
;
965 struct tcp_sock
*tp
= tcp_sk(sk
);
966 struct tcp_md5sig_info
*md5sig
;
968 key
= tcp_md5_do_lookup(sk
, addr
, family
);
970 /* Pre-existing entry - just update that one. */
971 memcpy(key
->key
, newkey
, newkeylen
);
972 key
->keylen
= newkeylen
;
976 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
977 sock_owned_by_user(sk
));
979 md5sig
= kmalloc(sizeof(*md5sig
), gfp
);
983 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
984 INIT_HLIST_HEAD(&md5sig
->head
);
985 rcu_assign_pointer(tp
->md5sig_info
, md5sig
);
988 key
= sock_kmalloc(sk
, sizeof(*key
), gfp
);
991 if (!tcp_alloc_md5sig_pool()) {
992 sock_kfree_s(sk
, key
, sizeof(*key
));
996 memcpy(key
->key
, newkey
, newkeylen
);
997 key
->keylen
= newkeylen
;
998 key
->family
= family
;
999 memcpy(&key
->addr
, addr
,
1000 (family
== AF_INET6
) ? sizeof(struct in6_addr
) :
1001 sizeof(struct in_addr
));
1002 hlist_add_head_rcu(&key
->node
, &md5sig
->head
);
1005 EXPORT_SYMBOL(tcp_md5_do_add
);
1007 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
, int family
)
1009 struct tcp_md5sig_key
*key
;
1011 key
= tcp_md5_do_lookup(sk
, addr
, family
);
1014 hlist_del_rcu(&key
->node
);
1015 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1016 kfree_rcu(key
, rcu
);
1019 EXPORT_SYMBOL(tcp_md5_do_del
);
1021 static void tcp_clear_md5_list(struct sock
*sk
)
1023 struct tcp_sock
*tp
= tcp_sk(sk
);
1024 struct tcp_md5sig_key
*key
;
1025 struct hlist_node
*n
;
1026 struct tcp_md5sig_info
*md5sig
;
1028 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
1030 hlist_for_each_entry_safe(key
, n
, &md5sig
->head
, node
) {
1031 hlist_del_rcu(&key
->node
);
1032 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1033 kfree_rcu(key
, rcu
);
1037 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1040 struct tcp_md5sig cmd
;
1041 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1043 if (optlen
< sizeof(cmd
))
1046 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1049 if (sin
->sin_family
!= AF_INET
)
1052 if (!cmd
.tcpm_keylen
)
1053 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1056 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1059 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1060 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1064 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1065 __be32 daddr
, __be32 saddr
, int nbytes
)
1067 struct tcp4_pseudohdr
*bp
;
1068 struct scatterlist sg
;
1070 bp
= &hp
->md5_blk
.ip4
;
1073 * 1. the TCP pseudo-header (in the order: source IP address,
1074 * destination IP address, zero-padded protocol number, and
1080 bp
->protocol
= IPPROTO_TCP
;
1081 bp
->len
= cpu_to_be16(nbytes
);
1083 sg_init_one(&sg
, bp
, sizeof(*bp
));
1084 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1087 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1088 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1090 struct tcp_md5sig_pool
*hp
;
1091 struct hash_desc
*desc
;
1093 hp
= tcp_get_md5sig_pool();
1095 goto clear_hash_noput
;
1096 desc
= &hp
->md5_desc
;
1098 if (crypto_hash_init(desc
))
1100 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1102 if (tcp_md5_hash_header(hp
, th
))
1104 if (tcp_md5_hash_key(hp
, key
))
1106 if (crypto_hash_final(desc
, md5_hash
))
1109 tcp_put_md5sig_pool();
1113 tcp_put_md5sig_pool();
1115 memset(md5_hash
, 0, 16);
1119 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1120 const struct sock
*sk
, const struct request_sock
*req
,
1121 const struct sk_buff
*skb
)
1123 struct tcp_md5sig_pool
*hp
;
1124 struct hash_desc
*desc
;
1125 const struct tcphdr
*th
= tcp_hdr(skb
);
1126 __be32 saddr
, daddr
;
1129 saddr
= inet_sk(sk
)->inet_saddr
;
1130 daddr
= inet_sk(sk
)->inet_daddr
;
1132 saddr
= inet_rsk(req
)->ir_loc_addr
;
1133 daddr
= inet_rsk(req
)->ir_rmt_addr
;
1135 const struct iphdr
*iph
= ip_hdr(skb
);
1140 hp
= tcp_get_md5sig_pool();
1142 goto clear_hash_noput
;
1143 desc
= &hp
->md5_desc
;
1145 if (crypto_hash_init(desc
))
1148 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1150 if (tcp_md5_hash_header(hp
, th
))
1152 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1154 if (tcp_md5_hash_key(hp
, key
))
1156 if (crypto_hash_final(desc
, md5_hash
))
1159 tcp_put_md5sig_pool();
1163 tcp_put_md5sig_pool();
1165 memset(md5_hash
, 0, 16);
1168 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1170 static bool tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1173 * This gets called for each TCP segment that arrives
1174 * so we want to be efficient.
1175 * We have 3 drop cases:
1176 * o No MD5 hash and one expected.
1177 * o MD5 hash and we're not expecting one.
1178 * o MD5 hash and its wrong.
1180 const __u8
*hash_location
= NULL
;
1181 struct tcp_md5sig_key
*hash_expected
;
1182 const struct iphdr
*iph
= ip_hdr(skb
);
1183 const struct tcphdr
*th
= tcp_hdr(skb
);
1185 unsigned char newhash
[16];
1187 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1189 hash_location
= tcp_parse_md5sig_option(th
);
1191 /* We've parsed the options - do we have a hash? */
1192 if (!hash_expected
&& !hash_location
)
1195 if (hash_expected
&& !hash_location
) {
1196 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1200 if (!hash_expected
&& hash_location
) {
1201 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1205 /* Okay, so this is hash_expected and hash_location -
1206 * so we need to calculate the checksum.
1208 genhash
= tcp_v4_md5_hash_skb(newhash
,
1212 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1213 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1214 &iph
->saddr
, ntohs(th
->source
),
1215 &iph
->daddr
, ntohs(th
->dest
),
1216 genhash
? " tcp_v4_calc_md5_hash failed"
1225 static void tcp_v4_init_req(struct request_sock
*req
, struct sock
*sk
,
1226 struct sk_buff
*skb
)
1228 struct inet_request_sock
*ireq
= inet_rsk(req
);
1230 ireq
->ir_loc_addr
= ip_hdr(skb
)->daddr
;
1231 ireq
->ir_rmt_addr
= ip_hdr(skb
)->saddr
;
1232 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1233 ireq
->opt
= tcp_v4_save_options(skb
);
1236 static struct dst_entry
*tcp_v4_route_req(struct sock
*sk
, struct flowi
*fl
,
1237 const struct request_sock
*req
,
1240 struct dst_entry
*dst
= inet_csk_route_req(sk
, &fl
->u
.ip4
, req
);
1243 if (fl
->u
.ip4
.daddr
== inet_rsk(req
)->ir_rmt_addr
)
1252 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1254 .obj_size
= sizeof(struct tcp_request_sock
),
1255 .rtx_syn_ack
= tcp_rtx_synack
,
1256 .send_ack
= tcp_v4_reqsk_send_ack
,
1257 .destructor
= tcp_v4_reqsk_destructor
,
1258 .send_reset
= tcp_v4_send_reset
,
1259 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1262 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1263 .mss_clamp
= TCP_MSS_DEFAULT
,
1264 #ifdef CONFIG_TCP_MD5SIG
1265 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1266 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1268 .init_req
= tcp_v4_init_req
,
1269 #ifdef CONFIG_SYN_COOKIES
1270 .cookie_init_seq
= cookie_v4_init_sequence
,
1272 .route_req
= tcp_v4_route_req
,
1273 .init_seq
= tcp_v4_init_sequence
,
1274 .send_synack
= tcp_v4_send_synack
,
1275 .queue_hash_add
= inet_csk_reqsk_queue_hash_add
,
1278 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1280 /* Never answer to SYNs send to broadcast or multicast */
1281 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1284 return tcp_conn_request(&tcp_request_sock_ops
,
1285 &tcp_request_sock_ipv4_ops
, sk
, skb
);
1288 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1291 EXPORT_SYMBOL(tcp_v4_conn_request
);
1295 * The three way handshake has completed - we got a valid synack -
1296 * now create the new socket.
1298 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1299 struct request_sock
*req
,
1300 struct dst_entry
*dst
)
1302 struct inet_request_sock
*ireq
;
1303 struct inet_sock
*newinet
;
1304 struct tcp_sock
*newtp
;
1306 #ifdef CONFIG_TCP_MD5SIG
1307 struct tcp_md5sig_key
*key
;
1309 struct ip_options_rcu
*inet_opt
;
1311 if (sk_acceptq_is_full(sk
))
1314 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1318 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1319 inet_sk_rx_dst_set(newsk
, skb
);
1321 newtp
= tcp_sk(newsk
);
1322 newinet
= inet_sk(newsk
);
1323 ireq
= inet_rsk(req
);
1324 newinet
->inet_daddr
= ireq
->ir_rmt_addr
;
1325 newinet
->inet_rcv_saddr
= ireq
->ir_loc_addr
;
1326 newinet
->inet_saddr
= ireq
->ir_loc_addr
;
1327 inet_opt
= ireq
->opt
;
1328 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1330 newinet
->mc_index
= inet_iif(skb
);
1331 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1332 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1333 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1334 inet_set_txhash(newsk
);
1336 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1337 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1340 dst
= inet_csk_route_child_sock(sk
, newsk
, req
);
1344 /* syncookie case : see end of cookie_v4_check() */
1346 sk_setup_caps(newsk
, dst
);
1348 tcp_sync_mss(newsk
, dst_mtu(dst
));
1349 newtp
->advmss
= dst_metric_advmss(dst
);
1350 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1351 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1352 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1354 tcp_initialize_rcv_mss(newsk
);
1356 #ifdef CONFIG_TCP_MD5SIG
1357 /* Copy over the MD5 key from the original socket */
1358 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1362 * We're using one, so create a matching key
1363 * on the newsk structure. If we fail to get
1364 * memory, then we end up not copying the key
1367 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1368 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1369 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1373 if (__inet_inherit_port(sk
, newsk
) < 0)
1375 __inet_hash_nolisten(newsk
, NULL
);
1380 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1384 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1387 inet_csk_prepare_forced_close(newsk
);
1391 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1393 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1395 struct tcphdr
*th
= tcp_hdr(skb
);
1396 const struct iphdr
*iph
= ip_hdr(skb
);
1398 struct request_sock
**prev
;
1399 /* Find possible connection requests. */
1400 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1401 iph
->saddr
, iph
->daddr
);
1403 return tcp_check_req(sk
, skb
, req
, prev
, false);
1405 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1406 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1409 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1413 inet_twsk_put(inet_twsk(nsk
));
1417 #ifdef CONFIG_SYN_COOKIES
1419 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1424 /* The socket must have it's spinlock held when we get
1427 * We have a potential double-lock case here, so even when
1428 * doing backlog processing we use the BH locking scheme.
1429 * This is because we cannot sleep with the original spinlock
1432 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1435 #ifdef CONFIG_TCP_MD5SIG
1437 * We really want to reject the packet as early as possible
1439 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1440 * o There is an MD5 option and we're not expecting one
1442 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1446 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1447 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1449 sock_rps_save_rxhash(sk
, skb
);
1451 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1452 dst
->ops
->check(dst
, 0) == NULL
) {
1454 sk
->sk_rx_dst
= NULL
;
1457 tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
);
1461 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1464 if (sk
->sk_state
== TCP_LISTEN
) {
1465 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1470 sock_rps_save_rxhash(nsk
, skb
);
1471 if (tcp_child_process(sk
, nsk
, skb
)) {
1478 sock_rps_save_rxhash(sk
, skb
);
1480 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1487 tcp_v4_send_reset(rsk
, skb
);
1490 /* Be careful here. If this function gets more complicated and
1491 * gcc suffers from register pressure on the x86, sk (in %ebx)
1492 * might be destroyed here. This current version compiles correctly,
1493 * but you have been warned.
1498 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1499 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1502 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1504 void tcp_v4_early_demux(struct sk_buff
*skb
)
1506 const struct iphdr
*iph
;
1507 const struct tcphdr
*th
;
1510 if (skb
->pkt_type
!= PACKET_HOST
)
1513 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1519 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1522 sk
= __inet_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1523 iph
->saddr
, th
->source
,
1524 iph
->daddr
, ntohs(th
->dest
),
1528 skb
->destructor
= sock_edemux
;
1529 if (sk
->sk_state
!= TCP_TIME_WAIT
) {
1530 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1533 dst
= dst_check(dst
, 0);
1535 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1536 skb_dst_set_noref(skb
, dst
);
1541 /* Packet is added to VJ-style prequeue for processing in process
1542 * context, if a reader task is waiting. Apparently, this exciting
1543 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1544 * failed somewhere. Latency? Burstiness? Well, at least now we will
1545 * see, why it failed. 8)8) --ANK
1548 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1550 struct tcp_sock
*tp
= tcp_sk(sk
);
1552 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
1555 if (skb
->len
<= tcp_hdrlen(skb
) &&
1556 skb_queue_len(&tp
->ucopy
.prequeue
) == 0)
1560 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1561 tp
->ucopy
.memory
+= skb
->truesize
;
1562 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1563 struct sk_buff
*skb1
;
1565 BUG_ON(sock_owned_by_user(sk
));
1567 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1568 sk_backlog_rcv(sk
, skb1
);
1569 NET_INC_STATS_BH(sock_net(sk
),
1570 LINUX_MIB_TCPPREQUEUEDROPPED
);
1573 tp
->ucopy
.memory
= 0;
1574 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1575 wake_up_interruptible_sync_poll(sk_sleep(sk
),
1576 POLLIN
| POLLRDNORM
| POLLRDBAND
);
1577 if (!inet_csk_ack_scheduled(sk
))
1578 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
1579 (3 * tcp_rto_min(sk
)) / 4,
1584 EXPORT_SYMBOL(tcp_prequeue
);
1590 int tcp_v4_rcv(struct sk_buff
*skb
)
1592 const struct iphdr
*iph
;
1593 const struct tcphdr
*th
;
1596 struct net
*net
= dev_net(skb
->dev
);
1598 if (skb
->pkt_type
!= PACKET_HOST
)
1601 /* Count it even if it's bad */
1602 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1604 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1609 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1611 if (!pskb_may_pull(skb
, th
->doff
* 4))
1614 /* An explanation is required here, I think.
1615 * Packet length and doff are validated by header prediction,
1616 * provided case of th->doff==0 is eliminated.
1617 * So, we defer the checks. */
1619 if (skb_checksum_init(skb
, IPPROTO_TCP
, inet_compute_pseudo
))
1624 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1625 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1626 skb
->len
- th
->doff
* 4);
1627 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1628 TCP_SKB_CB(skb
)->when
= 0;
1629 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
1630 TCP_SKB_CB(skb
)->sacked
= 0;
1632 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1637 if (sk
->sk_state
== TCP_TIME_WAIT
)
1640 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1641 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1642 goto discard_and_relse
;
1645 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1646 goto discard_and_relse
;
1649 if (sk_filter(sk
, skb
))
1650 goto discard_and_relse
;
1652 sk_mark_napi_id(sk
, skb
);
1655 bh_lock_sock_nested(sk
);
1657 if (!sock_owned_by_user(sk
)) {
1658 #ifdef CONFIG_NET_DMA
1659 struct tcp_sock
*tp
= tcp_sk(sk
);
1660 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1661 tp
->ucopy
.dma_chan
= net_dma_find_channel();
1662 if (tp
->ucopy
.dma_chan
)
1663 ret
= tcp_v4_do_rcv(sk
, skb
);
1667 if (!tcp_prequeue(sk
, skb
))
1668 ret
= tcp_v4_do_rcv(sk
, skb
);
1670 } else if (unlikely(sk_add_backlog(sk
, skb
,
1671 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
1673 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1674 goto discard_and_relse
;
1683 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1686 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1688 TCP_INC_STATS_BH(net
, TCP_MIB_CSUMERRORS
);
1690 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1692 tcp_v4_send_reset(NULL
, skb
);
1696 /* Discard frame. */
1705 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1706 inet_twsk_put(inet_twsk(sk
));
1710 if (skb
->len
< (th
->doff
<< 2)) {
1711 inet_twsk_put(inet_twsk(sk
));
1714 if (tcp_checksum_complete(skb
)) {
1715 inet_twsk_put(inet_twsk(sk
));
1718 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1720 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1722 iph
->saddr
, th
->source
,
1723 iph
->daddr
, th
->dest
,
1726 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1727 inet_twsk_put(inet_twsk(sk
));
1731 /* Fall through to ACK */
1734 tcp_v4_timewait_ack(sk
, skb
);
1738 case TCP_TW_SUCCESS
:;
1743 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1744 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1745 .twsk_unique
= tcp_twsk_unique
,
1746 .twsk_destructor
= tcp_twsk_destructor
,
1749 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
1751 struct dst_entry
*dst
= skb_dst(skb
);
1754 sk
->sk_rx_dst
= dst
;
1755 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
1757 EXPORT_SYMBOL(inet_sk_rx_dst_set
);
1759 const struct inet_connection_sock_af_ops ipv4_specific
= {
1760 .queue_xmit
= ip_queue_xmit
,
1761 .send_check
= tcp_v4_send_check
,
1762 .rebuild_header
= inet_sk_rebuild_header
,
1763 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
1764 .conn_request
= tcp_v4_conn_request
,
1765 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1766 .net_header_len
= sizeof(struct iphdr
),
1767 .setsockopt
= ip_setsockopt
,
1768 .getsockopt
= ip_getsockopt
,
1769 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1770 .sockaddr_len
= sizeof(struct sockaddr_in
),
1771 .bind_conflict
= inet_csk_bind_conflict
,
1772 #ifdef CONFIG_COMPAT
1773 .compat_setsockopt
= compat_ip_setsockopt
,
1774 .compat_getsockopt
= compat_ip_getsockopt
,
1777 EXPORT_SYMBOL(ipv4_specific
);
1779 #ifdef CONFIG_TCP_MD5SIG
1780 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1781 .md5_lookup
= tcp_v4_md5_lookup
,
1782 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1783 .md5_parse
= tcp_v4_parse_md5_keys
,
1787 /* NOTE: A lot of things set to zero explicitly by call to
1788 * sk_alloc() so need not be done here.
1790 static int tcp_v4_init_sock(struct sock
*sk
)
1792 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1796 icsk
->icsk_af_ops
= &ipv4_specific
;
1798 #ifdef CONFIG_TCP_MD5SIG
1799 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv4_specific
;
1805 void tcp_v4_destroy_sock(struct sock
*sk
)
1807 struct tcp_sock
*tp
= tcp_sk(sk
);
1809 tcp_clear_xmit_timers(sk
);
1811 tcp_cleanup_congestion_control(sk
);
1813 /* Cleanup up the write buffer. */
1814 tcp_write_queue_purge(sk
);
1816 /* Cleans up our, hopefully empty, out_of_order_queue. */
1817 __skb_queue_purge(&tp
->out_of_order_queue
);
1819 #ifdef CONFIG_TCP_MD5SIG
1820 /* Clean up the MD5 key list, if any */
1821 if (tp
->md5sig_info
) {
1822 tcp_clear_md5_list(sk
);
1823 kfree_rcu(tp
->md5sig_info
, rcu
);
1824 tp
->md5sig_info
= NULL
;
1828 #ifdef CONFIG_NET_DMA
1829 /* Cleans up our sk_async_wait_queue */
1830 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1833 /* Clean prequeue, it must be empty really */
1834 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1836 /* Clean up a referenced TCP bind bucket. */
1837 if (inet_csk(sk
)->icsk_bind_hash
)
1840 BUG_ON(tp
->fastopen_rsk
!= NULL
);
1842 /* If socket is aborted during connect operation */
1843 tcp_free_fastopen_req(tp
);
1845 sk_sockets_allocated_dec(sk
);
1846 sock_release_memcg(sk
);
1848 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1850 #ifdef CONFIG_PROC_FS
1851 /* Proc filesystem TCP sock list dumping. */
1854 * Get next listener socket follow cur. If cur is NULL, get first socket
1855 * starting from bucket given in st->bucket; when st->bucket is zero the
1856 * very first socket in the hash table is returned.
1858 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
1860 struct inet_connection_sock
*icsk
;
1861 struct hlist_nulls_node
*node
;
1862 struct sock
*sk
= cur
;
1863 struct inet_listen_hashbucket
*ilb
;
1864 struct tcp_iter_state
*st
= seq
->private;
1865 struct net
*net
= seq_file_net(seq
);
1868 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1869 spin_lock_bh(&ilb
->lock
);
1870 sk
= sk_nulls_head(&ilb
->head
);
1874 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1878 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
1879 struct request_sock
*req
= cur
;
1881 icsk
= inet_csk(st
->syn_wait_sk
);
1885 if (req
->rsk_ops
->family
== st
->family
) {
1891 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
1894 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
1896 sk
= sk_nulls_next(st
->syn_wait_sk
);
1897 st
->state
= TCP_SEQ_STATE_LISTENING
;
1898 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1900 icsk
= inet_csk(sk
);
1901 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1902 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
1904 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1905 sk
= sk_nulls_next(sk
);
1908 sk_nulls_for_each_from(sk
, node
) {
1909 if (!net_eq(sock_net(sk
), net
))
1911 if (sk
->sk_family
== st
->family
) {
1915 icsk
= inet_csk(sk
);
1916 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1917 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
1919 st
->uid
= sock_i_uid(sk
);
1920 st
->syn_wait_sk
= sk
;
1921 st
->state
= TCP_SEQ_STATE_OPENREQ
;
1925 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1927 spin_unlock_bh(&ilb
->lock
);
1929 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
1930 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1931 spin_lock_bh(&ilb
->lock
);
1932 sk
= sk_nulls_head(&ilb
->head
);
1940 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
1942 struct tcp_iter_state
*st
= seq
->private;
1947 rc
= listening_get_next(seq
, NULL
);
1949 while (rc
&& *pos
) {
1950 rc
= listening_get_next(seq
, rc
);
1956 static inline bool empty_bucket(const struct tcp_iter_state
*st
)
1958 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
1962 * Get first established socket starting from bucket given in st->bucket.
1963 * If st->bucket is zero, the very first socket in the hash is returned.
1965 static void *established_get_first(struct seq_file
*seq
)
1967 struct tcp_iter_state
*st
= seq
->private;
1968 struct net
*net
= seq_file_net(seq
);
1972 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
1974 struct hlist_nulls_node
*node
;
1975 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
1977 /* Lockless fast path for the common case of empty buckets */
1978 if (empty_bucket(st
))
1982 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
1983 if (sk
->sk_family
!= st
->family
||
1984 !net_eq(sock_net(sk
), net
)) {
1990 spin_unlock_bh(lock
);
1996 static void *established_get_next(struct seq_file
*seq
, void *cur
)
1998 struct sock
*sk
= cur
;
1999 struct hlist_nulls_node
*node
;
2000 struct tcp_iter_state
*st
= seq
->private;
2001 struct net
*net
= seq_file_net(seq
);
2006 sk
= sk_nulls_next(sk
);
2008 sk_nulls_for_each_from(sk
, node
) {
2009 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2013 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2015 return established_get_first(seq
);
2018 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2020 struct tcp_iter_state
*st
= seq
->private;
2024 rc
= established_get_first(seq
);
2027 rc
= established_get_next(seq
, rc
);
2033 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2036 struct tcp_iter_state
*st
= seq
->private;
2038 st
->state
= TCP_SEQ_STATE_LISTENING
;
2039 rc
= listening_get_idx(seq
, &pos
);
2042 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2043 rc
= established_get_idx(seq
, pos
);
2049 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2051 struct tcp_iter_state
*st
= seq
->private;
2052 int offset
= st
->offset
;
2053 int orig_num
= st
->num
;
2056 switch (st
->state
) {
2057 case TCP_SEQ_STATE_OPENREQ
:
2058 case TCP_SEQ_STATE_LISTENING
:
2059 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2061 st
->state
= TCP_SEQ_STATE_LISTENING
;
2062 rc
= listening_get_next(seq
, NULL
);
2063 while (offset
-- && rc
)
2064 rc
= listening_get_next(seq
, rc
);
2068 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2070 case TCP_SEQ_STATE_ESTABLISHED
:
2071 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2073 rc
= established_get_first(seq
);
2074 while (offset
-- && rc
)
2075 rc
= established_get_next(seq
, rc
);
2083 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2085 struct tcp_iter_state
*st
= seq
->private;
2088 if (*pos
&& *pos
== st
->last_pos
) {
2089 rc
= tcp_seek_last_pos(seq
);
2094 st
->state
= TCP_SEQ_STATE_LISTENING
;
2098 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2101 st
->last_pos
= *pos
;
2105 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2107 struct tcp_iter_state
*st
= seq
->private;
2110 if (v
== SEQ_START_TOKEN
) {
2111 rc
= tcp_get_idx(seq
, 0);
2115 switch (st
->state
) {
2116 case TCP_SEQ_STATE_OPENREQ
:
2117 case TCP_SEQ_STATE_LISTENING
:
2118 rc
= listening_get_next(seq
, v
);
2120 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2123 rc
= established_get_first(seq
);
2126 case TCP_SEQ_STATE_ESTABLISHED
:
2127 rc
= established_get_next(seq
, v
);
2132 st
->last_pos
= *pos
;
2136 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2138 struct tcp_iter_state
*st
= seq
->private;
2140 switch (st
->state
) {
2141 case TCP_SEQ_STATE_OPENREQ
:
2143 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2144 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2146 case TCP_SEQ_STATE_LISTENING
:
2147 if (v
!= SEQ_START_TOKEN
)
2148 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2150 case TCP_SEQ_STATE_ESTABLISHED
:
2152 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2157 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2159 struct tcp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2160 struct tcp_iter_state
*s
;
2163 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2164 sizeof(struct tcp_iter_state
));
2168 s
= ((struct seq_file
*)file
->private_data
)->private;
2169 s
->family
= afinfo
->family
;
2173 EXPORT_SYMBOL(tcp_seq_open
);
2175 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2178 struct proc_dir_entry
*p
;
2180 afinfo
->seq_ops
.start
= tcp_seq_start
;
2181 afinfo
->seq_ops
.next
= tcp_seq_next
;
2182 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2184 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2185 afinfo
->seq_fops
, afinfo
);
2190 EXPORT_SYMBOL(tcp_proc_register
);
2192 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2194 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2196 EXPORT_SYMBOL(tcp_proc_unregister
);
2198 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2199 struct seq_file
*f
, int i
, kuid_t uid
)
2201 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2202 long delta
= req
->expires
- jiffies
;
2204 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2205 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2208 ntohs(inet_sk(sk
)->inet_sport
),
2210 ntohs(ireq
->ir_rmt_port
),
2212 0, 0, /* could print option size, but that is af dependent. */
2213 1, /* timers active (only the expire timer) */
2214 jiffies_delta_to_clock_t(delta
),
2216 from_kuid_munged(seq_user_ns(f
), uid
),
2217 0, /* non standard timer */
2218 0, /* open_requests have no inode */
2219 atomic_read(&sk
->sk_refcnt
),
2223 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
)
2226 unsigned long timer_expires
;
2227 const struct tcp_sock
*tp
= tcp_sk(sk
);
2228 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2229 const struct inet_sock
*inet
= inet_sk(sk
);
2230 struct fastopen_queue
*fastopenq
= icsk
->icsk_accept_queue
.fastopenq
;
2231 __be32 dest
= inet
->inet_daddr
;
2232 __be32 src
= inet
->inet_rcv_saddr
;
2233 __u16 destp
= ntohs(inet
->inet_dport
);
2234 __u16 srcp
= ntohs(inet
->inet_sport
);
2237 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
2238 icsk
->icsk_pending
== ICSK_TIME_EARLY_RETRANS
||
2239 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
2241 timer_expires
= icsk
->icsk_timeout
;
2242 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2244 timer_expires
= icsk
->icsk_timeout
;
2245 } else if (timer_pending(&sk
->sk_timer
)) {
2247 timer_expires
= sk
->sk_timer
.expires
;
2250 timer_expires
= jiffies
;
2253 if (sk
->sk_state
== TCP_LISTEN
)
2254 rx_queue
= sk
->sk_ack_backlog
;
2257 * because we dont lock socket, we might find a transient negative value
2259 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2261 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2262 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2263 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2264 tp
->write_seq
- tp
->snd_una
,
2267 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
2268 icsk
->icsk_retransmits
,
2269 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sk
)),
2270 icsk
->icsk_probes_out
,
2272 atomic_read(&sk
->sk_refcnt
), sk
,
2273 jiffies_to_clock_t(icsk
->icsk_rto
),
2274 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2275 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2277 sk
->sk_state
== TCP_LISTEN
?
2278 (fastopenq
? fastopenq
->max_qlen
: 0) :
2279 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
));
2282 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2283 struct seq_file
*f
, int i
)
2287 s32 delta
= tw
->tw_ttd
- inet_tw_time_stamp();
2289 dest
= tw
->tw_daddr
;
2290 src
= tw
->tw_rcv_saddr
;
2291 destp
= ntohs(tw
->tw_dport
);
2292 srcp
= ntohs(tw
->tw_sport
);
2294 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2295 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2296 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2297 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
2298 atomic_read(&tw
->tw_refcnt
), tw
);
2303 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2305 struct tcp_iter_state
*st
;
2306 struct sock
*sk
= v
;
2308 seq_setwidth(seq
, TMPSZ
- 1);
2309 if (v
== SEQ_START_TOKEN
) {
2310 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2311 "rx_queue tr tm->when retrnsmt uid timeout "
2317 switch (st
->state
) {
2318 case TCP_SEQ_STATE_LISTENING
:
2319 case TCP_SEQ_STATE_ESTABLISHED
:
2320 if (sk
->sk_state
== TCP_TIME_WAIT
)
2321 get_timewait4_sock(v
, seq
, st
->num
);
2323 get_tcp4_sock(v
, seq
, st
->num
);
2325 case TCP_SEQ_STATE_OPENREQ
:
2326 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
);
2334 static const struct file_operations tcp_afinfo_seq_fops
= {
2335 .owner
= THIS_MODULE
,
2336 .open
= tcp_seq_open
,
2338 .llseek
= seq_lseek
,
2339 .release
= seq_release_net
2342 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2345 .seq_fops
= &tcp_afinfo_seq_fops
,
2347 .show
= tcp4_seq_show
,
2351 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2353 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2356 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2358 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2361 static struct pernet_operations tcp4_net_ops
= {
2362 .init
= tcp4_proc_init_net
,
2363 .exit
= tcp4_proc_exit_net
,
2366 int __init
tcp4_proc_init(void)
2368 return register_pernet_subsys(&tcp4_net_ops
);
2371 void tcp4_proc_exit(void)
2373 unregister_pernet_subsys(&tcp4_net_ops
);
2375 #endif /* CONFIG_PROC_FS */
2377 struct proto tcp_prot
= {
2379 .owner
= THIS_MODULE
,
2381 .connect
= tcp_v4_connect
,
2382 .disconnect
= tcp_disconnect
,
2383 .accept
= inet_csk_accept
,
2385 .init
= tcp_v4_init_sock
,
2386 .destroy
= tcp_v4_destroy_sock
,
2387 .shutdown
= tcp_shutdown
,
2388 .setsockopt
= tcp_setsockopt
,
2389 .getsockopt
= tcp_getsockopt
,
2390 .recvmsg
= tcp_recvmsg
,
2391 .sendmsg
= tcp_sendmsg
,
2392 .sendpage
= tcp_sendpage
,
2393 .backlog_rcv
= tcp_v4_do_rcv
,
2394 .release_cb
= tcp_release_cb
,
2395 .mtu_reduced
= tcp_v4_mtu_reduced
,
2397 .unhash
= inet_unhash
,
2398 .get_port
= inet_csk_get_port
,
2399 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2400 .stream_memory_free
= tcp_stream_memory_free
,
2401 .sockets_allocated
= &tcp_sockets_allocated
,
2402 .orphan_count
= &tcp_orphan_count
,
2403 .memory_allocated
= &tcp_memory_allocated
,
2404 .memory_pressure
= &tcp_memory_pressure
,
2405 .sysctl_mem
= sysctl_tcp_mem
,
2406 .sysctl_wmem
= sysctl_tcp_wmem
,
2407 .sysctl_rmem
= sysctl_tcp_rmem
,
2408 .max_header
= MAX_TCP_HEADER
,
2409 .obj_size
= sizeof(struct tcp_sock
),
2410 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2411 .twsk_prot
= &tcp_timewait_sock_ops
,
2412 .rsk_prot
= &tcp_request_sock_ops
,
2413 .h
.hashinfo
= &tcp_hashinfo
,
2414 .no_autobind
= true,
2415 #ifdef CONFIG_COMPAT
2416 .compat_setsockopt
= compat_tcp_setsockopt
,
2417 .compat_getsockopt
= compat_tcp_getsockopt
,
2419 #ifdef CONFIG_MEMCG_KMEM
2420 .init_cgroup
= tcp_init_cgroup
,
2421 .destroy_cgroup
= tcp_destroy_cgroup
,
2422 .proto_cgroup
= tcp_proto_cgroup
,
2425 EXPORT_SYMBOL(tcp_prot
);
2427 static int __net_init
tcp_sk_init(struct net
*net
)
2429 net
->ipv4
.sysctl_tcp_ecn
= 2;
2433 static void __net_exit
tcp_sk_exit(struct net
*net
)
2437 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2439 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2442 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2443 .init
= tcp_sk_init
,
2444 .exit
= tcp_sk_exit
,
2445 .exit_batch
= tcp_sk_exit_batch
,
2448 void __init
tcp_v4_init(void)
2450 inet_hashinfo_init(&tcp_hashinfo
);
2451 if (register_pernet_subsys(&tcp_sk_ops
))
2452 panic("Failed to create the TCP control socket.\n");