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/secure_seq.h>
76 #include <net/tcp_memcontrol.h>
77 #include <net/busy_poll.h>
79 #include <linux/inet.h>
80 #include <linux/ipv6.h>
81 #include <linux/stddef.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
85 #include <linux/crypto.h>
86 #include <linux/scatterlist.h>
88 int sysctl_tcp_tw_reuse __read_mostly
;
89 int sysctl_tcp_low_latency __read_mostly
;
90 EXPORT_SYMBOL(sysctl_tcp_low_latency
);
92 #ifdef CONFIG_TCP_MD5SIG
93 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
94 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
);
97 struct inet_hashinfo tcp_hashinfo
;
98 EXPORT_SYMBOL(tcp_hashinfo
);
100 static __u32
tcp_v4_init_sequence(const struct sk_buff
*skb
)
102 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
105 tcp_hdr(skb
)->source
);
108 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
)
110 const struct tcp_timewait_sock
*tcptw
= tcp_twsk(sktw
);
111 struct tcp_sock
*tp
= tcp_sk(sk
);
113 /* With PAWS, it is safe from the viewpoint
114 of data integrity. Even without PAWS it is safe provided sequence
115 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
117 Actually, the idea is close to VJ's one, only timestamp cache is
118 held not per host, but per port pair and TW bucket is used as state
121 If TW bucket has been already destroyed we fall back to VJ's scheme
122 and use initial timestamp retrieved from peer table.
124 if (tcptw
->tw_ts_recent_stamp
&&
125 (!twp
|| (sysctl_tcp_tw_reuse
&&
126 get_seconds() - tcptw
->tw_ts_recent_stamp
> 1))) {
127 tp
->write_seq
= tcptw
->tw_snd_nxt
+ 65535 + 2;
128 if (tp
->write_seq
== 0)
130 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
131 tp
->rx_opt
.ts_recent_stamp
= tcptw
->tw_ts_recent_stamp
;
138 EXPORT_SYMBOL_GPL(tcp_twsk_unique
);
140 /* This will initiate an outgoing connection. */
141 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
143 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
144 struct inet_sock
*inet
= inet_sk(sk
);
145 struct tcp_sock
*tp
= tcp_sk(sk
);
146 __be16 orig_sport
, orig_dport
;
147 __be32 daddr
, nexthop
;
151 struct ip_options_rcu
*inet_opt
;
153 if (addr_len
< sizeof(struct sockaddr_in
))
156 if (usin
->sin_family
!= AF_INET
)
157 return -EAFNOSUPPORT
;
159 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
160 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
161 sock_owned_by_user(sk
));
162 if (inet_opt
&& inet_opt
->opt
.srr
) {
165 nexthop
= inet_opt
->opt
.faddr
;
168 orig_sport
= inet
->inet_sport
;
169 orig_dport
= usin
->sin_port
;
170 fl4
= &inet
->cork
.fl
.u
.ip4
;
171 rt
= ip_route_connect(fl4
, nexthop
, inet
->inet_saddr
,
172 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
174 orig_sport
, orig_dport
, sk
);
177 if (err
== -ENETUNREACH
)
178 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
182 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
187 if (!inet_opt
|| !inet_opt
->opt
.srr
)
190 if (!inet
->inet_saddr
)
191 inet
->inet_saddr
= fl4
->saddr
;
192 sk_rcv_saddr_set(sk
, inet
->inet_saddr
);
194 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->inet_daddr
!= daddr
) {
195 /* Reset inherited state */
196 tp
->rx_opt
.ts_recent
= 0;
197 tp
->rx_opt
.ts_recent_stamp
= 0;
198 if (likely(!tp
->repair
))
202 if (tcp_death_row
.sysctl_tw_recycle
&&
203 !tp
->rx_opt
.ts_recent_stamp
&& fl4
->daddr
== daddr
)
204 tcp_fetch_timewait_stamp(sk
, &rt
->dst
);
206 inet
->inet_dport
= usin
->sin_port
;
207 sk_daddr_set(sk
, daddr
);
209 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
211 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
213 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
215 /* Socket identity is still unknown (sport may be zero).
216 * However we set state to SYN-SENT and not releasing socket
217 * lock select source port, enter ourselves into the hash tables and
218 * complete initialization after this.
220 tcp_set_state(sk
, TCP_SYN_SENT
);
221 err
= inet_hash_connect(&tcp_death_row
, sk
);
227 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
228 inet
->inet_sport
, inet
->inet_dport
, sk
);
234 /* OK, now commit destination to socket. */
235 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
236 sk_setup_caps(sk
, &rt
->dst
);
238 if (!tp
->write_seq
&& likely(!tp
->repair
))
239 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
244 inet
->inet_id
= tp
->write_seq
^ jiffies
;
246 err
= tcp_connect(sk
);
256 * This unhashes the socket and releases the local port,
259 tcp_set_state(sk
, TCP_CLOSE
);
261 sk
->sk_route_caps
= 0;
262 inet
->inet_dport
= 0;
265 EXPORT_SYMBOL(tcp_v4_connect
);
268 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
269 * It can be called through tcp_release_cb() if socket was owned by user
270 * at the time tcp_v4_err() was called to handle ICMP message.
272 void tcp_v4_mtu_reduced(struct sock
*sk
)
274 struct dst_entry
*dst
;
275 struct inet_sock
*inet
= inet_sk(sk
);
276 u32 mtu
= tcp_sk(sk
)->mtu_info
;
278 dst
= inet_csk_update_pmtu(sk
, mtu
);
282 /* Something is about to be wrong... Remember soft error
283 * for the case, if this connection will not able to recover.
285 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
286 sk
->sk_err_soft
= EMSGSIZE
;
290 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
291 ip_sk_accept_pmtu(sk
) &&
292 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
293 tcp_sync_mss(sk
, mtu
);
295 /* Resend the TCP packet because it's
296 * clear that the old packet has been
297 * dropped. This is the new "fast" path mtu
300 tcp_simple_retransmit(sk
);
301 } /* else let the usual retransmit timer handle it */
303 EXPORT_SYMBOL(tcp_v4_mtu_reduced
);
305 static void do_redirect(struct sk_buff
*skb
, struct sock
*sk
)
307 struct dst_entry
*dst
= __sk_dst_check(sk
, 0);
310 dst
->ops
->redirect(dst
, sk
, skb
);
314 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
315 void tcp_req_err(struct sock
*sk
, u32 seq
)
317 struct request_sock
*req
= inet_reqsk(sk
);
318 struct net
*net
= sock_net(sk
);
320 /* ICMPs are not backlogged, hence we cannot get
321 * an established socket here.
325 if (seq
!= tcp_rsk(req
)->snt_isn
) {
326 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
330 * Still in SYN_RECV, just remove it silently.
331 * There is no good way to pass the error to the newly
332 * created socket, and POSIX does not want network
333 * errors returned from accept().
335 NET_INC_STATS_BH(net
, LINUX_MIB_LISTENDROPS
);
336 inet_csk_reqsk_queue_drop(req
->rsk_listener
, req
);
339 EXPORT_SYMBOL(tcp_req_err
);
342 * This routine is called by the ICMP module when it gets some
343 * sort of error condition. If err < 0 then the socket should
344 * be closed and the error returned to the user. If err > 0
345 * it's just the icmp type << 8 | icmp code. After adjustment
346 * header points to the first 8 bytes of the tcp header. We need
347 * to find the appropriate port.
349 * The locking strategy used here is very "optimistic". When
350 * someone else accesses the socket the ICMP is just dropped
351 * and for some paths there is no check at all.
352 * A more general error queue to queue errors for later handling
353 * is probably better.
357 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
359 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
360 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
361 struct inet_connection_sock
*icsk
;
363 struct inet_sock
*inet
;
364 const int type
= icmp_hdr(icmp_skb
)->type
;
365 const int code
= icmp_hdr(icmp_skb
)->code
;
368 struct request_sock
*fastopen
;
372 struct net
*net
= dev_net(icmp_skb
->dev
);
374 sk
= __inet_lookup_established(net
, &tcp_hashinfo
, iph
->daddr
,
375 th
->dest
, iph
->saddr
, ntohs(th
->source
),
378 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
381 if (sk
->sk_state
== TCP_TIME_WAIT
) {
382 inet_twsk_put(inet_twsk(sk
));
385 seq
= ntohl(th
->seq
);
386 if (sk
->sk_state
== TCP_NEW_SYN_RECV
)
387 return tcp_req_err(sk
, seq
);
390 /* If too many ICMPs get dropped on busy
391 * servers this needs to be solved differently.
392 * We do take care of PMTU discovery (RFC1191) special case :
393 * we can receive locally generated ICMP messages while socket is held.
395 if (sock_owned_by_user(sk
)) {
396 if (!(type
== ICMP_DEST_UNREACH
&& code
== ICMP_FRAG_NEEDED
))
397 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
399 if (sk
->sk_state
== TCP_CLOSE
)
402 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
403 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
409 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
410 fastopen
= tp
->fastopen_rsk
;
411 snd_una
= fastopen
? tcp_rsk(fastopen
)->snt_isn
: tp
->snd_una
;
412 if (sk
->sk_state
!= TCP_LISTEN
&&
413 !between(seq
, snd_una
, tp
->snd_nxt
)) {
414 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
420 do_redirect(icmp_skb
, sk
);
422 case ICMP_SOURCE_QUENCH
:
423 /* Just silently ignore these. */
425 case ICMP_PARAMETERPROB
:
428 case ICMP_DEST_UNREACH
:
429 if (code
> NR_ICMP_UNREACH
)
432 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
433 /* We are not interested in TCP_LISTEN and open_requests
434 * (SYN-ACKs send out by Linux are always <576bytes so
435 * they should go through unfragmented).
437 if (sk
->sk_state
== TCP_LISTEN
)
441 if (!sock_owned_by_user(sk
)) {
442 tcp_v4_mtu_reduced(sk
);
444 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED
, &tp
->tsq_flags
))
450 err
= icmp_err_convert
[code
].errno
;
451 /* check if icmp_skb allows revert of backoff
452 * (see draft-zimmermann-tcp-lcd) */
453 if (code
!= ICMP_NET_UNREACH
&& code
!= ICMP_HOST_UNREACH
)
455 if (seq
!= tp
->snd_una
|| !icsk
->icsk_retransmits
||
456 !icsk
->icsk_backoff
|| fastopen
)
459 if (sock_owned_by_user(sk
))
462 icsk
->icsk_backoff
--;
463 icsk
->icsk_rto
= tp
->srtt_us
? __tcp_set_rto(tp
) :
465 icsk
->icsk_rto
= inet_csk_rto_backoff(icsk
, TCP_RTO_MAX
);
467 skb
= tcp_write_queue_head(sk
);
470 remaining
= icsk
->icsk_rto
-
472 tcp_time_stamp
- tcp_skb_timestamp(skb
));
475 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
476 remaining
, TCP_RTO_MAX
);
478 /* RTO revert clocked out retransmission.
479 * Will retransmit now */
480 tcp_retransmit_timer(sk
);
484 case ICMP_TIME_EXCEEDED
:
491 switch (sk
->sk_state
) {
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
)
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 sk not NULL, it means we did a successful lookup and incoming
603 * route had to be correct. prequeue might have dropped our dst.
605 if (!sk
&& skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
608 /* Swap the send and the receive. */
609 memset(&rep
, 0, sizeof(rep
));
610 rep
.th
.dest
= th
->source
;
611 rep
.th
.source
= th
->dest
;
612 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
616 rep
.th
.seq
= th
->ack_seq
;
619 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
620 skb
->len
- (th
->doff
<< 2));
623 memset(&arg
, 0, sizeof(arg
));
624 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
625 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
627 net
= sk
? sock_net(sk
) : dev_net(skb_dst(skb
)->dev
);
628 #ifdef CONFIG_TCP_MD5SIG
629 hash_location
= tcp_parse_md5sig_option(th
);
630 if (!sk
&& hash_location
) {
632 * active side is lost. Try to find listening socket through
633 * source port, and then find md5 key through listening socket.
634 * we are not loose security here:
635 * Incoming packet is checked with md5 hash with finding key,
636 * no RST generated if md5 hash doesn't match.
638 sk1
= __inet_lookup_listener(net
,
639 &tcp_hashinfo
, ip_hdr(skb
)->saddr
,
640 th
->source
, ip_hdr(skb
)->daddr
,
641 ntohs(th
->source
), inet_iif(skb
));
642 /* don't send rst if it can't find key */
646 key
= tcp_md5_do_lookup(sk1
, (union tcp_md5_addr
*)
647 &ip_hdr(skb
)->saddr
, AF_INET
);
651 genhash
= tcp_v4_md5_hash_skb(newhash
, key
, NULL
, skb
);
652 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
655 key
= sk
? tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)
661 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
663 (TCPOPT_MD5SIG
<< 8) |
665 /* Update length and the length the header thinks exists */
666 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
667 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
669 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
670 key
, ip_hdr(skb
)->saddr
,
671 ip_hdr(skb
)->daddr
, &rep
.th
);
674 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
675 ip_hdr(skb
)->saddr
, /* XXX */
676 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
677 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
678 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
679 /* When socket is gone, all binding information is lost.
680 * routing might fail in this case. No choice here, if we choose to force
681 * input interface, we will misroute in case of asymmetric route.
684 arg
.bound_dev_if
= sk
->sk_bound_dev_if
;
686 arg
.tos
= ip_hdr(skb
)->tos
;
687 ip_send_unicast_reply(*this_cpu_ptr(net
->ipv4
.tcp_sk
),
688 skb
, &TCP_SKB_CB(skb
)->header
.h4
.opt
,
689 ip_hdr(skb
)->saddr
, ip_hdr(skb
)->daddr
,
690 &arg
, arg
.iov
[0].iov_len
);
692 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
693 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
695 #ifdef CONFIG_TCP_MD5SIG
704 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
705 outside socket context is ugly, certainly. What can I do?
708 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
709 u32 win
, u32 tsval
, u32 tsecr
, int oif
,
710 struct tcp_md5sig_key
*key
,
711 int reply_flags
, u8 tos
)
713 const struct tcphdr
*th
= tcp_hdr(skb
);
716 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
717 #ifdef CONFIG_TCP_MD5SIG
718 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
722 struct ip_reply_arg arg
;
723 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
725 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
726 memset(&arg
, 0, sizeof(arg
));
728 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
729 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
731 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
732 (TCPOPT_TIMESTAMP
<< 8) |
734 rep
.opt
[1] = htonl(tsval
);
735 rep
.opt
[2] = htonl(tsecr
);
736 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
739 /* Swap the send and the receive. */
740 rep
.th
.dest
= th
->source
;
741 rep
.th
.source
= th
->dest
;
742 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
743 rep
.th
.seq
= htonl(seq
);
744 rep
.th
.ack_seq
= htonl(ack
);
746 rep
.th
.window
= htons(win
);
748 #ifdef CONFIG_TCP_MD5SIG
750 int offset
= (tsecr
) ? 3 : 0;
752 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
754 (TCPOPT_MD5SIG
<< 8) |
756 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
757 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
759 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
760 key
, ip_hdr(skb
)->saddr
,
761 ip_hdr(skb
)->daddr
, &rep
.th
);
764 arg
.flags
= reply_flags
;
765 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
766 ip_hdr(skb
)->saddr
, /* XXX */
767 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
768 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
770 arg
.bound_dev_if
= oif
;
772 ip_send_unicast_reply(*this_cpu_ptr(net
->ipv4
.tcp_sk
),
773 skb
, &TCP_SKB_CB(skb
)->header
.h4
.opt
,
774 ip_hdr(skb
)->saddr
, ip_hdr(skb
)->daddr
,
775 &arg
, arg
.iov
[0].iov_len
);
777 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
780 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
782 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
783 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
785 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
786 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
787 tcp_time_stamp
+ tcptw
->tw_ts_offset
,
790 tcp_twsk_md5_key(tcptw
),
791 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
798 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
799 struct request_sock
*req
)
801 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
802 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
804 tcp_v4_send_ack(skb
, (sk
->sk_state
== TCP_LISTEN
) ?
805 tcp_rsk(req
)->snt_isn
+ 1 : tcp_sk(sk
)->snd_nxt
,
806 tcp_rsk(req
)->rcv_nxt
, req
->rcv_wnd
,
810 tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&ip_hdr(skb
)->daddr
,
812 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
817 * Send a SYN-ACK after having received a SYN.
818 * This still operates on a request_sock only, not on a big
821 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
823 struct request_sock
*req
,
825 struct tcp_fastopen_cookie
*foc
)
827 const struct inet_request_sock
*ireq
= inet_rsk(req
);
832 /* First, grab a route. */
833 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
836 skb
= tcp_make_synack(sk
, dst
, req
, foc
);
839 __tcp_v4_send_check(skb
, ireq
->ir_loc_addr
, ireq
->ir_rmt_addr
);
841 skb_set_queue_mapping(skb
, queue_mapping
);
842 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->ir_loc_addr
,
845 err
= net_xmit_eval(err
);
852 * IPv4 request_sock destructor.
854 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
856 kfree(inet_rsk(req
)->opt
);
860 #ifdef CONFIG_TCP_MD5SIG
862 * RFC2385 MD5 checksumming requires a mapping of
863 * IP address->MD5 Key.
864 * We need to maintain these in the sk structure.
867 /* Find the Key structure for an address. */
868 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
869 const union tcp_md5_addr
*addr
,
872 const struct tcp_sock
*tp
= tcp_sk(sk
);
873 struct tcp_md5sig_key
*key
;
874 unsigned int size
= sizeof(struct in_addr
);
875 const struct tcp_md5sig_info
*md5sig
;
877 /* caller either holds rcu_read_lock() or socket lock */
878 md5sig
= rcu_dereference_check(tp
->md5sig_info
,
879 sock_owned_by_user(sk
) ||
880 lockdep_is_held(&sk
->sk_lock
.slock
));
883 #if IS_ENABLED(CONFIG_IPV6)
884 if (family
== AF_INET6
)
885 size
= sizeof(struct in6_addr
);
887 hlist_for_each_entry_rcu(key
, &md5sig
->head
, node
) {
888 if (key
->family
!= family
)
890 if (!memcmp(&key
->addr
, addr
, size
))
895 EXPORT_SYMBOL(tcp_md5_do_lookup
);
897 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
898 const struct sock
*addr_sk
)
900 const union tcp_md5_addr
*addr
;
902 addr
= (const union tcp_md5_addr
*)&addr_sk
->sk_daddr
;
903 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
905 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
907 /* This can be called on a newly created socket, from other files */
908 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
909 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
)
911 /* Add Key to the list */
912 struct tcp_md5sig_key
*key
;
913 struct tcp_sock
*tp
= tcp_sk(sk
);
914 struct tcp_md5sig_info
*md5sig
;
916 key
= tcp_md5_do_lookup(sk
, addr
, family
);
918 /* Pre-existing entry - just update that one. */
919 memcpy(key
->key
, newkey
, newkeylen
);
920 key
->keylen
= newkeylen
;
924 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
925 sock_owned_by_user(sk
));
927 md5sig
= kmalloc(sizeof(*md5sig
), gfp
);
931 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
932 INIT_HLIST_HEAD(&md5sig
->head
);
933 rcu_assign_pointer(tp
->md5sig_info
, md5sig
);
936 key
= sock_kmalloc(sk
, sizeof(*key
), gfp
);
939 if (!tcp_alloc_md5sig_pool()) {
940 sock_kfree_s(sk
, key
, sizeof(*key
));
944 memcpy(key
->key
, newkey
, newkeylen
);
945 key
->keylen
= newkeylen
;
946 key
->family
= family
;
947 memcpy(&key
->addr
, addr
,
948 (family
== AF_INET6
) ? sizeof(struct in6_addr
) :
949 sizeof(struct in_addr
));
950 hlist_add_head_rcu(&key
->node
, &md5sig
->head
);
953 EXPORT_SYMBOL(tcp_md5_do_add
);
955 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
, int family
)
957 struct tcp_md5sig_key
*key
;
959 key
= tcp_md5_do_lookup(sk
, addr
, family
);
962 hlist_del_rcu(&key
->node
);
963 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
967 EXPORT_SYMBOL(tcp_md5_do_del
);
969 static void tcp_clear_md5_list(struct sock
*sk
)
971 struct tcp_sock
*tp
= tcp_sk(sk
);
972 struct tcp_md5sig_key
*key
;
973 struct hlist_node
*n
;
974 struct tcp_md5sig_info
*md5sig
;
976 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
978 hlist_for_each_entry_safe(key
, n
, &md5sig
->head
, node
) {
979 hlist_del_rcu(&key
->node
);
980 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
985 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
988 struct tcp_md5sig cmd
;
989 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
991 if (optlen
< sizeof(cmd
))
994 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
997 if (sin
->sin_family
!= AF_INET
)
1000 if (!cmd
.tcpm_keylen
)
1001 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1004 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1007 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1008 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1012 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1013 __be32 daddr
, __be32 saddr
, int nbytes
)
1015 struct tcp4_pseudohdr
*bp
;
1016 struct scatterlist sg
;
1018 bp
= &hp
->md5_blk
.ip4
;
1021 * 1. the TCP pseudo-header (in the order: source IP address,
1022 * destination IP address, zero-padded protocol number, and
1028 bp
->protocol
= IPPROTO_TCP
;
1029 bp
->len
= cpu_to_be16(nbytes
);
1031 sg_init_one(&sg
, bp
, sizeof(*bp
));
1032 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1035 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1036 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1038 struct tcp_md5sig_pool
*hp
;
1039 struct hash_desc
*desc
;
1041 hp
= tcp_get_md5sig_pool();
1043 goto clear_hash_noput
;
1044 desc
= &hp
->md5_desc
;
1046 if (crypto_hash_init(desc
))
1048 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1050 if (tcp_md5_hash_header(hp
, th
))
1052 if (tcp_md5_hash_key(hp
, key
))
1054 if (crypto_hash_final(desc
, md5_hash
))
1057 tcp_put_md5sig_pool();
1061 tcp_put_md5sig_pool();
1063 memset(md5_hash
, 0, 16);
1067 int tcp_v4_md5_hash_skb(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1068 const struct sock
*sk
,
1069 const struct sk_buff
*skb
)
1071 struct tcp_md5sig_pool
*hp
;
1072 struct hash_desc
*desc
;
1073 const struct tcphdr
*th
= tcp_hdr(skb
);
1074 __be32 saddr
, daddr
;
1076 if (sk
) { /* valid for establish/request sockets */
1077 saddr
= sk
->sk_rcv_saddr
;
1078 daddr
= sk
->sk_daddr
;
1080 const struct iphdr
*iph
= ip_hdr(skb
);
1085 hp
= tcp_get_md5sig_pool();
1087 goto clear_hash_noput
;
1088 desc
= &hp
->md5_desc
;
1090 if (crypto_hash_init(desc
))
1093 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1095 if (tcp_md5_hash_header(hp
, th
))
1097 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1099 if (tcp_md5_hash_key(hp
, key
))
1101 if (crypto_hash_final(desc
, md5_hash
))
1104 tcp_put_md5sig_pool();
1108 tcp_put_md5sig_pool();
1110 memset(md5_hash
, 0, 16);
1113 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1115 /* Called with rcu_read_lock() */
1116 static bool tcp_v4_inbound_md5_hash(struct sock
*sk
,
1117 const struct sk_buff
*skb
)
1120 * This gets called for each TCP segment that arrives
1121 * so we want to be efficient.
1122 * We have 3 drop cases:
1123 * o No MD5 hash and one expected.
1124 * o MD5 hash and we're not expecting one.
1125 * o MD5 hash and its wrong.
1127 const __u8
*hash_location
= NULL
;
1128 struct tcp_md5sig_key
*hash_expected
;
1129 const struct iphdr
*iph
= ip_hdr(skb
);
1130 const struct tcphdr
*th
= tcp_hdr(skb
);
1132 unsigned char newhash
[16];
1134 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1136 hash_location
= tcp_parse_md5sig_option(th
);
1138 /* We've parsed the options - do we have a hash? */
1139 if (!hash_expected
&& !hash_location
)
1142 if (hash_expected
&& !hash_location
) {
1143 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1147 if (!hash_expected
&& hash_location
) {
1148 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1152 /* Okay, so this is hash_expected and hash_location -
1153 * so we need to calculate the checksum.
1155 genhash
= tcp_v4_md5_hash_skb(newhash
,
1159 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1160 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1161 &iph
->saddr
, ntohs(th
->source
),
1162 &iph
->daddr
, ntohs(th
->dest
),
1163 genhash
? " tcp_v4_calc_md5_hash failed"
1171 static void tcp_v4_init_req(struct request_sock
*req
, struct sock
*sk_listener
,
1172 struct sk_buff
*skb
)
1174 struct inet_request_sock
*ireq
= inet_rsk(req
);
1176 sk_rcv_saddr_set(req_to_sk(req
), ip_hdr(skb
)->daddr
);
1177 sk_daddr_set(req_to_sk(req
), ip_hdr(skb
)->saddr
);
1178 ireq
->no_srccheck
= inet_sk(sk_listener
)->transparent
;
1179 ireq
->opt
= tcp_v4_save_options(skb
);
1182 static struct dst_entry
*tcp_v4_route_req(struct sock
*sk
, struct flowi
*fl
,
1183 const struct request_sock
*req
,
1186 struct dst_entry
*dst
= inet_csk_route_req(sk
, &fl
->u
.ip4
, req
);
1189 if (fl
->u
.ip4
.daddr
== inet_rsk(req
)->ir_rmt_addr
)
1198 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1200 .obj_size
= sizeof(struct tcp_request_sock
),
1201 .rtx_syn_ack
= tcp_rtx_synack
,
1202 .send_ack
= tcp_v4_reqsk_send_ack
,
1203 .destructor
= tcp_v4_reqsk_destructor
,
1204 .send_reset
= tcp_v4_send_reset
,
1205 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1208 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1209 .mss_clamp
= TCP_MSS_DEFAULT
,
1210 #ifdef CONFIG_TCP_MD5SIG
1211 .req_md5_lookup
= tcp_v4_md5_lookup
,
1212 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1214 .init_req
= tcp_v4_init_req
,
1215 #ifdef CONFIG_SYN_COOKIES
1216 .cookie_init_seq
= cookie_v4_init_sequence
,
1218 .route_req
= tcp_v4_route_req
,
1219 .init_seq
= tcp_v4_init_sequence
,
1220 .send_synack
= tcp_v4_send_synack
,
1221 .queue_hash_add
= inet_csk_reqsk_queue_hash_add
,
1224 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1226 /* Never answer to SYNs send to broadcast or multicast */
1227 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1230 return tcp_conn_request(&tcp_request_sock_ops
,
1231 &tcp_request_sock_ipv4_ops
, sk
, skb
);
1234 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1237 EXPORT_SYMBOL(tcp_v4_conn_request
);
1241 * The three way handshake has completed - we got a valid synack -
1242 * now create the new socket.
1244 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1245 struct request_sock
*req
,
1246 struct dst_entry
*dst
)
1248 struct inet_request_sock
*ireq
;
1249 struct inet_sock
*newinet
;
1250 struct tcp_sock
*newtp
;
1252 #ifdef CONFIG_TCP_MD5SIG
1253 struct tcp_md5sig_key
*key
;
1255 struct ip_options_rcu
*inet_opt
;
1257 if (sk_acceptq_is_full(sk
))
1260 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1264 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1265 inet_sk_rx_dst_set(newsk
, skb
);
1267 newtp
= tcp_sk(newsk
);
1268 newinet
= inet_sk(newsk
);
1269 ireq
= inet_rsk(req
);
1270 sk_daddr_set(newsk
, ireq
->ir_rmt_addr
);
1271 sk_rcv_saddr_set(newsk
, ireq
->ir_loc_addr
);
1272 newinet
->inet_saddr
= ireq
->ir_loc_addr
;
1273 inet_opt
= ireq
->opt
;
1274 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1276 newinet
->mc_index
= inet_iif(skb
);
1277 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1278 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1279 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1281 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1282 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1285 dst
= inet_csk_route_child_sock(sk
, newsk
, req
);
1289 /* syncookie case : see end of cookie_v4_check() */
1291 sk_setup_caps(newsk
, dst
);
1293 tcp_ca_openreq_child(newsk
, dst
);
1295 tcp_sync_mss(newsk
, dst_mtu(dst
));
1296 newtp
->advmss
= dst_metric_advmss(dst
);
1297 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1298 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1299 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1301 tcp_initialize_rcv_mss(newsk
);
1303 #ifdef CONFIG_TCP_MD5SIG
1304 /* Copy over the MD5 key from the original socket */
1305 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1309 * We're using one, so create a matching key
1310 * on the newsk structure. If we fail to get
1311 * memory, then we end up not copying the key
1314 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1315 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1316 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1320 if (__inet_inherit_port(sk
, newsk
) < 0)
1322 __inet_hash_nolisten(newsk
, NULL
);
1327 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1331 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1334 inet_csk_prepare_forced_close(newsk
);
1338 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1340 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1342 const struct tcphdr
*th
= tcp_hdr(skb
);
1343 const struct iphdr
*iph
= ip_hdr(skb
);
1344 struct request_sock
*req
;
1347 req
= inet_csk_search_req(sk
, th
->source
, iph
->saddr
, iph
->daddr
);
1349 nsk
= tcp_check_req(sk
, skb
, req
, false);
1350 if (!nsk
|| nsk
== sk
)
1355 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1356 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1359 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1363 inet_twsk_put(inet_twsk(nsk
));
1367 #ifdef CONFIG_SYN_COOKIES
1369 sk
= cookie_v4_check(sk
, skb
);
1374 /* The socket must have it's spinlock held when we get
1377 * We have a potential double-lock case here, so even when
1378 * doing backlog processing we use the BH locking scheme.
1379 * This is because we cannot sleep with the original spinlock
1382 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1386 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1387 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1389 sock_rps_save_rxhash(sk
, skb
);
1390 sk_mark_napi_id(sk
, skb
);
1392 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1393 !dst
->ops
->check(dst
, 0)) {
1395 sk
->sk_rx_dst
= NULL
;
1398 tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
);
1402 if (tcp_checksum_complete(skb
))
1405 if (sk
->sk_state
== TCP_LISTEN
) {
1406 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1411 sock_rps_save_rxhash(nsk
, skb
);
1412 sk_mark_napi_id(sk
, skb
);
1413 if (tcp_child_process(sk
, nsk
, skb
)) {
1420 sock_rps_save_rxhash(sk
, skb
);
1422 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1429 tcp_v4_send_reset(rsk
, skb
);
1432 /* Be careful here. If this function gets more complicated and
1433 * gcc suffers from register pressure on the x86, sk (in %ebx)
1434 * might be destroyed here. This current version compiles correctly,
1435 * but you have been warned.
1440 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1441 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1444 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1446 void tcp_v4_early_demux(struct sk_buff
*skb
)
1448 const struct iphdr
*iph
;
1449 const struct tcphdr
*th
;
1452 if (skb
->pkt_type
!= PACKET_HOST
)
1455 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1461 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1464 sk
= __inet_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1465 iph
->saddr
, th
->source
,
1466 iph
->daddr
, ntohs(th
->dest
),
1470 skb
->destructor
= sock_edemux
;
1471 if (sk_fullsock(sk
)) {
1472 struct dst_entry
*dst
= READ_ONCE(sk
->sk_rx_dst
);
1475 dst
= dst_check(dst
, 0);
1477 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1478 skb_dst_set_noref(skb
, dst
);
1483 /* Packet is added to VJ-style prequeue for processing in process
1484 * context, if a reader task is waiting. Apparently, this exciting
1485 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1486 * failed somewhere. Latency? Burstiness? Well, at least now we will
1487 * see, why it failed. 8)8) --ANK
1490 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1492 struct tcp_sock
*tp
= tcp_sk(sk
);
1494 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
1497 if (skb
->len
<= tcp_hdrlen(skb
) &&
1498 skb_queue_len(&tp
->ucopy
.prequeue
) == 0)
1501 /* Before escaping RCU protected region, we need to take care of skb
1502 * dst. Prequeue is only enabled for established sockets.
1503 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1504 * Instead of doing full sk_rx_dst validity here, let's perform
1505 * an optimistic check.
1507 if (likely(sk
->sk_rx_dst
))
1512 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1513 tp
->ucopy
.memory
+= skb
->truesize
;
1514 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1515 struct sk_buff
*skb1
;
1517 BUG_ON(sock_owned_by_user(sk
));
1519 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1520 sk_backlog_rcv(sk
, skb1
);
1521 NET_INC_STATS_BH(sock_net(sk
),
1522 LINUX_MIB_TCPPREQUEUEDROPPED
);
1525 tp
->ucopy
.memory
= 0;
1526 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1527 wake_up_interruptible_sync_poll(sk_sleep(sk
),
1528 POLLIN
| POLLRDNORM
| POLLRDBAND
);
1529 if (!inet_csk_ack_scheduled(sk
))
1530 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
1531 (3 * tcp_rto_min(sk
)) / 4,
1536 EXPORT_SYMBOL(tcp_prequeue
);
1542 int tcp_v4_rcv(struct sk_buff
*skb
)
1544 const struct iphdr
*iph
;
1545 const struct tcphdr
*th
;
1548 struct net
*net
= dev_net(skb
->dev
);
1550 if (skb
->pkt_type
!= PACKET_HOST
)
1553 /* Count it even if it's bad */
1554 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1556 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1561 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1563 if (!pskb_may_pull(skb
, th
->doff
* 4))
1566 /* An explanation is required here, I think.
1567 * Packet length and doff are validated by header prediction,
1568 * provided case of th->doff==0 is eliminated.
1569 * So, we defer the checks. */
1571 if (skb_checksum_init(skb
, IPPROTO_TCP
, inet_compute_pseudo
))
1576 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1577 * barrier() makes sure compiler wont play fool^Waliasing games.
1579 memmove(&TCP_SKB_CB(skb
)->header
.h4
, IPCB(skb
),
1580 sizeof(struct inet_skb_parm
));
1583 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1584 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1585 skb
->len
- th
->doff
* 4);
1586 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1587 TCP_SKB_CB(skb
)->tcp_flags
= tcp_flag_byte(th
);
1588 TCP_SKB_CB(skb
)->tcp_tw_isn
= 0;
1589 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
1590 TCP_SKB_CB(skb
)->sacked
= 0;
1592 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1597 if (sk
->sk_state
== TCP_TIME_WAIT
)
1600 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1601 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1602 goto discard_and_relse
;
1605 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1606 goto discard_and_relse
;
1608 #ifdef CONFIG_TCP_MD5SIG
1610 * We really want to reject the packet as early as possible
1612 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1613 * o There is an MD5 option and we're not expecting one
1615 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1616 goto discard_and_relse
;
1621 if (sk_filter(sk
, skb
))
1622 goto discard_and_relse
;
1624 sk_incoming_cpu_update(sk
);
1627 bh_lock_sock_nested(sk
);
1628 tcp_sk(sk
)->segs_in
+= max_t(u16
, 1, skb_shinfo(skb
)->gso_segs
);
1630 if (!sock_owned_by_user(sk
)) {
1631 if (!tcp_prequeue(sk
, skb
))
1632 ret
= tcp_v4_do_rcv(sk
, skb
);
1633 } else if (unlikely(sk_add_backlog(sk
, skb
,
1634 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
1636 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1637 goto discard_and_relse
;
1646 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1649 if (tcp_checksum_complete(skb
)) {
1651 TCP_INC_STATS_BH(net
, TCP_MIB_CSUMERRORS
);
1653 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1655 tcp_v4_send_reset(NULL
, skb
);
1659 /* Discard frame. */
1668 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1669 inet_twsk_put(inet_twsk(sk
));
1673 if (tcp_checksum_complete(skb
)) {
1674 inet_twsk_put(inet_twsk(sk
));
1677 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1679 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1681 iph
->saddr
, th
->source
,
1682 iph
->daddr
, th
->dest
,
1685 inet_twsk_deschedule_put(inet_twsk(sk
));
1689 /* Fall through to ACK */
1692 tcp_v4_timewait_ack(sk
, skb
);
1696 case TCP_TW_SUCCESS
:;
1701 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1702 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1703 .twsk_unique
= tcp_twsk_unique
,
1704 .twsk_destructor
= tcp_twsk_destructor
,
1707 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
1709 struct dst_entry
*dst
= skb_dst(skb
);
1713 sk
->sk_rx_dst
= dst
;
1714 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
1717 EXPORT_SYMBOL(inet_sk_rx_dst_set
);
1719 const struct inet_connection_sock_af_ops ipv4_specific
= {
1720 .queue_xmit
= ip_queue_xmit
,
1721 .send_check
= tcp_v4_send_check
,
1722 .rebuild_header
= inet_sk_rebuild_header
,
1723 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
1724 .conn_request
= tcp_v4_conn_request
,
1725 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1726 .net_header_len
= sizeof(struct iphdr
),
1727 .setsockopt
= ip_setsockopt
,
1728 .getsockopt
= ip_getsockopt
,
1729 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1730 .sockaddr_len
= sizeof(struct sockaddr_in
),
1731 .bind_conflict
= inet_csk_bind_conflict
,
1732 #ifdef CONFIG_COMPAT
1733 .compat_setsockopt
= compat_ip_setsockopt
,
1734 .compat_getsockopt
= compat_ip_getsockopt
,
1736 .mtu_reduced
= tcp_v4_mtu_reduced
,
1738 EXPORT_SYMBOL(ipv4_specific
);
1740 #ifdef CONFIG_TCP_MD5SIG
1741 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1742 .md5_lookup
= tcp_v4_md5_lookup
,
1743 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1744 .md5_parse
= tcp_v4_parse_md5_keys
,
1748 /* NOTE: A lot of things set to zero explicitly by call to
1749 * sk_alloc() so need not be done here.
1751 static int tcp_v4_init_sock(struct sock
*sk
)
1753 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1757 icsk
->icsk_af_ops
= &ipv4_specific
;
1759 #ifdef CONFIG_TCP_MD5SIG
1760 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv4_specific
;
1766 void tcp_v4_destroy_sock(struct sock
*sk
)
1768 struct tcp_sock
*tp
= tcp_sk(sk
);
1770 tcp_clear_xmit_timers(sk
);
1772 tcp_cleanup_congestion_control(sk
);
1774 /* Cleanup up the write buffer. */
1775 tcp_write_queue_purge(sk
);
1777 /* Cleans up our, hopefully empty, out_of_order_queue. */
1778 __skb_queue_purge(&tp
->out_of_order_queue
);
1780 #ifdef CONFIG_TCP_MD5SIG
1781 /* Clean up the MD5 key list, if any */
1782 if (tp
->md5sig_info
) {
1783 tcp_clear_md5_list(sk
);
1784 kfree_rcu(tp
->md5sig_info
, rcu
);
1785 tp
->md5sig_info
= NULL
;
1789 /* Clean prequeue, it must be empty really */
1790 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1792 /* Clean up a referenced TCP bind bucket. */
1793 if (inet_csk(sk
)->icsk_bind_hash
)
1796 BUG_ON(tp
->fastopen_rsk
);
1798 /* If socket is aborted during connect operation */
1799 tcp_free_fastopen_req(tp
);
1800 tcp_saved_syn_free(tp
);
1802 sk_sockets_allocated_dec(sk
);
1803 sock_release_memcg(sk
);
1805 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1807 #ifdef CONFIG_PROC_FS
1808 /* Proc filesystem TCP sock list dumping. */
1811 * Get next listener socket follow cur. If cur is NULL, get first socket
1812 * starting from bucket given in st->bucket; when st->bucket is zero the
1813 * very first socket in the hash table is returned.
1815 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
1817 struct inet_connection_sock
*icsk
;
1818 struct hlist_nulls_node
*node
;
1819 struct sock
*sk
= cur
;
1820 struct inet_listen_hashbucket
*ilb
;
1821 struct tcp_iter_state
*st
= seq
->private;
1822 struct net
*net
= seq_file_net(seq
);
1825 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1826 spin_lock_bh(&ilb
->lock
);
1827 sk
= sk_nulls_head(&ilb
->head
);
1831 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1835 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
1836 struct request_sock
*req
= cur
;
1838 icsk
= inet_csk(st
->syn_wait_sk
);
1842 if (req
->rsk_ops
->family
== st
->family
) {
1848 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
1851 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
1853 sk
= sk_nulls_next(st
->syn_wait_sk
);
1854 st
->state
= TCP_SEQ_STATE_LISTENING
;
1855 spin_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1857 icsk
= inet_csk(sk
);
1858 spin_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1859 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
1861 spin_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1862 sk
= sk_nulls_next(sk
);
1865 sk_nulls_for_each_from(sk
, node
) {
1866 if (!net_eq(sock_net(sk
), net
))
1868 if (sk
->sk_family
== st
->family
) {
1872 icsk
= inet_csk(sk
);
1873 spin_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1874 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
1876 st
->uid
= sock_i_uid(sk
);
1877 st
->syn_wait_sk
= sk
;
1878 st
->state
= TCP_SEQ_STATE_OPENREQ
;
1882 spin_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
1884 spin_unlock_bh(&ilb
->lock
);
1886 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
1887 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1888 spin_lock_bh(&ilb
->lock
);
1889 sk
= sk_nulls_head(&ilb
->head
);
1897 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
1899 struct tcp_iter_state
*st
= seq
->private;
1904 rc
= listening_get_next(seq
, NULL
);
1906 while (rc
&& *pos
) {
1907 rc
= listening_get_next(seq
, rc
);
1913 static inline bool empty_bucket(const struct tcp_iter_state
*st
)
1915 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
1919 * Get first established socket starting from bucket given in st->bucket.
1920 * If st->bucket is zero, the very first socket in the hash is returned.
1922 static void *established_get_first(struct seq_file
*seq
)
1924 struct tcp_iter_state
*st
= seq
->private;
1925 struct net
*net
= seq_file_net(seq
);
1929 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
1931 struct hlist_nulls_node
*node
;
1932 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
1934 /* Lockless fast path for the common case of empty buckets */
1935 if (empty_bucket(st
))
1939 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
1940 if (sk
->sk_family
!= st
->family
||
1941 !net_eq(sock_net(sk
), net
)) {
1947 spin_unlock_bh(lock
);
1953 static void *established_get_next(struct seq_file
*seq
, void *cur
)
1955 struct sock
*sk
= cur
;
1956 struct hlist_nulls_node
*node
;
1957 struct tcp_iter_state
*st
= seq
->private;
1958 struct net
*net
= seq_file_net(seq
);
1963 sk
= sk_nulls_next(sk
);
1965 sk_nulls_for_each_from(sk
, node
) {
1966 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
1970 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
1972 return established_get_first(seq
);
1975 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
1977 struct tcp_iter_state
*st
= seq
->private;
1981 rc
= established_get_first(seq
);
1984 rc
= established_get_next(seq
, rc
);
1990 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
1993 struct tcp_iter_state
*st
= seq
->private;
1995 st
->state
= TCP_SEQ_STATE_LISTENING
;
1996 rc
= listening_get_idx(seq
, &pos
);
1999 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2000 rc
= established_get_idx(seq
, pos
);
2006 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2008 struct tcp_iter_state
*st
= seq
->private;
2009 int offset
= st
->offset
;
2010 int orig_num
= st
->num
;
2013 switch (st
->state
) {
2014 case TCP_SEQ_STATE_OPENREQ
:
2015 case TCP_SEQ_STATE_LISTENING
:
2016 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2018 st
->state
= TCP_SEQ_STATE_LISTENING
;
2019 rc
= listening_get_next(seq
, NULL
);
2020 while (offset
-- && rc
)
2021 rc
= listening_get_next(seq
, rc
);
2025 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2027 case TCP_SEQ_STATE_ESTABLISHED
:
2028 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2030 rc
= established_get_first(seq
);
2031 while (offset
-- && rc
)
2032 rc
= established_get_next(seq
, rc
);
2040 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2042 struct tcp_iter_state
*st
= seq
->private;
2045 if (*pos
&& *pos
== st
->last_pos
) {
2046 rc
= tcp_seek_last_pos(seq
);
2051 st
->state
= TCP_SEQ_STATE_LISTENING
;
2055 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2058 st
->last_pos
= *pos
;
2062 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2064 struct tcp_iter_state
*st
= seq
->private;
2067 if (v
== SEQ_START_TOKEN
) {
2068 rc
= tcp_get_idx(seq
, 0);
2072 switch (st
->state
) {
2073 case TCP_SEQ_STATE_OPENREQ
:
2074 case TCP_SEQ_STATE_LISTENING
:
2075 rc
= listening_get_next(seq
, v
);
2077 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2080 rc
= established_get_first(seq
);
2083 case TCP_SEQ_STATE_ESTABLISHED
:
2084 rc
= established_get_next(seq
, v
);
2089 st
->last_pos
= *pos
;
2093 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2095 struct tcp_iter_state
*st
= seq
->private;
2097 switch (st
->state
) {
2098 case TCP_SEQ_STATE_OPENREQ
:
2100 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2101 spin_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2103 case TCP_SEQ_STATE_LISTENING
:
2104 if (v
!= SEQ_START_TOKEN
)
2105 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2107 case TCP_SEQ_STATE_ESTABLISHED
:
2109 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2114 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2116 struct tcp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2117 struct tcp_iter_state
*s
;
2120 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2121 sizeof(struct tcp_iter_state
));
2125 s
= ((struct seq_file
*)file
->private_data
)->private;
2126 s
->family
= afinfo
->family
;
2130 EXPORT_SYMBOL(tcp_seq_open
);
2132 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2135 struct proc_dir_entry
*p
;
2137 afinfo
->seq_ops
.start
= tcp_seq_start
;
2138 afinfo
->seq_ops
.next
= tcp_seq_next
;
2139 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2141 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2142 afinfo
->seq_fops
, afinfo
);
2147 EXPORT_SYMBOL(tcp_proc_register
);
2149 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2151 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2153 EXPORT_SYMBOL(tcp_proc_unregister
);
2155 static void get_openreq4(const struct request_sock
*req
,
2156 struct seq_file
*f
, int i
, kuid_t uid
)
2158 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2159 long delta
= req
->rsk_timer
.expires
- jiffies
;
2161 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2162 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2167 ntohs(ireq
->ir_rmt_port
),
2169 0, 0, /* could print option size, but that is af dependent. */
2170 1, /* timers active (only the expire timer) */
2171 jiffies_delta_to_clock_t(delta
),
2173 from_kuid_munged(seq_user_ns(f
), uid
),
2174 0, /* non standard timer */
2175 0, /* open_requests have no inode */
2180 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
)
2183 unsigned long timer_expires
;
2184 const struct tcp_sock
*tp
= tcp_sk(sk
);
2185 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2186 const struct inet_sock
*inet
= inet_sk(sk
);
2187 struct fastopen_queue
*fastopenq
= icsk
->icsk_accept_queue
.fastopenq
;
2188 __be32 dest
= inet
->inet_daddr
;
2189 __be32 src
= inet
->inet_rcv_saddr
;
2190 __u16 destp
= ntohs(inet
->inet_dport
);
2191 __u16 srcp
= ntohs(inet
->inet_sport
);
2194 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
2195 icsk
->icsk_pending
== ICSK_TIME_EARLY_RETRANS
||
2196 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
2198 timer_expires
= icsk
->icsk_timeout
;
2199 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2201 timer_expires
= icsk
->icsk_timeout
;
2202 } else if (timer_pending(&sk
->sk_timer
)) {
2204 timer_expires
= sk
->sk_timer
.expires
;
2207 timer_expires
= jiffies
;
2210 if (sk
->sk_state
== TCP_LISTEN
)
2211 rx_queue
= sk
->sk_ack_backlog
;
2214 * because we dont lock socket, we might find a transient negative value
2216 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2218 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2219 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2220 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2221 tp
->write_seq
- tp
->snd_una
,
2224 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
2225 icsk
->icsk_retransmits
,
2226 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sk
)),
2227 icsk
->icsk_probes_out
,
2229 atomic_read(&sk
->sk_refcnt
), sk
,
2230 jiffies_to_clock_t(icsk
->icsk_rto
),
2231 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2232 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2234 sk
->sk_state
== TCP_LISTEN
?
2235 (fastopenq
? fastopenq
->max_qlen
: 0) :
2236 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
));
2239 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2240 struct seq_file
*f
, int i
)
2242 long delta
= tw
->tw_timer
.expires
- jiffies
;
2246 dest
= tw
->tw_daddr
;
2247 src
= tw
->tw_rcv_saddr
;
2248 destp
= ntohs(tw
->tw_dport
);
2249 srcp
= ntohs(tw
->tw_sport
);
2251 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2252 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2253 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2254 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
2255 atomic_read(&tw
->tw_refcnt
), tw
);
2260 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2262 struct tcp_iter_state
*st
;
2263 struct sock
*sk
= v
;
2265 seq_setwidth(seq
, TMPSZ
- 1);
2266 if (v
== SEQ_START_TOKEN
) {
2267 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2268 "rx_queue tr tm->when retrnsmt uid timeout "
2274 switch (st
->state
) {
2275 case TCP_SEQ_STATE_LISTENING
:
2276 case TCP_SEQ_STATE_ESTABLISHED
:
2277 if (sk
->sk_state
== TCP_TIME_WAIT
)
2278 get_timewait4_sock(v
, seq
, st
->num
);
2280 get_tcp4_sock(v
, seq
, st
->num
);
2282 case TCP_SEQ_STATE_OPENREQ
:
2283 get_openreq4(v
, seq
, st
->num
, st
->uid
);
2291 static const struct file_operations tcp_afinfo_seq_fops
= {
2292 .owner
= THIS_MODULE
,
2293 .open
= tcp_seq_open
,
2295 .llseek
= seq_lseek
,
2296 .release
= seq_release_net
2299 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2302 .seq_fops
= &tcp_afinfo_seq_fops
,
2304 .show
= tcp4_seq_show
,
2308 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2310 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2313 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2315 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2318 static struct pernet_operations tcp4_net_ops
= {
2319 .init
= tcp4_proc_init_net
,
2320 .exit
= tcp4_proc_exit_net
,
2323 int __init
tcp4_proc_init(void)
2325 return register_pernet_subsys(&tcp4_net_ops
);
2328 void tcp4_proc_exit(void)
2330 unregister_pernet_subsys(&tcp4_net_ops
);
2332 #endif /* CONFIG_PROC_FS */
2334 struct proto tcp_prot
= {
2336 .owner
= THIS_MODULE
,
2338 .connect
= tcp_v4_connect
,
2339 .disconnect
= tcp_disconnect
,
2340 .accept
= inet_csk_accept
,
2342 .init
= tcp_v4_init_sock
,
2343 .destroy
= tcp_v4_destroy_sock
,
2344 .shutdown
= tcp_shutdown
,
2345 .setsockopt
= tcp_setsockopt
,
2346 .getsockopt
= tcp_getsockopt
,
2347 .recvmsg
= tcp_recvmsg
,
2348 .sendmsg
= tcp_sendmsg
,
2349 .sendpage
= tcp_sendpage
,
2350 .backlog_rcv
= tcp_v4_do_rcv
,
2351 .release_cb
= tcp_release_cb
,
2353 .unhash
= inet_unhash
,
2354 .get_port
= inet_csk_get_port
,
2355 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2356 .stream_memory_free
= tcp_stream_memory_free
,
2357 .sockets_allocated
= &tcp_sockets_allocated
,
2358 .orphan_count
= &tcp_orphan_count
,
2359 .memory_allocated
= &tcp_memory_allocated
,
2360 .memory_pressure
= &tcp_memory_pressure
,
2361 .sysctl_mem
= sysctl_tcp_mem
,
2362 .sysctl_wmem
= sysctl_tcp_wmem
,
2363 .sysctl_rmem
= sysctl_tcp_rmem
,
2364 .max_header
= MAX_TCP_HEADER
,
2365 .obj_size
= sizeof(struct tcp_sock
),
2366 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2367 .twsk_prot
= &tcp_timewait_sock_ops
,
2368 .rsk_prot
= &tcp_request_sock_ops
,
2369 .h
.hashinfo
= &tcp_hashinfo
,
2370 .no_autobind
= true,
2371 #ifdef CONFIG_COMPAT
2372 .compat_setsockopt
= compat_tcp_setsockopt
,
2373 .compat_getsockopt
= compat_tcp_getsockopt
,
2375 #ifdef CONFIG_MEMCG_KMEM
2376 .init_cgroup
= tcp_init_cgroup
,
2377 .destroy_cgroup
= tcp_destroy_cgroup
,
2378 .proto_cgroup
= tcp_proto_cgroup
,
2381 EXPORT_SYMBOL(tcp_prot
);
2383 static void __net_exit
tcp_sk_exit(struct net
*net
)
2387 for_each_possible_cpu(cpu
)
2388 inet_ctl_sock_destroy(*per_cpu_ptr(net
->ipv4
.tcp_sk
, cpu
));
2389 free_percpu(net
->ipv4
.tcp_sk
);
2392 static int __net_init
tcp_sk_init(struct net
*net
)
2396 net
->ipv4
.tcp_sk
= alloc_percpu(struct sock
*);
2397 if (!net
->ipv4
.tcp_sk
)
2400 for_each_possible_cpu(cpu
) {
2403 res
= inet_ctl_sock_create(&sk
, PF_INET
, SOCK_RAW
,
2407 *per_cpu_ptr(net
->ipv4
.tcp_sk
, cpu
) = sk
;
2410 net
->ipv4
.sysctl_tcp_ecn
= 2;
2411 net
->ipv4
.sysctl_tcp_ecn_fallback
= 1;
2413 net
->ipv4
.sysctl_tcp_base_mss
= TCP_BASE_MSS
;
2414 net
->ipv4
.sysctl_tcp_probe_threshold
= TCP_PROBE_THRESHOLD
;
2415 net
->ipv4
.sysctl_tcp_probe_interval
= TCP_PROBE_INTERVAL
;
2424 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2426 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2429 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2430 .init
= tcp_sk_init
,
2431 .exit
= tcp_sk_exit
,
2432 .exit_batch
= tcp_sk_exit_batch
,
2435 void __init
tcp_v4_init(void)
2437 inet_hashinfo_init(&tcp_hashinfo
);
2438 if (register_pernet_subsys(&tcp_sk_ops
))
2439 panic("Failed to create the TCP control socket.\n");