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 inline __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
;
211 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
213 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
215 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
217 /* Socket identity is still unknown (sport may be zero).
218 * However we set state to SYN-SENT and not releasing socket
219 * lock select source port, enter ourselves into the hash tables and
220 * complete initialization after this.
222 tcp_set_state(sk
, TCP_SYN_SENT
);
223 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 static 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 */
304 static void do_redirect(struct sk_buff
*skb
, struct sock
*sk
)
306 struct dst_entry
*dst
= __sk_dst_check(sk
, 0);
309 dst
->ops
->redirect(dst
, sk
, skb
);
313 * This routine is called by the ICMP module when it gets some
314 * sort of error condition. If err < 0 then the socket should
315 * be closed and the error returned to the user. If err > 0
316 * it's just the icmp type << 8 | icmp code. After adjustment
317 * header points to the first 8 bytes of the tcp header. We need
318 * to find the appropriate port.
320 * The locking strategy used here is very "optimistic". When
321 * someone else accesses the socket the ICMP is just dropped
322 * and for some paths there is no check at all.
323 * A more general error queue to queue errors for later handling
324 * is probably better.
328 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
330 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
331 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
332 struct inet_connection_sock
*icsk
;
334 struct inet_sock
*inet
;
335 const int type
= icmp_hdr(icmp_skb
)->type
;
336 const int code
= icmp_hdr(icmp_skb
)->code
;
339 struct request_sock
*req
;
343 struct net
*net
= dev_net(icmp_skb
->dev
);
345 if (icmp_skb
->len
< (iph
->ihl
<< 2) + 8) {
346 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
350 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
351 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
353 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
356 if (sk
->sk_state
== TCP_TIME_WAIT
) {
357 inet_twsk_put(inet_twsk(sk
));
362 /* If too many ICMPs get dropped on busy
363 * servers this needs to be solved differently.
364 * We do take care of PMTU discovery (RFC1191) special case :
365 * we can receive locally generated ICMP messages while socket is held.
367 if (sock_owned_by_user(sk
)) {
368 if (!(type
== ICMP_DEST_UNREACH
&& code
== ICMP_FRAG_NEEDED
))
369 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
371 if (sk
->sk_state
== TCP_CLOSE
)
374 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
375 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
381 req
= tp
->fastopen_rsk
;
382 seq
= ntohl(th
->seq
);
383 if (sk
->sk_state
!= TCP_LISTEN
&&
384 !between(seq
, tp
->snd_una
, tp
->snd_nxt
) &&
385 (req
== NULL
|| seq
!= tcp_rsk(req
)->snt_isn
)) {
386 /* For a Fast Open socket, allow seq to be snt_isn. */
387 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
393 do_redirect(icmp_skb
, sk
);
395 case ICMP_SOURCE_QUENCH
:
396 /* Just silently ignore these. */
398 case ICMP_PARAMETERPROB
:
401 case ICMP_DEST_UNREACH
:
402 if (code
> NR_ICMP_UNREACH
)
405 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
406 /* We are not interested in TCP_LISTEN and open_requests
407 * (SYN-ACKs send out by Linux are always <576bytes so
408 * they should go through unfragmented).
410 if (sk
->sk_state
== TCP_LISTEN
)
414 if (!sock_owned_by_user(sk
)) {
415 tcp_v4_mtu_reduced(sk
);
417 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED
, &tp
->tsq_flags
))
423 err
= icmp_err_convert
[code
].errno
;
424 /* check if icmp_skb allows revert of backoff
425 * (see draft-zimmermann-tcp-lcd) */
426 if (code
!= ICMP_NET_UNREACH
&& code
!= ICMP_HOST_UNREACH
)
428 if (seq
!= tp
->snd_una
|| !icsk
->icsk_retransmits
||
432 /* XXX (TFO) - revisit the following logic for TFO */
434 if (sock_owned_by_user(sk
))
437 icsk
->icsk_backoff
--;
438 inet_csk(sk
)->icsk_rto
= (tp
->srtt
? __tcp_set_rto(tp
) :
439 TCP_TIMEOUT_INIT
) << icsk
->icsk_backoff
;
442 skb
= tcp_write_queue_head(sk
);
445 remaining
= icsk
->icsk_rto
- min(icsk
->icsk_rto
,
446 tcp_time_stamp
- TCP_SKB_CB(skb
)->when
);
449 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
450 remaining
, TCP_RTO_MAX
);
452 /* RTO revert clocked out retransmission.
453 * Will retransmit now */
454 tcp_retransmit_timer(sk
);
458 case ICMP_TIME_EXCEEDED
:
465 /* XXX (TFO) - if it's a TFO socket and has been accepted, rather
466 * than following the TCP_SYN_RECV case and closing the socket,
467 * we ignore the ICMP error and keep trying like a fully established
468 * socket. Is this the right thing to do?
470 if (req
&& req
->sk
== NULL
)
473 switch (sk
->sk_state
) {
474 struct request_sock
*req
, **prev
;
476 if (sock_owned_by_user(sk
))
479 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
480 iph
->daddr
, iph
->saddr
);
484 /* ICMPs are not backlogged, hence we cannot get
485 an established socket here.
489 if (seq
!= tcp_rsk(req
)->snt_isn
) {
490 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
495 * Still in SYN_RECV, just remove it silently.
496 * There is no good way to pass the error to the newly
497 * created socket, and POSIX does not want network
498 * errors returned from accept().
500 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
501 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
505 case TCP_SYN_RECV
: /* Cannot happen.
506 It can f.e. if SYNs crossed,
509 if (!sock_owned_by_user(sk
)) {
512 sk
->sk_error_report(sk
);
516 sk
->sk_err_soft
= err
;
521 /* If we've already connected we will keep trying
522 * until we time out, or the user gives up.
524 * rfc1122 4.2.3.9 allows to consider as hard errors
525 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
526 * but it is obsoleted by pmtu discovery).
528 * Note, that in modern internet, where routing is unreliable
529 * and in each dark corner broken firewalls sit, sending random
530 * errors ordered by their masters even this two messages finally lose
531 * their original sense (even Linux sends invalid PORT_UNREACHs)
533 * Now we are in compliance with RFCs.
538 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
540 sk
->sk_error_report(sk
);
541 } else { /* Only an error on timeout */
542 sk
->sk_err_soft
= err
;
550 void __tcp_v4_send_check(struct sk_buff
*skb
, __be32 saddr
, __be32 daddr
)
552 struct tcphdr
*th
= tcp_hdr(skb
);
554 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
555 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
556 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
557 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
559 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
566 /* This routine computes an IPv4 TCP checksum. */
567 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
569 const struct inet_sock
*inet
= inet_sk(sk
);
571 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
573 EXPORT_SYMBOL(tcp_v4_send_check
);
576 * This routine will send an RST to the other tcp.
578 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
580 * Answer: if a packet caused RST, it is not for a socket
581 * existing in our system, if it is matched to a socket,
582 * it is just duplicate segment or bug in other side's TCP.
583 * So that we build reply only basing on parameters
584 * arrived with segment.
585 * Exception: precedence violation. We do not implement it in any case.
588 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
590 const struct tcphdr
*th
= tcp_hdr(skb
);
593 #ifdef CONFIG_TCP_MD5SIG
594 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
597 struct ip_reply_arg arg
;
598 #ifdef CONFIG_TCP_MD5SIG
599 struct tcp_md5sig_key
*key
;
600 const __u8
*hash_location
= NULL
;
601 unsigned char newhash
[16];
603 struct sock
*sk1
= NULL
;
607 /* Never send a reset in response to a reset. */
611 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
614 /* Swap the send and the receive. */
615 memset(&rep
, 0, sizeof(rep
));
616 rep
.th
.dest
= th
->source
;
617 rep
.th
.source
= th
->dest
;
618 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
622 rep
.th
.seq
= th
->ack_seq
;
625 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
626 skb
->len
- (th
->doff
<< 2));
629 memset(&arg
, 0, sizeof(arg
));
630 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
631 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
633 #ifdef CONFIG_TCP_MD5SIG
634 hash_location
= tcp_parse_md5sig_option(th
);
635 if (!sk
&& hash_location
) {
637 * active side is lost. Try to find listening socket through
638 * source port, and then find md5 key through listening socket.
639 * we are not loose security here:
640 * Incoming packet is checked with md5 hash with finding key,
641 * no RST generated if md5 hash doesn't match.
643 sk1
= __inet_lookup_listener(dev_net(skb_dst(skb
)->dev
),
644 &tcp_hashinfo
, ip_hdr(skb
)->saddr
,
645 th
->source
, ip_hdr(skb
)->daddr
,
646 ntohs(th
->source
), inet_iif(skb
));
647 /* don't send rst if it can't find key */
651 key
= tcp_md5_do_lookup(sk1
, (union tcp_md5_addr
*)
652 &ip_hdr(skb
)->saddr
, AF_INET
);
656 genhash
= tcp_v4_md5_hash_skb(newhash
, key
, NULL
, NULL
, skb
);
657 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
660 key
= sk
? tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)
666 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
668 (TCPOPT_MD5SIG
<< 8) |
670 /* Update length and the length the header thinks exists */
671 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
672 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
674 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
675 key
, ip_hdr(skb
)->saddr
,
676 ip_hdr(skb
)->daddr
, &rep
.th
);
679 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
680 ip_hdr(skb
)->saddr
, /* XXX */
681 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
682 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
683 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
684 /* When socket is gone, all binding information is lost.
685 * routing might fail in this case. No choice here, if we choose to force
686 * input interface, we will misroute in case of asymmetric route.
689 arg
.bound_dev_if
= sk
->sk_bound_dev_if
;
691 net
= dev_net(skb_dst(skb
)->dev
);
692 arg
.tos
= ip_hdr(skb
)->tos
;
693 ip_send_unicast_reply(net
, skb
, ip_hdr(skb
)->saddr
,
694 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
696 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
697 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
699 #ifdef CONFIG_TCP_MD5SIG
708 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
709 outside socket context is ugly, certainly. What can I do?
712 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
713 u32 win
, u32 tsval
, u32 tsecr
, int oif
,
714 struct tcp_md5sig_key
*key
,
715 int reply_flags
, u8 tos
)
717 const struct tcphdr
*th
= tcp_hdr(skb
);
720 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
721 #ifdef CONFIG_TCP_MD5SIG
722 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
726 struct ip_reply_arg arg
;
727 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
729 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
730 memset(&arg
, 0, sizeof(arg
));
732 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
733 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
735 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
736 (TCPOPT_TIMESTAMP
<< 8) |
738 rep
.opt
[1] = htonl(tsval
);
739 rep
.opt
[2] = htonl(tsecr
);
740 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
743 /* Swap the send and the receive. */
744 rep
.th
.dest
= th
->source
;
745 rep
.th
.source
= th
->dest
;
746 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
747 rep
.th
.seq
= htonl(seq
);
748 rep
.th
.ack_seq
= htonl(ack
);
750 rep
.th
.window
= htons(win
);
752 #ifdef CONFIG_TCP_MD5SIG
754 int offset
= (tsecr
) ? 3 : 0;
756 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
758 (TCPOPT_MD5SIG
<< 8) |
760 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
761 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
763 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
764 key
, ip_hdr(skb
)->saddr
,
765 ip_hdr(skb
)->daddr
, &rep
.th
);
768 arg
.flags
= reply_flags
;
769 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
770 ip_hdr(skb
)->saddr
, /* XXX */
771 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
772 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
774 arg
.bound_dev_if
= oif
;
776 ip_send_unicast_reply(net
, skb
, ip_hdr(skb
)->saddr
,
777 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
779 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
782 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
784 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
785 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
787 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
788 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
789 tcp_time_stamp
+ tcptw
->tw_ts_offset
,
792 tcp_twsk_md5_key(tcptw
),
793 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
800 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
801 struct request_sock
*req
)
803 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
804 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
806 tcp_v4_send_ack(skb
, (sk
->sk_state
== TCP_LISTEN
) ?
807 tcp_rsk(req
)->snt_isn
+ 1 : tcp_sk(sk
)->snd_nxt
,
808 tcp_rsk(req
)->rcv_nxt
, req
->rcv_wnd
,
812 tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&ip_hdr(skb
)->daddr
,
814 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
819 * Send a SYN-ACK after having received a SYN.
820 * This still operates on a request_sock only, not on a big
823 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
824 struct request_sock
*req
,
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
, NULL
);
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
);
846 if (!tcp_rsk(req
)->snt_synack
&& !err
)
847 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
853 static int tcp_v4_rtx_synack(struct sock
*sk
, struct request_sock
*req
)
855 int res
= tcp_v4_send_synack(sk
, NULL
, req
, 0);
858 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
863 * IPv4 request_sock destructor.
865 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
867 kfree(inet_rsk(req
)->opt
);
871 * Return true if a syncookie should be sent
873 bool tcp_syn_flood_action(struct sock
*sk
,
874 const struct sk_buff
*skb
,
877 const char *msg
= "Dropping request";
878 bool want_cookie
= false;
879 struct listen_sock
*lopt
;
883 #ifdef CONFIG_SYN_COOKIES
884 if (sysctl_tcp_syncookies
) {
885 msg
= "Sending cookies";
887 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
890 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
892 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
893 if (!lopt
->synflood_warned
&& sysctl_tcp_syncookies
!= 2) {
894 lopt
->synflood_warned
= 1;
895 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
896 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
900 EXPORT_SYMBOL(tcp_syn_flood_action
);
903 * Save and compile IPv4 options into the request_sock if needed.
905 static struct ip_options_rcu
*tcp_v4_save_options(struct sk_buff
*skb
)
907 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
908 struct ip_options_rcu
*dopt
= NULL
;
910 if (opt
&& opt
->optlen
) {
911 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
913 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
915 if (ip_options_echo(&dopt
->opt
, skb
)) {
924 #ifdef CONFIG_TCP_MD5SIG
926 * RFC2385 MD5 checksumming requires a mapping of
927 * IP address->MD5 Key.
928 * We need to maintain these in the sk structure.
931 /* Find the Key structure for an address. */
932 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
933 const union tcp_md5_addr
*addr
,
936 struct tcp_sock
*tp
= tcp_sk(sk
);
937 struct tcp_md5sig_key
*key
;
938 unsigned int size
= sizeof(struct in_addr
);
939 struct tcp_md5sig_info
*md5sig
;
941 /* caller either holds rcu_read_lock() or socket lock */
942 md5sig
= rcu_dereference_check(tp
->md5sig_info
,
943 sock_owned_by_user(sk
) ||
944 lockdep_is_held(&sk
->sk_lock
.slock
));
947 #if IS_ENABLED(CONFIG_IPV6)
948 if (family
== AF_INET6
)
949 size
= sizeof(struct in6_addr
);
951 hlist_for_each_entry_rcu(key
, &md5sig
->head
, node
) {
952 if (key
->family
!= family
)
954 if (!memcmp(&key
->addr
, addr
, size
))
959 EXPORT_SYMBOL(tcp_md5_do_lookup
);
961 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
962 struct sock
*addr_sk
)
964 union tcp_md5_addr
*addr
;
966 addr
= (union tcp_md5_addr
*)&inet_sk(addr_sk
)->inet_daddr
;
967 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
969 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
971 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
972 struct request_sock
*req
)
974 union tcp_md5_addr
*addr
;
976 addr
= (union tcp_md5_addr
*)&inet_rsk(req
)->ir_rmt_addr
;
977 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
980 /* This can be called on a newly created socket, from other files */
981 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
982 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
)
984 /* Add Key to the list */
985 struct tcp_md5sig_key
*key
;
986 struct tcp_sock
*tp
= tcp_sk(sk
);
987 struct tcp_md5sig_info
*md5sig
;
989 key
= tcp_md5_do_lookup(sk
, addr
, family
);
991 /* Pre-existing entry - just update that one. */
992 memcpy(key
->key
, newkey
, newkeylen
);
993 key
->keylen
= newkeylen
;
997 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
998 sock_owned_by_user(sk
));
1000 md5sig
= kmalloc(sizeof(*md5sig
), gfp
);
1004 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
1005 INIT_HLIST_HEAD(&md5sig
->head
);
1006 rcu_assign_pointer(tp
->md5sig_info
, md5sig
);
1009 key
= sock_kmalloc(sk
, sizeof(*key
), gfp
);
1012 if (!tcp_alloc_md5sig_pool()) {
1013 sock_kfree_s(sk
, key
, sizeof(*key
));
1017 memcpy(key
->key
, newkey
, newkeylen
);
1018 key
->keylen
= newkeylen
;
1019 key
->family
= family
;
1020 memcpy(&key
->addr
, addr
,
1021 (family
== AF_INET6
) ? sizeof(struct in6_addr
) :
1022 sizeof(struct in_addr
));
1023 hlist_add_head_rcu(&key
->node
, &md5sig
->head
);
1026 EXPORT_SYMBOL(tcp_md5_do_add
);
1028 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
, int family
)
1030 struct tcp_md5sig_key
*key
;
1032 key
= tcp_md5_do_lookup(sk
, addr
, family
);
1035 hlist_del_rcu(&key
->node
);
1036 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1037 kfree_rcu(key
, rcu
);
1040 EXPORT_SYMBOL(tcp_md5_do_del
);
1042 static void tcp_clear_md5_list(struct sock
*sk
)
1044 struct tcp_sock
*tp
= tcp_sk(sk
);
1045 struct tcp_md5sig_key
*key
;
1046 struct hlist_node
*n
;
1047 struct tcp_md5sig_info
*md5sig
;
1049 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
1051 hlist_for_each_entry_safe(key
, n
, &md5sig
->head
, node
) {
1052 hlist_del_rcu(&key
->node
);
1053 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1054 kfree_rcu(key
, rcu
);
1058 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1061 struct tcp_md5sig cmd
;
1062 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1064 if (optlen
< sizeof(cmd
))
1067 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1070 if (sin
->sin_family
!= AF_INET
)
1073 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
)
1074 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1077 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1080 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1081 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1085 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1086 __be32 daddr
, __be32 saddr
, int nbytes
)
1088 struct tcp4_pseudohdr
*bp
;
1089 struct scatterlist sg
;
1091 bp
= &hp
->md5_blk
.ip4
;
1094 * 1. the TCP pseudo-header (in the order: source IP address,
1095 * destination IP address, zero-padded protocol number, and
1101 bp
->protocol
= IPPROTO_TCP
;
1102 bp
->len
= cpu_to_be16(nbytes
);
1104 sg_init_one(&sg
, bp
, sizeof(*bp
));
1105 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1108 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1109 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1111 struct tcp_md5sig_pool
*hp
;
1112 struct hash_desc
*desc
;
1114 hp
= tcp_get_md5sig_pool();
1116 goto clear_hash_noput
;
1117 desc
= &hp
->md5_desc
;
1119 if (crypto_hash_init(desc
))
1121 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1123 if (tcp_md5_hash_header(hp
, th
))
1125 if (tcp_md5_hash_key(hp
, key
))
1127 if (crypto_hash_final(desc
, md5_hash
))
1130 tcp_put_md5sig_pool();
1134 tcp_put_md5sig_pool();
1136 memset(md5_hash
, 0, 16);
1140 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1141 const struct sock
*sk
, const struct request_sock
*req
,
1142 const struct sk_buff
*skb
)
1144 struct tcp_md5sig_pool
*hp
;
1145 struct hash_desc
*desc
;
1146 const struct tcphdr
*th
= tcp_hdr(skb
);
1147 __be32 saddr
, daddr
;
1150 saddr
= inet_sk(sk
)->inet_saddr
;
1151 daddr
= inet_sk(sk
)->inet_daddr
;
1153 saddr
= inet_rsk(req
)->ir_loc_addr
;
1154 daddr
= inet_rsk(req
)->ir_rmt_addr
;
1156 const struct iphdr
*iph
= ip_hdr(skb
);
1161 hp
= tcp_get_md5sig_pool();
1163 goto clear_hash_noput
;
1164 desc
= &hp
->md5_desc
;
1166 if (crypto_hash_init(desc
))
1169 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1171 if (tcp_md5_hash_header(hp
, th
))
1173 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1175 if (tcp_md5_hash_key(hp
, key
))
1177 if (crypto_hash_final(desc
, md5_hash
))
1180 tcp_put_md5sig_pool();
1184 tcp_put_md5sig_pool();
1186 memset(md5_hash
, 0, 16);
1189 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1191 static bool tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1194 * This gets called for each TCP segment that arrives
1195 * so we want to be efficient.
1196 * We have 3 drop cases:
1197 * o No MD5 hash and one expected.
1198 * o MD5 hash and we're not expecting one.
1199 * o MD5 hash and its wrong.
1201 const __u8
*hash_location
= NULL
;
1202 struct tcp_md5sig_key
*hash_expected
;
1203 const struct iphdr
*iph
= ip_hdr(skb
);
1204 const struct tcphdr
*th
= tcp_hdr(skb
);
1206 unsigned char newhash
[16];
1208 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1210 hash_location
= tcp_parse_md5sig_option(th
);
1212 /* We've parsed the options - do we have a hash? */
1213 if (!hash_expected
&& !hash_location
)
1216 if (hash_expected
&& !hash_location
) {
1217 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1221 if (!hash_expected
&& hash_location
) {
1222 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1226 /* Okay, so this is hash_expected and hash_location -
1227 * so we need to calculate the checksum.
1229 genhash
= tcp_v4_md5_hash_skb(newhash
,
1233 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1234 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1235 &iph
->saddr
, ntohs(th
->source
),
1236 &iph
->daddr
, ntohs(th
->dest
),
1237 genhash
? " tcp_v4_calc_md5_hash failed"
1246 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1248 .obj_size
= sizeof(struct tcp_request_sock
),
1249 .rtx_syn_ack
= tcp_v4_rtx_synack
,
1250 .send_ack
= tcp_v4_reqsk_send_ack
,
1251 .destructor
= tcp_v4_reqsk_destructor
,
1252 .send_reset
= tcp_v4_send_reset
,
1253 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1256 #ifdef CONFIG_TCP_MD5SIG
1257 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1258 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1259 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1263 static bool tcp_fastopen_check(struct sock
*sk
, struct sk_buff
*skb
,
1264 struct request_sock
*req
,
1265 struct tcp_fastopen_cookie
*foc
,
1266 struct tcp_fastopen_cookie
*valid_foc
)
1268 bool skip_cookie
= false;
1269 struct fastopen_queue
*fastopenq
;
1271 if (likely(!fastopen_cookie_present(foc
))) {
1272 /* See include/net/tcp.h for the meaning of these knobs */
1273 if ((sysctl_tcp_fastopen
& TFO_SERVER_ALWAYS
) ||
1274 ((sysctl_tcp_fastopen
& TFO_SERVER_COOKIE_NOT_REQD
) &&
1275 (TCP_SKB_CB(skb
)->end_seq
!= TCP_SKB_CB(skb
)->seq
+ 1)))
1276 skip_cookie
= true; /* no cookie to validate */
1280 fastopenq
= inet_csk(sk
)->icsk_accept_queue
.fastopenq
;
1281 /* A FO option is present; bump the counter. */
1282 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENPASSIVE
);
1284 /* Make sure the listener has enabled fastopen, and we don't
1285 * exceed the max # of pending TFO requests allowed before trying
1286 * to validating the cookie in order to avoid burning CPU cycles
1289 * XXX (TFO) - The implication of checking the max_qlen before
1290 * processing a cookie request is that clients can't differentiate
1291 * between qlen overflow causing Fast Open to be disabled
1292 * temporarily vs a server not supporting Fast Open at all.
1294 if ((sysctl_tcp_fastopen
& TFO_SERVER_ENABLE
) == 0 ||
1295 fastopenq
== NULL
|| fastopenq
->max_qlen
== 0)
1298 if (fastopenq
->qlen
>= fastopenq
->max_qlen
) {
1299 struct request_sock
*req1
;
1300 spin_lock(&fastopenq
->lock
);
1301 req1
= fastopenq
->rskq_rst_head
;
1302 if ((req1
== NULL
) || time_after(req1
->expires
, jiffies
)) {
1303 spin_unlock(&fastopenq
->lock
);
1304 NET_INC_STATS_BH(sock_net(sk
),
1305 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW
);
1306 /* Avoid bumping LINUX_MIB_TCPFASTOPENPASSIVEFAIL*/
1310 fastopenq
->rskq_rst_head
= req1
->dl_next
;
1312 spin_unlock(&fastopenq
->lock
);
1316 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1320 if (foc
->len
== TCP_FASTOPEN_COOKIE_SIZE
) {
1321 if ((sysctl_tcp_fastopen
& TFO_SERVER_COOKIE_NOT_CHKED
) == 0) {
1322 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
,
1323 ip_hdr(skb
)->daddr
, valid_foc
);
1324 if ((valid_foc
->len
!= TCP_FASTOPEN_COOKIE_SIZE
) ||
1325 memcmp(&foc
->val
[0], &valid_foc
->val
[0],
1326 TCP_FASTOPEN_COOKIE_SIZE
) != 0)
1328 valid_foc
->len
= -1;
1330 /* Acknowledge the data received from the peer. */
1331 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1333 } else if (foc
->len
== 0) { /* Client requesting a cookie */
1334 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
,
1335 ip_hdr(skb
)->daddr
, valid_foc
);
1336 NET_INC_STATS_BH(sock_net(sk
),
1337 LINUX_MIB_TCPFASTOPENCOOKIEREQD
);
1339 /* Client sent a cookie with wrong size. Treat it
1340 * the same as invalid and return a valid one.
1342 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
,
1343 ip_hdr(skb
)->daddr
, valid_foc
);
1348 static int tcp_v4_conn_req_fastopen(struct sock
*sk
,
1349 struct sk_buff
*skb
,
1350 struct sk_buff
*skb_synack
,
1351 struct request_sock
*req
)
1353 struct tcp_sock
*tp
= tcp_sk(sk
);
1354 struct request_sock_queue
*queue
= &inet_csk(sk
)->icsk_accept_queue
;
1355 const struct inet_request_sock
*ireq
= inet_rsk(req
);
1359 req
->num_retrans
= 0;
1360 req
->num_timeout
= 0;
1363 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
);
1364 if (child
== NULL
) {
1365 NET_INC_STATS_BH(sock_net(sk
),
1366 LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
1367 kfree_skb(skb_synack
);
1370 err
= ip_build_and_send_pkt(skb_synack
, sk
, ireq
->ir_loc_addr
,
1371 ireq
->ir_rmt_addr
, ireq
->opt
);
1372 err
= net_xmit_eval(err
);
1374 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1375 /* XXX (TFO) - is it ok to ignore error and continue? */
1377 spin_lock(&queue
->fastopenq
->lock
);
1378 queue
->fastopenq
->qlen
++;
1379 spin_unlock(&queue
->fastopenq
->lock
);
1381 /* Initialize the child socket. Have to fix some values to take
1382 * into account the child is a Fast Open socket and is created
1383 * only out of the bits carried in the SYN packet.
1387 tp
->fastopen_rsk
= req
;
1388 /* Do a hold on the listner sk so that if the listener is being
1389 * closed, the child that has been accepted can live on and still
1390 * access listen_lock.
1393 tcp_rsk(req
)->listener
= sk
;
1395 /* RFC1323: The window in SYN & SYN/ACK segments is never
1396 * scaled. So correct it appropriately.
1398 tp
->snd_wnd
= ntohs(tcp_hdr(skb
)->window
);
1400 /* Activate the retrans timer so that SYNACK can be retransmitted.
1401 * The request socket is not added to the SYN table of the parent
1402 * because it's been added to the accept queue directly.
1404 inet_csk_reset_xmit_timer(child
, ICSK_TIME_RETRANS
,
1405 TCP_TIMEOUT_INIT
, TCP_RTO_MAX
);
1407 /* Add the child socket directly into the accept queue */
1408 inet_csk_reqsk_queue_add(sk
, req
, child
);
1410 /* Now finish processing the fastopen child socket. */
1411 inet_csk(child
)->icsk_af_ops
->rebuild_header(child
);
1412 tcp_init_congestion_control(child
);
1413 tcp_mtup_init(child
);
1414 tcp_init_metrics(child
);
1415 tcp_init_buffer_space(child
);
1417 /* Queue the data carried in the SYN packet. We need to first
1418 * bump skb's refcnt because the caller will attempt to free it.
1420 * XXX (TFO) - we honor a zero-payload TFO request for now.
1421 * (Any reason not to?)
1423 if (TCP_SKB_CB(skb
)->end_seq
== TCP_SKB_CB(skb
)->seq
+ 1) {
1424 /* Don't queue the skb if there is no payload in SYN.
1425 * XXX (TFO) - How about SYN+FIN?
1427 tp
->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1431 __skb_pull(skb
, tcp_hdr(skb
)->doff
* 4);
1432 skb_set_owner_r(skb
, child
);
1433 __skb_queue_tail(&child
->sk_receive_queue
, skb
);
1434 tp
->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1435 tp
->syn_data_acked
= 1;
1437 sk
->sk_data_ready(sk
, 0);
1438 bh_unlock_sock(child
);
1440 WARN_ON(req
->sk
== NULL
);
1444 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1446 struct tcp_options_received tmp_opt
;
1447 struct request_sock
*req
;
1448 struct inet_request_sock
*ireq
;
1449 struct tcp_sock
*tp
= tcp_sk(sk
);
1450 struct dst_entry
*dst
= NULL
;
1451 __be32 saddr
= ip_hdr(skb
)->saddr
;
1452 __be32 daddr
= ip_hdr(skb
)->daddr
;
1453 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1454 bool want_cookie
= false;
1456 struct tcp_fastopen_cookie foc
= { .len
= -1 };
1457 struct tcp_fastopen_cookie valid_foc
= { .len
= -1 };
1458 struct sk_buff
*skb_synack
;
1461 /* Never answer to SYNs send to broadcast or multicast */
1462 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1465 /* TW buckets are converted to open requests without
1466 * limitations, they conserve resources and peer is
1467 * evidently real one.
1469 if ((sysctl_tcp_syncookies
== 2 ||
1470 inet_csk_reqsk_queue_is_full(sk
)) && !isn
) {
1471 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCP");
1476 /* Accept backlog is full. If we have already queued enough
1477 * of warm entries in syn queue, drop request. It is better than
1478 * clogging syn queue with openreqs with exponentially increasing
1481 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1) {
1482 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1486 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1490 #ifdef CONFIG_TCP_MD5SIG
1491 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1494 tcp_clear_options(&tmp_opt
);
1495 tmp_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
1496 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
1497 tcp_parse_options(skb
, &tmp_opt
, 0, want_cookie
? NULL
: &foc
);
1499 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1500 tcp_clear_options(&tmp_opt
);
1502 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1503 tcp_openreq_init(req
, &tmp_opt
, skb
);
1505 ireq
= inet_rsk(req
);
1506 ireq
->ir_loc_addr
= daddr
;
1507 ireq
->ir_rmt_addr
= saddr
;
1508 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1509 ireq
->opt
= tcp_v4_save_options(skb
);
1511 if (security_inet_conn_request(sk
, skb
, req
))
1514 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1515 TCP_ECN_create_request(req
, skb
, sock_net(sk
));
1518 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1519 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1521 /* VJ's idea. We save last timestamp seen
1522 * from the destination in peer table, when entering
1523 * state TIME-WAIT, and check against it before
1524 * accepting new connection request.
1526 * If "isn" is not zero, this request hit alive
1527 * timewait bucket, so that all the necessary checks
1528 * are made in the function processing timewait state.
1530 if (tmp_opt
.saw_tstamp
&&
1531 tcp_death_row
.sysctl_tw_recycle
&&
1532 (dst
= inet_csk_route_req(sk
, &fl4
, req
)) != NULL
&&
1533 fl4
.daddr
== saddr
) {
1534 if (!tcp_peer_is_proven(req
, dst
, true)) {
1535 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1536 goto drop_and_release
;
1539 /* Kill the following clause, if you dislike this way. */
1540 else if (!sysctl_tcp_syncookies
&&
1541 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1542 (sysctl_max_syn_backlog
>> 2)) &&
1543 !tcp_peer_is_proven(req
, dst
, false)) {
1544 /* Without syncookies last quarter of
1545 * backlog is filled with destinations,
1546 * proven to be alive.
1547 * It means that we continue to communicate
1548 * to destinations, already remembered
1549 * to the moment of synflood.
1551 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("drop open request from %pI4/%u\n"),
1552 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1553 goto drop_and_release
;
1556 isn
= tcp_v4_init_sequence(skb
);
1558 tcp_rsk(req
)->snt_isn
= isn
;
1561 dst
= inet_csk_route_req(sk
, &fl4
, req
);
1565 do_fastopen
= tcp_fastopen_check(sk
, skb
, req
, &foc
, &valid_foc
);
1567 /* We don't call tcp_v4_send_synack() directly because we need
1568 * to make sure a child socket can be created successfully before
1569 * sending back synack!
1571 * XXX (TFO) - Ideally one would simply call tcp_v4_send_synack()
1572 * (or better yet, call tcp_send_synack() in the child context
1573 * directly, but will have to fix bunch of other code first)
1574 * after syn_recv_sock() except one will need to first fix the
1575 * latter to remove its dependency on the current implementation
1576 * of tcp_v4_send_synack()->tcp_select_initial_window().
1578 skb_synack
= tcp_make_synack(sk
, dst
, req
,
1579 fastopen_cookie_present(&valid_foc
) ? &valid_foc
: NULL
);
1582 __tcp_v4_send_check(skb_synack
, ireq
->ir_loc_addr
, ireq
->ir_rmt_addr
);
1583 skb_set_queue_mapping(skb_synack
, skb_get_queue_mapping(skb
));
1587 if (likely(!do_fastopen
)) {
1589 err
= ip_build_and_send_pkt(skb_synack
, sk
, ireq
->ir_loc_addr
,
1590 ireq
->ir_rmt_addr
, ireq
->opt
);
1591 err
= net_xmit_eval(err
);
1592 if (err
|| want_cookie
)
1595 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1596 tcp_rsk(req
)->listener
= NULL
;
1597 /* Add the request_sock to the SYN table */
1598 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1599 if (fastopen_cookie_present(&foc
) && foc
.len
!= 0)
1600 NET_INC_STATS_BH(sock_net(sk
),
1601 LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
1602 } else if (tcp_v4_conn_req_fastopen(sk
, skb
, skb_synack
, req
))
1612 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1615 EXPORT_SYMBOL(tcp_v4_conn_request
);
1619 * The three way handshake has completed - we got a valid synack -
1620 * now create the new socket.
1622 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1623 struct request_sock
*req
,
1624 struct dst_entry
*dst
)
1626 struct inet_request_sock
*ireq
;
1627 struct inet_sock
*newinet
;
1628 struct tcp_sock
*newtp
;
1630 #ifdef CONFIG_TCP_MD5SIG
1631 struct tcp_md5sig_key
*key
;
1633 struct ip_options_rcu
*inet_opt
;
1635 if (sk_acceptq_is_full(sk
))
1638 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1642 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1643 inet_sk_rx_dst_set(newsk
, skb
);
1645 newtp
= tcp_sk(newsk
);
1646 newinet
= inet_sk(newsk
);
1647 ireq
= inet_rsk(req
);
1648 newinet
->inet_daddr
= ireq
->ir_rmt_addr
;
1649 newinet
->inet_rcv_saddr
= ireq
->ir_loc_addr
;
1650 newinet
->inet_saddr
= ireq
->ir_loc_addr
;
1651 inet_opt
= ireq
->opt
;
1652 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1654 newinet
->mc_index
= inet_iif(skb
);
1655 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1656 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1657 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1659 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1660 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1663 dst
= inet_csk_route_child_sock(sk
, newsk
, req
);
1667 /* syncookie case : see end of cookie_v4_check() */
1669 sk_setup_caps(newsk
, dst
);
1671 tcp_sync_mss(newsk
, dst_mtu(dst
));
1672 newtp
->advmss
= dst_metric_advmss(dst
);
1673 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1674 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1675 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1677 tcp_initialize_rcv_mss(newsk
);
1679 #ifdef CONFIG_TCP_MD5SIG
1680 /* Copy over the MD5 key from the original socket */
1681 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1685 * We're using one, so create a matching key
1686 * on the newsk structure. If we fail to get
1687 * memory, then we end up not copying the key
1690 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1691 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1692 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1696 if (__inet_inherit_port(sk
, newsk
) < 0)
1698 __inet_hash_nolisten(newsk
, NULL
);
1703 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1707 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1710 inet_csk_prepare_forced_close(newsk
);
1714 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1716 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1718 struct tcphdr
*th
= tcp_hdr(skb
);
1719 const struct iphdr
*iph
= ip_hdr(skb
);
1721 struct request_sock
**prev
;
1722 /* Find possible connection requests. */
1723 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1724 iph
->saddr
, iph
->daddr
);
1726 return tcp_check_req(sk
, skb
, req
, prev
, false);
1728 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1729 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1732 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1736 inet_twsk_put(inet_twsk(nsk
));
1740 #ifdef CONFIG_SYN_COOKIES
1742 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1747 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1749 const struct iphdr
*iph
= ip_hdr(skb
);
1751 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1752 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1753 iph
->daddr
, skb
->csum
)) {
1754 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1759 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1760 skb
->len
, IPPROTO_TCP
, 0);
1762 if (skb
->len
<= 76) {
1763 return __skb_checksum_complete(skb
);
1769 /* The socket must have it's spinlock held when we get
1772 * We have a potential double-lock case here, so even when
1773 * doing backlog processing we use the BH locking scheme.
1774 * This is because we cannot sleep with the original spinlock
1777 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1780 #ifdef CONFIG_TCP_MD5SIG
1782 * We really want to reject the packet as early as possible
1784 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1785 * o There is an MD5 option and we're not expecting one
1787 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1791 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1792 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1794 sock_rps_save_rxhash(sk
, skb
);
1796 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1797 dst
->ops
->check(dst
, 0) == NULL
) {
1799 sk
->sk_rx_dst
= NULL
;
1802 tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
);
1806 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1809 if (sk
->sk_state
== TCP_LISTEN
) {
1810 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1815 sock_rps_save_rxhash(nsk
, skb
);
1816 if (tcp_child_process(sk
, nsk
, skb
)) {
1823 sock_rps_save_rxhash(sk
, skb
);
1825 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1832 tcp_v4_send_reset(rsk
, skb
);
1835 /* Be careful here. If this function gets more complicated and
1836 * gcc suffers from register pressure on the x86, sk (in %ebx)
1837 * might be destroyed here. This current version compiles correctly,
1838 * but you have been warned.
1843 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1844 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1847 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1849 void tcp_v4_early_demux(struct sk_buff
*skb
)
1851 const struct iphdr
*iph
;
1852 const struct tcphdr
*th
;
1855 if (skb
->pkt_type
!= PACKET_HOST
)
1858 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1864 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1867 sk
= __inet_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1868 iph
->saddr
, th
->source
,
1869 iph
->daddr
, ntohs(th
->dest
),
1873 skb
->destructor
= sock_edemux
;
1874 if (sk
->sk_state
!= TCP_TIME_WAIT
) {
1875 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1878 dst
= dst_check(dst
, 0);
1880 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1881 skb_dst_set_noref(skb
, dst
);
1886 /* Packet is added to VJ-style prequeue for processing in process
1887 * context, if a reader task is waiting. Apparently, this exciting
1888 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1889 * failed somewhere. Latency? Burstiness? Well, at least now we will
1890 * see, why it failed. 8)8) --ANK
1893 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1895 struct tcp_sock
*tp
= tcp_sk(sk
);
1897 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
1900 if (skb
->len
<= tcp_hdrlen(skb
) &&
1901 skb_queue_len(&tp
->ucopy
.prequeue
) == 0)
1905 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1906 tp
->ucopy
.memory
+= skb
->truesize
;
1907 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1908 struct sk_buff
*skb1
;
1910 BUG_ON(sock_owned_by_user(sk
));
1912 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1913 sk_backlog_rcv(sk
, skb1
);
1914 NET_INC_STATS_BH(sock_net(sk
),
1915 LINUX_MIB_TCPPREQUEUEDROPPED
);
1918 tp
->ucopy
.memory
= 0;
1919 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1920 wake_up_interruptible_sync_poll(sk_sleep(sk
),
1921 POLLIN
| POLLRDNORM
| POLLRDBAND
);
1922 if (!inet_csk_ack_scheduled(sk
))
1923 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
1924 (3 * tcp_rto_min(sk
)) / 4,
1929 EXPORT_SYMBOL(tcp_prequeue
);
1935 int tcp_v4_rcv(struct sk_buff
*skb
)
1937 const struct iphdr
*iph
;
1938 const struct tcphdr
*th
;
1941 struct net
*net
= dev_net(skb
->dev
);
1943 if (skb
->pkt_type
!= PACKET_HOST
)
1946 /* Count it even if it's bad */
1947 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1949 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1954 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1956 if (!pskb_may_pull(skb
, th
->doff
* 4))
1959 /* An explanation is required here, I think.
1960 * Packet length and doff are validated by header prediction,
1961 * provided case of th->doff==0 is eliminated.
1962 * So, we defer the checks. */
1963 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
1968 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1969 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1970 skb
->len
- th
->doff
* 4);
1971 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1972 TCP_SKB_CB(skb
)->when
= 0;
1973 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
1974 TCP_SKB_CB(skb
)->sacked
= 0;
1976 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1981 if (sk
->sk_state
== TCP_TIME_WAIT
)
1984 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1985 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1986 goto discard_and_relse
;
1989 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1990 goto discard_and_relse
;
1993 if (sk_filter(sk
, skb
))
1994 goto discard_and_relse
;
1996 sk_mark_napi_id(sk
, skb
);
1999 bh_lock_sock_nested(sk
);
2001 if (!sock_owned_by_user(sk
)) {
2002 #ifdef CONFIG_NET_DMA
2003 struct tcp_sock
*tp
= tcp_sk(sk
);
2004 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
2005 tp
->ucopy
.dma_chan
= net_dma_find_channel();
2006 if (tp
->ucopy
.dma_chan
)
2007 ret
= tcp_v4_do_rcv(sk
, skb
);
2011 if (!tcp_prequeue(sk
, skb
))
2012 ret
= tcp_v4_do_rcv(sk
, skb
);
2014 } else if (unlikely(sk_add_backlog(sk
, skb
,
2015 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
2017 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
2018 goto discard_and_relse
;
2027 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
2030 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
2032 TCP_INC_STATS_BH(net
, TCP_MIB_CSUMERRORS
);
2034 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
2036 tcp_v4_send_reset(NULL
, skb
);
2040 /* Discard frame. */
2049 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
2050 inet_twsk_put(inet_twsk(sk
));
2054 if (skb
->len
< (th
->doff
<< 2)) {
2055 inet_twsk_put(inet_twsk(sk
));
2058 if (tcp_checksum_complete(skb
)) {
2059 inet_twsk_put(inet_twsk(sk
));
2062 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
2064 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
2066 iph
->saddr
, th
->source
,
2067 iph
->daddr
, th
->dest
,
2070 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
2071 inet_twsk_put(inet_twsk(sk
));
2075 /* Fall through to ACK */
2078 tcp_v4_timewait_ack(sk
, skb
);
2082 case TCP_TW_SUCCESS
:;
2087 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
2088 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
2089 .twsk_unique
= tcp_twsk_unique
,
2090 .twsk_destructor
= tcp_twsk_destructor
,
2093 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
2095 struct dst_entry
*dst
= skb_dst(skb
);
2098 sk
->sk_rx_dst
= dst
;
2099 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
2101 EXPORT_SYMBOL(inet_sk_rx_dst_set
);
2103 const struct inet_connection_sock_af_ops ipv4_specific
= {
2104 .queue_xmit
= ip_queue_xmit
,
2105 .send_check
= tcp_v4_send_check
,
2106 .rebuild_header
= inet_sk_rebuild_header
,
2107 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
2108 .conn_request
= tcp_v4_conn_request
,
2109 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
2110 .net_header_len
= sizeof(struct iphdr
),
2111 .setsockopt
= ip_setsockopt
,
2112 .getsockopt
= ip_getsockopt
,
2113 .addr2sockaddr
= inet_csk_addr2sockaddr
,
2114 .sockaddr_len
= sizeof(struct sockaddr_in
),
2115 .bind_conflict
= inet_csk_bind_conflict
,
2116 #ifdef CONFIG_COMPAT
2117 .compat_setsockopt
= compat_ip_setsockopt
,
2118 .compat_getsockopt
= compat_ip_getsockopt
,
2121 EXPORT_SYMBOL(ipv4_specific
);
2123 #ifdef CONFIG_TCP_MD5SIG
2124 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
2125 .md5_lookup
= tcp_v4_md5_lookup
,
2126 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
2127 .md5_parse
= tcp_v4_parse_md5_keys
,
2131 /* NOTE: A lot of things set to zero explicitly by call to
2132 * sk_alloc() so need not be done here.
2134 static int tcp_v4_init_sock(struct sock
*sk
)
2136 struct inet_connection_sock
*icsk
= inet_csk(sk
);
2140 icsk
->icsk_af_ops
= &ipv4_specific
;
2142 #ifdef CONFIG_TCP_MD5SIG
2143 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv4_specific
;
2149 void tcp_v4_destroy_sock(struct sock
*sk
)
2151 struct tcp_sock
*tp
= tcp_sk(sk
);
2153 tcp_clear_xmit_timers(sk
);
2155 tcp_cleanup_congestion_control(sk
);
2157 /* Cleanup up the write buffer. */
2158 tcp_write_queue_purge(sk
);
2160 /* Cleans up our, hopefully empty, out_of_order_queue. */
2161 __skb_queue_purge(&tp
->out_of_order_queue
);
2163 #ifdef CONFIG_TCP_MD5SIG
2164 /* Clean up the MD5 key list, if any */
2165 if (tp
->md5sig_info
) {
2166 tcp_clear_md5_list(sk
);
2167 kfree_rcu(tp
->md5sig_info
, rcu
);
2168 tp
->md5sig_info
= NULL
;
2172 #ifdef CONFIG_NET_DMA
2173 /* Cleans up our sk_async_wait_queue */
2174 __skb_queue_purge(&sk
->sk_async_wait_queue
);
2177 /* Clean prequeue, it must be empty really */
2178 __skb_queue_purge(&tp
->ucopy
.prequeue
);
2180 /* Clean up a referenced TCP bind bucket. */
2181 if (inet_csk(sk
)->icsk_bind_hash
)
2184 BUG_ON(tp
->fastopen_rsk
!= NULL
);
2186 /* If socket is aborted during connect operation */
2187 tcp_free_fastopen_req(tp
);
2189 sk_sockets_allocated_dec(sk
);
2190 sock_release_memcg(sk
);
2192 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
2194 #ifdef CONFIG_PROC_FS
2195 /* Proc filesystem TCP sock list dumping. */
2198 * Get next listener socket follow cur. If cur is NULL, get first socket
2199 * starting from bucket given in st->bucket; when st->bucket is zero the
2200 * very first socket in the hash table is returned.
2202 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
2204 struct inet_connection_sock
*icsk
;
2205 struct hlist_nulls_node
*node
;
2206 struct sock
*sk
= cur
;
2207 struct inet_listen_hashbucket
*ilb
;
2208 struct tcp_iter_state
*st
= seq
->private;
2209 struct net
*net
= seq_file_net(seq
);
2212 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2213 spin_lock_bh(&ilb
->lock
);
2214 sk
= sk_nulls_head(&ilb
->head
);
2218 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2222 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
2223 struct request_sock
*req
= cur
;
2225 icsk
= inet_csk(st
->syn_wait_sk
);
2229 if (req
->rsk_ops
->family
== st
->family
) {
2235 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
2238 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
2240 sk
= sk_nulls_next(st
->syn_wait_sk
);
2241 st
->state
= TCP_SEQ_STATE_LISTENING
;
2242 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2244 icsk
= inet_csk(sk
);
2245 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2246 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
2248 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2249 sk
= sk_nulls_next(sk
);
2252 sk_nulls_for_each_from(sk
, node
) {
2253 if (!net_eq(sock_net(sk
), net
))
2255 if (sk
->sk_family
== st
->family
) {
2259 icsk
= inet_csk(sk
);
2260 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2261 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2263 st
->uid
= sock_i_uid(sk
);
2264 st
->syn_wait_sk
= sk
;
2265 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2269 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2271 spin_unlock_bh(&ilb
->lock
);
2273 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2274 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2275 spin_lock_bh(&ilb
->lock
);
2276 sk
= sk_nulls_head(&ilb
->head
);
2284 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2286 struct tcp_iter_state
*st
= seq
->private;
2291 rc
= listening_get_next(seq
, NULL
);
2293 while (rc
&& *pos
) {
2294 rc
= listening_get_next(seq
, rc
);
2300 static inline bool empty_bucket(const struct tcp_iter_state
*st
)
2302 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2306 * Get first established socket starting from bucket given in st->bucket.
2307 * If st->bucket is zero, the very first socket in the hash is returned.
2309 static void *established_get_first(struct seq_file
*seq
)
2311 struct tcp_iter_state
*st
= seq
->private;
2312 struct net
*net
= seq_file_net(seq
);
2316 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2318 struct hlist_nulls_node
*node
;
2319 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2321 /* Lockless fast path for the common case of empty buckets */
2322 if (empty_bucket(st
))
2326 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2327 if (sk
->sk_family
!= st
->family
||
2328 !net_eq(sock_net(sk
), net
)) {
2334 spin_unlock_bh(lock
);
2340 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2342 struct sock
*sk
= cur
;
2343 struct hlist_nulls_node
*node
;
2344 struct tcp_iter_state
*st
= seq
->private;
2345 struct net
*net
= seq_file_net(seq
);
2350 sk
= sk_nulls_next(sk
);
2352 sk_nulls_for_each_from(sk
, node
) {
2353 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2357 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2359 return established_get_first(seq
);
2362 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2364 struct tcp_iter_state
*st
= seq
->private;
2368 rc
= established_get_first(seq
);
2371 rc
= established_get_next(seq
, rc
);
2377 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2380 struct tcp_iter_state
*st
= seq
->private;
2382 st
->state
= TCP_SEQ_STATE_LISTENING
;
2383 rc
= listening_get_idx(seq
, &pos
);
2386 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2387 rc
= established_get_idx(seq
, pos
);
2393 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2395 struct tcp_iter_state
*st
= seq
->private;
2396 int offset
= st
->offset
;
2397 int orig_num
= st
->num
;
2400 switch (st
->state
) {
2401 case TCP_SEQ_STATE_OPENREQ
:
2402 case TCP_SEQ_STATE_LISTENING
:
2403 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2405 st
->state
= TCP_SEQ_STATE_LISTENING
;
2406 rc
= listening_get_next(seq
, NULL
);
2407 while (offset
-- && rc
)
2408 rc
= listening_get_next(seq
, rc
);
2412 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2414 case TCP_SEQ_STATE_ESTABLISHED
:
2415 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2417 rc
= established_get_first(seq
);
2418 while (offset
-- && rc
)
2419 rc
= established_get_next(seq
, rc
);
2427 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2429 struct tcp_iter_state
*st
= seq
->private;
2432 if (*pos
&& *pos
== st
->last_pos
) {
2433 rc
= tcp_seek_last_pos(seq
);
2438 st
->state
= TCP_SEQ_STATE_LISTENING
;
2442 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2445 st
->last_pos
= *pos
;
2449 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2451 struct tcp_iter_state
*st
= seq
->private;
2454 if (v
== SEQ_START_TOKEN
) {
2455 rc
= tcp_get_idx(seq
, 0);
2459 switch (st
->state
) {
2460 case TCP_SEQ_STATE_OPENREQ
:
2461 case TCP_SEQ_STATE_LISTENING
:
2462 rc
= listening_get_next(seq
, v
);
2464 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2467 rc
= established_get_first(seq
);
2470 case TCP_SEQ_STATE_ESTABLISHED
:
2471 rc
= established_get_next(seq
, v
);
2476 st
->last_pos
= *pos
;
2480 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2482 struct tcp_iter_state
*st
= seq
->private;
2484 switch (st
->state
) {
2485 case TCP_SEQ_STATE_OPENREQ
:
2487 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2488 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2490 case TCP_SEQ_STATE_LISTENING
:
2491 if (v
!= SEQ_START_TOKEN
)
2492 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2494 case TCP_SEQ_STATE_ESTABLISHED
:
2496 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2501 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2503 struct tcp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2504 struct tcp_iter_state
*s
;
2507 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2508 sizeof(struct tcp_iter_state
));
2512 s
= ((struct seq_file
*)file
->private_data
)->private;
2513 s
->family
= afinfo
->family
;
2517 EXPORT_SYMBOL(tcp_seq_open
);
2519 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2522 struct proc_dir_entry
*p
;
2524 afinfo
->seq_ops
.start
= tcp_seq_start
;
2525 afinfo
->seq_ops
.next
= tcp_seq_next
;
2526 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2528 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2529 afinfo
->seq_fops
, afinfo
);
2534 EXPORT_SYMBOL(tcp_proc_register
);
2536 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2538 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2540 EXPORT_SYMBOL(tcp_proc_unregister
);
2542 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2543 struct seq_file
*f
, int i
, kuid_t uid
)
2545 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2546 long delta
= req
->expires
- jiffies
;
2548 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2549 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2552 ntohs(inet_sk(sk
)->inet_sport
),
2554 ntohs(ireq
->ir_rmt_port
),
2556 0, 0, /* could print option size, but that is af dependent. */
2557 1, /* timers active (only the expire timer) */
2558 jiffies_delta_to_clock_t(delta
),
2560 from_kuid_munged(seq_user_ns(f
), uid
),
2561 0, /* non standard timer */
2562 0, /* open_requests have no inode */
2563 atomic_read(&sk
->sk_refcnt
),
2567 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
)
2570 unsigned long timer_expires
;
2571 const struct tcp_sock
*tp
= tcp_sk(sk
);
2572 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2573 const struct inet_sock
*inet
= inet_sk(sk
);
2574 struct fastopen_queue
*fastopenq
= icsk
->icsk_accept_queue
.fastopenq
;
2575 __be32 dest
= inet
->inet_daddr
;
2576 __be32 src
= inet
->inet_rcv_saddr
;
2577 __u16 destp
= ntohs(inet
->inet_dport
);
2578 __u16 srcp
= ntohs(inet
->inet_sport
);
2581 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
2582 icsk
->icsk_pending
== ICSK_TIME_EARLY_RETRANS
||
2583 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
2585 timer_expires
= icsk
->icsk_timeout
;
2586 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2588 timer_expires
= icsk
->icsk_timeout
;
2589 } else if (timer_pending(&sk
->sk_timer
)) {
2591 timer_expires
= sk
->sk_timer
.expires
;
2594 timer_expires
= jiffies
;
2597 if (sk
->sk_state
== TCP_LISTEN
)
2598 rx_queue
= sk
->sk_ack_backlog
;
2601 * because we dont lock socket, we might find a transient negative value
2603 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2605 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2606 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2607 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2608 tp
->write_seq
- tp
->snd_una
,
2611 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
2612 icsk
->icsk_retransmits
,
2613 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sk
)),
2614 icsk
->icsk_probes_out
,
2616 atomic_read(&sk
->sk_refcnt
), sk
,
2617 jiffies_to_clock_t(icsk
->icsk_rto
),
2618 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2619 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2621 sk
->sk_state
== TCP_LISTEN
?
2622 (fastopenq
? fastopenq
->max_qlen
: 0) :
2623 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
));
2626 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2627 struct seq_file
*f
, int i
)
2631 long delta
= tw
->tw_ttd
- jiffies
;
2633 dest
= tw
->tw_daddr
;
2634 src
= tw
->tw_rcv_saddr
;
2635 destp
= ntohs(tw
->tw_dport
);
2636 srcp
= ntohs(tw
->tw_sport
);
2638 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2639 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2640 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2641 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
2642 atomic_read(&tw
->tw_refcnt
), tw
);
2647 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2649 struct tcp_iter_state
*st
;
2650 struct sock
*sk
= v
;
2652 seq_setwidth(seq
, TMPSZ
- 1);
2653 if (v
== SEQ_START_TOKEN
) {
2654 seq_puts(seq
, " sl local_address rem_address st tx_queue "
2655 "rx_queue tr tm->when retrnsmt uid timeout "
2661 switch (st
->state
) {
2662 case TCP_SEQ_STATE_LISTENING
:
2663 case TCP_SEQ_STATE_ESTABLISHED
:
2664 if (sk
->sk_state
== TCP_TIME_WAIT
)
2665 get_timewait4_sock(v
, seq
, st
->num
);
2667 get_tcp4_sock(v
, seq
, st
->num
);
2669 case TCP_SEQ_STATE_OPENREQ
:
2670 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
);
2678 static const struct file_operations tcp_afinfo_seq_fops
= {
2679 .owner
= THIS_MODULE
,
2680 .open
= tcp_seq_open
,
2682 .llseek
= seq_lseek
,
2683 .release
= seq_release_net
2686 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2689 .seq_fops
= &tcp_afinfo_seq_fops
,
2691 .show
= tcp4_seq_show
,
2695 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2697 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2700 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2702 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2705 static struct pernet_operations tcp4_net_ops
= {
2706 .init
= tcp4_proc_init_net
,
2707 .exit
= tcp4_proc_exit_net
,
2710 int __init
tcp4_proc_init(void)
2712 return register_pernet_subsys(&tcp4_net_ops
);
2715 void tcp4_proc_exit(void)
2717 unregister_pernet_subsys(&tcp4_net_ops
);
2719 #endif /* CONFIG_PROC_FS */
2721 struct proto tcp_prot
= {
2723 .owner
= THIS_MODULE
,
2725 .connect
= tcp_v4_connect
,
2726 .disconnect
= tcp_disconnect
,
2727 .accept
= inet_csk_accept
,
2729 .init
= tcp_v4_init_sock
,
2730 .destroy
= tcp_v4_destroy_sock
,
2731 .shutdown
= tcp_shutdown
,
2732 .setsockopt
= tcp_setsockopt
,
2733 .getsockopt
= tcp_getsockopt
,
2734 .recvmsg
= tcp_recvmsg
,
2735 .sendmsg
= tcp_sendmsg
,
2736 .sendpage
= tcp_sendpage
,
2737 .backlog_rcv
= tcp_v4_do_rcv
,
2738 .release_cb
= tcp_release_cb
,
2739 .mtu_reduced
= tcp_v4_mtu_reduced
,
2741 .unhash
= inet_unhash
,
2742 .get_port
= inet_csk_get_port
,
2743 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2744 .stream_memory_free
= tcp_stream_memory_free
,
2745 .sockets_allocated
= &tcp_sockets_allocated
,
2746 .orphan_count
= &tcp_orphan_count
,
2747 .memory_allocated
= &tcp_memory_allocated
,
2748 .memory_pressure
= &tcp_memory_pressure
,
2749 .sysctl_mem
= sysctl_tcp_mem
,
2750 .sysctl_wmem
= sysctl_tcp_wmem
,
2751 .sysctl_rmem
= sysctl_tcp_rmem
,
2752 .max_header
= MAX_TCP_HEADER
,
2753 .obj_size
= sizeof(struct tcp_sock
),
2754 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2755 .twsk_prot
= &tcp_timewait_sock_ops
,
2756 .rsk_prot
= &tcp_request_sock_ops
,
2757 .h
.hashinfo
= &tcp_hashinfo
,
2758 .no_autobind
= true,
2759 #ifdef CONFIG_COMPAT
2760 .compat_setsockopt
= compat_tcp_setsockopt
,
2761 .compat_getsockopt
= compat_tcp_getsockopt
,
2763 #ifdef CONFIG_MEMCG_KMEM
2764 .init_cgroup
= tcp_init_cgroup
,
2765 .destroy_cgroup
= tcp_destroy_cgroup
,
2766 .proto_cgroup
= tcp_proto_cgroup
,
2769 EXPORT_SYMBOL(tcp_prot
);
2771 static int __net_init
tcp_sk_init(struct net
*net
)
2773 net
->ipv4
.sysctl_tcp_ecn
= 2;
2777 static void __net_exit
tcp_sk_exit(struct net
*net
)
2781 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2783 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2786 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2787 .init
= tcp_sk_init
,
2788 .exit
= tcp_sk_exit
,
2789 .exit_batch
= tcp_sk_exit_batch
,
2792 void __init
tcp_v4_init(void)
2794 inet_hashinfo_init(&tcp_hashinfo
);
2795 if (register_pernet_subsys(&tcp_sk_ops
))
2796 panic("Failed to create the TCP control socket.\n");