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>
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
);
93 #ifdef CONFIG_TCP_MD5SIG
94 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
95 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
);
98 struct inet_hashinfo tcp_hashinfo
;
99 EXPORT_SYMBOL(tcp_hashinfo
);
101 static inline __u32
tcp_v4_init_sequence(const struct sk_buff
*skb
)
103 return secure_tcp_sequence_number(ip_hdr(skb
)->daddr
,
106 tcp_hdr(skb
)->source
);
109 int tcp_twsk_unique(struct sock
*sk
, struct sock
*sktw
, void *twp
)
111 const struct tcp_timewait_sock
*tcptw
= tcp_twsk(sktw
);
112 struct tcp_sock
*tp
= tcp_sk(sk
);
114 /* With PAWS, it is safe from the viewpoint
115 of data integrity. Even without PAWS it is safe provided sequence
116 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
118 Actually, the idea is close to VJ's one, only timestamp cache is
119 held not per host, but per port pair and TW bucket is used as state
122 If TW bucket has been already destroyed we fall back to VJ's scheme
123 and use initial timestamp retrieved from peer table.
125 if (tcptw
->tw_ts_recent_stamp
&&
126 (twp
== NULL
|| (sysctl_tcp_tw_reuse
&&
127 get_seconds() - tcptw
->tw_ts_recent_stamp
> 1))) {
128 tp
->write_seq
= tcptw
->tw_snd_nxt
+ 65535 + 2;
129 if (tp
->write_seq
== 0)
131 tp
->rx_opt
.ts_recent
= tcptw
->tw_ts_recent
;
132 tp
->rx_opt
.ts_recent_stamp
= tcptw
->tw_ts_recent_stamp
;
139 EXPORT_SYMBOL_GPL(tcp_twsk_unique
);
141 static int tcp_repair_connect(struct sock
*sk
)
143 tcp_connect_init(sk
);
144 tcp_finish_connect(sk
, NULL
);
149 /* This will initiate an outgoing connection. */
150 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
152 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
153 struct inet_sock
*inet
= inet_sk(sk
);
154 struct tcp_sock
*tp
= tcp_sk(sk
);
155 __be16 orig_sport
, orig_dport
;
156 __be32 daddr
, nexthop
;
160 struct ip_options_rcu
*inet_opt
;
162 if (addr_len
< sizeof(struct sockaddr_in
))
165 if (usin
->sin_family
!= AF_INET
)
166 return -EAFNOSUPPORT
;
168 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
169 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
170 sock_owned_by_user(sk
));
171 if (inet_opt
&& inet_opt
->opt
.srr
) {
174 nexthop
= inet_opt
->opt
.faddr
;
177 orig_sport
= inet
->inet_sport
;
178 orig_dport
= usin
->sin_port
;
179 fl4
= &inet
->cork
.fl
.u
.ip4
;
180 rt
= ip_route_connect(fl4
, nexthop
, inet
->inet_saddr
,
181 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
183 orig_sport
, orig_dport
, sk
, true);
186 if (err
== -ENETUNREACH
)
187 IP_INC_STATS_BH(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
191 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
196 if (!inet_opt
|| !inet_opt
->opt
.srr
)
199 if (!inet
->inet_saddr
)
200 inet
->inet_saddr
= fl4
->saddr
;
201 inet
->inet_rcv_saddr
= inet
->inet_saddr
;
203 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->inet_daddr
!= daddr
) {
204 /* Reset inherited state */
205 tp
->rx_opt
.ts_recent
= 0;
206 tp
->rx_opt
.ts_recent_stamp
= 0;
207 if (likely(!tp
->repair
))
211 if (tcp_death_row
.sysctl_tw_recycle
&&
212 !tp
->rx_opt
.ts_recent_stamp
&& fl4
->daddr
== daddr
) {
213 struct inet_peer
*peer
= rt_get_peer(rt
, fl4
->daddr
);
215 * VJ's idea. We save last timestamp seen from
216 * the destination in peer table, when entering state
217 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
218 * when trying new connection.
221 inet_peer_refcheck(peer
);
222 if ((u32
)get_seconds() - peer
->tcp_ts_stamp
<= TCP_PAWS_MSL
) {
223 tp
->rx_opt
.ts_recent_stamp
= peer
->tcp_ts_stamp
;
224 tp
->rx_opt
.ts_recent
= peer
->tcp_ts
;
229 inet
->inet_dport
= usin
->sin_port
;
230 inet
->inet_daddr
= daddr
;
232 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
234 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
236 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
238 /* Socket identity is still unknown (sport may be zero).
239 * However we set state to SYN-SENT and not releasing socket
240 * lock select source port, enter ourselves into the hash tables and
241 * complete initialization after this.
243 tcp_set_state(sk
, TCP_SYN_SENT
);
244 err
= inet_hash_connect(&tcp_death_row
, sk
);
248 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
249 inet
->inet_sport
, inet
->inet_dport
, sk
);
255 /* OK, now commit destination to socket. */
256 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
257 sk_setup_caps(sk
, &rt
->dst
);
259 if (!tp
->write_seq
&& likely(!tp
->repair
))
260 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
265 inet
->inet_id
= tp
->write_seq
^ jiffies
;
267 if (likely(!tp
->repair
))
268 err
= tcp_connect(sk
);
270 err
= tcp_repair_connect(sk
);
280 * This unhashes the socket and releases the local port,
283 tcp_set_state(sk
, TCP_CLOSE
);
285 sk
->sk_route_caps
= 0;
286 inet
->inet_dport
= 0;
289 EXPORT_SYMBOL(tcp_v4_connect
);
292 * This routine does path mtu discovery as defined in RFC1191.
294 static void do_pmtu_discovery(struct sock
*sk
, const struct iphdr
*iph
, u32 mtu
)
296 struct dst_entry
*dst
;
297 struct inet_sock
*inet
= inet_sk(sk
);
299 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
300 * send out by Linux are always <576bytes so they should go through
303 if (sk
->sk_state
== TCP_LISTEN
)
306 /* We don't check in the destentry if pmtu discovery is forbidden
307 * on this route. We just assume that no packet_to_big packets
308 * are send back when pmtu discovery is not active.
309 * There is a small race when the user changes this flag in the
310 * route, but I think that's acceptable.
312 if ((dst
= __sk_dst_check(sk
, 0)) == NULL
)
315 dst
->ops
->update_pmtu(dst
, mtu
);
317 /* Something is about to be wrong... Remember soft error
318 * for the case, if this connection will not able to recover.
320 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
321 sk
->sk_err_soft
= EMSGSIZE
;
325 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
326 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
327 tcp_sync_mss(sk
, mtu
);
329 /* Resend the TCP packet because it's
330 * clear that the old packet has been
331 * dropped. This is the new "fast" path mtu
334 tcp_simple_retransmit(sk
);
335 } /* else let the usual retransmit timer handle it */
339 * This routine is called by the ICMP module when it gets some
340 * sort of error condition. If err < 0 then the socket should
341 * be closed and the error returned to the user. If err > 0
342 * it's just the icmp type << 8 | icmp code. After adjustment
343 * header points to the first 8 bytes of the tcp header. We need
344 * to find the appropriate port.
346 * The locking strategy used here is very "optimistic". When
347 * someone else accesses the socket the ICMP is just dropped
348 * and for some paths there is no check at all.
349 * A more general error queue to queue errors for later handling
350 * is probably better.
354 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
356 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
357 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
358 struct inet_connection_sock
*icsk
;
360 struct inet_sock
*inet
;
361 const int type
= icmp_hdr(icmp_skb
)->type
;
362 const int code
= icmp_hdr(icmp_skb
)->code
;
368 struct net
*net
= dev_net(icmp_skb
->dev
);
370 if (icmp_skb
->len
< (iph
->ihl
<< 2) + 8) {
371 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
375 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
376 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
378 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
381 if (sk
->sk_state
== TCP_TIME_WAIT
) {
382 inet_twsk_put(inet_twsk(sk
));
387 /* If too many ICMPs get dropped on busy
388 * servers this needs to be solved differently.
390 if (sock_owned_by_user(sk
))
391 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
393 if (sk
->sk_state
== TCP_CLOSE
)
396 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
397 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
403 seq
= ntohl(th
->seq
);
404 if (sk
->sk_state
!= TCP_LISTEN
&&
405 !between(seq
, tp
->snd_una
, tp
->snd_nxt
)) {
406 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
411 case ICMP_SOURCE_QUENCH
:
412 /* Just silently ignore these. */
414 case ICMP_PARAMETERPROB
:
417 case ICMP_DEST_UNREACH
:
418 if (code
> NR_ICMP_UNREACH
)
421 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
422 if (!sock_owned_by_user(sk
))
423 do_pmtu_discovery(sk
, iph
, info
);
427 err
= icmp_err_convert
[code
].errno
;
428 /* check if icmp_skb allows revert of backoff
429 * (see draft-zimmermann-tcp-lcd) */
430 if (code
!= ICMP_NET_UNREACH
&& code
!= ICMP_HOST_UNREACH
)
432 if (seq
!= tp
->snd_una
|| !icsk
->icsk_retransmits
||
436 if (sock_owned_by_user(sk
))
439 icsk
->icsk_backoff
--;
440 inet_csk(sk
)->icsk_rto
= (tp
->srtt
? __tcp_set_rto(tp
) :
441 TCP_TIMEOUT_INIT
) << icsk
->icsk_backoff
;
444 skb
= tcp_write_queue_head(sk
);
447 remaining
= icsk
->icsk_rto
- min(icsk
->icsk_rto
,
448 tcp_time_stamp
- TCP_SKB_CB(skb
)->when
);
451 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
452 remaining
, TCP_RTO_MAX
);
454 /* RTO revert clocked out retransmission.
455 * Will retransmit now */
456 tcp_retransmit_timer(sk
);
460 case ICMP_TIME_EXCEEDED
:
467 switch (sk
->sk_state
) {
468 struct request_sock
*req
, **prev
;
470 if (sock_owned_by_user(sk
))
473 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
474 iph
->daddr
, iph
->saddr
);
478 /* ICMPs are not backlogged, hence we cannot get
479 an established socket here.
483 if (seq
!= tcp_rsk(req
)->snt_isn
) {
484 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
489 * Still in SYN_RECV, just remove it silently.
490 * There is no good way to pass the error to the newly
491 * created socket, and POSIX does not want network
492 * errors returned from accept().
494 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
498 case TCP_SYN_RECV
: /* Cannot happen.
499 It can f.e. if SYNs crossed.
501 if (!sock_owned_by_user(sk
)) {
504 sk
->sk_error_report(sk
);
508 sk
->sk_err_soft
= err
;
513 /* If we've already connected we will keep trying
514 * until we time out, or the user gives up.
516 * rfc1122 4.2.3.9 allows to consider as hard errors
517 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
518 * but it is obsoleted by pmtu discovery).
520 * Note, that in modern internet, where routing is unreliable
521 * and in each dark corner broken firewalls sit, sending random
522 * errors ordered by their masters even this two messages finally lose
523 * their original sense (even Linux sends invalid PORT_UNREACHs)
525 * Now we are in compliance with RFCs.
530 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
532 sk
->sk_error_report(sk
);
533 } else { /* Only an error on timeout */
534 sk
->sk_err_soft
= err
;
542 static void __tcp_v4_send_check(struct sk_buff
*skb
,
543 __be32 saddr
, __be32 daddr
)
545 struct tcphdr
*th
= tcp_hdr(skb
);
547 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
548 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
549 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
550 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
552 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
559 /* This routine computes an IPv4 TCP checksum. */
560 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
562 const struct inet_sock
*inet
= inet_sk(sk
);
564 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
566 EXPORT_SYMBOL(tcp_v4_send_check
);
568 int tcp_v4_gso_send_check(struct sk_buff
*skb
)
570 const struct iphdr
*iph
;
573 if (!pskb_may_pull(skb
, sizeof(*th
)))
580 skb
->ip_summed
= CHECKSUM_PARTIAL
;
581 __tcp_v4_send_check(skb
, iph
->saddr
, iph
->daddr
);
586 * This routine will send an RST to the other tcp.
588 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
590 * Answer: if a packet caused RST, it is not for a socket
591 * existing in our system, if it is matched to a socket,
592 * it is just duplicate segment or bug in other side's TCP.
593 * So that we build reply only basing on parameters
594 * arrived with segment.
595 * Exception: precedence violation. We do not implement it in any case.
598 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
600 const struct tcphdr
*th
= tcp_hdr(skb
);
603 #ifdef CONFIG_TCP_MD5SIG
604 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
607 struct ip_reply_arg arg
;
608 #ifdef CONFIG_TCP_MD5SIG
609 struct tcp_md5sig_key
*key
;
610 const __u8
*hash_location
= NULL
;
611 unsigned char newhash
[16];
613 struct sock
*sk1
= NULL
;
617 /* Never send a reset in response to a reset. */
621 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
624 /* Swap the send and the receive. */
625 memset(&rep
, 0, sizeof(rep
));
626 rep
.th
.dest
= th
->source
;
627 rep
.th
.source
= th
->dest
;
628 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
632 rep
.th
.seq
= th
->ack_seq
;
635 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
636 skb
->len
- (th
->doff
<< 2));
639 memset(&arg
, 0, sizeof(arg
));
640 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
641 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
643 #ifdef CONFIG_TCP_MD5SIG
644 hash_location
= tcp_parse_md5sig_option(th
);
645 if (!sk
&& hash_location
) {
647 * active side is lost. Try to find listening socket through
648 * source port, and then find md5 key through listening socket.
649 * we are not loose security here:
650 * Incoming packet is checked with md5 hash with finding key,
651 * no RST generated if md5 hash doesn't match.
653 sk1
= __inet_lookup_listener(dev_net(skb_dst(skb
)->dev
),
654 &tcp_hashinfo
, ip_hdr(skb
)->daddr
,
655 ntohs(th
->source
), inet_iif(skb
));
656 /* don't send rst if it can't find key */
660 key
= tcp_md5_do_lookup(sk1
, (union tcp_md5_addr
*)
661 &ip_hdr(skb
)->saddr
, AF_INET
);
665 genhash
= tcp_v4_md5_hash_skb(newhash
, key
, NULL
, NULL
, skb
);
666 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
669 key
= sk
? tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)
675 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
677 (TCPOPT_MD5SIG
<< 8) |
679 /* Update length and the length the header thinks exists */
680 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
681 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
683 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
684 key
, ip_hdr(skb
)->saddr
,
685 ip_hdr(skb
)->daddr
, &rep
.th
);
688 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
689 ip_hdr(skb
)->saddr
, /* XXX */
690 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
691 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
692 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
693 /* When socket is gone, all binding information is lost.
694 * routing might fail in this case. using iif for oif to
695 * make sure we can deliver it
697 arg
.bound_dev_if
= sk
? sk
->sk_bound_dev_if
: inet_iif(skb
);
699 net
= dev_net(skb_dst(skb
)->dev
);
700 arg
.tos
= ip_hdr(skb
)->tos
;
701 ip_send_reply(net
->ipv4
.tcp_sock
, skb
, ip_hdr(skb
)->saddr
,
702 &arg
, arg
.iov
[0].iov_len
);
704 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
705 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
707 #ifdef CONFIG_TCP_MD5SIG
716 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
717 outside socket context is ugly, certainly. What can I do?
720 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
721 u32 win
, u32 ts
, int oif
,
722 struct tcp_md5sig_key
*key
,
723 int reply_flags
, u8 tos
)
725 const struct tcphdr
*th
= tcp_hdr(skb
);
728 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
729 #ifdef CONFIG_TCP_MD5SIG
730 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
734 struct ip_reply_arg arg
;
735 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
737 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
738 memset(&arg
, 0, sizeof(arg
));
740 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
741 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
743 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
744 (TCPOPT_TIMESTAMP
<< 8) |
746 rep
.opt
[1] = htonl(tcp_time_stamp
);
747 rep
.opt
[2] = htonl(ts
);
748 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
751 /* Swap the send and the receive. */
752 rep
.th
.dest
= th
->source
;
753 rep
.th
.source
= th
->dest
;
754 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
755 rep
.th
.seq
= htonl(seq
);
756 rep
.th
.ack_seq
= htonl(ack
);
758 rep
.th
.window
= htons(win
);
760 #ifdef CONFIG_TCP_MD5SIG
762 int offset
= (ts
) ? 3 : 0;
764 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
766 (TCPOPT_MD5SIG
<< 8) |
768 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
769 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
771 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
772 key
, ip_hdr(skb
)->saddr
,
773 ip_hdr(skb
)->daddr
, &rep
.th
);
776 arg
.flags
= reply_flags
;
777 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
778 ip_hdr(skb
)->saddr
, /* XXX */
779 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
780 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
782 arg
.bound_dev_if
= oif
;
784 ip_send_reply(net
->ipv4
.tcp_sock
, skb
, ip_hdr(skb
)->saddr
,
785 &arg
, arg
.iov
[0].iov_len
);
787 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
790 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
792 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
793 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
795 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
796 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
799 tcp_twsk_md5_key(tcptw
),
800 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
807 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
808 struct request_sock
*req
)
810 tcp_v4_send_ack(skb
, tcp_rsk(req
)->snt_isn
+ 1,
811 tcp_rsk(req
)->rcv_isn
+ 1, req
->rcv_wnd
,
814 tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&ip_hdr(skb
)->daddr
,
816 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
821 * Send a SYN-ACK after having received a SYN.
822 * This still operates on a request_sock only, not on a big
825 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
826 struct request_sock
*req
,
827 struct request_values
*rvp
)
829 const struct inet_request_sock
*ireq
= inet_rsk(req
);
832 struct sk_buff
* skb
;
834 /* First, grab a route. */
835 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
838 skb
= tcp_make_synack(sk
, dst
, req
, rvp
);
841 __tcp_v4_send_check(skb
, ireq
->loc_addr
, ireq
->rmt_addr
);
843 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
846 err
= net_xmit_eval(err
);
853 static int tcp_v4_rtx_synack(struct sock
*sk
, struct request_sock
*req
,
854 struct request_values
*rvp
)
856 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
857 return tcp_v4_send_synack(sk
, NULL
, req
, rvp
);
861 * IPv4 request_sock destructor.
863 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
865 kfree(inet_rsk(req
)->opt
);
869 * Return true if a syncookie should be sent
871 bool tcp_syn_flood_action(struct sock
*sk
,
872 const struct sk_buff
*skb
,
875 const char *msg
= "Dropping request";
876 bool want_cookie
= false;
877 struct listen_sock
*lopt
;
881 #ifdef CONFIG_SYN_COOKIES
882 if (sysctl_tcp_syncookies
) {
883 msg
= "Sending cookies";
885 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
888 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
890 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
891 if (!lopt
->synflood_warned
) {
892 lopt
->synflood_warned
= 1;
893 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
894 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
898 EXPORT_SYMBOL(tcp_syn_flood_action
);
901 * Save and compile IPv4 options into the request_sock if needed.
903 static struct ip_options_rcu
*tcp_v4_save_options(struct sock
*sk
,
906 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
907 struct ip_options_rcu
*dopt
= NULL
;
909 if (opt
&& opt
->optlen
) {
910 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
912 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
914 if (ip_options_echo(&dopt
->opt
, skb
)) {
923 #ifdef CONFIG_TCP_MD5SIG
925 * RFC2385 MD5 checksumming requires a mapping of
926 * IP address->MD5 Key.
927 * We need to maintain these in the sk structure.
930 /* Find the Key structure for an address. */
931 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
932 const union tcp_md5_addr
*addr
,
935 struct tcp_sock
*tp
= tcp_sk(sk
);
936 struct tcp_md5sig_key
*key
;
937 struct hlist_node
*pos
;
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
, pos
, &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
)->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
, (union tcp_md5_addr
*)&addr
, AF_INET
);
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 (hlist_empty(&md5sig
->head
) && !tcp_alloc_md5sig_pool(sk
)) {
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_sock
*tp
= tcp_sk(sk
);
1031 struct tcp_md5sig_key
*key
;
1032 struct tcp_md5sig_info
*md5sig
;
1034 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&addr
, AF_INET
);
1037 hlist_del_rcu(&key
->node
);
1038 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1039 kfree_rcu(key
, rcu
);
1040 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
1041 sock_owned_by_user(sk
));
1042 if (hlist_empty(&md5sig
->head
))
1043 tcp_free_md5sig_pool();
1046 EXPORT_SYMBOL(tcp_md5_do_del
);
1048 void tcp_clear_md5_list(struct sock
*sk
)
1050 struct tcp_sock
*tp
= tcp_sk(sk
);
1051 struct tcp_md5sig_key
*key
;
1052 struct hlist_node
*pos
, *n
;
1053 struct tcp_md5sig_info
*md5sig
;
1055 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
1057 if (!hlist_empty(&md5sig
->head
))
1058 tcp_free_md5sig_pool();
1059 hlist_for_each_entry_safe(key
, pos
, n
, &md5sig
->head
, node
) {
1060 hlist_del_rcu(&key
->node
);
1061 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1062 kfree_rcu(key
, rcu
);
1066 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1069 struct tcp_md5sig cmd
;
1070 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1072 if (optlen
< sizeof(cmd
))
1075 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1078 if (sin
->sin_family
!= AF_INET
)
1081 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
)
1082 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1085 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1088 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1089 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1093 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1094 __be32 daddr
, __be32 saddr
, int nbytes
)
1096 struct tcp4_pseudohdr
*bp
;
1097 struct scatterlist sg
;
1099 bp
= &hp
->md5_blk
.ip4
;
1102 * 1. the TCP pseudo-header (in the order: source IP address,
1103 * destination IP address, zero-padded protocol number, and
1109 bp
->protocol
= IPPROTO_TCP
;
1110 bp
->len
= cpu_to_be16(nbytes
);
1112 sg_init_one(&sg
, bp
, sizeof(*bp
));
1113 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1116 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1117 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1119 struct tcp_md5sig_pool
*hp
;
1120 struct hash_desc
*desc
;
1122 hp
= tcp_get_md5sig_pool();
1124 goto clear_hash_noput
;
1125 desc
= &hp
->md5_desc
;
1127 if (crypto_hash_init(desc
))
1129 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1131 if (tcp_md5_hash_header(hp
, th
))
1133 if (tcp_md5_hash_key(hp
, key
))
1135 if (crypto_hash_final(desc
, md5_hash
))
1138 tcp_put_md5sig_pool();
1142 tcp_put_md5sig_pool();
1144 memset(md5_hash
, 0, 16);
1148 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1149 const struct sock
*sk
, const struct request_sock
*req
,
1150 const struct sk_buff
*skb
)
1152 struct tcp_md5sig_pool
*hp
;
1153 struct hash_desc
*desc
;
1154 const struct tcphdr
*th
= tcp_hdr(skb
);
1155 __be32 saddr
, daddr
;
1158 saddr
= inet_sk(sk
)->inet_saddr
;
1159 daddr
= inet_sk(sk
)->inet_daddr
;
1161 saddr
= inet_rsk(req
)->loc_addr
;
1162 daddr
= inet_rsk(req
)->rmt_addr
;
1164 const struct iphdr
*iph
= ip_hdr(skb
);
1169 hp
= tcp_get_md5sig_pool();
1171 goto clear_hash_noput
;
1172 desc
= &hp
->md5_desc
;
1174 if (crypto_hash_init(desc
))
1177 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1179 if (tcp_md5_hash_header(hp
, th
))
1181 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1183 if (tcp_md5_hash_key(hp
, key
))
1185 if (crypto_hash_final(desc
, md5_hash
))
1188 tcp_put_md5sig_pool();
1192 tcp_put_md5sig_pool();
1194 memset(md5_hash
, 0, 16);
1197 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1199 static bool tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1202 * This gets called for each TCP segment that arrives
1203 * so we want to be efficient.
1204 * We have 3 drop cases:
1205 * o No MD5 hash and one expected.
1206 * o MD5 hash and we're not expecting one.
1207 * o MD5 hash and its wrong.
1209 const __u8
*hash_location
= NULL
;
1210 struct tcp_md5sig_key
*hash_expected
;
1211 const struct iphdr
*iph
= ip_hdr(skb
);
1212 const struct tcphdr
*th
= tcp_hdr(skb
);
1214 unsigned char newhash
[16];
1216 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1218 hash_location
= tcp_parse_md5sig_option(th
);
1220 /* We've parsed the options - do we have a hash? */
1221 if (!hash_expected
&& !hash_location
)
1224 if (hash_expected
&& !hash_location
) {
1225 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1229 if (!hash_expected
&& hash_location
) {
1230 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1234 /* Okay, so this is hash_expected and hash_location -
1235 * so we need to calculate the checksum.
1237 genhash
= tcp_v4_md5_hash_skb(newhash
,
1241 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1242 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1243 &iph
->saddr
, ntohs(th
->source
),
1244 &iph
->daddr
, ntohs(th
->dest
),
1245 genhash
? " tcp_v4_calc_md5_hash failed"
1254 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1256 .obj_size
= sizeof(struct tcp_request_sock
),
1257 .rtx_syn_ack
= tcp_v4_rtx_synack
,
1258 .send_ack
= tcp_v4_reqsk_send_ack
,
1259 .destructor
= tcp_v4_reqsk_destructor
,
1260 .send_reset
= tcp_v4_send_reset
,
1261 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1264 #ifdef CONFIG_TCP_MD5SIG
1265 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1266 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1267 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1271 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1273 struct tcp_extend_values tmp_ext
;
1274 struct tcp_options_received tmp_opt
;
1275 const u8
*hash_location
;
1276 struct request_sock
*req
;
1277 struct inet_request_sock
*ireq
;
1278 struct tcp_sock
*tp
= tcp_sk(sk
);
1279 struct dst_entry
*dst
= NULL
;
1280 __be32 saddr
= ip_hdr(skb
)->saddr
;
1281 __be32 daddr
= ip_hdr(skb
)->daddr
;
1282 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1283 bool want_cookie
= false;
1285 /* Never answer to SYNs send to broadcast or multicast */
1286 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1289 /* TW buckets are converted to open requests without
1290 * limitations, they conserve resources and peer is
1291 * evidently real one.
1293 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
1294 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCP");
1299 /* Accept backlog is full. If we have already queued enough
1300 * of warm entries in syn queue, drop request. It is better than
1301 * clogging syn queue with openreqs with exponentially increasing
1304 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1)
1307 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1311 #ifdef CONFIG_TCP_MD5SIG
1312 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1315 tcp_clear_options(&tmp_opt
);
1316 tmp_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
1317 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
1318 tcp_parse_options(skb
, &tmp_opt
, &hash_location
, 0);
1320 if (tmp_opt
.cookie_plus
> 0 &&
1321 tmp_opt
.saw_tstamp
&&
1322 !tp
->rx_opt
.cookie_out_never
&&
1323 (sysctl_tcp_cookie_size
> 0 ||
1324 (tp
->cookie_values
!= NULL
&&
1325 tp
->cookie_values
->cookie_desired
> 0))) {
1327 u32
*mess
= &tmp_ext
.cookie_bakery
[COOKIE_DIGEST_WORDS
];
1328 int l
= tmp_opt
.cookie_plus
- TCPOLEN_COOKIE_BASE
;
1330 if (tcp_cookie_generator(&tmp_ext
.cookie_bakery
[0]) != 0)
1331 goto drop_and_release
;
1333 /* Secret recipe starts with IP addresses */
1334 *mess
++ ^= (__force u32
)daddr
;
1335 *mess
++ ^= (__force u32
)saddr
;
1337 /* plus variable length Initiator Cookie */
1340 *c
++ ^= *hash_location
++;
1342 want_cookie
= false; /* not our kind of cookie */
1343 tmp_ext
.cookie_out_never
= 0; /* false */
1344 tmp_ext
.cookie_plus
= tmp_opt
.cookie_plus
;
1345 } else if (!tp
->rx_opt
.cookie_in_always
) {
1346 /* redundant indications, but ensure initialization. */
1347 tmp_ext
.cookie_out_never
= 1; /* true */
1348 tmp_ext
.cookie_plus
= 0;
1350 goto drop_and_release
;
1352 tmp_ext
.cookie_in_always
= tp
->rx_opt
.cookie_in_always
;
1354 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1355 tcp_clear_options(&tmp_opt
);
1357 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1358 tcp_openreq_init(req
, &tmp_opt
, skb
);
1360 ireq
= inet_rsk(req
);
1361 ireq
->loc_addr
= daddr
;
1362 ireq
->rmt_addr
= saddr
;
1363 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1364 ireq
->opt
= tcp_v4_save_options(sk
, skb
);
1366 if (security_inet_conn_request(sk
, skb
, req
))
1369 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1370 TCP_ECN_create_request(req
, skb
);
1373 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1374 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1376 struct inet_peer
*peer
= NULL
;
1379 /* VJ's idea. We save last timestamp seen
1380 * from the destination in peer table, when entering
1381 * state TIME-WAIT, and check against it before
1382 * accepting new connection request.
1384 * If "isn" is not zero, this request hit alive
1385 * timewait bucket, so that all the necessary checks
1386 * are made in the function processing timewait state.
1388 if (tmp_opt
.saw_tstamp
&&
1389 tcp_death_row
.sysctl_tw_recycle
&&
1390 (dst
= inet_csk_route_req(sk
, &fl4
, req
)) != NULL
&&
1391 fl4
.daddr
== saddr
&&
1392 (peer
= rt_get_peer((struct rtable
*)dst
, fl4
.daddr
)) != NULL
) {
1393 inet_peer_refcheck(peer
);
1394 if ((u32
)get_seconds() - peer
->tcp_ts_stamp
< TCP_PAWS_MSL
&&
1395 (s32
)(peer
->tcp_ts
- req
->ts_recent
) >
1397 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1398 goto drop_and_release
;
1401 /* Kill the following clause, if you dislike this way. */
1402 else if (!sysctl_tcp_syncookies
&&
1403 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1404 (sysctl_max_syn_backlog
>> 2)) &&
1405 (!peer
|| !peer
->tcp_ts_stamp
) &&
1406 (!dst
|| !dst_metric(dst
, RTAX_RTT
))) {
1407 /* Without syncookies last quarter of
1408 * backlog is filled with destinations,
1409 * proven to be alive.
1410 * It means that we continue to communicate
1411 * to destinations, already remembered
1412 * to the moment of synflood.
1414 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("drop open request from %pI4/%u\n"),
1415 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1416 goto drop_and_release
;
1419 isn
= tcp_v4_init_sequence(skb
);
1421 tcp_rsk(req
)->snt_isn
= isn
;
1422 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1424 if (tcp_v4_send_synack(sk
, dst
, req
,
1425 (struct request_values
*)&tmp_ext
) ||
1429 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1439 EXPORT_SYMBOL(tcp_v4_conn_request
);
1443 * The three way handshake has completed - we got a valid synack -
1444 * now create the new socket.
1446 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1447 struct request_sock
*req
,
1448 struct dst_entry
*dst
)
1450 struct inet_request_sock
*ireq
;
1451 struct inet_sock
*newinet
;
1452 struct tcp_sock
*newtp
;
1454 #ifdef CONFIG_TCP_MD5SIG
1455 struct tcp_md5sig_key
*key
;
1457 struct ip_options_rcu
*inet_opt
;
1459 if (sk_acceptq_is_full(sk
))
1462 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1466 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1468 newtp
= tcp_sk(newsk
);
1469 newinet
= inet_sk(newsk
);
1470 ireq
= inet_rsk(req
);
1471 newinet
->inet_daddr
= ireq
->rmt_addr
;
1472 newinet
->inet_rcv_saddr
= ireq
->loc_addr
;
1473 newinet
->inet_saddr
= ireq
->loc_addr
;
1474 inet_opt
= ireq
->opt
;
1475 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1477 newinet
->mc_index
= inet_iif(skb
);
1478 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1479 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1480 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1482 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1483 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1486 dst
= inet_csk_route_child_sock(sk
, newsk
, req
);
1490 /* syncookie case : see end of cookie_v4_check() */
1492 sk_setup_caps(newsk
, dst
);
1494 tcp_mtup_init(newsk
);
1495 tcp_sync_mss(newsk
, dst_mtu(dst
));
1496 newtp
->advmss
= dst_metric_advmss(dst
);
1497 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1498 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1499 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1501 tcp_initialize_rcv_mss(newsk
);
1502 if (tcp_rsk(req
)->snt_synack
)
1503 tcp_valid_rtt_meas(newsk
,
1504 tcp_time_stamp
- tcp_rsk(req
)->snt_synack
);
1505 newtp
->total_retrans
= req
->retrans
;
1507 #ifdef CONFIG_TCP_MD5SIG
1508 /* Copy over the MD5 key from the original socket */
1509 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1513 * We're using one, so create a matching key
1514 * on the newsk structure. If we fail to get
1515 * memory, then we end up not copying the key
1518 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1519 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1520 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1524 if (__inet_inherit_port(sk
, newsk
) < 0)
1526 __inet_hash_nolisten(newsk
, NULL
);
1531 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1535 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1538 tcp_clear_xmit_timers(newsk
);
1539 tcp_cleanup_congestion_control(newsk
);
1540 bh_unlock_sock(newsk
);
1544 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1546 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1548 struct tcphdr
*th
= tcp_hdr(skb
);
1549 const struct iphdr
*iph
= ip_hdr(skb
);
1551 struct request_sock
**prev
;
1552 /* Find possible connection requests. */
1553 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1554 iph
->saddr
, iph
->daddr
);
1556 return tcp_check_req(sk
, skb
, req
, prev
);
1558 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1559 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1562 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1566 inet_twsk_put(inet_twsk(nsk
));
1570 #ifdef CONFIG_SYN_COOKIES
1572 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1577 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1579 const struct iphdr
*iph
= ip_hdr(skb
);
1581 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1582 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1583 iph
->daddr
, skb
->csum
)) {
1584 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1589 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1590 skb
->len
, IPPROTO_TCP
, 0);
1592 if (skb
->len
<= 76) {
1593 return __skb_checksum_complete(skb
);
1599 /* The socket must have it's spinlock held when we get
1602 * We have a potential double-lock case here, so even when
1603 * doing backlog processing we use the BH locking scheme.
1604 * This is because we cannot sleep with the original spinlock
1607 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1610 #ifdef CONFIG_TCP_MD5SIG
1612 * We really want to reject the packet as early as possible
1614 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1615 * o There is an MD5 option and we're not expecting one
1617 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1621 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1622 sock_rps_save_rxhash(sk
, skb
);
1623 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1630 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1633 if (sk
->sk_state
== TCP_LISTEN
) {
1634 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1639 sock_rps_save_rxhash(nsk
, skb
);
1640 if (tcp_child_process(sk
, nsk
, skb
)) {
1647 sock_rps_save_rxhash(sk
, skb
);
1649 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1656 tcp_v4_send_reset(rsk
, skb
);
1659 /* Be careful here. If this function gets more complicated and
1660 * gcc suffers from register pressure on the x86, sk (in %ebx)
1661 * might be destroyed here. This current version compiles correctly,
1662 * but you have been warned.
1667 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1670 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1676 int tcp_v4_rcv(struct sk_buff
*skb
)
1678 const struct iphdr
*iph
;
1679 const struct tcphdr
*th
;
1682 struct net
*net
= dev_net(skb
->dev
);
1684 if (skb
->pkt_type
!= PACKET_HOST
)
1687 /* Count it even if it's bad */
1688 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1690 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1695 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1697 if (!pskb_may_pull(skb
, th
->doff
* 4))
1700 /* An explanation is required here, I think.
1701 * Packet length and doff are validated by header prediction,
1702 * provided case of th->doff==0 is eliminated.
1703 * So, we defer the checks. */
1704 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
1709 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1710 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1711 skb
->len
- th
->doff
* 4);
1712 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1713 TCP_SKB_CB(skb
)->when
= 0;
1714 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
1715 TCP_SKB_CB(skb
)->sacked
= 0;
1717 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1722 if (sk
->sk_state
== TCP_TIME_WAIT
)
1725 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1726 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1727 goto discard_and_relse
;
1730 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1731 goto discard_and_relse
;
1734 if (sk_filter(sk
, skb
))
1735 goto discard_and_relse
;
1739 bh_lock_sock_nested(sk
);
1741 if (!sock_owned_by_user(sk
)) {
1742 #ifdef CONFIG_NET_DMA
1743 struct tcp_sock
*tp
= tcp_sk(sk
);
1744 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1745 tp
->ucopy
.dma_chan
= net_dma_find_channel();
1746 if (tp
->ucopy
.dma_chan
)
1747 ret
= tcp_v4_do_rcv(sk
, skb
);
1751 if (!tcp_prequeue(sk
, skb
))
1752 ret
= tcp_v4_do_rcv(sk
, skb
);
1754 } else if (unlikely(sk_add_backlog(sk
, skb
,
1755 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
1757 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1758 goto discard_and_relse
;
1767 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1770 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1772 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1774 tcp_v4_send_reset(NULL
, skb
);
1778 /* Discard frame. */
1787 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1788 inet_twsk_put(inet_twsk(sk
));
1792 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1793 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1794 inet_twsk_put(inet_twsk(sk
));
1797 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1799 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1801 iph
->daddr
, th
->dest
,
1804 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1805 inet_twsk_put(inet_twsk(sk
));
1809 /* Fall through to ACK */
1812 tcp_v4_timewait_ack(sk
, skb
);
1816 case TCP_TW_SUCCESS
:;
1821 struct inet_peer
*tcp_v4_get_peer(struct sock
*sk
, bool *release_it
)
1823 struct rtable
*rt
= (struct rtable
*) __sk_dst_get(sk
);
1824 struct inet_sock
*inet
= inet_sk(sk
);
1825 struct inet_peer
*peer
;
1828 inet
->cork
.fl
.u
.ip4
.daddr
!= inet
->inet_daddr
) {
1829 peer
= inet_getpeer_v4(inet
->inet_daddr
, 1);
1833 rt_bind_peer(rt
, inet
->inet_daddr
, 1);
1835 *release_it
= false;
1840 EXPORT_SYMBOL(tcp_v4_get_peer
);
1842 void *tcp_v4_tw_get_peer(struct sock
*sk
)
1844 const struct inet_timewait_sock
*tw
= inet_twsk(sk
);
1846 return inet_getpeer_v4(tw
->tw_daddr
, 1);
1848 EXPORT_SYMBOL(tcp_v4_tw_get_peer
);
1850 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1851 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1852 .twsk_unique
= tcp_twsk_unique
,
1853 .twsk_destructor
= tcp_twsk_destructor
,
1854 .twsk_getpeer
= tcp_v4_tw_get_peer
,
1857 const struct inet_connection_sock_af_ops ipv4_specific
= {
1858 .queue_xmit
= ip_queue_xmit
,
1859 .send_check
= tcp_v4_send_check
,
1860 .rebuild_header
= inet_sk_rebuild_header
,
1861 .conn_request
= tcp_v4_conn_request
,
1862 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1863 .get_peer
= tcp_v4_get_peer
,
1864 .net_header_len
= sizeof(struct iphdr
),
1865 .setsockopt
= ip_setsockopt
,
1866 .getsockopt
= ip_getsockopt
,
1867 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1868 .sockaddr_len
= sizeof(struct sockaddr_in
),
1869 .bind_conflict
= inet_csk_bind_conflict
,
1870 #ifdef CONFIG_COMPAT
1871 .compat_setsockopt
= compat_ip_setsockopt
,
1872 .compat_getsockopt
= compat_ip_getsockopt
,
1875 EXPORT_SYMBOL(ipv4_specific
);
1877 #ifdef CONFIG_TCP_MD5SIG
1878 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1879 .md5_lookup
= tcp_v4_md5_lookup
,
1880 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1881 .md5_parse
= tcp_v4_parse_md5_keys
,
1885 /* NOTE: A lot of things set to zero explicitly by call to
1886 * sk_alloc() so need not be done here.
1888 static int tcp_v4_init_sock(struct sock
*sk
)
1890 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1894 icsk
->icsk_af_ops
= &ipv4_specific
;
1896 #ifdef CONFIG_TCP_MD5SIG
1897 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv4_specific
;
1903 void tcp_v4_destroy_sock(struct sock
*sk
)
1905 struct tcp_sock
*tp
= tcp_sk(sk
);
1907 tcp_clear_xmit_timers(sk
);
1909 tcp_cleanup_congestion_control(sk
);
1911 /* Cleanup up the write buffer. */
1912 tcp_write_queue_purge(sk
);
1914 /* Cleans up our, hopefully empty, out_of_order_queue. */
1915 __skb_queue_purge(&tp
->out_of_order_queue
);
1917 #ifdef CONFIG_TCP_MD5SIG
1918 /* Clean up the MD5 key list, if any */
1919 if (tp
->md5sig_info
) {
1920 tcp_clear_md5_list(sk
);
1921 kfree_rcu(tp
->md5sig_info
, rcu
);
1922 tp
->md5sig_info
= NULL
;
1926 #ifdef CONFIG_NET_DMA
1927 /* Cleans up our sk_async_wait_queue */
1928 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1931 /* Clean prequeue, it must be empty really */
1932 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1934 /* Clean up a referenced TCP bind bucket. */
1935 if (inet_csk(sk
)->icsk_bind_hash
)
1939 * If sendmsg cached page exists, toss it.
1941 if (sk
->sk_sndmsg_page
) {
1942 __free_page(sk
->sk_sndmsg_page
);
1943 sk
->sk_sndmsg_page
= NULL
;
1946 /* TCP Cookie Transactions */
1947 if (tp
->cookie_values
!= NULL
) {
1948 kref_put(&tp
->cookie_values
->kref
,
1949 tcp_cookie_values_release
);
1950 tp
->cookie_values
= NULL
;
1953 sk_sockets_allocated_dec(sk
);
1954 sock_release_memcg(sk
);
1956 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1958 #ifdef CONFIG_PROC_FS
1959 /* Proc filesystem TCP sock list dumping. */
1961 static inline struct inet_timewait_sock
*tw_head(struct hlist_nulls_head
*head
)
1963 return hlist_nulls_empty(head
) ? NULL
:
1964 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
1967 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
1969 return !is_a_nulls(tw
->tw_node
.next
) ?
1970 hlist_nulls_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
1974 * Get next listener socket follow cur. If cur is NULL, get first socket
1975 * starting from bucket given in st->bucket; when st->bucket is zero the
1976 * very first socket in the hash table is returned.
1978 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
1980 struct inet_connection_sock
*icsk
;
1981 struct hlist_nulls_node
*node
;
1982 struct sock
*sk
= cur
;
1983 struct inet_listen_hashbucket
*ilb
;
1984 struct tcp_iter_state
*st
= seq
->private;
1985 struct net
*net
= seq_file_net(seq
);
1988 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1989 spin_lock_bh(&ilb
->lock
);
1990 sk
= sk_nulls_head(&ilb
->head
);
1994 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
1998 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
1999 struct request_sock
*req
= cur
;
2001 icsk
= inet_csk(st
->syn_wait_sk
);
2005 if (req
->rsk_ops
->family
== st
->family
) {
2011 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
2014 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
2016 sk
= sk_nulls_next(st
->syn_wait_sk
);
2017 st
->state
= TCP_SEQ_STATE_LISTENING
;
2018 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2020 icsk
= inet_csk(sk
);
2021 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2022 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
2024 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2025 sk
= sk_nulls_next(sk
);
2028 sk_nulls_for_each_from(sk
, node
) {
2029 if (!net_eq(sock_net(sk
), net
))
2031 if (sk
->sk_family
== st
->family
) {
2035 icsk
= inet_csk(sk
);
2036 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2037 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2039 st
->uid
= sock_i_uid(sk
);
2040 st
->syn_wait_sk
= sk
;
2041 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2045 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2047 spin_unlock_bh(&ilb
->lock
);
2049 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2050 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2051 spin_lock_bh(&ilb
->lock
);
2052 sk
= sk_nulls_head(&ilb
->head
);
2060 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2062 struct tcp_iter_state
*st
= seq
->private;
2067 rc
= listening_get_next(seq
, NULL
);
2069 while (rc
&& *pos
) {
2070 rc
= listening_get_next(seq
, rc
);
2076 static inline bool empty_bucket(struct tcp_iter_state
*st
)
2078 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
) &&
2079 hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2083 * Get first established socket starting from bucket given in st->bucket.
2084 * If st->bucket is zero, the very first socket in the hash is returned.
2086 static void *established_get_first(struct seq_file
*seq
)
2088 struct tcp_iter_state
*st
= seq
->private;
2089 struct net
*net
= seq_file_net(seq
);
2093 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2095 struct hlist_nulls_node
*node
;
2096 struct inet_timewait_sock
*tw
;
2097 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2099 /* Lockless fast path for the common case of empty buckets */
2100 if (empty_bucket(st
))
2104 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2105 if (sk
->sk_family
!= st
->family
||
2106 !net_eq(sock_net(sk
), net
)) {
2112 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2113 inet_twsk_for_each(tw
, node
,
2114 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
2115 if (tw
->tw_family
!= st
->family
||
2116 !net_eq(twsk_net(tw
), net
)) {
2122 spin_unlock_bh(lock
);
2123 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2129 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2131 struct sock
*sk
= cur
;
2132 struct inet_timewait_sock
*tw
;
2133 struct hlist_nulls_node
*node
;
2134 struct tcp_iter_state
*st
= seq
->private;
2135 struct net
*net
= seq_file_net(seq
);
2140 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2144 while (tw
&& (tw
->tw_family
!= st
->family
|| !net_eq(twsk_net(tw
), net
))) {
2151 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2152 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2154 /* Look for next non empty bucket */
2156 while (++st
->bucket
<= tcp_hashinfo
.ehash_mask
&&
2159 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2162 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2163 sk
= sk_nulls_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2165 sk
= sk_nulls_next(sk
);
2167 sk_nulls_for_each_from(sk
, node
) {
2168 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2172 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2173 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2181 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2183 struct tcp_iter_state
*st
= seq
->private;
2187 rc
= established_get_first(seq
);
2190 rc
= established_get_next(seq
, rc
);
2196 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2199 struct tcp_iter_state
*st
= seq
->private;
2201 st
->state
= TCP_SEQ_STATE_LISTENING
;
2202 rc
= listening_get_idx(seq
, &pos
);
2205 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2206 rc
= established_get_idx(seq
, pos
);
2212 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2214 struct tcp_iter_state
*st
= seq
->private;
2215 int offset
= st
->offset
;
2216 int orig_num
= st
->num
;
2219 switch (st
->state
) {
2220 case TCP_SEQ_STATE_OPENREQ
:
2221 case TCP_SEQ_STATE_LISTENING
:
2222 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2224 st
->state
= TCP_SEQ_STATE_LISTENING
;
2225 rc
= listening_get_next(seq
, NULL
);
2226 while (offset
-- && rc
)
2227 rc
= listening_get_next(seq
, rc
);
2232 case TCP_SEQ_STATE_ESTABLISHED
:
2233 case TCP_SEQ_STATE_TIME_WAIT
:
2234 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2235 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2237 rc
= established_get_first(seq
);
2238 while (offset
-- && rc
)
2239 rc
= established_get_next(seq
, rc
);
2247 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2249 struct tcp_iter_state
*st
= seq
->private;
2252 if (*pos
&& *pos
== st
->last_pos
) {
2253 rc
= tcp_seek_last_pos(seq
);
2258 st
->state
= TCP_SEQ_STATE_LISTENING
;
2262 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2265 st
->last_pos
= *pos
;
2269 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2271 struct tcp_iter_state
*st
= seq
->private;
2274 if (v
== SEQ_START_TOKEN
) {
2275 rc
= tcp_get_idx(seq
, 0);
2279 switch (st
->state
) {
2280 case TCP_SEQ_STATE_OPENREQ
:
2281 case TCP_SEQ_STATE_LISTENING
:
2282 rc
= listening_get_next(seq
, v
);
2284 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2287 rc
= established_get_first(seq
);
2290 case TCP_SEQ_STATE_ESTABLISHED
:
2291 case TCP_SEQ_STATE_TIME_WAIT
:
2292 rc
= established_get_next(seq
, v
);
2297 st
->last_pos
= *pos
;
2301 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2303 struct tcp_iter_state
*st
= seq
->private;
2305 switch (st
->state
) {
2306 case TCP_SEQ_STATE_OPENREQ
:
2308 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2309 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2311 case TCP_SEQ_STATE_LISTENING
:
2312 if (v
!= SEQ_START_TOKEN
)
2313 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2315 case TCP_SEQ_STATE_TIME_WAIT
:
2316 case TCP_SEQ_STATE_ESTABLISHED
:
2318 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2323 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2325 struct tcp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
2326 struct tcp_iter_state
*s
;
2329 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2330 sizeof(struct tcp_iter_state
));
2334 s
= ((struct seq_file
*)file
->private_data
)->private;
2335 s
->family
= afinfo
->family
;
2339 EXPORT_SYMBOL(tcp_seq_open
);
2341 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2344 struct proc_dir_entry
*p
;
2346 afinfo
->seq_ops
.start
= tcp_seq_start
;
2347 afinfo
->seq_ops
.next
= tcp_seq_next
;
2348 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2350 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2351 afinfo
->seq_fops
, afinfo
);
2356 EXPORT_SYMBOL(tcp_proc_register
);
2358 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2360 proc_net_remove(net
, afinfo
->name
);
2362 EXPORT_SYMBOL(tcp_proc_unregister
);
2364 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2365 struct seq_file
*f
, int i
, int uid
, int *len
)
2367 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2368 int ttd
= req
->expires
- jiffies
;
2370 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2371 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2374 ntohs(inet_sk(sk
)->inet_sport
),
2376 ntohs(ireq
->rmt_port
),
2378 0, 0, /* could print option size, but that is af dependent. */
2379 1, /* timers active (only the expire timer) */
2380 jiffies_to_clock_t(ttd
),
2383 0, /* non standard timer */
2384 0, /* open_requests have no inode */
2385 atomic_read(&sk
->sk_refcnt
),
2390 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
, int *len
)
2393 unsigned long timer_expires
;
2394 const struct tcp_sock
*tp
= tcp_sk(sk
);
2395 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2396 const struct inet_sock
*inet
= inet_sk(sk
);
2397 __be32 dest
= inet
->inet_daddr
;
2398 __be32 src
= inet
->inet_rcv_saddr
;
2399 __u16 destp
= ntohs(inet
->inet_dport
);
2400 __u16 srcp
= ntohs(inet
->inet_sport
);
2403 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
) {
2405 timer_expires
= icsk
->icsk_timeout
;
2406 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2408 timer_expires
= icsk
->icsk_timeout
;
2409 } else if (timer_pending(&sk
->sk_timer
)) {
2411 timer_expires
= sk
->sk_timer
.expires
;
2414 timer_expires
= jiffies
;
2417 if (sk
->sk_state
== TCP_LISTEN
)
2418 rx_queue
= sk
->sk_ack_backlog
;
2421 * because we dont lock socket, we might find a transient negative value
2423 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2425 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2426 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2427 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2428 tp
->write_seq
- tp
->snd_una
,
2431 jiffies_to_clock_t(timer_expires
- jiffies
),
2432 icsk
->icsk_retransmits
,
2434 icsk
->icsk_probes_out
,
2436 atomic_read(&sk
->sk_refcnt
), sk
,
2437 jiffies_to_clock_t(icsk
->icsk_rto
),
2438 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2439 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2441 tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
,
2445 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2446 struct seq_file
*f
, int i
, int *len
)
2450 int ttd
= tw
->tw_ttd
- jiffies
;
2455 dest
= tw
->tw_daddr
;
2456 src
= tw
->tw_rcv_saddr
;
2457 destp
= ntohs(tw
->tw_dport
);
2458 srcp
= ntohs(tw
->tw_sport
);
2460 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2461 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2462 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2463 3, jiffies_to_clock_t(ttd
), 0, 0, 0, 0,
2464 atomic_read(&tw
->tw_refcnt
), tw
, len
);
2469 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2471 struct tcp_iter_state
*st
;
2474 if (v
== SEQ_START_TOKEN
) {
2475 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2476 " sl local_address rem_address st tx_queue "
2477 "rx_queue tr tm->when retrnsmt uid timeout "
2483 switch (st
->state
) {
2484 case TCP_SEQ_STATE_LISTENING
:
2485 case TCP_SEQ_STATE_ESTABLISHED
:
2486 get_tcp4_sock(v
, seq
, st
->num
, &len
);
2488 case TCP_SEQ_STATE_OPENREQ
:
2489 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
, &len
);
2491 case TCP_SEQ_STATE_TIME_WAIT
:
2492 get_timewait4_sock(v
, seq
, st
->num
, &len
);
2495 seq_printf(seq
, "%*s\n", TMPSZ
- 1 - len
, "");
2500 static const struct file_operations tcp_afinfo_seq_fops
= {
2501 .owner
= THIS_MODULE
,
2502 .open
= tcp_seq_open
,
2504 .llseek
= seq_lseek
,
2505 .release
= seq_release_net
2508 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2511 .seq_fops
= &tcp_afinfo_seq_fops
,
2513 .show
= tcp4_seq_show
,
2517 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2519 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2522 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2524 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2527 static struct pernet_operations tcp4_net_ops
= {
2528 .init
= tcp4_proc_init_net
,
2529 .exit
= tcp4_proc_exit_net
,
2532 int __init
tcp4_proc_init(void)
2534 return register_pernet_subsys(&tcp4_net_ops
);
2537 void tcp4_proc_exit(void)
2539 unregister_pernet_subsys(&tcp4_net_ops
);
2541 #endif /* CONFIG_PROC_FS */
2543 struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2545 const struct iphdr
*iph
= skb_gro_network_header(skb
);
2547 switch (skb
->ip_summed
) {
2548 case CHECKSUM_COMPLETE
:
2549 if (!tcp_v4_check(skb_gro_len(skb
), iph
->saddr
, iph
->daddr
,
2551 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2557 NAPI_GRO_CB(skb
)->flush
= 1;
2561 return tcp_gro_receive(head
, skb
);
2564 int tcp4_gro_complete(struct sk_buff
*skb
)
2566 const struct iphdr
*iph
= ip_hdr(skb
);
2567 struct tcphdr
*th
= tcp_hdr(skb
);
2569 th
->check
= ~tcp_v4_check(skb
->len
- skb_transport_offset(skb
),
2570 iph
->saddr
, iph
->daddr
, 0);
2571 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
2573 return tcp_gro_complete(skb
);
2576 struct proto tcp_prot
= {
2578 .owner
= THIS_MODULE
,
2580 .connect
= tcp_v4_connect
,
2581 .disconnect
= tcp_disconnect
,
2582 .accept
= inet_csk_accept
,
2584 .init
= tcp_v4_init_sock
,
2585 .destroy
= tcp_v4_destroy_sock
,
2586 .shutdown
= tcp_shutdown
,
2587 .setsockopt
= tcp_setsockopt
,
2588 .getsockopt
= tcp_getsockopt
,
2589 .recvmsg
= tcp_recvmsg
,
2590 .sendmsg
= tcp_sendmsg
,
2591 .sendpage
= tcp_sendpage
,
2592 .backlog_rcv
= tcp_v4_do_rcv
,
2594 .unhash
= inet_unhash
,
2595 .get_port
= inet_csk_get_port
,
2596 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2597 .sockets_allocated
= &tcp_sockets_allocated
,
2598 .orphan_count
= &tcp_orphan_count
,
2599 .memory_allocated
= &tcp_memory_allocated
,
2600 .memory_pressure
= &tcp_memory_pressure
,
2601 .sysctl_wmem
= sysctl_tcp_wmem
,
2602 .sysctl_rmem
= sysctl_tcp_rmem
,
2603 .max_header
= MAX_TCP_HEADER
,
2604 .obj_size
= sizeof(struct tcp_sock
),
2605 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2606 .twsk_prot
= &tcp_timewait_sock_ops
,
2607 .rsk_prot
= &tcp_request_sock_ops
,
2608 .h
.hashinfo
= &tcp_hashinfo
,
2609 .no_autobind
= true,
2610 #ifdef CONFIG_COMPAT
2611 .compat_setsockopt
= compat_tcp_setsockopt
,
2612 .compat_getsockopt
= compat_tcp_getsockopt
,
2614 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2615 .init_cgroup
= tcp_init_cgroup
,
2616 .destroy_cgroup
= tcp_destroy_cgroup
,
2617 .proto_cgroup
= tcp_proto_cgroup
,
2620 EXPORT_SYMBOL(tcp_prot
);
2622 static int __net_init
tcp_sk_init(struct net
*net
)
2624 return inet_ctl_sock_create(&net
->ipv4
.tcp_sock
,
2625 PF_INET
, SOCK_RAW
, IPPROTO_TCP
, net
);
2628 static void __net_exit
tcp_sk_exit(struct net
*net
)
2630 inet_ctl_sock_destroy(net
->ipv4
.tcp_sock
);
2633 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2635 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2638 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2639 .init
= tcp_sk_init
,
2640 .exit
= tcp_sk_exit
,
2641 .exit_batch
= tcp_sk_exit_batch
,
2644 void __init
tcp_v4_init(void)
2646 inet_hashinfo_init(&tcp_hashinfo
);
2647 if (register_pernet_subsys(&tcp_sk_ops
))
2648 panic("Failed to create the TCP control socket.\n");