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 tcp_fetch_timewait_stamp(sk
, &rt
->dst
);
215 inet
->inet_dport
= usin
->sin_port
;
216 inet
->inet_daddr
= daddr
;
218 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
220 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
222 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
224 /* Socket identity is still unknown (sport may be zero).
225 * However we set state to SYN-SENT and not releasing socket
226 * lock select source port, enter ourselves into the hash tables and
227 * complete initialization after this.
229 tcp_set_state(sk
, TCP_SYN_SENT
);
230 err
= inet_hash_connect(&tcp_death_row
, sk
);
234 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
235 inet
->inet_sport
, inet
->inet_dport
, sk
);
241 /* OK, now commit destination to socket. */
242 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
243 sk_setup_caps(sk
, &rt
->dst
);
245 if (!tp
->write_seq
&& likely(!tp
->repair
))
246 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
251 inet
->inet_id
= tp
->write_seq
^ jiffies
;
253 if (likely(!tp
->repair
))
254 err
= tcp_connect(sk
);
256 err
= tcp_repair_connect(sk
);
266 * This unhashes the socket and releases the local port,
269 tcp_set_state(sk
, TCP_CLOSE
);
271 sk
->sk_route_caps
= 0;
272 inet
->inet_dport
= 0;
275 EXPORT_SYMBOL(tcp_v4_connect
);
278 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
279 * It can be called through tcp_release_cb() if socket was owned by user
280 * at the time tcp_v4_err() was called to handle ICMP message.
282 static void tcp_v4_mtu_reduced(struct sock
*sk
)
284 struct dst_entry
*dst
;
285 struct inet_sock
*inet
= inet_sk(sk
);
286 u32 mtu
= tcp_sk(sk
)->mtu_info
;
288 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
289 * send out by Linux are always <576bytes so they should go through
292 if (sk
->sk_state
== TCP_LISTEN
)
295 dst
= inet_csk_update_pmtu(sk
, mtu
);
299 /* Something is about to be wrong... Remember soft error
300 * for the case, if this connection will not able to recover.
302 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
303 sk
->sk_err_soft
= EMSGSIZE
;
307 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
308 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
309 tcp_sync_mss(sk
, mtu
);
311 /* Resend the TCP packet because it's
312 * clear that the old packet has been
313 * dropped. This is the new "fast" path mtu
316 tcp_simple_retransmit(sk
);
317 } /* else let the usual retransmit timer handle it */
320 static void do_redirect(struct sk_buff
*skb
, struct sock
*sk
)
322 struct dst_entry
*dst
= __sk_dst_check(sk
, 0);
325 dst
->ops
->redirect(dst
, sk
, skb
);
329 * This routine is called by the ICMP module when it gets some
330 * sort of error condition. If err < 0 then the socket should
331 * be closed and the error returned to the user. If err > 0
332 * it's just the icmp type << 8 | icmp code. After adjustment
333 * header points to the first 8 bytes of the tcp header. We need
334 * to find the appropriate port.
336 * The locking strategy used here is very "optimistic". When
337 * someone else accesses the socket the ICMP is just dropped
338 * and for some paths there is no check at all.
339 * A more general error queue to queue errors for later handling
340 * is probably better.
344 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
346 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
347 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
348 struct inet_connection_sock
*icsk
;
350 struct inet_sock
*inet
;
351 const int type
= icmp_hdr(icmp_skb
)->type
;
352 const int code
= icmp_hdr(icmp_skb
)->code
;
358 struct net
*net
= dev_net(icmp_skb
->dev
);
360 if (icmp_skb
->len
< (iph
->ihl
<< 2) + 8) {
361 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
365 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
366 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
368 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
371 if (sk
->sk_state
== TCP_TIME_WAIT
) {
372 inet_twsk_put(inet_twsk(sk
));
377 /* If too many ICMPs get dropped on busy
378 * servers this needs to be solved differently.
379 * We do take care of PMTU discovery (RFC1191) special case :
380 * we can receive locally generated ICMP messages while socket is held.
382 if (sock_owned_by_user(sk
) &&
383 type
!= ICMP_DEST_UNREACH
&&
384 code
!= ICMP_FRAG_NEEDED
)
385 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
387 if (sk
->sk_state
== TCP_CLOSE
)
390 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
391 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
397 seq
= ntohl(th
->seq
);
398 if (sk
->sk_state
!= TCP_LISTEN
&&
399 !between(seq
, tp
->snd_una
, tp
->snd_nxt
)) {
400 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
406 do_redirect(icmp_skb
, sk
);
408 case ICMP_SOURCE_QUENCH
:
409 /* Just silently ignore these. */
411 case ICMP_PARAMETERPROB
:
414 case ICMP_DEST_UNREACH
:
415 if (code
> NR_ICMP_UNREACH
)
418 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
420 if (!sock_owned_by_user(sk
))
421 tcp_v4_mtu_reduced(sk
);
423 set_bit(TCP_MTU_REDUCED_DEFERRED
, &tp
->tsq_flags
);
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_unicast_reply(net
, skb
, ip_hdr(skb
)->saddr
,
702 ip_hdr(skb
)->daddr
, &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_unicast_reply(net
, skb
, ip_hdr(skb
)->saddr
,
785 ip_hdr(skb
)->daddr
, &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
,
831 const struct inet_request_sock
*ireq
= inet_rsk(req
);
834 struct sk_buff
* skb
;
836 /* First, grab a route. */
837 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
840 skb
= tcp_make_synack(sk
, dst
, req
, rvp
);
843 __tcp_v4_send_check(skb
, ireq
->loc_addr
, ireq
->rmt_addr
);
845 skb_set_queue_mapping(skb
, queue_mapping
);
846 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
849 err
= net_xmit_eval(err
);
855 static int tcp_v4_rtx_synack(struct sock
*sk
, struct request_sock
*req
,
856 struct request_values
*rvp
)
858 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
859 return tcp_v4_send_synack(sk
, NULL
, req
, rvp
, 0, false);
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
) {
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 sock
*sk
,
908 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
909 struct ip_options_rcu
*dopt
= NULL
;
911 if (opt
&& opt
->optlen
) {
912 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
914 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
916 if (ip_options_echo(&dopt
->opt
, skb
)) {
925 #ifdef CONFIG_TCP_MD5SIG
927 * RFC2385 MD5 checksumming requires a mapping of
928 * IP address->MD5 Key.
929 * We need to maintain these in the sk structure.
932 /* Find the Key structure for an address. */
933 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
934 const union tcp_md5_addr
*addr
,
937 struct tcp_sock
*tp
= tcp_sk(sk
);
938 struct tcp_md5sig_key
*key
;
939 struct hlist_node
*pos
;
940 unsigned int size
= sizeof(struct in_addr
);
941 struct tcp_md5sig_info
*md5sig
;
943 /* caller either holds rcu_read_lock() or socket lock */
944 md5sig
= rcu_dereference_check(tp
->md5sig_info
,
945 sock_owned_by_user(sk
) ||
946 lockdep_is_held(&sk
->sk_lock
.slock
));
949 #if IS_ENABLED(CONFIG_IPV6)
950 if (family
== AF_INET6
)
951 size
= sizeof(struct in6_addr
);
953 hlist_for_each_entry_rcu(key
, pos
, &md5sig
->head
, node
) {
954 if (key
->family
!= family
)
956 if (!memcmp(&key
->addr
, addr
, size
))
961 EXPORT_SYMBOL(tcp_md5_do_lookup
);
963 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
964 struct sock
*addr_sk
)
966 union tcp_md5_addr
*addr
;
968 addr
= (union tcp_md5_addr
*)&inet_sk(addr_sk
)->inet_daddr
;
969 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
971 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
973 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
974 struct request_sock
*req
)
976 union tcp_md5_addr
*addr
;
978 addr
= (union tcp_md5_addr
*)&inet_rsk(req
)->rmt_addr
;
979 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
982 /* This can be called on a newly created socket, from other files */
983 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
984 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
)
986 /* Add Key to the list */
987 struct tcp_md5sig_key
*key
;
988 struct tcp_sock
*tp
= tcp_sk(sk
);
989 struct tcp_md5sig_info
*md5sig
;
991 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&addr
, AF_INET
);
993 /* Pre-existing entry - just update that one. */
994 memcpy(key
->key
, newkey
, newkeylen
);
995 key
->keylen
= newkeylen
;
999 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
1000 sock_owned_by_user(sk
));
1002 md5sig
= kmalloc(sizeof(*md5sig
), gfp
);
1006 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
1007 INIT_HLIST_HEAD(&md5sig
->head
);
1008 rcu_assign_pointer(tp
->md5sig_info
, md5sig
);
1011 key
= sock_kmalloc(sk
, sizeof(*key
), gfp
);
1014 if (hlist_empty(&md5sig
->head
) && !tcp_alloc_md5sig_pool(sk
)) {
1015 sock_kfree_s(sk
, key
, sizeof(*key
));
1019 memcpy(key
->key
, newkey
, newkeylen
);
1020 key
->keylen
= newkeylen
;
1021 key
->family
= family
;
1022 memcpy(&key
->addr
, addr
,
1023 (family
== AF_INET6
) ? sizeof(struct in6_addr
) :
1024 sizeof(struct in_addr
));
1025 hlist_add_head_rcu(&key
->node
, &md5sig
->head
);
1028 EXPORT_SYMBOL(tcp_md5_do_add
);
1030 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
, int family
)
1032 struct tcp_sock
*tp
= tcp_sk(sk
);
1033 struct tcp_md5sig_key
*key
;
1034 struct tcp_md5sig_info
*md5sig
;
1036 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&addr
, AF_INET
);
1039 hlist_del_rcu(&key
->node
);
1040 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1041 kfree_rcu(key
, rcu
);
1042 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
1043 sock_owned_by_user(sk
));
1044 if (hlist_empty(&md5sig
->head
))
1045 tcp_free_md5sig_pool();
1048 EXPORT_SYMBOL(tcp_md5_do_del
);
1050 void tcp_clear_md5_list(struct sock
*sk
)
1052 struct tcp_sock
*tp
= tcp_sk(sk
);
1053 struct tcp_md5sig_key
*key
;
1054 struct hlist_node
*pos
, *n
;
1055 struct tcp_md5sig_info
*md5sig
;
1057 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
1059 if (!hlist_empty(&md5sig
->head
))
1060 tcp_free_md5sig_pool();
1061 hlist_for_each_entry_safe(key
, pos
, n
, &md5sig
->head
, node
) {
1062 hlist_del_rcu(&key
->node
);
1063 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1064 kfree_rcu(key
, rcu
);
1068 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1071 struct tcp_md5sig cmd
;
1072 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1074 if (optlen
< sizeof(cmd
))
1077 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1080 if (sin
->sin_family
!= AF_INET
)
1083 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
)
1084 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1087 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1090 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1091 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1095 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1096 __be32 daddr
, __be32 saddr
, int nbytes
)
1098 struct tcp4_pseudohdr
*bp
;
1099 struct scatterlist sg
;
1101 bp
= &hp
->md5_blk
.ip4
;
1104 * 1. the TCP pseudo-header (in the order: source IP address,
1105 * destination IP address, zero-padded protocol number, and
1111 bp
->protocol
= IPPROTO_TCP
;
1112 bp
->len
= cpu_to_be16(nbytes
);
1114 sg_init_one(&sg
, bp
, sizeof(*bp
));
1115 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1118 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1119 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1121 struct tcp_md5sig_pool
*hp
;
1122 struct hash_desc
*desc
;
1124 hp
= tcp_get_md5sig_pool();
1126 goto clear_hash_noput
;
1127 desc
= &hp
->md5_desc
;
1129 if (crypto_hash_init(desc
))
1131 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1133 if (tcp_md5_hash_header(hp
, th
))
1135 if (tcp_md5_hash_key(hp
, key
))
1137 if (crypto_hash_final(desc
, md5_hash
))
1140 tcp_put_md5sig_pool();
1144 tcp_put_md5sig_pool();
1146 memset(md5_hash
, 0, 16);
1150 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1151 const struct sock
*sk
, const struct request_sock
*req
,
1152 const struct sk_buff
*skb
)
1154 struct tcp_md5sig_pool
*hp
;
1155 struct hash_desc
*desc
;
1156 const struct tcphdr
*th
= tcp_hdr(skb
);
1157 __be32 saddr
, daddr
;
1160 saddr
= inet_sk(sk
)->inet_saddr
;
1161 daddr
= inet_sk(sk
)->inet_daddr
;
1163 saddr
= inet_rsk(req
)->loc_addr
;
1164 daddr
= inet_rsk(req
)->rmt_addr
;
1166 const struct iphdr
*iph
= ip_hdr(skb
);
1171 hp
= tcp_get_md5sig_pool();
1173 goto clear_hash_noput
;
1174 desc
= &hp
->md5_desc
;
1176 if (crypto_hash_init(desc
))
1179 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1181 if (tcp_md5_hash_header(hp
, th
))
1183 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1185 if (tcp_md5_hash_key(hp
, key
))
1187 if (crypto_hash_final(desc
, md5_hash
))
1190 tcp_put_md5sig_pool();
1194 tcp_put_md5sig_pool();
1196 memset(md5_hash
, 0, 16);
1199 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1201 static bool tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1204 * This gets called for each TCP segment that arrives
1205 * so we want to be efficient.
1206 * We have 3 drop cases:
1207 * o No MD5 hash and one expected.
1208 * o MD5 hash and we're not expecting one.
1209 * o MD5 hash and its wrong.
1211 const __u8
*hash_location
= NULL
;
1212 struct tcp_md5sig_key
*hash_expected
;
1213 const struct iphdr
*iph
= ip_hdr(skb
);
1214 const struct tcphdr
*th
= tcp_hdr(skb
);
1216 unsigned char newhash
[16];
1218 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1220 hash_location
= tcp_parse_md5sig_option(th
);
1222 /* We've parsed the options - do we have a hash? */
1223 if (!hash_expected
&& !hash_location
)
1226 if (hash_expected
&& !hash_location
) {
1227 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1231 if (!hash_expected
&& hash_location
) {
1232 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1236 /* Okay, so this is hash_expected and hash_location -
1237 * so we need to calculate the checksum.
1239 genhash
= tcp_v4_md5_hash_skb(newhash
,
1243 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1244 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1245 &iph
->saddr
, ntohs(th
->source
),
1246 &iph
->daddr
, ntohs(th
->dest
),
1247 genhash
? " tcp_v4_calc_md5_hash failed"
1256 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1258 .obj_size
= sizeof(struct tcp_request_sock
),
1259 .rtx_syn_ack
= tcp_v4_rtx_synack
,
1260 .send_ack
= tcp_v4_reqsk_send_ack
,
1261 .destructor
= tcp_v4_reqsk_destructor
,
1262 .send_reset
= tcp_v4_send_reset
,
1263 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1266 #ifdef CONFIG_TCP_MD5SIG
1267 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1268 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1269 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1273 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1275 struct tcp_extend_values tmp_ext
;
1276 struct tcp_options_received tmp_opt
;
1277 const u8
*hash_location
;
1278 struct request_sock
*req
;
1279 struct inet_request_sock
*ireq
;
1280 struct tcp_sock
*tp
= tcp_sk(sk
);
1281 struct dst_entry
*dst
= NULL
;
1282 __be32 saddr
= ip_hdr(skb
)->saddr
;
1283 __be32 daddr
= ip_hdr(skb
)->daddr
;
1284 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1285 bool want_cookie
= false;
1287 /* Never answer to SYNs send to broadcast or multicast */
1288 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1291 /* TW buckets are converted to open requests without
1292 * limitations, they conserve resources and peer is
1293 * evidently real one.
1295 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
1296 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCP");
1301 /* Accept backlog is full. If we have already queued enough
1302 * of warm entries in syn queue, drop request. It is better than
1303 * clogging syn queue with openreqs with exponentially increasing
1306 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1)
1309 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1313 #ifdef CONFIG_TCP_MD5SIG
1314 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1317 tcp_clear_options(&tmp_opt
);
1318 tmp_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
1319 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
1320 tcp_parse_options(skb
, &tmp_opt
, &hash_location
, 0, NULL
);
1322 if (tmp_opt
.cookie_plus
> 0 &&
1323 tmp_opt
.saw_tstamp
&&
1324 !tp
->rx_opt
.cookie_out_never
&&
1325 (sysctl_tcp_cookie_size
> 0 ||
1326 (tp
->cookie_values
!= NULL
&&
1327 tp
->cookie_values
->cookie_desired
> 0))) {
1329 u32
*mess
= &tmp_ext
.cookie_bakery
[COOKIE_DIGEST_WORDS
];
1330 int l
= tmp_opt
.cookie_plus
- TCPOLEN_COOKIE_BASE
;
1332 if (tcp_cookie_generator(&tmp_ext
.cookie_bakery
[0]) != 0)
1333 goto drop_and_release
;
1335 /* Secret recipe starts with IP addresses */
1336 *mess
++ ^= (__force u32
)daddr
;
1337 *mess
++ ^= (__force u32
)saddr
;
1339 /* plus variable length Initiator Cookie */
1342 *c
++ ^= *hash_location
++;
1344 want_cookie
= false; /* not our kind of cookie */
1345 tmp_ext
.cookie_out_never
= 0; /* false */
1346 tmp_ext
.cookie_plus
= tmp_opt
.cookie_plus
;
1347 } else if (!tp
->rx_opt
.cookie_in_always
) {
1348 /* redundant indications, but ensure initialization. */
1349 tmp_ext
.cookie_out_never
= 1; /* true */
1350 tmp_ext
.cookie_plus
= 0;
1352 goto drop_and_release
;
1354 tmp_ext
.cookie_in_always
= tp
->rx_opt
.cookie_in_always
;
1356 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1357 tcp_clear_options(&tmp_opt
);
1359 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1360 tcp_openreq_init(req
, &tmp_opt
, skb
);
1362 ireq
= inet_rsk(req
);
1363 ireq
->loc_addr
= daddr
;
1364 ireq
->rmt_addr
= saddr
;
1365 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1366 ireq
->opt
= tcp_v4_save_options(sk
, skb
);
1368 if (security_inet_conn_request(sk
, skb
, req
))
1371 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1372 TCP_ECN_create_request(req
, skb
);
1375 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1376 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1380 /* VJ's idea. We save last timestamp seen
1381 * from the destination in peer table, when entering
1382 * state TIME-WAIT, and check against it before
1383 * accepting new connection request.
1385 * If "isn" is not zero, this request hit alive
1386 * timewait bucket, so that all the necessary checks
1387 * are made in the function processing timewait state.
1389 if (tmp_opt
.saw_tstamp
&&
1390 tcp_death_row
.sysctl_tw_recycle
&&
1391 (dst
= inet_csk_route_req(sk
, &fl4
, req
)) != NULL
&&
1392 fl4
.daddr
== saddr
) {
1393 if (!tcp_peer_is_proven(req
, dst
, true)) {
1394 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1395 goto drop_and_release
;
1398 /* Kill the following clause, if you dislike this way. */
1399 else if (!sysctl_tcp_syncookies
&&
1400 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1401 (sysctl_max_syn_backlog
>> 2)) &&
1402 !tcp_peer_is_proven(req
, dst
, false)) {
1403 /* Without syncookies last quarter of
1404 * backlog is filled with destinations,
1405 * proven to be alive.
1406 * It means that we continue to communicate
1407 * to destinations, already remembered
1408 * to the moment of synflood.
1410 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("drop open request from %pI4/%u\n"),
1411 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1412 goto drop_and_release
;
1415 isn
= tcp_v4_init_sequence(skb
);
1417 tcp_rsk(req
)->snt_isn
= isn
;
1418 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1420 if (tcp_v4_send_synack(sk
, dst
, req
,
1421 (struct request_values
*)&tmp_ext
,
1422 skb_get_queue_mapping(skb
),
1427 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1437 EXPORT_SYMBOL(tcp_v4_conn_request
);
1441 * The three way handshake has completed - we got a valid synack -
1442 * now create the new socket.
1444 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1445 struct request_sock
*req
,
1446 struct dst_entry
*dst
)
1448 struct inet_request_sock
*ireq
;
1449 struct inet_sock
*newinet
;
1450 struct tcp_sock
*newtp
;
1452 #ifdef CONFIG_TCP_MD5SIG
1453 struct tcp_md5sig_key
*key
;
1455 struct ip_options_rcu
*inet_opt
;
1457 if (sk_acceptq_is_full(sk
))
1460 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1464 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1466 newtp
= tcp_sk(newsk
);
1467 newinet
= inet_sk(newsk
);
1468 ireq
= inet_rsk(req
);
1469 newinet
->inet_daddr
= ireq
->rmt_addr
;
1470 newinet
->inet_rcv_saddr
= ireq
->loc_addr
;
1471 newinet
->inet_saddr
= ireq
->loc_addr
;
1472 inet_opt
= ireq
->opt
;
1473 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1475 newinet
->mc_index
= inet_iif(skb
);
1476 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1477 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1478 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1480 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1481 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1484 dst
= inet_csk_route_child_sock(sk
, newsk
, req
);
1488 /* syncookie case : see end of cookie_v4_check() */
1490 sk_setup_caps(newsk
, dst
);
1492 tcp_mtup_init(newsk
);
1493 tcp_sync_mss(newsk
, dst_mtu(dst
));
1494 newtp
->advmss
= dst_metric_advmss(dst
);
1495 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1496 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1497 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1499 tcp_initialize_rcv_mss(newsk
);
1500 if (tcp_rsk(req
)->snt_synack
)
1501 tcp_valid_rtt_meas(newsk
,
1502 tcp_time_stamp
- tcp_rsk(req
)->snt_synack
);
1503 newtp
->total_retrans
= req
->retrans
;
1505 #ifdef CONFIG_TCP_MD5SIG
1506 /* Copy over the MD5 key from the original socket */
1507 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1511 * We're using one, so create a matching key
1512 * on the newsk structure. If we fail to get
1513 * memory, then we end up not copying the key
1516 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1517 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1518 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1522 if (__inet_inherit_port(sk
, newsk
) < 0)
1524 __inet_hash_nolisten(newsk
, NULL
);
1529 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1533 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1536 tcp_clear_xmit_timers(newsk
);
1537 tcp_cleanup_congestion_control(newsk
);
1538 bh_unlock_sock(newsk
);
1542 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1544 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1546 struct tcphdr
*th
= tcp_hdr(skb
);
1547 const struct iphdr
*iph
= ip_hdr(skb
);
1549 struct request_sock
**prev
;
1550 /* Find possible connection requests. */
1551 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1552 iph
->saddr
, iph
->daddr
);
1554 return tcp_check_req(sk
, skb
, req
, prev
);
1556 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1557 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1560 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1564 inet_twsk_put(inet_twsk(nsk
));
1568 #ifdef CONFIG_SYN_COOKIES
1570 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1575 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1577 const struct iphdr
*iph
= ip_hdr(skb
);
1579 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1580 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1581 iph
->daddr
, skb
->csum
)) {
1582 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1587 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1588 skb
->len
, IPPROTO_TCP
, 0);
1590 if (skb
->len
<= 76) {
1591 return __skb_checksum_complete(skb
);
1597 /* The socket must have it's spinlock held when we get
1600 * We have a potential double-lock case here, so even when
1601 * doing backlog processing we use the BH locking scheme.
1602 * This is because we cannot sleep with the original spinlock
1605 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1608 #ifdef CONFIG_TCP_MD5SIG
1610 * We really want to reject the packet as early as possible
1612 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1613 * o There is an MD5 option and we're not expecting one
1615 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1619 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1620 sock_rps_save_rxhash(sk
, skb
);
1621 if (sk
->sk_rx_dst
) {
1622 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1623 if (dst
->ops
->check(dst
, 0) == NULL
) {
1625 sk
->sk_rx_dst
= NULL
;
1628 if (unlikely(sk
->sk_rx_dst
== NULL
)) {
1629 struct inet_sock
*icsk
= inet_sk(sk
);
1630 struct rtable
*rt
= skb_rtable(skb
);
1632 sk
->sk_rx_dst
= dst_clone(&rt
->dst
);
1633 icsk
->rx_dst_ifindex
= inet_iif(skb
);
1635 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1642 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1645 if (sk
->sk_state
== TCP_LISTEN
) {
1646 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1651 sock_rps_save_rxhash(nsk
, skb
);
1652 if (tcp_child_process(sk
, nsk
, skb
)) {
1659 sock_rps_save_rxhash(sk
, skb
);
1661 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1668 tcp_v4_send_reset(rsk
, skb
);
1671 /* Be careful here. If this function gets more complicated and
1672 * gcc suffers from register pressure on the x86, sk (in %ebx)
1673 * might be destroyed here. This current version compiles correctly,
1674 * but you have been warned.
1679 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1682 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1684 void tcp_v4_early_demux(struct sk_buff
*skb
)
1686 struct net
*net
= dev_net(skb
->dev
);
1687 const struct iphdr
*iph
;
1688 const struct tcphdr
*th
;
1691 if (skb
->pkt_type
!= PACKET_HOST
)
1694 if (!pskb_may_pull(skb
, ip_hdrlen(skb
) + sizeof(struct tcphdr
)))
1698 th
= (struct tcphdr
*) ((char *)iph
+ ip_hdrlen(skb
));
1700 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1703 sk
= __inet_lookup_established(net
, &tcp_hashinfo
,
1704 iph
->saddr
, th
->source
,
1705 iph
->daddr
, ntohs(th
->dest
),
1709 skb
->destructor
= sock_edemux
;
1710 if (sk
->sk_state
!= TCP_TIME_WAIT
) {
1711 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1712 struct inet_sock
*icsk
= inet_sk(sk
);
1714 dst
= dst_check(dst
, 0);
1716 icsk
->rx_dst_ifindex
== skb
->skb_iif
)
1717 skb_dst_set_noref(skb
, dst
);
1726 int tcp_v4_rcv(struct sk_buff
*skb
)
1728 const struct iphdr
*iph
;
1729 const struct tcphdr
*th
;
1732 struct net
*net
= dev_net(skb
->dev
);
1734 if (skb
->pkt_type
!= PACKET_HOST
)
1737 /* Count it even if it's bad */
1738 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1740 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
1745 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1747 if (!pskb_may_pull(skb
, th
->doff
* 4))
1750 /* An explanation is required here, I think.
1751 * Packet length and doff are validated by header prediction,
1752 * provided case of th->doff==0 is eliminated.
1753 * So, we defer the checks. */
1754 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
1759 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
1760 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
1761 skb
->len
- th
->doff
* 4);
1762 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
1763 TCP_SKB_CB(skb
)->when
= 0;
1764 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
1765 TCP_SKB_CB(skb
)->sacked
= 0;
1767 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
1772 if (sk
->sk_state
== TCP_TIME_WAIT
)
1775 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
1776 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
1777 goto discard_and_relse
;
1780 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
1781 goto discard_and_relse
;
1784 if (sk_filter(sk
, skb
))
1785 goto discard_and_relse
;
1789 bh_lock_sock_nested(sk
);
1791 if (!sock_owned_by_user(sk
)) {
1792 #ifdef CONFIG_NET_DMA
1793 struct tcp_sock
*tp
= tcp_sk(sk
);
1794 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
1795 tp
->ucopy
.dma_chan
= net_dma_find_channel();
1796 if (tp
->ucopy
.dma_chan
)
1797 ret
= tcp_v4_do_rcv(sk
, skb
);
1801 if (!tcp_prequeue(sk
, skb
))
1802 ret
= tcp_v4_do_rcv(sk
, skb
);
1804 } else if (unlikely(sk_add_backlog(sk
, skb
,
1805 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
1807 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
1808 goto discard_and_relse
;
1817 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
1820 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1822 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1824 tcp_v4_send_reset(NULL
, skb
);
1828 /* Discard frame. */
1837 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
1838 inet_twsk_put(inet_twsk(sk
));
1842 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
1843 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
1844 inet_twsk_put(inet_twsk(sk
));
1847 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
1849 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
1851 iph
->daddr
, th
->dest
,
1854 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
1855 inet_twsk_put(inet_twsk(sk
));
1859 /* Fall through to ACK */
1862 tcp_v4_timewait_ack(sk
, skb
);
1866 case TCP_TW_SUCCESS
:;
1871 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
1872 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
1873 .twsk_unique
= tcp_twsk_unique
,
1874 .twsk_destructor
= tcp_twsk_destructor
,
1877 const struct inet_connection_sock_af_ops ipv4_specific
= {
1878 .queue_xmit
= ip_queue_xmit
,
1879 .send_check
= tcp_v4_send_check
,
1880 .rebuild_header
= inet_sk_rebuild_header
,
1881 .conn_request
= tcp_v4_conn_request
,
1882 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
1883 .net_header_len
= sizeof(struct iphdr
),
1884 .setsockopt
= ip_setsockopt
,
1885 .getsockopt
= ip_getsockopt
,
1886 .addr2sockaddr
= inet_csk_addr2sockaddr
,
1887 .sockaddr_len
= sizeof(struct sockaddr_in
),
1888 .bind_conflict
= inet_csk_bind_conflict
,
1889 #ifdef CONFIG_COMPAT
1890 .compat_setsockopt
= compat_ip_setsockopt
,
1891 .compat_getsockopt
= compat_ip_getsockopt
,
1894 EXPORT_SYMBOL(ipv4_specific
);
1896 #ifdef CONFIG_TCP_MD5SIG
1897 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
1898 .md5_lookup
= tcp_v4_md5_lookup
,
1899 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1900 .md5_parse
= tcp_v4_parse_md5_keys
,
1904 /* NOTE: A lot of things set to zero explicitly by call to
1905 * sk_alloc() so need not be done here.
1907 static int tcp_v4_init_sock(struct sock
*sk
)
1909 struct inet_connection_sock
*icsk
= inet_csk(sk
);
1913 icsk
->icsk_af_ops
= &ipv4_specific
;
1915 #ifdef CONFIG_TCP_MD5SIG
1916 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv4_specific
;
1922 void tcp_v4_destroy_sock(struct sock
*sk
)
1924 struct tcp_sock
*tp
= tcp_sk(sk
);
1926 tcp_clear_xmit_timers(sk
);
1928 tcp_cleanup_congestion_control(sk
);
1930 /* Cleanup up the write buffer. */
1931 tcp_write_queue_purge(sk
);
1933 /* Cleans up our, hopefully empty, out_of_order_queue. */
1934 __skb_queue_purge(&tp
->out_of_order_queue
);
1936 #ifdef CONFIG_TCP_MD5SIG
1937 /* Clean up the MD5 key list, if any */
1938 if (tp
->md5sig_info
) {
1939 tcp_clear_md5_list(sk
);
1940 kfree_rcu(tp
->md5sig_info
, rcu
);
1941 tp
->md5sig_info
= NULL
;
1945 #ifdef CONFIG_NET_DMA
1946 /* Cleans up our sk_async_wait_queue */
1947 __skb_queue_purge(&sk
->sk_async_wait_queue
);
1950 /* Clean prequeue, it must be empty really */
1951 __skb_queue_purge(&tp
->ucopy
.prequeue
);
1953 /* Clean up a referenced TCP bind bucket. */
1954 if (inet_csk(sk
)->icsk_bind_hash
)
1958 * If sendmsg cached page exists, toss it.
1960 if (sk
->sk_sndmsg_page
) {
1961 __free_page(sk
->sk_sndmsg_page
);
1962 sk
->sk_sndmsg_page
= NULL
;
1965 /* TCP Cookie Transactions */
1966 if (tp
->cookie_values
!= NULL
) {
1967 kref_put(&tp
->cookie_values
->kref
,
1968 tcp_cookie_values_release
);
1969 tp
->cookie_values
= NULL
;
1972 /* If socket is aborted during connect operation */
1973 tcp_free_fastopen_req(tp
);
1975 sk_sockets_allocated_dec(sk
);
1976 sock_release_memcg(sk
);
1978 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
1980 #ifdef CONFIG_PROC_FS
1981 /* Proc filesystem TCP sock list dumping. */
1983 static inline struct inet_timewait_sock
*tw_head(struct hlist_nulls_head
*head
)
1985 return hlist_nulls_empty(head
) ? NULL
:
1986 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
1989 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
1991 return !is_a_nulls(tw
->tw_node
.next
) ?
1992 hlist_nulls_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
1996 * Get next listener socket follow cur. If cur is NULL, get first socket
1997 * starting from bucket given in st->bucket; when st->bucket is zero the
1998 * very first socket in the hash table is returned.
2000 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
2002 struct inet_connection_sock
*icsk
;
2003 struct hlist_nulls_node
*node
;
2004 struct sock
*sk
= cur
;
2005 struct inet_listen_hashbucket
*ilb
;
2006 struct tcp_iter_state
*st
= seq
->private;
2007 struct net
*net
= seq_file_net(seq
);
2010 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2011 spin_lock_bh(&ilb
->lock
);
2012 sk
= sk_nulls_head(&ilb
->head
);
2016 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2020 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
2021 struct request_sock
*req
= cur
;
2023 icsk
= inet_csk(st
->syn_wait_sk
);
2027 if (req
->rsk_ops
->family
== st
->family
) {
2033 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
2036 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
2038 sk
= sk_nulls_next(st
->syn_wait_sk
);
2039 st
->state
= TCP_SEQ_STATE_LISTENING
;
2040 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2042 icsk
= inet_csk(sk
);
2043 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2044 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
2046 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2047 sk
= sk_nulls_next(sk
);
2050 sk_nulls_for_each_from(sk
, node
) {
2051 if (!net_eq(sock_net(sk
), net
))
2053 if (sk
->sk_family
== st
->family
) {
2057 icsk
= inet_csk(sk
);
2058 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2059 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2061 st
->uid
= sock_i_uid(sk
);
2062 st
->syn_wait_sk
= sk
;
2063 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2067 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2069 spin_unlock_bh(&ilb
->lock
);
2071 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2072 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2073 spin_lock_bh(&ilb
->lock
);
2074 sk
= sk_nulls_head(&ilb
->head
);
2082 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2084 struct tcp_iter_state
*st
= seq
->private;
2089 rc
= listening_get_next(seq
, NULL
);
2091 while (rc
&& *pos
) {
2092 rc
= listening_get_next(seq
, rc
);
2098 static inline bool empty_bucket(struct tcp_iter_state
*st
)
2100 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
) &&
2101 hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2105 * Get first established socket starting from bucket given in st->bucket.
2106 * If st->bucket is zero, the very first socket in the hash is returned.
2108 static void *established_get_first(struct seq_file
*seq
)
2110 struct tcp_iter_state
*st
= seq
->private;
2111 struct net
*net
= seq_file_net(seq
);
2115 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2117 struct hlist_nulls_node
*node
;
2118 struct inet_timewait_sock
*tw
;
2119 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2121 /* Lockless fast path for the common case of empty buckets */
2122 if (empty_bucket(st
))
2126 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2127 if (sk
->sk_family
!= st
->family
||
2128 !net_eq(sock_net(sk
), net
)) {
2134 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2135 inet_twsk_for_each(tw
, node
,
2136 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
2137 if (tw
->tw_family
!= st
->family
||
2138 !net_eq(twsk_net(tw
), net
)) {
2144 spin_unlock_bh(lock
);
2145 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2151 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2153 struct sock
*sk
= cur
;
2154 struct inet_timewait_sock
*tw
;
2155 struct hlist_nulls_node
*node
;
2156 struct tcp_iter_state
*st
= seq
->private;
2157 struct net
*net
= seq_file_net(seq
);
2162 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2166 while (tw
&& (tw
->tw_family
!= st
->family
|| !net_eq(twsk_net(tw
), net
))) {
2173 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2174 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2176 /* Look for next non empty bucket */
2178 while (++st
->bucket
<= tcp_hashinfo
.ehash_mask
&&
2181 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2184 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2185 sk
= sk_nulls_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2187 sk
= sk_nulls_next(sk
);
2189 sk_nulls_for_each_from(sk
, node
) {
2190 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2194 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2195 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2203 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2205 struct tcp_iter_state
*st
= seq
->private;
2209 rc
= established_get_first(seq
);
2212 rc
= established_get_next(seq
, rc
);
2218 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2221 struct tcp_iter_state
*st
= seq
->private;
2223 st
->state
= TCP_SEQ_STATE_LISTENING
;
2224 rc
= listening_get_idx(seq
, &pos
);
2227 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2228 rc
= established_get_idx(seq
, pos
);
2234 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2236 struct tcp_iter_state
*st
= seq
->private;
2237 int offset
= st
->offset
;
2238 int orig_num
= st
->num
;
2241 switch (st
->state
) {
2242 case TCP_SEQ_STATE_OPENREQ
:
2243 case TCP_SEQ_STATE_LISTENING
:
2244 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2246 st
->state
= TCP_SEQ_STATE_LISTENING
;
2247 rc
= listening_get_next(seq
, NULL
);
2248 while (offset
-- && rc
)
2249 rc
= listening_get_next(seq
, rc
);
2254 case TCP_SEQ_STATE_ESTABLISHED
:
2255 case TCP_SEQ_STATE_TIME_WAIT
:
2256 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2257 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2259 rc
= established_get_first(seq
);
2260 while (offset
-- && rc
)
2261 rc
= established_get_next(seq
, rc
);
2269 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2271 struct tcp_iter_state
*st
= seq
->private;
2274 if (*pos
&& *pos
== st
->last_pos
) {
2275 rc
= tcp_seek_last_pos(seq
);
2280 st
->state
= TCP_SEQ_STATE_LISTENING
;
2284 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2287 st
->last_pos
= *pos
;
2291 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2293 struct tcp_iter_state
*st
= seq
->private;
2296 if (v
== SEQ_START_TOKEN
) {
2297 rc
= tcp_get_idx(seq
, 0);
2301 switch (st
->state
) {
2302 case TCP_SEQ_STATE_OPENREQ
:
2303 case TCP_SEQ_STATE_LISTENING
:
2304 rc
= listening_get_next(seq
, v
);
2306 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2309 rc
= established_get_first(seq
);
2312 case TCP_SEQ_STATE_ESTABLISHED
:
2313 case TCP_SEQ_STATE_TIME_WAIT
:
2314 rc
= established_get_next(seq
, v
);
2319 st
->last_pos
= *pos
;
2323 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2325 struct tcp_iter_state
*st
= seq
->private;
2327 switch (st
->state
) {
2328 case TCP_SEQ_STATE_OPENREQ
:
2330 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2331 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2333 case TCP_SEQ_STATE_LISTENING
:
2334 if (v
!= SEQ_START_TOKEN
)
2335 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2337 case TCP_SEQ_STATE_TIME_WAIT
:
2338 case TCP_SEQ_STATE_ESTABLISHED
:
2340 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2345 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2347 struct tcp_seq_afinfo
*afinfo
= PDE(inode
)->data
;
2348 struct tcp_iter_state
*s
;
2351 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2352 sizeof(struct tcp_iter_state
));
2356 s
= ((struct seq_file
*)file
->private_data
)->private;
2357 s
->family
= afinfo
->family
;
2361 EXPORT_SYMBOL(tcp_seq_open
);
2363 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2366 struct proc_dir_entry
*p
;
2368 afinfo
->seq_ops
.start
= tcp_seq_start
;
2369 afinfo
->seq_ops
.next
= tcp_seq_next
;
2370 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2372 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2373 afinfo
->seq_fops
, afinfo
);
2378 EXPORT_SYMBOL(tcp_proc_register
);
2380 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2382 proc_net_remove(net
, afinfo
->name
);
2384 EXPORT_SYMBOL(tcp_proc_unregister
);
2386 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2387 struct seq_file
*f
, int i
, int uid
, int *len
)
2389 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2390 int ttd
= req
->expires
- jiffies
;
2392 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2393 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2396 ntohs(inet_sk(sk
)->inet_sport
),
2398 ntohs(ireq
->rmt_port
),
2400 0, 0, /* could print option size, but that is af dependent. */
2401 1, /* timers active (only the expire timer) */
2402 jiffies_to_clock_t(ttd
),
2405 0, /* non standard timer */
2406 0, /* open_requests have no inode */
2407 atomic_read(&sk
->sk_refcnt
),
2412 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
, int *len
)
2415 unsigned long timer_expires
;
2416 const struct tcp_sock
*tp
= tcp_sk(sk
);
2417 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2418 const struct inet_sock
*inet
= inet_sk(sk
);
2419 __be32 dest
= inet
->inet_daddr
;
2420 __be32 src
= inet
->inet_rcv_saddr
;
2421 __u16 destp
= ntohs(inet
->inet_dport
);
2422 __u16 srcp
= ntohs(inet
->inet_sport
);
2425 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
) {
2427 timer_expires
= icsk
->icsk_timeout
;
2428 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2430 timer_expires
= icsk
->icsk_timeout
;
2431 } else if (timer_pending(&sk
->sk_timer
)) {
2433 timer_expires
= sk
->sk_timer
.expires
;
2436 timer_expires
= jiffies
;
2439 if (sk
->sk_state
== TCP_LISTEN
)
2440 rx_queue
= sk
->sk_ack_backlog
;
2443 * because we dont lock socket, we might find a transient negative value
2445 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2447 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2448 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2449 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2450 tp
->write_seq
- tp
->snd_una
,
2453 jiffies_to_clock_t(timer_expires
- jiffies
),
2454 icsk
->icsk_retransmits
,
2456 icsk
->icsk_probes_out
,
2458 atomic_read(&sk
->sk_refcnt
), sk
,
2459 jiffies_to_clock_t(icsk
->icsk_rto
),
2460 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2461 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2463 tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
,
2467 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2468 struct seq_file
*f
, int i
, int *len
)
2472 int ttd
= tw
->tw_ttd
- jiffies
;
2477 dest
= tw
->tw_daddr
;
2478 src
= tw
->tw_rcv_saddr
;
2479 destp
= ntohs(tw
->tw_dport
);
2480 srcp
= ntohs(tw
->tw_sport
);
2482 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2483 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2484 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2485 3, jiffies_to_clock_t(ttd
), 0, 0, 0, 0,
2486 atomic_read(&tw
->tw_refcnt
), tw
, len
);
2491 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2493 struct tcp_iter_state
*st
;
2496 if (v
== SEQ_START_TOKEN
) {
2497 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2498 " sl local_address rem_address st tx_queue "
2499 "rx_queue tr tm->when retrnsmt uid timeout "
2505 switch (st
->state
) {
2506 case TCP_SEQ_STATE_LISTENING
:
2507 case TCP_SEQ_STATE_ESTABLISHED
:
2508 get_tcp4_sock(v
, seq
, st
->num
, &len
);
2510 case TCP_SEQ_STATE_OPENREQ
:
2511 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
, &len
);
2513 case TCP_SEQ_STATE_TIME_WAIT
:
2514 get_timewait4_sock(v
, seq
, st
->num
, &len
);
2517 seq_printf(seq
, "%*s\n", TMPSZ
- 1 - len
, "");
2522 static const struct file_operations tcp_afinfo_seq_fops
= {
2523 .owner
= THIS_MODULE
,
2524 .open
= tcp_seq_open
,
2526 .llseek
= seq_lseek
,
2527 .release
= seq_release_net
2530 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2533 .seq_fops
= &tcp_afinfo_seq_fops
,
2535 .show
= tcp4_seq_show
,
2539 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2541 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2544 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2546 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2549 static struct pernet_operations tcp4_net_ops
= {
2550 .init
= tcp4_proc_init_net
,
2551 .exit
= tcp4_proc_exit_net
,
2554 int __init
tcp4_proc_init(void)
2556 return register_pernet_subsys(&tcp4_net_ops
);
2559 void tcp4_proc_exit(void)
2561 unregister_pernet_subsys(&tcp4_net_ops
);
2563 #endif /* CONFIG_PROC_FS */
2565 struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2567 const struct iphdr
*iph
= skb_gro_network_header(skb
);
2569 switch (skb
->ip_summed
) {
2570 case CHECKSUM_COMPLETE
:
2571 if (!tcp_v4_check(skb_gro_len(skb
), iph
->saddr
, iph
->daddr
,
2573 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2579 NAPI_GRO_CB(skb
)->flush
= 1;
2583 return tcp_gro_receive(head
, skb
);
2586 int tcp4_gro_complete(struct sk_buff
*skb
)
2588 const struct iphdr
*iph
= ip_hdr(skb
);
2589 struct tcphdr
*th
= tcp_hdr(skb
);
2591 th
->check
= ~tcp_v4_check(skb
->len
- skb_transport_offset(skb
),
2592 iph
->saddr
, iph
->daddr
, 0);
2593 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
2595 return tcp_gro_complete(skb
);
2598 struct proto tcp_prot
= {
2600 .owner
= THIS_MODULE
,
2602 .connect
= tcp_v4_connect
,
2603 .disconnect
= tcp_disconnect
,
2604 .accept
= inet_csk_accept
,
2606 .init
= tcp_v4_init_sock
,
2607 .destroy
= tcp_v4_destroy_sock
,
2608 .shutdown
= tcp_shutdown
,
2609 .setsockopt
= tcp_setsockopt
,
2610 .getsockopt
= tcp_getsockopt
,
2611 .recvmsg
= tcp_recvmsg
,
2612 .sendmsg
= tcp_sendmsg
,
2613 .sendpage
= tcp_sendpage
,
2614 .backlog_rcv
= tcp_v4_do_rcv
,
2615 .release_cb
= tcp_release_cb
,
2616 .mtu_reduced
= tcp_v4_mtu_reduced
,
2618 .unhash
= inet_unhash
,
2619 .get_port
= inet_csk_get_port
,
2620 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2621 .sockets_allocated
= &tcp_sockets_allocated
,
2622 .orphan_count
= &tcp_orphan_count
,
2623 .memory_allocated
= &tcp_memory_allocated
,
2624 .memory_pressure
= &tcp_memory_pressure
,
2625 .sysctl_wmem
= sysctl_tcp_wmem
,
2626 .sysctl_rmem
= sysctl_tcp_rmem
,
2627 .max_header
= MAX_TCP_HEADER
,
2628 .obj_size
= sizeof(struct tcp_sock
),
2629 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2630 .twsk_prot
= &tcp_timewait_sock_ops
,
2631 .rsk_prot
= &tcp_request_sock_ops
,
2632 .h
.hashinfo
= &tcp_hashinfo
,
2633 .no_autobind
= true,
2634 #ifdef CONFIG_COMPAT
2635 .compat_setsockopt
= compat_tcp_setsockopt
,
2636 .compat_getsockopt
= compat_tcp_getsockopt
,
2638 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2639 .init_cgroup
= tcp_init_cgroup
,
2640 .destroy_cgroup
= tcp_destroy_cgroup
,
2641 .proto_cgroup
= tcp_proto_cgroup
,
2644 EXPORT_SYMBOL(tcp_prot
);
2646 static int __net_init
tcp_sk_init(struct net
*net
)
2651 static void __net_exit
tcp_sk_exit(struct net
*net
)
2655 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2657 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2660 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2661 .init
= tcp_sk_init
,
2662 .exit
= tcp_sk_exit
,
2663 .exit_batch
= tcp_sk_exit_batch
,
2666 void __init
tcp_v4_init(void)
2668 inet_hashinfo_init(&tcp_hashinfo
);
2669 if (register_pernet_subsys(&tcp_sk_ops
))
2670 panic("Failed to create the TCP control socket.\n");