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 /* This will initiate an outgoing connection. */
142 int tcp_v4_connect(struct sock
*sk
, struct sockaddr
*uaddr
, int addr_len
)
144 struct sockaddr_in
*usin
= (struct sockaddr_in
*)uaddr
;
145 struct inet_sock
*inet
= inet_sk(sk
);
146 struct tcp_sock
*tp
= tcp_sk(sk
);
147 __be16 orig_sport
, orig_dport
;
148 __be32 daddr
, nexthop
;
152 struct ip_options_rcu
*inet_opt
;
154 if (addr_len
< sizeof(struct sockaddr_in
))
157 if (usin
->sin_family
!= AF_INET
)
158 return -EAFNOSUPPORT
;
160 nexthop
= daddr
= usin
->sin_addr
.s_addr
;
161 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
162 sock_owned_by_user(sk
));
163 if (inet_opt
&& inet_opt
->opt
.srr
) {
166 nexthop
= inet_opt
->opt
.faddr
;
169 orig_sport
= inet
->inet_sport
;
170 orig_dport
= usin
->sin_port
;
171 fl4
= &inet
->cork
.fl
.u
.ip4
;
172 rt
= ip_route_connect(fl4
, nexthop
, inet
->inet_saddr
,
173 RT_CONN_FLAGS(sk
), sk
->sk_bound_dev_if
,
175 orig_sport
, orig_dport
, sk
, true);
178 if (err
== -ENETUNREACH
)
179 IP_INC_STATS(sock_net(sk
), IPSTATS_MIB_OUTNOROUTES
);
183 if (rt
->rt_flags
& (RTCF_MULTICAST
| RTCF_BROADCAST
)) {
188 if (!inet_opt
|| !inet_opt
->opt
.srr
)
191 if (!inet
->inet_saddr
)
192 inet
->inet_saddr
= fl4
->saddr
;
193 inet
->inet_rcv_saddr
= inet
->inet_saddr
;
195 if (tp
->rx_opt
.ts_recent_stamp
&& inet
->inet_daddr
!= daddr
) {
196 /* Reset inherited state */
197 tp
->rx_opt
.ts_recent
= 0;
198 tp
->rx_opt
.ts_recent_stamp
= 0;
199 if (likely(!tp
->repair
))
203 if (tcp_death_row
.sysctl_tw_recycle
&&
204 !tp
->rx_opt
.ts_recent_stamp
&& fl4
->daddr
== daddr
)
205 tcp_fetch_timewait_stamp(sk
, &rt
->dst
);
207 inet
->inet_dport
= usin
->sin_port
;
208 inet
->inet_daddr
= daddr
;
210 inet_csk(sk
)->icsk_ext_hdr_len
= 0;
212 inet_csk(sk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
214 tp
->rx_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
216 /* Socket identity is still unknown (sport may be zero).
217 * However we set state to SYN-SENT and not releasing socket
218 * lock select source port, enter ourselves into the hash tables and
219 * complete initialization after this.
221 tcp_set_state(sk
, TCP_SYN_SENT
);
222 err
= inet_hash_connect(&tcp_death_row
, sk
);
226 rt
= ip_route_newports(fl4
, rt
, orig_sport
, orig_dport
,
227 inet
->inet_sport
, inet
->inet_dport
, sk
);
233 /* OK, now commit destination to socket. */
234 sk
->sk_gso_type
= SKB_GSO_TCPV4
;
235 sk_setup_caps(sk
, &rt
->dst
);
237 if (!tp
->write_seq
&& likely(!tp
->repair
))
238 tp
->write_seq
= secure_tcp_sequence_number(inet
->inet_saddr
,
243 inet
->inet_id
= tp
->write_seq
^ jiffies
;
245 err
= tcp_connect(sk
);
255 * This unhashes the socket and releases the local port,
258 tcp_set_state(sk
, TCP_CLOSE
);
260 sk
->sk_route_caps
= 0;
261 inet
->inet_dport
= 0;
264 EXPORT_SYMBOL(tcp_v4_connect
);
267 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
268 * It can be called through tcp_release_cb() if socket was owned by user
269 * at the time tcp_v4_err() was called to handle ICMP message.
271 void tcp_v4_mtu_reduced(struct sock
*sk
)
273 struct dst_entry
*dst
;
274 struct inet_sock
*inet
= inet_sk(sk
);
275 u32 mtu
= tcp_sk(sk
)->mtu_info
;
277 dst
= inet_csk_update_pmtu(sk
, mtu
);
281 /* Something is about to be wrong... Remember soft error
282 * for the case, if this connection will not able to recover.
284 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
285 sk
->sk_err_soft
= EMSGSIZE
;
289 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
290 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
291 tcp_sync_mss(sk
, mtu
);
293 /* Resend the TCP packet because it's
294 * clear that the old packet has been
295 * dropped. This is the new "fast" path mtu
298 tcp_simple_retransmit(sk
);
299 } /* else let the usual retransmit timer handle it */
301 EXPORT_SYMBOL(tcp_v4_mtu_reduced
);
303 static void do_redirect(struct sk_buff
*skb
, struct sock
*sk
)
305 struct dst_entry
*dst
= __sk_dst_check(sk
, 0);
308 dst
->ops
->redirect(dst
, sk
, skb
);
312 * This routine is called by the ICMP module when it gets some
313 * sort of error condition. If err < 0 then the socket should
314 * be closed and the error returned to the user. If err > 0
315 * it's just the icmp type << 8 | icmp code. After adjustment
316 * header points to the first 8 bytes of the tcp header. We need
317 * to find the appropriate port.
319 * The locking strategy used here is very "optimistic". When
320 * someone else accesses the socket the ICMP is just dropped
321 * and for some paths there is no check at all.
322 * A more general error queue to queue errors for later handling
323 * is probably better.
327 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
329 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
330 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
331 struct inet_connection_sock
*icsk
;
333 struct inet_sock
*inet
;
334 const int type
= icmp_hdr(icmp_skb
)->type
;
335 const int code
= icmp_hdr(icmp_skb
)->code
;
338 struct request_sock
*req
;
342 struct net
*net
= dev_net(icmp_skb
->dev
);
344 if (icmp_skb
->len
< (iph
->ihl
<< 2) + 8) {
345 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
349 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
350 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
352 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
355 if (sk
->sk_state
== TCP_TIME_WAIT
) {
356 inet_twsk_put(inet_twsk(sk
));
361 /* If too many ICMPs get dropped on busy
362 * servers this needs to be solved differently.
363 * We do take care of PMTU discovery (RFC1191) special case :
364 * we can receive locally generated ICMP messages while socket is held.
366 if (sock_owned_by_user(sk
)) {
367 if (!(type
== ICMP_DEST_UNREACH
&& code
== ICMP_FRAG_NEEDED
))
368 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
370 if (sk
->sk_state
== TCP_CLOSE
)
373 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
374 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
380 req
= tp
->fastopen_rsk
;
381 seq
= ntohl(th
->seq
);
382 if (sk
->sk_state
!= TCP_LISTEN
&&
383 !between(seq
, tp
->snd_una
, tp
->snd_nxt
) &&
384 (req
== NULL
|| seq
!= tcp_rsk(req
)->snt_isn
)) {
385 /* For a Fast Open socket, allow seq to be snt_isn. */
386 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
392 if (!sock_owned_by_user(sk
))
393 do_redirect(icmp_skb
, sk
);
395 case ICMP_SOURCE_QUENCH
:
396 /* Just silently ignore these. */
398 case ICMP_PARAMETERPROB
:
401 case ICMP_DEST_UNREACH
:
402 if (code
> NR_ICMP_UNREACH
)
405 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
406 /* We are not interested in TCP_LISTEN and open_requests
407 * (SYN-ACKs send out by Linux are always <576bytes so
408 * they should go through unfragmented).
410 if (sk
->sk_state
== TCP_LISTEN
)
414 if (!sock_owned_by_user(sk
)) {
415 tcp_v4_mtu_reduced(sk
);
417 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED
, &tp
->tsq_flags
))
423 err
= icmp_err_convert
[code
].errno
;
424 /* check if icmp_skb allows revert of backoff
425 * (see draft-zimmermann-tcp-lcd) */
426 if (code
!= ICMP_NET_UNREACH
&& code
!= ICMP_HOST_UNREACH
)
428 if (seq
!= tp
->snd_una
|| !icsk
->icsk_retransmits
||
432 /* XXX (TFO) - revisit the following logic for TFO */
434 if (sock_owned_by_user(sk
))
437 icsk
->icsk_backoff
--;
438 inet_csk(sk
)->icsk_rto
= (tp
->srtt
? __tcp_set_rto(tp
) :
439 TCP_TIMEOUT_INIT
) << icsk
->icsk_backoff
;
442 skb
= tcp_write_queue_head(sk
);
445 remaining
= icsk
->icsk_rto
- min(icsk
->icsk_rto
,
446 tcp_time_stamp
- TCP_SKB_CB(skb
)->when
);
449 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
450 remaining
, TCP_RTO_MAX
);
452 /* RTO revert clocked out retransmission.
453 * Will retransmit now */
454 tcp_retransmit_timer(sk
);
458 case ICMP_TIME_EXCEEDED
:
465 /* XXX (TFO) - if it's a TFO socket and has been accepted, rather
466 * than following the TCP_SYN_RECV case and closing the socket,
467 * we ignore the ICMP error and keep trying like a fully established
468 * socket. Is this the right thing to do?
470 if (req
&& req
->sk
== NULL
)
473 switch (sk
->sk_state
) {
474 struct request_sock
*req
, **prev
;
476 if (sock_owned_by_user(sk
))
479 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
480 iph
->daddr
, iph
->saddr
);
484 /* ICMPs are not backlogged, hence we cannot get
485 an established socket here.
489 if (seq
!= tcp_rsk(req
)->snt_isn
) {
490 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
495 * Still in SYN_RECV, just remove it silently.
496 * There is no good way to pass the error to the newly
497 * created socket, and POSIX does not want network
498 * errors returned from accept().
500 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
501 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
505 case TCP_SYN_RECV
: /* Cannot happen.
506 It can f.e. if SYNs crossed,
509 if (!sock_owned_by_user(sk
)) {
512 sk
->sk_error_report(sk
);
516 sk
->sk_err_soft
= err
;
521 /* If we've already connected we will keep trying
522 * until we time out, or the user gives up.
524 * rfc1122 4.2.3.9 allows to consider as hard errors
525 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
526 * but it is obsoleted by pmtu discovery).
528 * Note, that in modern internet, where routing is unreliable
529 * and in each dark corner broken firewalls sit, sending random
530 * errors ordered by their masters even this two messages finally lose
531 * their original sense (even Linux sends invalid PORT_UNREACHs)
533 * Now we are in compliance with RFCs.
538 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
540 sk
->sk_error_report(sk
);
541 } else { /* Only an error on timeout */
542 sk
->sk_err_soft
= err
;
550 static void __tcp_v4_send_check(struct sk_buff
*skb
,
551 __be32 saddr
, __be32 daddr
)
553 struct tcphdr
*th
= tcp_hdr(skb
);
555 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
556 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
557 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
558 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
560 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
567 /* This routine computes an IPv4 TCP checksum. */
568 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
570 const struct inet_sock
*inet
= inet_sk(sk
);
572 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
574 EXPORT_SYMBOL(tcp_v4_send_check
);
576 int tcp_v4_gso_send_check(struct sk_buff
*skb
)
578 const struct iphdr
*iph
;
581 if (!pskb_may_pull(skb
, sizeof(*th
)))
588 skb
->ip_summed
= CHECKSUM_PARTIAL
;
589 __tcp_v4_send_check(skb
, iph
->saddr
, iph
->daddr
);
594 * This routine will send an RST to the other tcp.
596 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
598 * Answer: if a packet caused RST, it is not for a socket
599 * existing in our system, if it is matched to a socket,
600 * it is just duplicate segment or bug in other side's TCP.
601 * So that we build reply only basing on parameters
602 * arrived with segment.
603 * Exception: precedence violation. We do not implement it in any case.
606 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
608 const struct tcphdr
*th
= tcp_hdr(skb
);
611 #ifdef CONFIG_TCP_MD5SIG
612 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
615 struct ip_reply_arg arg
;
616 #ifdef CONFIG_TCP_MD5SIG
617 struct tcp_md5sig_key
*key
;
618 const __u8
*hash_location
= NULL
;
619 unsigned char newhash
[16];
621 struct sock
*sk1
= NULL
;
625 /* Never send a reset in response to a reset. */
629 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
632 /* Swap the send and the receive. */
633 memset(&rep
, 0, sizeof(rep
));
634 rep
.th
.dest
= th
->source
;
635 rep
.th
.source
= th
->dest
;
636 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
640 rep
.th
.seq
= th
->ack_seq
;
643 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
644 skb
->len
- (th
->doff
<< 2));
647 memset(&arg
, 0, sizeof(arg
));
648 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
649 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
651 #ifdef CONFIG_TCP_MD5SIG
652 hash_location
= tcp_parse_md5sig_option(th
);
653 if (!sk
&& hash_location
) {
655 * active side is lost. Try to find listening socket through
656 * source port, and then find md5 key through listening socket.
657 * we are not loose security here:
658 * Incoming packet is checked with md5 hash with finding key,
659 * no RST generated if md5 hash doesn't match.
661 sk1
= __inet_lookup_listener(dev_net(skb_dst(skb
)->dev
),
662 &tcp_hashinfo
, ip_hdr(skb
)->saddr
,
663 th
->source
, ip_hdr(skb
)->daddr
,
664 ntohs(th
->source
), inet_iif(skb
));
665 /* don't send rst if it can't find key */
669 key
= tcp_md5_do_lookup(sk1
, (union tcp_md5_addr
*)
670 &ip_hdr(skb
)->saddr
, AF_INET
);
674 genhash
= tcp_v4_md5_hash_skb(newhash
, key
, NULL
, NULL
, skb
);
675 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
678 key
= sk
? tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)
684 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
686 (TCPOPT_MD5SIG
<< 8) |
688 /* Update length and the length the header thinks exists */
689 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
690 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
692 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
693 key
, ip_hdr(skb
)->saddr
,
694 ip_hdr(skb
)->daddr
, &rep
.th
);
697 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
698 ip_hdr(skb
)->saddr
, /* XXX */
699 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
700 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
701 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
702 /* When socket is gone, all binding information is lost.
703 * routing might fail in this case. No choice here, if we choose to force
704 * input interface, we will misroute in case of asymmetric route.
707 arg
.bound_dev_if
= sk
->sk_bound_dev_if
;
709 net
= dev_net(skb_dst(skb
)->dev
);
710 arg
.tos
= ip_hdr(skb
)->tos
;
711 ip_send_unicast_reply(*this_cpu_ptr(net
->ipv4
.tcp_sk
),
712 skb
, ip_hdr(skb
)->saddr
,
713 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
715 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
716 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
718 #ifdef CONFIG_TCP_MD5SIG
727 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
728 outside socket context is ugly, certainly. What can I do?
731 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
732 u32 win
, u32 tsval
, u32 tsecr
, int oif
,
733 struct tcp_md5sig_key
*key
,
734 int reply_flags
, u8 tos
)
736 const struct tcphdr
*th
= tcp_hdr(skb
);
739 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
740 #ifdef CONFIG_TCP_MD5SIG
741 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
745 struct ip_reply_arg arg
;
746 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
748 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
749 memset(&arg
, 0, sizeof(arg
));
751 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
752 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
754 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
755 (TCPOPT_TIMESTAMP
<< 8) |
757 rep
.opt
[1] = htonl(tsval
);
758 rep
.opt
[2] = htonl(tsecr
);
759 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
762 /* Swap the send and the receive. */
763 rep
.th
.dest
= th
->source
;
764 rep
.th
.source
= th
->dest
;
765 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
766 rep
.th
.seq
= htonl(seq
);
767 rep
.th
.ack_seq
= htonl(ack
);
769 rep
.th
.window
= htons(win
);
771 #ifdef CONFIG_TCP_MD5SIG
773 int offset
= (tsecr
) ? 3 : 0;
775 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
777 (TCPOPT_MD5SIG
<< 8) |
779 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
780 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
782 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
783 key
, ip_hdr(skb
)->saddr
,
784 ip_hdr(skb
)->daddr
, &rep
.th
);
787 arg
.flags
= reply_flags
;
788 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
789 ip_hdr(skb
)->saddr
, /* XXX */
790 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
791 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
793 arg
.bound_dev_if
= oif
;
795 ip_send_unicast_reply(*this_cpu_ptr(net
->ipv4
.tcp_sk
),
796 skb
, ip_hdr(skb
)->saddr
,
797 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
799 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
802 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
804 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
805 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
807 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
808 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
809 tcp_time_stamp
+ tcptw
->tw_ts_offset
,
812 tcp_twsk_md5_key(tcptw
),
813 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
820 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
821 struct request_sock
*req
)
823 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
824 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
826 tcp_v4_send_ack(skb
, (sk
->sk_state
== TCP_LISTEN
) ?
827 tcp_rsk(req
)->snt_isn
+ 1 : tcp_sk(sk
)->snd_nxt
,
828 tcp_rsk(req
)->rcv_nxt
,
829 req
->rcv_wnd
>> inet_rsk(req
)->rcv_wscale
,
833 tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&ip_hdr(skb
)->daddr
,
835 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
840 * Send a SYN-ACK after having received a SYN.
841 * This still operates on a request_sock only, not on a big
844 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
845 struct request_sock
*req
,
849 const struct inet_request_sock
*ireq
= inet_rsk(req
);
852 struct sk_buff
* skb
;
854 /* First, grab a route. */
855 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
858 skb
= tcp_make_synack(sk
, dst
, req
, NULL
);
861 __tcp_v4_send_check(skb
, ireq
->loc_addr
, ireq
->rmt_addr
);
863 skb_set_queue_mapping(skb
, queue_mapping
);
864 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
867 err
= net_xmit_eval(err
);
868 if (!tcp_rsk(req
)->snt_synack
&& !err
)
869 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
875 static int tcp_v4_rtx_synack(struct sock
*sk
, struct request_sock
*req
)
877 int res
= tcp_v4_send_synack(sk
, NULL
, req
, 0, false);
880 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
885 * IPv4 request_sock destructor.
887 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
889 kfree(inet_rsk(req
)->opt
);
893 * Return true if a syncookie should be sent
895 bool tcp_syn_flood_action(struct sock
*sk
,
896 const struct sk_buff
*skb
,
899 const char *msg
= "Dropping request";
900 bool want_cookie
= false;
901 struct listen_sock
*lopt
;
905 #ifdef CONFIG_SYN_COOKIES
906 if (sysctl_tcp_syncookies
) {
907 msg
= "Sending cookies";
909 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
912 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
914 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
915 if (!lopt
->synflood_warned
) {
916 lopt
->synflood_warned
= 1;
917 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
918 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
922 EXPORT_SYMBOL(tcp_syn_flood_action
);
925 * Save and compile IPv4 options into the request_sock if needed.
927 static struct ip_options_rcu
*tcp_v4_save_options(struct sk_buff
*skb
)
929 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
930 struct ip_options_rcu
*dopt
= NULL
;
932 if (opt
&& opt
->optlen
) {
933 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
935 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
937 if (ip_options_echo(&dopt
->opt
, skb
)) {
946 #ifdef CONFIG_TCP_MD5SIG
948 * RFC2385 MD5 checksumming requires a mapping of
949 * IP address->MD5 Key.
950 * We need to maintain these in the sk structure.
953 /* Find the Key structure for an address. */
954 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
955 const union tcp_md5_addr
*addr
,
958 struct tcp_sock
*tp
= tcp_sk(sk
);
959 struct tcp_md5sig_key
*key
;
960 unsigned int size
= sizeof(struct in_addr
);
961 struct tcp_md5sig_info
*md5sig
;
963 /* caller either holds rcu_read_lock() or socket lock */
964 md5sig
= rcu_dereference_check(tp
->md5sig_info
,
965 sock_owned_by_user(sk
) ||
966 lockdep_is_held(&sk
->sk_lock
.slock
));
969 #if IS_ENABLED(CONFIG_IPV6)
970 if (family
== AF_INET6
)
971 size
= sizeof(struct in6_addr
);
973 hlist_for_each_entry_rcu(key
, &md5sig
->head
, node
) {
974 if (key
->family
!= family
)
976 if (!memcmp(&key
->addr
, addr
, size
))
981 EXPORT_SYMBOL(tcp_md5_do_lookup
);
983 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
984 struct sock
*addr_sk
)
986 union tcp_md5_addr
*addr
;
988 addr
= (union tcp_md5_addr
*)&inet_sk(addr_sk
)->inet_daddr
;
989 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
991 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
993 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
994 struct request_sock
*req
)
996 union tcp_md5_addr
*addr
;
998 addr
= (union tcp_md5_addr
*)&inet_rsk(req
)->rmt_addr
;
999 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
1002 /* This can be called on a newly created socket, from other files */
1003 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1004 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
)
1006 /* Add Key to the list */
1007 struct tcp_md5sig_key
*key
;
1008 struct tcp_sock
*tp
= tcp_sk(sk
);
1009 struct tcp_md5sig_info
*md5sig
;
1011 key
= tcp_md5_do_lookup(sk
, addr
, family
);
1013 /* Pre-existing entry - just update that one. */
1014 memcpy(key
->key
, newkey
, newkeylen
);
1015 key
->keylen
= newkeylen
;
1019 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
1020 sock_owned_by_user(sk
) ||
1021 lockdep_is_held(&sk
->sk_lock
.slock
));
1023 md5sig
= kmalloc(sizeof(*md5sig
), gfp
);
1027 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
1028 INIT_HLIST_HEAD(&md5sig
->head
);
1029 rcu_assign_pointer(tp
->md5sig_info
, md5sig
);
1032 key
= sock_kmalloc(sk
, sizeof(*key
), gfp
);
1035 if (hlist_empty(&md5sig
->head
) && !tcp_alloc_md5sig_pool(sk
)) {
1036 sock_kfree_s(sk
, key
, sizeof(*key
));
1040 memcpy(key
->key
, newkey
, newkeylen
);
1041 key
->keylen
= newkeylen
;
1042 key
->family
= family
;
1043 memcpy(&key
->addr
, addr
,
1044 (family
== AF_INET6
) ? sizeof(struct in6_addr
) :
1045 sizeof(struct in_addr
));
1046 hlist_add_head_rcu(&key
->node
, &md5sig
->head
);
1049 EXPORT_SYMBOL(tcp_md5_do_add
);
1051 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
, int family
)
1053 struct tcp_sock
*tp
= tcp_sk(sk
);
1054 struct tcp_md5sig_key
*key
;
1055 struct tcp_md5sig_info
*md5sig
;
1057 key
= tcp_md5_do_lookup(sk
, addr
, family
);
1060 hlist_del_rcu(&key
->node
);
1061 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1062 kfree_rcu(key
, rcu
);
1063 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
1064 sock_owned_by_user(sk
));
1065 if (hlist_empty(&md5sig
->head
))
1066 tcp_free_md5sig_pool();
1069 EXPORT_SYMBOL(tcp_md5_do_del
);
1071 static void tcp_clear_md5_list(struct sock
*sk
)
1073 struct tcp_sock
*tp
= tcp_sk(sk
);
1074 struct tcp_md5sig_key
*key
;
1075 struct hlist_node
*n
;
1076 struct tcp_md5sig_info
*md5sig
;
1078 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
1080 if (!hlist_empty(&md5sig
->head
))
1081 tcp_free_md5sig_pool();
1082 hlist_for_each_entry_safe(key
, n
, &md5sig
->head
, node
) {
1083 hlist_del_rcu(&key
->node
);
1084 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1085 kfree_rcu(key
, rcu
);
1089 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1092 struct tcp_md5sig cmd
;
1093 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1095 if (optlen
< sizeof(cmd
))
1098 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1101 if (sin
->sin_family
!= AF_INET
)
1104 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
)
1105 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1108 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1111 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1112 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1116 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1117 __be32 daddr
, __be32 saddr
, int nbytes
)
1119 struct tcp4_pseudohdr
*bp
;
1120 struct scatterlist sg
;
1122 bp
= &hp
->md5_blk
.ip4
;
1125 * 1. the TCP pseudo-header (in the order: source IP address,
1126 * destination IP address, zero-padded protocol number, and
1132 bp
->protocol
= IPPROTO_TCP
;
1133 bp
->len
= cpu_to_be16(nbytes
);
1135 sg_init_one(&sg
, bp
, sizeof(*bp
));
1136 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1139 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1140 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1142 struct tcp_md5sig_pool
*hp
;
1143 struct hash_desc
*desc
;
1145 hp
= tcp_get_md5sig_pool();
1147 goto clear_hash_noput
;
1148 desc
= &hp
->md5_desc
;
1150 if (crypto_hash_init(desc
))
1152 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1154 if (tcp_md5_hash_header(hp
, th
))
1156 if (tcp_md5_hash_key(hp
, key
))
1158 if (crypto_hash_final(desc
, md5_hash
))
1161 tcp_put_md5sig_pool();
1165 tcp_put_md5sig_pool();
1167 memset(md5_hash
, 0, 16);
1171 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1172 const struct sock
*sk
, const struct request_sock
*req
,
1173 const struct sk_buff
*skb
)
1175 struct tcp_md5sig_pool
*hp
;
1176 struct hash_desc
*desc
;
1177 const struct tcphdr
*th
= tcp_hdr(skb
);
1178 __be32 saddr
, daddr
;
1181 saddr
= inet_sk(sk
)->inet_saddr
;
1182 daddr
= inet_sk(sk
)->inet_daddr
;
1184 saddr
= inet_rsk(req
)->loc_addr
;
1185 daddr
= inet_rsk(req
)->rmt_addr
;
1187 const struct iphdr
*iph
= ip_hdr(skb
);
1192 hp
= tcp_get_md5sig_pool();
1194 goto clear_hash_noput
;
1195 desc
= &hp
->md5_desc
;
1197 if (crypto_hash_init(desc
))
1200 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1202 if (tcp_md5_hash_header(hp
, th
))
1204 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1206 if (tcp_md5_hash_key(hp
, key
))
1208 if (crypto_hash_final(desc
, md5_hash
))
1211 tcp_put_md5sig_pool();
1215 tcp_put_md5sig_pool();
1217 memset(md5_hash
, 0, 16);
1220 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1222 static bool tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1225 * This gets called for each TCP segment that arrives
1226 * so we want to be efficient.
1227 * We have 3 drop cases:
1228 * o No MD5 hash and one expected.
1229 * o MD5 hash and we're not expecting one.
1230 * o MD5 hash and its wrong.
1232 const __u8
*hash_location
= NULL
;
1233 struct tcp_md5sig_key
*hash_expected
;
1234 const struct iphdr
*iph
= ip_hdr(skb
);
1235 const struct tcphdr
*th
= tcp_hdr(skb
);
1237 unsigned char newhash
[16];
1239 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1241 hash_location
= tcp_parse_md5sig_option(th
);
1243 /* We've parsed the options - do we have a hash? */
1244 if (!hash_expected
&& !hash_location
)
1247 if (hash_expected
&& !hash_location
) {
1248 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1252 if (!hash_expected
&& hash_location
) {
1253 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1257 /* Okay, so this is hash_expected and hash_location -
1258 * so we need to calculate the checksum.
1260 genhash
= tcp_v4_md5_hash_skb(newhash
,
1264 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1265 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1266 &iph
->saddr
, ntohs(th
->source
),
1267 &iph
->daddr
, ntohs(th
->dest
),
1268 genhash
? " tcp_v4_calc_md5_hash failed"
1277 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1279 .obj_size
= sizeof(struct tcp_request_sock
),
1280 .rtx_syn_ack
= tcp_v4_rtx_synack
,
1281 .send_ack
= tcp_v4_reqsk_send_ack
,
1282 .destructor
= tcp_v4_reqsk_destructor
,
1283 .send_reset
= tcp_v4_send_reset
,
1284 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1287 #ifdef CONFIG_TCP_MD5SIG
1288 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1289 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1290 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1294 static bool tcp_fastopen_check(struct sock
*sk
, struct sk_buff
*skb
,
1295 struct request_sock
*req
,
1296 struct tcp_fastopen_cookie
*foc
,
1297 struct tcp_fastopen_cookie
*valid_foc
)
1299 bool skip_cookie
= false;
1300 struct fastopen_queue
*fastopenq
;
1302 if (likely(!fastopen_cookie_present(foc
))) {
1303 /* See include/net/tcp.h for the meaning of these knobs */
1304 if ((sysctl_tcp_fastopen
& TFO_SERVER_ALWAYS
) ||
1305 ((sysctl_tcp_fastopen
& TFO_SERVER_COOKIE_NOT_REQD
) &&
1306 (TCP_SKB_CB(skb
)->end_seq
!= TCP_SKB_CB(skb
)->seq
+ 1)))
1307 skip_cookie
= true; /* no cookie to validate */
1311 fastopenq
= inet_csk(sk
)->icsk_accept_queue
.fastopenq
;
1312 /* A FO option is present; bump the counter. */
1313 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENPASSIVE
);
1315 /* Make sure the listener has enabled fastopen, and we don't
1316 * exceed the max # of pending TFO requests allowed before trying
1317 * to validating the cookie in order to avoid burning CPU cycles
1320 * XXX (TFO) - The implication of checking the max_qlen before
1321 * processing a cookie request is that clients can't differentiate
1322 * between qlen overflow causing Fast Open to be disabled
1323 * temporarily vs a server not supporting Fast Open at all.
1325 if ((sysctl_tcp_fastopen
& TFO_SERVER_ENABLE
) == 0 ||
1326 fastopenq
== NULL
|| fastopenq
->max_qlen
== 0)
1329 if (fastopenq
->qlen
>= fastopenq
->max_qlen
) {
1330 struct request_sock
*req1
;
1331 spin_lock(&fastopenq
->lock
);
1332 req1
= fastopenq
->rskq_rst_head
;
1333 if ((req1
== NULL
) || time_after(req1
->expires
, jiffies
)) {
1334 spin_unlock(&fastopenq
->lock
);
1335 NET_INC_STATS_BH(sock_net(sk
),
1336 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW
);
1337 /* Avoid bumping LINUX_MIB_TCPFASTOPENPASSIVEFAIL*/
1341 fastopenq
->rskq_rst_head
= req1
->dl_next
;
1343 spin_unlock(&fastopenq
->lock
);
1347 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1350 if (foc
->len
== TCP_FASTOPEN_COOKIE_SIZE
) {
1351 if ((sysctl_tcp_fastopen
& TFO_SERVER_COOKIE_NOT_CHKED
) == 0) {
1352 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
, valid_foc
);
1353 if ((valid_foc
->len
!= TCP_FASTOPEN_COOKIE_SIZE
) ||
1354 memcmp(&foc
->val
[0], &valid_foc
->val
[0],
1355 TCP_FASTOPEN_COOKIE_SIZE
) != 0)
1357 valid_foc
->len
= -1;
1359 /* Acknowledge the data received from the peer. */
1360 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1362 } else if (foc
->len
== 0) { /* Client requesting a cookie */
1363 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
, valid_foc
);
1364 NET_INC_STATS_BH(sock_net(sk
),
1365 LINUX_MIB_TCPFASTOPENCOOKIEREQD
);
1367 /* Client sent a cookie with wrong size. Treat it
1368 * the same as invalid and return a valid one.
1370 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
, valid_foc
);
1375 static int tcp_v4_conn_req_fastopen(struct sock
*sk
,
1376 struct sk_buff
*skb
,
1377 struct sk_buff
*skb_synack
,
1378 struct request_sock
*req
)
1380 struct tcp_sock
*tp
= tcp_sk(sk
);
1381 struct request_sock_queue
*queue
= &inet_csk(sk
)->icsk_accept_queue
;
1382 const struct inet_request_sock
*ireq
= inet_rsk(req
);
1386 req
->num_retrans
= 0;
1387 req
->num_timeout
= 0;
1390 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
);
1391 if (child
== NULL
) {
1392 NET_INC_STATS_BH(sock_net(sk
),
1393 LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
1394 kfree_skb(skb_synack
);
1397 err
= ip_build_and_send_pkt(skb_synack
, sk
, ireq
->loc_addr
,
1398 ireq
->rmt_addr
, ireq
->opt
);
1399 err
= net_xmit_eval(err
);
1401 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1402 /* XXX (TFO) - is it ok to ignore error and continue? */
1404 spin_lock(&queue
->fastopenq
->lock
);
1405 queue
->fastopenq
->qlen
++;
1406 spin_unlock(&queue
->fastopenq
->lock
);
1408 /* Initialize the child socket. Have to fix some values to take
1409 * into account the child is a Fast Open socket and is created
1410 * only out of the bits carried in the SYN packet.
1414 tp
->fastopen_rsk
= req
;
1415 /* Do a hold on the listner sk so that if the listener is being
1416 * closed, the child that has been accepted can live on and still
1417 * access listen_lock.
1420 tcp_rsk(req
)->listener
= sk
;
1422 /* RFC1323: The window in SYN & SYN/ACK segments is never
1423 * scaled. So correct it appropriately.
1425 tp
->snd_wnd
= ntohs(tcp_hdr(skb
)->window
);
1426 tp
->max_window
= tp
->snd_wnd
;
1428 /* Activate the retrans timer so that SYNACK can be retransmitted.
1429 * The request socket is not added to the SYN table of the parent
1430 * because it's been added to the accept queue directly.
1432 inet_csk_reset_xmit_timer(child
, ICSK_TIME_RETRANS
,
1433 TCP_TIMEOUT_INIT
, TCP_RTO_MAX
);
1435 /* Add the child socket directly into the accept queue */
1436 inet_csk_reqsk_queue_add(sk
, req
, child
);
1438 /* Now finish processing the fastopen child socket. */
1439 inet_csk(child
)->icsk_af_ops
->rebuild_header(child
);
1440 tcp_init_congestion_control(child
);
1441 tcp_mtup_init(child
);
1442 tcp_init_buffer_space(child
);
1443 tcp_init_metrics(child
);
1445 /* Queue the data carried in the SYN packet. We need to first
1446 * bump skb's refcnt because the caller will attempt to free it.
1448 * XXX (TFO) - we honor a zero-payload TFO request for now.
1449 * (Any reason not to?)
1451 if (TCP_SKB_CB(skb
)->end_seq
== TCP_SKB_CB(skb
)->seq
+ 1) {
1452 /* Don't queue the skb if there is no payload in SYN.
1453 * XXX (TFO) - How about SYN+FIN?
1455 tp
->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1459 __skb_pull(skb
, tcp_hdr(skb
)->doff
* 4);
1460 skb_set_owner_r(skb
, child
);
1461 __skb_queue_tail(&child
->sk_receive_queue
, skb
);
1462 tp
->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1463 tp
->syn_data_acked
= 1;
1465 sk
->sk_data_ready(sk
, 0);
1466 bh_unlock_sock(child
);
1468 WARN_ON(req
->sk
== NULL
);
1472 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1474 struct tcp_options_received tmp_opt
;
1475 struct request_sock
*req
;
1476 struct inet_request_sock
*ireq
;
1477 struct tcp_sock
*tp
= tcp_sk(sk
);
1478 struct dst_entry
*dst
= NULL
;
1479 __be32 saddr
= ip_hdr(skb
)->saddr
;
1480 __be32 daddr
= ip_hdr(skb
)->daddr
;
1481 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1482 bool want_cookie
= false;
1484 struct tcp_fastopen_cookie foc
= { .len
= -1 };
1485 struct tcp_fastopen_cookie valid_foc
= { .len
= -1 };
1486 struct sk_buff
*skb_synack
;
1489 /* Never answer to SYNs send to broadcast or multicast */
1490 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1493 /* TW buckets are converted to open requests without
1494 * limitations, they conserve resources and peer is
1495 * evidently real one.
1497 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
1498 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCP");
1503 /* Accept backlog is full. If we have already queued enough
1504 * of warm entries in syn queue, drop request. It is better than
1505 * clogging syn queue with openreqs with exponentially increasing
1508 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1) {
1509 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1513 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1517 #ifdef CONFIG_TCP_MD5SIG
1518 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1521 tcp_clear_options(&tmp_opt
);
1522 tmp_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
1523 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
1524 tcp_parse_options(skb
, &tmp_opt
, 0, want_cookie
? NULL
: &foc
);
1526 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1527 tcp_clear_options(&tmp_opt
);
1529 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1530 tcp_openreq_init(req
, &tmp_opt
, skb
);
1532 ireq
= inet_rsk(req
);
1533 ireq
->loc_addr
= daddr
;
1534 ireq
->rmt_addr
= saddr
;
1535 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1536 ireq
->opt
= tcp_v4_save_options(skb
);
1538 if (security_inet_conn_request(sk
, skb
, req
))
1541 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1542 TCP_ECN_create_request(req
, skb
, sock_net(sk
));
1545 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1546 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1548 /* VJ's idea. We save last timestamp seen
1549 * from the destination in peer table, when entering
1550 * state TIME-WAIT, and check against it before
1551 * accepting new connection request.
1553 * If "isn" is not zero, this request hit alive
1554 * timewait bucket, so that all the necessary checks
1555 * are made in the function processing timewait state.
1557 if (tmp_opt
.saw_tstamp
&&
1558 tcp_death_row
.sysctl_tw_recycle
&&
1559 (dst
= inet_csk_route_req(sk
, &fl4
, req
)) != NULL
&&
1560 fl4
.daddr
== saddr
) {
1561 if (!tcp_peer_is_proven(req
, dst
, true)) {
1562 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1563 goto drop_and_release
;
1566 /* Kill the following clause, if you dislike this way. */
1567 else if (!sysctl_tcp_syncookies
&&
1568 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1569 (sysctl_max_syn_backlog
>> 2)) &&
1570 !tcp_peer_is_proven(req
, dst
, false)) {
1571 /* Without syncookies last quarter of
1572 * backlog is filled with destinations,
1573 * proven to be alive.
1574 * It means that we continue to communicate
1575 * to destinations, already remembered
1576 * to the moment of synflood.
1578 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("drop open request from %pI4/%u\n"),
1579 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1580 goto drop_and_release
;
1583 isn
= tcp_v4_init_sequence(skb
);
1585 tcp_rsk(req
)->snt_isn
= isn
;
1588 dst
= inet_csk_route_req(sk
, &fl4
, req
);
1592 do_fastopen
= tcp_fastopen_check(sk
, skb
, req
, &foc
, &valid_foc
);
1594 /* We don't call tcp_v4_send_synack() directly because we need
1595 * to make sure a child socket can be created successfully before
1596 * sending back synack!
1598 * XXX (TFO) - Ideally one would simply call tcp_v4_send_synack()
1599 * (or better yet, call tcp_send_synack() in the child context
1600 * directly, but will have to fix bunch of other code first)
1601 * after syn_recv_sock() except one will need to first fix the
1602 * latter to remove its dependency on the current implementation
1603 * of tcp_v4_send_synack()->tcp_select_initial_window().
1605 skb_synack
= tcp_make_synack(sk
, dst
, req
,
1606 fastopen_cookie_present(&valid_foc
) ? &valid_foc
: NULL
);
1609 __tcp_v4_send_check(skb_synack
, ireq
->loc_addr
, ireq
->rmt_addr
);
1610 skb_set_queue_mapping(skb_synack
, skb_get_queue_mapping(skb
));
1614 if (likely(!do_fastopen
)) {
1616 err
= ip_build_and_send_pkt(skb_synack
, sk
, ireq
->loc_addr
,
1617 ireq
->rmt_addr
, ireq
->opt
);
1618 err
= net_xmit_eval(err
);
1619 if (err
|| want_cookie
)
1622 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1623 tcp_rsk(req
)->listener
= NULL
;
1624 /* Add the request_sock to the SYN table */
1625 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1626 if (fastopen_cookie_present(&foc
) && foc
.len
!= 0)
1627 NET_INC_STATS_BH(sock_net(sk
),
1628 LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
1629 } else if (tcp_v4_conn_req_fastopen(sk
, skb
, skb_synack
, req
))
1639 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1642 EXPORT_SYMBOL(tcp_v4_conn_request
);
1646 * The three way handshake has completed - we got a valid synack -
1647 * now create the new socket.
1649 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1650 struct request_sock
*req
,
1651 struct dst_entry
*dst
)
1653 struct inet_request_sock
*ireq
;
1654 struct inet_sock
*newinet
;
1655 struct tcp_sock
*newtp
;
1657 #ifdef CONFIG_TCP_MD5SIG
1658 struct tcp_md5sig_key
*key
;
1660 struct ip_options_rcu
*inet_opt
;
1662 if (sk_acceptq_is_full(sk
))
1665 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1669 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1670 inet_sk_rx_dst_set(newsk
, skb
);
1672 newtp
= tcp_sk(newsk
);
1673 newinet
= inet_sk(newsk
);
1674 ireq
= inet_rsk(req
);
1675 newinet
->inet_daddr
= ireq
->rmt_addr
;
1676 newinet
->inet_rcv_saddr
= ireq
->loc_addr
;
1677 newinet
->inet_saddr
= ireq
->loc_addr
;
1678 inet_opt
= ireq
->opt
;
1679 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1681 newinet
->mc_index
= inet_iif(skb
);
1682 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1683 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1684 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1686 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1687 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1690 dst
= inet_csk_route_child_sock(sk
, newsk
, req
);
1694 /* syncookie case : see end of cookie_v4_check() */
1696 sk_setup_caps(newsk
, dst
);
1698 tcp_mtup_init(newsk
);
1699 tcp_sync_mss(newsk
, dst_mtu(dst
));
1700 newtp
->advmss
= dst_metric_advmss(dst
);
1701 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1702 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1703 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1705 tcp_initialize_rcv_mss(newsk
);
1706 tcp_synack_rtt_meas(newsk
, req
);
1707 newtp
->total_retrans
= req
->num_retrans
;
1709 #ifdef CONFIG_TCP_MD5SIG
1710 /* Copy over the MD5 key from the original socket */
1711 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1715 * We're using one, so create a matching key
1716 * on the newsk structure. If we fail to get
1717 * memory, then we end up not copying the key
1720 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1721 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1722 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1726 if (__inet_inherit_port(sk
, newsk
) < 0)
1728 __inet_hash_nolisten(newsk
, NULL
);
1733 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1737 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1740 inet_csk_prepare_forced_close(newsk
);
1744 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1746 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1748 struct tcphdr
*th
= tcp_hdr(skb
);
1749 const struct iphdr
*iph
= ip_hdr(skb
);
1751 struct request_sock
**prev
;
1752 /* Find possible connection requests. */
1753 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1754 iph
->saddr
, iph
->daddr
);
1756 return tcp_check_req(sk
, skb
, req
, prev
, false);
1758 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1759 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1762 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1766 inet_twsk_put(inet_twsk(nsk
));
1770 #ifdef CONFIG_SYN_COOKIES
1772 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1777 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1779 const struct iphdr
*iph
= ip_hdr(skb
);
1781 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1782 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1783 iph
->daddr
, skb
->csum
)) {
1784 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1789 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1790 skb
->len
, IPPROTO_TCP
, 0);
1792 if (skb
->len
<= 76) {
1793 return __skb_checksum_complete(skb
);
1799 /* The socket must have it's spinlock held when we get
1802 * We have a potential double-lock case here, so even when
1803 * doing backlog processing we use the BH locking scheme.
1804 * This is because we cannot sleep with the original spinlock
1807 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1810 #ifdef CONFIG_TCP_MD5SIG
1812 * We really want to reject the packet as early as possible
1814 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1815 * o There is an MD5 option and we're not expecting one
1817 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1821 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1822 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1824 sock_rps_save_rxhash(sk
, skb
);
1826 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1827 dst
->ops
->check(dst
, 0) == NULL
) {
1829 sk
->sk_rx_dst
= NULL
;
1832 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1839 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1842 if (sk
->sk_state
== TCP_LISTEN
) {
1843 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1848 sock_rps_save_rxhash(nsk
, skb
);
1849 if (tcp_child_process(sk
, nsk
, skb
)) {
1856 sock_rps_save_rxhash(sk
, skb
);
1858 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1865 tcp_v4_send_reset(rsk
, skb
);
1868 /* Be careful here. If this function gets more complicated and
1869 * gcc suffers from register pressure on the x86, sk (in %ebx)
1870 * might be destroyed here. This current version compiles correctly,
1871 * but you have been warned.
1876 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1877 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1880 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1882 void tcp_v4_early_demux(struct sk_buff
*skb
)
1884 const struct iphdr
*iph
;
1885 const struct tcphdr
*th
;
1888 if (skb
->pkt_type
!= PACKET_HOST
)
1891 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1897 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1900 sk
= __inet_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1901 iph
->saddr
, th
->source
,
1902 iph
->daddr
, ntohs(th
->dest
),
1906 skb
->destructor
= sock_edemux
;
1907 if (sk
->sk_state
!= TCP_TIME_WAIT
) {
1908 struct dst_entry
*dst
= ACCESS_ONCE(sk
->sk_rx_dst
);
1911 dst
= dst_check(dst
, 0);
1913 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1914 skb_dst_set_noref(skb
, dst
);
1919 /* Packet is added to VJ-style prequeue for processing in process
1920 * context, if a reader task is waiting. Apparently, this exciting
1921 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1922 * failed somewhere. Latency? Burstiness? Well, at least now we will
1923 * see, why it failed. 8)8) --ANK
1926 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1928 struct tcp_sock
*tp
= tcp_sk(sk
);
1930 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
1933 if (skb
->len
<= tcp_hdrlen(skb
) &&
1934 skb_queue_len(&tp
->ucopy
.prequeue
) == 0)
1938 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1939 tp
->ucopy
.memory
+= skb
->truesize
;
1940 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1941 struct sk_buff
*skb1
;
1943 BUG_ON(sock_owned_by_user(sk
));
1945 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1946 sk_backlog_rcv(sk
, skb1
);
1947 NET_INC_STATS_BH(sock_net(sk
),
1948 LINUX_MIB_TCPPREQUEUEDROPPED
);
1951 tp
->ucopy
.memory
= 0;
1952 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1953 wake_up_interruptible_sync_poll(sk_sleep(sk
),
1954 POLLIN
| POLLRDNORM
| POLLRDBAND
);
1955 if (!inet_csk_ack_scheduled(sk
))
1956 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
1957 (3 * tcp_rto_min(sk
)) / 4,
1962 EXPORT_SYMBOL(tcp_prequeue
);
1964 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
)
1966 struct tcphdr
*th
= (struct tcphdr
*)skb
->data
;
1967 unsigned int eaten
= skb
->len
;
1970 err
= sk_filter_trim_cap(sk
, skb
, th
->doff
* 4);
1973 TCP_SKB_CB(skb
)->end_seq
-= eaten
;
1977 EXPORT_SYMBOL(tcp_filter
);
1983 int tcp_v4_rcv(struct sk_buff
*skb
)
1985 const struct iphdr
*iph
;
1986 const struct tcphdr
*th
;
1989 struct net
*net
= dev_net(skb
->dev
);
1991 if (skb
->pkt_type
!= PACKET_HOST
)
1994 /* Count it even if it's bad */
1995 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
1997 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
2002 if (th
->doff
< sizeof(struct tcphdr
) / 4)
2004 if (!pskb_may_pull(skb
, th
->doff
* 4))
2007 /* An explanation is required here, I think.
2008 * Packet length and doff are validated by header prediction,
2009 * provided case of th->doff==0 is eliminated.
2010 * So, we defer the checks. */
2011 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
2016 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
2017 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
2018 skb
->len
- th
->doff
* 4);
2019 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
2020 TCP_SKB_CB(skb
)->when
= 0;
2021 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
2022 TCP_SKB_CB(skb
)->sacked
= 0;
2024 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
2029 if (sk
->sk_state
== TCP_TIME_WAIT
)
2032 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
2033 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
2034 goto discard_and_relse
;
2037 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
2038 goto discard_and_relse
;
2041 if (tcp_filter(sk
, skb
))
2042 goto discard_and_relse
;
2043 th
= (const struct tcphdr
*)skb
->data
;
2048 bh_lock_sock_nested(sk
);
2050 if (!sock_owned_by_user(sk
)) {
2051 #ifdef CONFIG_NET_DMA
2052 struct tcp_sock
*tp
= tcp_sk(sk
);
2053 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
2054 tp
->ucopy
.dma_chan
= net_dma_find_channel();
2055 if (tp
->ucopy
.dma_chan
)
2056 ret
= tcp_v4_do_rcv(sk
, skb
);
2060 if (!tcp_prequeue(sk
, skb
))
2061 ret
= tcp_v4_do_rcv(sk
, skb
);
2063 } else if (unlikely(sk_add_backlog(sk
, skb
,
2064 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
2066 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
2067 goto discard_and_relse
;
2076 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
2079 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
2081 TCP_INC_STATS_BH(net
, TCP_MIB_CSUMERRORS
);
2083 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
2085 tcp_v4_send_reset(NULL
, skb
);
2089 /* Discard frame. */
2098 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
2099 inet_twsk_put(inet_twsk(sk
));
2103 if (skb
->len
< (th
->doff
<< 2)) {
2104 inet_twsk_put(inet_twsk(sk
));
2107 if (tcp_checksum_complete(skb
)) {
2108 inet_twsk_put(inet_twsk(sk
));
2111 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
2113 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
2115 iph
->saddr
, th
->source
,
2116 iph
->daddr
, th
->dest
,
2119 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
2120 inet_twsk_put(inet_twsk(sk
));
2124 /* Fall through to ACK */
2127 tcp_v4_timewait_ack(sk
, skb
);
2131 case TCP_TW_SUCCESS
:;
2136 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
2137 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
2138 .twsk_unique
= tcp_twsk_unique
,
2139 .twsk_destructor
= tcp_twsk_destructor
,
2142 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
2144 struct dst_entry
*dst
= skb_dst(skb
);
2147 sk
->sk_rx_dst
= dst
;
2148 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
2150 EXPORT_SYMBOL(inet_sk_rx_dst_set
);
2152 const struct inet_connection_sock_af_ops ipv4_specific
= {
2153 .queue_xmit
= ip_queue_xmit
,
2154 .send_check
= tcp_v4_send_check
,
2155 .rebuild_header
= inet_sk_rebuild_header
,
2156 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
2157 .conn_request
= tcp_v4_conn_request
,
2158 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
2159 .net_header_len
= sizeof(struct iphdr
),
2160 .setsockopt
= ip_setsockopt
,
2161 .getsockopt
= ip_getsockopt
,
2162 .addr2sockaddr
= inet_csk_addr2sockaddr
,
2163 .sockaddr_len
= sizeof(struct sockaddr_in
),
2164 .bind_conflict
= inet_csk_bind_conflict
,
2165 #ifdef CONFIG_COMPAT
2166 .compat_setsockopt
= compat_ip_setsockopt
,
2167 .compat_getsockopt
= compat_ip_getsockopt
,
2169 .mtu_reduced
= tcp_v4_mtu_reduced
,
2171 EXPORT_SYMBOL(ipv4_specific
);
2173 #ifdef CONFIG_TCP_MD5SIG
2174 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
2175 .md5_lookup
= tcp_v4_md5_lookup
,
2176 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
2177 .md5_parse
= tcp_v4_parse_md5_keys
,
2181 /* NOTE: A lot of things set to zero explicitly by call to
2182 * sk_alloc() so need not be done here.
2184 static int tcp_v4_init_sock(struct sock
*sk
)
2186 struct inet_connection_sock
*icsk
= inet_csk(sk
);
2190 icsk
->icsk_af_ops
= &ipv4_specific
;
2192 #ifdef CONFIG_TCP_MD5SIG
2193 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv4_specific
;
2199 void tcp_v4_destroy_sock(struct sock
*sk
)
2201 struct tcp_sock
*tp
= tcp_sk(sk
);
2203 tcp_clear_xmit_timers(sk
);
2205 tcp_cleanup_congestion_control(sk
);
2207 /* Cleanup up the write buffer. */
2208 tcp_write_queue_purge(sk
);
2210 /* Cleans up our, hopefully empty, out_of_order_queue. */
2211 __skb_queue_purge(&tp
->out_of_order_queue
);
2213 #ifdef CONFIG_TCP_MD5SIG
2214 /* Clean up the MD5 key list, if any */
2215 if (tp
->md5sig_info
) {
2216 tcp_clear_md5_list(sk
);
2217 kfree_rcu(tp
->md5sig_info
, rcu
);
2218 tp
->md5sig_info
= NULL
;
2222 #ifdef CONFIG_NET_DMA
2223 /* Cleans up our sk_async_wait_queue */
2224 __skb_queue_purge(&sk
->sk_async_wait_queue
);
2227 /* Clean prequeue, it must be empty really */
2228 __skb_queue_purge(&tp
->ucopy
.prequeue
);
2230 /* Clean up a referenced TCP bind bucket. */
2231 if (inet_csk(sk
)->icsk_bind_hash
)
2234 BUG_ON(tp
->fastopen_rsk
!= NULL
);
2236 /* If socket is aborted during connect operation */
2237 tcp_free_fastopen_req(tp
);
2239 sk_sockets_allocated_dec(sk
);
2240 sock_release_memcg(sk
);
2242 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
2244 #ifdef CONFIG_PROC_FS
2245 /* Proc filesystem TCP sock list dumping. */
2247 static inline struct inet_timewait_sock
*tw_head(struct hlist_nulls_head
*head
)
2249 return hlist_nulls_empty(head
) ? NULL
:
2250 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
2253 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
2255 return !is_a_nulls(tw
->tw_node
.next
) ?
2256 hlist_nulls_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
2260 * Get next listener socket follow cur. If cur is NULL, get first socket
2261 * starting from bucket given in st->bucket; when st->bucket is zero the
2262 * very first socket in the hash table is returned.
2264 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
2266 struct inet_connection_sock
*icsk
;
2267 struct hlist_nulls_node
*node
;
2268 struct sock
*sk
= cur
;
2269 struct inet_listen_hashbucket
*ilb
;
2270 struct tcp_iter_state
*st
= seq
->private;
2271 struct net
*net
= seq_file_net(seq
);
2274 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2275 spin_lock_bh(&ilb
->lock
);
2276 sk
= sk_nulls_head(&ilb
->head
);
2280 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2284 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
2285 struct request_sock
*req
= cur
;
2287 icsk
= inet_csk(st
->syn_wait_sk
);
2291 if (req
->rsk_ops
->family
== st
->family
) {
2297 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
2300 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
2302 sk
= sk_nulls_next(st
->syn_wait_sk
);
2303 st
->state
= TCP_SEQ_STATE_LISTENING
;
2304 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2306 icsk
= inet_csk(sk
);
2307 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2308 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
2310 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2311 sk
= sk_nulls_next(sk
);
2314 sk_nulls_for_each_from(sk
, node
) {
2315 if (!net_eq(sock_net(sk
), net
))
2317 if (sk
->sk_family
== st
->family
) {
2321 icsk
= inet_csk(sk
);
2322 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2323 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2325 st
->uid
= sock_i_uid(sk
);
2326 st
->syn_wait_sk
= sk
;
2327 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2331 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2333 spin_unlock_bh(&ilb
->lock
);
2335 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2336 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2337 spin_lock_bh(&ilb
->lock
);
2338 sk
= sk_nulls_head(&ilb
->head
);
2346 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2348 struct tcp_iter_state
*st
= seq
->private;
2353 rc
= listening_get_next(seq
, NULL
);
2355 while (rc
&& *pos
) {
2356 rc
= listening_get_next(seq
, rc
);
2362 static inline bool empty_bucket(struct tcp_iter_state
*st
)
2364 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
) &&
2365 hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2369 * Get first established socket starting from bucket given in st->bucket.
2370 * If st->bucket is zero, the very first socket in the hash is returned.
2372 static void *established_get_first(struct seq_file
*seq
)
2374 struct tcp_iter_state
*st
= seq
->private;
2375 struct net
*net
= seq_file_net(seq
);
2379 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2381 struct hlist_nulls_node
*node
;
2382 struct inet_timewait_sock
*tw
;
2383 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2385 /* Lockless fast path for the common case of empty buckets */
2386 if (empty_bucket(st
))
2390 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2391 if (sk
->sk_family
!= st
->family
||
2392 !net_eq(sock_net(sk
), net
)) {
2398 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2399 inet_twsk_for_each(tw
, node
,
2400 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
2401 if (tw
->tw_family
!= st
->family
||
2402 !net_eq(twsk_net(tw
), net
)) {
2408 spin_unlock_bh(lock
);
2409 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2415 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2417 struct sock
*sk
= cur
;
2418 struct inet_timewait_sock
*tw
;
2419 struct hlist_nulls_node
*node
;
2420 struct tcp_iter_state
*st
= seq
->private;
2421 struct net
*net
= seq_file_net(seq
);
2426 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2430 while (tw
&& (tw
->tw_family
!= st
->family
|| !net_eq(twsk_net(tw
), net
))) {
2437 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2438 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2440 /* Look for next non empty bucket */
2442 while (++st
->bucket
<= tcp_hashinfo
.ehash_mask
&&
2445 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2448 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2449 sk
= sk_nulls_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2451 sk
= sk_nulls_next(sk
);
2453 sk_nulls_for_each_from(sk
, node
) {
2454 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2458 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2459 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2467 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2469 struct tcp_iter_state
*st
= seq
->private;
2473 rc
= established_get_first(seq
);
2476 rc
= established_get_next(seq
, rc
);
2482 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2485 struct tcp_iter_state
*st
= seq
->private;
2487 st
->state
= TCP_SEQ_STATE_LISTENING
;
2488 rc
= listening_get_idx(seq
, &pos
);
2491 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2492 rc
= established_get_idx(seq
, pos
);
2498 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2500 struct tcp_iter_state
*st
= seq
->private;
2501 int offset
= st
->offset
;
2502 int orig_num
= st
->num
;
2505 switch (st
->state
) {
2506 case TCP_SEQ_STATE_OPENREQ
:
2507 case TCP_SEQ_STATE_LISTENING
:
2508 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2510 st
->state
= TCP_SEQ_STATE_LISTENING
;
2511 rc
= listening_get_next(seq
, NULL
);
2512 while (offset
-- && rc
)
2513 rc
= listening_get_next(seq
, rc
);
2518 case TCP_SEQ_STATE_ESTABLISHED
:
2519 case TCP_SEQ_STATE_TIME_WAIT
:
2520 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2521 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2523 rc
= established_get_first(seq
);
2524 while (offset
-- && rc
)
2525 rc
= established_get_next(seq
, rc
);
2533 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2535 struct tcp_iter_state
*st
= seq
->private;
2538 if (*pos
&& *pos
== st
->last_pos
) {
2539 rc
= tcp_seek_last_pos(seq
);
2544 st
->state
= TCP_SEQ_STATE_LISTENING
;
2548 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2551 st
->last_pos
= *pos
;
2555 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2557 struct tcp_iter_state
*st
= seq
->private;
2560 if (v
== SEQ_START_TOKEN
) {
2561 rc
= tcp_get_idx(seq
, 0);
2565 switch (st
->state
) {
2566 case TCP_SEQ_STATE_OPENREQ
:
2567 case TCP_SEQ_STATE_LISTENING
:
2568 rc
= listening_get_next(seq
, v
);
2570 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2573 rc
= established_get_first(seq
);
2576 case TCP_SEQ_STATE_ESTABLISHED
:
2577 case TCP_SEQ_STATE_TIME_WAIT
:
2578 rc
= established_get_next(seq
, v
);
2583 st
->last_pos
= *pos
;
2587 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2589 struct tcp_iter_state
*st
= seq
->private;
2591 switch (st
->state
) {
2592 case TCP_SEQ_STATE_OPENREQ
:
2594 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2595 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2597 case TCP_SEQ_STATE_LISTENING
:
2598 if (v
!= SEQ_START_TOKEN
)
2599 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2601 case TCP_SEQ_STATE_TIME_WAIT
:
2602 case TCP_SEQ_STATE_ESTABLISHED
:
2604 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2609 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2611 struct tcp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2612 struct tcp_iter_state
*s
;
2615 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2616 sizeof(struct tcp_iter_state
));
2620 s
= ((struct seq_file
*)file
->private_data
)->private;
2621 s
->family
= afinfo
->family
;
2625 EXPORT_SYMBOL(tcp_seq_open
);
2627 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2630 struct proc_dir_entry
*p
;
2632 afinfo
->seq_ops
.start
= tcp_seq_start
;
2633 afinfo
->seq_ops
.next
= tcp_seq_next
;
2634 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2636 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2637 afinfo
->seq_fops
, afinfo
);
2642 EXPORT_SYMBOL(tcp_proc_register
);
2644 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2646 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2648 EXPORT_SYMBOL(tcp_proc_unregister
);
2650 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2651 struct seq_file
*f
, int i
, kuid_t uid
, int *len
)
2653 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2654 long delta
= req
->expires
- jiffies
;
2656 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2657 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2660 ntohs(inet_sk(sk
)->inet_sport
),
2662 ntohs(ireq
->rmt_port
),
2664 0, 0, /* could print option size, but that is af dependent. */
2665 1, /* timers active (only the expire timer) */
2666 jiffies_delta_to_clock_t(delta
),
2668 from_kuid_munged(seq_user_ns(f
), uid
),
2669 0, /* non standard timer */
2670 0, /* open_requests have no inode */
2671 atomic_read(&sk
->sk_refcnt
),
2676 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
, int *len
)
2679 unsigned long timer_expires
;
2680 const struct tcp_sock
*tp
= tcp_sk(sk
);
2681 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2682 const struct inet_sock
*inet
= inet_sk(sk
);
2683 struct fastopen_queue
*fastopenq
= icsk
->icsk_accept_queue
.fastopenq
;
2684 __be32 dest
= inet
->inet_daddr
;
2685 __be32 src
= inet
->inet_rcv_saddr
;
2686 __u16 destp
= ntohs(inet
->inet_dport
);
2687 __u16 srcp
= ntohs(inet
->inet_sport
);
2690 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
2691 icsk
->icsk_pending
== ICSK_TIME_EARLY_RETRANS
||
2692 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
2694 timer_expires
= icsk
->icsk_timeout
;
2695 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2697 timer_expires
= icsk
->icsk_timeout
;
2698 } else if (timer_pending(&sk
->sk_timer
)) {
2700 timer_expires
= sk
->sk_timer
.expires
;
2703 timer_expires
= jiffies
;
2706 if (sk
->sk_state
== TCP_LISTEN
)
2707 rx_queue
= sk
->sk_ack_backlog
;
2710 * because we dont lock socket, we might find a transient negative value
2712 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2714 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2715 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2716 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2717 tp
->write_seq
- tp
->snd_una
,
2720 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
2721 icsk
->icsk_retransmits
,
2722 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sk
)),
2723 icsk
->icsk_probes_out
,
2725 atomic_read(&sk
->sk_refcnt
), sk
,
2726 jiffies_to_clock_t(icsk
->icsk_rto
),
2727 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2728 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2730 sk
->sk_state
== TCP_LISTEN
?
2731 (fastopenq
? fastopenq
->max_qlen
: 0) :
2732 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
),
2736 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2737 struct seq_file
*f
, int i
, int *len
)
2741 long delta
= tw
->tw_ttd
- jiffies
;
2743 dest
= tw
->tw_daddr
;
2744 src
= tw
->tw_rcv_saddr
;
2745 destp
= ntohs(tw
->tw_dport
);
2746 srcp
= ntohs(tw
->tw_sport
);
2748 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2749 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2750 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2751 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
2752 atomic_read(&tw
->tw_refcnt
), tw
, len
);
2757 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2759 struct tcp_iter_state
*st
;
2762 if (v
== SEQ_START_TOKEN
) {
2763 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2764 " sl local_address rem_address st tx_queue "
2765 "rx_queue tr tm->when retrnsmt uid timeout "
2771 switch (st
->state
) {
2772 case TCP_SEQ_STATE_LISTENING
:
2773 case TCP_SEQ_STATE_ESTABLISHED
:
2774 get_tcp4_sock(v
, seq
, st
->num
, &len
);
2776 case TCP_SEQ_STATE_OPENREQ
:
2777 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
, &len
);
2779 case TCP_SEQ_STATE_TIME_WAIT
:
2780 get_timewait4_sock(v
, seq
, st
->num
, &len
);
2783 seq_printf(seq
, "%*s\n", TMPSZ
- 1 - len
, "");
2788 static const struct file_operations tcp_afinfo_seq_fops
= {
2789 .owner
= THIS_MODULE
,
2790 .open
= tcp_seq_open
,
2792 .llseek
= seq_lseek
,
2793 .release
= seq_release_net
2796 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2799 .seq_fops
= &tcp_afinfo_seq_fops
,
2801 .show
= tcp4_seq_show
,
2805 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2807 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2810 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2812 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2815 static struct pernet_operations tcp4_net_ops
= {
2816 .init
= tcp4_proc_init_net
,
2817 .exit
= tcp4_proc_exit_net
,
2820 int __init
tcp4_proc_init(void)
2822 return register_pernet_subsys(&tcp4_net_ops
);
2825 void tcp4_proc_exit(void)
2827 unregister_pernet_subsys(&tcp4_net_ops
);
2829 #endif /* CONFIG_PROC_FS */
2831 struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2833 const struct iphdr
*iph
= skb_gro_network_header(skb
);
2837 switch (skb
->ip_summed
) {
2838 case CHECKSUM_COMPLETE
:
2839 if (!tcp_v4_check(skb_gro_len(skb
), iph
->saddr
, iph
->daddr
,
2841 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2845 NAPI_GRO_CB(skb
)->flush
= 1;
2849 wsum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
2850 skb_gro_len(skb
), IPPROTO_TCP
, 0);
2851 sum
= csum_fold(skb_checksum(skb
,
2852 skb_gro_offset(skb
),
2858 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2862 return tcp_gro_receive(head
, skb
);
2865 int tcp4_gro_complete(struct sk_buff
*skb
)
2867 const struct iphdr
*iph
= ip_hdr(skb
);
2868 struct tcphdr
*th
= tcp_hdr(skb
);
2870 th
->check
= ~tcp_v4_check(skb
->len
- skb_transport_offset(skb
),
2871 iph
->saddr
, iph
->daddr
, 0);
2872 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
2874 return tcp_gro_complete(skb
);
2877 struct proto tcp_prot
= {
2879 .owner
= THIS_MODULE
,
2881 .connect
= tcp_v4_connect
,
2882 .disconnect
= tcp_disconnect
,
2883 .accept
= inet_csk_accept
,
2885 .init
= tcp_v4_init_sock
,
2886 .destroy
= tcp_v4_destroy_sock
,
2887 .shutdown
= tcp_shutdown
,
2888 .setsockopt
= tcp_setsockopt
,
2889 .getsockopt
= tcp_getsockopt
,
2890 .recvmsg
= tcp_recvmsg
,
2891 .sendmsg
= tcp_sendmsg
,
2892 .sendpage
= tcp_sendpage
,
2893 .backlog_rcv
= tcp_v4_do_rcv
,
2894 .release_cb
= tcp_release_cb
,
2896 .unhash
= inet_unhash
,
2897 .get_port
= inet_csk_get_port
,
2898 .enter_memory_pressure
= tcp_enter_memory_pressure
,
2899 .sockets_allocated
= &tcp_sockets_allocated
,
2900 .orphan_count
= &tcp_orphan_count
,
2901 .memory_allocated
= &tcp_memory_allocated
,
2902 .memory_pressure
= &tcp_memory_pressure
,
2903 .sysctl_wmem
= sysctl_tcp_wmem
,
2904 .sysctl_rmem
= sysctl_tcp_rmem
,
2905 .max_header
= MAX_TCP_HEADER
,
2906 .obj_size
= sizeof(struct tcp_sock
),
2907 .slab_flags
= SLAB_DESTROY_BY_RCU
,
2908 .twsk_prot
= &tcp_timewait_sock_ops
,
2909 .rsk_prot
= &tcp_request_sock_ops
,
2910 .h
.hashinfo
= &tcp_hashinfo
,
2911 .no_autobind
= true,
2912 #ifdef CONFIG_COMPAT
2913 .compat_setsockopt
= compat_tcp_setsockopt
,
2914 .compat_getsockopt
= compat_tcp_getsockopt
,
2916 #ifdef CONFIG_MEMCG_KMEM
2917 .init_cgroup
= tcp_init_cgroup
,
2918 .destroy_cgroup
= tcp_destroy_cgroup
,
2919 .proto_cgroup
= tcp_proto_cgroup
,
2922 EXPORT_SYMBOL(tcp_prot
);
2924 static void __net_exit
tcp_sk_exit(struct net
*net
)
2928 for_each_possible_cpu(cpu
)
2929 inet_ctl_sock_destroy(*per_cpu_ptr(net
->ipv4
.tcp_sk
, cpu
));
2930 free_percpu(net
->ipv4
.tcp_sk
);
2933 static int __net_init
tcp_sk_init(struct net
*net
)
2937 net
->ipv4
.tcp_sk
= alloc_percpu(struct sock
*);
2938 if (!net
->ipv4
.tcp_sk
)
2941 for_each_possible_cpu(cpu
) {
2944 res
= inet_ctl_sock_create(&sk
, PF_INET
, SOCK_RAW
,
2948 *per_cpu_ptr(net
->ipv4
.tcp_sk
, cpu
) = sk
;
2950 net
->ipv4
.sysctl_tcp_ecn
= 2;
2959 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
2961 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
2964 static struct pernet_operations __net_initdata tcp_sk_ops
= {
2965 .init
= tcp_sk_init
,
2966 .exit
= tcp_sk_exit
,
2967 .exit_batch
= tcp_sk_exit_batch
,
2970 void __init
tcp_v4_init(void)
2972 inet_hashinfo_init(&tcp_hashinfo
);
2973 if (register_pernet_subsys(&tcp_sk_ops
))
2974 panic("Failed to create the TCP control socket.\n");