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
);
236 printk(KERN_INFO
"[socket_conn]IPV4 socket[%lu] sport:%u \n", SOCK_INODE(sk
->sk_socket
)->i_ino
, ntohs(inet
->inet_sport
));
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 inet_sock
*inet
= inet_sk(sk
);
274 struct dst_entry
*dst
;
277 if ((1 << sk
->sk_state
) & (TCPF_LISTEN
| TCPF_CLOSE
))
279 mtu
= tcp_sk(sk
)->mtu_info
;
280 dst
= inet_csk_update_pmtu(sk
, mtu
);
284 /* Something is about to be wrong... Remember soft error
285 * for the case, if this connection will not able to recover.
287 if (mtu
< dst_mtu(dst
) && ip_dont_fragment(sk
, dst
))
288 sk
->sk_err_soft
= EMSGSIZE
;
292 if (inet
->pmtudisc
!= IP_PMTUDISC_DONT
&&
293 inet_csk(sk
)->icsk_pmtu_cookie
> mtu
) {
294 tcp_sync_mss(sk
, mtu
);
296 /* Resend the TCP packet because it's
297 * clear that the old packet has been
298 * dropped. This is the new "fast" path mtu
301 tcp_simple_retransmit(sk
);
302 } /* else let the usual retransmit timer handle it */
304 EXPORT_SYMBOL(tcp_v4_mtu_reduced
);
306 static void do_redirect(struct sk_buff
*skb
, struct sock
*sk
)
308 struct dst_entry
*dst
= __sk_dst_check(sk
, 0);
311 dst
->ops
->redirect(dst
, sk
, skb
);
315 * This routine is called by the ICMP module when it gets some
316 * sort of error condition. If err < 0 then the socket should
317 * be closed and the error returned to the user. If err > 0
318 * it's just the icmp type << 8 | icmp code. After adjustment
319 * header points to the first 8 bytes of the tcp header. We need
320 * to find the appropriate port.
322 * The locking strategy used here is very "optimistic". When
323 * someone else accesses the socket the ICMP is just dropped
324 * and for some paths there is no check at all.
325 * A more general error queue to queue errors for later handling
326 * is probably better.
330 void tcp_v4_err(struct sk_buff
*icmp_skb
, u32 info
)
332 const struct iphdr
*iph
= (const struct iphdr
*)icmp_skb
->data
;
333 struct tcphdr
*th
= (struct tcphdr
*)(icmp_skb
->data
+ (iph
->ihl
<< 2));
334 struct inet_connection_sock
*icsk
;
336 struct inet_sock
*inet
;
337 const int type
= icmp_hdr(icmp_skb
)->type
;
338 const int code
= icmp_hdr(icmp_skb
)->code
;
341 struct request_sock
*req
;
345 struct net
*net
= dev_net(icmp_skb
->dev
);
347 if (icmp_skb
->len
< (iph
->ihl
<< 2) + 8) {
348 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
352 sk
= inet_lookup(net
, &tcp_hashinfo
, iph
->daddr
, th
->dest
,
353 iph
->saddr
, th
->source
, inet_iif(icmp_skb
));
355 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
358 if (sk
->sk_state
== TCP_TIME_WAIT
) {
359 inet_twsk_put(inet_twsk(sk
));
364 /* If too many ICMPs get dropped on busy
365 * servers this needs to be solved differently.
366 * We do take care of PMTU discovery (RFC1191) special case :
367 * we can receive locally generated ICMP messages while socket is held.
369 if (sock_owned_by_user(sk
)) {
370 if (!(type
== ICMP_DEST_UNREACH
&& code
== ICMP_FRAG_NEEDED
))
371 NET_INC_STATS_BH(net
, LINUX_MIB_LOCKDROPPEDICMPS
);
373 if (sk
->sk_state
== TCP_CLOSE
)
376 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
377 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
383 req
= tp
->fastopen_rsk
;
384 seq
= ntohl(th
->seq
);
385 if (sk
->sk_state
!= TCP_LISTEN
&&
386 !between(seq
, tp
->snd_una
, tp
->snd_nxt
) &&
387 (req
== NULL
|| seq
!= tcp_rsk(req
)->snt_isn
)) {
388 /* For a Fast Open socket, allow seq to be snt_isn. */
389 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
395 if (!sock_owned_by_user(sk
))
396 do_redirect(icmp_skb
, sk
);
398 case ICMP_SOURCE_QUENCH
:
399 /* Just silently ignore these. */
401 case ICMP_PARAMETERPROB
:
404 case ICMP_DEST_UNREACH
:
405 if (code
> NR_ICMP_UNREACH
)
408 if (code
== ICMP_FRAG_NEEDED
) { /* PMTU discovery (RFC1191) */
409 /* We are not interested in TCP_LISTEN and open_requests
410 * (SYN-ACKs send out by Linux are always <576bytes so
411 * they should go through unfragmented).
413 if (sk
->sk_state
== TCP_LISTEN
)
417 if (!sock_owned_by_user(sk
)) {
418 tcp_v4_mtu_reduced(sk
);
420 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED
, &tp
->tsq_flags
))
426 err
= icmp_err_convert
[code
].errno
;
427 /* check if icmp_skb allows revert of backoff
428 * (see draft-zimmermann-tcp-lcd) */
429 if (code
!= ICMP_NET_UNREACH
&& code
!= ICMP_HOST_UNREACH
)
431 if (seq
!= tp
->snd_una
|| !icsk
->icsk_retransmits
||
435 /* XXX (TFO) - revisit the following logic for TFO */
437 if (sock_owned_by_user(sk
))
440 icsk
->icsk_backoff
--;
441 inet_csk(sk
)->icsk_rto
= (tp
->srtt
? __tcp_set_rto(tp
) :
442 TCP_TIMEOUT_INIT
) << icsk
->icsk_backoff
;
445 skb
= tcp_write_queue_head(sk
);
448 remaining
= icsk
->icsk_rto
- min(icsk
->icsk_rto
,
449 tcp_time_stamp
- TCP_SKB_CB(skb
)->when
);
452 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
453 remaining
, sysctl_tcp_rto_max
);
455 /* RTO revert clocked out retransmission.
456 * Will retransmit now */
457 tcp_retransmit_timer(sk
);
461 case ICMP_TIME_EXCEEDED
:
468 /* XXX (TFO) - if it's a TFO socket and has been accepted, rather
469 * than following the TCP_SYN_RECV case and closing the socket,
470 * we ignore the ICMP error and keep trying like a fully established
471 * socket. Is this the right thing to do?
473 if (req
&& req
->sk
== NULL
)
476 switch (sk
->sk_state
) {
477 struct request_sock
*req
, **prev
;
479 if (sock_owned_by_user(sk
))
482 req
= inet_csk_search_req(sk
, &prev
, th
->dest
,
483 iph
->daddr
, iph
->saddr
);
487 /* ICMPs are not backlogged, hence we cannot get
488 an established socket here.
492 if (seq
!= tcp_rsk(req
)->snt_isn
) {
493 NET_INC_STATS_BH(net
, LINUX_MIB_OUTOFWINDOWICMPS
);
498 * Still in SYN_RECV, just remove it silently.
499 * There is no good way to pass the error to the newly
500 * created socket, and POSIX does not want network
501 * errors returned from accept().
503 inet_csk_reqsk_queue_drop(sk
, req
, prev
);
504 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
508 case TCP_SYN_RECV
: /* Cannot happen.
509 It can f.e. if SYNs crossed,
512 if (!sock_owned_by_user(sk
)) {
515 sk
->sk_error_report(sk
);
519 sk
->sk_err_soft
= err
;
524 /* If we've already connected we will keep trying
525 * until we time out, or the user gives up.
527 * rfc1122 4.2.3.9 allows to consider as hard errors
528 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
529 * but it is obsoleted by pmtu discovery).
531 * Note, that in modern internet, where routing is unreliable
532 * and in each dark corner broken firewalls sit, sending random
533 * errors ordered by their masters even this two messages finally lose
534 * their original sense (even Linux sends invalid PORT_UNREACHs)
536 * Now we are in compliance with RFCs.
541 if (!sock_owned_by_user(sk
) && inet
->recverr
) {
543 sk
->sk_error_report(sk
);
544 } else { /* Only an error on timeout */
545 sk
->sk_err_soft
= err
;
553 static void __tcp_v4_send_check(struct sk_buff
*skb
,
554 __be32 saddr
, __be32 daddr
)
556 struct tcphdr
*th
= tcp_hdr(skb
);
558 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
559 th
->check
= ~tcp_v4_check(skb
->len
, saddr
, daddr
, 0);
560 skb
->csum_start
= skb_transport_header(skb
) - skb
->head
;
561 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
563 th
->check
= tcp_v4_check(skb
->len
, saddr
, daddr
,
570 /* This routine computes an IPv4 TCP checksum. */
571 void tcp_v4_send_check(struct sock
*sk
, struct sk_buff
*skb
)
573 const struct inet_sock
*inet
= inet_sk(sk
);
575 __tcp_v4_send_check(skb
, inet
->inet_saddr
, inet
->inet_daddr
);
577 EXPORT_SYMBOL(tcp_v4_send_check
);
579 int tcp_v4_gso_send_check(struct sk_buff
*skb
)
581 const struct iphdr
*iph
;
584 if (!pskb_may_pull(skb
, sizeof(*th
)))
591 skb
->ip_summed
= CHECKSUM_PARTIAL
;
592 __tcp_v4_send_check(skb
, iph
->saddr
, iph
->daddr
);
597 * This routine will send an RST to the other tcp.
599 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
601 * Answer: if a packet caused RST, it is not for a socket
602 * existing in our system, if it is matched to a socket,
603 * it is just duplicate segment or bug in other side's TCP.
604 * So that we build reply only basing on parameters
605 * arrived with segment.
606 * Exception: precedence violation. We do not implement it in any case.
609 static void tcp_v4_send_reset(struct sock
*sk
, struct sk_buff
*skb
)
611 const struct tcphdr
*th
= tcp_hdr(skb
);
614 #ifdef CONFIG_TCP_MD5SIG
615 __be32 opt
[(TCPOLEN_MD5SIG_ALIGNED
>> 2)];
618 struct ip_reply_arg arg
;
619 #ifdef CONFIG_TCP_MD5SIG
620 struct tcp_md5sig_key
*key
;
621 const __u8
*hash_location
= NULL
;
622 unsigned char newhash
[16];
624 struct sock
*sk1
= NULL
;
628 /* Never send a reset in response to a reset. */
632 if (skb_rtable(skb
)->rt_type
!= RTN_LOCAL
)
635 /* Swap the send and the receive. */
636 memset(&rep
, 0, sizeof(rep
));
637 rep
.th
.dest
= th
->source
;
638 rep
.th
.source
= th
->dest
;
639 rep
.th
.doff
= sizeof(struct tcphdr
) / 4;
643 rep
.th
.seq
= th
->ack_seq
;
646 rep
.th
.ack_seq
= htonl(ntohl(th
->seq
) + th
->syn
+ th
->fin
+
647 skb
->len
- (th
->doff
<< 2));
650 memset(&arg
, 0, sizeof(arg
));
651 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
652 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
654 #ifdef CONFIG_TCP_MD5SIG
655 hash_location
= tcp_parse_md5sig_option(th
);
656 if (!sk
&& hash_location
) {
658 * active side is lost. Try to find listening socket through
659 * source port, and then find md5 key through listening socket.
660 * we are not loose security here:
661 * Incoming packet is checked with md5 hash with finding key,
662 * no RST generated if md5 hash doesn't match.
664 sk1
= __inet_lookup_listener(dev_net(skb_dst(skb
)->dev
),
665 &tcp_hashinfo
, ip_hdr(skb
)->saddr
,
666 th
->source
, ip_hdr(skb
)->daddr
,
667 ntohs(th
->source
), inet_iif(skb
));
668 /* don't send rst if it can't find key */
672 key
= tcp_md5_do_lookup(sk1
, (union tcp_md5_addr
*)
673 &ip_hdr(skb
)->saddr
, AF_INET
);
677 genhash
= tcp_v4_md5_hash_skb(newhash
, key
, NULL
, NULL
, skb
);
678 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0)
681 key
= sk
? tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)
687 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) |
689 (TCPOPT_MD5SIG
<< 8) |
691 /* Update length and the length the header thinks exists */
692 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
693 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
695 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[1],
696 key
, ip_hdr(skb
)->saddr
,
697 ip_hdr(skb
)->daddr
, &rep
.th
);
700 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
701 ip_hdr(skb
)->saddr
, /* XXX */
702 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
703 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
704 arg
.flags
= (sk
&& inet_sk(sk
)->transparent
) ? IP_REPLY_ARG_NOSRCCHECK
: 0;
705 /* When socket is gone, all binding information is lost.
706 * routing might fail in this case. No choice here, if we choose to force
707 * input interface, we will misroute in case of asymmetric route.
710 arg
.bound_dev_if
= sk
->sk_bound_dev_if
;
712 net
= dev_net(skb_dst(skb
)->dev
);
713 arg
.tos
= ip_hdr(skb
)->tos
;
714 ip_send_unicast_reply(*this_cpu_ptr(net
->ipv4
.tcp_sk
),
715 skb
, ip_hdr(skb
)->saddr
,
716 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
718 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
719 TCP_INC_STATS_BH(net
, TCP_MIB_OUTRSTS
);
721 #ifdef CONFIG_TCP_MD5SIG
730 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
731 outside socket context is ugly, certainly. What can I do?
734 static void tcp_v4_send_ack(struct sk_buff
*skb
, u32 seq
, u32 ack
,
735 u32 win
, u32 tsval
, u32 tsecr
, int oif
,
736 struct tcp_md5sig_key
*key
,
737 int reply_flags
, u8 tos
)
739 const struct tcphdr
*th
= tcp_hdr(skb
);
742 __be32 opt
[(TCPOLEN_TSTAMP_ALIGNED
>> 2)
743 #ifdef CONFIG_TCP_MD5SIG
744 + (TCPOLEN_MD5SIG_ALIGNED
>> 2)
748 struct ip_reply_arg arg
;
749 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
751 memset(&rep
.th
, 0, sizeof(struct tcphdr
));
752 memset(&arg
, 0, sizeof(arg
));
754 arg
.iov
[0].iov_base
= (unsigned char *)&rep
;
755 arg
.iov
[0].iov_len
= sizeof(rep
.th
);
757 rep
.opt
[0] = htonl((TCPOPT_NOP
<< 24) | (TCPOPT_NOP
<< 16) |
758 (TCPOPT_TIMESTAMP
<< 8) |
760 rep
.opt
[1] = htonl(tsval
);
761 rep
.opt
[2] = htonl(tsecr
);
762 arg
.iov
[0].iov_len
+= TCPOLEN_TSTAMP_ALIGNED
;
765 /* Swap the send and the receive. */
766 rep
.th
.dest
= th
->source
;
767 rep
.th
.source
= th
->dest
;
768 rep
.th
.doff
= arg
.iov
[0].iov_len
/ 4;
769 rep
.th
.seq
= htonl(seq
);
770 rep
.th
.ack_seq
= htonl(ack
);
772 rep
.th
.window
= htons(win
);
774 #ifdef CONFIG_TCP_MD5SIG
776 int offset
= (tsecr
) ? 3 : 0;
778 rep
.opt
[offset
++] = htonl((TCPOPT_NOP
<< 24) |
780 (TCPOPT_MD5SIG
<< 8) |
782 arg
.iov
[0].iov_len
+= TCPOLEN_MD5SIG_ALIGNED
;
783 rep
.th
.doff
= arg
.iov
[0].iov_len
/4;
785 tcp_v4_md5_hash_hdr((__u8
*) &rep
.opt
[offset
],
786 key
, ip_hdr(skb
)->saddr
,
787 ip_hdr(skb
)->daddr
, &rep
.th
);
790 arg
.flags
= reply_flags
;
791 arg
.csum
= csum_tcpudp_nofold(ip_hdr(skb
)->daddr
,
792 ip_hdr(skb
)->saddr
, /* XXX */
793 arg
.iov
[0].iov_len
, IPPROTO_TCP
, 0);
794 arg
.csumoffset
= offsetof(struct tcphdr
, check
) / 2;
796 arg
.bound_dev_if
= oif
;
798 ip_send_unicast_reply(*this_cpu_ptr(net
->ipv4
.tcp_sk
),
799 skb
, ip_hdr(skb
)->saddr
,
800 ip_hdr(skb
)->daddr
, &arg
, arg
.iov
[0].iov_len
);
802 TCP_INC_STATS_BH(net
, TCP_MIB_OUTSEGS
);
805 static void tcp_v4_timewait_ack(struct sock
*sk
, struct sk_buff
*skb
)
807 struct inet_timewait_sock
*tw
= inet_twsk(sk
);
808 struct tcp_timewait_sock
*tcptw
= tcp_twsk(sk
);
810 tcp_v4_send_ack(skb
, tcptw
->tw_snd_nxt
, tcptw
->tw_rcv_nxt
,
811 tcptw
->tw_rcv_wnd
>> tw
->tw_rcv_wscale
,
812 tcp_time_stamp
+ tcptw
->tw_ts_offset
,
815 tcp_twsk_md5_key(tcptw
),
816 tw
->tw_transparent
? IP_REPLY_ARG_NOSRCCHECK
: 0,
823 static void tcp_v4_reqsk_send_ack(struct sock
*sk
, struct sk_buff
*skb
,
824 struct request_sock
*req
)
826 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
827 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
829 tcp_v4_send_ack(skb
, (sk
->sk_state
== TCP_LISTEN
) ?
830 tcp_rsk(req
)->snt_isn
+ 1 : tcp_sk(sk
)->snd_nxt
,
831 tcp_rsk(req
)->rcv_nxt
,
832 req
->rcv_wnd
>> inet_rsk(req
)->rcv_wscale
,
836 tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&ip_hdr(skb
)->daddr
,
838 inet_rsk(req
)->no_srccheck
? IP_REPLY_ARG_NOSRCCHECK
: 0,
843 * Send a SYN-ACK after having received a SYN.
844 * This still operates on a request_sock only, not on a big
847 static int tcp_v4_send_synack(struct sock
*sk
, struct dst_entry
*dst
,
848 struct request_sock
*req
,
852 const struct inet_request_sock
*ireq
= inet_rsk(req
);
855 struct sk_buff
* skb
;
857 /* First, grab a route. */
858 if (!dst
&& (dst
= inet_csk_route_req(sk
, &fl4
, req
)) == NULL
)
861 skb
= tcp_make_synack(sk
, dst
, req
, NULL
);
864 __tcp_v4_send_check(skb
, ireq
->loc_addr
, ireq
->rmt_addr
);
866 skb_set_queue_mapping(skb
, queue_mapping
);
867 err
= ip_build_and_send_pkt(skb
, sk
, ireq
->loc_addr
,
870 err
= net_xmit_eval(err
);
871 if (!tcp_rsk(req
)->snt_synack
&& !err
)
872 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
878 static int tcp_v4_rtx_synack(struct sock
*sk
, struct request_sock
*req
)
880 int res
= tcp_v4_send_synack(sk
, NULL
, req
, 0, false);
883 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_RETRANSSEGS
);
888 * IPv4 request_sock destructor.
890 static void tcp_v4_reqsk_destructor(struct request_sock
*req
)
892 kfree(inet_rsk(req
)->opt
);
896 * Return true if a syncookie should be sent
898 bool tcp_syn_flood_action(struct sock
*sk
,
899 const struct sk_buff
*skb
,
902 const char *msg
= "Dropping request";
903 bool want_cookie
= false;
904 struct listen_sock
*lopt
;
908 #ifdef CONFIG_SYN_COOKIES
909 if (sysctl_tcp_syncookies
) {
910 msg
= "Sending cookies";
912 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDOCOOKIES
);
915 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPREQQFULLDROP
);
917 lopt
= inet_csk(sk
)->icsk_accept_queue
.listen_opt
;
918 if (!lopt
->synflood_warned
) {
919 lopt
->synflood_warned
= 1;
920 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
921 proto
, ntohs(tcp_hdr(skb
)->dest
), msg
);
925 EXPORT_SYMBOL(tcp_syn_flood_action
);
928 * Save and compile IPv4 options into the request_sock if needed.
930 static struct ip_options_rcu
*tcp_v4_save_options(struct sk_buff
*skb
)
932 const struct ip_options
*opt
= &(IPCB(skb
)->opt
);
933 struct ip_options_rcu
*dopt
= NULL
;
935 if (opt
&& opt
->optlen
) {
936 int opt_size
= sizeof(*dopt
) + opt
->optlen
;
938 dopt
= kmalloc(opt_size
, GFP_ATOMIC
);
940 if (ip_options_echo(&dopt
->opt
, skb
)) {
949 #ifdef CONFIG_TCP_MD5SIG
951 * RFC2385 MD5 checksumming requires a mapping of
952 * IP address->MD5 Key.
953 * We need to maintain these in the sk structure.
956 /* Find the Key structure for an address. */
957 struct tcp_md5sig_key
*tcp_md5_do_lookup(struct sock
*sk
,
958 const union tcp_md5_addr
*addr
,
961 struct tcp_sock
*tp
= tcp_sk(sk
);
962 struct tcp_md5sig_key
*key
;
963 unsigned int size
= sizeof(struct in_addr
);
964 struct tcp_md5sig_info
*md5sig
;
966 /* caller either holds rcu_read_lock() or socket lock */
967 md5sig
= rcu_dereference_check(tp
->md5sig_info
,
968 sock_owned_by_user(sk
) ||
969 lockdep_is_held(&sk
->sk_lock
.slock
));
972 #if IS_ENABLED(CONFIG_IPV6)
973 if (family
== AF_INET6
)
974 size
= sizeof(struct in6_addr
);
976 hlist_for_each_entry_rcu(key
, &md5sig
->head
, node
) {
977 if (key
->family
!= family
)
979 if (!memcmp(&key
->addr
, addr
, size
))
984 EXPORT_SYMBOL(tcp_md5_do_lookup
);
986 struct tcp_md5sig_key
*tcp_v4_md5_lookup(struct sock
*sk
,
987 struct sock
*addr_sk
)
989 union tcp_md5_addr
*addr
;
991 addr
= (union tcp_md5_addr
*)&inet_sk(addr_sk
)->inet_daddr
;
992 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
994 EXPORT_SYMBOL(tcp_v4_md5_lookup
);
996 static struct tcp_md5sig_key
*tcp_v4_reqsk_md5_lookup(struct sock
*sk
,
997 struct request_sock
*req
)
999 union tcp_md5_addr
*addr
;
1001 addr
= (union tcp_md5_addr
*)&inet_rsk(req
)->rmt_addr
;
1002 return tcp_md5_do_lookup(sk
, addr
, AF_INET
);
1005 /* This can be called on a newly created socket, from other files */
1006 int tcp_md5_do_add(struct sock
*sk
, const union tcp_md5_addr
*addr
,
1007 int family
, const u8
*newkey
, u8 newkeylen
, gfp_t gfp
)
1009 /* Add Key to the list */
1010 struct tcp_md5sig_key
*key
;
1011 struct tcp_sock
*tp
= tcp_sk(sk
);
1012 struct tcp_md5sig_info
*md5sig
;
1014 key
= tcp_md5_do_lookup(sk
, addr
, family
);
1016 /* Pre-existing entry - just update that one. */
1017 memcpy(key
->key
, newkey
, newkeylen
);
1018 key
->keylen
= newkeylen
;
1022 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
1023 sock_owned_by_user(sk
) ||
1024 lockdep_is_held(&sk
->sk_lock
.slock
));
1026 md5sig
= kmalloc(sizeof(*md5sig
), gfp
);
1030 sk_nocaps_add(sk
, NETIF_F_GSO_MASK
);
1031 INIT_HLIST_HEAD(&md5sig
->head
);
1032 rcu_assign_pointer(tp
->md5sig_info
, md5sig
);
1035 key
= sock_kmalloc(sk
, sizeof(*key
), gfp
);
1038 if (hlist_empty(&md5sig
->head
) && !tcp_alloc_md5sig_pool(sk
)) {
1039 sock_kfree_s(sk
, key
, sizeof(*key
));
1043 memcpy(key
->key
, newkey
, newkeylen
);
1044 key
->keylen
= newkeylen
;
1045 key
->family
= family
;
1046 memcpy(&key
->addr
, addr
,
1047 (family
== AF_INET6
) ? sizeof(struct in6_addr
) :
1048 sizeof(struct in_addr
));
1049 hlist_add_head_rcu(&key
->node
, &md5sig
->head
);
1052 EXPORT_SYMBOL(tcp_md5_do_add
);
1054 int tcp_md5_do_del(struct sock
*sk
, const union tcp_md5_addr
*addr
, int family
)
1056 struct tcp_sock
*tp
= tcp_sk(sk
);
1057 struct tcp_md5sig_key
*key
;
1058 struct tcp_md5sig_info
*md5sig
;
1060 key
= tcp_md5_do_lookup(sk
, addr
, family
);
1063 hlist_del_rcu(&key
->node
);
1064 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1065 kfree_rcu(key
, rcu
);
1066 md5sig
= rcu_dereference_protected(tp
->md5sig_info
,
1067 sock_owned_by_user(sk
));
1068 if (hlist_empty(&md5sig
->head
))
1069 tcp_free_md5sig_pool();
1072 EXPORT_SYMBOL(tcp_md5_do_del
);
1074 static void tcp_clear_md5_list(struct sock
*sk
)
1076 struct tcp_sock
*tp
= tcp_sk(sk
);
1077 struct tcp_md5sig_key
*key
;
1078 struct hlist_node
*n
;
1079 struct tcp_md5sig_info
*md5sig
;
1081 md5sig
= rcu_dereference_protected(tp
->md5sig_info
, 1);
1083 if (!hlist_empty(&md5sig
->head
))
1084 tcp_free_md5sig_pool();
1085 hlist_for_each_entry_safe(key
, n
, &md5sig
->head
, node
) {
1086 hlist_del_rcu(&key
->node
);
1087 atomic_sub(sizeof(*key
), &sk
->sk_omem_alloc
);
1088 kfree_rcu(key
, rcu
);
1092 static int tcp_v4_parse_md5_keys(struct sock
*sk
, char __user
*optval
,
1095 struct tcp_md5sig cmd
;
1096 struct sockaddr_in
*sin
= (struct sockaddr_in
*)&cmd
.tcpm_addr
;
1098 if (optlen
< sizeof(cmd
))
1101 if (copy_from_user(&cmd
, optval
, sizeof(cmd
)))
1104 if (sin
->sin_family
!= AF_INET
)
1107 if (!cmd
.tcpm_key
|| !cmd
.tcpm_keylen
)
1108 return tcp_md5_do_del(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1111 if (cmd
.tcpm_keylen
> TCP_MD5SIG_MAXKEYLEN
)
1114 return tcp_md5_do_add(sk
, (union tcp_md5_addr
*)&sin
->sin_addr
.s_addr
,
1115 AF_INET
, cmd
.tcpm_key
, cmd
.tcpm_keylen
,
1119 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool
*hp
,
1120 __be32 daddr
, __be32 saddr
, int nbytes
)
1122 struct tcp4_pseudohdr
*bp
;
1123 struct scatterlist sg
;
1125 bp
= &hp
->md5_blk
.ip4
;
1128 * 1. the TCP pseudo-header (in the order: source IP address,
1129 * destination IP address, zero-padded protocol number, and
1135 bp
->protocol
= IPPROTO_TCP
;
1136 bp
->len
= cpu_to_be16(nbytes
);
1138 sg_init_one(&sg
, bp
, sizeof(*bp
));
1139 return crypto_hash_update(&hp
->md5_desc
, &sg
, sizeof(*bp
));
1142 static int tcp_v4_md5_hash_hdr(char *md5_hash
, const struct tcp_md5sig_key
*key
,
1143 __be32 daddr
, __be32 saddr
, const struct tcphdr
*th
)
1145 struct tcp_md5sig_pool
*hp
;
1146 struct hash_desc
*desc
;
1148 hp
= tcp_get_md5sig_pool();
1150 goto clear_hash_noput
;
1151 desc
= &hp
->md5_desc
;
1153 if (crypto_hash_init(desc
))
1155 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, th
->doff
<< 2))
1157 if (tcp_md5_hash_header(hp
, th
))
1159 if (tcp_md5_hash_key(hp
, key
))
1161 if (crypto_hash_final(desc
, md5_hash
))
1164 tcp_put_md5sig_pool();
1168 tcp_put_md5sig_pool();
1170 memset(md5_hash
, 0, 16);
1174 int tcp_v4_md5_hash_skb(char *md5_hash
, struct tcp_md5sig_key
*key
,
1175 const struct sock
*sk
, const struct request_sock
*req
,
1176 const struct sk_buff
*skb
)
1178 struct tcp_md5sig_pool
*hp
;
1179 struct hash_desc
*desc
;
1180 const struct tcphdr
*th
= tcp_hdr(skb
);
1181 __be32 saddr
, daddr
;
1184 saddr
= inet_sk(sk
)->inet_saddr
;
1185 daddr
= inet_sk(sk
)->inet_daddr
;
1187 saddr
= inet_rsk(req
)->loc_addr
;
1188 daddr
= inet_rsk(req
)->rmt_addr
;
1190 const struct iphdr
*iph
= ip_hdr(skb
);
1195 hp
= tcp_get_md5sig_pool();
1197 goto clear_hash_noput
;
1198 desc
= &hp
->md5_desc
;
1200 if (crypto_hash_init(desc
))
1203 if (tcp_v4_md5_hash_pseudoheader(hp
, daddr
, saddr
, skb
->len
))
1205 if (tcp_md5_hash_header(hp
, th
))
1207 if (tcp_md5_hash_skb_data(hp
, skb
, th
->doff
<< 2))
1209 if (tcp_md5_hash_key(hp
, key
))
1211 if (crypto_hash_final(desc
, md5_hash
))
1214 tcp_put_md5sig_pool();
1218 tcp_put_md5sig_pool();
1220 memset(md5_hash
, 0, 16);
1223 EXPORT_SYMBOL(tcp_v4_md5_hash_skb
);
1225 static bool tcp_v4_inbound_md5_hash(struct sock
*sk
, const struct sk_buff
*skb
)
1228 * This gets called for each TCP segment that arrives
1229 * so we want to be efficient.
1230 * We have 3 drop cases:
1231 * o No MD5 hash and one expected.
1232 * o MD5 hash and we're not expecting one.
1233 * o MD5 hash and its wrong.
1235 const __u8
*hash_location
= NULL
;
1236 struct tcp_md5sig_key
*hash_expected
;
1237 const struct iphdr
*iph
= ip_hdr(skb
);
1238 const struct tcphdr
*th
= tcp_hdr(skb
);
1240 unsigned char newhash
[16];
1242 hash_expected
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&iph
->saddr
,
1244 hash_location
= tcp_parse_md5sig_option(th
);
1246 /* We've parsed the options - do we have a hash? */
1247 if (!hash_expected
&& !hash_location
)
1250 if (hash_expected
&& !hash_location
) {
1251 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5NOTFOUND
);
1255 if (!hash_expected
&& hash_location
) {
1256 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPMD5UNEXPECTED
);
1260 /* Okay, so this is hash_expected and hash_location -
1261 * so we need to calculate the checksum.
1263 genhash
= tcp_v4_md5_hash_skb(newhash
,
1267 if (genhash
|| memcmp(hash_location
, newhash
, 16) != 0) {
1268 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1269 &iph
->saddr
, ntohs(th
->source
),
1270 &iph
->daddr
, ntohs(th
->dest
),
1271 genhash
? " tcp_v4_calc_md5_hash failed"
1280 struct request_sock_ops tcp_request_sock_ops __read_mostly
= {
1282 .obj_size
= sizeof(struct tcp_request_sock
),
1283 .rtx_syn_ack
= tcp_v4_rtx_synack
,
1284 .send_ack
= tcp_v4_reqsk_send_ack
,
1285 .destructor
= tcp_v4_reqsk_destructor
,
1286 .send_reset
= tcp_v4_send_reset
,
1287 .syn_ack_timeout
= tcp_syn_ack_timeout
,
1290 #ifdef CONFIG_TCP_MD5SIG
1291 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops
= {
1292 .md5_lookup
= tcp_v4_reqsk_md5_lookup
,
1293 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
1297 static bool tcp_fastopen_check(struct sock
*sk
, struct sk_buff
*skb
,
1298 struct request_sock
*req
,
1299 struct tcp_fastopen_cookie
*foc
,
1300 struct tcp_fastopen_cookie
*valid_foc
)
1302 bool skip_cookie
= false;
1303 struct fastopen_queue
*fastopenq
;
1305 if (likely(!fastopen_cookie_present(foc
))) {
1306 /* See include/net/tcp.h for the meaning of these knobs */
1307 if ((sysctl_tcp_fastopen
& TFO_SERVER_ALWAYS
) ||
1308 ((sysctl_tcp_fastopen
& TFO_SERVER_COOKIE_NOT_REQD
) &&
1309 (TCP_SKB_CB(skb
)->end_seq
!= TCP_SKB_CB(skb
)->seq
+ 1)))
1310 skip_cookie
= true; /* no cookie to validate */
1314 fastopenq
= inet_csk(sk
)->icsk_accept_queue
.fastopenq
;
1315 /* A FO option is present; bump the counter. */
1316 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_TCPFASTOPENPASSIVE
);
1318 /* Make sure the listener has enabled fastopen, and we don't
1319 * exceed the max # of pending TFO requests allowed before trying
1320 * to validating the cookie in order to avoid burning CPU cycles
1323 * XXX (TFO) - The implication of checking the max_qlen before
1324 * processing a cookie request is that clients can't differentiate
1325 * between qlen overflow causing Fast Open to be disabled
1326 * temporarily vs a server not supporting Fast Open at all.
1328 if ((sysctl_tcp_fastopen
& TFO_SERVER_ENABLE
) == 0 ||
1329 fastopenq
== NULL
|| fastopenq
->max_qlen
== 0)
1332 if (fastopenq
->qlen
>= fastopenq
->max_qlen
) {
1333 struct request_sock
*req1
;
1334 spin_lock(&fastopenq
->lock
);
1335 req1
= fastopenq
->rskq_rst_head
;
1336 if ((req1
== NULL
) || time_after(req1
->expires
, jiffies
)) {
1337 spin_unlock(&fastopenq
->lock
);
1338 NET_INC_STATS_BH(sock_net(sk
),
1339 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW
);
1340 /* Avoid bumping LINUX_MIB_TCPFASTOPENPASSIVEFAIL*/
1344 fastopenq
->rskq_rst_head
= req1
->dl_next
;
1346 spin_unlock(&fastopenq
->lock
);
1350 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1353 if (foc
->len
== TCP_FASTOPEN_COOKIE_SIZE
) {
1354 if ((sysctl_tcp_fastopen
& TFO_SERVER_COOKIE_NOT_CHKED
) == 0) {
1355 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
, valid_foc
);
1356 if ((valid_foc
->len
!= TCP_FASTOPEN_COOKIE_SIZE
) ||
1357 memcmp(&foc
->val
[0], &valid_foc
->val
[0],
1358 TCP_FASTOPEN_COOKIE_SIZE
) != 0)
1360 valid_foc
->len
= -1;
1362 /* Acknowledge the data received from the peer. */
1363 tcp_rsk(req
)->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1365 } else if (foc
->len
== 0) { /* Client requesting a cookie */
1366 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
, valid_foc
);
1367 NET_INC_STATS_BH(sock_net(sk
),
1368 LINUX_MIB_TCPFASTOPENCOOKIEREQD
);
1370 /* Client sent a cookie with wrong size. Treat it
1371 * the same as invalid and return a valid one.
1373 tcp_fastopen_cookie_gen(ip_hdr(skb
)->saddr
, valid_foc
);
1378 static int tcp_v4_conn_req_fastopen(struct sock
*sk
,
1379 struct sk_buff
*skb
,
1380 struct sk_buff
*skb_synack
,
1381 struct request_sock
*req
)
1383 struct tcp_sock
*tp
= tcp_sk(sk
);
1384 struct request_sock_queue
*queue
= &inet_csk(sk
)->icsk_accept_queue
;
1385 const struct inet_request_sock
*ireq
= inet_rsk(req
);
1389 req
->num_retrans
= 0;
1390 req
->num_timeout
= 0;
1393 child
= inet_csk(sk
)->icsk_af_ops
->syn_recv_sock(sk
, skb
, req
, NULL
);
1394 if (child
== NULL
) {
1395 NET_INC_STATS_BH(sock_net(sk
),
1396 LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
1397 kfree_skb(skb_synack
);
1400 err
= ip_build_and_send_pkt(skb_synack
, sk
, ireq
->loc_addr
,
1401 ireq
->rmt_addr
, ireq
->opt
);
1402 err
= net_xmit_eval(err
);
1404 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1405 /* XXX (TFO) - is it ok to ignore error and continue? */
1407 spin_lock(&queue
->fastopenq
->lock
);
1408 queue
->fastopenq
->qlen
++;
1409 spin_unlock(&queue
->fastopenq
->lock
);
1411 /* Initialize the child socket. Have to fix some values to take
1412 * into account the child is a Fast Open socket and is created
1413 * only out of the bits carried in the SYN packet.
1417 tp
->fastopen_rsk
= req
;
1418 /* Do a hold on the listner sk so that if the listener is being
1419 * closed, the child that has been accepted can live on and still
1420 * access listen_lock.
1423 tcp_rsk(req
)->listener
= sk
;
1425 /* RFC1323: The window in SYN & SYN/ACK segments is never
1426 * scaled. So correct it appropriately.
1428 tp
->snd_wnd
= ntohs(tcp_hdr(skb
)->window
);
1429 tp
->max_window
= tp
->snd_wnd
;
1431 /* Activate the retrans timer so that SYNACK can be retransmitted.
1432 * The request socket is not added to the SYN table of the parent
1433 * because it's been added to the accept queue directly.
1435 inet_csk_reset_xmit_timer(child
, ICSK_TIME_RETRANS
,
1436 TCP_TIMEOUT_INIT
, sysctl_tcp_rto_max
);
1438 /* Add the child socket directly into the accept queue */
1439 inet_csk_reqsk_queue_add(sk
, req
, child
);
1441 /* Now finish processing the fastopen child socket. */
1442 inet_csk(child
)->icsk_af_ops
->rebuild_header(child
);
1443 tcp_init_congestion_control(child
);
1444 tcp_mtup_init(child
);
1445 tcp_init_buffer_space(child
);
1446 tcp_init_metrics(child
);
1448 /* Queue the data carried in the SYN packet. We need to first
1449 * bump skb's refcnt because the caller will attempt to free it.
1451 * XXX (TFO) - we honor a zero-payload TFO request for now.
1452 * (Any reason not to?)
1454 if (TCP_SKB_CB(skb
)->end_seq
== TCP_SKB_CB(skb
)->seq
+ 1) {
1455 /* Don't queue the skb if there is no payload in SYN.
1456 * XXX (TFO) - How about SYN+FIN?
1458 tp
->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1462 __skb_pull(skb
, tcp_hdr(skb
)->doff
* 4);
1463 skb_set_owner_r(skb
, child
);
1464 __skb_queue_tail(&child
->sk_receive_queue
, skb
);
1465 tp
->rcv_nxt
= TCP_SKB_CB(skb
)->end_seq
;
1466 tp
->syn_data_acked
= 1;
1468 sk
->sk_data_ready(sk
, 0);
1469 bh_unlock_sock(child
);
1471 WARN_ON(req
->sk
== NULL
);
1475 int tcp_v4_conn_request(struct sock
*sk
, struct sk_buff
*skb
)
1477 struct tcp_options_received tmp_opt
;
1478 struct request_sock
*req
;
1479 struct inet_request_sock
*ireq
;
1480 struct tcp_sock
*tp
= tcp_sk(sk
);
1481 struct dst_entry
*dst
= NULL
;
1482 __be32 saddr
= ip_hdr(skb
)->saddr
;
1483 __be32 daddr
= ip_hdr(skb
)->daddr
;
1484 __u32 isn
= TCP_SKB_CB(skb
)->when
;
1485 bool want_cookie
= false;
1487 struct tcp_fastopen_cookie foc
= { .len
= -1 };
1488 struct tcp_fastopen_cookie valid_foc
= { .len
= -1 };
1489 struct sk_buff
*skb_synack
;
1492 /* Never answer to SYNs send to broadcast or multicast */
1493 if (skb_rtable(skb
)->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
1496 /* TW buckets are converted to open requests without
1497 * limitations, they conserve resources and peer is
1498 * evidently real one.
1500 if (inet_csk_reqsk_queue_is_full(sk
) && !isn
) {
1501 want_cookie
= tcp_syn_flood_action(sk
, skb
, "TCP");
1506 /* Accept backlog is full. If we have already queued enough
1507 * of warm entries in syn queue, drop request. It is better than
1508 * clogging syn queue with openreqs with exponentially increasing
1511 if (sk_acceptq_is_full(sk
) && inet_csk_reqsk_queue_young(sk
) > 1) {
1512 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1516 req
= inet_reqsk_alloc(&tcp_request_sock_ops
);
1520 #ifdef CONFIG_TCP_MD5SIG
1521 tcp_rsk(req
)->af_specific
= &tcp_request_sock_ipv4_ops
;
1524 tcp_clear_options(&tmp_opt
);
1525 tmp_opt
.mss_clamp
= TCP_MSS_DEFAULT
;
1526 tmp_opt
.user_mss
= tp
->rx_opt
.user_mss
;
1527 tcp_parse_options(skb
, &tmp_opt
, 0, want_cookie
? NULL
: &foc
);
1529 if (want_cookie
&& !tmp_opt
.saw_tstamp
)
1530 tcp_clear_options(&tmp_opt
);
1532 tmp_opt
.tstamp_ok
= tmp_opt
.saw_tstamp
;
1533 tcp_openreq_init(req
, &tmp_opt
, skb
);
1535 ireq
= inet_rsk(req
);
1536 ireq
->loc_addr
= daddr
;
1537 ireq
->rmt_addr
= saddr
;
1538 ireq
->no_srccheck
= inet_sk(sk
)->transparent
;
1539 ireq
->opt
= tcp_v4_save_options(skb
);
1540 ireq
->ir_mark
= inet_request_mark(sk
, skb
);
1542 if (security_inet_conn_request(sk
, skb
, req
))
1545 if (!want_cookie
|| tmp_opt
.tstamp_ok
)
1546 TCP_ECN_create_request(req
, skb
, sock_net(sk
));
1549 isn
= cookie_v4_init_sequence(sk
, skb
, &req
->mss
);
1550 req
->cookie_ts
= tmp_opt
.tstamp_ok
;
1552 /* VJ's idea. We save last timestamp seen
1553 * from the destination in peer table, when entering
1554 * state TIME-WAIT, and check against it before
1555 * accepting new connection request.
1557 * If "isn" is not zero, this request hit alive
1558 * timewait bucket, so that all the necessary checks
1559 * are made in the function processing timewait state.
1561 if (tmp_opt
.saw_tstamp
&&
1562 tcp_death_row
.sysctl_tw_recycle
&&
1563 (dst
= inet_csk_route_req(sk
, &fl4
, req
)) != NULL
&&
1564 fl4
.daddr
== saddr
) {
1565 if (!tcp_peer_is_proven(req
, dst
, true)) {
1566 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_PAWSPASSIVEREJECTED
);
1567 goto drop_and_release
;
1570 /* Kill the following clause, if you dislike this way. */
1571 else if (!sysctl_tcp_syncookies
&&
1572 (sysctl_max_syn_backlog
- inet_csk_reqsk_queue_len(sk
) <
1573 (sysctl_max_syn_backlog
>> 2)) &&
1574 !tcp_peer_is_proven(req
, dst
, false)) {
1575 /* Without syncookies last quarter of
1576 * backlog is filled with destinations,
1577 * proven to be alive.
1578 * It means that we continue to communicate
1579 * to destinations, already remembered
1580 * to the moment of synflood.
1582 LIMIT_NETDEBUG(KERN_DEBUG
pr_fmt("drop open request from %pI4/%u\n"),
1583 &saddr
, ntohs(tcp_hdr(skb
)->source
));
1584 goto drop_and_release
;
1587 isn
= tcp_v4_init_sequence(skb
);
1589 tcp_rsk(req
)->snt_isn
= isn
;
1592 dst
= inet_csk_route_req(sk
, &fl4
, req
);
1596 do_fastopen
= tcp_fastopen_check(sk
, skb
, req
, &foc
, &valid_foc
);
1598 /* We don't call tcp_v4_send_synack() directly because we need
1599 * to make sure a child socket can be created successfully before
1600 * sending back synack!
1602 * XXX (TFO) - Ideally one would simply call tcp_v4_send_synack()
1603 * (or better yet, call tcp_send_synack() in the child context
1604 * directly, but will have to fix bunch of other code first)
1605 * after syn_recv_sock() except one will need to first fix the
1606 * latter to remove its dependency on the current implementation
1607 * of tcp_v4_send_synack()->tcp_select_initial_window().
1609 skb_synack
= tcp_make_synack(sk
, dst
, req
,
1610 fastopen_cookie_present(&valid_foc
) ? &valid_foc
: NULL
);
1613 __tcp_v4_send_check(skb_synack
, ireq
->loc_addr
, ireq
->rmt_addr
);
1614 skb_set_queue_mapping(skb_synack
, skb_get_queue_mapping(skb
));
1618 if (likely(!do_fastopen
)) {
1620 err
= ip_build_and_send_pkt(skb_synack
, sk
, ireq
->loc_addr
,
1621 ireq
->rmt_addr
, ireq
->opt
);
1622 err
= net_xmit_eval(err
);
1623 if (err
|| want_cookie
)
1626 tcp_rsk(req
)->snt_synack
= tcp_time_stamp
;
1627 tcp_rsk(req
)->listener
= NULL
;
1628 /* Add the request_sock to the SYN table */
1629 inet_csk_reqsk_queue_hash_add(sk
, req
, TCP_TIMEOUT_INIT
);
1630 if (fastopen_cookie_present(&foc
) && foc
.len
!= 0)
1631 NET_INC_STATS_BH(sock_net(sk
),
1632 LINUX_MIB_TCPFASTOPENPASSIVEFAIL
);
1633 } else if (tcp_v4_conn_req_fastopen(sk
, skb
, skb_synack
, req
))
1643 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1646 EXPORT_SYMBOL(tcp_v4_conn_request
);
1650 * The three way handshake has completed - we got a valid synack -
1651 * now create the new socket.
1653 struct sock
*tcp_v4_syn_recv_sock(struct sock
*sk
, struct sk_buff
*skb
,
1654 struct request_sock
*req
,
1655 struct dst_entry
*dst
)
1657 struct inet_request_sock
*ireq
;
1658 struct inet_sock
*newinet
;
1659 struct tcp_sock
*newtp
;
1661 #ifdef CONFIG_TCP_MD5SIG
1662 struct tcp_md5sig_key
*key
;
1664 struct ip_options_rcu
*inet_opt
;
1666 if (sk_acceptq_is_full(sk
))
1669 newsk
= tcp_create_openreq_child(sk
, req
, skb
);
1673 newsk
->sk_gso_type
= SKB_GSO_TCPV4
;
1674 inet_sk_rx_dst_set(newsk
, skb
);
1676 newtp
= tcp_sk(newsk
);
1677 newinet
= inet_sk(newsk
);
1678 ireq
= inet_rsk(req
);
1679 newinet
->inet_daddr
= ireq
->rmt_addr
;
1680 newinet
->inet_rcv_saddr
= ireq
->loc_addr
;
1681 newinet
->inet_saddr
= ireq
->loc_addr
;
1682 inet_opt
= ireq
->opt
;
1683 rcu_assign_pointer(newinet
->inet_opt
, inet_opt
);
1685 newinet
->mc_index
= inet_iif(skb
);
1686 newinet
->mc_ttl
= ip_hdr(skb
)->ttl
;
1687 newinet
->rcv_tos
= ip_hdr(skb
)->tos
;
1688 inet_csk(newsk
)->icsk_ext_hdr_len
= 0;
1690 inet_csk(newsk
)->icsk_ext_hdr_len
= inet_opt
->opt
.optlen
;
1691 newinet
->inet_id
= newtp
->write_seq
^ jiffies
;
1694 dst
= inet_csk_route_child_sock(sk
, newsk
, req
);
1698 /* syncookie case : see end of cookie_v4_check() */
1700 sk_setup_caps(newsk
, dst
);
1702 tcp_mtup_init(newsk
);
1703 tcp_sync_mss(newsk
, dst_mtu(dst
));
1704 newtp
->advmss
= dst_metric_advmss(dst
);
1705 if (tcp_sk(sk
)->rx_opt
.user_mss
&&
1706 tcp_sk(sk
)->rx_opt
.user_mss
< newtp
->advmss
)
1707 newtp
->advmss
= tcp_sk(sk
)->rx_opt
.user_mss
;
1709 tcp_initialize_rcv_mss(newsk
);
1710 tcp_synack_rtt_meas(newsk
, req
);
1711 newtp
->total_retrans
= req
->num_retrans
;
1713 #ifdef CONFIG_TCP_MD5SIG
1714 /* Copy over the MD5 key from the original socket */
1715 key
= tcp_md5_do_lookup(sk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1719 * We're using one, so create a matching key
1720 * on the newsk structure. If we fail to get
1721 * memory, then we end up not copying the key
1724 tcp_md5_do_add(newsk
, (union tcp_md5_addr
*)&newinet
->inet_daddr
,
1725 AF_INET
, key
->key
, key
->keylen
, GFP_ATOMIC
);
1726 sk_nocaps_add(newsk
, NETIF_F_GSO_MASK
);
1730 if (__inet_inherit_port(sk
, newsk
) < 0)
1732 __inet_hash_nolisten(newsk
, NULL
);
1737 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENOVERFLOWS
);
1741 NET_INC_STATS_BH(sock_net(sk
), LINUX_MIB_LISTENDROPS
);
1744 inet_csk_prepare_forced_close(newsk
);
1748 EXPORT_SYMBOL(tcp_v4_syn_recv_sock
);
1750 static struct sock
*tcp_v4_hnd_req(struct sock
*sk
, struct sk_buff
*skb
)
1752 struct tcphdr
*th
= tcp_hdr(skb
);
1753 const struct iphdr
*iph
= ip_hdr(skb
);
1755 struct request_sock
**prev
;
1756 /* Find possible connection requests. */
1757 struct request_sock
*req
= inet_csk_search_req(sk
, &prev
, th
->source
,
1758 iph
->saddr
, iph
->daddr
);
1760 return tcp_check_req(sk
, skb
, req
, prev
, false);
1762 nsk
= inet_lookup_established(sock_net(sk
), &tcp_hashinfo
, iph
->saddr
,
1763 th
->source
, iph
->daddr
, th
->dest
, inet_iif(skb
));
1766 if (nsk
->sk_state
!= TCP_TIME_WAIT
) {
1770 inet_twsk_put(inet_twsk(nsk
));
1774 #ifdef CONFIG_SYN_COOKIES
1776 sk
= cookie_v4_check(sk
, skb
, &(IPCB(skb
)->opt
));
1781 static __sum16
tcp_v4_checksum_init(struct sk_buff
*skb
)
1783 const struct iphdr
*iph
= ip_hdr(skb
);
1785 if (skb
->ip_summed
== CHECKSUM_COMPLETE
) {
1786 if (!tcp_v4_check(skb
->len
, iph
->saddr
,
1787 iph
->daddr
, skb
->csum
)) {
1788 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1793 skb
->csum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
1794 skb
->len
, IPPROTO_TCP
, 0);
1796 if (skb
->len
<= 76) {
1797 return __skb_checksum_complete(skb
);
1803 /* The socket must have it's spinlock held when we get
1806 * We have a potential double-lock case here, so even when
1807 * doing backlog processing we use the BH locking scheme.
1808 * This is because we cannot sleep with the original spinlock
1811 int tcp_v4_do_rcv(struct sock
*sk
, struct sk_buff
*skb
)
1814 #ifdef CONFIG_TCP_MD5SIG
1816 * We really want to reject the packet as early as possible
1818 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1819 * o There is an MD5 option and we're not expecting one
1821 if (tcp_v4_inbound_md5_hash(sk
, skb
))
1825 if (sk
->sk_state
== TCP_ESTABLISHED
) { /* Fast path */
1826 struct dst_entry
*dst
= sk
->sk_rx_dst
;
1828 sock_rps_save_rxhash(sk
, skb
);
1830 if (inet_sk(sk
)->rx_dst_ifindex
!= skb
->skb_iif
||
1831 dst
->ops
->check(dst
, 0) == NULL
) {
1833 sk
->sk_rx_dst
= NULL
;
1836 if (tcp_rcv_established(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1843 if (skb
->len
< tcp_hdrlen(skb
) || tcp_checksum_complete(skb
))
1846 if (sk
->sk_state
== TCP_LISTEN
) {
1847 struct sock
*nsk
= tcp_v4_hnd_req(sk
, skb
);
1852 sock_rps_save_rxhash(nsk
, skb
);
1853 if (tcp_child_process(sk
, nsk
, skb
)) {
1860 sock_rps_save_rxhash(sk
, skb
);
1862 if (tcp_rcv_state_process(sk
, skb
, tcp_hdr(skb
), skb
->len
)) {
1869 tcp_v4_send_reset(rsk
, skb
);
1872 /* Be careful here. If this function gets more complicated and
1873 * gcc suffers from register pressure on the x86, sk (in %ebx)
1874 * might be destroyed here. This current version compiles correctly,
1875 * but you have been warned.
1880 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_CSUMERRORS
);
1881 TCP_INC_STATS_BH(sock_net(sk
), TCP_MIB_INERRS
);
1884 EXPORT_SYMBOL(tcp_v4_do_rcv
);
1886 void tcp_v4_early_demux(struct sk_buff
*skb
)
1888 const struct iphdr
*iph
;
1889 const struct tcphdr
*th
;
1892 if (skb
->pkt_type
!= PACKET_HOST
)
1895 if (!pskb_may_pull(skb
, skb_transport_offset(skb
) + sizeof(struct tcphdr
)))
1901 if (th
->doff
< sizeof(struct tcphdr
) / 4)
1904 sk
= __inet_lookup_established(dev_net(skb
->dev
), &tcp_hashinfo
,
1905 iph
->saddr
, th
->source
,
1906 iph
->daddr
, ntohs(th
->dest
),
1910 skb
->destructor
= sock_edemux
;
1911 if (sk
->sk_state
!= TCP_TIME_WAIT
) {
1912 struct dst_entry
*dst
= ACCESS_ONCE(sk
->sk_rx_dst
);
1915 dst
= dst_check(dst
, 0);
1917 inet_sk(sk
)->rx_dst_ifindex
== skb
->skb_iif
)
1918 skb_dst_set_noref(skb
, dst
);
1923 /* Packet is added to VJ-style prequeue for processing in process
1924 * context, if a reader task is waiting. Apparently, this exciting
1925 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1926 * failed somewhere. Latency? Burstiness? Well, at least now we will
1927 * see, why it failed. 8)8) --ANK
1930 bool tcp_prequeue(struct sock
*sk
, struct sk_buff
*skb
)
1932 struct tcp_sock
*tp
= tcp_sk(sk
);
1934 if (sysctl_tcp_low_latency
|| !tp
->ucopy
.task
)
1937 if (skb
->len
<= tcp_hdrlen(skb
) &&
1938 skb_queue_len(&tp
->ucopy
.prequeue
) == 0)
1942 __skb_queue_tail(&tp
->ucopy
.prequeue
, skb
);
1943 tp
->ucopy
.memory
+= skb
->truesize
;
1944 if (tp
->ucopy
.memory
> sk
->sk_rcvbuf
) {
1945 struct sk_buff
*skb1
;
1947 BUG_ON(sock_owned_by_user(sk
));
1949 while ((skb1
= __skb_dequeue(&tp
->ucopy
.prequeue
)) != NULL
) {
1950 sk_backlog_rcv(sk
, skb1
);
1951 NET_INC_STATS_BH(sock_net(sk
),
1952 LINUX_MIB_TCPPREQUEUEDROPPED
);
1955 tp
->ucopy
.memory
= 0;
1956 } else if (skb_queue_len(&tp
->ucopy
.prequeue
) == 1) {
1957 wake_up_interruptible_sync_poll(sk_sleep(sk
),
1958 POLLIN
| POLLRDNORM
| POLLRDBAND
);
1959 if (!inet_csk_ack_scheduled(sk
))
1960 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_DACK
,
1961 (3 * tcp_rto_min(sk
)) / 4,
1962 sysctl_tcp_rto_max
);
1966 EXPORT_SYMBOL(tcp_prequeue
);
1968 int tcp_filter(struct sock
*sk
, struct sk_buff
*skb
)
1970 struct tcphdr
*th
= (struct tcphdr
*)skb
->data
;
1971 unsigned int eaten
= skb
->len
;
1974 err
= sk_filter_trim_cap(sk
, skb
, th
->doff
* 4);
1977 TCP_SKB_CB(skb
)->end_seq
-= eaten
;
1981 EXPORT_SYMBOL(tcp_filter
);
1987 int tcp_v4_rcv(struct sk_buff
*skb
)
1989 const struct iphdr
*iph
;
1990 const struct tcphdr
*th
;
1993 struct net
*net
= dev_net(skb
->dev
);
1995 if (skb
->pkt_type
!= PACKET_HOST
)
1998 /* Count it even if it's bad */
1999 TCP_INC_STATS_BH(net
, TCP_MIB_INSEGS
);
2001 if (!pskb_may_pull(skb
, sizeof(struct tcphdr
)))
2006 if (th
->doff
< sizeof(struct tcphdr
) / 4)
2008 if (!pskb_may_pull(skb
, th
->doff
* 4))
2011 /* An explanation is required here, I think.
2012 * Packet length and doff are validated by header prediction,
2013 * provided case of th->doff==0 is eliminated.
2014 * So, we defer the checks. */
2015 if (!skb_csum_unnecessary(skb
) && tcp_v4_checksum_init(skb
))
2020 TCP_SKB_CB(skb
)->seq
= ntohl(th
->seq
);
2021 TCP_SKB_CB(skb
)->end_seq
= (TCP_SKB_CB(skb
)->seq
+ th
->syn
+ th
->fin
+
2022 skb
->len
- th
->doff
* 4);
2023 TCP_SKB_CB(skb
)->ack_seq
= ntohl(th
->ack_seq
);
2024 TCP_SKB_CB(skb
)->when
= 0;
2025 TCP_SKB_CB(skb
)->ip_dsfield
= ipv4_get_dsfield(iph
);
2026 TCP_SKB_CB(skb
)->sacked
= 0;
2028 sk
= __inet_lookup_skb(&tcp_hashinfo
, skb
, th
->source
, th
->dest
);
2033 if (sk
->sk_state
== TCP_TIME_WAIT
)
2036 if (unlikely(iph
->ttl
< inet_sk(sk
)->min_ttl
)) {
2037 NET_INC_STATS_BH(net
, LINUX_MIB_TCPMINTTLDROP
);
2038 goto discard_and_relse
;
2041 if (!xfrm4_policy_check(sk
, XFRM_POLICY_IN
, skb
))
2042 goto discard_and_relse
;
2045 if (tcp_filter(sk
, skb
))
2046 goto discard_and_relse
;
2047 th
= (const struct tcphdr
*)skb
->data
;
2052 bh_lock_sock_nested(sk
);
2054 if (!sock_owned_by_user(sk
)) {
2055 #ifdef CONFIG_NET_DMA
2056 struct tcp_sock
*tp
= tcp_sk(sk
);
2057 if (!tp
->ucopy
.dma_chan
&& tp
->ucopy
.pinned_list
)
2058 tp
->ucopy
.dma_chan
= net_dma_find_channel();
2059 if (tp
->ucopy
.dma_chan
)
2060 ret
= tcp_v4_do_rcv(sk
, skb
);
2064 if (!tcp_prequeue(sk
, skb
))
2065 ret
= tcp_v4_do_rcv(sk
, skb
);
2067 } else if (unlikely(sk_add_backlog(sk
, skb
,
2068 sk
->sk_rcvbuf
+ sk
->sk_sndbuf
))) {
2070 NET_INC_STATS_BH(net
, LINUX_MIB_TCPBACKLOGDROP
);
2071 goto discard_and_relse
;
2080 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
))
2083 if (skb
->len
< (th
->doff
<< 2) || tcp_checksum_complete(skb
)) {
2085 TCP_INC_STATS_BH(net
, TCP_MIB_CSUMERRORS
);
2087 TCP_INC_STATS_BH(net
, TCP_MIB_INERRS
);
2089 tcp_v4_send_reset(NULL
, skb
);
2093 /* Discard frame. */
2102 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
2103 inet_twsk_put(inet_twsk(sk
));
2107 if (skb
->len
< (th
->doff
<< 2)) {
2108 inet_twsk_put(inet_twsk(sk
));
2111 if (tcp_checksum_complete(skb
)) {
2112 inet_twsk_put(inet_twsk(sk
));
2115 switch (tcp_timewait_state_process(inet_twsk(sk
), skb
, th
)) {
2117 struct sock
*sk2
= inet_lookup_listener(dev_net(skb
->dev
),
2119 iph
->saddr
, th
->source
,
2120 iph
->daddr
, th
->dest
,
2123 inet_twsk_deschedule(inet_twsk(sk
), &tcp_death_row
);
2124 inet_twsk_put(inet_twsk(sk
));
2128 /* Fall through to ACK */
2131 tcp_v4_timewait_ack(sk
, skb
);
2135 case TCP_TW_SUCCESS
:;
2140 static struct timewait_sock_ops tcp_timewait_sock_ops
= {
2141 .twsk_obj_size
= sizeof(struct tcp_timewait_sock
),
2142 .twsk_unique
= tcp_twsk_unique
,
2143 .twsk_destructor
= tcp_twsk_destructor
,
2146 void inet_sk_rx_dst_set(struct sock
*sk
, const struct sk_buff
*skb
)
2148 struct dst_entry
*dst
= skb_dst(skb
);
2151 sk
->sk_rx_dst
= dst
;
2152 inet_sk(sk
)->rx_dst_ifindex
= skb
->skb_iif
;
2154 EXPORT_SYMBOL(inet_sk_rx_dst_set
);
2156 const struct inet_connection_sock_af_ops ipv4_specific
= {
2157 .queue_xmit
= ip_queue_xmit
,
2158 .send_check
= tcp_v4_send_check
,
2159 .rebuild_header
= inet_sk_rebuild_header
,
2160 .sk_rx_dst_set
= inet_sk_rx_dst_set
,
2161 .conn_request
= tcp_v4_conn_request
,
2162 .syn_recv_sock
= tcp_v4_syn_recv_sock
,
2163 .net_header_len
= sizeof(struct iphdr
),
2164 .setsockopt
= ip_setsockopt
,
2165 .getsockopt
= ip_getsockopt
,
2166 .addr2sockaddr
= inet_csk_addr2sockaddr
,
2167 .sockaddr_len
= sizeof(struct sockaddr_in
),
2168 .bind_conflict
= inet_csk_bind_conflict
,
2169 #ifdef CONFIG_COMPAT
2170 .compat_setsockopt
= compat_ip_setsockopt
,
2171 .compat_getsockopt
= compat_ip_getsockopt
,
2173 .mtu_reduced
= tcp_v4_mtu_reduced
,
2175 EXPORT_SYMBOL(ipv4_specific
);
2177 #ifdef CONFIG_TCP_MD5SIG
2178 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific
= {
2179 .md5_lookup
= tcp_v4_md5_lookup
,
2180 .calc_md5_hash
= tcp_v4_md5_hash_skb
,
2181 .md5_parse
= tcp_v4_parse_md5_keys
,
2185 /* NOTE: A lot of things set to zero explicitly by call to
2186 * sk_alloc() so need not be done here.
2188 static int tcp_v4_init_sock(struct sock
*sk
)
2190 struct inet_connection_sock
*icsk
= inet_csk(sk
);
2193 icsk
->icsk_MMSRB
= 0;
2195 icsk
->icsk_af_ops
= &ipv4_specific
;
2197 #ifdef CONFIG_TCP_MD5SIG
2198 tcp_sk(sk
)->af_specific
= &tcp_sock_ipv4_specific
;
2204 void tcp_v4_destroy_sock(struct sock
*sk
)
2206 struct tcp_sock
*tp
= tcp_sk(sk
);
2208 tcp_clear_xmit_timers(sk
);
2210 tcp_cleanup_congestion_control(sk
);
2212 /* Cleanup up the write buffer. */
2213 tcp_write_queue_purge(sk
);
2215 /* Cleans up our, hopefully empty, out_of_order_queue. */
2216 __skb_queue_purge(&tp
->out_of_order_queue
);
2218 #ifdef CONFIG_TCP_MD5SIG
2219 /* Clean up the MD5 key list, if any */
2220 if (tp
->md5sig_info
) {
2221 tcp_clear_md5_list(sk
);
2222 kfree_rcu(tp
->md5sig_info
, rcu
);
2223 tp
->md5sig_info
= NULL
;
2227 #ifdef CONFIG_NET_DMA
2228 /* Cleans up our sk_async_wait_queue */
2229 __skb_queue_purge(&sk
->sk_async_wait_queue
);
2232 /* Clean prequeue, it must be empty really */
2233 __skb_queue_purge(&tp
->ucopy
.prequeue
);
2235 /* Clean up a referenced TCP bind bucket. */
2236 if (inet_csk(sk
)->icsk_bind_hash
)
2239 BUG_ON(tp
->fastopen_rsk
!= NULL
);
2241 /* If socket is aborted during connect operation */
2242 tcp_free_fastopen_req(tp
);
2244 sk_sockets_allocated_dec(sk
);
2245 sock_release_memcg(sk
);
2247 EXPORT_SYMBOL(tcp_v4_destroy_sock
);
2249 void tcp_v4_handle_retrans_time_by_uid(struct uid_err uid_e
)
2251 unsigned int bucket
;
2252 uid_t skuid
= (uid_t
)(uid_e
.appuid
);
2253 struct inet_connection_sock
*icsk
= NULL
;//inet_csk(sk);
2256 for (bucket
= 0; bucket
< tcp_hashinfo
.ehash_mask
; bucket
++) {
2257 struct hlist_nulls_node
*node
;
2259 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, bucket
);
2262 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[bucket
].chain
) {
2264 if (sysctl_ip_dynaddr
&& sk
->sk_state
== TCP_SYN_SENT
)
2266 if (sock_flag(sk
, SOCK_DEAD
))
2270 if(SOCK_INODE(sk
->sk_socket
)->i_uid
!= skuid
)
2273 printk("[mmspb] tcp_v4_handle_retrans_time_by_uid socket uid(%d) match!",
2274 SOCK_INODE(sk
->sk_socket
)->i_uid
);
2280 spin_unlock_bh(lock
);
2285 // update sk time out value
2286 icsk
= inet_csk(sk
);
2287 printk("[mmspb] tcp_v4_handle_retrans_time_by_uid update timer\n");
2289 sk_reset_timer(sk
, &icsk
->icsk_retransmit_timer
, jiffies
+ 2);
2290 icsk
->icsk_rto
= sysctl_tcp_rto_min
* 30;
2291 icsk
->icsk_MMSRB
= 1;
2299 spin_unlock_bh(lock
);
2306 * tcp_v4_nuke_addr_by_uid - destroy all sockets of spcial uid
2308 void tcp_v4_reset_connections_by_uid(struct uid_err uid_e
)
2310 unsigned int bucket
;
2311 uid_t skuid
= (uid_t
)(uid_e
.appuid
);
2313 for (bucket
= 0; bucket
< tcp_hashinfo
.ehash_mask
; bucket
++) {
2314 struct hlist_nulls_node
*node
;
2316 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, bucket
);
2320 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[bucket
].chain
) {
2322 if (sysctl_ip_dynaddr
&& sk
->sk_state
== TCP_SYN_SENT
)
2324 if (sock_flag(sk
, SOCK_DEAD
))
2328 if(SOCK_INODE(sk
->sk_socket
)->i_uid
!= skuid
)
2331 printk(KERN_INFO
"SIOCKILLSOCK socket uid(%d) match!",
2332 SOCK_INODE(sk
->sk_socket
)->i_uid
);
2338 spin_unlock_bh(lock
);
2342 sk
->sk_err
= uid_e
.errNum
;
2343 printk(KERN_INFO
"SIOCKILLSOCK set sk err == %d!! \n", sk
->sk_err
);
2344 sk
->sk_error_report(sk
);
2353 spin_unlock_bh(lock
);
2358 #ifdef CONFIG_PROC_FS
2359 /* Proc filesystem TCP sock list dumping. */
2361 static inline struct inet_timewait_sock
*tw_head(struct hlist_nulls_head
*head
)
2363 return hlist_nulls_empty(head
) ? NULL
:
2364 list_entry(head
->first
, struct inet_timewait_sock
, tw_node
);
2367 static inline struct inet_timewait_sock
*tw_next(struct inet_timewait_sock
*tw
)
2369 return !is_a_nulls(tw
->tw_node
.next
) ?
2370 hlist_nulls_entry(tw
->tw_node
.next
, typeof(*tw
), tw_node
) : NULL
;
2374 * Get next listener socket follow cur. If cur is NULL, get first socket
2375 * starting from bucket given in st->bucket; when st->bucket is zero the
2376 * very first socket in the hash table is returned.
2378 static void *listening_get_next(struct seq_file
*seq
, void *cur
)
2380 struct inet_connection_sock
*icsk
;
2381 struct hlist_nulls_node
*node
;
2382 struct sock
*sk
= cur
;
2383 struct inet_listen_hashbucket
*ilb
;
2384 struct tcp_iter_state
*st
= seq
->private;
2385 struct net
*net
= seq_file_net(seq
);
2388 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2389 spin_lock_bh(&ilb
->lock
);
2390 sk
= sk_nulls_head(&ilb
->head
);
2394 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2398 if (st
->state
== TCP_SEQ_STATE_OPENREQ
) {
2399 struct request_sock
*req
= cur
;
2401 icsk
= inet_csk(st
->syn_wait_sk
);
2405 if (req
->rsk_ops
->family
== st
->family
) {
2411 if (++st
->sbucket
>= icsk
->icsk_accept_queue
.listen_opt
->nr_table_entries
)
2414 req
= icsk
->icsk_accept_queue
.listen_opt
->syn_table
[st
->sbucket
];
2416 sk
= sk_nulls_next(st
->syn_wait_sk
);
2417 st
->state
= TCP_SEQ_STATE_LISTENING
;
2418 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2420 icsk
= inet_csk(sk
);
2421 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2422 if (reqsk_queue_len(&icsk
->icsk_accept_queue
))
2424 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2425 sk
= sk_nulls_next(sk
);
2428 sk_nulls_for_each_from(sk
, node
) {
2429 if (!net_eq(sock_net(sk
), net
))
2431 if (sk
->sk_family
== st
->family
) {
2435 icsk
= inet_csk(sk
);
2436 read_lock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2437 if (reqsk_queue_len(&icsk
->icsk_accept_queue
)) {
2439 st
->uid
= sock_i_uid(sk
);
2440 st
->syn_wait_sk
= sk
;
2441 st
->state
= TCP_SEQ_STATE_OPENREQ
;
2445 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2447 spin_unlock_bh(&ilb
->lock
);
2449 if (++st
->bucket
< INET_LHTABLE_SIZE
) {
2450 ilb
= &tcp_hashinfo
.listening_hash
[st
->bucket
];
2451 spin_lock_bh(&ilb
->lock
);
2452 sk
= sk_nulls_head(&ilb
->head
);
2460 static void *listening_get_idx(struct seq_file
*seq
, loff_t
*pos
)
2462 struct tcp_iter_state
*st
= seq
->private;
2467 rc
= listening_get_next(seq
, NULL
);
2469 while (rc
&& *pos
) {
2470 rc
= listening_get_next(seq
, rc
);
2476 static inline bool empty_bucket(struct tcp_iter_state
*st
)
2478 return hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].chain
) &&
2479 hlist_nulls_empty(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2483 * Get first established socket starting from bucket given in st->bucket.
2484 * If st->bucket is zero, the very first socket in the hash is returned.
2486 static void *established_get_first(struct seq_file
*seq
)
2488 struct tcp_iter_state
*st
= seq
->private;
2489 struct net
*net
= seq_file_net(seq
);
2493 for (; st
->bucket
<= tcp_hashinfo
.ehash_mask
; ++st
->bucket
) {
2495 struct hlist_nulls_node
*node
;
2496 struct inet_timewait_sock
*tw
;
2497 spinlock_t
*lock
= inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
);
2499 /* Lockless fast path for the common case of empty buckets */
2500 if (empty_bucket(st
))
2504 sk_nulls_for_each(sk
, node
, &tcp_hashinfo
.ehash
[st
->bucket
].chain
) {
2505 if (sk
->sk_family
!= st
->family
||
2506 !net_eq(sock_net(sk
), net
)) {
2512 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2513 inet_twsk_for_each(tw
, node
,
2514 &tcp_hashinfo
.ehash
[st
->bucket
].twchain
) {
2515 if (tw
->tw_family
!= st
->family
||
2516 !net_eq(twsk_net(tw
), net
)) {
2522 spin_unlock_bh(lock
);
2523 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2529 static void *established_get_next(struct seq_file
*seq
, void *cur
)
2531 struct sock
*sk
= cur
;
2532 struct inet_timewait_sock
*tw
;
2533 struct hlist_nulls_node
*node
;
2534 struct tcp_iter_state
*st
= seq
->private;
2535 struct net
*net
= seq_file_net(seq
);
2540 if (st
->state
== TCP_SEQ_STATE_TIME_WAIT
) {
2544 while (tw
&& (tw
->tw_family
!= st
->family
|| !net_eq(twsk_net(tw
), net
))) {
2551 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2552 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2554 /* Look for next non empty bucket */
2556 while (++st
->bucket
<= tcp_hashinfo
.ehash_mask
&&
2559 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2562 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2563 sk
= sk_nulls_head(&tcp_hashinfo
.ehash
[st
->bucket
].chain
);
2565 sk
= sk_nulls_next(sk
);
2567 sk_nulls_for_each_from(sk
, node
) {
2568 if (sk
->sk_family
== st
->family
&& net_eq(sock_net(sk
), net
))
2572 st
->state
= TCP_SEQ_STATE_TIME_WAIT
;
2573 tw
= tw_head(&tcp_hashinfo
.ehash
[st
->bucket
].twchain
);
2581 static void *established_get_idx(struct seq_file
*seq
, loff_t pos
)
2583 struct tcp_iter_state
*st
= seq
->private;
2587 rc
= established_get_first(seq
);
2590 rc
= established_get_next(seq
, rc
);
2596 static void *tcp_get_idx(struct seq_file
*seq
, loff_t pos
)
2599 struct tcp_iter_state
*st
= seq
->private;
2601 st
->state
= TCP_SEQ_STATE_LISTENING
;
2602 rc
= listening_get_idx(seq
, &pos
);
2605 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2606 rc
= established_get_idx(seq
, pos
);
2612 static void *tcp_seek_last_pos(struct seq_file
*seq
)
2614 struct tcp_iter_state
*st
= seq
->private;
2615 int offset
= st
->offset
;
2616 int orig_num
= st
->num
;
2619 switch (st
->state
) {
2620 case TCP_SEQ_STATE_OPENREQ
:
2621 case TCP_SEQ_STATE_LISTENING
:
2622 if (st
->bucket
>= INET_LHTABLE_SIZE
)
2624 st
->state
= TCP_SEQ_STATE_LISTENING
;
2625 rc
= listening_get_next(seq
, NULL
);
2626 while (offset
-- && rc
)
2627 rc
= listening_get_next(seq
, rc
);
2632 case TCP_SEQ_STATE_ESTABLISHED
:
2633 case TCP_SEQ_STATE_TIME_WAIT
:
2634 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2635 if (st
->bucket
> tcp_hashinfo
.ehash_mask
)
2637 rc
= established_get_first(seq
);
2638 while (offset
-- && rc
)
2639 rc
= established_get_next(seq
, rc
);
2647 static void *tcp_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2649 struct tcp_iter_state
*st
= seq
->private;
2652 if (*pos
&& *pos
== st
->last_pos
) {
2653 rc
= tcp_seek_last_pos(seq
);
2658 st
->state
= TCP_SEQ_STATE_LISTENING
;
2662 rc
= *pos
? tcp_get_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2665 st
->last_pos
= *pos
;
2669 static void *tcp_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2671 struct tcp_iter_state
*st
= seq
->private;
2674 if (v
== SEQ_START_TOKEN
) {
2675 rc
= tcp_get_idx(seq
, 0);
2679 switch (st
->state
) {
2680 case TCP_SEQ_STATE_OPENREQ
:
2681 case TCP_SEQ_STATE_LISTENING
:
2682 rc
= listening_get_next(seq
, v
);
2684 st
->state
= TCP_SEQ_STATE_ESTABLISHED
;
2687 rc
= established_get_first(seq
);
2690 case TCP_SEQ_STATE_ESTABLISHED
:
2691 case TCP_SEQ_STATE_TIME_WAIT
:
2692 rc
= established_get_next(seq
, v
);
2697 st
->last_pos
= *pos
;
2701 static void tcp_seq_stop(struct seq_file
*seq
, void *v
)
2703 struct tcp_iter_state
*st
= seq
->private;
2705 switch (st
->state
) {
2706 case TCP_SEQ_STATE_OPENREQ
:
2708 struct inet_connection_sock
*icsk
= inet_csk(st
->syn_wait_sk
);
2709 read_unlock_bh(&icsk
->icsk_accept_queue
.syn_wait_lock
);
2711 case TCP_SEQ_STATE_LISTENING
:
2712 if (v
!= SEQ_START_TOKEN
)
2713 spin_unlock_bh(&tcp_hashinfo
.listening_hash
[st
->bucket
].lock
);
2715 case TCP_SEQ_STATE_TIME_WAIT
:
2716 case TCP_SEQ_STATE_ESTABLISHED
:
2718 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo
, st
->bucket
));
2723 int tcp_seq_open(struct inode
*inode
, struct file
*file
)
2725 struct tcp_seq_afinfo
*afinfo
= PDE_DATA(inode
);
2726 struct tcp_iter_state
*s
;
2729 err
= seq_open_net(inode
, file
, &afinfo
->seq_ops
,
2730 sizeof(struct tcp_iter_state
));
2734 s
= ((struct seq_file
*)file
->private_data
)->private;
2735 s
->family
= afinfo
->family
;
2739 EXPORT_SYMBOL(tcp_seq_open
);
2741 int tcp_proc_register(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2744 struct proc_dir_entry
*p
;
2746 afinfo
->seq_ops
.start
= tcp_seq_start
;
2747 afinfo
->seq_ops
.next
= tcp_seq_next
;
2748 afinfo
->seq_ops
.stop
= tcp_seq_stop
;
2750 p
= proc_create_data(afinfo
->name
, S_IRUGO
, net
->proc_net
,
2751 afinfo
->seq_fops
, afinfo
);
2756 EXPORT_SYMBOL(tcp_proc_register
);
2758 void tcp_proc_unregister(struct net
*net
, struct tcp_seq_afinfo
*afinfo
)
2760 remove_proc_entry(afinfo
->name
, net
->proc_net
);
2762 EXPORT_SYMBOL(tcp_proc_unregister
);
2764 static void get_openreq4(const struct sock
*sk
, const struct request_sock
*req
,
2765 struct seq_file
*f
, int i
, kuid_t uid
, int *len
)
2767 const struct inet_request_sock
*ireq
= inet_rsk(req
);
2768 long delta
= req
->expires
- jiffies
;
2770 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2771 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2774 ntohs(inet_sk(sk
)->inet_sport
),
2776 ntohs(ireq
->rmt_port
),
2778 0, 0, /* could print option size, but that is af dependent. */
2779 1, /* timers active (only the expire timer) */
2780 jiffies_delta_to_clock_t(delta
),
2782 from_kuid_munged(seq_user_ns(f
), uid
),
2783 0, /* non standard timer */
2784 0, /* open_requests have no inode */
2785 atomic_read(&sk
->sk_refcnt
),
2790 static void get_tcp4_sock(struct sock
*sk
, struct seq_file
*f
, int i
, int *len
)
2793 unsigned long timer_expires
;
2794 const struct tcp_sock
*tp
= tcp_sk(sk
);
2795 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
2796 const struct inet_sock
*inet
= inet_sk(sk
);
2797 struct fastopen_queue
*fastopenq
= icsk
->icsk_accept_queue
.fastopenq
;
2798 __be32 dest
= inet
->inet_daddr
;
2799 __be32 src
= inet
->inet_rcv_saddr
;
2800 __u16 destp
= ntohs(inet
->inet_dport
);
2801 __u16 srcp
= ntohs(inet
->inet_sport
);
2804 if (icsk
->icsk_pending
== ICSK_TIME_RETRANS
||
2805 icsk
->icsk_pending
== ICSK_TIME_EARLY_RETRANS
||
2806 icsk
->icsk_pending
== ICSK_TIME_LOSS_PROBE
) {
2808 timer_expires
= icsk
->icsk_timeout
;
2809 } else if (icsk
->icsk_pending
== ICSK_TIME_PROBE0
) {
2811 timer_expires
= icsk
->icsk_timeout
;
2812 } else if (timer_pending(&sk
->sk_timer
)) {
2814 timer_expires
= sk
->sk_timer
.expires
;
2817 timer_expires
= jiffies
;
2820 if (sk
->sk_state
== TCP_LISTEN
)
2821 rx_queue
= sk
->sk_ack_backlog
;
2824 * because we dont lock socket, we might find a transient negative value
2826 rx_queue
= max_t(int, tp
->rcv_nxt
- tp
->copied_seq
, 0);
2828 seq_printf(f
, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2829 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2830 i
, src
, srcp
, dest
, destp
, sk
->sk_state
,
2831 tp
->write_seq
- tp
->snd_una
,
2834 jiffies_delta_to_clock_t(timer_expires
- jiffies
),
2835 icsk
->icsk_retransmits
,
2836 from_kuid_munged(seq_user_ns(f
), sock_i_uid(sk
)),
2837 icsk
->icsk_probes_out
,
2839 atomic_read(&sk
->sk_refcnt
), sk
,
2840 jiffies_to_clock_t(icsk
->icsk_rto
),
2841 jiffies_to_clock_t(icsk
->icsk_ack
.ato
),
2842 (icsk
->icsk_ack
.quick
<< 1) | icsk
->icsk_ack
.pingpong
,
2844 sk
->sk_state
== TCP_LISTEN
?
2845 (fastopenq
? fastopenq
->max_qlen
: 0) :
2846 (tcp_in_initial_slowstart(tp
) ? -1 : tp
->snd_ssthresh
),
2850 static void get_timewait4_sock(const struct inet_timewait_sock
*tw
,
2851 struct seq_file
*f
, int i
, int *len
)
2855 long delta
= tw
->tw_ttd
- jiffies
;
2857 dest
= tw
->tw_daddr
;
2858 src
= tw
->tw_rcv_saddr
;
2859 destp
= ntohs(tw
->tw_dport
);
2860 srcp
= ntohs(tw
->tw_sport
);
2862 seq_printf(f
, "%4d: %08X:%04X %08X:%04X"
2863 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2864 i
, src
, srcp
, dest
, destp
, tw
->tw_substate
, 0, 0,
2865 3, jiffies_delta_to_clock_t(delta
), 0, 0, 0, 0,
2866 atomic_read(&tw
->tw_refcnt
), tw
, len
);
2871 static int tcp4_seq_show(struct seq_file
*seq
, void *v
)
2873 struct tcp_iter_state
*st
;
2876 if (v
== SEQ_START_TOKEN
) {
2877 seq_printf(seq
, "%-*s\n", TMPSZ
- 1,
2878 " sl local_address rem_address st tx_queue "
2879 "rx_queue tr tm->when retrnsmt uid timeout "
2885 switch (st
->state
) {
2886 case TCP_SEQ_STATE_LISTENING
:
2887 case TCP_SEQ_STATE_ESTABLISHED
:
2888 get_tcp4_sock(v
, seq
, st
->num
, &len
);
2890 case TCP_SEQ_STATE_OPENREQ
:
2891 get_openreq4(st
->syn_wait_sk
, v
, seq
, st
->num
, st
->uid
, &len
);
2893 case TCP_SEQ_STATE_TIME_WAIT
:
2894 get_timewait4_sock(v
, seq
, st
->num
, &len
);
2897 seq_printf(seq
, "%*s\n", TMPSZ
- 1 - len
, "");
2902 static const struct file_operations tcp_afinfo_seq_fops
= {
2903 .owner
= THIS_MODULE
,
2904 .open
= tcp_seq_open
,
2906 .llseek
= seq_lseek
,
2907 .release
= seq_release_net
2910 static struct tcp_seq_afinfo tcp4_seq_afinfo
= {
2913 .seq_fops
= &tcp_afinfo_seq_fops
,
2915 .show
= tcp4_seq_show
,
2919 static int __net_init
tcp4_proc_init_net(struct net
*net
)
2921 return tcp_proc_register(net
, &tcp4_seq_afinfo
);
2924 static void __net_exit
tcp4_proc_exit_net(struct net
*net
)
2926 tcp_proc_unregister(net
, &tcp4_seq_afinfo
);
2929 static struct pernet_operations tcp4_net_ops
= {
2930 .init
= tcp4_proc_init_net
,
2931 .exit
= tcp4_proc_exit_net
,
2934 int __init
tcp4_proc_init(void)
2936 return register_pernet_subsys(&tcp4_net_ops
);
2939 void tcp4_proc_exit(void)
2941 unregister_pernet_subsys(&tcp4_net_ops
);
2943 #endif /* CONFIG_PROC_FS */
2945 struct sk_buff
**tcp4_gro_receive(struct sk_buff
**head
, struct sk_buff
*skb
)
2947 const struct iphdr
*iph
= skb_gro_network_header(skb
);
2951 switch (skb
->ip_summed
) {
2952 case CHECKSUM_COMPLETE
:
2953 if (!tcp_v4_check(skb_gro_len(skb
), iph
->saddr
, iph
->daddr
,
2955 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2959 NAPI_GRO_CB(skb
)->flush
= 1;
2963 wsum
= csum_tcpudp_nofold(iph
->saddr
, iph
->daddr
,
2964 skb_gro_len(skb
), IPPROTO_TCP
, 0);
2965 sum
= csum_fold(skb_checksum(skb
,
2966 skb_gro_offset(skb
),
2972 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
2976 return tcp_gro_receive(head
, skb
);
2979 int tcp4_gro_complete(struct sk_buff
*skb
)
2981 const struct iphdr
*iph
= ip_hdr(skb
);
2982 struct tcphdr
*th
= tcp_hdr(skb
);
2984 th
->check
= ~tcp_v4_check(skb
->len
- skb_transport_offset(skb
),
2985 iph
->saddr
, iph
->daddr
, 0);
2986 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
2988 return tcp_gro_complete(skb
);
2991 struct proto tcp_prot
= {
2993 .owner
= THIS_MODULE
,
2995 .connect
= tcp_v4_connect
,
2996 .disconnect
= tcp_disconnect
,
2997 .accept
= inet_csk_accept
,
2999 .init
= tcp_v4_init_sock
,
3000 .destroy
= tcp_v4_destroy_sock
,
3001 .shutdown
= tcp_shutdown
,
3002 .setsockopt
= tcp_setsockopt
,
3003 .getsockopt
= tcp_getsockopt
,
3004 .recvmsg
= tcp_recvmsg
,
3005 .sendmsg
= tcp_sendmsg
,
3006 .sendpage
= tcp_sendpage
,
3007 .backlog_rcv
= tcp_v4_do_rcv
,
3008 .release_cb
= tcp_release_cb
,
3010 .unhash
= inet_unhash
,
3011 .get_port
= inet_csk_get_port
,
3012 .enter_memory_pressure
= tcp_enter_memory_pressure
,
3013 .sockets_allocated
= &tcp_sockets_allocated
,
3014 .orphan_count
= &tcp_orphan_count
,
3015 .memory_allocated
= &tcp_memory_allocated
,
3016 .memory_pressure
= &tcp_memory_pressure
,
3017 .sysctl_wmem
= sysctl_tcp_wmem
,
3018 .sysctl_rmem
= sysctl_tcp_rmem
,
3019 .max_header
= MAX_TCP_HEADER
,
3020 .obj_size
= sizeof(struct tcp_sock
),
3021 .slab_flags
= SLAB_DESTROY_BY_RCU
,
3022 .twsk_prot
= &tcp_timewait_sock_ops
,
3023 .rsk_prot
= &tcp_request_sock_ops
,
3024 .h
.hashinfo
= &tcp_hashinfo
,
3025 .no_autobind
= true,
3026 #ifdef CONFIG_COMPAT
3027 .compat_setsockopt
= compat_tcp_setsockopt
,
3028 .compat_getsockopt
= compat_tcp_getsockopt
,
3030 #ifdef CONFIG_MEMCG_KMEM
3031 .init_cgroup
= tcp_init_cgroup
,
3032 .destroy_cgroup
= tcp_destroy_cgroup
,
3033 .proto_cgroup
= tcp_proto_cgroup
,
3036 EXPORT_SYMBOL(tcp_prot
);
3038 static void __net_exit
tcp_sk_exit(struct net
*net
)
3042 for_each_possible_cpu(cpu
)
3043 inet_ctl_sock_destroy(*per_cpu_ptr(net
->ipv4
.tcp_sk
, cpu
));
3044 free_percpu(net
->ipv4
.tcp_sk
);
3047 static int __net_init
tcp_sk_init(struct net
*net
)
3051 net
->ipv4
.tcp_sk
= alloc_percpu(struct sock
*);
3052 if (!net
->ipv4
.tcp_sk
)
3055 for_each_possible_cpu(cpu
) {
3058 res
= inet_ctl_sock_create(&sk
, PF_INET
, SOCK_RAW
,
3062 *per_cpu_ptr(net
->ipv4
.tcp_sk
, cpu
) = sk
;
3064 net
->ipv4
.sysctl_tcp_ecn
= 2;
3073 static void __net_exit
tcp_sk_exit_batch(struct list_head
*net_exit_list
)
3075 inet_twsk_purge(&tcp_hashinfo
, &tcp_death_row
, AF_INET
);
3078 static struct pernet_operations __net_initdata tcp_sk_ops
= {
3079 .init
= tcp_sk_init
,
3080 .exit
= tcp_sk_exit
,
3081 .exit_batch
= tcp_sk_exit_batch
,
3084 void __init
tcp_v4_init(void)
3086 inet_hashinfo_init(&tcp_hashinfo
);
3087 if (register_pernet_subsys(&tcp_sk_ops
))
3088 panic("Failed to create the TCP control socket.\n");