blob: db003438aaf5f6a2b27319c4669be1b1f62c7ae3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * IPv4 specific functions
9 *
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 *
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.
22 */
23
24/*
25 * Changes:
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
34 * ACK bit.
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -070037 * request_sock handling and moved
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080040 * Added new listen semantics.
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 * 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
45 * coma.
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.
51 */
52
Joe Perchesafd465032012-03-12 07:03:32 +000053#define pr_fmt(fmt) "TCP: " fmt
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Herbert Xueb4dea52008-12-29 23:04:08 -080055#include <linux/bottom_half.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090064#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020066#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <net/icmp.h>
Arnaldo Carvalho de Melo304a1612005-08-09 19:59:20 -070068#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include <net/tcp.h>
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -030070#include <net/transp_v6.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#include <net/ipv6.h>
72#include <net/inet_common.h>
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -080073#include <net/timewait_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <net/xfrm.h>
David S. Miller6e5714e2011-08-03 20:50:44 -070075#include <net/secure_seq.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +000076#include <net/tcp_memcontrol.h>
Eliezer Tamir076bb0c2013-07-10 17:13:17 +030077#include <net/busy_poll.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
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>
84
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080085#include <linux/crypto.h>
86#include <linux/scatterlist.h>
87
Brian Haleyab32ea52006-09-22 14:15:41 -070088int sysctl_tcp_tw_reuse __read_mostly;
89int sysctl_tcp_low_latency __read_mostly;
Eric Dumazet4bc2f182010-07-09 21:22:10 +000090EXPORT_SYMBOL(sysctl_tcp_low_latency);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080092#ifdef CONFIG_TCP_MD5SIG
Eric Dumazeta915da9b2012-01-31 05:18:33 +000093static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
Eric Dumazet318cf7a2011-10-24 02:46:04 -040094 __be32 daddr, __be32 saddr, const struct tcphdr *th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080095#endif
96
Eric Dumazet5caea4e2008-11-20 00:40:07 -080097struct inet_hashinfo tcp_hashinfo;
Eric Dumazet4bc2f182010-07-09 21:22:10 +000098EXPORT_SYMBOL(tcp_hashinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Octavian Purdila936b8bd2014-06-25 17:09:57 +0300100static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700102 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
103 ip_hdr(skb)->saddr,
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700104 tcp_hdr(skb)->dest,
105 tcp_hdr(skb)->source);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106}
107
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800108int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
109{
110 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
111 struct tcp_sock *tp = tcp_sk(sk);
112
113 /* With PAWS, it is safe from the viewpoint
114 of data integrity. Even without PAWS it is safe provided sequence
115 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
116
117 Actually, the idea is close to VJ's one, only timestamp cache is
118 held not per host, but per port pair and TW bucket is used as state
119 holder.
120
121 If TW bucket has been already destroyed we fall back to VJ's scheme
122 and use initial timestamp retrieved from peer table.
123 */
124 if (tcptw->tw_ts_recent_stamp &&
Ian Morris51456b22015-04-03 09:17:26 +0100125 (!twp || (sysctl_tcp_tw_reuse &&
James Morris9d729f72007-03-04 16:12:44 -0800126 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800127 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
128 if (tp->write_seq == 0)
129 tp->write_seq = 1;
130 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
131 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
132 sock_hold(sktw);
133 return 1;
134 }
135
136 return 0;
137}
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800138EXPORT_SYMBOL_GPL(tcp_twsk_unique);
139
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140/* This will initiate an outgoing connection. */
141int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
142{
David S. Miller2d7192d2011-04-26 13:28:44 -0700143 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 struct inet_sock *inet = inet_sk(sk);
145 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerdca8b082011-02-24 13:38:12 -0800146 __be16 orig_sport, orig_dport;
Al Virobada8ad2006-09-26 21:27:15 -0700147 __be32 daddr, nexthop;
David S. Millerda905bd2011-05-06 16:11:19 -0700148 struct flowi4 *fl4;
David S. Miller2d7192d2011-04-26 13:28:44 -0700149 struct rtable *rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 int err;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000151 struct ip_options_rcu *inet_opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
153 if (addr_len < sizeof(struct sockaddr_in))
154 return -EINVAL;
155
156 if (usin->sin_family != AF_INET)
157 return -EAFNOSUPPORT;
158
159 nexthop = daddr = usin->sin_addr.s_addr;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000160 inet_opt = rcu_dereference_protected(inet->inet_opt,
161 sock_owned_by_user(sk));
162 if (inet_opt && inet_opt->opt.srr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 if (!daddr)
164 return -EINVAL;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000165 nexthop = inet_opt->opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 }
167
David S. Millerdca8b082011-02-24 13:38:12 -0800168 orig_sport = inet->inet_sport;
169 orig_dport = usin->sin_port;
David S. Millerda905bd2011-05-06 16:11:19 -0700170 fl4 = &inet->cork.fl.u.ip4;
171 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
David S. Millerb23dd4f2011-03-02 14:31:35 -0800172 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
173 IPPROTO_TCP,
Steffen Klassert0e0d44a2013-08-28 08:04:14 +0200174 orig_sport, orig_dport, sk);
David S. Millerb23dd4f2011-03-02 14:31:35 -0800175 if (IS_ERR(rt)) {
176 err = PTR_ERR(rt);
177 if (err == -ENETUNREACH)
Eric Dumazetf1d8cba2013-11-28 09:51:22 -0800178 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
David S. Millerb23dd4f2011-03-02 14:31:35 -0800179 return err;
Wei Dong584bdf82007-05-31 22:49:28 -0700180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
182 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
183 ip_rt_put(rt);
184 return -ENETUNREACH;
185 }
186
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000187 if (!inet_opt || !inet_opt->opt.srr)
David S. Millerda905bd2011-05-06 16:11:19 -0700188 daddr = fl4->daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000190 if (!inet->inet_saddr)
David S. Millerda905bd2011-05-06 16:11:19 -0700191 inet->inet_saddr = fl4->saddr;
Eric Dumazetd1e559d2015-03-18 14:05:35 -0700192 sk_rcv_saddr_set(sk, inet->inet_saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000194 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 /* Reset inherited state */
196 tp->rx_opt.ts_recent = 0;
197 tp->rx_opt.ts_recent_stamp = 0;
Pavel Emelyanovee995282012-04-19 03:40:39 +0000198 if (likely(!tp->repair))
199 tp->write_seq = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
201
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700202 if (tcp_death_row.sysctl_tw_recycle &&
David S. Miller81166dd2012-07-10 03:14:24 -0700203 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
204 tcp_fetch_timewait_stamp(sk, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000206 inet->inet_dport = usin->sin_port;
Eric Dumazetd1e559d2015-03-18 14:05:35 -0700207 sk_daddr_set(sk, daddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800209 inet_csk(sk)->icsk_ext_hdr_len = 0;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000210 if (inet_opt)
211 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
William Allen Simpsonbee7ca92009-11-10 09:51:18 +0000213 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
215 /* Socket identity is still unknown (sport may be zero).
216 * However we set state to SYN-SENT and not releasing socket
217 * lock select source port, enter ourselves into the hash tables and
218 * complete initialization after this.
219 */
220 tcp_set_state(sk, TCP_SYN_SENT);
Arnaldo Carvalho de Meloa7f5e7f2005-12-13 23:25:31 -0800221 err = inet_hash_connect(&tcp_death_row, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 if (err)
223 goto failure;
224
Tom Herbert877d1f62015-07-28 16:02:05 -0700225 sk_set_txhash(sk);
Sathya Perla9e7ceb02014-10-22 21:42:01 +0530226
David S. Millerda905bd2011-05-06 16:11:19 -0700227 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
David S. Millerb23dd4f2011-03-02 14:31:35 -0800228 inet->inet_sport, inet->inet_dport, sk);
229 if (IS_ERR(rt)) {
230 err = PTR_ERR(rt);
231 rt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 goto failure;
David S. Millerb23dd4f2011-03-02 14:31:35 -0800233 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 /* OK, now commit destination to socket. */
Herbert Xubcd76112006-06-30 13:36:35 -0700235 sk->sk_gso_type = SKB_GSO_TCPV4;
Changli Gaod8d1f302010-06-10 23:31:35 -0700236 sk_setup_caps(sk, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Pavel Emelyanovee995282012-04-19 03:40:39 +0000238 if (!tp->write_seq && likely(!tp->repair))
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000239 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
240 inet->inet_daddr,
241 inet->inet_sport,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 usin->sin_port);
243
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000244 inet->inet_id = tp->write_seq ^ jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Andrey Vagin2b916472012-11-22 01:13:58 +0000246 err = tcp_connect(sk);
Pavel Emelyanovee995282012-04-19 03:40:39 +0000247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 rt = NULL;
249 if (err)
250 goto failure;
251
252 return 0;
253
254failure:
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200255 /*
256 * This unhashes the socket and releases the local port,
257 * if necessary.
258 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 tcp_set_state(sk, TCP_CLOSE);
260 ip_rt_put(rt);
261 sk->sk_route_caps = 0;
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000262 inet->inet_dport = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return err;
264}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000265EXPORT_SYMBOL(tcp_v4_connect);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267/*
Eric Dumazet563d34d2012-07-23 09:48:52 +0200268 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
269 * It can be called through tcp_release_cb() if socket was owned by user
270 * at the time tcp_v4_err() was called to handle ICMP message.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 */
Neal Cardwell4fab9072014-08-14 12:40:05 -0400272void tcp_v4_mtu_reduced(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 struct dst_entry *dst;
275 struct inet_sock *inet = inet_sk(sk);
Eric Dumazet563d34d2012-07-23 09:48:52 +0200276 u32 mtu = tcp_sk(sk)->mtu_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
David S. Miller80d0a692012-07-16 03:28:06 -0700278 dst = inet_csk_update_pmtu(sk, mtu);
279 if (!dst)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 return;
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 /* Something is about to be wrong... Remember soft error
283 * for the case, if this connection will not able to recover.
284 */
285 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
286 sk->sk_err_soft = EMSGSIZE;
287
288 mtu = dst_mtu(dst);
289
290 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +0100291 ip_sk_accept_pmtu(sk) &&
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -0800292 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 tcp_sync_mss(sk, mtu);
294
295 /* Resend the TCP packet because it's
296 * clear that the old packet has been
297 * dropped. This is the new "fast" path mtu
298 * discovery.
299 */
300 tcp_simple_retransmit(sk);
301 } /* else let the usual retransmit timer handle it */
302}
Neal Cardwell4fab9072014-08-14 12:40:05 -0400303EXPORT_SYMBOL(tcp_v4_mtu_reduced);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
David S. Miller55be7a92012-07-11 21:27:49 -0700305static void do_redirect(struct sk_buff *skb, struct sock *sk)
306{
307 struct dst_entry *dst = __sk_dst_check(sk, 0);
308
David S. Miller1ed5c482012-07-12 00:41:25 -0700309 if (dst)
David S. Miller6700c272012-07-17 03:29:28 -0700310 dst->ops->redirect(dst, sk, skb);
David S. Miller55be7a92012-07-11 21:27:49 -0700311}
312
Eric Dumazet26e37362015-03-22 10:22:22 -0700313
314/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
315void tcp_req_err(struct sock *sk, u32 seq)
316{
317 struct request_sock *req = inet_reqsk(sk);
318 struct net *net = sock_net(sk);
319
320 /* ICMPs are not backlogged, hence we cannot get
321 * an established socket here.
322 */
323 WARN_ON(req->sk);
324
325 if (seq != tcp_rsk(req)->snt_isn) {
326 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
327 } else {
328 /*
329 * Still in SYN_RECV, just remove it silently.
330 * There is no good way to pass the error to the newly
331 * created socket, and POSIX does not want network
332 * errors returned from accept().
333 */
Fan Duc6973662015-03-23 15:00:41 -0700334 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
Eric Dumazetef84d8c2015-10-14 11:16:26 -0700335 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS);
Eric Dumazet26e37362015-03-22 10:22:22 -0700336 }
Eric Dumazetef84d8c2015-10-14 11:16:26 -0700337 reqsk_put(req);
Eric Dumazet26e37362015-03-22 10:22:22 -0700338}
339EXPORT_SYMBOL(tcp_req_err);
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341/*
342 * This routine is called by the ICMP module when it gets some
343 * sort of error condition. If err < 0 then the socket should
344 * be closed and the error returned to the user. If err > 0
345 * it's just the icmp type << 8 | icmp code. After adjustment
346 * header points to the first 8 bytes of the tcp header. We need
347 * to find the appropriate port.
348 *
349 * The locking strategy used here is very "optimistic". When
350 * someone else accesses the socket the ICMP is just dropped
351 * and for some paths there is no check at all.
352 * A more general error queue to queue errors for later handling
353 * is probably better.
354 *
355 */
356
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000357void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
Eric Dumazetb71d1d42011-04-22 04:53:02 +0000359 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000360 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000361 struct inet_connection_sock *icsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 struct tcp_sock *tp;
363 struct inet_sock *inet;
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000364 const int type = icmp_hdr(icmp_skb)->type;
365 const int code = icmp_hdr(icmp_skb)->code;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 struct sock *sk;
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000367 struct sk_buff *skb;
Yuchung Cheng0a672f72014-05-11 20:22:12 -0700368 struct request_sock *fastopen;
369 __u32 seq, snd_una;
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000370 __u32 remaining;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 int err;
Damian Lukowski4d1a2d92009-08-26 00:16:27 +0000372 struct net *net = dev_net(icmp_skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Eric Dumazet26e37362015-03-22 10:22:22 -0700374 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
375 th->dest, iph->saddr, ntohs(th->source),
376 inet_iif(icmp_skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 if (!sk) {
Pavel Emelyanovdcfc23c2008-07-14 23:03:00 -0700378 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 return;
380 }
381 if (sk->sk_state == TCP_TIME_WAIT) {
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -0700382 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 return;
384 }
Eric Dumazet26e37362015-03-22 10:22:22 -0700385 seq = ntohl(th->seq);
386 if (sk->sk_state == TCP_NEW_SYN_RECV)
387 return tcp_req_err(sk, seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389 bh_lock_sock(sk);
390 /* If too many ICMPs get dropped on busy
391 * servers this needs to be solved differently.
Eric Dumazet563d34d2012-07-23 09:48:52 +0200392 * We do take care of PMTU discovery (RFC1191) special case :
393 * we can receive locally generated ICMP messages while socket is held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 */
Eric Dumazetb74aa932013-01-19 16:10:37 +0000395 if (sock_owned_by_user(sk)) {
396 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
397 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 if (sk->sk_state == TCP_CLOSE)
400 goto out;
401
stephen hemminger97e3ecd12010-03-18 11:27:32 +0000402 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
403 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
404 goto out;
405 }
406
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000407 icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 tp = tcp_sk(sk);
Yuchung Cheng0a672f72014-05-11 20:22:12 -0700409 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
410 fastopen = tp->fastopen_rsk;
411 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 if (sk->sk_state != TCP_LISTEN &&
Yuchung Cheng0a672f72014-05-11 20:22:12 -0700413 !between(seq, snd_una, tp->snd_nxt)) {
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700414 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 goto out;
416 }
417
418 switch (type) {
David S. Miller55be7a92012-07-11 21:27:49 -0700419 case ICMP_REDIRECT:
420 do_redirect(icmp_skb, sk);
421 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 case ICMP_SOURCE_QUENCH:
423 /* Just silently ignore these. */
424 goto out;
425 case ICMP_PARAMETERPROB:
426 err = EPROTO;
427 break;
428 case ICMP_DEST_UNREACH:
429 if (code > NR_ICMP_UNREACH)
430 goto out;
431
432 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
Eric Dumazet0d4f0602013-03-18 07:01:28 +0000433 /* We are not interested in TCP_LISTEN and open_requests
434 * (SYN-ACKs send out by Linux are always <576bytes so
435 * they should go through unfragmented).
436 */
437 if (sk->sk_state == TCP_LISTEN)
438 goto out;
439
Eric Dumazet563d34d2012-07-23 09:48:52 +0200440 tp->mtu_info = info;
Eric Dumazet144d56e2012-08-20 00:22:46 +0000441 if (!sock_owned_by_user(sk)) {
Eric Dumazet563d34d2012-07-23 09:48:52 +0200442 tcp_v4_mtu_reduced(sk);
Eric Dumazet144d56e2012-08-20 00:22:46 +0000443 } else {
444 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
445 sock_hold(sk);
446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 goto out;
448 }
449
450 err = icmp_err_convert[code].errno;
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000451 /* check if icmp_skb allows revert of backoff
452 * (see draft-zimmermann-tcp-lcd) */
453 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
454 break;
455 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
Yuchung Cheng0a672f72014-05-11 20:22:12 -0700456 !icsk->icsk_backoff || fastopen)
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000457 break;
458
David S. Miller8f49c272010-11-12 13:35:00 -0800459 if (sock_owned_by_user(sk))
460 break;
461
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000462 icsk->icsk_backoff--;
Eric Dumazetfcdd1cf2014-09-22 13:19:44 -0700463 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
464 TCP_TIMEOUT_INIT;
465 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000466
467 skb = tcp_write_queue_head(sk);
468 BUG_ON(!skb);
469
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700470 remaining = icsk->icsk_rto -
471 min(icsk->icsk_rto,
472 tcp_time_stamp - tcp_skb_timestamp(skb));
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000473
474 if (remaining) {
475 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
476 remaining, TCP_RTO_MAX);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000477 } else {
478 /* RTO revert clocked out retransmission.
479 * Will retransmit now */
480 tcp_retransmit_timer(sk);
481 }
482
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 break;
484 case ICMP_TIME_EXCEEDED:
485 err = EHOSTUNREACH;
486 break;
487 default:
488 goto out;
489 }
490
491 switch (sk->sk_state) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 case TCP_SYN_SENT:
Yuchung Cheng0a672f72014-05-11 20:22:12 -0700493 case TCP_SYN_RECV:
494 /* Only in fast or simultaneous open. If a fast open socket is
495 * is already accepted it is treated as a connected one below.
496 */
Ian Morris51456b22015-04-03 09:17:26 +0100497 if (fastopen && !fastopen->sk)
Yuchung Cheng0a672f72014-05-11 20:22:12 -0700498 break;
499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (!sock_owned_by_user(sk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 sk->sk_err = err;
502
503 sk->sk_error_report(sk);
504
505 tcp_done(sk);
506 } else {
507 sk->sk_err_soft = err;
508 }
509 goto out;
510 }
511
512 /* If we've already connected we will keep trying
513 * until we time out, or the user gives up.
514 *
515 * rfc1122 4.2.3.9 allows to consider as hard errors
516 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
517 * but it is obsoleted by pmtu discovery).
518 *
519 * Note, that in modern internet, where routing is unreliable
520 * and in each dark corner broken firewalls sit, sending random
521 * errors ordered by their masters even this two messages finally lose
522 * their original sense (even Linux sends invalid PORT_UNREACHs)
523 *
524 * Now we are in compliance with RFCs.
525 * --ANK (980905)
526 */
527
528 inet = inet_sk(sk);
529 if (!sock_owned_by_user(sk) && inet->recverr) {
530 sk->sk_err = err;
531 sk->sk_error_report(sk);
532 } else { /* Only an error on timeout */
533 sk->sk_err_soft = err;
534 }
535
536out:
537 bh_unlock_sock(sk);
538 sock_put(sk);
539}
540
Daniel Borkmann28850dc2013-06-07 05:11:46 +0000541void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700543 struct tcphdr *th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Patrick McHardy84fa7932006-08-29 16:44:56 -0700545 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Herbert Xu419f9f82010-04-11 02:15:53 +0000546 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
Herbert Xu663ead32007-04-09 11:59:07 -0700547 skb->csum_start = skb_transport_header(skb) - skb->head;
Al Viroff1dcad2006-11-20 18:07:29 -0800548 skb->csum_offset = offsetof(struct tcphdr, check);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 } else {
Herbert Xu419f9f82010-04-11 02:15:53 +0000550 th->check = tcp_v4_check(skb->len, saddr, daddr,
Joe Perches07f07572008-11-19 15:44:53 -0800551 csum_partial(th,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 th->doff << 2,
553 skb->csum));
554 }
555}
556
Herbert Xu419f9f82010-04-11 02:15:53 +0000557/* This routine computes an IPv4 TCP checksum. */
Herbert Xubb296242010-04-11 02:15:55 +0000558void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
Herbert Xu419f9f82010-04-11 02:15:53 +0000559{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400560 const struct inet_sock *inet = inet_sk(sk);
Herbert Xu419f9f82010-04-11 02:15:53 +0000561
562 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
563}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000564EXPORT_SYMBOL(tcp_v4_send_check);
Herbert Xu419f9f82010-04-11 02:15:53 +0000565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566/*
567 * This routine will send an RST to the other tcp.
568 *
569 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
570 * for reset.
571 * Answer: if a packet caused RST, it is not for a socket
572 * existing in our system, if it is matched to a socket,
573 * it is just duplicate segment or bug in other side's TCP.
574 * So that we build reply only basing on parameters
575 * arrived with segment.
576 * Exception: precedence violation. We do not implement it in any case.
577 */
578
Eric Dumazeta00e7442015-09-29 07:42:39 -0700579static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400581 const struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800582 struct {
583 struct tcphdr th;
584#ifdef CONFIG_TCP_MD5SIG
Al Viro714e85b2006-11-14 20:51:49 -0800585 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800586#endif
587 } rep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 struct ip_reply_arg arg;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800589#ifdef CONFIG_TCP_MD5SIG
590 struct tcp_md5sig_key *key;
Shawn Lu658ddaa2012-01-31 22:35:48 +0000591 const __u8 *hash_location = NULL;
592 unsigned char newhash[16];
593 int genhash;
594 struct sock *sk1 = NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800595#endif
Pavel Emelyanova86b1e32008-07-16 20:20:58 -0700596 struct net *net;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
598 /* Never send a reset in response to a reset. */
599 if (th->rst)
600 return;
601
Eric Dumazetc3658e82014-11-25 07:40:04 -0800602 /* If sk not NULL, it means we did a successful lookup and incoming
603 * route had to be correct. prequeue might have dropped our dst.
604 */
605 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 return;
607
608 /* Swap the send and the receive. */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800609 memset(&rep, 0, sizeof(rep));
610 rep.th.dest = th->source;
611 rep.th.source = th->dest;
612 rep.th.doff = sizeof(struct tcphdr) / 4;
613 rep.th.rst = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615 if (th->ack) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800616 rep.th.seq = th->ack_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 } else {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800618 rep.th.ack = 1;
619 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
620 skb->len - (th->doff << 2));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 }
622
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200623 memset(&arg, 0, sizeof(arg));
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800624 arg.iov[0].iov_base = (unsigned char *)&rep;
625 arg.iov[0].iov_len = sizeof(rep.th);
626
Eric Dumazet0f85fea2014-12-09 09:56:08 -0800627 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800628#ifdef CONFIG_TCP_MD5SIG
Shawn Lu658ddaa2012-01-31 22:35:48 +0000629 hash_location = tcp_parse_md5sig_option(th);
630 if (!sk && hash_location) {
631 /*
632 * active side is lost. Try to find listening socket through
633 * source port, and then find md5 key through listening socket.
634 * we are not loose security here:
635 * Incoming packet is checked with md5 hash with finding key,
636 * no RST generated if md5 hash doesn't match.
637 */
Eric Dumazet0f85fea2014-12-09 09:56:08 -0800638 sk1 = __inet_lookup_listener(net,
Tom Herbertda5e3632013-01-22 09:50:24 +0000639 &tcp_hashinfo, ip_hdr(skb)->saddr,
640 th->source, ip_hdr(skb)->daddr,
Shawn Lu658ddaa2012-01-31 22:35:48 +0000641 ntohs(th->source), inet_iif(skb));
642 /* don't send rst if it can't find key */
643 if (!sk1)
644 return;
645 rcu_read_lock();
646 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
647 &ip_hdr(skb)->saddr, AF_INET);
648 if (!key)
649 goto release_sk1;
650
Eric Dumazet39f8e582015-03-24 15:58:55 -0700651 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
Shawn Lu658ddaa2012-01-31 22:35:48 +0000652 if (genhash || memcmp(hash_location, newhash, 16) != 0)
653 goto release_sk1;
654 } else {
655 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
656 &ip_hdr(skb)->saddr,
657 AF_INET) : NULL;
658 }
659
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800660 if (key) {
661 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
662 (TCPOPT_NOP << 16) |
663 (TCPOPT_MD5SIG << 8) |
664 TCPOLEN_MD5SIG);
665 /* Update length and the length the header thinks exists */
666 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
667 rep.th.doff = arg.iov[0].iov_len / 4;
668
Adam Langley49a72df2008-07-19 00:01:42 -0700669 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
Ilpo Järvinen78e645cb2008-10-09 14:37:47 -0700670 key, ip_hdr(skb)->saddr,
671 ip_hdr(skb)->daddr, &rep.th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800672 }
673#endif
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700674 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
675 ip_hdr(skb)->saddr, /* XXX */
Ilpo Järvinen52cd5752008-10-08 11:34:06 -0700676 arg.iov[0].iov_len, IPPROTO_TCP, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700678 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
Shawn Lue2446ea2012-02-04 12:38:09 +0000679 /* When socket is gone, all binding information is lost.
Alexey Kuznetsov4c675252012-10-12 04:34:17 +0000680 * routing might fail in this case. No choice here, if we choose to force
681 * input interface, we will misroute in case of asymmetric route.
Shawn Lue2446ea2012-02-04 12:38:09 +0000682 */
Alexey Kuznetsov4c675252012-10-12 04:34:17 +0000683 if (sk)
684 arg.bound_dev_if = sk->sk_bound_dev_if;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Eric Dumazet66b13d92011-10-24 03:06:21 -0400686 arg.tos = ip_hdr(skb)->tos;
Eric Dumazetbdbbb852015-01-29 21:35:05 -0800687 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
688 skb, &TCP_SKB_CB(skb)->header.h4.opt,
Eric Dumazet24a2d432014-09-27 09:50:55 -0700689 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
690 &arg, arg.iov[0].iov_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700692 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
693 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
Shawn Lu658ddaa2012-01-31 22:35:48 +0000694
695#ifdef CONFIG_TCP_MD5SIG
696release_sk1:
697 if (sk1) {
698 rcu_read_unlock();
699 sock_put(sk1);
700 }
701#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
703
704/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
705 outside socket context is ugly, certainly. What can I do?
706 */
707
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900708static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
Andrey Vaginee684b62013-02-11 05:50:19 +0000709 u32 win, u32 tsval, u32 tsecr, int oif,
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700710 struct tcp_md5sig_key *key,
Eric Dumazet66b13d92011-10-24 03:06:21 -0400711 int reply_flags, u8 tos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400713 const struct tcphdr *th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 struct {
715 struct tcphdr th;
Al Viro714e85b2006-11-14 20:51:49 -0800716 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800717#ifdef CONFIG_TCP_MD5SIG
Al Viro714e85b2006-11-14 20:51:49 -0800718 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800719#endif
720 ];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 } rep;
722 struct ip_reply_arg arg;
Eric Dumazetadf30902009-06-02 05:19:30 +0000723 struct net *net = dev_net(skb_dst(skb)->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
725 memset(&rep.th, 0, sizeof(struct tcphdr));
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200726 memset(&arg, 0, sizeof(arg));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728 arg.iov[0].iov_base = (unsigned char *)&rep;
729 arg.iov[0].iov_len = sizeof(rep.th);
Andrey Vaginee684b62013-02-11 05:50:19 +0000730 if (tsecr) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800731 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
732 (TCPOPT_TIMESTAMP << 8) |
733 TCPOLEN_TIMESTAMP);
Andrey Vaginee684b62013-02-11 05:50:19 +0000734 rep.opt[1] = htonl(tsval);
735 rep.opt[2] = htonl(tsecr);
Craig Schlentercb48cfe2007-01-09 00:11:15 -0800736 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
738
739 /* Swap the send and the receive. */
740 rep.th.dest = th->source;
741 rep.th.source = th->dest;
742 rep.th.doff = arg.iov[0].iov_len / 4;
743 rep.th.seq = htonl(seq);
744 rep.th.ack_seq = htonl(ack);
745 rep.th.ack = 1;
746 rep.th.window = htons(win);
747
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800748#ifdef CONFIG_TCP_MD5SIG
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800749 if (key) {
Andrey Vaginee684b62013-02-11 05:50:19 +0000750 int offset = (tsecr) ? 3 : 0;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800751
752 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
753 (TCPOPT_NOP << 16) |
754 (TCPOPT_MD5SIG << 8) |
755 TCPOLEN_MD5SIG);
756 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
757 rep.th.doff = arg.iov[0].iov_len/4;
758
Adam Langley49a72df2008-07-19 00:01:42 -0700759 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
Adam Langley90b7e112008-07-31 20:49:48 -0700760 key, ip_hdr(skb)->saddr,
761 ip_hdr(skb)->daddr, &rep.th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800762 }
763#endif
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700764 arg.flags = reply_flags;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700765 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
766 ip_hdr(skb)->saddr, /* XXX */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 arg.iov[0].iov_len, IPPROTO_TCP, 0);
768 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900769 if (oif)
770 arg.bound_dev_if = oif;
Eric Dumazet66b13d92011-10-24 03:06:21 -0400771 arg.tos = tos;
Eric Dumazetbdbbb852015-01-29 21:35:05 -0800772 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
773 skb, &TCP_SKB_CB(skb)->header.h4.opt,
Eric Dumazet24a2d432014-09-27 09:50:55 -0700774 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
775 &arg, arg.iov[0].iov_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700777 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
780static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
781{
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700782 struct inet_timewait_sock *tw = inet_twsk(sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800783 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900785 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200786 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
Andrey Vaginee684b62013-02-11 05:50:19 +0000787 tcp_time_stamp + tcptw->tw_ts_offset,
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900788 tcptw->tw_ts_recent,
789 tw->tw_bound_dev_if,
KOVACS Krisztian88ef4a52008-10-01 07:41:00 -0700790 tcp_twsk_md5_key(tcptw),
Eric Dumazet66b13d92011-10-24 03:06:21 -0400791 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
792 tw->tw_tos
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900793 );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700795 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
797
Eric Dumazeta00e7442015-09-29 07:42:39 -0700798static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200799 struct request_sock *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
Jerry Chu168a8f52012-08-31 12:29:13 +0000801 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
802 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
803 */
804 tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
805 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
Eric Dumazeted53d0a2015-10-08 19:33:23 -0700806 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
Andrey Vaginee684b62013-02-11 05:50:19 +0000807 tcp_time_stamp,
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +0900808 req->ts_recent,
809 0,
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000810 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
811 AF_INET),
Eric Dumazet66b13d92011-10-24 03:06:21 -0400812 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
813 ip_hdr(skb)->tos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
815
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816/*
Kris Katterjohn9bf1d832008-02-17 22:29:19 -0800817 * Send a SYN-ACK after having received a SYN.
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700818 * This still operates on a request_sock only, not on a big
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 * socket.
820 */
Eric Dumazet0f935dbe2015-09-25 07:39:21 -0700821static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
Octavian Purdilad6274bd2014-06-25 17:09:58 +0300822 struct flowi *fl,
Octavian Purdila72659ec2010-01-17 19:09:39 -0800823 struct request_sock *req,
Eric Dumazetca6fb062015-10-02 11:43:35 -0700824 struct tcp_fastopen_cookie *foc,
825 bool attach_req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700827 const struct inet_request_sock *ireq = inet_rsk(req);
David S. Miller6bd023f2011-05-18 18:32:03 -0400828 struct flowi4 fl4;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 int err = -1;
Weilong Chend41db5a2013-12-23 14:37:28 +0800830 struct sk_buff *skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
832 /* First, grab a route. */
David S. Millerba3f7f02012-07-17 14:02:46 -0700833 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
Denis V. Lunevfd80eb92008-02-29 11:43:03 -0800834 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Eric Dumazetca6fb062015-10-02 11:43:35 -0700836 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
838 if (skb) {
Eric Dumazet634fb9792013-10-09 15:21:29 -0700839 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Eric Dumazet634fb9792013-10-09 15:21:29 -0700841 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
842 ireq->ir_rmt_addr,
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700843 ireq->opt);
Gerrit Renkerb9df3cb2006-11-14 11:21:36 -0200844 err = net_xmit_eval(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 }
846
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 return err;
848}
849
850/*
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700851 * IPv4 request_sock destructor.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 */
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700853static void tcp_v4_reqsk_destructor(struct request_sock *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
Jesper Juhla51482b2005-11-08 09:41:34 -0800855 kfree(inet_rsk(req)->opt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800859#ifdef CONFIG_TCP_MD5SIG
860/*
861 * RFC2385 MD5 checksumming requires a mapping of
862 * IP address->MD5 Key.
863 * We need to maintain these in the sk structure.
864 */
865
866/* Find the Key structure for an address. */
Eric Dumazetb83e3de2015-09-25 07:39:15 -0700867struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000868 const union tcp_md5_addr *addr,
869 int family)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800870{
Eric Dumazetfd3a1542015-03-24 15:58:56 -0700871 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000872 struct tcp_md5sig_key *key;
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000873 unsigned int size = sizeof(struct in_addr);
Eric Dumazetfd3a1542015-03-24 15:58:56 -0700874 const struct tcp_md5sig_info *md5sig;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800875
Eric Dumazeta8afca02012-01-31 18:45:40 +0000876 /* caller either holds rcu_read_lock() or socket lock */
877 md5sig = rcu_dereference_check(tp->md5sig_info,
Eric Dumazetb4fb05e2012-03-07 04:45:43 +0000878 sock_owned_by_user(sk) ||
Eric Dumazetb83e3de2015-09-25 07:39:15 -0700879 lockdep_is_held((spinlock_t *)&sk->sk_lock.slock));
Eric Dumazeta8afca02012-01-31 18:45:40 +0000880 if (!md5sig)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800881 return NULL;
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000882#if IS_ENABLED(CONFIG_IPV6)
883 if (family == AF_INET6)
884 size = sizeof(struct in6_addr);
885#endif
Sasha Levinb67bfe02013-02-27 17:06:00 -0800886 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000887 if (key->family != family)
888 continue;
889 if (!memcmp(&key->addr, addr, size))
890 return key;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800891 }
892 return NULL;
893}
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000894EXPORT_SYMBOL(tcp_md5_do_lookup);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800895
Eric Dumazetb83e3de2015-09-25 07:39:15 -0700896struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -0700897 const struct sock *addr_sk)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800898{
Eric Dumazetb52e6922015-04-09 14:36:42 -0700899 const union tcp_md5_addr *addr;
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000900
Eric Dumazetb52e6922015-04-09 14:36:42 -0700901 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000902 return tcp_md5_do_lookup(sk, addr, AF_INET);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800903}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800904EXPORT_SYMBOL(tcp_v4_md5_lookup);
905
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800906/* This can be called on a newly created socket, from other files */
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000907int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
908 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800909{
910 /* Add Key to the list */
Matthias M. Dellwegb0a713e2007-10-29 20:55:27 -0700911 struct tcp_md5sig_key *key;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800912 struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000913 struct tcp_md5sig_info *md5sig;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800914
Aydin Arikc0353c72013-06-14 18:56:31 +1200915 key = tcp_md5_do_lookup(sk, addr, family);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800916 if (key) {
917 /* Pre-existing entry - just update that one. */
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000918 memcpy(key->key, newkey, newkeylen);
Matthias M. Dellwegb0a713e2007-10-29 20:55:27 -0700919 key->keylen = newkeylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000920 return 0;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800921 }
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000922
Eric Dumazeta8afca02012-01-31 18:45:40 +0000923 md5sig = rcu_dereference_protected(tp->md5sig_info,
Eric Dumazet1b8e6a02015-11-18 12:40:13 -0800924 sock_owned_by_user(sk) ||
925 lockdep_is_held(&sk->sk_lock.slock));
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000926 if (!md5sig) {
927 md5sig = kmalloc(sizeof(*md5sig), gfp);
928 if (!md5sig)
929 return -ENOMEM;
930
931 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
932 INIT_HLIST_HEAD(&md5sig->head);
Eric Dumazeta8afca02012-01-31 18:45:40 +0000933 rcu_assign_pointer(tp->md5sig_info, md5sig);
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000934 }
935
Eric Dumazet5f3d9cb2012-01-31 10:56:48 +0000936 key = sock_kmalloc(sk, sizeof(*key), gfp);
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000937 if (!key)
938 return -ENOMEM;
Eric Dumazet71cea172013-05-20 06:52:26 +0000939 if (!tcp_alloc_md5sig_pool()) {
Eric Dumazet5f3d9cb2012-01-31 10:56:48 +0000940 sock_kfree_s(sk, key, sizeof(*key));
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000941 return -ENOMEM;
942 }
943
944 memcpy(key->key, newkey, newkeylen);
945 key->keylen = newkeylen;
946 key->family = family;
947 memcpy(&key->addr, addr,
948 (family == AF_INET6) ? sizeof(struct in6_addr) :
949 sizeof(struct in_addr));
950 hlist_add_head_rcu(&key->node, &md5sig->head);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800951 return 0;
952}
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000953EXPORT_SYMBOL(tcp_md5_do_add);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800954
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000955int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800956{
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000957 struct tcp_md5sig_key *key;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800958
Aydin Arikc0353c72013-06-14 18:56:31 +1200959 key = tcp_md5_do_lookup(sk, addr, family);
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000960 if (!key)
961 return -ENOENT;
962 hlist_del_rcu(&key->node);
Eric Dumazet5f3d9cb2012-01-31 10:56:48 +0000963 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000964 kfree_rcu(key, rcu);
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000965 return 0;
966}
967EXPORT_SYMBOL(tcp_md5_do_del);
968
stephen hemmingere0683e702012-10-26 14:31:40 +0000969static void tcp_clear_md5_list(struct sock *sk)
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000970{
971 struct tcp_sock *tp = tcp_sk(sk);
972 struct tcp_md5sig_key *key;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800973 struct hlist_node *n;
Eric Dumazeta8afca02012-01-31 18:45:40 +0000974 struct tcp_md5sig_info *md5sig;
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000975
Eric Dumazeta8afca02012-01-31 18:45:40 +0000976 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
977
Sasha Levinb67bfe02013-02-27 17:06:00 -0800978 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000979 hlist_del_rcu(&key->node);
Eric Dumazet5f3d9cb2012-01-31 10:56:48 +0000980 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000981 kfree_rcu(key, rcu);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800982 }
983}
984
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200985static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
986 int optlen)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800987{
988 struct tcp_md5sig cmd;
989 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800990
991 if (optlen < sizeof(cmd))
992 return -EINVAL;
993
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -0200994 if (copy_from_user(&cmd, optval, sizeof(cmd)))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800995 return -EFAULT;
996
997 if (sin->sin_family != AF_INET)
998 return -EINVAL;
999
Dmitry Popov64a124e2014-08-03 22:45:19 +04001000 if (!cmd.tcpm_keylen)
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001001 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1002 AF_INET);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001003
1004 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1005 return -EINVAL;
1006
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001007 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1008 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1009 GFP_KERNEL);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001010}
1011
Adam Langley49a72df2008-07-19 00:01:42 -07001012static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1013 __be32 daddr, __be32 saddr, int nbytes)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001014{
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001015 struct tcp4_pseudohdr *bp;
Adam Langley49a72df2008-07-19 00:01:42 -07001016 struct scatterlist sg;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001017
1018 bp = &hp->md5_blk.ip4;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001019
1020 /*
Adam Langley49a72df2008-07-19 00:01:42 -07001021 * 1. the TCP pseudo-header (in the order: source IP address,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001022 * destination IP address, zero-padded protocol number, and
1023 * segment length)
1024 */
1025 bp->saddr = saddr;
1026 bp->daddr = daddr;
1027 bp->pad = 0;
YOSHIFUJI Hideaki076fb722008-04-17 12:48:12 +09001028 bp->protocol = IPPROTO_TCP;
Adam Langley49a72df2008-07-19 00:01:42 -07001029 bp->len = cpu_to_be16(nbytes);
David S. Millerc7da57a2007-10-26 00:41:21 -07001030
Adam Langley49a72df2008-07-19 00:01:42 -07001031 sg_init_one(&sg, bp, sizeof(*bp));
1032 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1033}
1034
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001035static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
Eric Dumazet318cf7a2011-10-24 02:46:04 -04001036 __be32 daddr, __be32 saddr, const struct tcphdr *th)
Adam Langley49a72df2008-07-19 00:01:42 -07001037{
1038 struct tcp_md5sig_pool *hp;
1039 struct hash_desc *desc;
1040
1041 hp = tcp_get_md5sig_pool();
1042 if (!hp)
1043 goto clear_hash_noput;
1044 desc = &hp->md5_desc;
1045
1046 if (crypto_hash_init(desc))
1047 goto clear_hash;
1048 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1049 goto clear_hash;
1050 if (tcp_md5_hash_header(hp, th))
1051 goto clear_hash;
1052 if (tcp_md5_hash_key(hp, key))
1053 goto clear_hash;
1054 if (crypto_hash_final(desc, md5_hash))
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001055 goto clear_hash;
1056
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001057 tcp_put_md5sig_pool();
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001058 return 0;
Adam Langley49a72df2008-07-19 00:01:42 -07001059
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001060clear_hash:
1061 tcp_put_md5sig_pool();
1062clear_hash_noput:
1063 memset(md5_hash, 0, 16);
Adam Langley49a72df2008-07-19 00:01:42 -07001064 return 1;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001065}
1066
Eric Dumazet39f8e582015-03-24 15:58:55 -07001067int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1068 const struct sock *sk,
Eric Dumazet318cf7a2011-10-24 02:46:04 -04001069 const struct sk_buff *skb)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001070{
Adam Langley49a72df2008-07-19 00:01:42 -07001071 struct tcp_md5sig_pool *hp;
1072 struct hash_desc *desc;
Eric Dumazet318cf7a2011-10-24 02:46:04 -04001073 const struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001074 __be32 saddr, daddr;
1075
Eric Dumazet39f8e582015-03-24 15:58:55 -07001076 if (sk) { /* valid for establish/request sockets */
1077 saddr = sk->sk_rcv_saddr;
1078 daddr = sk->sk_daddr;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001079 } else {
Adam Langley49a72df2008-07-19 00:01:42 -07001080 const struct iphdr *iph = ip_hdr(skb);
1081 saddr = iph->saddr;
1082 daddr = iph->daddr;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001083 }
Adam Langley49a72df2008-07-19 00:01:42 -07001084
1085 hp = tcp_get_md5sig_pool();
1086 if (!hp)
1087 goto clear_hash_noput;
1088 desc = &hp->md5_desc;
1089
1090 if (crypto_hash_init(desc))
1091 goto clear_hash;
1092
1093 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1094 goto clear_hash;
1095 if (tcp_md5_hash_header(hp, th))
1096 goto clear_hash;
1097 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1098 goto clear_hash;
1099 if (tcp_md5_hash_key(hp, key))
1100 goto clear_hash;
1101 if (crypto_hash_final(desc, md5_hash))
1102 goto clear_hash;
1103
1104 tcp_put_md5sig_pool();
1105 return 0;
1106
1107clear_hash:
1108 tcp_put_md5sig_pool();
1109clear_hash_noput:
1110 memset(md5_hash, 0, 16);
1111 return 1;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001112}
Adam Langley49a72df2008-07-19 00:01:42 -07001113EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001114
Eric Dumazetba8e2752015-10-02 11:43:28 -07001115#endif
1116
Eric Dumazetff74e232015-03-24 15:58:54 -07001117/* Called with rcu_read_lock() */
Eric Dumazetba8e2752015-10-02 11:43:28 -07001118static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
Eric Dumazetff74e232015-03-24 15:58:54 -07001119 const struct sk_buff *skb)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001120{
Eric Dumazetba8e2752015-10-02 11:43:28 -07001121#ifdef CONFIG_TCP_MD5SIG
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001122 /*
1123 * This gets called for each TCP segment that arrives
1124 * so we want to be efficient.
1125 * We have 3 drop cases:
1126 * o No MD5 hash and one expected.
1127 * o MD5 hash and we're not expecting one.
1128 * o MD5 hash and its wrong.
1129 */
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001130 const __u8 *hash_location = NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001131 struct tcp_md5sig_key *hash_expected;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001132 const struct iphdr *iph = ip_hdr(skb);
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001133 const struct tcphdr *th = tcp_hdr(skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001134 int genhash;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001135 unsigned char newhash[16];
1136
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001137 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1138 AF_INET);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +09001139 hash_location = tcp_parse_md5sig_option(th);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001140
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001141 /* We've parsed the options - do we have a hash? */
1142 if (!hash_expected && !hash_location)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001143 return false;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001144
1145 if (hash_expected && !hash_location) {
David S. Miller785957d2008-07-30 03:03:15 -07001146 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001147 return true;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001148 }
1149
1150 if (!hash_expected && hash_location) {
David S. Miller785957d2008-07-30 03:03:15 -07001151 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001152 return true;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001153 }
1154
1155 /* Okay, so this is hash_expected and hash_location -
1156 * so we need to calculate the checksum.
1157 */
Adam Langley49a72df2008-07-19 00:01:42 -07001158 genhash = tcp_v4_md5_hash_skb(newhash,
1159 hash_expected,
Eric Dumazet39f8e582015-03-24 15:58:55 -07001160 NULL, skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001161
1162 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
Joe Perchese87cc472012-05-13 21:56:26 +00001163 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1164 &iph->saddr, ntohs(th->source),
1165 &iph->daddr, ntohs(th->dest),
1166 genhash ? " tcp_v4_calc_md5_hash failed"
1167 : "");
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001168 return true;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001169 }
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001170 return false;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001171#endif
Eric Dumazetba8e2752015-10-02 11:43:28 -07001172 return false;
1173}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001174
Eric Dumazetb40cf182015-09-25 07:39:08 -07001175static void tcp_v4_init_req(struct request_sock *req,
1176 const struct sock *sk_listener,
Octavian Purdila16bea702014-06-25 17:09:53 +03001177 struct sk_buff *skb)
1178{
1179 struct inet_request_sock *ireq = inet_rsk(req);
1180
Eric Dumazet08d2cc3b2015-03-18 14:05:38 -07001181 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1182 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1183 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
Octavian Purdila16bea702014-06-25 17:09:53 +03001184 ireq->opt = tcp_v4_save_options(skb);
1185}
1186
Eric Dumazetf9646292015-09-29 07:42:50 -07001187static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1188 struct flowi *fl,
Octavian Purdilad94e0412014-06-25 17:09:55 +03001189 const struct request_sock *req,
1190 bool *strict)
1191{
1192 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1193
1194 if (strict) {
1195 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1196 *strict = true;
1197 else
1198 *strict = false;
1199 }
1200
1201 return dst;
1202}
1203
Eric Dumazet72a3eff2006-11-16 02:30:37 -08001204struct request_sock_ops tcp_request_sock_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 .family = PF_INET,
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001206 .obj_size = sizeof(struct tcp_request_sock),
Octavian Purdila5db92c92014-06-25 17:09:59 +03001207 .rtx_syn_ack = tcp_rtx_synack,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001208 .send_ack = tcp_v4_reqsk_send_ack,
1209 .destructor = tcp_v4_reqsk_destructor,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 .send_reset = tcp_v4_send_reset,
stephen hemminger688d1942014-08-29 23:32:05 -07001211 .syn_ack_timeout = tcp_syn_ack_timeout,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212};
1213
Stephen Hemmingerb2e4b3d2009-09-01 19:25:03 +00001214static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001215 .mss_clamp = TCP_MSS_DEFAULT,
Octavian Purdila16bea702014-06-25 17:09:53 +03001216#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001217 .req_md5_lookup = tcp_v4_md5_lookup,
John Dykstrae3afe7b2009-07-16 05:04:51 +00001218 .calc_md5_hash = tcp_v4_md5_hash_skb,
Andrew Mortonb6332e62006-11-30 19:16:28 -08001219#endif
Octavian Purdila16bea702014-06-25 17:09:53 +03001220 .init_req = tcp_v4_init_req,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001221#ifdef CONFIG_SYN_COOKIES
1222 .cookie_init_seq = cookie_v4_init_sequence,
1223#endif
Octavian Purdilad94e0412014-06-25 17:09:55 +03001224 .route_req = tcp_v4_route_req,
Octavian Purdila936b8bd2014-06-25 17:09:57 +03001225 .init_seq = tcp_v4_init_sequence,
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001226 .send_synack = tcp_v4_send_synack,
Octavian Purdila16bea702014-06-25 17:09:53 +03001227};
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001228
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1230{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 /* Never answer to SYNs send to broadcast or multicast */
Eric Dumazet511c3f92009-06-02 05:14:27 +00001232 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 goto drop;
1234
Octavian Purdila1fb6f152014-06-25 17:10:02 +03001235 return tcp_conn_request(&tcp_request_sock_ops,
1236 &tcp_request_sock_ipv4_ops, sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238drop:
Vijay Subramanian848bf152013-01-31 08:24:06 +00001239 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 return 0;
1241}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001242EXPORT_SYMBOL(tcp_v4_conn_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
1244
1245/*
1246 * The three way handshake has completed - we got a valid synack -
1247 * now create the new socket.
1248 */
Eric Dumazet0c271712015-09-29 07:42:48 -07001249struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001250 struct request_sock *req,
Eric Dumazet5e0724d2015-10-22 08:20:46 -07001251 struct dst_entry *dst,
1252 struct request_sock *req_unhash,
1253 bool *own_req)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001255 struct inet_request_sock *ireq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 struct inet_sock *newinet;
1257 struct tcp_sock *newtp;
1258 struct sock *newsk;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001259#ifdef CONFIG_TCP_MD5SIG
1260 struct tcp_md5sig_key *key;
1261#endif
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001262 struct ip_options_rcu *inet_opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264 if (sk_acceptq_is_full(sk))
1265 goto exit_overflow;
1266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 newsk = tcp_create_openreq_child(sk, req, skb);
1268 if (!newsk)
Balazs Scheidler093d2822010-10-21 13:06:43 +02001269 goto exit_nonewsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Herbert Xubcd76112006-06-30 13:36:35 -07001271 newsk->sk_gso_type = SKB_GSO_TCPV4;
Neal Cardwellfae6ef82012-08-19 03:30:38 +00001272 inet_sk_rx_dst_set(newsk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
1274 newtp = tcp_sk(newsk);
1275 newinet = inet_sk(newsk);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001276 ireq = inet_rsk(req);
Eric Dumazetd1e559d2015-03-18 14:05:35 -07001277 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1278 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
Eric Dumazet634fb9792013-10-09 15:21:29 -07001279 newinet->inet_saddr = ireq->ir_loc_addr;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001280 inet_opt = ireq->opt;
1281 rcu_assign_pointer(newinet->inet_opt, inet_opt);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001282 ireq->opt = NULL;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001283 newinet->mc_index = inet_iif(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001284 newinet->mc_ttl = ip_hdr(skb)->ttl;
Jiri Benc4c507d22012-02-09 09:35:49 +00001285 newinet->rcv_tos = ip_hdr(skb)->tos;
Arnaldo Carvalho de Melod83d8462005-12-13 23:26:10 -08001286 inet_csk(newsk)->icsk_ext_hdr_len = 0;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001287 if (inet_opt)
1288 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00001289 newinet->inet_id = newtp->write_seq ^ jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
Eric Dumazetdfd25ff2012-03-10 09:20:21 +00001291 if (!dst) {
1292 dst = inet_csk_route_child_sock(sk, newsk, req);
1293 if (!dst)
1294 goto put_and_exit;
1295 } else {
1296 /* syncookie case : see end of cookie_v4_check() */
1297 }
David S. Miller0e734412011-05-08 15:28:03 -07001298 sk_setup_caps(newsk, dst);
1299
Daniel Borkmann81164412015-01-05 23:57:48 +01001300 tcp_ca_openreq_child(newsk, dst);
1301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 tcp_sync_mss(newsk, dst_mtu(dst));
David S. Miller0dbaee32010-12-13 12:52:14 -08001303 newtp->advmss = dst_metric_advmss(dst);
Tom Quetchenbachf5fff5d2008-09-21 00:21:51 -07001304 if (tcp_sk(sk)->rx_opt.user_mss &&
1305 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1306 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1307
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 tcp_initialize_rcv_mss(newsk);
1309
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001310#ifdef CONFIG_TCP_MD5SIG
1311 /* Copy over the MD5 key from the original socket */
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001312 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1313 AF_INET);
Ian Morris00db4122015-04-03 09:17:27 +01001314 if (key) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001315 /*
1316 * We're using one, so create a matching key
1317 * on the newsk structure. If we fail to get
1318 * memory, then we end up not copying the key
1319 * across. Shucks.
1320 */
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001321 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1322 AF_INET, key->key, key->keylen, GFP_ATOMIC);
Eric Dumazeta4654192010-05-16 00:36:33 -07001323 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001324 }
1325#endif
1326
David S. Miller0e734412011-05-08 15:28:03 -07001327 if (__inet_inherit_port(sk, newsk) < 0)
1328 goto put_and_exit;
Eric Dumazet5e0724d2015-10-22 08:20:46 -07001329 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
Eric Dumazet805c4bc2015-11-05 11:07:13 -08001330 if (*own_req)
Eric Dumazet49a496c2015-11-05 12:50:19 -08001331 tcp_move_syn(newtp, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 return newsk;
1334
1335exit_overflow:
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001336 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
Balazs Scheidler093d2822010-10-21 13:06:43 +02001337exit_nonewsk:
1338 dst_release(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339exit:
Pavel Emelyanovde0744a2008-07-16 20:31:16 -07001340 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 return NULL;
David S. Miller0e734412011-05-08 15:28:03 -07001342put_and_exit:
Christoph Paasche337e242012-12-14 04:07:58 +00001343 inet_csk_prepare_forced_close(newsk);
1344 tcp_done(newsk);
David S. Miller0e734412011-05-08 15:28:03 -07001345 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001347EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Eric Dumazet079096f2015-10-02 11:43:32 -07001349static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351#ifdef CONFIG_SYN_COOKIES
Eric Dumazet079096f2015-10-02 11:43:32 -07001352 const struct tcphdr *th = tcp_hdr(skb);
1353
Florian Westphalaf9b4732010-06-03 00:43:44 +00001354 if (!th->syn)
Cong Wang461b74c2014-10-15 14:33:22 -07001355 sk = cookie_v4_check(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356#endif
1357 return sk;
1358}
1359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360/* The socket must have it's spinlock held when we get
Eric Dumazete994b2f2015-10-02 11:43:39 -07001361 * here, unless it is a TCP_LISTEN socket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 *
1363 * We have a potential double-lock case here, so even when
1364 * doing backlog processing we use the BH locking scheme.
1365 * This is because we cannot sleep with the original spinlock
1366 * held.
1367 */
1368int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1369{
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001370 struct sock *rsk;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
Eric Dumazet404e0a82012-07-29 23:20:37 +00001373 struct dst_entry *dst = sk->sk_rx_dst;
1374
Tom Herbertbdeab992011-08-14 19:45:55 +00001375 sock_rps_save_rxhash(sk, skb);
Eric Dumazet3d973792014-11-11 05:54:27 -08001376 sk_mark_napi_id(sk, skb);
Eric Dumazet404e0a82012-07-29 23:20:37 +00001377 if (dst) {
Eric Dumazet505fbcf2012-07-27 06:23:40 +00001378 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
Ian Morris51456b22015-04-03 09:17:26 +01001379 !dst->ops->check(dst, 0)) {
David S. Miller92101b32012-07-23 16:29:00 -07001380 dst_release(dst);
1381 sk->sk_rx_dst = NULL;
1382 }
1383 }
Vijay Subramanianc995ae22013-09-03 12:23:22 -07001384 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 return 0;
1386 }
1387
Eric Dumazet12e25e12015-06-03 23:49:21 -07001388 if (tcp_checksum_complete(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 goto csum_err;
1390
1391 if (sk->sk_state == TCP_LISTEN) {
Eric Dumazet079096f2015-10-02 11:43:32 -07001392 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 if (!nsk)
1395 goto discard;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 if (nsk != sk) {
Tom Herbertbdeab992011-08-14 19:45:55 +00001397 sock_rps_save_rxhash(nsk, skb);
Eric Dumazet38cb5242015-10-02 11:43:26 -07001398 sk_mark_napi_id(nsk, skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001399 if (tcp_child_process(sk, nsk, skb)) {
1400 rsk = nsk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 goto reset;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001402 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 return 0;
1404 }
Eric Dumazetca551582010-06-03 09:03:58 +00001405 } else
Tom Herbertbdeab992011-08-14 19:45:55 +00001406 sock_rps_save_rxhash(sk, skb);
Eric Dumazetca551582010-06-03 09:03:58 +00001407
Eric Dumazet72ab4a82015-09-29 07:42:41 -07001408 if (tcp_rcv_state_process(sk, skb)) {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001409 rsk = sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 goto reset;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 return 0;
1413
1414reset:
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001415 tcp_v4_send_reset(rsk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416discard:
1417 kfree_skb(skb);
1418 /* Be careful here. If this function gets more complicated and
1419 * gcc suffers from register pressure on the x86, sk (in %ebx)
1420 * might be destroyed here. This current version compiles correctly,
1421 * but you have been warned.
1422 */
1423 return 0;
1424
1425csum_err:
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001426 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001427 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 goto discard;
1429}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001430EXPORT_SYMBOL(tcp_v4_do_rcv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
David S. Miller160eb5a2012-06-27 22:01:22 -07001432void tcp_v4_early_demux(struct sk_buff *skb)
David S. Miller41063e92012-06-19 21:22:05 -07001433{
David S. Miller41063e92012-06-19 21:22:05 -07001434 const struct iphdr *iph;
1435 const struct tcphdr *th;
1436 struct sock *sk;
David S. Miller41063e92012-06-19 21:22:05 -07001437
David S. Miller41063e92012-06-19 21:22:05 -07001438 if (skb->pkt_type != PACKET_HOST)
David S. Miller160eb5a2012-06-27 22:01:22 -07001439 return;
David S. Miller41063e92012-06-19 21:22:05 -07001440
Eric Dumazet45f00f92012-10-22 21:42:47 +00001441 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
David S. Miller160eb5a2012-06-27 22:01:22 -07001442 return;
David S. Miller41063e92012-06-19 21:22:05 -07001443
1444 iph = ip_hdr(skb);
Eric Dumazet45f00f92012-10-22 21:42:47 +00001445 th = tcp_hdr(skb);
David S. Miller41063e92012-06-19 21:22:05 -07001446
1447 if (th->doff < sizeof(struct tcphdr) / 4)
David S. Miller160eb5a2012-06-27 22:01:22 -07001448 return;
David S. Miller41063e92012-06-19 21:22:05 -07001449
Eric Dumazet45f00f92012-10-22 21:42:47 +00001450 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
David S. Miller41063e92012-06-19 21:22:05 -07001451 iph->saddr, th->source,
Vijay Subramanian7011d082012-06-23 17:38:10 +00001452 iph->daddr, ntohs(th->dest),
Eric Dumazet9cb429d2012-07-24 01:19:31 +00001453 skb->skb_iif);
David S. Miller41063e92012-06-19 21:22:05 -07001454 if (sk) {
1455 skb->sk = sk;
1456 skb->destructor = sock_edemux;
Eric Dumazetf7e4eb02015-03-15 21:12:13 -07001457 if (sk_fullsock(sk)) {
Michal Kubečekd0c294c2015-03-23 15:14:00 +01001458 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
Eric Dumazet505fbcf2012-07-27 06:23:40 +00001459
David S. Miller41063e92012-06-19 21:22:05 -07001460 if (dst)
1461 dst = dst_check(dst, 0);
David S. Miller92101b32012-07-23 16:29:00 -07001462 if (dst &&
Eric Dumazet505fbcf2012-07-27 06:23:40 +00001463 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
David S. Miller92101b32012-07-23 16:29:00 -07001464 skb_dst_set_noref(skb, dst);
David S. Miller41063e92012-06-19 21:22:05 -07001465 }
1466 }
David S. Miller41063e92012-06-19 21:22:05 -07001467}
1468
Eric Dumazetb2fb4f52013-03-06 12:58:01 +00001469/* Packet is added to VJ-style prequeue for processing in process
1470 * context, if a reader task is waiting. Apparently, this exciting
1471 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1472 * failed somewhere. Latency? Burstiness? Well, at least now we will
1473 * see, why it failed. 8)8) --ANK
1474 *
1475 */
1476bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1477{
1478 struct tcp_sock *tp = tcp_sk(sk);
1479
1480 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1481 return false;
1482
1483 if (skb->len <= tcp_hdrlen(skb) &&
1484 skb_queue_len(&tp->ucopy.prequeue) == 0)
1485 return false;
1486
Eric Dumazetca777ef2014-09-08 08:06:07 -07001487 /* Before escaping RCU protected region, we need to take care of skb
1488 * dst. Prequeue is only enabled for established sockets.
1489 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1490 * Instead of doing full sk_rx_dst validity here, let's perform
1491 * an optimistic check.
1492 */
1493 if (likely(sk->sk_rx_dst))
1494 skb_dst_drop(skb);
1495 else
1496 skb_dst_force(skb);
1497
Eric Dumazetb2fb4f52013-03-06 12:58:01 +00001498 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1499 tp->ucopy.memory += skb->truesize;
1500 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1501 struct sk_buff *skb1;
1502
1503 BUG_ON(sock_owned_by_user(sk));
1504
1505 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1506 sk_backlog_rcv(sk, skb1);
1507 NET_INC_STATS_BH(sock_net(sk),
1508 LINUX_MIB_TCPPREQUEUEDROPPED);
1509 }
1510
1511 tp->ucopy.memory = 0;
1512 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1513 wake_up_interruptible_sync_poll(sk_sleep(sk),
1514 POLLIN | POLLRDNORM | POLLRDBAND);
1515 if (!inet_csk_ack_scheduled(sk))
1516 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1517 (3 * tcp_rto_min(sk)) / 4,
1518 TCP_RTO_MAX);
1519 }
1520 return true;
1521}
1522EXPORT_SYMBOL(tcp_prequeue);
1523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524/*
1525 * From tcp_input.c
1526 */
1527
1528int tcp_v4_rcv(struct sk_buff *skb)
1529{
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001530 const struct iphdr *iph;
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001531 const struct tcphdr *th;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 struct sock *sk;
1533 int ret;
Pavel Emelyanova86b1e32008-07-16 20:20:58 -07001534 struct net *net = dev_net(skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
1536 if (skb->pkt_type != PACKET_HOST)
1537 goto discard_it;
1538
1539 /* Count it even if it's bad */
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001540 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
1542 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1543 goto discard_it;
1544
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001545 th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
1547 if (th->doff < sizeof(struct tcphdr) / 4)
1548 goto bad_packet;
1549 if (!pskb_may_pull(skb, th->doff * 4))
1550 goto discard_it;
1551
1552 /* An explanation is required here, I think.
1553 * Packet length and doff are validated by header prediction,
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -08001554 * provided case of th->doff==0 is eliminated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 * So, we defer the checks. */
Tom Herberted70fcf2014-05-02 16:29:38 -07001556
1557 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001558 goto csum_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001560 th = tcp_hdr(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001561 iph = ip_hdr(skb);
Eric Dumazet971f10e2014-09-27 09:50:57 -07001562 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1563 * barrier() makes sure compiler wont play fool^Waliasing games.
1564 */
1565 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1566 sizeof(struct inet_skb_parm));
1567 barrier();
1568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1570 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1571 skb->len - th->doff * 4);
1572 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
Eric Dumazete11ecdd2014-09-15 04:19:51 -07001573 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
Eric Dumazet04317da2014-09-05 15:33:32 -07001574 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
Eric Dumazetb82d1bb2011-09-27 02:20:08 -04001575 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 TCP_SKB_CB(skb)->sacked = 0;
1577
Eric Dumazet4bdc3d62015-10-13 17:12:54 -07001578lookup:
Arnaldo Carvalho de Melo9a1f27c2008-10-07 11:41:57 -07001579 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 if (!sk)
1581 goto no_tcp_socket;
1582
Eric Dumazetbb134d52010-03-09 05:55:56 +00001583process:
1584 if (sk->sk_state == TCP_TIME_WAIT)
1585 goto do_time_wait;
1586
Eric Dumazet079096f2015-10-02 11:43:32 -07001587 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1588 struct request_sock *req = inet_reqsk(sk);
1589 struct sock *nsk = NULL;
1590
1591 sk = req->rsk_listener;
1592 if (tcp_v4_inbound_md5_hash(sk, skb))
1593 goto discard_and_relse;
Eric Dumazet4bdc3d62015-10-13 17:12:54 -07001594 if (likely(sk->sk_state == TCP_LISTEN)) {
Eric Dumazet079096f2015-10-02 11:43:32 -07001595 nsk = tcp_check_req(sk, skb, req, false);
Eric Dumazet4bdc3d62015-10-13 17:12:54 -07001596 } else {
Eric Dumazetf03f2e12015-10-14 11:16:27 -07001597 inet_csk_reqsk_queue_drop_and_put(sk, req);
Eric Dumazet4bdc3d62015-10-13 17:12:54 -07001598 goto lookup;
1599 }
Eric Dumazet079096f2015-10-02 11:43:32 -07001600 if (!nsk) {
1601 reqsk_put(req);
1602 goto discard_it;
1603 }
1604 if (nsk == sk) {
1605 sock_hold(sk);
1606 reqsk_put(req);
1607 } else if (tcp_child_process(sk, nsk, skb)) {
1608 tcp_v4_send_reset(nsk, skb);
1609 goto discard_it;
1610 } else {
1611 return 0;
1612 }
1613 }
Eric Dumazet6cce09f2010-03-07 23:21:57 +00001614 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1615 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
Stephen Hemmingerd218d112010-01-11 16:28:01 -08001616 goto discard_and_relse;
Eric Dumazet6cce09f2010-03-07 23:21:57 +00001617 }
Stephen Hemmingerd218d112010-01-11 16:28:01 -08001618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1620 goto discard_and_relse;
Dmitry Popov9ea88a12014-08-07 02:38:22 +04001621
Dmitry Popov9ea88a12014-08-07 02:38:22 +04001622 if (tcp_v4_inbound_md5_hash(sk, skb))
1623 goto discard_and_relse;
Dmitry Popov9ea88a12014-08-07 02:38:22 +04001624
Patrick McHardyb59c2702006-01-06 23:06:10 -08001625 nf_reset(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07001627 if (sk_filter(sk, skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 goto discard_and_relse;
1629
1630 skb->dev = NULL;
1631
Eric Dumazete994b2f2015-10-02 11:43:39 -07001632 if (sk->sk_state == TCP_LISTEN) {
1633 ret = tcp_v4_do_rcv(sk, skb);
1634 goto put_and_return;
1635 }
1636
1637 sk_incoming_cpu_update(sk);
1638
Ingo Molnarc6366182006-07-03 00:25:13 -07001639 bh_lock_sock_nested(sk);
Marcelo Ricardo Leitner2efd0552015-05-20 16:35:41 -07001640 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 ret = 0;
1642 if (!sock_owned_by_user(sk)) {
Dan Williams7bced392013-12-30 12:37:29 -08001643 if (!tcp_prequeue(sk, skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 ret = tcp_v4_do_rcv(sk, skb);
Eric Dumazetda882c12012-04-22 23:38:54 +00001645 } else if (unlikely(sk_add_backlog(sk, skb,
1646 sk->sk_rcvbuf + sk->sk_sndbuf))) {
Zhu Yi6b03a532010-03-04 18:01:41 +00001647 bh_unlock_sock(sk);
Eric Dumazet6cce09f2010-03-07 23:21:57 +00001648 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
Zhu Yi6b03a532010-03-04 18:01:41 +00001649 goto discard_and_relse;
1650 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 bh_unlock_sock(sk);
1652
Eric Dumazete994b2f2015-10-02 11:43:39 -07001653put_and_return:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 sock_put(sk);
1655
1656 return ret;
1657
1658no_tcp_socket:
1659 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1660 goto discard_it;
1661
Eric Dumazet12e25e12015-06-03 23:49:21 -07001662 if (tcp_checksum_complete(skb)) {
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001663csum_error:
1664 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665bad_packet:
Pavel Emelyanov63231bd2008-07-16 20:22:25 -07001666 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 } else {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001668 tcp_v4_send_reset(NULL, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 }
1670
1671discard_it:
1672 /* Discard frame. */
1673 kfree_skb(skb);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001674 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676discard_and_relse:
1677 sock_put(sk);
1678 goto discard_it;
1679
1680do_time_wait:
1681 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001682 inet_twsk_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 goto discard_it;
1684 }
1685
Eric Dumazet6a5dc9e2013-04-29 08:39:56 +00001686 if (tcp_checksum_complete(skb)) {
1687 inet_twsk_put(inet_twsk(sk));
1688 goto csum_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 }
YOSHIFUJI Hideaki9469c7b2006-10-10 19:41:46 -07001690 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 case TCP_TW_SYN: {
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09001692 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
Pavel Emelyanovc67499c2008-01-31 05:06:40 -08001693 &tcp_hashinfo,
Tom Herbertda5e3632013-01-22 09:50:24 +00001694 iph->saddr, th->source,
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001695 iph->daddr, th->dest,
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001696 inet_iif(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 if (sk2) {
Eric Dumazetdbe7faa2015-07-08 14:28:30 -07001698 inet_twsk_deschedule_put(inet_twsk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 sk = sk2;
1700 goto process;
1701 }
1702 /* Fall through to ACK */
1703 }
1704 case TCP_TW_ACK:
1705 tcp_v4_timewait_ack(sk, skb);
1706 break;
1707 case TCP_TW_RST:
1708 goto no_tcp_socket;
1709 case TCP_TW_SUCCESS:;
1710 }
1711 goto discard_it;
1712}
1713
David S. Millerccb7c412010-12-01 18:09:13 -08001714static struct timewait_sock_ops tcp_timewait_sock_ops = {
1715 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1716 .twsk_unique = tcp_twsk_unique,
1717 .twsk_destructor= tcp_twsk_destructor,
David S. Millerccb7c412010-12-01 18:09:13 -08001718};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
Eric Dumazet63d02d12012-08-09 14:11:00 +00001720void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
Eric Dumazet5d299f32012-08-06 05:09:33 +00001721{
1722 struct dst_entry *dst = skb_dst(skb);
1723
Eric Dumazetca777ef2014-09-08 08:06:07 -07001724 if (dst) {
1725 dst_hold(dst);
1726 sk->sk_rx_dst = dst;
1727 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1728 }
Eric Dumazet5d299f32012-08-06 05:09:33 +00001729}
Eric Dumazet63d02d12012-08-09 14:11:00 +00001730EXPORT_SYMBOL(inet_sk_rx_dst_set);
Eric Dumazet5d299f32012-08-06 05:09:33 +00001731
Stephen Hemminger3b401a82009-09-01 19:25:04 +00001732const struct inet_connection_sock_af_ops ipv4_specific = {
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001733 .queue_xmit = ip_queue_xmit,
1734 .send_check = tcp_v4_send_check,
1735 .rebuild_header = inet_sk_rebuild_header,
Eric Dumazet5d299f32012-08-06 05:09:33 +00001736 .sk_rx_dst_set = inet_sk_rx_dst_set,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001737 .conn_request = tcp_v4_conn_request,
1738 .syn_recv_sock = tcp_v4_syn_recv_sock,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001739 .net_header_len = sizeof(struct iphdr),
1740 .setsockopt = ip_setsockopt,
1741 .getsockopt = ip_getsockopt,
1742 .addr2sockaddr = inet_csk_addr2sockaddr,
1743 .sockaddr_len = sizeof(struct sockaddr_in),
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001744 .bind_conflict = inet_csk_bind_conflict,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001745#ifdef CONFIG_COMPAT
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08001746 .compat_setsockopt = compat_ip_setsockopt,
1747 .compat_getsockopt = compat_ip_getsockopt,
Dmitry Mishin3fdadf72006-03-20 22:45:21 -08001748#endif
Neal Cardwell4fab9072014-08-14 12:40:05 -04001749 .mtu_reduced = tcp_v4_mtu_reduced,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750};
Eric Dumazet4bc2f182010-07-09 21:22:10 +00001751EXPORT_SYMBOL(ipv4_specific);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001753#ifdef CONFIG_TCP_MD5SIG
Stephen Hemmingerb2e4b3d2009-09-01 19:25:03 +00001754static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001755 .md5_lookup = tcp_v4_md5_lookup,
Adam Langley49a72df2008-07-19 00:01:42 -07001756 .calc_md5_hash = tcp_v4_md5_hash_skb,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001757 .md5_parse = tcp_v4_parse_md5_keys,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001758};
Andrew Mortonb6332e62006-11-30 19:16:28 -08001759#endif
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761/* NOTE: A lot of things set to zero explicitly by call to
1762 * sk_alloc() so need not be done here.
1763 */
1764static int tcp_v4_init_sock(struct sock *sk)
1765{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001766 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
Neal Cardwell900f65d2012-04-19 09:55:21 +00001768 tcp_init_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -08001770 icsk->icsk_af_ops = &ipv4_specific;
Neal Cardwell900f65d2012-04-19 09:55:21 +00001771
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001772#ifdef CONFIG_TCP_MD5SIG
David S. Millerac807fa2012-04-23 03:21:58 -04001773 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001774#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 return 0;
1777}
1778
Brian Haley7d06b2e2008-06-14 17:04:49 -07001779void tcp_v4_destroy_sock(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780{
1781 struct tcp_sock *tp = tcp_sk(sk);
1782
1783 tcp_clear_xmit_timers(sk);
1784
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001785 tcp_cleanup_congestion_control(sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001786
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 /* Cleanup up the write buffer. */
David S. Millerfe067e82007-03-07 12:12:44 -08001788 tcp_write_queue_purge(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
1790 /* Cleans up our, hopefully empty, out_of_order_queue. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001791 __skb_queue_purge(&tp->out_of_order_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001793#ifdef CONFIG_TCP_MD5SIG
1794 /* Clean up the MD5 key list, if any */
1795 if (tp->md5sig_info) {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001796 tcp_clear_md5_list(sk);
Eric Dumazeta8afca02012-01-31 18:45:40 +00001797 kfree_rcu(tp->md5sig_info, rcu);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001798 tp->md5sig_info = NULL;
1799 }
1800#endif
1801
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 /* Clean prequeue, it must be empty really */
1803 __skb_queue_purge(&tp->ucopy.prequeue);
1804
1805 /* Clean up a referenced TCP bind bucket. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001806 if (inet_csk(sk)->icsk_bind_hash)
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08001807 inet_put_port(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Ian Morris00db4122015-04-03 09:17:27 +01001809 BUG_ON(tp->fastopen_rsk);
William Allen Simpson435cf552009-12-02 18:17:05 +00001810
Yuchung Chengcf60af02012-07-19 06:43:09 +00001811 /* If socket is aborted during connect operation */
1812 tcp_free_fastopen_req(tp);
Eric Dumazetcd8ae852015-05-03 21:34:46 -07001813 tcp_saved_syn_free(tp);
Yuchung Chengcf60af02012-07-19 06:43:09 +00001814
Glauber Costa180d8cd2011-12-11 21:47:02 +00001815 sk_sockets_allocated_dec(sk);
Glauber Costad1a4c0b2011-12-11 21:47:04 +00001816 sock_release_memcg(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818EXPORT_SYMBOL(tcp_v4_destroy_sock);
1819
1820#ifdef CONFIG_PROC_FS
1821/* Proc filesystem TCP sock list dumping. */
1822
Tom Herberta8b690f2010-06-07 00:43:42 -07001823/*
1824 * Get next listener socket follow cur. If cur is NULL, get first socket
1825 * starting from bucket given in st->bucket; when st->bucket is zero the
1826 * very first socket in the hash table is returned.
1827 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828static void *listening_get_next(struct seq_file *seq, void *cur)
1829{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001830 struct inet_connection_sock *icsk;
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001831 struct hlist_nulls_node *node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 struct sock *sk = cur;
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001833 struct inet_listen_hashbucket *ilb;
Jianjun Kong5799de02008-11-03 02:49:10 -08001834 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07001835 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
1837 if (!sk) {
Tom Herberta8b690f2010-06-07 00:43:42 -07001838 ilb = &tcp_hashinfo.listening_hash[st->bucket];
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001839 spin_lock_bh(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001840 sk = sk_nulls_head(&ilb->head);
Tom Herberta8b690f2010-06-07 00:43:42 -07001841 st->offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 goto get_sk;
1843 }
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001844 ilb = &tcp_hashinfo.listening_hash[st->bucket];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 ++st->num;
Tom Herberta8b690f2010-06-07 00:43:42 -07001846 ++st->offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
Eric Dumazet079096f2015-10-02 11:43:32 -07001848 sk = sk_nulls_next(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849get_sk:
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001850 sk_nulls_for_each_from(sk, node) {
Pavel Emelyanov8475ef92010-11-22 03:26:12 +00001851 if (!net_eq(sock_net(sk), net))
1852 continue;
1853 if (sk->sk_family == st->family) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 cur = sk;
1855 goto out;
1856 }
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001857 icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 }
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001859 spin_unlock_bh(&ilb->lock);
Tom Herberta8b690f2010-06-07 00:43:42 -07001860 st->offset = 0;
Arnaldo Carvalho de Melo0f7ff922005-08-09 19:59:44 -07001861 if (++st->bucket < INET_LHTABLE_SIZE) {
Eric Dumazet5caea4e2008-11-20 00:40:07 -08001862 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1863 spin_lock_bh(&ilb->lock);
Eric Dumazetc25eb3b2008-11-23 17:22:55 -08001864 sk = sk_nulls_head(&ilb->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 goto get_sk;
1866 }
1867 cur = NULL;
1868out:
1869 return cur;
1870}
1871
1872static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1873{
Tom Herberta8b690f2010-06-07 00:43:42 -07001874 struct tcp_iter_state *st = seq->private;
1875 void *rc;
1876
1877 st->bucket = 0;
1878 st->offset = 0;
1879 rc = listening_get_next(seq, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
1881 while (rc && *pos) {
1882 rc = listening_get_next(seq, rc);
1883 --*pos;
1884 }
1885 return rc;
1886}
1887
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07001888static inline bool empty_bucket(const struct tcp_iter_state *st)
Andi Kleen6eac5602008-08-28 01:08:02 -07001889{
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07001890 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
Andi Kleen6eac5602008-08-28 01:08:02 -07001891}
1892
Tom Herberta8b690f2010-06-07 00:43:42 -07001893/*
1894 * Get first established socket starting from bucket given in st->bucket.
1895 * If st->bucket is zero, the very first socket in the hash is returned.
1896 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897static void *established_get_first(struct seq_file *seq)
1898{
Jianjun Kong5799de02008-11-03 02:49:10 -08001899 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07001900 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 void *rc = NULL;
1902
Tom Herberta8b690f2010-06-07 00:43:42 -07001903 st->offset = 0;
1904 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 struct sock *sk;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001906 struct hlist_nulls_node *node;
Eric Dumazet9db66bd2008-11-20 20:39:09 -08001907 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Andi Kleen6eac5602008-08-28 01:08:02 -07001909 /* Lockless fast path for the common case of empty buckets */
1910 if (empty_bucket(st))
1911 continue;
1912
Eric Dumazet9db66bd2008-11-20 20:39:09 -08001913 spin_lock_bh(lock);
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001914 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
Daniel Lezcanof40c8172008-03-21 04:13:54 -07001915 if (sk->sk_family != st->family ||
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001916 !net_eq(sock_net(sk), net)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 continue;
1918 }
1919 rc = sk;
1920 goto out;
1921 }
Eric Dumazet9db66bd2008-11-20 20:39:09 -08001922 spin_unlock_bh(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 }
1924out:
1925 return rc;
1926}
1927
1928static void *established_get_next(struct seq_file *seq, void *cur)
1929{
1930 struct sock *sk = cur;
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001931 struct hlist_nulls_node *node;
Jianjun Kong5799de02008-11-03 02:49:10 -08001932 struct tcp_iter_state *st = seq->private;
Denis V. Luneva4146b12008-04-13 22:11:14 -07001933 struct net *net = seq_file_net(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
1935 ++st->num;
Tom Herberta8b690f2010-06-07 00:43:42 -07001936 ++st->offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07001938 sk = sk_nulls_next(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08001940 sk_nulls_for_each_from(sk, node) {
YOSHIFUJI Hideaki878628f2008-03-26 03:57:35 +09001941 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07001942 return sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 }
1944
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07001945 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1946 ++st->bucket;
1947 return established_get_first(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948}
1949
1950static void *established_get_idx(struct seq_file *seq, loff_t pos)
1951{
Tom Herberta8b690f2010-06-07 00:43:42 -07001952 struct tcp_iter_state *st = seq->private;
1953 void *rc;
1954
1955 st->bucket = 0;
1956 rc = established_get_first(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
1958 while (rc && pos) {
1959 rc = established_get_next(seq, rc);
1960 --pos;
Arnaldo Carvalho de Melo71742592006-11-17 10:57:30 -02001961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 return rc;
1963}
1964
1965static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1966{
1967 void *rc;
Jianjun Kong5799de02008-11-03 02:49:10 -08001968 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 st->state = TCP_SEQ_STATE_LISTENING;
1971 rc = listening_get_idx(seq, &pos);
1972
1973 if (!rc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 st->state = TCP_SEQ_STATE_ESTABLISHED;
1975 rc = established_get_idx(seq, pos);
1976 }
1977
1978 return rc;
1979}
1980
Tom Herberta8b690f2010-06-07 00:43:42 -07001981static void *tcp_seek_last_pos(struct seq_file *seq)
1982{
1983 struct tcp_iter_state *st = seq->private;
1984 int offset = st->offset;
1985 int orig_num = st->num;
1986 void *rc = NULL;
1987
1988 switch (st->state) {
Tom Herberta8b690f2010-06-07 00:43:42 -07001989 case TCP_SEQ_STATE_LISTENING:
1990 if (st->bucket >= INET_LHTABLE_SIZE)
1991 break;
1992 st->state = TCP_SEQ_STATE_LISTENING;
1993 rc = listening_get_next(seq, NULL);
1994 while (offset-- && rc)
1995 rc = listening_get_next(seq, rc);
1996 if (rc)
1997 break;
1998 st->bucket = 0;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07001999 st->state = TCP_SEQ_STATE_ESTABLISHED;
Tom Herberta8b690f2010-06-07 00:43:42 -07002000 /* Fallthrough */
2001 case TCP_SEQ_STATE_ESTABLISHED:
Tom Herberta8b690f2010-06-07 00:43:42 -07002002 if (st->bucket > tcp_hashinfo.ehash_mask)
2003 break;
2004 rc = established_get_first(seq);
2005 while (offset-- && rc)
2006 rc = established_get_next(seq, rc);
2007 }
2008
2009 st->num = orig_num;
2010
2011 return rc;
2012}
2013
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2015{
Jianjun Kong5799de02008-11-03 02:49:10 -08002016 struct tcp_iter_state *st = seq->private;
Tom Herberta8b690f2010-06-07 00:43:42 -07002017 void *rc;
2018
2019 if (*pos && *pos == st->last_pos) {
2020 rc = tcp_seek_last_pos(seq);
2021 if (rc)
2022 goto out;
2023 }
2024
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 st->state = TCP_SEQ_STATE_LISTENING;
2026 st->num = 0;
Tom Herberta8b690f2010-06-07 00:43:42 -07002027 st->bucket = 0;
2028 st->offset = 0;
2029 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2030
2031out:
2032 st->last_pos = *pos;
2033 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034}
2035
2036static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2037{
Tom Herberta8b690f2010-06-07 00:43:42 -07002038 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 void *rc = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
2041 if (v == SEQ_START_TOKEN) {
2042 rc = tcp_get_idx(seq, 0);
2043 goto out;
2044 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
2046 switch (st->state) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 case TCP_SEQ_STATE_LISTENING:
2048 rc = listening_get_next(seq, v);
2049 if (!rc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 st->state = TCP_SEQ_STATE_ESTABLISHED;
Tom Herberta8b690f2010-06-07 00:43:42 -07002051 st->bucket = 0;
2052 st->offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 rc = established_get_first(seq);
2054 }
2055 break;
2056 case TCP_SEQ_STATE_ESTABLISHED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 rc = established_get_next(seq, v);
2058 break;
2059 }
2060out:
2061 ++*pos;
Tom Herberta8b690f2010-06-07 00:43:42 -07002062 st->last_pos = *pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 return rc;
2064}
2065
2066static void tcp_seq_stop(struct seq_file *seq, void *v)
2067{
Jianjun Kong5799de02008-11-03 02:49:10 -08002068 struct tcp_iter_state *st = seq->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
2070 switch (st->state) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 case TCP_SEQ_STATE_LISTENING:
2072 if (v != SEQ_START_TOKEN)
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002073 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 case TCP_SEQ_STATE_ESTABLISHED:
2076 if (v)
Eric Dumazet9db66bd2008-11-20 20:39:09 -08002077 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 break;
2079 }
2080}
2081
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002082int tcp_seq_open(struct inode *inode, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
Al Virod9dda782013-03-31 18:16:14 -04002084 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 struct tcp_iter_state *s;
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002086 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002088 err = seq_open_net(inode, file, &afinfo->seq_ops,
2089 sizeof(struct tcp_iter_state));
2090 if (err < 0)
2091 return err;
Daniel Lezcanof40c8172008-03-21 04:13:54 -07002092
Denis V. Lunev52d6f3f2008-04-13 22:12:41 -07002093 s = ((struct seq_file *)file->private_data)->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 s->family = afinfo->family;
stephen hemminger688d1942014-08-29 23:32:05 -07002095 s->last_pos = 0;
Daniel Lezcanof40c8172008-03-21 04:13:54 -07002096 return 0;
2097}
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002098EXPORT_SYMBOL(tcp_seq_open);
Daniel Lezcanof40c8172008-03-21 04:13:54 -07002099
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07002100int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101{
2102 int rc = 0;
2103 struct proc_dir_entry *p;
2104
Denis V. Lunev9427c4b2008-04-13 22:12:13 -07002105 afinfo->seq_ops.start = tcp_seq_start;
2106 afinfo->seq_ops.next = tcp_seq_next;
2107 afinfo->seq_ops.stop = tcp_seq_stop;
2108
Denis V. Lunev84841c32008-05-02 04:10:08 -07002109 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002110 afinfo->seq_fops, afinfo);
Denis V. Lunev84841c32008-05-02 04:10:08 -07002111 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 rc = -ENOMEM;
2113 return rc;
2114}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002115EXPORT_SYMBOL(tcp_proc_register);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07002117void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118{
Gao fengece31ff2013-02-18 01:34:56 +00002119 remove_proc_entry(afinfo->name, net->proc_net);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120}
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002121EXPORT_SYMBOL(tcp_proc_unregister);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122
Eric Dumazetd4f06872015-03-12 16:44:09 -07002123static void get_openreq4(const struct request_sock *req,
Eric Dumazetaa3a0c82015-10-02 11:43:30 -07002124 struct seq_file *f, int i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07002126 const struct inet_request_sock *ireq = inet_rsk(req);
Eric Dumazetfa76ce732015-03-19 19:04:20 -07002127 long delta = req->rsk_timer.expires - jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002129 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
Tetsuo Handa652586d2013-11-14 14:31:57 -08002130 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 i,
Eric Dumazet634fb9792013-10-09 15:21:29 -07002132 ireq->ir_loc_addr,
Eric Dumazetd4f06872015-03-12 16:44:09 -07002133 ireq->ir_num,
Eric Dumazet634fb9792013-10-09 15:21:29 -07002134 ireq->ir_rmt_addr,
2135 ntohs(ireq->ir_rmt_port),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 TCP_SYN_RECV,
2137 0, 0, /* could print option size, but that is af dependent. */
2138 1, /* timers active (only the expire timer) */
Eric Dumazeta399a802012-08-08 21:13:53 +00002139 jiffies_delta_to_clock_t(delta),
Eric Dumazete6c022a2012-10-27 23:16:46 +00002140 req->num_timeout,
Eric Dumazetaa3a0c82015-10-02 11:43:30 -07002141 from_kuid_munged(seq_user_ns(f),
2142 sock_i_uid(req->rsk_listener)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 0, /* non standard timer */
2144 0, /* open_requests have no inode */
Eric Dumazetd4f06872015-03-12 16:44:09 -07002145 0,
Tetsuo Handa652586d2013-11-14 14:31:57 -08002146 req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147}
2148
Tetsuo Handa652586d2013-11-14 14:31:57 -08002149static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150{
2151 int timer_active;
2152 unsigned long timer_expires;
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002153 const struct tcp_sock *tp = tcp_sk(sk);
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002154 const struct inet_connection_sock *icsk = inet_csk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002155 const struct inet_sock *inet = inet_sk(sk);
Eric Dumazet0536fcc2015-09-29 07:42:52 -07002156 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
Eric Dumazetc720c7e2009-10-15 06:30:45 +00002157 __be32 dest = inet->inet_daddr;
2158 __be32 src = inet->inet_rcv_saddr;
2159 __u16 destp = ntohs(inet->inet_dport);
2160 __u16 srcp = ntohs(inet->inet_sport);
Eric Dumazet49d09002009-12-03 16:06:13 -08002161 int rx_queue;
Eric Dumazet00fd38d2015-11-12 08:43:18 -08002162 int state;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +00002164 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2165 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2166 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 timer_active = 1;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002168 timer_expires = icsk->icsk_timeout;
2169 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 timer_active = 4;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002171 timer_expires = icsk->icsk_timeout;
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002172 } else if (timer_pending(&sk->sk_timer)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 timer_active = 2;
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002174 timer_expires = sk->sk_timer.expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 } else {
2176 timer_active = 0;
2177 timer_expires = jiffies;
2178 }
2179
Eric Dumazet00fd38d2015-11-12 08:43:18 -08002180 state = sk_state_load(sk);
2181 if (state == TCP_LISTEN)
Eric Dumazet49d09002009-12-03 16:06:13 -08002182 rx_queue = sk->sk_ack_backlog;
2183 else
Eric Dumazet00fd38d2015-11-12 08:43:18 -08002184 /* Because we don't lock the socket,
2185 * we might find a transient negative value.
Eric Dumazet49d09002009-12-03 16:06:13 -08002186 */
2187 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2188
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002189 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
Tetsuo Handa652586d2013-11-14 14:31:57 -08002190 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
Eric Dumazet00fd38d2015-11-12 08:43:18 -08002191 i, src, srcp, dest, destp, state,
Sridhar Samudrala47da8ee2006-06-27 13:29:00 -07002192 tp->write_seq - tp->snd_una,
Eric Dumazet49d09002009-12-03 16:06:13 -08002193 rx_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 timer_active,
Eric Dumazeta399a802012-08-08 21:13:53 +00002195 jiffies_delta_to_clock_t(timer_expires - jiffies),
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002196 icsk->icsk_retransmits,
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06002197 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03002198 icsk->icsk_probes_out,
Ilpo Järvinencf4c6bf2007-02-22 01:13:58 -08002199 sock_i_ino(sk),
2200 atomic_read(&sk->sk_refcnt), sk,
Stephen Hemminger7be87352008-06-27 20:00:19 -07002201 jiffies_to_clock_t(icsk->icsk_rto),
2202 jiffies_to_clock_t(icsk->icsk_ack.ato),
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002203 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 tp->snd_cwnd,
Eric Dumazet00fd38d2015-11-12 08:43:18 -08002205 state == TCP_LISTEN ?
2206 fastopenq->max_qlen :
Tetsuo Handa652586d2013-11-14 14:31:57 -08002207 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208}
2209
Eric Dumazetcf533ea2011-10-21 05:22:42 -04002210static void get_timewait4_sock(const struct inet_timewait_sock *tw,
Tetsuo Handa652586d2013-11-14 14:31:57 -08002211 struct seq_file *f, int i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
Eric Dumazet789f5582015-04-12 18:51:09 -07002213 long delta = tw->tw_timer.expires - jiffies;
Al Viro23f33c22006-09-27 18:43:50 -07002214 __be32 dest, src;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 __u16 destp, srcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216
2217 dest = tw->tw_daddr;
2218 src = tw->tw_rcv_saddr;
2219 destp = ntohs(tw->tw_dport);
2220 srcp = ntohs(tw->tw_sport);
2221
Pavel Emelyanov5e659e42008-04-24 01:02:16 -07002222 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
Tetsuo Handa652586d2013-11-14 14:31:57 -08002223 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
Eric Dumazeta399a802012-08-08 21:13:53 +00002225 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
Tetsuo Handa652586d2013-11-14 14:31:57 -08002226 atomic_read(&tw->tw_refcnt), tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227}
2228
2229#define TMPSZ 150
2230
2231static int tcp4_seq_show(struct seq_file *seq, void *v)
2232{
Jianjun Kong5799de02008-11-03 02:49:10 -08002233 struct tcp_iter_state *st;
Eric Dumazet05dbc7b2013-10-03 00:22:02 -07002234 struct sock *sk = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235
Tetsuo Handa652586d2013-11-14 14:31:57 -08002236 seq_setwidth(seq, TMPSZ - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 if (v == SEQ_START_TOKEN) {
Tetsuo Handa652586d2013-11-14 14:31:57 -08002238 seq_puts(seq, " sl local_address rem_address st tx_queue "
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 "rx_queue tr tm->when retrnsmt uid timeout "
2240 "inode");
2241 goto out;
2242 }
2243 st = seq->private;
2244
Eric Dumazet079096f2015-10-02 11:43:32 -07002245 if (sk->sk_state == TCP_TIME_WAIT)
2246 get_timewait4_sock(v, seq, st->num);
2247 else if (sk->sk_state == TCP_NEW_SYN_RECV)
Eric Dumazetaa3a0c82015-10-02 11:43:30 -07002248 get_openreq4(v, seq, st->num);
Eric Dumazet079096f2015-10-02 11:43:32 -07002249 else
2250 get_tcp4_sock(v, seq, st->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251out:
Tetsuo Handa652586d2013-11-14 14:31:57 -08002252 seq_pad(seq, '\n');
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 return 0;
2254}
2255
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002256static const struct file_operations tcp_afinfo_seq_fops = {
2257 .owner = THIS_MODULE,
2258 .open = tcp_seq_open,
2259 .read = seq_read,
2260 .llseek = seq_lseek,
2261 .release = seq_release_net
2262};
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264static struct tcp_seq_afinfo tcp4_seq_afinfo = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 .name = "tcp",
2266 .family = AF_INET,
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002267 .seq_fops = &tcp_afinfo_seq_fops,
Denis V. Lunev9427c4b2008-04-13 22:12:13 -07002268 .seq_ops = {
2269 .show = tcp4_seq_show,
2270 },
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271};
2272
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002273static int __net_init tcp4_proc_init_net(struct net *net)
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002274{
2275 return tcp_proc_register(net, &tcp4_seq_afinfo);
2276}
2277
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00002278static void __net_exit tcp4_proc_exit_net(struct net *net)
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002279{
2280 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2281}
2282
2283static struct pernet_operations tcp4_net_ops = {
2284 .init = tcp4_proc_init_net,
2285 .exit = tcp4_proc_exit_net,
2286};
2287
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288int __init tcp4_proc_init(void)
2289{
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002290 return register_pernet_subsys(&tcp4_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291}
2292
2293void tcp4_proc_exit(void)
2294{
Pavel Emelyanov757764f2008-03-24 14:56:02 -07002295 unregister_pernet_subsys(&tcp4_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296}
2297#endif /* CONFIG_PROC_FS */
2298
2299struct proto tcp_prot = {
2300 .name = "TCP",
2301 .owner = THIS_MODULE,
2302 .close = tcp_close,
2303 .connect = tcp_v4_connect,
2304 .disconnect = tcp_disconnect,
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07002305 .accept = inet_csk_accept,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 .ioctl = tcp_ioctl,
2307 .init = tcp_v4_init_sock,
2308 .destroy = tcp_v4_destroy_sock,
2309 .shutdown = tcp_shutdown,
2310 .setsockopt = tcp_setsockopt,
2311 .getsockopt = tcp_getsockopt,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 .recvmsg = tcp_recvmsg,
Changli Gao7ba42912010-07-10 20:41:55 +00002313 .sendmsg = tcp_sendmsg,
2314 .sendpage = tcp_sendpage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 .backlog_rcv = tcp_v4_do_rcv,
Eric Dumazet46d3cea2012-07-11 05:50:31 +00002316 .release_cb = tcp_release_cb,
Arnaldo Carvalho de Meloab1e0a12008-02-03 04:06:04 -08002317 .hash = inet_hash,
2318 .unhash = inet_unhash,
2319 .get_port = inet_csk_get_port,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 .enter_memory_pressure = tcp_enter_memory_pressure,
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07002321 .stream_memory_free = tcp_stream_memory_free,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 .sockets_allocated = &tcp_sockets_allocated,
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07002323 .orphan_count = &tcp_orphan_count,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 .memory_allocated = &tcp_memory_allocated,
2325 .memory_pressure = &tcp_memory_pressure,
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -07002326 .sysctl_mem = sysctl_tcp_mem,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 .sysctl_wmem = sysctl_tcp_wmem,
2328 .sysctl_rmem = sysctl_tcp_rmem,
2329 .max_header = MAX_TCP_HEADER,
2330 .obj_size = sizeof(struct tcp_sock),
Eric Dumazet3ab5aee2008-11-16 19:40:17 -08002331 .slab_flags = SLAB_DESTROY_BY_RCU,
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -08002332 .twsk_prot = &tcp_timewait_sock_ops,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07002333 .rsk_prot = &tcp_request_sock_ops,
Pavel Emelyanov39d8cda2008-03-22 16:50:58 -07002334 .h.hashinfo = &tcp_hashinfo,
Changli Gao7ba42912010-07-10 20:41:55 +00002335 .no_autobind = true,
Arnaldo Carvalho de Melo543d9cf2006-03-20 22:48:35 -08002336#ifdef CONFIG_COMPAT
2337 .compat_setsockopt = compat_tcp_setsockopt,
2338 .compat_getsockopt = compat_tcp_getsockopt,
2339#endif
Andrew Mortonc255a452012-07-31 16:43:02 -07002340#ifdef CONFIG_MEMCG_KMEM
Glauber Costad1a4c0b2011-12-11 21:47:04 +00002341 .init_cgroup = tcp_init_cgroup,
2342 .destroy_cgroup = tcp_destroy_cgroup,
2343 .proto_cgroup = tcp_proto_cgroup,
2344#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345};
Eric Dumazet4bc2f182010-07-09 21:22:10 +00002346EXPORT_SYMBOL(tcp_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
Denis V. Lunev046ee902008-04-03 14:31:33 -07002348static void __net_exit tcp_sk_exit(struct net *net)
2349{
Eric Dumazetbdbbb852015-01-29 21:35:05 -08002350 int cpu;
2351
2352 for_each_possible_cpu(cpu)
2353 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2354 free_percpu(net->ipv4.tcp_sk);
2355}
2356
2357static int __net_init tcp_sk_init(struct net *net)
2358{
2359 int res, cpu;
2360
2361 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2362 if (!net->ipv4.tcp_sk)
2363 return -ENOMEM;
2364
2365 for_each_possible_cpu(cpu) {
2366 struct sock *sk;
2367
2368 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2369 IPPROTO_TCP, net);
2370 if (res)
2371 goto fail;
2372 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2373 }
Daniel Borkmann49213552015-05-19 21:04:22 +02002374
Eric Dumazetbdbbb852015-01-29 21:35:05 -08002375 net->ipv4.sysctl_tcp_ecn = 2;
Daniel Borkmann49213552015-05-19 21:04:22 +02002376 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2377
Fan Dub0f9ca52015-02-10 09:53:16 +08002378 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
Fan Du6b58e0a2015-03-06 11:18:23 +08002379 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
Fan Du05cbc0d2015-03-06 11:18:24 +08002380 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
Eric Dumazetbdbbb852015-01-29 21:35:05 -08002381
Daniel Borkmann49213552015-05-19 21:04:22 +02002382 return 0;
Eric Dumazetbdbbb852015-01-29 21:35:05 -08002383fail:
2384 tcp_sk_exit(net);
2385
2386 return res;
Eric W. Biedermanb099ce22009-12-03 02:29:09 +00002387}
2388
2389static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2390{
2391 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
Denis V. Lunev046ee902008-04-03 14:31:33 -07002392}
2393
2394static struct pernet_operations __net_initdata tcp_sk_ops = {
Eric W. Biedermanb099ce22009-12-03 02:29:09 +00002395 .init = tcp_sk_init,
2396 .exit = tcp_sk_exit,
2397 .exit_batch = tcp_sk_exit_batch,
Denis V. Lunev046ee902008-04-03 14:31:33 -07002398};
2399
Denis V. Lunev9b0f9762008-02-29 11:13:15 -08002400void __init tcp_v4_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401{
Eric Dumazet5caea4e2008-11-20 00:40:07 -08002402 inet_hashinfo_init(&tcp_hashinfo);
Eric W. Biederman6a1b3052009-02-22 00:10:18 -08002403 if (register_pernet_subsys(&tcp_sk_ops))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 panic("Failed to create the TCP control socket.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405}