blob: 4c53e7c865864df29f0980c4c756e6daee966c7f [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 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07008 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Corey Minyard <wf-rch!minyard@relay.EU.net>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
14 * Linus Torvalds, <torvalds@cs.helsinki.fi>
15 * Alan Cox, <gw4pts@gw4pts.ampr.org>
16 * Matthew Dillon, <dillon@apollo.west.oic.com>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18 * Jorge Cwik, <jorge@laser.satlink.net>
19 */
20
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mm.h>
22#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/sysctl.h>
25#include <linux/workqueue.h>
26#include <net/tcp.h>
27#include <net/inet_common.h>
28#include <net/xfrm.h>
29
Brian Haleyab32ea52006-09-22 14:15:41 -070030int sysctl_tcp_abort_on_overflow __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070032struct inet_timewait_death_row tcp_death_row = {
33 .sysctl_max_tw_buckets = NR_FILE * 2,
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070034 .hashinfo = &tcp_hashinfo,
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070035};
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070036EXPORT_SYMBOL_GPL(tcp_death_row);
37
Eric Dumazeta2a385d2012-05-16 23:15:34 +000038static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -070039{
40 if (seq == s_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +000041 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 if (after(end_seq, s_win) && before(seq, e_win))
Eric Dumazeta2a385d2012-05-16 23:15:34 +000043 return true;
Eric Dumazeta02cec22010-09-22 20:43:57 +000044 return seq == e_win && seq == end_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045}
46
Neal Cardwell4fb17a62015-02-06 16:04:41 -050047static enum tcp_tw_status
48tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
49 const struct sk_buff *skb, int mib_idx)
50{
51 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
52
53 if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
54 &tcptw->tw_last_oow_ack_time)) {
55 /* Send ACK. Note, we do not put the bucket,
56 * it will be released by caller.
57 */
58 return TCP_TW_ACK;
59 }
60
61 /* We are rate-limiting, so just release the tw sock and drop skb. */
62 inet_twsk_put(tw);
63 return TCP_TW_SUCCESS;
64}
65
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090066/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * * Main purpose of TIME-WAIT state is to close connection gracefully,
68 * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
69 * (and, probably, tail of data) and one or more our ACKs are lost.
70 * * What is TIME-WAIT timeout? It is associated with maximal packet
71 * lifetime in the internet, which results in wrong conclusion, that
72 * it is set to catch "old duplicate segments" wandering out of their path.
73 * It is not quite correct. This timeout is calculated so that it exceeds
74 * maximal retransmission timeout enough to allow to lose one (or more)
75 * segments sent by peer and our ACKs. This time may be calculated from RTO.
76 * * When TIME-WAIT socket receives RST, it means that another end
77 * finally closed and we are allowed to kill TIME-WAIT too.
78 * * Second purpose of TIME-WAIT is catching old duplicate segments.
79 * Well, certainly it is pure paranoia, but if we load TIME-WAIT
80 * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
81 * * If we invented some more clever way to catch duplicates
82 * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
83 *
84 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
85 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
86 * from the very beginning.
87 *
88 * NOTE. With recycling (and later with fin-wait-2) TW bucket
89 * is _not_ stateless. It means, that strictly speaking we must
90 * spinlock it. I do not want! Well, probability of misbehaviour
91 * is ridiculously low and, seems, we could use some mb() tricks
92 * to avoid misread sequence numbers, states etc. --ANK
Alan Cox4308fc52012-09-19 15:46:06 +010093 *
94 * We don't need to initialize tmp_out.sack_ok as we don't use the results
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 */
96enum tcp_tw_status
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -070097tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
98 const struct tcphdr *th)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099{
100 struct tcp_options_received tmp_opt;
William Allen Simpson4957faade2009-12-02 18:25:27 +0000101 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000102 bool paws_reject = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
David S. Millerbb5b7c12009-12-15 20:56:42 -0800104 tmp_opt.saw_tstamp = 0;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700105 if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000106 tcp_parse_options(skb, &tmp_opt, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
108 if (tmp_opt.saw_tstamp) {
Andrey Vaginee684b62013-02-11 05:50:19 +0000109 tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700110 tmp_opt.ts_recent = tcptw->tw_ts_recent;
111 tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +0000112 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 }
114 }
115
116 if (tw->tw_substate == TCP_FIN_WAIT2) {
117 /* Just repeat all the checks of tcp_rcv_state_process() */
118
119 /* Out of window, send ACK */
120 if (paws_reject ||
121 !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700122 tcptw->tw_rcv_nxt,
123 tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
Neal Cardwell4fb17a62015-02-06 16:04:41 -0500124 return tcp_timewait_check_oow_rate_limit(
125 tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127 if (th->rst)
128 goto kill;
129
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700130 if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
Florian Westphal271c3b92015-12-21 21:29:26 +0100131 return TCP_TW_RST;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133 /* Dup ACK? */
Wei Yongjun1ac530b2009-06-24 22:29:31 +0000134 if (!th->ack ||
135 !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700137 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 return TCP_TW_SUCCESS;
139 }
140
141 /* New data or FIN. If new data arrive after half-duplex close,
142 * reset.
143 */
144 if (!th->fin ||
Florian Westphal271c3b92015-12-21 21:29:26 +0100145 TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 return TCP_TW_RST;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148 /* FIN arrived, enter true time-wait state. */
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700149 tw->tw_substate = TCP_TIME_WAIT;
150 tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 if (tmp_opt.saw_tstamp) {
James Morris9d729f72007-03-04 16:12:44 -0800152 tcptw->tw_ts_recent_stamp = get_seconds();
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700153 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 }
155
David S. Millerccb7c412010-12-01 18:09:13 -0800156 if (tcp_death_row.sysctl_tw_recycle &&
157 tcptw->tw_ts_recent_stamp &&
158 tcp_tw_remember_stamp(tw))
Eric Dumazeted2e9232015-09-19 09:08:34 -0700159 inet_twsk_reschedule(tw, tw->tw_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 else
Eric Dumazeted2e9232015-09-19 09:08:34 -0700161 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 return TCP_TW_ACK;
163 }
164
165 /*
166 * Now real TIME-WAIT state.
167 *
168 * RFC 1122:
169 * "When a connection is [...] on TIME-WAIT state [...]
170 * [a TCP] MAY accept a new SYN from the remote TCP to
171 * reopen the connection directly, if it:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900172 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 * (1) assigns its initial sequence number for the new
174 * connection to be larger than the largest sequence
175 * number it used on the previous connection incarnation,
176 * and
177 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900178 * (2) returns to TIME-WAIT state if the SYN turns out
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 * to be an old duplicate".
180 */
181
182 if (!paws_reject &&
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700183 (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
185 /* In window segment, it may be only reset or bare ack. */
186
187 if (th->rst) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800188 /* This is TIME_WAIT assassination, in two flavors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 * Oh well... nobody has a sufficient solution to this
190 * protocol bug yet.
191 */
192 if (sysctl_tcp_rfc1337 == 0) {
193kill:
Eric Dumazetdbe7faa2015-07-08 14:28:30 -0700194 inet_twsk_deschedule_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 return TCP_TW_SUCCESS;
196 }
197 }
Eric Dumazeted2e9232015-09-19 09:08:34 -0700198 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
200 if (tmp_opt.saw_tstamp) {
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700201 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
James Morris9d729f72007-03-04 16:12:44 -0800202 tcptw->tw_ts_recent_stamp = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 }
204
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700205 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 return TCP_TW_SUCCESS;
207 }
208
209 /* Out of window segment.
210
211 All the segments are ACKed immediately.
212
213 The only exception is new SYN. We accept it, if it is
214 not old duplicate and we are not in danger to be killed
215 by delayed old duplicates. RFC check is that it has
216 newer sequence number works at rates <40Mbit/sec.
217 However, if paws works, it is reliable AND even more,
218 we even may relax silly seq space cutoff.
219
220 RED-PEN: we violate main RFC requirement, if this SYN will appear
221 old duplicate (i.e. we receive RST in reply to SYN-ACK),
222 we must return socket to time-wait state. It is not good,
223 but not fatal yet.
224 */
225
226 if (th->syn && !th->rst && !th->ack && !paws_reject &&
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700227 (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
228 (tmp_opt.saw_tstamp &&
229 (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
230 u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 if (isn == 0)
232 isn++;
Eric Dumazet04317da2014-09-05 15:33:32 -0700233 TCP_SKB_CB(skb)->tcp_tw_isn = isn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return TCP_TW_SYN;
235 }
236
237 if (paws_reject)
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700238 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800240 if (!th->rst) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 /* In this case we must reset the TIMEWAIT timer.
242 *
243 * If it is ACKless SYN it may be both old duplicate
244 * and new good SYN with random sequence number <rcv_nxt.
245 * Do not reschedule in the last case.
246 */
247 if (paws_reject || th->ack)
Eric Dumazeted2e9232015-09-19 09:08:34 -0700248 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Neal Cardwell4fb17a62015-02-06 16:04:41 -0500250 return tcp_timewait_check_oow_rate_limit(
251 tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 }
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700253 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 return TCP_TW_SUCCESS;
255}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000256EXPORT_SYMBOL(tcp_timewait_state_process);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900258/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 * Move a socket to time-wait or dead fin-wait-2 state.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900260 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261void tcp_time_wait(struct sock *sk, int state, int timeo)
262{
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800263 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700264 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazet789f5582015-04-12 18:51:09 -0700265 struct inet_timewait_sock *tw;
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000266 bool recycle_ok = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
David S. Millerb6242b92012-07-10 03:27:56 -0700268 if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
David S. Miller3f419d22010-11-29 13:37:14 -0800269 recycle_ok = tcp_remember_stamp(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
Eric Dumazet789f5582015-04-12 18:51:09 -0700271 tw = inet_twsk_alloc(sk, &tcp_death_row, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Ian Morris00db4122015-04-03 09:17:27 +0100273 if (tw) {
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700274 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700275 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
David S. Miller23978492012-06-09 14:56:12 -0700276 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700277
David S. Miller23978492012-06-09 14:56:12 -0700278 tw->tw_transparent = inet->transparent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700280 tcptw->tw_rcv_nxt = tp->rcv_nxt;
281 tcptw->tw_snd_nxt = tp->snd_nxt;
282 tcptw->tw_rcv_wnd = tcp_receive_window(tp);
283 tcptw->tw_ts_recent = tp->rx_opt.ts_recent;
284 tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000285 tcptw->tw_ts_offset = tp->tsoffset;
Neal Cardwell4fb17a62015-02-06 16:04:41 -0500286 tcptw->tw_last_oow_ack_time = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000288#if IS_ENABLED(CONFIG_IPV6)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 if (tw->tw_family == PF_INET6) {
290 struct ipv6_pinfo *np = inet6_sk(sk);
291
Eric Dumazetefe42082013-10-03 15:42:29 -0700292 tw->tw_v6_daddr = sk->sk_v6_daddr;
293 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
Eric Dumazetb903d322011-10-27 00:44:35 -0400294 tw->tw_tclass = np->tclass;
Florent Fourcot21858cd2015-05-16 00:24:59 +0200295 tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
Eric Dumazet9fe516b2014-06-27 08:36:16 -0700296 tw->tw_ipv6only = sk->sk_ipv6only;
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298#endif
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800299
300#ifdef CONFIG_TCP_MD5SIG
301 /*
302 * The timewait bucket does not have the key DB from the
303 * sock structure. We just make a quick copy of the
304 * md5 key being used (if indeed we are using one)
305 * so the timewait ack generating code has the key.
306 */
307 do {
308 struct tcp_md5sig_key *key;
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000309 tcptw->tw_md5_key = NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800310 key = tp->af_specific->md5_lookup(sk, sk);
Ian Morris00db4122015-04-03 09:17:27 +0100311 if (key) {
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000312 tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
Eric Dumazet71cea172013-05-20 06:52:26 +0000313 if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool())
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800314 BUG();
315 }
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800316 } while (0);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800317#endif
318
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 /* Get the TIME_WAIT timeout firing. */
320 if (timeo < rto)
321 timeo = rto;
322
323 if (recycle_ok) {
324 tw->tw_timeout = rto;
325 } else {
326 tw->tw_timeout = TCP_TIMEWAIT_LEN;
327 if (state == TCP_TIME_WAIT)
328 timeo = TCP_TIMEWAIT_LEN;
329 }
330
Eric Dumazet789f5582015-04-12 18:51:09 -0700331 inet_twsk_schedule(tw, timeo);
Eric Dumazeted2e9232015-09-19 09:08:34 -0700332 /* Linkage updates. */
333 __inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700334 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 } else {
336 /* Sorry, if we're out of memory, just CLOSE this
337 * socket up. We've got bigger problems than
338 * non-graceful socket closings.
339 */
Tom Herbert67631512010-12-08 12:16:33 -0800340 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 }
342
343 tcp_update_metrics(sk);
344 tcp_done(sk);
345}
346
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800347void tcp_twsk_destructor(struct sock *sk)
348{
David S. Millerb6242b92012-07-10 03:27:56 -0700349#ifdef CONFIG_TCP_MD5SIG
David S. Millera9286302006-11-14 19:53:22 -0800350 struct tcp_timewait_sock *twsk = tcp_twsk(sk);
David S. Miller23978492012-06-09 14:56:12 -0700351
Eric Dumazet71cea172013-05-20 06:52:26 +0000352 if (twsk->tw_md5_key)
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000353 kfree_rcu(twsk->tw_md5_key, rcu);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800354#endif
355}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800356EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
357
Eric Dumazetb1964b52015-09-25 07:39:09 -0700358/* Warning : This function is called without sk_listener being locked.
359 * Be sure to read socket fields once, as their value could change under us.
360 */
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700361void tcp_openreq_init_rwin(struct request_sock *req,
Eric Dumazetb1964b52015-09-25 07:39:09 -0700362 const struct sock *sk_listener,
363 const struct dst_entry *dst)
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700364{
365 struct inet_request_sock *ireq = inet_rsk(req);
Eric Dumazetb1964b52015-09-25 07:39:09 -0700366 const struct tcp_sock *tp = tcp_sk(sk_listener);
367 u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
368 int full_space = tcp_full_space(sk_listener);
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700369 int mss = dst_metric_advmss(dst);
Eric Dumazetb1964b52015-09-25 07:39:09 -0700370 u32 window_clamp;
371 __u8 rcv_wscale;
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700372
Eric Dumazetb1964b52015-09-25 07:39:09 -0700373 if (user_mss && user_mss < mss)
374 mss = user_mss;
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700375
Eric Dumazetb1964b52015-09-25 07:39:09 -0700376 window_clamp = READ_ONCE(tp->window_clamp);
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700377 /* Set this up on the first call only */
Eric Dumazeted53d0a2015-10-08 19:33:23 -0700378 req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW);
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700379
380 /* limit the window selection if the user enforce a smaller rx buffer */
Eric Dumazetb1964b52015-09-25 07:39:09 -0700381 if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK &&
Eric Dumazeted53d0a2015-10-08 19:33:23 -0700382 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
383 req->rsk_window_clamp = full_space;
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700384
385 /* tcp_full_space because it is guaranteed to be the first packet */
Eric Dumazetb1964b52015-09-25 07:39:09 -0700386 tcp_select_initial_window(full_space,
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700387 mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
Eric Dumazeted53d0a2015-10-08 19:33:23 -0700388 &req->rsk_rcv_wnd,
389 &req->rsk_window_clamp,
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700390 ireq->wscale_ok,
391 &rcv_wscale,
392 dst_metric(dst, RTAX_INITRWND));
393 ireq->rcv_wscale = rcv_wscale;
394}
395EXPORT_SYMBOL(tcp_openreq_init_rwin);
396
Florian Westphal735d3832014-09-29 13:08:30 +0200397static void tcp_ecn_openreq_child(struct tcp_sock *tp,
398 const struct request_sock *req)
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700399{
400 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
401}
402
Daniel Borkmann81164412015-01-05 23:57:48 +0100403void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
404{
405 struct inet_connection_sock *icsk = inet_csk(sk);
406 u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
407 bool ca_got_dst = false;
408
409 if (ca_key != TCP_CA_UNSPEC) {
410 const struct tcp_congestion_ops *ca;
411
412 rcu_read_lock();
413 ca = tcp_ca_find_key(ca_key);
414 if (likely(ca && try_module_get(ca->owner))) {
415 icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
416 icsk->icsk_ca_ops = ca;
417 ca_got_dst = true;
418 }
419 rcu_read_unlock();
420 }
421
Neal Cardwell9f950412015-05-29 13:47:07 -0400422 /* If no valid choice made yet, assign current system default ca. */
423 if (!ca_got_dst &&
424 (!icsk->icsk_ca_setsockopt ||
425 !try_module_get(icsk->icsk_ca_ops->owner)))
Daniel Borkmann81164412015-01-05 23:57:48 +0100426 tcp_assign_congestion_control(sk);
427
428 tcp_set_ca_state(sk, TCP_CA_Open);
429}
430EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432/* This is not only more efficient than what we used to do, it eliminates
433 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
434 *
435 * Actually, we could lots of memory writes here. tp of listening
436 * socket contains all necessary default parameters.
437 */
Eric Dumazetc28c6f02015-09-29 07:42:47 -0700438struct sock *tcp_create_openreq_child(const struct sock *sk,
439 struct request_sock *req,
440 struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
Eric Dumazete56c57d2011-11-08 17:07:07 -0500442 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
Ian Morris00db4122015-04-03 09:17:27 +0100444 if (newsk) {
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700445 const struct inet_request_sock *ireq = inet_rsk(req);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700446 struct tcp_request_sock *treq = tcp_rsk(req);
Arnaldo Carvalho de Meloa9948a72007-02-28 11:05:56 -0800447 struct inet_connection_sock *newicsk = inet_csk(newsk);
William Allen Simpson435cf552009-12-02 18:17:05 +0000448 struct tcp_sock *newtp = tcp_sk(newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 /* Now setup tcp_sock */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 newtp->pred_flags = 0;
William Allen Simpson435cf552009-12-02 18:17:05 +0000452
453 newtp->rcv_wup = newtp->copied_seq =
454 newtp->rcv_nxt = treq->rcv_isn + 1;
Eric Dumazeta9d99ce2016-03-06 09:29:21 -0800455 newtp->segs_in = 1;
William Allen Simpson435cf552009-12-02 18:17:05 +0000456
457 newtp->snd_sml = newtp->snd_una =
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000458 newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
460 tcp_prequeue_init(newtp);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000461 INIT_LIST_HEAD(&newtp->tsq_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Hantzis Fotisee7537b2009-03-02 22:42:02 -0800463 tcp_init_wl(newtp, treq->rcv_isn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Eric Dumazet740b0f12014-02-26 14:02:48 -0800465 newtp->srtt_us = 0;
466 newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
Yuchung Chengf6722582015-10-16 21:57:42 -0700467 newtp->rtt_min[0].rtt = ~0U;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700468 newicsk->icsk_rto = TCP_TIMEOUT_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 newtp->packets_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 newtp->retrans_out = 0;
472 newtp->sacked_out = 0;
473 newtp->fackets_out = 0;
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -0700474 newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000475 tcp_enable_early_retrans(newtp);
Nandita Dukkipati9b717a82013-03-11 10:00:44 +0000476 newtp->tlp_high_seq = 0;
Yuchung Cheng0f1c28a2015-09-18 11:36:14 -0700477 newtp->lsndtime = treq->snt_synack.stamp_jiffies;
Eric Dumazetd8ed6252015-09-22 20:44:17 -0700478 newsk->sk_txhash = treq->txhash;
Neal Cardwellf2b2c582015-02-06 16:04:40 -0500479 newtp->last_oow_ack_time = 0;
Yuchung Cheng375fe022013-07-22 16:20:45 -0700480 newtp->total_retrans = req->num_retrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 /* So many TCP implementations out there (incorrectly) count the
483 * initial SYN frame in their delayed-ACK and congestion control
484 * algorithms that we must have the following bandaid to talk
485 * efficiently to them. -DaveM
486 */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000487 newtp->snd_cwnd = TCP_INIT_CWND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 newtp->snd_cwnd_cnt = 0;
489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 tcp_init_xmit_timers(newsk);
Eric Dumazet996b1752014-01-06 09:36:12 -0800491 __skb_queue_head_init(&newtp->out_of_order_queue);
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000492 newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 newtp->rx_opt.saw_tstamp = 0;
495
496 newtp->rx_opt.dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 newtp->rx_opt.num_sacks = 0;
Ilpo Järvinencabeccb2009-02-28 04:44:38 +0000498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 newtp->urg_data = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 if (sock_flag(newsk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700502 inet_csk_reset_keepalive_timer(newsk,
503 keepalive_time_when(newtp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700505 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800506 if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (sysctl_tcp_fack)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300508 tcp_enable_fack(newtp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 }
Eric Dumazeted53d0a2015-10-08 19:33:23 -0700510 newtp->window_clamp = req->rsk_window_clamp;
511 newtp->rcv_ssthresh = req->rsk_rcv_wnd;
512 newtp->rcv_wnd = req->rsk_rcv_wnd;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700513 newtp->rx_opt.wscale_ok = ireq->wscale_ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (newtp->rx_opt.wscale_ok) {
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700515 newtp->rx_opt.snd_wscale = ireq->snd_wscale;
516 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 } else {
518 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
519 newtp->window_clamp = min(newtp->window_clamp, 65535U);
520 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700521 newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
522 newtp->rx_opt.snd_wscale);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 newtp->max_window = newtp->snd_wnd;
524
525 if (newtp->rx_opt.tstamp_ok) {
526 newtp->rx_opt.ts_recent = req->ts_recent;
James Morris9d729f72007-03-04 16:12:44 -0800527 newtp->rx_opt.ts_recent_stamp = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
529 } else {
530 newtp->rx_opt.ts_recent_stamp = 0;
531 newtp->tcp_header_len = sizeof(struct tcphdr);
532 }
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000533 newtp->tsoffset = 0;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800534#ifdef CONFIG_TCP_MD5SIG
535 newtp->md5sig_info = NULL; /*XXX*/
536 if (newtp->af_specific->md5_lookup(sk, newsk))
537 newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
538#endif
William Allen Simpsonbee7ca92009-11-10 09:51:18 +0000539 if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700540 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 newtp->rx_opt.mss_clamp = req->mss;
Florian Westphal735d3832014-09-29 13:08:30 +0200542 tcp_ecn_openreq_child(newtp, req);
Jerry Chu83368862012-08-31 12:29:12 +0000543 newtp->fastopen_rsk = NULL;
Yuchung Cheng6f736012012-10-19 15:14:44 +0000544 newtp->syn_data_acked = 0;
Yuchung Cheng659a8ad2015-10-16 21:57:46 -0700545 newtp->rack.mstamp.v64 = 0;
546 newtp->rack.advanced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700548 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
550 return newsk;
551}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000552EXPORT_SYMBOL(tcp_create_openreq_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900554/*
Jerry Chu83368862012-08-31 12:29:12 +0000555 * Process an incoming packet for SYN_RECV sockets represented as a
556 * request_sock. Normally sk is the listener socket but for TFO it
557 * points to the child socket.
558 *
559 * XXX (TFO) - The current impl contains a special check for ack
560 * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
Alan Cox4308fc52012-09-19 15:46:06 +0100561 *
562 * We don't need to initialize tmp_opt.sack_ok as we don't use the results
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 */
564
Jianjun Kong5a5f3a82008-11-03 00:24:34 -0800565struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700566 struct request_sock *req,
Jerry Chu83368862012-08-31 12:29:12 +0000567 bool fastopen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
William Allen Simpson4957faade2009-12-02 18:25:27 +0000569 struct tcp_options_received tmp_opt;
William Allen Simpson4957faade2009-12-02 18:25:27 +0000570 struct sock *child;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700571 const struct tcphdr *th = tcp_hdr(skb);
Al Viro714e85b2006-11-14 20:51:49 -0800572 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000573 bool paws_reject = false;
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700574 bool own_req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
David S. Millerbb5b7c12009-12-15 20:56:42 -0800576 tmp_opt.saw_tstamp = 0;
577 if (th->doff > (sizeof(struct tcphdr)>>2)) {
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000578 tcp_parse_options(skb, &tmp_opt, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 if (tmp_opt.saw_tstamp) {
581 tmp_opt.ts_recent = req->ts_recent;
582 /* We do not store true stamp, but it is not required,
583 * it can be estimated (approximately)
584 * from another data.
585 */
Eric Dumazete6c022a2012-10-27 23:16:46 +0000586 tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +0000587 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
589 }
590
591 /* Check for pure retransmitted SYN. */
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700592 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 flg == TCP_FLAG_SYN &&
594 !paws_reject) {
595 /*
596 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
597 * this case on figure 6 and figure 8, but formal
598 * protocol description says NOTHING.
599 * To be more exact, it says that we should send ACK,
600 * because this segment (at least, if it has no data)
601 * is out of window.
602 *
603 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT
604 * describe SYN-RECV state. All the description
605 * is wrong, we cannot believe to it and should
606 * rely only on common sense and implementation
607 * experience.
608 *
609 * Enforce "SYN-ACK" according to figure 8, figure 6
610 * of RFC793, fixed by RFC1122.
Jerry Chu83368862012-08-31 12:29:12 +0000611 *
612 * Note that even if there is new data in the SYN packet
613 * they will be thrown away too.
Yuchung Chengcd75eff2013-04-29 08:44:51 +0000614 *
615 * Reset timer after retransmitting SYNACK, similar to
616 * the idea of fast retransmit in recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 */
Neal Cardwella9b2c062015-02-06 16:04:39 -0500618 if (!tcp_oow_rate_limited(sock_net(sk), skb,
619 LINUX_MIB_TCPACKSKIPPEDSYNRECV,
620 &tcp_rsk(req)->last_oow_ack_time) &&
621
Eric Dumazetdd929c12015-04-08 15:34:04 -0700622 !inet_rtx_syn_ack(sk, req)) {
623 unsigned long expires = jiffies;
624
625 expires += min(TCP_TIMEOUT_INIT << req->num_timeout,
626 TCP_RTO_MAX);
627 if (!fastopen)
628 mod_timer_pending(&req->rsk_timer, expires);
629 else
630 req->rsk_timer.expires = expires;
631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 return NULL;
633 }
634
635 /* Further reproduces section "SEGMENT ARRIVES"
636 for state SYN-RECEIVED of RFC793.
637 It is broken, however, it does not work only
638 when SYNs are crossed.
639
640 You would think that SYN crossing is impossible here, since
641 we should have a SYN_SENT socket (from connect()) on our end,
642 but this is not true if the crossed SYNs were sent to both
643 ends by a malicious third party. We must defend against this,
644 and to do that we first verify the ACK (as per RFC793, page
645 36) and reset if it is invalid. Is this a true full defense?
646 To convince ourselves, let us consider a way in which the ACK
647 test can still pass in this 'malicious crossed SYNs' case.
648 Malicious sender sends identical SYNs (and thus identical sequence
649 numbers) to both A and B:
650
651 A: gets SYN, seq=7
652 B: gets SYN, seq=7
653
654 By our good fortune, both A and B select the same initial
655 send sequence number of seven :-)
656
657 A: sends SYN|ACK, seq=7, ack_seq=8
658 B: sends SYN|ACK, seq=7, ack_seq=8
659
660 So we are now A eating this SYN|ACK, ACK test passes. So
661 does sequence test, SYN is truncated, and thus we consider
662 it a bare ACK.
663
David S. Millerec0a1962008-06-12 16:31:35 -0700664 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
665 bare ACK. Otherwise, we create an established connection. Both
666 ends (listening sockets) accept the new incoming connection and try
667 to talk to each other. 8-)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
669 Note: This case is both harmless, and rare. Possibility is about the
670 same as us discovering intelligent life on another plant tomorrow.
671
672 But generally, we should (RFC lies!) to accept ACK
673 from SYNACK both here and in tcp_rcv_state_process().
674 tcp_rcv_state_process() does not, hence, we do not too.
675
676 Note that the case is absolutely generic:
677 we cannot optimize anything here without
678 violating protocol. All the checks must be made
679 before attempt to create socket.
680 */
681
682 /* RFC793 page 36: "If the connection is in any non-synchronized state ...
683 * and the incoming segment acknowledges something not yet
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800684 * sent (the segment carries an unacceptable ACK) ...
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 * a reset is sent."
686 *
Jerry Chu83368862012-08-31 12:29:12 +0000687 * Invalid ACK: reset will be sent by listening socket.
688 * Note that the ACK validity check for a Fast Open socket is done
689 * elsewhere and is checked directly against the child socket rather
690 * than req because user data may have been sent out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 */
Jerry Chu83368862012-08-31 12:29:12 +0000692 if ((flg & TCP_FLAG_ACK) && !fastopen &&
William Allen Simpson435cf552009-12-02 18:17:05 +0000693 (TCP_SKB_CB(skb)->ack_seq !=
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000694 tcp_rsk(req)->snt_isn + 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 return sk;
696
697 /* Also, it would be not so bad idea to check rcv_tsecr, which
698 * is essentially ACK extension and too early or too late values
699 * should cause reset in unsynchronized states.
700 */
701
702 /* RFC793: "first check sequence number". */
703
704 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
Eric Dumazeted53d0a2015-10-08 19:33:23 -0700705 tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 /* Out of window: send ACK and drop. */
Eric Dumazet4ce7e932016-04-01 08:52:22 -0700707 if (!(flg & TCP_FLAG_RST) &&
708 !tcp_oow_rate_limited(sock_net(sk), skb,
709 LINUX_MIB_TCPACKSKIPPEDSYNRECV,
710 &tcp_rsk(req)->last_oow_ack_time))
Gui Jianfeng6edafaa2008-08-06 23:50:04 -0700711 req->rsk_ops->send_ack(sk, skb, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 if (paws_reject)
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700713 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 return NULL;
715 }
716
717 /* In sequence, PAWS is OK. */
718
Jerry Chu83368862012-08-31 12:29:12 +0000719 if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
Adam Langley2aaab9a2008-08-07 20:27:45 -0700720 req->ts_recent = tmp_opt.rcv_tsval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Adam Langley2aaab9a2008-08-07 20:27:45 -0700722 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
723 /* Truncate SYN, it is out of window starting
724 at tcp_rsk(req)->rcv_isn + 1. */
725 flg &= ~TCP_FLAG_SYN;
726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Adam Langley2aaab9a2008-08-07 20:27:45 -0700728 /* RFC793: "second check the RST bit" and
729 * "fourth, check the SYN bit"
730 */
731 if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
732 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
733 goto embryonic_reset;
734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Adam Langley2aaab9a2008-08-07 20:27:45 -0700736 /* ACK sequence verified above, just make sure ACK is
737 * set. If ACK not set, just silently drop the packet.
Jerry Chu83368862012-08-31 12:29:12 +0000738 *
739 * XXX (TFO) - if we ever allow "data after SYN", the
740 * following check needs to be removed.
Adam Langley2aaab9a2008-08-07 20:27:45 -0700741 */
742 if (!(flg & TCP_FLAG_ACK))
743 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Jerry Chu83368862012-08-31 12:29:12 +0000745 /* For Fast Open no more processing is needed (sk is the
746 * child socket).
747 */
748 if (fastopen)
749 return sk;
750
Julian Anastasovd1b99ba2009-10-19 10:01:56 +0000751 /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
Eric Dumazete6c022a2012-10-27 23:16:46 +0000752 if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
Adam Langley2aaab9a2008-08-07 20:27:45 -0700753 TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
754 inet_rsk(req)->acked = 1;
Eric Dumazet907cdda2010-03-19 05:37:18 +0000755 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
Adam Langley2aaab9a2008-08-07 20:27:45 -0700756 return NULL;
757 }
David S. Millerec0a1962008-06-12 16:31:35 -0700758
Adam Langley2aaab9a2008-08-07 20:27:45 -0700759 /* OK, ACK is valid, create big socket and
760 * feed this segment to it. It will repeat all
761 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
762 * ESTABLISHED STATE. If it will be dropped after
763 * socket is created, wait for troubles.
764 */
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700765 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
766 req, &own_req);
Ian Morris51456b22015-04-03 09:17:26 +0100767 if (!child)
Adam Langley2aaab9a2008-08-07 20:27:45 -0700768 goto listen_overflow;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Eric Dumazet6bcfd7f2015-10-08 11:16:48 -0700770 sock_rps_save_rxhash(child, skb);
Yuchung Cheng0f1c28a2015-09-18 11:36:14 -0700771 tcp_synack_rtt_meas(child, req);
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700772 return inet_csk_complete_hashdance(sk, child, req, own_req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Adam Langley2aaab9a2008-08-07 20:27:45 -0700774listen_overflow:
775 if (!sysctl_tcp_abort_on_overflow) {
776 inet_rsk(req)->acked = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 return NULL;
Adam Langley2aaab9a2008-08-07 20:27:45 -0700778 }
779
780embryonic_reset:
Jerry Chu83368862012-08-31 12:29:12 +0000781 if (!(flg & TCP_FLAG_RST)) {
782 /* Received a bad SYN pkt - for TFO We try not to reset
783 * the local connection unless it's really necessary to
784 * avoid becoming vulnerable to outside attack aiming at
785 * resetting legit local connections.
786 */
Adam Langley2aaab9a2008-08-07 20:27:45 -0700787 req->rsk_ops->send_reset(sk, skb);
Jerry Chu83368862012-08-31 12:29:12 +0000788 } else if (fastopen) { /* received a valid RST pkt */
789 reqsk_fastopen_remove(sk, req, true);
790 tcp_reset(sk);
791 }
792 if (!fastopen) {
Eric Dumazet52452c52015-03-19 19:04:19 -0700793 inet_csk_reqsk_queue_drop(sk, req);
Jerry Chu83368862012-08-31 12:29:12 +0000794 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
795 }
Adam Langley2aaab9a2008-08-07 20:27:45 -0700796 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000798EXPORT_SYMBOL(tcp_check_req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
800/*
801 * Queue segment on the new socket if the new socket is active,
802 * otherwise we just shortcircuit this and continue with
803 * the new socket.
Jerry Chu83368862012-08-31 12:29:12 +0000804 *
805 * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
806 * when entering. But other states are possible due to a race condition
807 * where after __inet_lookup_established() fails but before the listener
808 * locked is obtained, other packets cause the same connection to
809 * be created.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 */
811
812int tcp_child_process(struct sock *parent, struct sock *child,
813 struct sk_buff *skb)
814{
815 int ret = 0;
816 int state = child->sk_state;
817
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -0700818 tcp_segs_in(tcp_sk(child), skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 if (!sock_owned_by_user(child)) {
Eric Dumazet72ab4a82015-09-29 07:42:41 -0700820 ret = tcp_rcv_state_process(child, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 /* Wakeup parent, send SIGIO */
822 if (state == TCP_SYN_RECV && child->sk_state != state)
David S. Miller676d2362014-04-11 16:15:36 -0400823 parent->sk_data_ready(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 } else {
825 /* Alas, it is possible again, because we do lookup
826 * in main socket hash table and lock on listening
827 * socket does not protect us more.
828 */
Zhu Yia3a858f2010-03-04 18:01:47 +0000829 __sk_add_backlog(child, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 }
831
832 bh_unlock_sock(child);
833 sock_put(child);
834 return ret;
835}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836EXPORT_SYMBOL(tcp_child_process);