blob: dd11ac7798c626d9abe3fbada06fadc99eafe378 [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
David S. Millere994b7c2009-11-21 11:22:25 -080030int sysctl_tcp_syncookies __read_mostly = 1;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080031EXPORT_SYMBOL(sysctl_tcp_syncookies);
32
Brian Haleyab32ea52006-09-22 14:15:41 -070033int sysctl_tcp_abort_on_overflow __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070035struct inet_timewait_death_row tcp_death_row = {
36 .sysctl_max_tw_buckets = NR_FILE * 2,
37 .period = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
Ingo Molnare4d91912006-07-03 00:24:34 -070038 .death_lock = __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock),
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070039 .hashinfo = &tcp_hashinfo,
40 .tw_timer = TIMER_INITIALIZER(inet_twdr_hangman, 0,
41 (unsigned long)&tcp_death_row),
42 .twkill_work = __WORK_INITIALIZER(tcp_death_row.twkill_work,
David Howells65f27f32006-11-22 14:55:48 +000043 inet_twdr_twkill_work),
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070044/* Short-time timewait calendar */
45
46 .twcal_hand = -1,
47 .twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
48 (unsigned long)&tcp_death_row),
49};
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070050EXPORT_SYMBOL_GPL(tcp_death_row);
51
Eric Dumazeta2a385d2012-05-16 23:15:34 +000052static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053{
54 if (seq == s_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +000055 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 if (after(end_seq, s_win) && before(seq, e_win))
Eric Dumazeta2a385d2012-05-16 23:15:34 +000057 return true;
Eric Dumazeta02cec22010-09-22 20:43:57 +000058 return seq == e_win && seq == end_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059}
60
Neal Cardwell4fb17a62015-02-06 16:04:41 -050061static enum tcp_tw_status
62tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
63 const struct sk_buff *skb, int mib_idx)
64{
65 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
66
67 if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
68 &tcptw->tw_last_oow_ack_time)) {
69 /* Send ACK. Note, we do not put the bucket,
70 * it will be released by caller.
71 */
72 return TCP_TW_ACK;
73 }
74
75 /* We are rate-limiting, so just release the tw sock and drop skb. */
76 inet_twsk_put(tw);
77 return TCP_TW_SUCCESS;
78}
79
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090080/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 * * Main purpose of TIME-WAIT state is to close connection gracefully,
82 * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
83 * (and, probably, tail of data) and one or more our ACKs are lost.
84 * * What is TIME-WAIT timeout? It is associated with maximal packet
85 * lifetime in the internet, which results in wrong conclusion, that
86 * it is set to catch "old duplicate segments" wandering out of their path.
87 * It is not quite correct. This timeout is calculated so that it exceeds
88 * maximal retransmission timeout enough to allow to lose one (or more)
89 * segments sent by peer and our ACKs. This time may be calculated from RTO.
90 * * When TIME-WAIT socket receives RST, it means that another end
91 * finally closed and we are allowed to kill TIME-WAIT too.
92 * * Second purpose of TIME-WAIT is catching old duplicate segments.
93 * Well, certainly it is pure paranoia, but if we load TIME-WAIT
94 * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
95 * * If we invented some more clever way to catch duplicates
96 * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
97 *
98 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
99 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
100 * from the very beginning.
101 *
102 * NOTE. With recycling (and later with fin-wait-2) TW bucket
103 * is _not_ stateless. It means, that strictly speaking we must
104 * spinlock it. I do not want! Well, probability of misbehaviour
105 * is ridiculously low and, seems, we could use some mb() tricks
106 * to avoid misread sequence numbers, states etc. --ANK
Alan Cox4308fc52012-09-19 15:46:06 +0100107 *
108 * We don't need to initialize tmp_out.sack_ok as we don't use the results
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 */
110enum tcp_tw_status
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700111tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
112 const struct tcphdr *th)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
114 struct tcp_options_received tmp_opt;
William Allen Simpson4957faade2009-12-02 18:25:27 +0000115 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000116 bool paws_reject = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
David S. Millerbb5b7c12009-12-15 20:56:42 -0800118 tmp_opt.saw_tstamp = 0;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700119 if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000120 tcp_parse_options(skb, &tmp_opt, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
122 if (tmp_opt.saw_tstamp) {
Andrey Vaginee684b62013-02-11 05:50:19 +0000123 tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700124 tmp_opt.ts_recent = tcptw->tw_ts_recent;
125 tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +0000126 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 }
128 }
129
130 if (tw->tw_substate == TCP_FIN_WAIT2) {
131 /* Just repeat all the checks of tcp_rcv_state_process() */
132
133 /* Out of window, send ACK */
134 if (paws_reject ||
135 !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700136 tcptw->tw_rcv_nxt,
137 tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
Neal Cardwell4fb17a62015-02-06 16:04:41 -0500138 return tcp_timewait_check_oow_rate_limit(
139 tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
141 if (th->rst)
142 goto kill;
143
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700144 if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 goto kill_with_rst;
146
147 /* Dup ACK? */
Wei Yongjun1ac530b2009-06-24 22:29:31 +0000148 if (!th->ack ||
149 !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700151 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 return TCP_TW_SUCCESS;
153 }
154
155 /* New data or FIN. If new data arrive after half-duplex close,
156 * reset.
157 */
158 if (!th->fin ||
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700159 TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160kill_with_rst:
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700161 inet_twsk_deschedule(tw, &tcp_death_row);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700162 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 return TCP_TW_RST;
164 }
165
166 /* FIN arrived, enter true time-wait state. */
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700167 tw->tw_substate = TCP_TIME_WAIT;
168 tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 if (tmp_opt.saw_tstamp) {
James Morris9d729f72007-03-04 16:12:44 -0800170 tcptw->tw_ts_recent_stamp = get_seconds();
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700171 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 }
173
David S. Millerccb7c412010-12-01 18:09:13 -0800174 if (tcp_death_row.sysctl_tw_recycle &&
175 tcptw->tw_ts_recent_stamp &&
176 tcp_tw_remember_stamp(tw))
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700177 inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout,
178 TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 else
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700180 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
181 TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 return TCP_TW_ACK;
183 }
184
185 /*
186 * Now real TIME-WAIT state.
187 *
188 * RFC 1122:
189 * "When a connection is [...] on TIME-WAIT state [...]
190 * [a TCP] MAY accept a new SYN from the remote TCP to
191 * reopen the connection directly, if it:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900192 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 * (1) assigns its initial sequence number for the new
194 * connection to be larger than the largest sequence
195 * number it used on the previous connection incarnation,
196 * and
197 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900198 * (2) returns to TIME-WAIT state if the SYN turns out
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 * to be an old duplicate".
200 */
201
202 if (!paws_reject &&
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700203 (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
205 /* In window segment, it may be only reset or bare ack. */
206
207 if (th->rst) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800208 /* This is TIME_WAIT assassination, in two flavors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 * Oh well... nobody has a sufficient solution to this
210 * protocol bug yet.
211 */
212 if (sysctl_tcp_rfc1337 == 0) {
213kill:
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700214 inet_twsk_deschedule(tw, &tcp_death_row);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700215 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 return TCP_TW_SUCCESS;
217 }
218 }
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700219 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
220 TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
222 if (tmp_opt.saw_tstamp) {
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700223 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
James Morris9d729f72007-03-04 16:12:44 -0800224 tcptw->tw_ts_recent_stamp = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 }
226
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700227 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 return TCP_TW_SUCCESS;
229 }
230
231 /* Out of window segment.
232
233 All the segments are ACKed immediately.
234
235 The only exception is new SYN. We accept it, if it is
236 not old duplicate and we are not in danger to be killed
237 by delayed old duplicates. RFC check is that it has
238 newer sequence number works at rates <40Mbit/sec.
239 However, if paws works, it is reliable AND even more,
240 we even may relax silly seq space cutoff.
241
242 RED-PEN: we violate main RFC requirement, if this SYN will appear
243 old duplicate (i.e. we receive RST in reply to SYN-ACK),
244 we must return socket to time-wait state. It is not good,
245 but not fatal yet.
246 */
247
248 if (th->syn && !th->rst && !th->ack && !paws_reject &&
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700249 (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
250 (tmp_opt.saw_tstamp &&
251 (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
252 u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 if (isn == 0)
254 isn++;
Eric Dumazet04317da2014-09-05 15:33:32 -0700255 TCP_SKB_CB(skb)->tcp_tw_isn = isn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 return TCP_TW_SYN;
257 }
258
259 if (paws_reject)
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700260 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800262 if (!th->rst) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 /* In this case we must reset the TIMEWAIT timer.
264 *
265 * If it is ACKless SYN it may be both old duplicate
266 * and new good SYN with random sequence number <rcv_nxt.
267 * Do not reschedule in the last case.
268 */
269 if (paws_reject || th->ack)
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700270 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
271 TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Neal Cardwell4fb17a62015-02-06 16:04:41 -0500273 return tcp_timewait_check_oow_rate_limit(
274 tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 }
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700276 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 return TCP_TW_SUCCESS;
278}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000279EXPORT_SYMBOL(tcp_timewait_state_process);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900281/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 * Move a socket to time-wait or dead fin-wait-2 state.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900283 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284void tcp_time_wait(struct sock *sk, int state, int timeo)
285{
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700286 struct inet_timewait_sock *tw = NULL;
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800287 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700288 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000289 bool recycle_ok = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
David S. Millerb6242b92012-07-10 03:27:56 -0700291 if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
David S. Miller3f419d22010-11-29 13:37:14 -0800292 recycle_ok = tcp_remember_stamp(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700294 if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets)
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700295 tw = inet_twsk_alloc(sk, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700297 if (tw != NULL) {
298 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700299 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
David S. Miller23978492012-06-09 14:56:12 -0700300 struct inet_sock *inet = inet_sk(sk);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700301
David S. Miller23978492012-06-09 14:56:12 -0700302 tw->tw_transparent = inet->transparent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700304 tcptw->tw_rcv_nxt = tp->rcv_nxt;
305 tcptw->tw_snd_nxt = tp->snd_nxt;
306 tcptw->tw_rcv_wnd = tcp_receive_window(tp);
307 tcptw->tw_ts_recent = tp->rx_opt.ts_recent;
308 tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000309 tcptw->tw_ts_offset = tp->tsoffset;
Neal Cardwell4fb17a62015-02-06 16:04:41 -0500310 tcptw->tw_last_oow_ack_time = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000312#if IS_ENABLED(CONFIG_IPV6)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 if (tw->tw_family == PF_INET6) {
314 struct ipv6_pinfo *np = inet6_sk(sk);
315
Eric Dumazetefe42082013-10-03 15:42:29 -0700316 tw->tw_v6_daddr = sk->sk_v6_daddr;
317 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
Eric Dumazetb903d322011-10-27 00:44:35 -0400318 tw->tw_tclass = np->tclass;
Florent Fourcot1d13a962014-01-16 17:21:22 +0100319 tw->tw_flowlabel = np->flow_label >> 12;
Eric Dumazet9fe516b2014-06-27 08:36:16 -0700320 tw->tw_ipv6only = sk->sk_ipv6only;
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700321 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322#endif
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800323
324#ifdef CONFIG_TCP_MD5SIG
325 /*
326 * The timewait bucket does not have the key DB from the
327 * sock structure. We just make a quick copy of the
328 * md5 key being used (if indeed we are using one)
329 * so the timewait ack generating code has the key.
330 */
331 do {
332 struct tcp_md5sig_key *key;
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000333 tcptw->tw_md5_key = NULL;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800334 key = tp->af_specific->md5_lookup(sk, sk);
335 if (key != NULL) {
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000336 tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
Eric Dumazet71cea172013-05-20 06:52:26 +0000337 if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool())
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800338 BUG();
339 }
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800340 } while (0);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800341#endif
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 /* Linkage updates. */
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700344 __inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 /* Get the TIME_WAIT timeout firing. */
347 if (timeo < rto)
348 timeo = rto;
349
350 if (recycle_ok) {
351 tw->tw_timeout = rto;
352 } else {
353 tw->tw_timeout = TCP_TIMEWAIT_LEN;
354 if (state == TCP_TIME_WAIT)
355 timeo = TCP_TIMEWAIT_LEN;
356 }
357
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700358 inet_twsk_schedule(tw, &tcp_death_row, timeo,
359 TCP_TIMEWAIT_LEN);
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700360 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 } else {
362 /* Sorry, if we're out of memory, just CLOSE this
363 * socket up. We've got bigger problems than
364 * non-graceful socket closings.
365 */
Tom Herbert67631512010-12-08 12:16:33 -0800366 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
368
369 tcp_update_metrics(sk);
370 tcp_done(sk);
371}
372
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800373void tcp_twsk_destructor(struct sock *sk)
374{
David S. Millerb6242b92012-07-10 03:27:56 -0700375#ifdef CONFIG_TCP_MD5SIG
David S. Millera9286302006-11-14 19:53:22 -0800376 struct tcp_timewait_sock *twsk = tcp_twsk(sk);
David S. Miller23978492012-06-09 14:56:12 -0700377
Eric Dumazet71cea172013-05-20 06:52:26 +0000378 if (twsk->tw_md5_key)
Eric Dumazeta915da9b2012-01-31 05:18:33 +0000379 kfree_rcu(twsk->tw_md5_key, rcu);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800380#endif
381}
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800382EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
383
Yuchung Cheng843f4a52014-05-11 20:22:11 -0700384void tcp_openreq_init_rwin(struct request_sock *req,
385 struct sock *sk, struct dst_entry *dst)
386{
387 struct inet_request_sock *ireq = inet_rsk(req);
388 struct tcp_sock *tp = tcp_sk(sk);
389 __u8 rcv_wscale;
390 int mss = dst_metric_advmss(dst);
391
392 if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < mss)
393 mss = tp->rx_opt.user_mss;
394
395 /* Set this up on the first call only */
396 req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW);
397
398 /* limit the window selection if the user enforce a smaller rx buffer */
399 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
400 (req->window_clamp > tcp_full_space(sk) || req->window_clamp == 0))
401 req->window_clamp = tcp_full_space(sk);
402
403 /* tcp_full_space because it is guaranteed to be the first packet */
404 tcp_select_initial_window(tcp_full_space(sk),
405 mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
406 &req->rcv_wnd,
407 &req->window_clamp,
408 ireq->wscale_ok,
409 &rcv_wscale,
410 dst_metric(dst, RTAX_INITRWND));
411 ireq->rcv_wscale = rcv_wscale;
412}
413EXPORT_SYMBOL(tcp_openreq_init_rwin);
414
Florian Westphal735d3832014-09-29 13:08:30 +0200415static void tcp_ecn_openreq_child(struct tcp_sock *tp,
416 const struct request_sock *req)
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700417{
418 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
419}
420
Daniel Borkmann81164412015-01-05 23:57:48 +0100421void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
422{
423 struct inet_connection_sock *icsk = inet_csk(sk);
424 u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
425 bool ca_got_dst = false;
426
427 if (ca_key != TCP_CA_UNSPEC) {
428 const struct tcp_congestion_ops *ca;
429
430 rcu_read_lock();
431 ca = tcp_ca_find_key(ca_key);
432 if (likely(ca && try_module_get(ca->owner))) {
433 icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
434 icsk->icsk_ca_ops = ca;
435 ca_got_dst = true;
436 }
437 rcu_read_unlock();
438 }
439
440 if (!ca_got_dst && !try_module_get(icsk->icsk_ca_ops->owner))
441 tcp_assign_congestion_control(sk);
442
443 tcp_set_ca_state(sk, TCP_CA_Open);
444}
445EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447/* This is not only more efficient than what we used to do, it eliminates
448 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
449 *
450 * Actually, we could lots of memory writes here. tp of listening
451 * socket contains all necessary default parameters.
452 */
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700453struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
Eric Dumazete56c57d2011-11-08 17:07:07 -0500455 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -0700457 if (newsk != NULL) {
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700458 const struct inet_request_sock *ireq = inet_rsk(req);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700459 struct tcp_request_sock *treq = tcp_rsk(req);
Arnaldo Carvalho de Meloa9948a72007-02-28 11:05:56 -0800460 struct inet_connection_sock *newicsk = inet_csk(newsk);
William Allen Simpson435cf552009-12-02 18:17:05 +0000461 struct tcp_sock *newtp = tcp_sk(newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 /* Now setup tcp_sock */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 newtp->pred_flags = 0;
William Allen Simpson435cf552009-12-02 18:17:05 +0000465
466 newtp->rcv_wup = newtp->copied_seq =
467 newtp->rcv_nxt = treq->rcv_isn + 1;
468
469 newtp->snd_sml = newtp->snd_una =
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000470 newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 tcp_prequeue_init(newtp);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000473 INIT_LIST_HEAD(&newtp->tsq_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
Hantzis Fotisee7537b2009-03-02 22:42:02 -0800475 tcp_init_wl(newtp, treq->rcv_isn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Eric Dumazet740b0f12014-02-26 14:02:48 -0800477 newtp->srtt_us = 0;
478 newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700479 newicsk->icsk_rto = TCP_TIMEOUT_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481 newtp->packets_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 newtp->retrans_out = 0;
483 newtp->sacked_out = 0;
484 newtp->fackets_out = 0;
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -0700485 newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000486 tcp_enable_early_retrans(newtp);
Nandita Dukkipati9b717a82013-03-11 10:00:44 +0000487 newtp->tlp_high_seq = 0;
Yuchung Cheng375fe022013-07-22 16:20:45 -0700488 newtp->lsndtime = treq->snt_synack;
Neal Cardwellf2b2c582015-02-06 16:04:40 -0500489 newtp->last_oow_ack_time = 0;
Yuchung Cheng375fe022013-07-22 16:20:45 -0700490 newtp->total_retrans = req->num_retrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
492 /* So many TCP implementations out there (incorrectly) count the
493 * initial SYN frame in their delayed-ACK and congestion control
494 * algorithms that we must have the following bandaid to talk
495 * efficiently to them. -DaveM
496 */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000497 newtp->snd_cwnd = TCP_INIT_CWND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 newtp->snd_cwnd_cnt = 0;
499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 tcp_init_xmit_timers(newsk);
Eric Dumazet996b1752014-01-06 09:36:12 -0800501 __skb_queue_head_init(&newtp->out_of_order_queue);
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000502 newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 newtp->rx_opt.saw_tstamp = 0;
505
506 newtp->rx_opt.dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 newtp->rx_opt.num_sacks = 0;
Ilpo Järvinencabeccb2009-02-28 04:44:38 +0000508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 newtp->urg_data = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 if (sock_flag(newsk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700512 inet_csk_reset_keepalive_timer(newsk,
513 keepalive_time_when(newtp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700515 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800516 if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 if (sysctl_tcp_fack)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300518 tcp_enable_fack(newtp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 }
520 newtp->window_clamp = req->window_clamp;
521 newtp->rcv_ssthresh = req->rcv_wnd;
522 newtp->rcv_wnd = req->rcv_wnd;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700523 newtp->rx_opt.wscale_ok = ireq->wscale_ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (newtp->rx_opt.wscale_ok) {
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700525 newtp->rx_opt.snd_wscale = ireq->snd_wscale;
526 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 } else {
528 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
529 newtp->window_clamp = min(newtp->window_clamp, 65535U);
530 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700531 newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
532 newtp->rx_opt.snd_wscale);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 newtp->max_window = newtp->snd_wnd;
534
535 if (newtp->rx_opt.tstamp_ok) {
536 newtp->rx_opt.ts_recent = req->ts_recent;
James Morris9d729f72007-03-04 16:12:44 -0800537 newtp->rx_opt.ts_recent_stamp = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
539 } else {
540 newtp->rx_opt.ts_recent_stamp = 0;
541 newtp->tcp_header_len = sizeof(struct tcphdr);
542 }
Andrey Vaginceaa1fe2013-02-11 05:50:17 +0000543 newtp->tsoffset = 0;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800544#ifdef CONFIG_TCP_MD5SIG
545 newtp->md5sig_info = NULL; /*XXX*/
546 if (newtp->af_specific->md5_lookup(sk, newsk))
547 newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
548#endif
William Allen Simpsonbee7ca92009-11-10 09:51:18 +0000549 if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700550 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 newtp->rx_opt.mss_clamp = req->mss;
Florian Westphal735d3832014-09-29 13:08:30 +0200552 tcp_ecn_openreq_child(newtp, req);
Jerry Chu83368862012-08-31 12:29:12 +0000553 newtp->fastopen_rsk = NULL;
Yuchung Cheng6f736012012-10-19 15:14:44 +0000554 newtp->syn_data_acked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700556 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
558 return newsk;
559}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000560EXPORT_SYMBOL(tcp_create_openreq_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900562/*
Jerry Chu83368862012-08-31 12:29:12 +0000563 * Process an incoming packet for SYN_RECV sockets represented as a
564 * request_sock. Normally sk is the listener socket but for TFO it
565 * points to the child socket.
566 *
567 * XXX (TFO) - The current impl contains a special check for ack
568 * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
Alan Cox4308fc52012-09-19 15:46:06 +0100569 *
570 * We don't need to initialize tmp_opt.sack_ok as we don't use the results
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 */
572
Jianjun Kong5a5f3a82008-11-03 00:24:34 -0800573struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700574 struct request_sock *req,
Jerry Chu83368862012-08-31 12:29:12 +0000575 struct request_sock **prev,
576 bool fastopen)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
William Allen Simpson4957faade2009-12-02 18:25:27 +0000578 struct tcp_options_received tmp_opt;
William Allen Simpson4957faade2009-12-02 18:25:27 +0000579 struct sock *child;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700580 const struct tcphdr *th = tcp_hdr(skb);
Al Viro714e85b2006-11-14 20:51:49 -0800581 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000582 bool paws_reject = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Jerry Chu83368862012-08-31 12:29:12 +0000584 BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN));
585
David S. Millerbb5b7c12009-12-15 20:56:42 -0800586 tmp_opt.saw_tstamp = 0;
587 if (th->doff > (sizeof(struct tcphdr)>>2)) {
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000588 tcp_parse_options(skb, &tmp_opt, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 if (tmp_opt.saw_tstamp) {
591 tmp_opt.ts_recent = req->ts_recent;
592 /* We do not store true stamp, but it is not required,
593 * it can be estimated (approximately)
594 * from another data.
595 */
Eric Dumazete6c022a2012-10-27 23:16:46 +0000596 tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +0000597 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 }
599 }
600
601 /* Check for pure retransmitted SYN. */
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700602 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 flg == TCP_FLAG_SYN &&
604 !paws_reject) {
605 /*
606 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
607 * this case on figure 6 and figure 8, but formal
608 * protocol description says NOTHING.
609 * To be more exact, it says that we should send ACK,
610 * because this segment (at least, if it has no data)
611 * is out of window.
612 *
613 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT
614 * describe SYN-RECV state. All the description
615 * is wrong, we cannot believe to it and should
616 * rely only on common sense and implementation
617 * experience.
618 *
619 * Enforce "SYN-ACK" according to figure 8, figure 6
620 * of RFC793, fixed by RFC1122.
Jerry Chu83368862012-08-31 12:29:12 +0000621 *
622 * Note that even if there is new data in the SYN packet
623 * they will be thrown away too.
Yuchung Chengcd75eff2013-04-29 08:44:51 +0000624 *
625 * Reset timer after retransmitting SYNACK, similar to
626 * the idea of fast retransmit in recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 */
Neal Cardwella9b2c062015-02-06 16:04:39 -0500628 if (!tcp_oow_rate_limited(sock_net(sk), skb,
629 LINUX_MIB_TCPACKSKIPPEDSYNRECV,
630 &tcp_rsk(req)->last_oow_ack_time) &&
631
632 !inet_rtx_syn_ack(sk, req))
Yuchung Chengcd75eff2013-04-29 08:44:51 +0000633 req->expires = min(TCP_TIMEOUT_INIT << req->num_timeout,
634 TCP_RTO_MAX) + jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 return NULL;
636 }
637
638 /* Further reproduces section "SEGMENT ARRIVES"
639 for state SYN-RECEIVED of RFC793.
640 It is broken, however, it does not work only
641 when SYNs are crossed.
642
643 You would think that SYN crossing is impossible here, since
644 we should have a SYN_SENT socket (from connect()) on our end,
645 but this is not true if the crossed SYNs were sent to both
646 ends by a malicious third party. We must defend against this,
647 and to do that we first verify the ACK (as per RFC793, page
648 36) and reset if it is invalid. Is this a true full defense?
649 To convince ourselves, let us consider a way in which the ACK
650 test can still pass in this 'malicious crossed SYNs' case.
651 Malicious sender sends identical SYNs (and thus identical sequence
652 numbers) to both A and B:
653
654 A: gets SYN, seq=7
655 B: gets SYN, seq=7
656
657 By our good fortune, both A and B select the same initial
658 send sequence number of seven :-)
659
660 A: sends SYN|ACK, seq=7, ack_seq=8
661 B: sends SYN|ACK, seq=7, ack_seq=8
662
663 So we are now A eating this SYN|ACK, ACK test passes. So
664 does sequence test, SYN is truncated, and thus we consider
665 it a bare ACK.
666
David S. Millerec0a1962008-06-12 16:31:35 -0700667 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
668 bare ACK. Otherwise, we create an established connection. Both
669 ends (listening sockets) accept the new incoming connection and try
670 to talk to each other. 8-)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
672 Note: This case is both harmless, and rare. Possibility is about the
673 same as us discovering intelligent life on another plant tomorrow.
674
675 But generally, we should (RFC lies!) to accept ACK
676 from SYNACK both here and in tcp_rcv_state_process().
677 tcp_rcv_state_process() does not, hence, we do not too.
678
679 Note that the case is absolutely generic:
680 we cannot optimize anything here without
681 violating protocol. All the checks must be made
682 before attempt to create socket.
683 */
684
685 /* RFC793 page 36: "If the connection is in any non-synchronized state ...
686 * and the incoming segment acknowledges something not yet
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800687 * sent (the segment carries an unacceptable ACK) ...
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 * a reset is sent."
689 *
Jerry Chu83368862012-08-31 12:29:12 +0000690 * Invalid ACK: reset will be sent by listening socket.
691 * Note that the ACK validity check for a Fast Open socket is done
692 * elsewhere and is checked directly against the child socket rather
693 * than req because user data may have been sent out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 */
Jerry Chu83368862012-08-31 12:29:12 +0000695 if ((flg & TCP_FLAG_ACK) && !fastopen &&
William Allen Simpson435cf552009-12-02 18:17:05 +0000696 (TCP_SKB_CB(skb)->ack_seq !=
Christoph Paasch1a2c6182013-03-17 08:23:34 +0000697 tcp_rsk(req)->snt_isn + 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 return sk;
699
700 /* Also, it would be not so bad idea to check rcv_tsecr, which
701 * is essentially ACK extension and too early or too late values
702 * should cause reset in unsynchronized states.
703 */
704
705 /* RFC793: "first check sequence number". */
706
707 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
Jerry Chu83368862012-08-31 12:29:12 +0000708 tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rcv_wnd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 /* Out of window: send ACK and drop. */
710 if (!(flg & TCP_FLAG_RST))
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 */
765 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
766 if (child == NULL)
767 goto listen_overflow;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Adam Langley2aaab9a2008-08-07 20:27:45 -0700769 inet_csk_reqsk_queue_unlink(sk, req, prev);
770 inet_csk_reqsk_queue_removed(sk, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Adam Langley2aaab9a2008-08-07 20:27:45 -0700772 inet_csk_reqsk_queue_add(sk, req, child);
773 return child;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Adam Langley2aaab9a2008-08-07 20:27:45 -0700775listen_overflow:
776 if (!sysctl_tcp_abort_on_overflow) {
777 inet_rsk(req)->acked = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 return NULL;
Adam Langley2aaab9a2008-08-07 20:27:45 -0700779 }
780
781embryonic_reset:
Jerry Chu83368862012-08-31 12:29:12 +0000782 if (!(flg & TCP_FLAG_RST)) {
783 /* Received a bad SYN pkt - for TFO We try not to reset
784 * the local connection unless it's really necessary to
785 * avoid becoming vulnerable to outside attack aiming at
786 * resetting legit local connections.
787 */
Adam Langley2aaab9a2008-08-07 20:27:45 -0700788 req->rsk_ops->send_reset(sk, skb);
Jerry Chu83368862012-08-31 12:29:12 +0000789 } else if (fastopen) { /* received a valid RST pkt */
790 reqsk_fastopen_remove(sk, req, true);
791 tcp_reset(sk);
792 }
793 if (!fastopen) {
794 inet_csk_reqsk_queue_drop(sk, req, prev);
795 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
796 }
Adam Langley2aaab9a2008-08-07 20:27:45 -0700797 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000799EXPORT_SYMBOL(tcp_check_req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801/*
802 * Queue segment on the new socket if the new socket is active,
803 * otherwise we just shortcircuit this and continue with
804 * the new socket.
Jerry Chu83368862012-08-31 12:29:12 +0000805 *
806 * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
807 * when entering. But other states are possible due to a race condition
808 * where after __inet_lookup_established() fails but before the listener
809 * locked is obtained, other packets cause the same connection to
810 * be created.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 */
812
813int tcp_child_process(struct sock *parent, struct sock *child,
814 struct sk_buff *skb)
815{
816 int ret = 0;
817 int state = child->sk_state;
818
819 if (!sock_owned_by_user(child)) {
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700820 ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
821 skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 /* Wakeup parent, send SIGIO */
823 if (state == TCP_SYN_RECV && child->sk_state != state)
David S. Miller676d2362014-04-11 16:15:36 -0400824 parent->sk_data_ready(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 } else {
826 /* Alas, it is possible again, because we do lookup
827 * in main socket hash table and lock on listening
828 * socket does not protect us more.
829 */
Zhu Yia3a858f2010-03-04 18:01:47 +0000830 __sk_add_backlog(child, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 }
832
833 bh_unlock_sock(child);
834 sock_put(child);
835 return ret;
836}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837EXPORT_SYMBOL(tcp_child_process);