blob: f976fc57892cf134e22f41255c4b8170f6dd02c9 [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>
23#include <linux/sysctl.h>
24#include <linux/workqueue.h>
25#include <net/tcp.h>
26#include <net/inet_common.h>
27#include <net/xfrm.h>
28
29#ifdef CONFIG_SYSCTL
30#define SYNC_INIT 0 /* let the user enable it */
31#else
32#define SYNC_INIT 1
33#endif
34
Brian Haleyab32ea52006-09-22 14:15:41 -070035int sysctl_tcp_syncookies __read_mostly = SYNC_INIT;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080036EXPORT_SYMBOL(sysctl_tcp_syncookies);
37
Brian Haleyab32ea52006-09-22 14:15:41 -070038int sysctl_tcp_abort_on_overflow __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070040struct inet_timewait_death_row tcp_death_row = {
41 .sysctl_max_tw_buckets = NR_FILE * 2,
42 .period = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
Ingo Molnare4d91912006-07-03 00:24:34 -070043 .death_lock = __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock),
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070044 .hashinfo = &tcp_hashinfo,
45 .tw_timer = TIMER_INITIALIZER(inet_twdr_hangman, 0,
46 (unsigned long)&tcp_death_row),
47 .twkill_work = __WORK_INITIALIZER(tcp_death_row.twkill_work,
David Howells65f27f32006-11-22 14:55:48 +000048 inet_twdr_twkill_work),
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070049/* Short-time timewait calendar */
50
51 .twcal_hand = -1,
52 .twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
53 (unsigned long)&tcp_death_row),
54};
55
56EXPORT_SYMBOL_GPL(tcp_death_row);
57
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
59{
60 if (seq == s_win)
61 return 1;
62 if (after(end_seq, s_win) && before(seq, e_win))
63 return 1;
64 return (seq == e_win && seq == end_seq);
65}
66
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090067/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * * Main purpose of TIME-WAIT state is to close connection gracefully,
69 * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
70 * (and, probably, tail of data) and one or more our ACKs are lost.
71 * * What is TIME-WAIT timeout? It is associated with maximal packet
72 * lifetime in the internet, which results in wrong conclusion, that
73 * it is set to catch "old duplicate segments" wandering out of their path.
74 * It is not quite correct. This timeout is calculated so that it exceeds
75 * maximal retransmission timeout enough to allow to lose one (or more)
76 * segments sent by peer and our ACKs. This time may be calculated from RTO.
77 * * When TIME-WAIT socket receives RST, it means that another end
78 * finally closed and we are allowed to kill TIME-WAIT too.
79 * * Second purpose of TIME-WAIT is catching old duplicate segments.
80 * Well, certainly it is pure paranoia, but if we load TIME-WAIT
81 * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
82 * * If we invented some more clever way to catch duplicates
83 * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
84 *
85 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
86 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
87 * from the very beginning.
88 *
89 * NOTE. With recycling (and later with fin-wait-2) TW bucket
90 * is _not_ stateless. It means, that strictly speaking we must
91 * spinlock it. I do not want! Well, probability of misbehaviour
92 * is ridiculously low and, seems, we could use some mb() tricks
93 * to avoid misread sequence numbers, states etc. --ANK
94 */
95enum tcp_tw_status
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -070096tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
97 const struct tcphdr *th)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -070099 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 struct tcp_options_received tmp_opt;
101 int paws_reject = 0;
102
103 tmp_opt.saw_tstamp = 0;
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700104 if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 tcp_parse_options(skb, &tmp_opt, 0);
106
107 if (tmp_opt.saw_tstamp) {
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700108 tmp_opt.ts_recent = tcptw->tw_ts_recent;
109 tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 paws_reject = tcp_paws_check(&tmp_opt, th->rst);
111 }
112 }
113
114 if (tw->tw_substate == TCP_FIN_WAIT2) {
115 /* Just repeat all the checks of tcp_rcv_state_process() */
116
117 /* Out of window, send ACK */
118 if (paws_reject ||
119 !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700120 tcptw->tw_rcv_nxt,
121 tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 return TCP_TW_ACK;
123
124 if (th->rst)
125 goto kill;
126
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700127 if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 goto kill_with_rst;
129
130 /* Dup ACK? */
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700131 if (!after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700133 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 return TCP_TW_SUCCESS;
135 }
136
137 /* New data or FIN. If new data arrive after half-duplex close,
138 * reset.
139 */
140 if (!th->fin ||
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700141 TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142kill_with_rst:
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700143 inet_twsk_deschedule(tw, &tcp_death_row);
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700144 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 return TCP_TW_RST;
146 }
147
148 /* FIN arrived, enter true time-wait state. */
Arnaldo Carvalho de Melo8feaf0c2005-08-09 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 Melo8feaf0c2005-08-09 20:09:30 -0700153 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 }
155
156 /* I am shamed, but failed to make it more elegant.
157 * Yes, it is direct reference to IP, which is impossible
158 * to generalize to IPv6. Taking into account that IPv6
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800159 * do not understand recycling in any case, it not
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 * a big problem in practice. --ANK */
161 if (tw->tw_family == AF_INET &&
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700162 tcp_death_row.sysctl_tw_recycle && tcptw->tw_ts_recent_stamp &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 tcp_v4_tw_remember_stamp(tw))
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700164 inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout,
165 TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 else
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700167 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
168 TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 return TCP_TW_ACK;
170 }
171
172 /*
173 * Now real TIME-WAIT state.
174 *
175 * RFC 1122:
176 * "When a connection is [...] on TIME-WAIT state [...]
177 * [a TCP] MAY accept a new SYN from the remote TCP to
178 * reopen the connection directly, if it:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900179 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 * (1) assigns its initial sequence number for the new
181 * connection to be larger than the largest sequence
182 * number it used on the previous connection incarnation,
183 * and
184 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900185 * (2) returns to TIME-WAIT state if the SYN turns out
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 * to be an old duplicate".
187 */
188
189 if (!paws_reject &&
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700190 (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
192 /* In window segment, it may be only reset or bare ack. */
193
194 if (th->rst) {
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800195 /* This is TIME_WAIT assassination, in two flavors.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 * Oh well... nobody has a sufficient solution to this
197 * protocol bug yet.
198 */
199 if (sysctl_tcp_rfc1337 == 0) {
200kill:
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700201 inet_twsk_deschedule(tw, &tcp_death_row);
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700202 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 return TCP_TW_SUCCESS;
204 }
205 }
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700206 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
207 TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
209 if (tmp_opt.saw_tstamp) {
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700210 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
James Morris9d729f72007-03-04 16:12:44 -0800211 tcptw->tw_ts_recent_stamp = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 }
213
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700214 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 return TCP_TW_SUCCESS;
216 }
217
218 /* Out of window segment.
219
220 All the segments are ACKed immediately.
221
222 The only exception is new SYN. We accept it, if it is
223 not old duplicate and we are not in danger to be killed
224 by delayed old duplicates. RFC check is that it has
225 newer sequence number works at rates <40Mbit/sec.
226 However, if paws works, it is reliable AND even more,
227 we even may relax silly seq space cutoff.
228
229 RED-PEN: we violate main RFC requirement, if this SYN will appear
230 old duplicate (i.e. we receive RST in reply to SYN-ACK),
231 we must return socket to time-wait state. It is not good,
232 but not fatal yet.
233 */
234
235 if (th->syn && !th->rst && !th->ack && !paws_reject &&
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700236 (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
237 (tmp_opt.saw_tstamp &&
238 (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
239 u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 if (isn == 0)
241 isn++;
242 TCP_SKB_CB(skb)->when = isn;
243 return TCP_TW_SYN;
244 }
245
246 if (paws_reject)
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700247 NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800249 if (!th->rst) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 /* In this case we must reset the TIMEWAIT timer.
251 *
252 * If it is ACKless SYN it may be both old duplicate
253 * and new good SYN with random sequence number <rcv_nxt.
254 * Do not reschedule in the last case.
255 */
256 if (paws_reject || th->ack)
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700257 inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
258 TCP_TIMEWAIT_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260 /* Send ACK. Note, we do not put the bucket,
261 * it will be released by caller.
262 */
263 return TCP_TW_ACK;
264 }
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700265 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 return TCP_TW_SUCCESS;
267}
268
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900269/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 * Move a socket to time-wait or dead fin-wait-2 state.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900271 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272void tcp_time_wait(struct sock *sk, int state, int timeo)
273{
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700274 struct inet_timewait_sock *tw = NULL;
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800275 const struct inet_connection_sock *icsk = inet_csk(sk);
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700276 const struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 int recycle_ok = 0;
278
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700279 if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800280 recycle_ok = icsk->icsk_af_ops->remember_stamp(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700282 if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets)
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700283 tw = inet_twsk_alloc(sk, state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700285 if (tw != NULL) {
286 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700287 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale;
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700290 tcptw->tw_rcv_nxt = tp->rcv_nxt;
291 tcptw->tw_snd_nxt = tp->snd_nxt;
292 tcptw->tw_rcv_wnd = tcp_receive_window(tp);
293 tcptw->tw_ts_recent = tp->rx_opt.ts_recent;
294 tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
296#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
297 if (tw->tw_family == PF_INET6) {
298 struct ipv6_pinfo *np = inet6_sk(sk);
Arnaldo Carvalho de Melo0fa1a532005-12-13 23:23:09 -0800299 struct inet6_timewait_sock *tw6;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
Arnaldo Carvalho de Melo0fa1a532005-12-13 23:23:09 -0800301 tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
302 tw6 = inet6_twsk((struct sock *)tw);
303 ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr);
304 ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr);
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700305 tw->tw_ipv6only = np->ipv6only;
Arnaldo Carvalho de Meloc6762702005-08-09 20:09:59 -0700306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307#endif
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800308
309#ifdef CONFIG_TCP_MD5SIG
310 /*
311 * The timewait bucket does not have the key DB from the
312 * sock structure. We just make a quick copy of the
313 * md5 key being used (if indeed we are using one)
314 * so the timewait ack generating code has the key.
315 */
316 do {
317 struct tcp_md5sig_key *key;
318 memset(tcptw->tw_md5_key, 0, sizeof(tcptw->tw_md5_key));
319 tcptw->tw_md5_keylen = 0;
320 key = tp->af_specific->md5_lookup(sk, sk);
321 if (key != NULL) {
322 memcpy(&tcptw->tw_md5_key, key->key, key->keylen);
323 tcptw->tw_md5_keylen = key->keylen;
324 if (tcp_alloc_md5sig_pool() == NULL)
325 BUG();
326 }
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800327 } while (0);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800328#endif
329
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 /* Linkage updates. */
Arnaldo Carvalho de Meloe48c4142005-08-09 20:09:46 -0700331 __inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
333 /* Get the TIME_WAIT timeout firing. */
334 if (timeo < rto)
335 timeo = rto;
336
337 if (recycle_ok) {
338 tw->tw_timeout = rto;
339 } else {
340 tw->tw_timeout = TCP_TIMEWAIT_LEN;
341 if (state == TCP_TIME_WAIT)
342 timeo = TCP_TIMEWAIT_LEN;
343 }
344
Arnaldo Carvalho de Melo696ab2d2005-08-09 20:45:03 -0700345 inet_twsk_schedule(tw, &tcp_death_row, timeo,
346 TCP_TIMEWAIT_LEN);
Arnaldo Carvalho de Melo8feaf0c2005-08-09 20:09:30 -0700347 inet_twsk_put(tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 } else {
349 /* Sorry, if we're out of memory, just CLOSE this
350 * socket up. We've got bigger problems than
351 * non-graceful socket closings.
352 */
Arnaldo Carvalho de Meloc6786242006-11-17 11:25:49 -0200353 LIMIT_NETDEBUG(KERN_INFO "TCP: time wait bucket table overflow\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 }
355
356 tcp_update_metrics(sk);
357 tcp_done(sk);
358}
359
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800360void tcp_twsk_destructor(struct sock *sk)
361{
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800362#ifdef CONFIG_TCP_MD5SIG
David S. Millera9286302006-11-14 19:53:22 -0800363 struct tcp_timewait_sock *twsk = tcp_twsk(sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800364 if (twsk->tw_md5_keylen)
365 tcp_put_md5sig_pool();
366#endif
367}
368
369EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
370
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700371static inline void TCP_ECN_openreq_child(struct tcp_sock *tp,
372 struct request_sock *req)
373{
374 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
375}
376
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377/* This is not only more efficient than what we used to do, it eliminates
378 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
379 *
380 * Actually, we could lots of memory writes here. tp of listening
381 * socket contains all necessary default parameters.
382 */
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700383struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384{
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700385 struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Arnaldo Carvalho de Melo87d11ce2005-08-09 20:10:12 -0700387 if (newsk != NULL) {
Arnaldo Carvalho de Melo9f1d2602005-08-09 20:11:24 -0700388 const struct inet_request_sock *ireq = inet_rsk(req);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700389 struct tcp_request_sock *treq = tcp_rsk(req);
Arnaldo Carvalho de Meloa9948a72007-02-28 11:05:56 -0800390 struct inet_connection_sock *newicsk = inet_csk(newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 struct tcp_sock *newtp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 /* Now setup tcp_sock */
394 newtp = tcp_sk(newsk);
395 newtp->pred_flags = 0;
Eric Dumazet54287cc2007-02-22 03:20:44 -0800396 newtp->rcv_wup = newtp->copied_seq = newtp->rcv_nxt = treq->rcv_isn + 1;
397 newtp->snd_sml = newtp->snd_una = newtp->snd_nxt = treq->snt_isn + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399 tcp_prequeue_init(newtp);
400
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700401 tcp_init_wl(newtp, treq->snt_isn, treq->rcv_isn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 newtp->srtt = 0;
404 newtp->mdev = TCP_TIMEOUT_INIT;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700405 newicsk->icsk_rto = TCP_TIMEOUT_INIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 newtp->packets_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 newtp->retrans_out = 0;
409 newtp->sacked_out = 0;
410 newtp->fackets_out = 0;
411 newtp->snd_ssthresh = 0x7fffffff;
412
413 /* So many TCP implementations out there (incorrectly) count the
414 * initial SYN frame in their delayed-ACK and congestion control
415 * algorithms that we must have the following bandaid to talk
416 * efficiently to them. -DaveM
417 */
418 newtp->snd_cwnd = 2;
419 newtp->snd_cwnd_cnt = 0;
Stephen Hemminger9772efb2005-11-10 17:09:53 -0800420 newtp->bytes_acked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422 newtp->frto_counter = 0;
423 newtp->frto_highmark = 0;
424
Stephen Hemminger7957aed2005-09-21 00:19:46 -0700425 newicsk->icsk_ca_ops = &tcp_init_congestion_ops;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700426
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300427 tcp_set_ca_state(newsk, TCP_CA_Open);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 tcp_init_xmit_timers(newsk);
429 skb_queue_head_init(&newtp->out_of_order_queue);
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700430 newtp->write_seq = treq->snt_isn + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 newtp->pushed_seq = newtp->write_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
433 newtp->rx_opt.saw_tstamp = 0;
434
435 newtp->rx_opt.dsack = 0;
436 newtp->rx_opt.eff_sacks = 0;
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 newtp->rx_opt.num_sacks = 0;
439 newtp->urg_data = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 if (sock_flag(newsk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700442 inet_csk_reset_keepalive_timer(newsk,
443 keepalive_time_when(newtp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700445 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800446 if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 if (sysctl_tcp_fack)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300448 tcp_enable_fack(newtp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 }
450 newtp->window_clamp = req->window_clamp;
451 newtp->rcv_ssthresh = req->rcv_wnd;
452 newtp->rcv_wnd = req->rcv_wnd;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700453 newtp->rx_opt.wscale_ok = ireq->wscale_ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 if (newtp->rx_opt.wscale_ok) {
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700455 newtp->rx_opt.snd_wscale = ireq->snd_wscale;
456 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 } else {
458 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
459 newtp->window_clamp = min(newtp->window_clamp, 65535U);
460 }
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700461 newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
462 newtp->rx_opt.snd_wscale);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 newtp->max_window = newtp->snd_wnd;
464
465 if (newtp->rx_opt.tstamp_ok) {
466 newtp->rx_opt.ts_recent = req->ts_recent;
James Morris9d729f72007-03-04 16:12:44 -0800467 newtp->rx_opt.ts_recent_stamp = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
469 } else {
470 newtp->rx_opt.ts_recent_stamp = 0;
471 newtp->tcp_header_len = sizeof(struct tcphdr);
472 }
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800473#ifdef CONFIG_TCP_MD5SIG
474 newtp->md5sig_info = NULL; /*XXX*/
475 if (newtp->af_specific->md5_lookup(sk, newsk))
476 newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
477#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 if (skb->len >= TCP_MIN_RCVMSS+newtp->tcp_header_len)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700479 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 newtp->rx_opt.mss_clamp = req->mss;
481 TCP_ECN_openreq_child(newtp, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Pavel Emelyanov63231bd2008-07-16 20:22:25 -0700483 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
485 return newsk;
486}
487
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900488/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 * Process an incoming packet for SYN_RECV sockets represented
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700490 * as a request_sock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 */
492
493struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700494 struct request_sock *req,
495 struct request_sock **prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700497 const struct tcphdr *th = tcp_hdr(skb);
Al Viro714e85b2006-11-14 20:51:49 -0800498 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 int paws_reject = 0;
500 struct tcp_options_received tmp_opt;
501 struct sock *child;
502
503 tmp_opt.saw_tstamp = 0;
504 if (th->doff > (sizeof(struct tcphdr)>>2)) {
505 tcp_parse_options(skb, &tmp_opt, 0);
506
507 if (tmp_opt.saw_tstamp) {
508 tmp_opt.ts_recent = req->ts_recent;
509 /* We do not store true stamp, but it is not required,
510 * it can be estimated (approximately)
511 * from another data.
512 */
James Morris9d729f72007-03-04 16:12:44 -0800513 tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 paws_reject = tcp_paws_check(&tmp_opt, th->rst);
515 }
516 }
517
518 /* Check for pure retransmitted SYN. */
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700519 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 flg == TCP_FLAG_SYN &&
521 !paws_reject) {
522 /*
523 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
524 * this case on figure 6 and figure 8, but formal
525 * protocol description says NOTHING.
526 * To be more exact, it says that we should send ACK,
527 * because this segment (at least, if it has no data)
528 * is out of window.
529 *
530 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT
531 * describe SYN-RECV state. All the description
532 * is wrong, we cannot believe to it and should
533 * rely only on common sense and implementation
534 * experience.
535 *
536 * Enforce "SYN-ACK" according to figure 8, figure 6
537 * of RFC793, fixed by RFC1122.
538 */
Denis V. Lunevfd80eb92008-02-29 11:43:03 -0800539 req->rsk_ops->rtx_syn_ack(sk, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 return NULL;
541 }
542
543 /* Further reproduces section "SEGMENT ARRIVES"
544 for state SYN-RECEIVED of RFC793.
545 It is broken, however, it does not work only
546 when SYNs are crossed.
547
548 You would think that SYN crossing is impossible here, since
549 we should have a SYN_SENT socket (from connect()) on our end,
550 but this is not true if the crossed SYNs were sent to both
551 ends by a malicious third party. We must defend against this,
552 and to do that we first verify the ACK (as per RFC793, page
553 36) and reset if it is invalid. Is this a true full defense?
554 To convince ourselves, let us consider a way in which the ACK
555 test can still pass in this 'malicious crossed SYNs' case.
556 Malicious sender sends identical SYNs (and thus identical sequence
557 numbers) to both A and B:
558
559 A: gets SYN, seq=7
560 B: gets SYN, seq=7
561
562 By our good fortune, both A and B select the same initial
563 send sequence number of seven :-)
564
565 A: sends SYN|ACK, seq=7, ack_seq=8
566 B: sends SYN|ACK, seq=7, ack_seq=8
567
568 So we are now A eating this SYN|ACK, ACK test passes. So
569 does sequence test, SYN is truncated, and thus we consider
570 it a bare ACK.
571
David S. Millerec0a1962008-06-12 16:31:35 -0700572 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
573 bare ACK. Otherwise, we create an established connection. Both
574 ends (listening sockets) accept the new incoming connection and try
575 to talk to each other. 8-)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577 Note: This case is both harmless, and rare. Possibility is about the
578 same as us discovering intelligent life on another plant tomorrow.
579
580 But generally, we should (RFC lies!) to accept ACK
581 from SYNACK both here and in tcp_rcv_state_process().
582 tcp_rcv_state_process() does not, hence, we do not too.
583
584 Note that the case is absolutely generic:
585 we cannot optimize anything here without
586 violating protocol. All the checks must be made
587 before attempt to create socket.
588 */
589
590 /* RFC793 page 36: "If the connection is in any non-synchronized state ...
591 * and the incoming segment acknowledges something not yet
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800592 * sent (the segment carries an unacceptable ACK) ...
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 * a reset is sent."
594 *
595 * Invalid ACK: reset will be sent by listening socket
596 */
597 if ((flg & TCP_FLAG_ACK) &&
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700598 (TCP_SKB_CB(skb)->ack_seq != tcp_rsk(req)->snt_isn + 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 return sk;
600
601 /* Also, it would be not so bad idea to check rcv_tsecr, which
602 * is essentially ACK extension and too early or too late values
603 * should cause reset in unsynchronized states.
604 */
605
606 /* RFC793: "first check sequence number". */
607
608 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700609 tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 /* Out of window: send ACK and drop. */
611 if (!(flg & TCP_FLAG_RST))
Gui Jianfeng6edafaa2008-08-06 23:50:04 -0700612 req->rsk_ops->send_ack(sk, skb, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 if (paws_reject)
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700614 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 return NULL;
616 }
617
618 /* In sequence, PAWS is OK. */
619
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700620 if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1))
Adam Langley2aaab9a2008-08-07 20:27:45 -0700621 req->ts_recent = tmp_opt.rcv_tsval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
Adam Langley2aaab9a2008-08-07 20:27:45 -0700623 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
624 /* Truncate SYN, it is out of window starting
625 at tcp_rsk(req)->rcv_isn + 1. */
626 flg &= ~TCP_FLAG_SYN;
627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Adam Langley2aaab9a2008-08-07 20:27:45 -0700629 /* RFC793: "second check the RST bit" and
630 * "fourth, check the SYN bit"
631 */
632 if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
633 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
634 goto embryonic_reset;
635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
Adam Langley2aaab9a2008-08-07 20:27:45 -0700637 /* ACK sequence verified above, just make sure ACK is
638 * set. If ACK not set, just silently drop the packet.
639 */
640 if (!(flg & TCP_FLAG_ACK))
641 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Adam Langley2aaab9a2008-08-07 20:27:45 -0700643 /* If TCP_DEFER_ACCEPT is set, drop bare ACK. */
644 if (inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
645 TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
646 inet_rsk(req)->acked = 1;
647 return NULL;
648 }
David S. Millerec0a1962008-06-12 16:31:35 -0700649
Adam Langley2aaab9a2008-08-07 20:27:45 -0700650 /* OK, ACK is valid, create big socket and
651 * feed this segment to it. It will repeat all
652 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
653 * ESTABLISHED STATE. If it will be dropped after
654 * socket is created, wait for troubles.
655 */
656 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
657 if (child == NULL)
658 goto listen_overflow;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800659#ifdef CONFIG_TCP_MD5SIG
Adam Langley2aaab9a2008-08-07 20:27:45 -0700660 else {
661 /* Copy over the MD5 key from the original socket */
662 struct tcp_md5sig_key *key;
663 struct tcp_sock *tp = tcp_sk(sk);
664 key = tp->af_specific->md5_lookup(sk, child);
665 if (key != NULL) {
666 /*
667 * We're using one, so create a matching key on the
668 * newsk structure. If we fail to get memory then we
669 * end up not copying the key across. Shucks.
670 */
671 char *newkey = kmemdup(key->key, key->keylen,
672 GFP_ATOMIC);
673 if (newkey) {
674 if (!tcp_alloc_md5sig_pool())
675 BUG();
676 tp->af_specific->md5_add(child, child, newkey,
677 key->keylen);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800678 }
679 }
Adam Langley2aaab9a2008-08-07 20:27:45 -0700680 }
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800681#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Adam Langley2aaab9a2008-08-07 20:27:45 -0700683 inet_csk_reqsk_queue_unlink(sk, req, prev);
684 inet_csk_reqsk_queue_removed(sk, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Adam Langley2aaab9a2008-08-07 20:27:45 -0700686 inet_csk_reqsk_queue_add(sk, req, child);
687 return child;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Adam Langley2aaab9a2008-08-07 20:27:45 -0700689listen_overflow:
690 if (!sysctl_tcp_abort_on_overflow) {
691 inet_rsk(req)->acked = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 return NULL;
Adam Langley2aaab9a2008-08-07 20:27:45 -0700693 }
694
695embryonic_reset:
696 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
697 if (!(flg & TCP_FLAG_RST))
698 req->rsk_ops->send_reset(sk, skb);
699
700 inet_csk_reqsk_queue_drop(sk, req, prev);
701 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
703
704/*
705 * Queue segment on the new socket if the new socket is active,
706 * otherwise we just shortcircuit this and continue with
707 * the new socket.
708 */
709
710int tcp_child_process(struct sock *parent, struct sock *child,
711 struct sk_buff *skb)
712{
713 int ret = 0;
714 int state = child->sk_state;
715
716 if (!sock_owned_by_user(child)) {
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700717 ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
718 skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 /* Wakeup parent, send SIGIO */
720 if (state == TCP_SYN_RECV && child->sk_state != state)
721 parent->sk_data_ready(parent, 0);
722 } else {
723 /* Alas, it is possible again, because we do lookup
724 * in main socket hash table and lock on listening
725 * socket does not protect us more.
726 */
727 sk_add_backlog(child, skb);
728 }
729
730 bh_unlock_sock(child);
731 sock_put(child);
732 return ret;
733}
734
735EXPORT_SYMBOL(tcp_check_req);
736EXPORT_SYMBOL(tcp_child_process);
737EXPORT_SYMBOL(tcp_create_openreq_child);
738EXPORT_SYMBOL(tcp_timewait_state_process);