]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | * | |
8 | * Version: $Id: tcp_minisocks.c,v 1.15 2002/02/01 22:01:04 davem Exp $ | |
9 | * | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
13 | * Corey Minyard <wf-rch!minyard@relay.EU.net> | |
14 | * Florian La Roche, <flla@stud.uni-sb.de> | |
15 | * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> | |
16 | * Linus Torvalds, <torvalds@cs.helsinki.fi> | |
17 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
18 | * Matthew Dillon, <dillon@apollo.west.oic.com> | |
19 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
20 | * Jorge Cwik, <jorge@laser.satlink.net> | |
21 | */ | |
22 | ||
1da177e4 LT |
23 | #include <linux/mm.h> |
24 | #include <linux/module.h> | |
25 | #include <linux/sysctl.h> | |
26 | #include <linux/workqueue.h> | |
27 | #include <net/tcp.h> | |
28 | #include <net/inet_common.h> | |
29 | #include <net/xfrm.h> | |
30 | ||
31 | #ifdef CONFIG_SYSCTL | |
32 | #define SYNC_INIT 0 /* let the user enable it */ | |
33 | #else | |
34 | #define SYNC_INIT 1 | |
35 | #endif | |
36 | ||
ab32ea5d BH |
37 | int sysctl_tcp_syncookies __read_mostly = SYNC_INIT; |
38 | int sysctl_tcp_abort_on_overflow __read_mostly; | |
1da177e4 | 39 | |
295ff7ed ACM |
40 | struct inet_timewait_death_row tcp_death_row = { |
41 | .sysctl_max_tw_buckets = NR_FILE * 2, | |
42 | .period = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS, | |
e4d91918 | 43 | .death_lock = __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock), |
295ff7ed ACM |
44 | .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, | |
65f27f38 | 48 | inet_twdr_twkill_work), |
295ff7ed ACM |
49 | /* 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 | ||
56 | EXPORT_SYMBOL_GPL(tcp_death_row); | |
57 | ||
1da177e4 LT |
58 | static __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 | ||
1da177e4 LT |
67 | /* |
68 | * * 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 | */ | |
95 | enum tcp_tw_status | |
8feaf0c0 ACM |
96 | tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb, |
97 | const struct tcphdr *th) | |
1da177e4 | 98 | { |
8feaf0c0 | 99 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
1da177e4 LT |
100 | struct tcp_options_received tmp_opt; |
101 | int paws_reject = 0; | |
102 | ||
103 | tmp_opt.saw_tstamp = 0; | |
8feaf0c0 | 104 | if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) { |
1da177e4 LT |
105 | tcp_parse_options(skb, &tmp_opt, 0); |
106 | ||
107 | if (tmp_opt.saw_tstamp) { | |
8feaf0c0 ACM |
108 | tmp_opt.ts_recent = tcptw->tw_ts_recent; |
109 | tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
1da177e4 LT |
110 | 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, | |
8feaf0c0 ACM |
120 | tcptw->tw_rcv_nxt, |
121 | tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd)) | |
1da177e4 LT |
122 | return TCP_TW_ACK; |
123 | ||
124 | if (th->rst) | |
125 | goto kill; | |
126 | ||
8feaf0c0 | 127 | if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt)) |
1da177e4 LT |
128 | goto kill_with_rst; |
129 | ||
130 | /* Dup ACK? */ | |
8feaf0c0 | 131 | if (!after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) || |
1da177e4 | 132 | TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) { |
8feaf0c0 | 133 | inet_twsk_put(tw); |
1da177e4 LT |
134 | 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 || | |
8feaf0c0 | 141 | TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) { |
1da177e4 | 142 | kill_with_rst: |
295ff7ed | 143 | inet_twsk_deschedule(tw, &tcp_death_row); |
8feaf0c0 | 144 | inet_twsk_put(tw); |
1da177e4 LT |
145 | return TCP_TW_RST; |
146 | } | |
147 | ||
148 | /* FIN arrived, enter true time-wait state. */ | |
8feaf0c0 ACM |
149 | tw->tw_substate = TCP_TIME_WAIT; |
150 | tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
1da177e4 | 151 | if (tmp_opt.saw_tstamp) { |
8feaf0c0 ACM |
152 | tcptw->tw_ts_recent_stamp = xtime.tv_sec; |
153 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; | |
1da177e4 LT |
154 | } |
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 | |
caa20d9a | 159 | * do not understand recycling in any case, it not |
1da177e4 LT |
160 | * a big problem in practice. --ANK */ |
161 | if (tw->tw_family == AF_INET && | |
295ff7ed | 162 | tcp_death_row.sysctl_tw_recycle && tcptw->tw_ts_recent_stamp && |
1da177e4 | 163 | tcp_v4_tw_remember_stamp(tw)) |
696ab2d3 ACM |
164 | inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout, |
165 | TCP_TIMEWAIT_LEN); | |
1da177e4 | 166 | else |
696ab2d3 ACM |
167 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, |
168 | TCP_TIMEWAIT_LEN); | |
1da177e4 LT |
169 | 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: | |
179 | * | |
180 | * (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 | * | |
185 | * (2) returns to TIME-WAIT state if the SYN turns out | |
186 | * to be an old duplicate". | |
187 | */ | |
188 | ||
189 | if (!paws_reject && | |
8feaf0c0 | 190 | (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt && |
1da177e4 LT |
191 | (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) { | |
caa20d9a | 195 | /* This is TIME_WAIT assassination, in two flavors. |
1da177e4 LT |
196 | * Oh well... nobody has a sufficient solution to this |
197 | * protocol bug yet. | |
198 | */ | |
199 | if (sysctl_tcp_rfc1337 == 0) { | |
200 | kill: | |
295ff7ed | 201 | inet_twsk_deschedule(tw, &tcp_death_row); |
8feaf0c0 | 202 | inet_twsk_put(tw); |
1da177e4 LT |
203 | return TCP_TW_SUCCESS; |
204 | } | |
205 | } | |
696ab2d3 ACM |
206 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, |
207 | TCP_TIMEWAIT_LEN); | |
1da177e4 LT |
208 | |
209 | if (tmp_opt.saw_tstamp) { | |
8feaf0c0 ACM |
210 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; |
211 | tcptw->tw_ts_recent_stamp = xtime.tv_sec; | |
1da177e4 LT |
212 | } |
213 | ||
8feaf0c0 | 214 | inet_twsk_put(tw); |
1da177e4 LT |
215 | 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 && | |
8feaf0c0 ACM |
236 | (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; | |
1da177e4 LT |
240 | if (isn == 0) |
241 | isn++; | |
242 | TCP_SKB_CB(skb)->when = isn; | |
243 | return TCP_TW_SYN; | |
244 | } | |
245 | ||
246 | if (paws_reject) | |
247 | NET_INC_STATS_BH(LINUX_MIB_PAWSESTABREJECTED); | |
248 | ||
249 | if(!th->rst) { | |
250 | /* 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) | |
696ab2d3 ACM |
257 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, |
258 | TCP_TIMEWAIT_LEN); | |
1da177e4 LT |
259 | |
260 | /* Send ACK. Note, we do not put the bucket, | |
261 | * it will be released by caller. | |
262 | */ | |
263 | return TCP_TW_ACK; | |
264 | } | |
8feaf0c0 | 265 | inet_twsk_put(tw); |
1da177e4 LT |
266 | return TCP_TW_SUCCESS; |
267 | } | |
268 | ||
1da177e4 LT |
269 | /* |
270 | * Move a socket to time-wait or dead fin-wait-2 state. | |
271 | */ | |
272 | void tcp_time_wait(struct sock *sk, int state, int timeo) | |
273 | { | |
8feaf0c0 | 274 | struct inet_timewait_sock *tw = NULL; |
8292a17a | 275 | const struct inet_connection_sock *icsk = inet_csk(sk); |
8feaf0c0 | 276 | const struct tcp_sock *tp = tcp_sk(sk); |
1da177e4 LT |
277 | int recycle_ok = 0; |
278 | ||
295ff7ed | 279 | if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp) |
8292a17a | 280 | recycle_ok = icsk->icsk_af_ops->remember_stamp(sk); |
1da177e4 | 281 | |
295ff7ed | 282 | if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets) |
c676270b | 283 | tw = inet_twsk_alloc(sk, state); |
1da177e4 | 284 | |
8feaf0c0 ACM |
285 | if (tw != NULL) { |
286 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); | |
463c84b9 | 287 | const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); |
8feaf0c0 | 288 | |
1da177e4 | 289 | tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale; |
8feaf0c0 ACM |
290 | 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; | |
1da177e4 LT |
295 | |
296 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
297 | if (tw->tw_family == PF_INET6) { | |
298 | struct ipv6_pinfo *np = inet6_sk(sk); | |
0fa1a53e | 299 | struct inet6_timewait_sock *tw6; |
1da177e4 | 300 | |
0fa1a53e ACM |
301 | 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); | |
8feaf0c0 | 305 | tw->tw_ipv6only = np->ipv6only; |
c676270b | 306 | } |
1da177e4 | 307 | #endif |
cfb6eeb4 YH |
308 | |
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 | } | |
327 | } while(0); | |
328 | #endif | |
329 | ||
1da177e4 | 330 | /* Linkage updates. */ |
e48c414e | 331 | __inet_twsk_hashdance(tw, sk, &tcp_hashinfo); |
1da177e4 LT |
332 | |
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 | ||
696ab2d3 ACM |
345 | inet_twsk_schedule(tw, &tcp_death_row, timeo, |
346 | TCP_TIMEWAIT_LEN); | |
8feaf0c0 | 347 | inet_twsk_put(tw); |
1da177e4 LT |
348 | } 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 | */ | |
c6786240 | 353 | LIMIT_NETDEBUG(KERN_INFO "TCP: time wait bucket table overflow\n"); |
1da177e4 LT |
354 | } |
355 | ||
356 | tcp_update_metrics(sk); | |
357 | tcp_done(sk); | |
358 | } | |
359 | ||
cfb6eeb4 YH |
360 | void tcp_twsk_destructor(struct sock *sk) |
361 | { | |
cfb6eeb4 | 362 | #ifdef CONFIG_TCP_MD5SIG |
a928630a | 363 | struct tcp_timewait_sock *twsk = tcp_twsk(sk); |
cfb6eeb4 YH |
364 | if (twsk->tw_md5_keylen) |
365 | tcp_put_md5sig_pool(); | |
366 | #endif | |
367 | } | |
368 | ||
369 | EXPORT_SYMBOL_GPL(tcp_twsk_destructor); | |
370 | ||
1da177e4 LT |
371 | /* This is not only more efficient than what we used to do, it eliminates |
372 | * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM | |
373 | * | |
374 | * Actually, we could lots of memory writes here. tp of listening | |
375 | * socket contains all necessary default parameters. | |
376 | */ | |
60236fdd | 377 | struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb) |
1da177e4 | 378 | { |
9f1d2604 | 379 | struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC); |
1da177e4 | 380 | |
87d11ceb | 381 | if (newsk != NULL) { |
9f1d2604 | 382 | const struct inet_request_sock *ireq = inet_rsk(req); |
2e6599cb | 383 | struct tcp_request_sock *treq = tcp_rsk(req); |
9f1d2604 | 384 | struct inet_connection_sock *newicsk = inet_csk(sk); |
1da177e4 | 385 | struct tcp_sock *newtp; |
1da177e4 | 386 | |
1da177e4 LT |
387 | /* Now setup tcp_sock */ |
388 | newtp = tcp_sk(newsk); | |
389 | newtp->pred_flags = 0; | |
2e6599cb | 390 | newtp->rcv_nxt = treq->rcv_isn + 1; |
87d11ceb | 391 | newtp->snd_nxt = newtp->snd_una = newtp->snd_sml = treq->snt_isn + 1; |
1da177e4 LT |
392 | |
393 | tcp_prequeue_init(newtp); | |
394 | ||
2e6599cb | 395 | tcp_init_wl(newtp, treq->snt_isn, treq->rcv_isn); |
1da177e4 | 396 | |
1da177e4 LT |
397 | newtp->srtt = 0; |
398 | newtp->mdev = TCP_TIMEOUT_INIT; | |
463c84b9 | 399 | newicsk->icsk_rto = TCP_TIMEOUT_INIT; |
1da177e4 LT |
400 | |
401 | newtp->packets_out = 0; | |
402 | newtp->left_out = 0; | |
403 | newtp->retrans_out = 0; | |
404 | newtp->sacked_out = 0; | |
405 | newtp->fackets_out = 0; | |
406 | newtp->snd_ssthresh = 0x7fffffff; | |
407 | ||
408 | /* So many TCP implementations out there (incorrectly) count the | |
409 | * initial SYN frame in their delayed-ACK and congestion control | |
410 | * algorithms that we must have the following bandaid to talk | |
411 | * efficiently to them. -DaveM | |
412 | */ | |
413 | newtp->snd_cwnd = 2; | |
414 | newtp->snd_cwnd_cnt = 0; | |
9772efb9 | 415 | newtp->bytes_acked = 0; |
1da177e4 LT |
416 | |
417 | newtp->frto_counter = 0; | |
418 | newtp->frto_highmark = 0; | |
419 | ||
7957aed7 | 420 | newicsk->icsk_ca_ops = &tcp_init_congestion_ops; |
317a76f9 | 421 | |
6687e988 | 422 | tcp_set_ca_state(newsk, TCP_CA_Open); |
1da177e4 LT |
423 | tcp_init_xmit_timers(newsk); |
424 | skb_queue_head_init(&newtp->out_of_order_queue); | |
2e6599cb ACM |
425 | newtp->rcv_wup = treq->rcv_isn + 1; |
426 | newtp->write_seq = treq->snt_isn + 1; | |
1da177e4 | 427 | newtp->pushed_seq = newtp->write_seq; |
2e6599cb | 428 | newtp->copied_seq = treq->rcv_isn + 1; |
1da177e4 LT |
429 | |
430 | newtp->rx_opt.saw_tstamp = 0; | |
431 | ||
432 | newtp->rx_opt.dsack = 0; | |
433 | newtp->rx_opt.eff_sacks = 0; | |
434 | ||
1da177e4 LT |
435 | newtp->rx_opt.num_sacks = 0; |
436 | newtp->urg_data = 0; | |
1da177e4 | 437 | |
1da177e4 | 438 | if (sock_flag(newsk, SOCK_KEEPOPEN)) |
463c84b9 ACM |
439 | inet_csk_reset_keepalive_timer(newsk, |
440 | keepalive_time_when(newtp)); | |
1da177e4 | 441 | |
2e6599cb ACM |
442 | newtp->rx_opt.tstamp_ok = ireq->tstamp_ok; |
443 | if((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) { | |
1da177e4 LT |
444 | if (sysctl_tcp_fack) |
445 | newtp->rx_opt.sack_ok |= 2; | |
446 | } | |
447 | newtp->window_clamp = req->window_clamp; | |
448 | newtp->rcv_ssthresh = req->rcv_wnd; | |
449 | newtp->rcv_wnd = req->rcv_wnd; | |
2e6599cb | 450 | newtp->rx_opt.wscale_ok = ireq->wscale_ok; |
1da177e4 | 451 | if (newtp->rx_opt.wscale_ok) { |
2e6599cb ACM |
452 | newtp->rx_opt.snd_wscale = ireq->snd_wscale; |
453 | newtp->rx_opt.rcv_wscale = ireq->rcv_wscale; | |
1da177e4 LT |
454 | } else { |
455 | newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0; | |
456 | newtp->window_clamp = min(newtp->window_clamp, 65535U); | |
457 | } | |
458 | newtp->snd_wnd = ntohs(skb->h.th->window) << newtp->rx_opt.snd_wscale; | |
459 | newtp->max_window = newtp->snd_wnd; | |
460 | ||
461 | if (newtp->rx_opt.tstamp_ok) { | |
462 | newtp->rx_opt.ts_recent = req->ts_recent; | |
463 | newtp->rx_opt.ts_recent_stamp = xtime.tv_sec; | |
464 | newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; | |
465 | } else { | |
466 | newtp->rx_opt.ts_recent_stamp = 0; | |
467 | newtp->tcp_header_len = sizeof(struct tcphdr); | |
468 | } | |
cfb6eeb4 YH |
469 | #ifdef CONFIG_TCP_MD5SIG |
470 | newtp->md5sig_info = NULL; /*XXX*/ | |
471 | if (newtp->af_specific->md5_lookup(sk, newsk)) | |
472 | newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED; | |
473 | #endif | |
1da177e4 | 474 | if (skb->len >= TCP_MIN_RCVMSS+newtp->tcp_header_len) |
463c84b9 | 475 | newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len; |
1da177e4 LT |
476 | newtp->rx_opt.mss_clamp = req->mss; |
477 | TCP_ECN_openreq_child(newtp, req); | |
1da177e4 | 478 | |
1da177e4 LT |
479 | TCP_INC_STATS_BH(TCP_MIB_PASSIVEOPENS); |
480 | } | |
481 | return newsk; | |
482 | } | |
483 | ||
484 | /* | |
485 | * Process an incoming packet for SYN_RECV sockets represented | |
60236fdd | 486 | * as a request_sock. |
1da177e4 LT |
487 | */ |
488 | ||
489 | struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb, | |
60236fdd ACM |
490 | struct request_sock *req, |
491 | struct request_sock **prev) | |
1da177e4 LT |
492 | { |
493 | struct tcphdr *th = skb->h.th; | |
714e85be | 494 | __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK); |
1da177e4 LT |
495 | int paws_reject = 0; |
496 | struct tcp_options_received tmp_opt; | |
497 | struct sock *child; | |
498 | ||
499 | tmp_opt.saw_tstamp = 0; | |
500 | if (th->doff > (sizeof(struct tcphdr)>>2)) { | |
501 | tcp_parse_options(skb, &tmp_opt, 0); | |
502 | ||
503 | if (tmp_opt.saw_tstamp) { | |
504 | tmp_opt.ts_recent = req->ts_recent; | |
505 | /* We do not store true stamp, but it is not required, | |
506 | * it can be estimated (approximately) | |
507 | * from another data. | |
508 | */ | |
509 | tmp_opt.ts_recent_stamp = xtime.tv_sec - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans); | |
510 | paws_reject = tcp_paws_check(&tmp_opt, th->rst); | |
511 | } | |
512 | } | |
513 | ||
514 | /* Check for pure retransmitted SYN. */ | |
2e6599cb | 515 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn && |
1da177e4 LT |
516 | flg == TCP_FLAG_SYN && |
517 | !paws_reject) { | |
518 | /* | |
519 | * RFC793 draws (Incorrectly! It was fixed in RFC1122) | |
520 | * this case on figure 6 and figure 8, but formal | |
521 | * protocol description says NOTHING. | |
522 | * To be more exact, it says that we should send ACK, | |
523 | * because this segment (at least, if it has no data) | |
524 | * is out of window. | |
525 | * | |
526 | * CONCLUSION: RFC793 (even with RFC1122) DOES NOT | |
527 | * describe SYN-RECV state. All the description | |
528 | * is wrong, we cannot believe to it and should | |
529 | * rely only on common sense and implementation | |
530 | * experience. | |
531 | * | |
532 | * Enforce "SYN-ACK" according to figure 8, figure 6 | |
533 | * of RFC793, fixed by RFC1122. | |
534 | */ | |
60236fdd | 535 | req->rsk_ops->rtx_syn_ack(sk, req, NULL); |
1da177e4 LT |
536 | return NULL; |
537 | } | |
538 | ||
539 | /* Further reproduces section "SEGMENT ARRIVES" | |
540 | for state SYN-RECEIVED of RFC793. | |
541 | It is broken, however, it does not work only | |
542 | when SYNs are crossed. | |
543 | ||
544 | You would think that SYN crossing is impossible here, since | |
545 | we should have a SYN_SENT socket (from connect()) on our end, | |
546 | but this is not true if the crossed SYNs were sent to both | |
547 | ends by a malicious third party. We must defend against this, | |
548 | and to do that we first verify the ACK (as per RFC793, page | |
549 | 36) and reset if it is invalid. Is this a true full defense? | |
550 | To convince ourselves, let us consider a way in which the ACK | |
551 | test can still pass in this 'malicious crossed SYNs' case. | |
552 | Malicious sender sends identical SYNs (and thus identical sequence | |
553 | numbers) to both A and B: | |
554 | ||
555 | A: gets SYN, seq=7 | |
556 | B: gets SYN, seq=7 | |
557 | ||
558 | By our good fortune, both A and B select the same initial | |
559 | send sequence number of seven :-) | |
560 | ||
561 | A: sends SYN|ACK, seq=7, ack_seq=8 | |
562 | B: sends SYN|ACK, seq=7, ack_seq=8 | |
563 | ||
564 | So we are now A eating this SYN|ACK, ACK test passes. So | |
565 | does sequence test, SYN is truncated, and thus we consider | |
566 | it a bare ACK. | |
567 | ||
295f7324 ACM |
568 | If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this |
569 | bare ACK. Otherwise, we create an established connection. Both | |
570 | ends (listening sockets) accept the new incoming connection and try | |
571 | to talk to each other. 8-) | |
1da177e4 LT |
572 | |
573 | Note: This case is both harmless, and rare. Possibility is about the | |
574 | same as us discovering intelligent life on another plant tomorrow. | |
575 | ||
576 | But generally, we should (RFC lies!) to accept ACK | |
577 | from SYNACK both here and in tcp_rcv_state_process(). | |
578 | tcp_rcv_state_process() does not, hence, we do not too. | |
579 | ||
580 | Note that the case is absolutely generic: | |
581 | we cannot optimize anything here without | |
582 | violating protocol. All the checks must be made | |
583 | before attempt to create socket. | |
584 | */ | |
585 | ||
586 | /* RFC793 page 36: "If the connection is in any non-synchronized state ... | |
587 | * and the incoming segment acknowledges something not yet | |
caa20d9a | 588 | * sent (the segment carries an unacceptable ACK) ... |
1da177e4 LT |
589 | * a reset is sent." |
590 | * | |
591 | * Invalid ACK: reset will be sent by listening socket | |
592 | */ | |
593 | if ((flg & TCP_FLAG_ACK) && | |
2e6599cb | 594 | (TCP_SKB_CB(skb)->ack_seq != tcp_rsk(req)->snt_isn + 1)) |
1da177e4 LT |
595 | return sk; |
596 | ||
597 | /* Also, it would be not so bad idea to check rcv_tsecr, which | |
598 | * is essentially ACK extension and too early or too late values | |
599 | * should cause reset in unsynchronized states. | |
600 | */ | |
601 | ||
602 | /* RFC793: "first check sequence number". */ | |
603 | ||
604 | if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
2e6599cb | 605 | tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) { |
1da177e4 LT |
606 | /* Out of window: send ACK and drop. */ |
607 | if (!(flg & TCP_FLAG_RST)) | |
60236fdd | 608 | req->rsk_ops->send_ack(skb, req); |
1da177e4 LT |
609 | if (paws_reject) |
610 | NET_INC_STATS_BH(LINUX_MIB_PAWSESTABREJECTED); | |
611 | return NULL; | |
612 | } | |
613 | ||
614 | /* In sequence, PAWS is OK. */ | |
615 | ||
2e6599cb | 616 | if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1)) |
1da177e4 LT |
617 | req->ts_recent = tmp_opt.rcv_tsval; |
618 | ||
2e6599cb | 619 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) { |
1da177e4 | 620 | /* Truncate SYN, it is out of window starting |
2e6599cb | 621 | at tcp_rsk(req)->rcv_isn + 1. */ |
1da177e4 LT |
622 | flg &= ~TCP_FLAG_SYN; |
623 | } | |
624 | ||
625 | /* RFC793: "second check the RST bit" and | |
626 | * "fourth, check the SYN bit" | |
627 | */ | |
3687b1dc WY |
628 | if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) { |
629 | TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS); | |
1da177e4 | 630 | goto embryonic_reset; |
3687b1dc | 631 | } |
1da177e4 LT |
632 | |
633 | /* ACK sequence verified above, just make sure ACK is | |
634 | * set. If ACK not set, just silently drop the packet. | |
635 | */ | |
636 | if (!(flg & TCP_FLAG_ACK)) | |
637 | return NULL; | |
638 | ||
639 | /* If TCP_DEFER_ACCEPT is set, drop bare ACK. */ | |
295f7324 ACM |
640 | if (inet_csk(sk)->icsk_accept_queue.rskq_defer_accept && |
641 | TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) { | |
2e6599cb | 642 | inet_rsk(req)->acked = 1; |
1da177e4 LT |
643 | return NULL; |
644 | } | |
645 | ||
646 | /* OK, ACK is valid, create big socket and | |
647 | * feed this segment to it. It will repeat all | |
648 | * the tests. THIS SEGMENT MUST MOVE SOCKET TO | |
649 | * ESTABLISHED STATE. If it will be dropped after | |
650 | * socket is created, wait for troubles. | |
651 | */ | |
8292a17a ACM |
652 | child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, |
653 | req, NULL); | |
1da177e4 LT |
654 | if (child == NULL) |
655 | goto listen_overflow; | |
cfb6eeb4 YH |
656 | #ifdef CONFIG_TCP_MD5SIG |
657 | else { | |
658 | /* Copy over the MD5 key from the original socket */ | |
659 | struct tcp_md5sig_key *key; | |
660 | struct tcp_sock *tp = tcp_sk(sk); | |
661 | key = tp->af_specific->md5_lookup(sk, child); | |
662 | if (key != NULL) { | |
663 | /* | |
664 | * We're using one, so create a matching key on the | |
665 | * newsk structure. If we fail to get memory then we | |
666 | * end up not copying the key across. Shucks. | |
667 | */ | |
c6786240 ACM |
668 | char *newkey = kmemdup(key->key, key->keylen, |
669 | GFP_ATOMIC); | |
cfb6eeb4 YH |
670 | if (newkey) { |
671 | if (!tcp_alloc_md5sig_pool()) | |
672 | BUG(); | |
cfb6eeb4 YH |
673 | tp->af_specific->md5_add(child, child, |
674 | newkey, | |
675 | key->keylen); | |
676 | } | |
677 | } | |
678 | } | |
679 | #endif | |
1da177e4 | 680 | |
463c84b9 ACM |
681 | inet_csk_reqsk_queue_unlink(sk, req, prev); |
682 | inet_csk_reqsk_queue_removed(sk, req); | |
1da177e4 | 683 | |
463c84b9 | 684 | inet_csk_reqsk_queue_add(sk, req, child); |
1da177e4 LT |
685 | return child; |
686 | ||
687 | listen_overflow: | |
688 | if (!sysctl_tcp_abort_on_overflow) { | |
2e6599cb | 689 | inet_rsk(req)->acked = 1; |
1da177e4 LT |
690 | return NULL; |
691 | } | |
692 | ||
693 | embryonic_reset: | |
694 | NET_INC_STATS_BH(LINUX_MIB_EMBRYONICRSTS); | |
695 | if (!(flg & TCP_FLAG_RST)) | |
cfb6eeb4 | 696 | req->rsk_ops->send_reset(sk, skb); |
1da177e4 | 697 | |
463c84b9 | 698 | inet_csk_reqsk_queue_drop(sk, req, prev); |
1da177e4 LT |
699 | return NULL; |
700 | } | |
701 | ||
702 | /* | |
703 | * Queue segment on the new socket if the new socket is active, | |
704 | * otherwise we just shortcircuit this and continue with | |
705 | * the new socket. | |
706 | */ | |
707 | ||
708 | int tcp_child_process(struct sock *parent, struct sock *child, | |
709 | struct sk_buff *skb) | |
710 | { | |
711 | int ret = 0; | |
712 | int state = child->sk_state; | |
713 | ||
714 | if (!sock_owned_by_user(child)) { | |
715 | ret = tcp_rcv_state_process(child, skb, skb->h.th, skb->len); | |
716 | ||
717 | /* Wakeup parent, send SIGIO */ | |
718 | if (state == TCP_SYN_RECV && child->sk_state != state) | |
719 | parent->sk_data_ready(parent, 0); | |
720 | } else { | |
721 | /* Alas, it is possible again, because we do lookup | |
722 | * in main socket hash table and lock on listening | |
723 | * socket does not protect us more. | |
724 | */ | |
725 | sk_add_backlog(child, skb); | |
726 | } | |
727 | ||
728 | bh_unlock_sock(child); | |
729 | sock_put(child); | |
730 | return ret; | |
731 | } | |
732 | ||
733 | EXPORT_SYMBOL(tcp_check_req); | |
734 | EXPORT_SYMBOL(tcp_child_process); | |
735 | EXPORT_SYMBOL(tcp_create_openreq_child); | |
736 | EXPORT_SYMBOL(tcp_timewait_state_process); |