]>
Commit | Line | Data |
---|---|---|
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 | * Authors: Ross Biro | |
9 | * 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 | ||
21 | #include <linux/mm.h> | |
22 | #include <linux/module.h> | |
23 | #include <linux/slab.h> | |
24 | #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 | ||
30 | int sysctl_tcp_syncookies __read_mostly = 1; | |
31 | EXPORT_SYMBOL(sysctl_tcp_syncookies); | |
32 | ||
33 | int sysctl_tcp_abort_on_overflow __read_mostly; | |
34 | ||
35 | struct inet_timewait_death_row tcp_death_row = { | |
36 | .sysctl_max_tw_buckets = NR_FILE * 2, | |
37 | .period = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS, | |
38 | .death_lock = __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock), | |
39 | .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, | |
43 | inet_twdr_twkill_work), | |
44 | /* 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 | }; | |
50 | ||
51 | EXPORT_SYMBOL_GPL(tcp_death_row); | |
52 | ||
53 | static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win) | |
54 | { | |
55 | if (seq == s_win) | |
56 | return 1; | |
57 | if (after(end_seq, s_win) && before(seq, e_win)) | |
58 | return 1; | |
59 | return (seq == e_win && seq == end_seq); | |
60 | } | |
61 | ||
62 | /* | |
63 | * * Main purpose of TIME-WAIT state is to close connection gracefully, | |
64 | * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN | |
65 | * (and, probably, tail of data) and one or more our ACKs are lost. | |
66 | * * What is TIME-WAIT timeout? It is associated with maximal packet | |
67 | * lifetime in the internet, which results in wrong conclusion, that | |
68 | * it is set to catch "old duplicate segments" wandering out of their path. | |
69 | * It is not quite correct. This timeout is calculated so that it exceeds | |
70 | * maximal retransmission timeout enough to allow to lose one (or more) | |
71 | * segments sent by peer and our ACKs. This time may be calculated from RTO. | |
72 | * * When TIME-WAIT socket receives RST, it means that another end | |
73 | * finally closed and we are allowed to kill TIME-WAIT too. | |
74 | * * Second purpose of TIME-WAIT is catching old duplicate segments. | |
75 | * Well, certainly it is pure paranoia, but if we load TIME-WAIT | |
76 | * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs. | |
77 | * * If we invented some more clever way to catch duplicates | |
78 | * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs. | |
79 | * | |
80 | * The algorithm below is based on FORMAL INTERPRETATION of RFCs. | |
81 | * When you compare it to RFCs, please, read section SEGMENT ARRIVES | |
82 | * from the very beginning. | |
83 | * | |
84 | * NOTE. With recycling (and later with fin-wait-2) TW bucket | |
85 | * is _not_ stateless. It means, that strictly speaking we must | |
86 | * spinlock it. I do not want! Well, probability of misbehaviour | |
87 | * is ridiculously low and, seems, we could use some mb() tricks | |
88 | * to avoid misread sequence numbers, states etc. --ANK | |
89 | */ | |
90 | enum tcp_tw_status | |
91 | tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb, | |
92 | const struct tcphdr *th) | |
93 | { | |
94 | struct tcp_options_received tmp_opt; | |
95 | u8 *hash_location; | |
96 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); | |
97 | int paws_reject = 0; | |
98 | ||
99 | tmp_opt.saw_tstamp = 0; | |
100 | if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) { | |
101 | tcp_parse_options(skb, &tmp_opt, &hash_location, 0); | |
102 | ||
103 | if (tmp_opt.saw_tstamp) { | |
104 | tmp_opt.ts_recent = tcptw->tw_ts_recent; | |
105 | tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
106 | paws_reject = tcp_paws_reject(&tmp_opt, th->rst); | |
107 | } | |
108 | } | |
109 | ||
110 | if (tw->tw_substate == TCP_FIN_WAIT2) { | |
111 | /* Just repeat all the checks of tcp_rcv_state_process() */ | |
112 | ||
113 | /* Out of window, send ACK */ | |
114 | if (paws_reject || | |
115 | !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
116 | tcptw->tw_rcv_nxt, | |
117 | tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd)) | |
118 | return TCP_TW_ACK; | |
119 | ||
120 | if (th->rst) | |
121 | goto kill; | |
122 | ||
123 | if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt)) | |
124 | goto kill_with_rst; | |
125 | ||
126 | /* Dup ACK? */ | |
127 | if (!th->ack || | |
128 | !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) || | |
129 | TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) { | |
130 | inet_twsk_put(tw); | |
131 | return TCP_TW_SUCCESS; | |
132 | } | |
133 | ||
134 | /* New data or FIN. If new data arrive after half-duplex close, | |
135 | * reset. | |
136 | */ | |
137 | if (!th->fin || | |
138 | TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) { | |
139 | kill_with_rst: | |
140 | inet_twsk_deschedule(tw, &tcp_death_row); | |
141 | inet_twsk_put(tw); | |
142 | return TCP_TW_RST; | |
143 | } | |
144 | ||
145 | /* FIN arrived, enter true time-wait state. */ | |
146 | tw->tw_substate = TCP_TIME_WAIT; | |
147 | tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
148 | if (tmp_opt.saw_tstamp) { | |
149 | tcptw->tw_ts_recent_stamp = get_seconds(); | |
150 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; | |
151 | } | |
152 | ||
153 | /* I am shamed, but failed to make it more elegant. | |
154 | * Yes, it is direct reference to IP, which is impossible | |
155 | * to generalize to IPv6. Taking into account that IPv6 | |
156 | * do not understand recycling in any case, it not | |
157 | * a big problem in practice. --ANK */ | |
158 | if (tw->tw_family == AF_INET && | |
159 | tcp_death_row.sysctl_tw_recycle && tcptw->tw_ts_recent_stamp && | |
160 | tcp_v4_tw_remember_stamp(tw)) | |
161 | inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout, | |
162 | TCP_TIMEWAIT_LEN); | |
163 | else | |
164 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, | |
165 | TCP_TIMEWAIT_LEN); | |
166 | return TCP_TW_ACK; | |
167 | } | |
168 | ||
169 | /* | |
170 | * Now real TIME-WAIT state. | |
171 | * | |
172 | * RFC 1122: | |
173 | * "When a connection is [...] on TIME-WAIT state [...] | |
174 | * [a TCP] MAY accept a new SYN from the remote TCP to | |
175 | * reopen the connection directly, if it: | |
176 | * | |
177 | * (1) assigns its initial sequence number for the new | |
178 | * connection to be larger than the largest sequence | |
179 | * number it used on the previous connection incarnation, | |
180 | * and | |
181 | * | |
182 | * (2) returns to TIME-WAIT state if the SYN turns out | |
183 | * to be an old duplicate". | |
184 | */ | |
185 | ||
186 | if (!paws_reject && | |
187 | (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt && | |
188 | (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) { | |
189 | /* In window segment, it may be only reset or bare ack. */ | |
190 | ||
191 | if (th->rst) { | |
192 | /* This is TIME_WAIT assassination, in two flavors. | |
193 | * Oh well... nobody has a sufficient solution to this | |
194 | * protocol bug yet. | |
195 | */ | |
196 | if (sysctl_tcp_rfc1337 == 0) { | |
197 | kill: | |
198 | inet_twsk_deschedule(tw, &tcp_death_row); | |
199 | inet_twsk_put(tw); | |
200 | return TCP_TW_SUCCESS; | |
201 | } | |
202 | } | |
203 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, | |
204 | TCP_TIMEWAIT_LEN); | |
205 | ||
206 | if (tmp_opt.saw_tstamp) { | |
207 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; | |
208 | tcptw->tw_ts_recent_stamp = get_seconds(); | |
209 | } | |
210 | ||
211 | inet_twsk_put(tw); | |
212 | return TCP_TW_SUCCESS; | |
213 | } | |
214 | ||
215 | /* Out of window segment. | |
216 | ||
217 | All the segments are ACKed immediately. | |
218 | ||
219 | The only exception is new SYN. We accept it, if it is | |
220 | not old duplicate and we are not in danger to be killed | |
221 | by delayed old duplicates. RFC check is that it has | |
222 | newer sequence number works at rates <40Mbit/sec. | |
223 | However, if paws works, it is reliable AND even more, | |
224 | we even may relax silly seq space cutoff. | |
225 | ||
226 | RED-PEN: we violate main RFC requirement, if this SYN will appear | |
227 | old duplicate (i.e. we receive RST in reply to SYN-ACK), | |
228 | we must return socket to time-wait state. It is not good, | |
229 | but not fatal yet. | |
230 | */ | |
231 | ||
232 | if (th->syn && !th->rst && !th->ack && !paws_reject && | |
233 | (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) || | |
234 | (tmp_opt.saw_tstamp && | |
235 | (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) { | |
236 | u32 isn = tcptw->tw_snd_nxt + 65535 + 2; | |
237 | if (isn == 0) | |
238 | isn++; | |
239 | TCP_SKB_CB(skb)->when = isn; | |
240 | return TCP_TW_SYN; | |
241 | } | |
242 | ||
243 | if (paws_reject) | |
244 | NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED); | |
245 | ||
246 | if (!th->rst) { | |
247 | /* In this case we must reset the TIMEWAIT timer. | |
248 | * | |
249 | * If it is ACKless SYN it may be both old duplicate | |
250 | * and new good SYN with random sequence number <rcv_nxt. | |
251 | * Do not reschedule in the last case. | |
252 | */ | |
253 | if (paws_reject || th->ack) | |
254 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN, | |
255 | TCP_TIMEWAIT_LEN); | |
256 | ||
257 | /* Send ACK. Note, we do not put the bucket, | |
258 | * it will be released by caller. | |
259 | */ | |
260 | return TCP_TW_ACK; | |
261 | } | |
262 | inet_twsk_put(tw); | |
263 | return TCP_TW_SUCCESS; | |
264 | } | |
265 | ||
266 | /* | |
267 | * Move a socket to time-wait or dead fin-wait-2 state. | |
268 | */ | |
269 | void tcp_time_wait(struct sock *sk, int state, int timeo) | |
270 | { | |
271 | struct inet_timewait_sock *tw = NULL; | |
272 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
273 | const struct tcp_sock *tp = tcp_sk(sk); | |
274 | int recycle_ok = 0; | |
275 | ||
276 | if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp) | |
277 | recycle_ok = icsk->icsk_af_ops->remember_stamp(sk); | |
278 | ||
279 | if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets) | |
280 | tw = inet_twsk_alloc(sk, state); | |
281 | ||
282 | if (tw != NULL) { | |
283 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); | |
284 | const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); | |
285 | ||
286 | tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale; | |
287 | tcptw->tw_rcv_nxt = tp->rcv_nxt; | |
288 | tcptw->tw_snd_nxt = tp->snd_nxt; | |
289 | tcptw->tw_rcv_wnd = tcp_receive_window(tp); | |
290 | tcptw->tw_ts_recent = tp->rx_opt.ts_recent; | |
291 | tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp; | |
292 | ||
293 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
294 | if (tw->tw_family == PF_INET6) { | |
295 | struct ipv6_pinfo *np = inet6_sk(sk); | |
296 | struct inet6_timewait_sock *tw6; | |
297 | ||
298 | tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot); | |
299 | tw6 = inet6_twsk((struct sock *)tw); | |
300 | ipv6_addr_copy(&tw6->tw_v6_daddr, &np->daddr); | |
301 | ipv6_addr_copy(&tw6->tw_v6_rcv_saddr, &np->rcv_saddr); | |
302 | tw->tw_ipv6only = np->ipv6only; | |
303 | } | |
304 | #endif | |
305 | ||
306 | #ifdef CONFIG_TCP_MD5SIG | |
307 | /* | |
308 | * The timewait bucket does not have the key DB from the | |
309 | * sock structure. We just make a quick copy of the | |
310 | * md5 key being used (if indeed we are using one) | |
311 | * so the timewait ack generating code has the key. | |
312 | */ | |
313 | do { | |
314 | struct tcp_md5sig_key *key; | |
315 | memset(tcptw->tw_md5_key, 0, sizeof(tcptw->tw_md5_key)); | |
316 | tcptw->tw_md5_keylen = 0; | |
317 | key = tp->af_specific->md5_lookup(sk, sk); | |
318 | if (key != NULL) { | |
319 | memcpy(&tcptw->tw_md5_key, key->key, key->keylen); | |
320 | tcptw->tw_md5_keylen = key->keylen; | |
321 | if (tcp_alloc_md5sig_pool(sk) == NULL) | |
322 | BUG(); | |
323 | } | |
324 | } while (0); | |
325 | #endif | |
326 | ||
327 | /* Linkage updates. */ | |
328 | __inet_twsk_hashdance(tw, sk, &tcp_hashinfo); | |
329 | ||
330 | /* Get the TIME_WAIT timeout firing. */ | |
331 | if (timeo < rto) | |
332 | timeo = rto; | |
333 | ||
334 | if (recycle_ok) { | |
335 | tw->tw_timeout = rto; | |
336 | } else { | |
337 | tw->tw_timeout = TCP_TIMEWAIT_LEN; | |
338 | if (state == TCP_TIME_WAIT) | |
339 | timeo = TCP_TIMEWAIT_LEN; | |
340 | } | |
341 | ||
342 | inet_twsk_schedule(tw, &tcp_death_row, timeo, | |
343 | TCP_TIMEWAIT_LEN); | |
344 | inet_twsk_put(tw); | |
345 | } else { | |
346 | /* Sorry, if we're out of memory, just CLOSE this | |
347 | * socket up. We've got bigger problems than | |
348 | * non-graceful socket closings. | |
349 | */ | |
350 | LIMIT_NETDEBUG(KERN_INFO "TCP: time wait bucket table overflow\n"); | |
351 | } | |
352 | ||
353 | tcp_update_metrics(sk); | |
354 | tcp_done(sk); | |
355 | } | |
356 | ||
357 | void tcp_twsk_destructor(struct sock *sk) | |
358 | { | |
359 | #ifdef CONFIG_TCP_MD5SIG | |
360 | struct tcp_timewait_sock *twsk = tcp_twsk(sk); | |
361 | if (twsk->tw_md5_keylen) | |
362 | tcp_free_md5sig_pool(); | |
363 | #endif | |
364 | } | |
365 | ||
366 | EXPORT_SYMBOL_GPL(tcp_twsk_destructor); | |
367 | ||
368 | static inline void TCP_ECN_openreq_child(struct tcp_sock *tp, | |
369 | struct request_sock *req) | |
370 | { | |
371 | tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0; | |
372 | } | |
373 | ||
374 | /* This is not only more efficient than what we used to do, it eliminates | |
375 | * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM | |
376 | * | |
377 | * Actually, we could lots of memory writes here. tp of listening | |
378 | * socket contains all necessary default parameters. | |
379 | */ | |
380 | struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb) | |
381 | { | |
382 | struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC); | |
383 | ||
384 | if (newsk != NULL) { | |
385 | const struct inet_request_sock *ireq = inet_rsk(req); | |
386 | struct tcp_request_sock *treq = tcp_rsk(req); | |
387 | struct inet_connection_sock *newicsk = inet_csk(newsk); | |
388 | struct tcp_sock *newtp = tcp_sk(newsk); | |
389 | struct tcp_sock *oldtp = tcp_sk(sk); | |
390 | struct tcp_cookie_values *oldcvp = oldtp->cookie_values; | |
391 | ||
392 | /* TCP Cookie Transactions require space for the cookie pair, | |
393 | * as it differs for each connection. There is no need to | |
394 | * copy any s_data_payload stored at the original socket. | |
395 | * Failure will prevent resuming the connection. | |
396 | * | |
397 | * Presumed copied, in order of appearance: | |
398 | * cookie_in_always, cookie_out_never | |
399 | */ | |
400 | if (oldcvp != NULL) { | |
401 | struct tcp_cookie_values *newcvp = | |
402 | kzalloc(sizeof(*newtp->cookie_values), | |
403 | GFP_ATOMIC); | |
404 | ||
405 | if (newcvp != NULL) { | |
406 | kref_init(&newcvp->kref); | |
407 | newcvp->cookie_desired = | |
408 | oldcvp->cookie_desired; | |
409 | newtp->cookie_values = newcvp; | |
410 | } else { | |
411 | /* Not Yet Implemented */ | |
412 | newtp->cookie_values = NULL; | |
413 | } | |
414 | } | |
415 | ||
416 | /* Now setup tcp_sock */ | |
417 | newtp->pred_flags = 0; | |
418 | ||
419 | newtp->rcv_wup = newtp->copied_seq = | |
420 | newtp->rcv_nxt = treq->rcv_isn + 1; | |
421 | ||
422 | newtp->snd_sml = newtp->snd_una = | |
423 | newtp->snd_nxt = newtp->snd_up = | |
424 | treq->snt_isn + 1 + tcp_s_data_size(oldtp); | |
425 | ||
426 | tcp_prequeue_init(newtp); | |
427 | ||
428 | tcp_init_wl(newtp, treq->rcv_isn); | |
429 | ||
430 | newtp->srtt = 0; | |
431 | newtp->mdev = TCP_TIMEOUT_INIT; | |
432 | newicsk->icsk_rto = TCP_TIMEOUT_INIT; | |
433 | ||
434 | newtp->packets_out = 0; | |
435 | newtp->retrans_out = 0; | |
436 | newtp->sacked_out = 0; | |
437 | newtp->fackets_out = 0; | |
438 | newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH; | |
439 | ||
440 | /* So many TCP implementations out there (incorrectly) count the | |
441 | * initial SYN frame in their delayed-ACK and congestion control | |
442 | * algorithms that we must have the following bandaid to talk | |
443 | * efficiently to them. -DaveM | |
444 | */ | |
445 | newtp->snd_cwnd = 2; | |
446 | newtp->snd_cwnd_cnt = 0; | |
447 | newtp->bytes_acked = 0; | |
448 | ||
449 | newtp->frto_counter = 0; | |
450 | newtp->frto_highmark = 0; | |
451 | ||
452 | newicsk->icsk_ca_ops = &tcp_init_congestion_ops; | |
453 | ||
454 | tcp_set_ca_state(newsk, TCP_CA_Open); | |
455 | tcp_init_xmit_timers(newsk); | |
456 | skb_queue_head_init(&newtp->out_of_order_queue); | |
457 | newtp->write_seq = newtp->pushed_seq = | |
458 | treq->snt_isn + 1 + tcp_s_data_size(oldtp); | |
459 | ||
460 | newtp->rx_opt.saw_tstamp = 0; | |
461 | ||
462 | newtp->rx_opt.dsack = 0; | |
463 | newtp->rx_opt.num_sacks = 0; | |
464 | ||
465 | newtp->urg_data = 0; | |
466 | ||
467 | if (sock_flag(newsk, SOCK_KEEPOPEN)) | |
468 | inet_csk_reset_keepalive_timer(newsk, | |
469 | keepalive_time_when(newtp)); | |
470 | ||
471 | newtp->rx_opt.tstamp_ok = ireq->tstamp_ok; | |
472 | if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) { | |
473 | if (sysctl_tcp_fack) | |
474 | tcp_enable_fack(newtp); | |
475 | } | |
476 | newtp->window_clamp = req->window_clamp; | |
477 | newtp->rcv_ssthresh = req->rcv_wnd; | |
478 | newtp->rcv_wnd = req->rcv_wnd; | |
479 | newtp->rx_opt.wscale_ok = ireq->wscale_ok; | |
480 | if (newtp->rx_opt.wscale_ok) { | |
481 | newtp->rx_opt.snd_wscale = ireq->snd_wscale; | |
482 | newtp->rx_opt.rcv_wscale = ireq->rcv_wscale; | |
483 | } else { | |
484 | newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0; | |
485 | newtp->window_clamp = min(newtp->window_clamp, 65535U); | |
486 | } | |
487 | newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) << | |
488 | newtp->rx_opt.snd_wscale); | |
489 | newtp->max_window = newtp->snd_wnd; | |
490 | ||
491 | if (newtp->rx_opt.tstamp_ok) { | |
492 | newtp->rx_opt.ts_recent = req->ts_recent; | |
493 | newtp->rx_opt.ts_recent_stamp = get_seconds(); | |
494 | newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; | |
495 | } else { | |
496 | newtp->rx_opt.ts_recent_stamp = 0; | |
497 | newtp->tcp_header_len = sizeof(struct tcphdr); | |
498 | } | |
499 | #ifdef CONFIG_TCP_MD5SIG | |
500 | newtp->md5sig_info = NULL; /*XXX*/ | |
501 | if (newtp->af_specific->md5_lookup(sk, newsk)) | |
502 | newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED; | |
503 | #endif | |
504 | if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len) | |
505 | newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len; | |
506 | newtp->rx_opt.mss_clamp = req->mss; | |
507 | TCP_ECN_openreq_child(newtp, req); | |
508 | ||
509 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS); | |
510 | } | |
511 | return newsk; | |
512 | } | |
513 | ||
514 | /* | |
515 | * Process an incoming packet for SYN_RECV sockets represented | |
516 | * as a request_sock. | |
517 | */ | |
518 | ||
519 | struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, | |
520 | struct request_sock *req, | |
521 | struct request_sock **prev) | |
522 | { | |
523 | struct tcp_options_received tmp_opt; | |
524 | u8 *hash_location; | |
525 | struct sock *child; | |
526 | const struct tcphdr *th = tcp_hdr(skb); | |
527 | __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK); | |
528 | int paws_reject = 0; | |
529 | ||
530 | tmp_opt.saw_tstamp = 0; | |
531 | if (th->doff > (sizeof(struct tcphdr)>>2)) { | |
532 | tcp_parse_options(skb, &tmp_opt, &hash_location, 0); | |
533 | ||
534 | if (tmp_opt.saw_tstamp) { | |
535 | tmp_opt.ts_recent = req->ts_recent; | |
536 | /* We do not store true stamp, but it is not required, | |
537 | * it can be estimated (approximately) | |
538 | * from another data. | |
539 | */ | |
540 | tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans); | |
541 | paws_reject = tcp_paws_reject(&tmp_opt, th->rst); | |
542 | } | |
543 | } | |
544 | ||
545 | /* Check for pure retransmitted SYN. */ | |
546 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn && | |
547 | flg == TCP_FLAG_SYN && | |
548 | !paws_reject) { | |
549 | /* | |
550 | * RFC793 draws (Incorrectly! It was fixed in RFC1122) | |
551 | * this case on figure 6 and figure 8, but formal | |
552 | * protocol description says NOTHING. | |
553 | * To be more exact, it says that we should send ACK, | |
554 | * because this segment (at least, if it has no data) | |
555 | * is out of window. | |
556 | * | |
557 | * CONCLUSION: RFC793 (even with RFC1122) DOES NOT | |
558 | * describe SYN-RECV state. All the description | |
559 | * is wrong, we cannot believe to it and should | |
560 | * rely only on common sense and implementation | |
561 | * experience. | |
562 | * | |
563 | * Enforce "SYN-ACK" according to figure 8, figure 6 | |
564 | * of RFC793, fixed by RFC1122. | |
565 | */ | |
566 | req->rsk_ops->rtx_syn_ack(sk, req, NULL); | |
567 | return NULL; | |
568 | } | |
569 | ||
570 | /* Further reproduces section "SEGMENT ARRIVES" | |
571 | for state SYN-RECEIVED of RFC793. | |
572 | It is broken, however, it does not work only | |
573 | when SYNs are crossed. | |
574 | ||
575 | You would think that SYN crossing is impossible here, since | |
576 | we should have a SYN_SENT socket (from connect()) on our end, | |
577 | but this is not true if the crossed SYNs were sent to both | |
578 | ends by a malicious third party. We must defend against this, | |
579 | and to do that we first verify the ACK (as per RFC793, page | |
580 | 36) and reset if it is invalid. Is this a true full defense? | |
581 | To convince ourselves, let us consider a way in which the ACK | |
582 | test can still pass in this 'malicious crossed SYNs' case. | |
583 | Malicious sender sends identical SYNs (and thus identical sequence | |
584 | numbers) to both A and B: | |
585 | ||
586 | A: gets SYN, seq=7 | |
587 | B: gets SYN, seq=7 | |
588 | ||
589 | By our good fortune, both A and B select the same initial | |
590 | send sequence number of seven :-) | |
591 | ||
592 | A: sends SYN|ACK, seq=7, ack_seq=8 | |
593 | B: sends SYN|ACK, seq=7, ack_seq=8 | |
594 | ||
595 | So we are now A eating this SYN|ACK, ACK test passes. So | |
596 | does sequence test, SYN is truncated, and thus we consider | |
597 | it a bare ACK. | |
598 | ||
599 | If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this | |
600 | bare ACK. Otherwise, we create an established connection. Both | |
601 | ends (listening sockets) accept the new incoming connection and try | |
602 | to talk to each other. 8-) | |
603 | ||
604 | Note: This case is both harmless, and rare. Possibility is about the | |
605 | same as us discovering intelligent life on another plant tomorrow. | |
606 | ||
607 | But generally, we should (RFC lies!) to accept ACK | |
608 | from SYNACK both here and in tcp_rcv_state_process(). | |
609 | tcp_rcv_state_process() does not, hence, we do not too. | |
610 | ||
611 | Note that the case is absolutely generic: | |
612 | we cannot optimize anything here without | |
613 | violating protocol. All the checks must be made | |
614 | before attempt to create socket. | |
615 | */ | |
616 | ||
617 | /* RFC793 page 36: "If the connection is in any non-synchronized state ... | |
618 | * and the incoming segment acknowledges something not yet | |
619 | * sent (the segment carries an unacceptable ACK) ... | |
620 | * a reset is sent." | |
621 | * | |
622 | * Invalid ACK: reset will be sent by listening socket | |
623 | */ | |
624 | if ((flg & TCP_FLAG_ACK) && | |
625 | (TCP_SKB_CB(skb)->ack_seq != | |
626 | tcp_rsk(req)->snt_isn + 1 + tcp_s_data_size(tcp_sk(sk)))) | |
627 | return sk; | |
628 | ||
629 | /* Also, it would be not so bad idea to check rcv_tsecr, which | |
630 | * is essentially ACK extension and too early or too late values | |
631 | * should cause reset in unsynchronized states. | |
632 | */ | |
633 | ||
634 | /* RFC793: "first check sequence number". */ | |
635 | ||
636 | if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
637 | tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) { | |
638 | /* Out of window: send ACK and drop. */ | |
639 | if (!(flg & TCP_FLAG_RST)) | |
640 | req->rsk_ops->send_ack(sk, skb, req); | |
641 | if (paws_reject) | |
642 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED); | |
643 | return NULL; | |
644 | } | |
645 | ||
646 | /* In sequence, PAWS is OK. */ | |
647 | ||
648 | if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1)) | |
649 | req->ts_recent = tmp_opt.rcv_tsval; | |
650 | ||
651 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) { | |
652 | /* Truncate SYN, it is out of window starting | |
653 | at tcp_rsk(req)->rcv_isn + 1. */ | |
654 | flg &= ~TCP_FLAG_SYN; | |
655 | } | |
656 | ||
657 | /* RFC793: "second check the RST bit" and | |
658 | * "fourth, check the SYN bit" | |
659 | */ | |
660 | if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) { | |
661 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS); | |
662 | goto embryonic_reset; | |
663 | } | |
664 | ||
665 | /* ACK sequence verified above, just make sure ACK is | |
666 | * set. If ACK not set, just silently drop the packet. | |
667 | */ | |
668 | if (!(flg & TCP_FLAG_ACK)) | |
669 | return NULL; | |
670 | ||
671 | /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */ | |
672 | if (req->retrans < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept && | |
673 | TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) { | |
674 | inet_rsk(req)->acked = 1; | |
675 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP); | |
676 | return NULL; | |
677 | } | |
678 | ||
679 | /* OK, ACK is valid, create big socket and | |
680 | * feed this segment to it. It will repeat all | |
681 | * the tests. THIS SEGMENT MUST MOVE SOCKET TO | |
682 | * ESTABLISHED STATE. If it will be dropped after | |
683 | * socket is created, wait for troubles. | |
684 | */ | |
685 | child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL); | |
686 | if (child == NULL) | |
687 | goto listen_overflow; | |
688 | ||
689 | inet_csk_reqsk_queue_unlink(sk, req, prev); | |
690 | inet_csk_reqsk_queue_removed(sk, req); | |
691 | ||
692 | inet_csk_reqsk_queue_add(sk, req, child); | |
693 | return child; | |
694 | ||
695 | listen_overflow: | |
696 | if (!sysctl_tcp_abort_on_overflow) { | |
697 | inet_rsk(req)->acked = 1; | |
698 | return NULL; | |
699 | } | |
700 | ||
701 | embryonic_reset: | |
702 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS); | |
703 | if (!(flg & TCP_FLAG_RST)) | |
704 | req->rsk_ops->send_reset(sk, skb); | |
705 | ||
706 | inet_csk_reqsk_queue_drop(sk, req, prev); | |
707 | return NULL; | |
708 | } | |
709 | ||
710 | /* | |
711 | * Queue segment on the new socket if the new socket is active, | |
712 | * otherwise we just shortcircuit this and continue with | |
713 | * the new socket. | |
714 | */ | |
715 | ||
716 | int tcp_child_process(struct sock *parent, struct sock *child, | |
717 | struct sk_buff *skb) | |
718 | { | |
719 | int ret = 0; | |
720 | int state = child->sk_state; | |
721 | ||
722 | if (!sock_owned_by_user(child)) { | |
723 | ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb), | |
724 | skb->len); | |
725 | /* Wakeup parent, send SIGIO */ | |
726 | if (state == TCP_SYN_RECV && child->sk_state != state) | |
727 | parent->sk_data_ready(parent, 0); | |
728 | } else { | |
729 | /* Alas, it is possible again, because we do lookup | |
730 | * in main socket hash table and lock on listening | |
731 | * socket does not protect us more. | |
732 | */ | |
733 | __sk_add_backlog(child, skb); | |
734 | } | |
735 | ||
736 | bh_unlock_sock(child); | |
737 | sock_put(child); | |
738 | return ret; | |
739 | } | |
740 | ||
741 | EXPORT_SYMBOL(tcp_check_req); | |
742 | EXPORT_SYMBOL(tcp_child_process); | |
743 | EXPORT_SYMBOL(tcp_create_openreq_child); | |
744 | EXPORT_SYMBOL(tcp_timewait_state_process); |