]>
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 | ||
23 | #include <linux/config.h> | |
24 | #include <linux/mm.h> | |
25 | #include <linux/module.h> | |
26 | #include <linux/sysctl.h> | |
27 | #include <linux/workqueue.h> | |
28 | #include <net/tcp.h> | |
29 | #include <net/inet_common.h> | |
30 | #include <net/xfrm.h> | |
31 | ||
32 | #ifdef CONFIG_SYSCTL | |
33 | #define SYNC_INIT 0 /* let the user enable it */ | |
34 | #else | |
35 | #define SYNC_INIT 1 | |
36 | #endif | |
37 | ||
295ff7ed ACM |
38 | /* New-style handling of TIME_WAIT sockets. */ |
39 | ||
40 | static void inet_twdr_hangman(unsigned long data); | |
41 | static void inet_twdr_twkill_work(void *data); | |
42 | static void inet_twdr_twcal_tick(unsigned long data); | |
1da177e4 LT |
43 | |
44 | int sysctl_tcp_syncookies = SYNC_INIT; | |
45 | int sysctl_tcp_abort_on_overflow; | |
46 | ||
295ff7ed ACM |
47 | struct inet_timewait_death_row tcp_death_row = { |
48 | .sysctl_max_tw_buckets = NR_FILE * 2, | |
49 | .period = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS, | |
50 | .death_lock = SPIN_LOCK_UNLOCKED, | |
51 | .hashinfo = &tcp_hashinfo, | |
52 | .tw_timer = TIMER_INITIALIZER(inet_twdr_hangman, 0, | |
53 | (unsigned long)&tcp_death_row), | |
54 | .twkill_work = __WORK_INITIALIZER(tcp_death_row.twkill_work, | |
55 | inet_twdr_twkill_work, | |
56 | &tcp_death_row), | |
57 | /* Short-time timewait calendar */ | |
58 | ||
59 | .twcal_hand = -1, | |
60 | .twcal_timer = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0, | |
61 | (unsigned long)&tcp_death_row), | |
62 | }; | |
63 | ||
64 | EXPORT_SYMBOL_GPL(tcp_death_row); | |
65 | ||
66 | static void inet_twsk_schedule(struct inet_timewait_sock *tw, | |
67 | struct inet_timewait_death_row *twdr, | |
68 | const int timeo); | |
1da177e4 LT |
69 | |
70 | static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win) | |
71 | { | |
72 | if (seq == s_win) | |
73 | return 1; | |
74 | if (after(end_seq, s_win) && before(seq, e_win)) | |
75 | return 1; | |
76 | return (seq == e_win && seq == end_seq); | |
77 | } | |
78 | ||
1da177e4 LT |
79 | /* |
80 | * * Main purpose of TIME-WAIT state is to close connection gracefully, | |
81 | * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN | |
82 | * (and, probably, tail of data) and one or more our ACKs are lost. | |
83 | * * What is TIME-WAIT timeout? It is associated with maximal packet | |
84 | * lifetime in the internet, which results in wrong conclusion, that | |
85 | * it is set to catch "old duplicate segments" wandering out of their path. | |
86 | * It is not quite correct. This timeout is calculated so that it exceeds | |
87 | * maximal retransmission timeout enough to allow to lose one (or more) | |
88 | * segments sent by peer and our ACKs. This time may be calculated from RTO. | |
89 | * * When TIME-WAIT socket receives RST, it means that another end | |
90 | * finally closed and we are allowed to kill TIME-WAIT too. | |
91 | * * Second purpose of TIME-WAIT is catching old duplicate segments. | |
92 | * Well, certainly it is pure paranoia, but if we load TIME-WAIT | |
93 | * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs. | |
94 | * * If we invented some more clever way to catch duplicates | |
95 | * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs. | |
96 | * | |
97 | * The algorithm below is based on FORMAL INTERPRETATION of RFCs. | |
98 | * When you compare it to RFCs, please, read section SEGMENT ARRIVES | |
99 | * from the very beginning. | |
100 | * | |
101 | * NOTE. With recycling (and later with fin-wait-2) TW bucket | |
102 | * is _not_ stateless. It means, that strictly speaking we must | |
103 | * spinlock it. I do not want! Well, probability of misbehaviour | |
104 | * is ridiculously low and, seems, we could use some mb() tricks | |
105 | * to avoid misread sequence numbers, states etc. --ANK | |
106 | */ | |
107 | enum tcp_tw_status | |
8feaf0c0 ACM |
108 | tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb, |
109 | const struct tcphdr *th) | |
1da177e4 | 110 | { |
8feaf0c0 | 111 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
1da177e4 LT |
112 | struct tcp_options_received tmp_opt; |
113 | int paws_reject = 0; | |
114 | ||
115 | tmp_opt.saw_tstamp = 0; | |
8feaf0c0 | 116 | if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) { |
1da177e4 LT |
117 | tcp_parse_options(skb, &tmp_opt, 0); |
118 | ||
119 | if (tmp_opt.saw_tstamp) { | |
8feaf0c0 ACM |
120 | tmp_opt.ts_recent = tcptw->tw_ts_recent; |
121 | tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
1da177e4 LT |
122 | paws_reject = tcp_paws_check(&tmp_opt, th->rst); |
123 | } | |
124 | } | |
125 | ||
126 | if (tw->tw_substate == TCP_FIN_WAIT2) { | |
127 | /* Just repeat all the checks of tcp_rcv_state_process() */ | |
128 | ||
129 | /* Out of window, send ACK */ | |
130 | if (paws_reject || | |
131 | !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
8feaf0c0 ACM |
132 | tcptw->tw_rcv_nxt, |
133 | tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd)) | |
1da177e4 LT |
134 | return TCP_TW_ACK; |
135 | ||
136 | if (th->rst) | |
137 | goto kill; | |
138 | ||
8feaf0c0 | 139 | if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt)) |
1da177e4 LT |
140 | goto kill_with_rst; |
141 | ||
142 | /* Dup ACK? */ | |
8feaf0c0 | 143 | if (!after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) || |
1da177e4 | 144 | TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) { |
8feaf0c0 | 145 | inet_twsk_put(tw); |
1da177e4 LT |
146 | return TCP_TW_SUCCESS; |
147 | } | |
148 | ||
149 | /* New data or FIN. If new data arrive after half-duplex close, | |
150 | * reset. | |
151 | */ | |
152 | if (!th->fin || | |
8feaf0c0 | 153 | TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) { |
1da177e4 | 154 | kill_with_rst: |
295ff7ed | 155 | inet_twsk_deschedule(tw, &tcp_death_row); |
8feaf0c0 | 156 | inet_twsk_put(tw); |
1da177e4 LT |
157 | return TCP_TW_RST; |
158 | } | |
159 | ||
160 | /* FIN arrived, enter true time-wait state. */ | |
8feaf0c0 ACM |
161 | tw->tw_substate = TCP_TIME_WAIT; |
162 | tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
1da177e4 | 163 | if (tmp_opt.saw_tstamp) { |
8feaf0c0 ACM |
164 | tcptw->tw_ts_recent_stamp = xtime.tv_sec; |
165 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; | |
1da177e4 LT |
166 | } |
167 | ||
168 | /* I am shamed, but failed to make it more elegant. | |
169 | * Yes, it is direct reference to IP, which is impossible | |
170 | * to generalize to IPv6. Taking into account that IPv6 | |
171 | * do not undertsnad recycling in any case, it not | |
172 | * a big problem in practice. --ANK */ | |
173 | if (tw->tw_family == AF_INET && | |
295ff7ed | 174 | tcp_death_row.sysctl_tw_recycle && tcptw->tw_ts_recent_stamp && |
1da177e4 | 175 | tcp_v4_tw_remember_stamp(tw)) |
295ff7ed | 176 | inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout); |
1da177e4 | 177 | else |
295ff7ed | 178 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN); |
1da177e4 LT |
179 | return TCP_TW_ACK; |
180 | } | |
181 | ||
182 | /* | |
183 | * Now real TIME-WAIT state. | |
184 | * | |
185 | * RFC 1122: | |
186 | * "When a connection is [...] on TIME-WAIT state [...] | |
187 | * [a TCP] MAY accept a new SYN from the remote TCP to | |
188 | * reopen the connection directly, if it: | |
189 | * | |
190 | * (1) assigns its initial sequence number for the new | |
191 | * connection to be larger than the largest sequence | |
192 | * number it used on the previous connection incarnation, | |
193 | * and | |
194 | * | |
195 | * (2) returns to TIME-WAIT state if the SYN turns out | |
196 | * to be an old duplicate". | |
197 | */ | |
198 | ||
199 | if (!paws_reject && | |
8feaf0c0 | 200 | (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt && |
1da177e4 LT |
201 | (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) { |
202 | /* In window segment, it may be only reset or bare ack. */ | |
203 | ||
204 | if (th->rst) { | |
205 | /* This is TIME_WAIT assasination, in two flavors. | |
206 | * Oh well... nobody has a sufficient solution to this | |
207 | * protocol bug yet. | |
208 | */ | |
209 | if (sysctl_tcp_rfc1337 == 0) { | |
210 | kill: | |
295ff7ed | 211 | inet_twsk_deschedule(tw, &tcp_death_row); |
8feaf0c0 | 212 | inet_twsk_put(tw); |
1da177e4 LT |
213 | return TCP_TW_SUCCESS; |
214 | } | |
215 | } | |
295ff7ed | 216 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN); |
1da177e4 LT |
217 | |
218 | if (tmp_opt.saw_tstamp) { | |
8feaf0c0 ACM |
219 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; |
220 | tcptw->tw_ts_recent_stamp = xtime.tv_sec; | |
1da177e4 LT |
221 | } |
222 | ||
8feaf0c0 | 223 | inet_twsk_put(tw); |
1da177e4 LT |
224 | return TCP_TW_SUCCESS; |
225 | } | |
226 | ||
227 | /* Out of window segment. | |
228 | ||
229 | All the segments are ACKed immediately. | |
230 | ||
231 | The only exception is new SYN. We accept it, if it is | |
232 | not old duplicate and we are not in danger to be killed | |
233 | by delayed old duplicates. RFC check is that it has | |
234 | newer sequence number works at rates <40Mbit/sec. | |
235 | However, if paws works, it is reliable AND even more, | |
236 | we even may relax silly seq space cutoff. | |
237 | ||
238 | RED-PEN: we violate main RFC requirement, if this SYN will appear | |
239 | old duplicate (i.e. we receive RST in reply to SYN-ACK), | |
240 | we must return socket to time-wait state. It is not good, | |
241 | but not fatal yet. | |
242 | */ | |
243 | ||
244 | if (th->syn && !th->rst && !th->ack && !paws_reject && | |
8feaf0c0 ACM |
245 | (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) || |
246 | (tmp_opt.saw_tstamp && | |
247 | (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) { | |
248 | u32 isn = tcptw->tw_snd_nxt + 65535 + 2; | |
1da177e4 LT |
249 | if (isn == 0) |
250 | isn++; | |
251 | TCP_SKB_CB(skb)->when = isn; | |
252 | return TCP_TW_SYN; | |
253 | } | |
254 | ||
255 | if (paws_reject) | |
256 | NET_INC_STATS_BH(LINUX_MIB_PAWSESTABREJECTED); | |
257 | ||
258 | if(!th->rst) { | |
259 | /* In this case we must reset the TIMEWAIT timer. | |
260 | * | |
261 | * If it is ACKless SYN it may be both old duplicate | |
262 | * and new good SYN with random sequence number <rcv_nxt. | |
263 | * Do not reschedule in the last case. | |
264 | */ | |
265 | if (paws_reject || th->ack) | |
295ff7ed | 266 | inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN); |
1da177e4 LT |
267 | |
268 | /* Send ACK. Note, we do not put the bucket, | |
269 | * it will be released by caller. | |
270 | */ | |
271 | return TCP_TW_ACK; | |
272 | } | |
8feaf0c0 | 273 | inet_twsk_put(tw); |
1da177e4 LT |
274 | return TCP_TW_SUCCESS; |
275 | } | |
276 | ||
1da177e4 LT |
277 | /* |
278 | * Move a socket to time-wait or dead fin-wait-2 state. | |
279 | */ | |
280 | void tcp_time_wait(struct sock *sk, int state, int timeo) | |
281 | { | |
8feaf0c0 ACM |
282 | struct inet_timewait_sock *tw = NULL; |
283 | const struct tcp_sock *tp = tcp_sk(sk); | |
1da177e4 LT |
284 | int recycle_ok = 0; |
285 | ||
295ff7ed | 286 | if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp) |
1da177e4 LT |
287 | recycle_ok = tp->af_specific->remember_stamp(sk); |
288 | ||
295ff7ed | 289 | if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets) |
c676270b | 290 | tw = inet_twsk_alloc(sk, state); |
1da177e4 | 291 | |
8feaf0c0 ACM |
292 | if (tw != NULL) { |
293 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); | |
463c84b9 ACM |
294 | const struct inet_connection_sock *icsk = inet_csk(sk); |
295 | const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); | |
8feaf0c0 | 296 | |
1da177e4 | 297 | tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale; |
8feaf0c0 ACM |
298 | tcptw->tw_rcv_nxt = tp->rcv_nxt; |
299 | tcptw->tw_snd_nxt = tp->snd_nxt; | |
300 | tcptw->tw_rcv_wnd = tcp_receive_window(tp); | |
301 | tcptw->tw_ts_recent = tp->rx_opt.ts_recent; | |
302 | tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp; | |
1da177e4 LT |
303 | |
304 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) | |
305 | if (tw->tw_family == PF_INET6) { | |
306 | struct ipv6_pinfo *np = inet6_sk(sk); | |
8feaf0c0 | 307 | struct tcp6_timewait_sock *tcp6tw = tcp6_twsk((struct sock *)tw); |
1da177e4 | 308 | |
8feaf0c0 ACM |
309 | ipv6_addr_copy(&tcp6tw->tw_v6_daddr, &np->daddr); |
310 | ipv6_addr_copy(&tcp6tw->tw_v6_rcv_saddr, &np->rcv_saddr); | |
311 | tw->tw_ipv6only = np->ipv6only; | |
c676270b | 312 | } |
1da177e4 LT |
313 | #endif |
314 | /* Linkage updates. */ | |
e48c414e | 315 | __inet_twsk_hashdance(tw, sk, &tcp_hashinfo); |
1da177e4 LT |
316 | |
317 | /* Get the TIME_WAIT timeout firing. */ | |
318 | if (timeo < rto) | |
319 | timeo = rto; | |
320 | ||
321 | if (recycle_ok) { | |
322 | tw->tw_timeout = rto; | |
323 | } else { | |
324 | tw->tw_timeout = TCP_TIMEWAIT_LEN; | |
325 | if (state == TCP_TIME_WAIT) | |
326 | timeo = TCP_TIMEWAIT_LEN; | |
327 | } | |
328 | ||
295ff7ed | 329 | inet_twsk_schedule(tw, &tcp_death_row, timeo); |
8feaf0c0 | 330 | inet_twsk_put(tw); |
1da177e4 LT |
331 | } else { |
332 | /* Sorry, if we're out of memory, just CLOSE this | |
333 | * socket up. We've got bigger problems than | |
334 | * non-graceful socket closings. | |
335 | */ | |
336 | if (net_ratelimit()) | |
337 | printk(KERN_INFO "TCP: time wait bucket table overflow\n"); | |
338 | } | |
339 | ||
340 | tcp_update_metrics(sk); | |
341 | tcp_done(sk); | |
342 | } | |
343 | ||
1da177e4 | 344 | /* Returns non-zero if quota exceeded. */ |
295ff7ed ACM |
345 | static int inet_twdr_do_twkill_work(struct inet_timewait_death_row *twdr, |
346 | const int slot) | |
1da177e4 | 347 | { |
8feaf0c0 | 348 | struct inet_timewait_sock *tw; |
1da177e4 LT |
349 | struct hlist_node *node; |
350 | unsigned int killed; | |
351 | int ret; | |
352 | ||
353 | /* NOTE: compare this to previous version where lock | |
354 | * was released after detaching chain. It was racy, | |
355 | * because tw buckets are scheduled in not serialized context | |
356 | * in 2.3 (with netfilter), and with softnet it is common, because | |
357 | * soft irqs are not sequenced. | |
358 | */ | |
359 | killed = 0; | |
360 | ret = 0; | |
361 | rescan: | |
295ff7ed | 362 | inet_twsk_for_each_inmate(tw, node, &twdr->cells[slot]) { |
8feaf0c0 | 363 | __inet_twsk_del_dead_node(tw); |
295ff7ed ACM |
364 | spin_unlock(&twdr->death_lock); |
365 | __inet_twsk_kill(tw, twdr->hashinfo); | |
8feaf0c0 | 366 | inet_twsk_put(tw); |
1da177e4 | 367 | killed++; |
295ff7ed ACM |
368 | spin_lock(&twdr->death_lock); |
369 | if (killed > INET_TWDR_TWKILL_QUOTA) { | |
1da177e4 LT |
370 | ret = 1; |
371 | break; | |
372 | } | |
373 | ||
295ff7ed | 374 | /* While we dropped twdr->death_lock, another cpu may have |
1da177e4 LT |
375 | * killed off the next TW bucket in the list, therefore |
376 | * do a fresh re-read of the hlist head node with the | |
377 | * lock reacquired. We still use the hlist traversal | |
378 | * macro in order to get the prefetches. | |
379 | */ | |
380 | goto rescan; | |
381 | } | |
382 | ||
295ff7ed | 383 | twdr->tw_count -= killed; |
1da177e4 LT |
384 | NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITED, killed); |
385 | ||
386 | return ret; | |
387 | } | |
388 | ||
295ff7ed | 389 | static void inet_twdr_hangman(unsigned long data) |
1da177e4 | 390 | { |
295ff7ed ACM |
391 | struct inet_timewait_death_row *twdr; |
392 | int unsigned need_timer; | |
1da177e4 | 393 | |
295ff7ed ACM |
394 | twdr = (struct inet_timewait_death_row *)data; |
395 | spin_lock(&twdr->death_lock); | |
1da177e4 | 396 | |
295ff7ed | 397 | if (twdr->tw_count == 0) |
1da177e4 LT |
398 | goto out; |
399 | ||
400 | need_timer = 0; | |
295ff7ed ACM |
401 | if (inet_twdr_do_twkill_work(twdr, twdr->slot)) { |
402 | twdr->thread_slots |= (1 << twdr->slot); | |
1da177e4 | 403 | mb(); |
295ff7ed | 404 | schedule_work(&twdr->twkill_work); |
1da177e4 LT |
405 | need_timer = 1; |
406 | } else { | |
407 | /* We purged the entire slot, anything left? */ | |
295ff7ed | 408 | if (twdr->tw_count) |
1da177e4 LT |
409 | need_timer = 1; |
410 | } | |
295ff7ed | 411 | twdr->slot = ((twdr->slot + 1) & (INET_TWDR_TWKILL_SLOTS - 1)); |
1da177e4 | 412 | if (need_timer) |
295ff7ed | 413 | mod_timer(&twdr->tw_timer, jiffies + twdr->period); |
1da177e4 | 414 | out: |
295ff7ed | 415 | spin_unlock(&twdr->death_lock); |
1da177e4 LT |
416 | } |
417 | ||
418 | extern void twkill_slots_invalid(void); | |
419 | ||
295ff7ed | 420 | static void inet_twdr_twkill_work(void *data) |
1da177e4 | 421 | { |
295ff7ed | 422 | struct inet_timewait_death_row *twdr = data; |
1da177e4 LT |
423 | int i; |
424 | ||
295ff7ed | 425 | if ((INET_TWDR_TWKILL_SLOTS - 1) > (sizeof(twdr->thread_slots) * 8)) |
1da177e4 LT |
426 | twkill_slots_invalid(); |
427 | ||
295ff7ed ACM |
428 | while (twdr->thread_slots) { |
429 | spin_lock_bh(&twdr->death_lock); | |
430 | for (i = 0; i < INET_TWDR_TWKILL_SLOTS; i++) { | |
431 | if (!(twdr->thread_slots & (1 << i))) | |
1da177e4 LT |
432 | continue; |
433 | ||
295ff7ed | 434 | while (inet_twdr_do_twkill_work(twdr, i) != 0) { |
1da177e4 | 435 | if (need_resched()) { |
295ff7ed | 436 | spin_unlock_bh(&twdr->death_lock); |
1da177e4 | 437 | schedule(); |
295ff7ed | 438 | spin_lock_bh(&twdr->death_lock); |
1da177e4 LT |
439 | } |
440 | } | |
441 | ||
295ff7ed | 442 | twdr->thread_slots &= ~(1 << i); |
1da177e4 | 443 | } |
295ff7ed | 444 | spin_unlock_bh(&twdr->death_lock); |
1da177e4 LT |
445 | } |
446 | } | |
447 | ||
448 | /* These are always called from BH context. See callers in | |
449 | * tcp_input.c to verify this. | |
450 | */ | |
451 | ||
452 | /* This is for handling early-kills of TIME_WAIT sockets. */ | |
295ff7ed ACM |
453 | void inet_twsk_deschedule(struct inet_timewait_sock *tw, |
454 | struct inet_timewait_death_row *twdr) | |
1da177e4 | 455 | { |
295ff7ed | 456 | spin_lock(&twdr->death_lock); |
8feaf0c0 ACM |
457 | if (inet_twsk_del_dead_node(tw)) { |
458 | inet_twsk_put(tw); | |
295ff7ed ACM |
459 | if (--twdr->tw_count == 0) |
460 | del_timer(&twdr->tw_timer); | |
1da177e4 | 461 | } |
295ff7ed ACM |
462 | spin_unlock(&twdr->death_lock); |
463 | __inet_twsk_kill(tw, twdr->hashinfo); | |
1da177e4 LT |
464 | } |
465 | ||
295ff7ed ACM |
466 | static void inet_twsk_schedule(struct inet_timewait_sock *tw, |
467 | struct inet_timewait_death_row *twdr, | |
468 | const int timeo) | |
1da177e4 LT |
469 | { |
470 | struct hlist_head *list; | |
471 | int slot; | |
472 | ||
473 | /* timeout := RTO * 3.5 | |
474 | * | |
475 | * 3.5 = 1+2+0.5 to wait for two retransmits. | |
476 | * | |
477 | * RATIONALE: if FIN arrived and we entered TIME-WAIT state, | |
478 | * our ACK acking that FIN can be lost. If N subsequent retransmitted | |
479 | * FINs (or previous seqments) are lost (probability of such event | |
480 | * is p^(N+1), where p is probability to lose single packet and | |
481 | * time to detect the loss is about RTO*(2^N - 1) with exponential | |
482 | * backoff). Normal timewait length is calculated so, that we | |
483 | * waited at least for one retransmitted FIN (maximal RTO is 120sec). | |
484 | * [ BTW Linux. following BSD, violates this requirement waiting | |
485 | * only for 60sec, we should wait at least for 240 secs. | |
486 | * Well, 240 consumes too much of resources 8) | |
487 | * ] | |
488 | * This interval is not reduced to catch old duplicate and | |
489 | * responces to our wandering segments living for two MSLs. | |
490 | * However, if we use PAWS to detect | |
491 | * old duplicates, we can reduce the interval to bounds required | |
492 | * by RTO, rather than MSL. So, if peer understands PAWS, we | |
493 | * kill tw bucket after 3.5*RTO (it is important that this number | |
494 | * is greater than TS tick!) and detect old duplicates with help | |
495 | * of PAWS. | |
496 | */ | |
295ff7ed | 497 | slot = (timeo + (1 << INET_TWDR_RECYCLE_TICK) - 1) >> INET_TWDR_RECYCLE_TICK; |
1da177e4 | 498 | |
295ff7ed | 499 | spin_lock(&twdr->death_lock); |
1da177e4 LT |
500 | |
501 | /* Unlink it, if it was scheduled */ | |
8feaf0c0 | 502 | if (inet_twsk_del_dead_node(tw)) |
295ff7ed | 503 | twdr->tw_count--; |
1da177e4 LT |
504 | else |
505 | atomic_inc(&tw->tw_refcnt); | |
506 | ||
295ff7ed | 507 | if (slot >= INET_TWDR_RECYCLE_SLOTS) { |
1da177e4 LT |
508 | /* Schedule to slow timer */ |
509 | if (timeo >= TCP_TIMEWAIT_LEN) { | |
295ff7ed | 510 | slot = INET_TWDR_TWKILL_SLOTS - 1; |
1da177e4 | 511 | } else { |
295ff7ed ACM |
512 | slot = (timeo + twdr->period - 1) / twdr->period; |
513 | if (slot >= INET_TWDR_TWKILL_SLOTS) | |
514 | slot = INET_TWDR_TWKILL_SLOTS - 1; | |
1da177e4 LT |
515 | } |
516 | tw->tw_ttd = jiffies + timeo; | |
295ff7ed ACM |
517 | slot = (twdr->slot + slot) & (INET_TWDR_TWKILL_SLOTS - 1); |
518 | list = &twdr->cells[slot]; | |
1da177e4 | 519 | } else { |
295ff7ed ACM |
520 | tw->tw_ttd = jiffies + (slot << INET_TWDR_RECYCLE_TICK); |
521 | ||
522 | if (twdr->twcal_hand < 0) { | |
523 | twdr->twcal_hand = 0; | |
524 | twdr->twcal_jiffie = jiffies; | |
525 | twdr->twcal_timer.expires = twdr->twcal_jiffie + | |
526 | (slot << INET_TWDR_RECYCLE_TICK); | |
527 | add_timer(&twdr->twcal_timer); | |
1da177e4 | 528 | } else { |
295ff7ed ACM |
529 | if (time_after(twdr->twcal_timer.expires, |
530 | jiffies + (slot << INET_TWDR_RECYCLE_TICK))) | |
531 | mod_timer(&twdr->twcal_timer, | |
532 | jiffies + (slot << INET_TWDR_RECYCLE_TICK)); | |
533 | slot = (twdr->twcal_hand + slot) & (INET_TWDR_RECYCLE_SLOTS - 1); | |
1da177e4 | 534 | } |
295ff7ed | 535 | list = &twdr->twcal_row[slot]; |
1da177e4 LT |
536 | } |
537 | ||
538 | hlist_add_head(&tw->tw_death_node, list); | |
539 | ||
295ff7ed ACM |
540 | if (twdr->tw_count++ == 0) |
541 | mod_timer(&twdr->tw_timer, jiffies + twdr->period); | |
542 | spin_unlock(&twdr->death_lock); | |
1da177e4 LT |
543 | } |
544 | ||
295ff7ed | 545 | void inet_twdr_twcal_tick(unsigned long data) |
1da177e4 | 546 | { |
295ff7ed | 547 | struct inet_timewait_death_row *twdr; |
1da177e4 LT |
548 | int n, slot; |
549 | unsigned long j; | |
550 | unsigned long now = jiffies; | |
551 | int killed = 0; | |
552 | int adv = 0; | |
553 | ||
295ff7ed ACM |
554 | twdr = (struct inet_timewait_death_row *)data; |
555 | ||
556 | spin_lock(&twdr->death_lock); | |
557 | if (twdr->twcal_hand < 0) | |
1da177e4 LT |
558 | goto out; |
559 | ||
295ff7ed ACM |
560 | slot = twdr->twcal_hand; |
561 | j = twdr->twcal_jiffie; | |
1da177e4 | 562 | |
295ff7ed | 563 | for (n = 0; n < INET_TWDR_RECYCLE_SLOTS; n++) { |
1da177e4 LT |
564 | if (time_before_eq(j, now)) { |
565 | struct hlist_node *node, *safe; | |
8feaf0c0 | 566 | struct inet_timewait_sock *tw; |
1da177e4 | 567 | |
8feaf0c0 | 568 | inet_twsk_for_each_inmate_safe(tw, node, safe, |
295ff7ed | 569 | &twdr->twcal_row[slot]) { |
8feaf0c0 | 570 | __inet_twsk_del_dead_node(tw); |
295ff7ed | 571 | __inet_twsk_kill(tw, twdr->hashinfo); |
8feaf0c0 | 572 | inet_twsk_put(tw); |
1da177e4 LT |
573 | killed++; |
574 | } | |
575 | } else { | |
576 | if (!adv) { | |
577 | adv = 1; | |
295ff7ed ACM |
578 | twdr->twcal_jiffie = j; |
579 | twdr->twcal_hand = slot; | |
1da177e4 LT |
580 | } |
581 | ||
295ff7ed ACM |
582 | if (!hlist_empty(&twdr->twcal_row[slot])) { |
583 | mod_timer(&twdr->twcal_timer, j); | |
1da177e4 LT |
584 | goto out; |
585 | } | |
586 | } | |
295ff7ed ACM |
587 | j += 1 << INET_TWDR_RECYCLE_TICK; |
588 | slot = (slot + 1) & (INET_TWDR_RECYCLE_SLOTS - 1); | |
1da177e4 | 589 | } |
295ff7ed | 590 | twdr->twcal_hand = -1; |
1da177e4 LT |
591 | |
592 | out: | |
295ff7ed ACM |
593 | if ((twdr->tw_count -= killed) == 0) |
594 | del_timer(&twdr->tw_timer); | |
1da177e4 | 595 | NET_ADD_STATS_BH(LINUX_MIB_TIMEWAITKILLED, killed); |
295ff7ed | 596 | spin_unlock(&twdr->death_lock); |
1da177e4 LT |
597 | } |
598 | ||
599 | /* This is not only more efficient than what we used to do, it eliminates | |
600 | * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM | |
601 | * | |
602 | * Actually, we could lots of memory writes here. tp of listening | |
603 | * socket contains all necessary default parameters. | |
604 | */ | |
60236fdd | 605 | struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb) |
1da177e4 | 606 | { |
9f1d2604 | 607 | struct sock *newsk = inet_csk_clone(sk, req, GFP_ATOMIC); |
1da177e4 | 608 | |
87d11ceb | 609 | if (newsk != NULL) { |
9f1d2604 | 610 | const struct inet_request_sock *ireq = inet_rsk(req); |
2e6599cb | 611 | struct tcp_request_sock *treq = tcp_rsk(req); |
9f1d2604 | 612 | struct inet_connection_sock *newicsk = inet_csk(sk); |
1da177e4 | 613 | struct tcp_sock *newtp; |
1da177e4 | 614 | |
1da177e4 LT |
615 | /* Now setup tcp_sock */ |
616 | newtp = tcp_sk(newsk); | |
617 | newtp->pred_flags = 0; | |
2e6599cb | 618 | newtp->rcv_nxt = treq->rcv_isn + 1; |
87d11ceb | 619 | newtp->snd_nxt = newtp->snd_una = newtp->snd_sml = treq->snt_isn + 1; |
1da177e4 LT |
620 | |
621 | tcp_prequeue_init(newtp); | |
622 | ||
2e6599cb | 623 | tcp_init_wl(newtp, treq->snt_isn, treq->rcv_isn); |
1da177e4 | 624 | |
1da177e4 LT |
625 | newtp->srtt = 0; |
626 | newtp->mdev = TCP_TIMEOUT_INIT; | |
463c84b9 | 627 | newicsk->icsk_rto = TCP_TIMEOUT_INIT; |
1da177e4 LT |
628 | |
629 | newtp->packets_out = 0; | |
630 | newtp->left_out = 0; | |
631 | newtp->retrans_out = 0; | |
632 | newtp->sacked_out = 0; | |
633 | newtp->fackets_out = 0; | |
634 | newtp->snd_ssthresh = 0x7fffffff; | |
635 | ||
636 | /* So many TCP implementations out there (incorrectly) count the | |
637 | * initial SYN frame in their delayed-ACK and congestion control | |
638 | * algorithms that we must have the following bandaid to talk | |
639 | * efficiently to them. -DaveM | |
640 | */ | |
641 | newtp->snd_cwnd = 2; | |
642 | newtp->snd_cwnd_cnt = 0; | |
643 | ||
644 | newtp->frto_counter = 0; | |
645 | newtp->frto_highmark = 0; | |
646 | ||
317a76f9 SH |
647 | newtp->ca_ops = &tcp_reno; |
648 | ||
1da177e4 LT |
649 | tcp_set_ca_state(newtp, TCP_CA_Open); |
650 | tcp_init_xmit_timers(newsk); | |
651 | skb_queue_head_init(&newtp->out_of_order_queue); | |
2e6599cb ACM |
652 | newtp->rcv_wup = treq->rcv_isn + 1; |
653 | newtp->write_seq = treq->snt_isn + 1; | |
1da177e4 | 654 | newtp->pushed_seq = newtp->write_seq; |
2e6599cb | 655 | newtp->copied_seq = treq->rcv_isn + 1; |
1da177e4 LT |
656 | |
657 | newtp->rx_opt.saw_tstamp = 0; | |
658 | ||
659 | newtp->rx_opt.dsack = 0; | |
660 | newtp->rx_opt.eff_sacks = 0; | |
661 | ||
662 | newtp->probes_out = 0; | |
663 | newtp->rx_opt.num_sacks = 0; | |
664 | newtp->urg_data = 0; | |
1da177e4 | 665 | |
1da177e4 | 666 | if (sock_flag(newsk, SOCK_KEEPOPEN)) |
463c84b9 ACM |
667 | inet_csk_reset_keepalive_timer(newsk, |
668 | keepalive_time_when(newtp)); | |
1da177e4 | 669 | |
2e6599cb ACM |
670 | newtp->rx_opt.tstamp_ok = ireq->tstamp_ok; |
671 | if((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) { | |
1da177e4 LT |
672 | if (sysctl_tcp_fack) |
673 | newtp->rx_opt.sack_ok |= 2; | |
674 | } | |
675 | newtp->window_clamp = req->window_clamp; | |
676 | newtp->rcv_ssthresh = req->rcv_wnd; | |
677 | newtp->rcv_wnd = req->rcv_wnd; | |
2e6599cb | 678 | newtp->rx_opt.wscale_ok = ireq->wscale_ok; |
1da177e4 | 679 | if (newtp->rx_opt.wscale_ok) { |
2e6599cb ACM |
680 | newtp->rx_opt.snd_wscale = ireq->snd_wscale; |
681 | newtp->rx_opt.rcv_wscale = ireq->rcv_wscale; | |
1da177e4 LT |
682 | } else { |
683 | newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0; | |
684 | newtp->window_clamp = min(newtp->window_clamp, 65535U); | |
685 | } | |
686 | newtp->snd_wnd = ntohs(skb->h.th->window) << newtp->rx_opt.snd_wscale; | |
687 | newtp->max_window = newtp->snd_wnd; | |
688 | ||
689 | if (newtp->rx_opt.tstamp_ok) { | |
690 | newtp->rx_opt.ts_recent = req->ts_recent; | |
691 | newtp->rx_opt.ts_recent_stamp = xtime.tv_sec; | |
692 | newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; | |
693 | } else { | |
694 | newtp->rx_opt.ts_recent_stamp = 0; | |
695 | newtp->tcp_header_len = sizeof(struct tcphdr); | |
696 | } | |
697 | if (skb->len >= TCP_MIN_RCVMSS+newtp->tcp_header_len) | |
463c84b9 | 698 | newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len; |
1da177e4 LT |
699 | newtp->rx_opt.mss_clamp = req->mss; |
700 | TCP_ECN_openreq_child(newtp, req); | |
701 | if (newtp->ecn_flags&TCP_ECN_OK) | |
702 | sock_set_flag(newsk, SOCK_NO_LARGESEND); | |
703 | ||
1da177e4 LT |
704 | TCP_INC_STATS_BH(TCP_MIB_PASSIVEOPENS); |
705 | } | |
706 | return newsk; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Process an incoming packet for SYN_RECV sockets represented | |
60236fdd | 711 | * as a request_sock. |
1da177e4 LT |
712 | */ |
713 | ||
714 | struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb, | |
60236fdd ACM |
715 | struct request_sock *req, |
716 | struct request_sock **prev) | |
1da177e4 LT |
717 | { |
718 | struct tcphdr *th = skb->h.th; | |
719 | struct tcp_sock *tp = tcp_sk(sk); | |
720 | u32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK); | |
721 | int paws_reject = 0; | |
722 | struct tcp_options_received tmp_opt; | |
723 | struct sock *child; | |
724 | ||
725 | tmp_opt.saw_tstamp = 0; | |
726 | if (th->doff > (sizeof(struct tcphdr)>>2)) { | |
727 | tcp_parse_options(skb, &tmp_opt, 0); | |
728 | ||
729 | if (tmp_opt.saw_tstamp) { | |
730 | tmp_opt.ts_recent = req->ts_recent; | |
731 | /* We do not store true stamp, but it is not required, | |
732 | * it can be estimated (approximately) | |
733 | * from another data. | |
734 | */ | |
735 | tmp_opt.ts_recent_stamp = xtime.tv_sec - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans); | |
736 | paws_reject = tcp_paws_check(&tmp_opt, th->rst); | |
737 | } | |
738 | } | |
739 | ||
740 | /* Check for pure retransmitted SYN. */ | |
2e6599cb | 741 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn && |
1da177e4 LT |
742 | flg == TCP_FLAG_SYN && |
743 | !paws_reject) { | |
744 | /* | |
745 | * RFC793 draws (Incorrectly! It was fixed in RFC1122) | |
746 | * this case on figure 6 and figure 8, but formal | |
747 | * protocol description says NOTHING. | |
748 | * To be more exact, it says that we should send ACK, | |
749 | * because this segment (at least, if it has no data) | |
750 | * is out of window. | |
751 | * | |
752 | * CONCLUSION: RFC793 (even with RFC1122) DOES NOT | |
753 | * describe SYN-RECV state. All the description | |
754 | * is wrong, we cannot believe to it and should | |
755 | * rely only on common sense and implementation | |
756 | * experience. | |
757 | * | |
758 | * Enforce "SYN-ACK" according to figure 8, figure 6 | |
759 | * of RFC793, fixed by RFC1122. | |
760 | */ | |
60236fdd | 761 | req->rsk_ops->rtx_syn_ack(sk, req, NULL); |
1da177e4 LT |
762 | return NULL; |
763 | } | |
764 | ||
765 | /* Further reproduces section "SEGMENT ARRIVES" | |
766 | for state SYN-RECEIVED of RFC793. | |
767 | It is broken, however, it does not work only | |
768 | when SYNs are crossed. | |
769 | ||
770 | You would think that SYN crossing is impossible here, since | |
771 | we should have a SYN_SENT socket (from connect()) on our end, | |
772 | but this is not true if the crossed SYNs were sent to both | |
773 | ends by a malicious third party. We must defend against this, | |
774 | and to do that we first verify the ACK (as per RFC793, page | |
775 | 36) and reset if it is invalid. Is this a true full defense? | |
776 | To convince ourselves, let us consider a way in which the ACK | |
777 | test can still pass in this 'malicious crossed SYNs' case. | |
778 | Malicious sender sends identical SYNs (and thus identical sequence | |
779 | numbers) to both A and B: | |
780 | ||
781 | A: gets SYN, seq=7 | |
782 | B: gets SYN, seq=7 | |
783 | ||
784 | By our good fortune, both A and B select the same initial | |
785 | send sequence number of seven :-) | |
786 | ||
787 | A: sends SYN|ACK, seq=7, ack_seq=8 | |
788 | B: sends SYN|ACK, seq=7, ack_seq=8 | |
789 | ||
790 | So we are now A eating this SYN|ACK, ACK test passes. So | |
791 | does sequence test, SYN is truncated, and thus we consider | |
792 | it a bare ACK. | |
793 | ||
295f7324 ACM |
794 | If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this |
795 | bare ACK. Otherwise, we create an established connection. Both | |
796 | ends (listening sockets) accept the new incoming connection and try | |
797 | to talk to each other. 8-) | |
1da177e4 LT |
798 | |
799 | Note: This case is both harmless, and rare. Possibility is about the | |
800 | same as us discovering intelligent life on another plant tomorrow. | |
801 | ||
802 | But generally, we should (RFC lies!) to accept ACK | |
803 | from SYNACK both here and in tcp_rcv_state_process(). | |
804 | tcp_rcv_state_process() does not, hence, we do not too. | |
805 | ||
806 | Note that the case is absolutely generic: | |
807 | we cannot optimize anything here without | |
808 | violating protocol. All the checks must be made | |
809 | before attempt to create socket. | |
810 | */ | |
811 | ||
812 | /* RFC793 page 36: "If the connection is in any non-synchronized state ... | |
813 | * and the incoming segment acknowledges something not yet | |
814 | * sent (the segment carries an unaccaptable ACK) ... | |
815 | * a reset is sent." | |
816 | * | |
817 | * Invalid ACK: reset will be sent by listening socket | |
818 | */ | |
819 | if ((flg & TCP_FLAG_ACK) && | |
2e6599cb | 820 | (TCP_SKB_CB(skb)->ack_seq != tcp_rsk(req)->snt_isn + 1)) |
1da177e4 LT |
821 | return sk; |
822 | ||
823 | /* Also, it would be not so bad idea to check rcv_tsecr, which | |
824 | * is essentially ACK extension and too early or too late values | |
825 | * should cause reset in unsynchronized states. | |
826 | */ | |
827 | ||
828 | /* RFC793: "first check sequence number". */ | |
829 | ||
830 | if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
2e6599cb | 831 | tcp_rsk(req)->rcv_isn + 1, tcp_rsk(req)->rcv_isn + 1 + req->rcv_wnd)) { |
1da177e4 LT |
832 | /* Out of window: send ACK and drop. */ |
833 | if (!(flg & TCP_FLAG_RST)) | |
60236fdd | 834 | req->rsk_ops->send_ack(skb, req); |
1da177e4 LT |
835 | if (paws_reject) |
836 | NET_INC_STATS_BH(LINUX_MIB_PAWSESTABREJECTED); | |
837 | return NULL; | |
838 | } | |
839 | ||
840 | /* In sequence, PAWS is OK. */ | |
841 | ||
2e6599cb | 842 | if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_isn + 1)) |
1da177e4 LT |
843 | req->ts_recent = tmp_opt.rcv_tsval; |
844 | ||
2e6599cb | 845 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) { |
1da177e4 | 846 | /* Truncate SYN, it is out of window starting |
2e6599cb | 847 | at tcp_rsk(req)->rcv_isn + 1. */ |
1da177e4 LT |
848 | flg &= ~TCP_FLAG_SYN; |
849 | } | |
850 | ||
851 | /* RFC793: "second check the RST bit" and | |
852 | * "fourth, check the SYN bit" | |
853 | */ | |
854 | if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) | |
855 | goto embryonic_reset; | |
856 | ||
857 | /* ACK sequence verified above, just make sure ACK is | |
858 | * set. If ACK not set, just silently drop the packet. | |
859 | */ | |
860 | if (!(flg & TCP_FLAG_ACK)) | |
861 | return NULL; | |
862 | ||
863 | /* If TCP_DEFER_ACCEPT is set, drop bare ACK. */ | |
295f7324 ACM |
864 | if (inet_csk(sk)->icsk_accept_queue.rskq_defer_accept && |
865 | TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) { | |
2e6599cb | 866 | inet_rsk(req)->acked = 1; |
1da177e4 LT |
867 | return NULL; |
868 | } | |
869 | ||
870 | /* OK, ACK is valid, create big socket and | |
871 | * feed this segment to it. It will repeat all | |
872 | * the tests. THIS SEGMENT MUST MOVE SOCKET TO | |
873 | * ESTABLISHED STATE. If it will be dropped after | |
874 | * socket is created, wait for troubles. | |
875 | */ | |
876 | child = tp->af_specific->syn_recv_sock(sk, skb, req, NULL); | |
877 | if (child == NULL) | |
878 | goto listen_overflow; | |
879 | ||
463c84b9 ACM |
880 | inet_csk_reqsk_queue_unlink(sk, req, prev); |
881 | inet_csk_reqsk_queue_removed(sk, req); | |
1da177e4 | 882 | |
463c84b9 | 883 | inet_csk_reqsk_queue_add(sk, req, child); |
1da177e4 LT |
884 | return child; |
885 | ||
886 | listen_overflow: | |
887 | if (!sysctl_tcp_abort_on_overflow) { | |
2e6599cb | 888 | inet_rsk(req)->acked = 1; |
1da177e4 LT |
889 | return NULL; |
890 | } | |
891 | ||
892 | embryonic_reset: | |
893 | NET_INC_STATS_BH(LINUX_MIB_EMBRYONICRSTS); | |
894 | if (!(flg & TCP_FLAG_RST)) | |
60236fdd | 895 | req->rsk_ops->send_reset(skb); |
1da177e4 | 896 | |
463c84b9 | 897 | inet_csk_reqsk_queue_drop(sk, req, prev); |
1da177e4 LT |
898 | return NULL; |
899 | } | |
900 | ||
901 | /* | |
902 | * Queue segment on the new socket if the new socket is active, | |
903 | * otherwise we just shortcircuit this and continue with | |
904 | * the new socket. | |
905 | */ | |
906 | ||
907 | int tcp_child_process(struct sock *parent, struct sock *child, | |
908 | struct sk_buff *skb) | |
909 | { | |
910 | int ret = 0; | |
911 | int state = child->sk_state; | |
912 | ||
913 | if (!sock_owned_by_user(child)) { | |
914 | ret = tcp_rcv_state_process(child, skb, skb->h.th, skb->len); | |
915 | ||
916 | /* Wakeup parent, send SIGIO */ | |
917 | if (state == TCP_SYN_RECV && child->sk_state != state) | |
918 | parent->sk_data_ready(parent, 0); | |
919 | } else { | |
920 | /* Alas, it is possible again, because we do lookup | |
921 | * in main socket hash table and lock on listening | |
922 | * socket does not protect us more. | |
923 | */ | |
924 | sk_add_backlog(child, skb); | |
925 | } | |
926 | ||
927 | bh_unlock_sock(child); | |
928 | sock_put(child); | |
929 | return ret; | |
930 | } | |
931 | ||
932 | EXPORT_SYMBOL(tcp_check_req); | |
933 | EXPORT_SYMBOL(tcp_child_process); | |
934 | EXPORT_SYMBOL(tcp_create_openreq_child); | |
935 | EXPORT_SYMBOL(tcp_timewait_state_process); | |
295ff7ed | 936 | EXPORT_SYMBOL(inet_twsk_deschedule); |