Commit | Line | Data |
---|---|---|
e48c414e ACM |
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 | * Generic TIME_WAIT sockets functions | |
7 | * | |
8 | * From code orinally in TCP | |
9 | */ | |
10 | ||
172589cc | 11 | #include <linux/kernel.h> |
9e337b0f | 12 | #include <linux/kmemcheck.h> |
e48c414e ACM |
13 | #include <net/inet_hashtables.h> |
14 | #include <net/inet_timewait_sock.h> | |
696ab2d3 | 15 | #include <net/ip.h> |
e48c414e ACM |
16 | |
17 | /* Must be called with locally disabled BHs. */ | |
acd159b6 AB |
18 | static void __inet_twsk_kill(struct inet_timewait_sock *tw, |
19 | struct inet_hashinfo *hashinfo) | |
e48c414e ACM |
20 | { |
21 | struct inet_bind_hashbucket *bhead; | |
22 | struct inet_bind_bucket *tb; | |
23 | /* Unlink from established hashes. */ | |
9db66bdc | 24 | spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash); |
e48c414e | 25 | |
9db66bdc | 26 | spin_lock(lock); |
3ab5aee7 | 27 | if (hlist_nulls_unhashed(&tw->tw_node)) { |
9db66bdc | 28 | spin_unlock(lock); |
e48c414e ACM |
29 | return; |
30 | } | |
3ab5aee7 ED |
31 | hlist_nulls_del_rcu(&tw->tw_node); |
32 | sk_nulls_node_init(&tw->tw_node); | |
9db66bdc | 33 | spin_unlock(lock); |
e48c414e ACM |
34 | |
35 | /* Disassociate with bind bucket. */ | |
7f635ab7 PE |
36 | bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num, |
37 | hashinfo->bhash_size)]; | |
e48c414e ACM |
38 | spin_lock(&bhead->lock); |
39 | tb = tw->tw_tb; | |
40 | __hlist_del(&tw->tw_bind_node); | |
41 | tw->tw_tb = NULL; | |
42 | inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb); | |
43 | spin_unlock(&bhead->lock); | |
44 | #ifdef SOCK_REFCNT_DEBUG | |
45 | if (atomic_read(&tw->tw_refcnt) != 1) { | |
46 | printk(KERN_DEBUG "%s timewait_sock %p refcnt=%d\n", | |
47 | tw->tw_prot->name, tw, atomic_read(&tw->tw_refcnt)); | |
48 | } | |
49 | #endif | |
50 | inet_twsk_put(tw); | |
51 | } | |
52 | ||
7054fb93 PE |
53 | void inet_twsk_put(struct inet_timewait_sock *tw) |
54 | { | |
55 | if (atomic_dec_and_test(&tw->tw_refcnt)) { | |
56 | struct module *owner = tw->tw_prot->owner; | |
57 | twsk_destructor((struct sock *)tw); | |
58 | #ifdef SOCK_REFCNT_DEBUG | |
59 | printk(KERN_DEBUG "%s timewait_sock %p released\n", | |
60 | tw->tw_prot->name, tw); | |
61 | #endif | |
cd5342d9 | 62 | release_net(twsk_net(tw)); |
7054fb93 PE |
63 | kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw); |
64 | module_put(owner); | |
65 | } | |
66 | } | |
67 | EXPORT_SYMBOL_GPL(inet_twsk_put); | |
68 | ||
e48c414e ACM |
69 | /* |
70 | * Enter the time wait state. This is called with locally disabled BH. | |
71 | * Essentially we whip up a timewait bucket, copy the relevant info into it | |
72 | * from the SK, and mess with hash chains and list linkage. | |
73 | */ | |
74 | void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk, | |
75 | struct inet_hashinfo *hashinfo) | |
76 | { | |
77 | const struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 78 | const struct inet_connection_sock *icsk = inet_csk(sk); |
81c3d547 | 79 | struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash); |
9db66bdc | 80 | spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash); |
e48c414e ACM |
81 | struct inet_bind_hashbucket *bhead; |
82 | /* Step 1: Put TW into bind hash. Original socket stays there too. | |
83 | Note, that any socket with inet->num != 0 MUST be bound in | |
84 | binding cache, even if it is closed. | |
85 | */ | |
7f635ab7 PE |
86 | bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->num, |
87 | hashinfo->bhash_size)]; | |
e48c414e | 88 | spin_lock(&bhead->lock); |
463c84b9 | 89 | tw->tw_tb = icsk->icsk_bind_hash; |
547b792c | 90 | WARN_ON(!icsk->icsk_bind_hash); |
e48c414e ACM |
91 | inet_twsk_add_bind_node(tw, &tw->tw_tb->owners); |
92 | spin_unlock(&bhead->lock); | |
93 | ||
9db66bdc | 94 | spin_lock(lock); |
e48c414e | 95 | |
3ab5aee7 ED |
96 | /* |
97 | * Step 2: Hash TW into TIMEWAIT chain. | |
98 | * Should be done before removing sk from established chain | |
99 | * because readers are lockless and search established first. | |
100 | */ | |
e48c414e | 101 | atomic_inc(&tw->tw_refcnt); |
3ab5aee7 ED |
102 | inet_twsk_add_node_rcu(tw, &ehead->twchain); |
103 | ||
104 | /* Step 3: Remove SK from established hash. */ | |
105 | if (__sk_nulls_del_node_init_rcu(sk)) | |
106 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); | |
e48c414e | 107 | |
9db66bdc | 108 | spin_unlock(lock); |
e48c414e | 109 | } |
c676270b | 110 | |
696ab2d3 ACM |
111 | EXPORT_SYMBOL_GPL(__inet_twsk_hashdance); |
112 | ||
c676270b ACM |
113 | struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, const int state) |
114 | { | |
6d6ee43e ACM |
115 | struct inet_timewait_sock *tw = |
116 | kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab, | |
54e6ecb2 | 117 | GFP_ATOMIC); |
c676270b ACM |
118 | if (tw != NULL) { |
119 | const struct inet_sock *inet = inet_sk(sk); | |
120 | ||
9e337b0f VN |
121 | kmemcheck_annotate_bitfield(tw, flags); |
122 | ||
c676270b ACM |
123 | /* Give us an identity. */ |
124 | tw->tw_daddr = inet->daddr; | |
125 | tw->tw_rcv_saddr = inet->rcv_saddr; | |
126 | tw->tw_bound_dev_if = sk->sk_bound_dev_if; | |
127 | tw->tw_num = inet->num; | |
128 | tw->tw_state = TCP_TIME_WAIT; | |
129 | tw->tw_substate = state; | |
130 | tw->tw_sport = inet->sport; | |
131 | tw->tw_dport = inet->dport; | |
132 | tw->tw_family = sk->sk_family; | |
133 | tw->tw_reuse = sk->sk_reuse; | |
81c3d547 | 134 | tw->tw_hash = sk->sk_hash; |
c676270b | 135 | tw->tw_ipv6only = 0; |
f5715aea | 136 | tw->tw_transparent = inet->transparent; |
c676270b | 137 | tw->tw_prot = sk->sk_prot_creator; |
cd5342d9 | 138 | twsk_net_set(tw, hold_net(sock_net(sk))); |
c676270b ACM |
139 | atomic_set(&tw->tw_refcnt, 1); |
140 | inet_twsk_dead_node_init(tw); | |
eeb2b856 | 141 | __module_get(tw->tw_prot->owner); |
c676270b ACM |
142 | } |
143 | ||
144 | return tw; | |
145 | } | |
696ab2d3 ACM |
146 | |
147 | EXPORT_SYMBOL_GPL(inet_twsk_alloc); | |
148 | ||
149 | /* Returns non-zero if quota exceeded. */ | |
150 | static int inet_twdr_do_twkill_work(struct inet_timewait_death_row *twdr, | |
151 | const int slot) | |
152 | { | |
153 | struct inet_timewait_sock *tw; | |
154 | struct hlist_node *node; | |
155 | unsigned int killed; | |
156 | int ret; | |
157 | ||
158 | /* NOTE: compare this to previous version where lock | |
159 | * was released after detaching chain. It was racy, | |
160 | * because tw buckets are scheduled in not serialized context | |
161 | * in 2.3 (with netfilter), and with softnet it is common, because | |
162 | * soft irqs are not sequenced. | |
163 | */ | |
164 | killed = 0; | |
165 | ret = 0; | |
166 | rescan: | |
167 | inet_twsk_for_each_inmate(tw, node, &twdr->cells[slot]) { | |
168 | __inet_twsk_del_dead_node(tw); | |
169 | spin_unlock(&twdr->death_lock); | |
170 | __inet_twsk_kill(tw, twdr->hashinfo); | |
f2bf415c PE |
171 | #ifdef CONFIG_NET_NS |
172 | NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITED); | |
173 | #endif | |
696ab2d3 ACM |
174 | inet_twsk_put(tw); |
175 | killed++; | |
176 | spin_lock(&twdr->death_lock); | |
177 | if (killed > INET_TWDR_TWKILL_QUOTA) { | |
178 | ret = 1; | |
179 | break; | |
180 | } | |
181 | ||
182 | /* While we dropped twdr->death_lock, another cpu may have | |
183 | * killed off the next TW bucket in the list, therefore | |
184 | * do a fresh re-read of the hlist head node with the | |
185 | * lock reacquired. We still use the hlist traversal | |
186 | * macro in order to get the prefetches. | |
187 | */ | |
188 | goto rescan; | |
189 | } | |
190 | ||
191 | twdr->tw_count -= killed; | |
f2bf415c PE |
192 | #ifndef CONFIG_NET_NS |
193 | NET_ADD_STATS_BH(&init_net, LINUX_MIB_TIMEWAITED, killed); | |
194 | #endif | |
696ab2d3 ACM |
195 | return ret; |
196 | } | |
197 | ||
198 | void inet_twdr_hangman(unsigned long data) | |
199 | { | |
200 | struct inet_timewait_death_row *twdr; | |
201 | int unsigned need_timer; | |
202 | ||
203 | twdr = (struct inet_timewait_death_row *)data; | |
204 | spin_lock(&twdr->death_lock); | |
205 | ||
206 | if (twdr->tw_count == 0) | |
207 | goto out; | |
208 | ||
209 | need_timer = 0; | |
210 | if (inet_twdr_do_twkill_work(twdr, twdr->slot)) { | |
211 | twdr->thread_slots |= (1 << twdr->slot); | |
696ab2d3 ACM |
212 | schedule_work(&twdr->twkill_work); |
213 | need_timer = 1; | |
214 | } else { | |
215 | /* We purged the entire slot, anything left? */ | |
216 | if (twdr->tw_count) | |
217 | need_timer = 1; | |
218 | } | |
219 | twdr->slot = ((twdr->slot + 1) & (INET_TWDR_TWKILL_SLOTS - 1)); | |
220 | if (need_timer) | |
221 | mod_timer(&twdr->tw_timer, jiffies + twdr->period); | |
222 | out: | |
223 | spin_unlock(&twdr->death_lock); | |
224 | } | |
225 | ||
226 | EXPORT_SYMBOL_GPL(inet_twdr_hangman); | |
227 | ||
65f27f38 | 228 | void inet_twdr_twkill_work(struct work_struct *work) |
696ab2d3 | 229 | { |
65f27f38 DH |
230 | struct inet_timewait_death_row *twdr = |
231 | container_of(work, struct inet_timewait_death_row, twkill_work); | |
696ab2d3 ACM |
232 | int i; |
233 | ||
95c9382a PE |
234 | BUILD_BUG_ON((INET_TWDR_TWKILL_SLOTS - 1) > |
235 | (sizeof(twdr->thread_slots) * 8)); | |
696ab2d3 ACM |
236 | |
237 | while (twdr->thread_slots) { | |
238 | spin_lock_bh(&twdr->death_lock); | |
239 | for (i = 0; i < INET_TWDR_TWKILL_SLOTS; i++) { | |
240 | if (!(twdr->thread_slots & (1 << i))) | |
241 | continue; | |
242 | ||
243 | while (inet_twdr_do_twkill_work(twdr, i) != 0) { | |
244 | if (need_resched()) { | |
245 | spin_unlock_bh(&twdr->death_lock); | |
246 | schedule(); | |
247 | spin_lock_bh(&twdr->death_lock); | |
248 | } | |
249 | } | |
250 | ||
251 | twdr->thread_slots &= ~(1 << i); | |
252 | } | |
253 | spin_unlock_bh(&twdr->death_lock); | |
254 | } | |
255 | } | |
256 | ||
257 | EXPORT_SYMBOL_GPL(inet_twdr_twkill_work); | |
258 | ||
259 | /* These are always called from BH context. See callers in | |
260 | * tcp_input.c to verify this. | |
261 | */ | |
262 | ||
263 | /* This is for handling early-kills of TIME_WAIT sockets. */ | |
264 | void inet_twsk_deschedule(struct inet_timewait_sock *tw, | |
265 | struct inet_timewait_death_row *twdr) | |
266 | { | |
267 | spin_lock(&twdr->death_lock); | |
268 | if (inet_twsk_del_dead_node(tw)) { | |
269 | inet_twsk_put(tw); | |
270 | if (--twdr->tw_count == 0) | |
271 | del_timer(&twdr->tw_timer); | |
272 | } | |
273 | spin_unlock(&twdr->death_lock); | |
274 | __inet_twsk_kill(tw, twdr->hashinfo); | |
275 | } | |
276 | ||
277 | EXPORT_SYMBOL(inet_twsk_deschedule); | |
278 | ||
279 | void inet_twsk_schedule(struct inet_timewait_sock *tw, | |
280 | struct inet_timewait_death_row *twdr, | |
281 | const int timeo, const int timewait_len) | |
282 | { | |
283 | struct hlist_head *list; | |
284 | int slot; | |
285 | ||
286 | /* timeout := RTO * 3.5 | |
287 | * | |
288 | * 3.5 = 1+2+0.5 to wait for two retransmits. | |
289 | * | |
290 | * RATIONALE: if FIN arrived and we entered TIME-WAIT state, | |
291 | * our ACK acking that FIN can be lost. If N subsequent retransmitted | |
292 | * FINs (or previous seqments) are lost (probability of such event | |
293 | * is p^(N+1), where p is probability to lose single packet and | |
294 | * time to detect the loss is about RTO*(2^N - 1) with exponential | |
295 | * backoff). Normal timewait length is calculated so, that we | |
296 | * waited at least for one retransmitted FIN (maximal RTO is 120sec). | |
297 | * [ BTW Linux. following BSD, violates this requirement waiting | |
298 | * only for 60sec, we should wait at least for 240 secs. | |
299 | * Well, 240 consumes too much of resources 8) | |
300 | * ] | |
301 | * This interval is not reduced to catch old duplicate and | |
302 | * responces to our wandering segments living for two MSLs. | |
303 | * However, if we use PAWS to detect | |
304 | * old duplicates, we can reduce the interval to bounds required | |
305 | * by RTO, rather than MSL. So, if peer understands PAWS, we | |
306 | * kill tw bucket after 3.5*RTO (it is important that this number | |
307 | * is greater than TS tick!) and detect old duplicates with help | |
308 | * of PAWS. | |
309 | */ | |
310 | slot = (timeo + (1 << INET_TWDR_RECYCLE_TICK) - 1) >> INET_TWDR_RECYCLE_TICK; | |
311 | ||
312 | spin_lock(&twdr->death_lock); | |
313 | ||
314 | /* Unlink it, if it was scheduled */ | |
315 | if (inet_twsk_del_dead_node(tw)) | |
316 | twdr->tw_count--; | |
317 | else | |
318 | atomic_inc(&tw->tw_refcnt); | |
319 | ||
320 | if (slot >= INET_TWDR_RECYCLE_SLOTS) { | |
321 | /* Schedule to slow timer */ | |
322 | if (timeo >= timewait_len) { | |
323 | slot = INET_TWDR_TWKILL_SLOTS - 1; | |
324 | } else { | |
172589cc | 325 | slot = DIV_ROUND_UP(timeo, twdr->period); |
696ab2d3 ACM |
326 | if (slot >= INET_TWDR_TWKILL_SLOTS) |
327 | slot = INET_TWDR_TWKILL_SLOTS - 1; | |
328 | } | |
329 | tw->tw_ttd = jiffies + timeo; | |
330 | slot = (twdr->slot + slot) & (INET_TWDR_TWKILL_SLOTS - 1); | |
331 | list = &twdr->cells[slot]; | |
332 | } else { | |
333 | tw->tw_ttd = jiffies + (slot << INET_TWDR_RECYCLE_TICK); | |
334 | ||
335 | if (twdr->twcal_hand < 0) { | |
336 | twdr->twcal_hand = 0; | |
337 | twdr->twcal_jiffie = jiffies; | |
338 | twdr->twcal_timer.expires = twdr->twcal_jiffie + | |
339 | (slot << INET_TWDR_RECYCLE_TICK); | |
340 | add_timer(&twdr->twcal_timer); | |
341 | } else { | |
342 | if (time_after(twdr->twcal_timer.expires, | |
343 | jiffies + (slot << INET_TWDR_RECYCLE_TICK))) | |
344 | mod_timer(&twdr->twcal_timer, | |
345 | jiffies + (slot << INET_TWDR_RECYCLE_TICK)); | |
346 | slot = (twdr->twcal_hand + slot) & (INET_TWDR_RECYCLE_SLOTS - 1); | |
347 | } | |
348 | list = &twdr->twcal_row[slot]; | |
349 | } | |
350 | ||
351 | hlist_add_head(&tw->tw_death_node, list); | |
352 | ||
353 | if (twdr->tw_count++ == 0) | |
354 | mod_timer(&twdr->tw_timer, jiffies + twdr->period); | |
355 | spin_unlock(&twdr->death_lock); | |
356 | } | |
357 | ||
358 | EXPORT_SYMBOL_GPL(inet_twsk_schedule); | |
359 | ||
360 | void inet_twdr_twcal_tick(unsigned long data) | |
361 | { | |
362 | struct inet_timewait_death_row *twdr; | |
363 | int n, slot; | |
364 | unsigned long j; | |
365 | unsigned long now = jiffies; | |
366 | int killed = 0; | |
367 | int adv = 0; | |
368 | ||
369 | twdr = (struct inet_timewait_death_row *)data; | |
370 | ||
371 | spin_lock(&twdr->death_lock); | |
372 | if (twdr->twcal_hand < 0) | |
373 | goto out; | |
374 | ||
375 | slot = twdr->twcal_hand; | |
376 | j = twdr->twcal_jiffie; | |
377 | ||
378 | for (n = 0; n < INET_TWDR_RECYCLE_SLOTS; n++) { | |
379 | if (time_before_eq(j, now)) { | |
380 | struct hlist_node *node, *safe; | |
381 | struct inet_timewait_sock *tw; | |
382 | ||
383 | inet_twsk_for_each_inmate_safe(tw, node, safe, | |
384 | &twdr->twcal_row[slot]) { | |
385 | __inet_twsk_del_dead_node(tw); | |
386 | __inet_twsk_kill(tw, twdr->hashinfo); | |
f2bf415c PE |
387 | #ifdef CONFIG_NET_NS |
388 | NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED); | |
389 | #endif | |
696ab2d3 ACM |
390 | inet_twsk_put(tw); |
391 | killed++; | |
392 | } | |
393 | } else { | |
394 | if (!adv) { | |
395 | adv = 1; | |
396 | twdr->twcal_jiffie = j; | |
397 | twdr->twcal_hand = slot; | |
398 | } | |
399 | ||
400 | if (!hlist_empty(&twdr->twcal_row[slot])) { | |
401 | mod_timer(&twdr->twcal_timer, j); | |
402 | goto out; | |
403 | } | |
404 | } | |
405 | j += 1 << INET_TWDR_RECYCLE_TICK; | |
406 | slot = (slot + 1) & (INET_TWDR_RECYCLE_SLOTS - 1); | |
407 | } | |
408 | twdr->twcal_hand = -1; | |
409 | ||
410 | out: | |
411 | if ((twdr->tw_count -= killed) == 0) | |
412 | del_timer(&twdr->tw_timer); | |
f2bf415c PE |
413 | #ifndef CONFIG_NET_NS |
414 | NET_ADD_STATS_BH(&init_net, LINUX_MIB_TIMEWAITKILLED, killed); | |
415 | #endif | |
696ab2d3 ACM |
416 | spin_unlock(&twdr->death_lock); |
417 | } | |
418 | ||
419 | EXPORT_SYMBOL_GPL(inet_twdr_twcal_tick); | |
d315492b DL |
420 | |
421 | void inet_twsk_purge(struct net *net, struct inet_hashinfo *hashinfo, | |
422 | struct inet_timewait_death_row *twdr, int family) | |
423 | { | |
424 | struct inet_timewait_sock *tw; | |
425 | struct sock *sk; | |
3ab5aee7 | 426 | struct hlist_nulls_node *node; |
d315492b DL |
427 | int h; |
428 | ||
429 | local_bh_disable(); | |
430 | for (h = 0; h < (hashinfo->ehash_size); h++) { | |
431 | struct inet_ehash_bucket *head = | |
432 | inet_ehash_bucket(hashinfo, h); | |
9db66bdc | 433 | spinlock_t *lock = inet_ehash_lockp(hashinfo, h); |
d315492b | 434 | restart: |
9db66bdc | 435 | spin_lock(lock); |
3ab5aee7 | 436 | sk_nulls_for_each(sk, node, &head->twchain) { |
d315492b DL |
437 | |
438 | tw = inet_twsk(sk); | |
439 | if (!net_eq(twsk_net(tw), net) || | |
440 | tw->tw_family != family) | |
441 | continue; | |
442 | ||
443 | atomic_inc(&tw->tw_refcnt); | |
9db66bdc | 444 | spin_unlock(lock); |
d315492b DL |
445 | inet_twsk_deschedule(tw, twdr); |
446 | inet_twsk_put(tw); | |
447 | ||
448 | goto restart; | |
449 | } | |
9db66bdc | 450 | spin_unlock(lock); |
d315492b DL |
451 | } |
452 | local_bh_enable(); | |
453 | } | |
454 | EXPORT_SYMBOL_GPL(inet_twsk_purge); |