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> |
5a0e3ad6 | 13 | #include <linux/slab.h> |
e48c414e ACM |
14 | #include <net/inet_hashtables.h> |
15 | #include <net/inet_timewait_sock.h> | |
696ab2d3 | 16 | #include <net/ip.h> |
e48c414e | 17 | |
13475a30 | 18 | |
2a8875e7 ED |
19 | /** |
20 | * inet_twsk_unhash - unhash a timewait socket from established hash | |
21 | * @tw: timewait socket | |
22 | * | |
23 | * unhash a timewait socket from established hash, if hashed. | |
24 | * ehash lock must be held by caller. | |
25 | * Returns 1 if caller should call inet_twsk_put() after lock release. | |
13475a30 ED |
26 | */ |
27 | int inet_twsk_unhash(struct inet_timewait_sock *tw) | |
28 | { | |
29 | if (hlist_nulls_unhashed(&tw->tw_node)) | |
30 | return 0; | |
31 | ||
32 | hlist_nulls_del_rcu(&tw->tw_node); | |
33 | sk_nulls_node_init(&tw->tw_node); | |
2a8875e7 ED |
34 | /* |
35 | * We cannot call inet_twsk_put() ourself under lock, | |
36 | * caller must call it for us. | |
37 | */ | |
13475a30 ED |
38 | return 1; |
39 | } | |
40 | ||
2a8875e7 ED |
41 | /** |
42 | * inet_twsk_bind_unhash - unhash a timewait socket from bind hash | |
43 | * @tw: timewait socket | |
44 | * @hashinfo: hashinfo pointer | |
45 | * | |
46 | * unhash a timewait socket from bind hash, if hashed. | |
47 | * bind hash lock must be held by caller. | |
48 | * Returns 1 if caller should call inet_twsk_put() after lock release. | |
3cdaedae ED |
49 | */ |
50 | int inet_twsk_bind_unhash(struct inet_timewait_sock *tw, | |
51 | struct inet_hashinfo *hashinfo) | |
52 | { | |
53 | struct inet_bind_bucket *tb = tw->tw_tb; | |
54 | ||
55 | if (!tb) | |
56 | return 0; | |
57 | ||
58 | __hlist_del(&tw->tw_bind_node); | |
59 | tw->tw_tb = NULL; | |
60 | inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb); | |
2a8875e7 ED |
61 | /* |
62 | * We cannot call inet_twsk_put() ourself under lock, | |
63 | * caller must call it for us. | |
64 | */ | |
3cdaedae ED |
65 | return 1; |
66 | } | |
67 | ||
e48c414e | 68 | /* Must be called with locally disabled BHs. */ |
acd159b6 AB |
69 | static void __inet_twsk_kill(struct inet_timewait_sock *tw, |
70 | struct inet_hashinfo *hashinfo) | |
e48c414e ACM |
71 | { |
72 | struct inet_bind_hashbucket *bhead; | |
13475a30 | 73 | int refcnt; |
e48c414e | 74 | /* Unlink from established hashes. */ |
9db66bdc | 75 | spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash); |
e48c414e | 76 | |
9db66bdc | 77 | spin_lock(lock); |
13475a30 | 78 | refcnt = inet_twsk_unhash(tw); |
9db66bdc | 79 | spin_unlock(lock); |
e48c414e ACM |
80 | |
81 | /* Disassociate with bind bucket. */ | |
7f635ab7 PE |
82 | bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num, |
83 | hashinfo->bhash_size)]; | |
3cdaedae | 84 | |
e48c414e | 85 | spin_lock(&bhead->lock); |
3cdaedae | 86 | refcnt += inet_twsk_bind_unhash(tw, hashinfo); |
e48c414e | 87 | spin_unlock(&bhead->lock); |
3cdaedae | 88 | |
e48c414e ACM |
89 | #ifdef SOCK_REFCNT_DEBUG |
90 | if (atomic_read(&tw->tw_refcnt) != 1) { | |
91 | printk(KERN_DEBUG "%s timewait_sock %p refcnt=%d\n", | |
92 | tw->tw_prot->name, tw, atomic_read(&tw->tw_refcnt)); | |
93 | } | |
94 | #endif | |
13475a30 ED |
95 | while (refcnt) { |
96 | inet_twsk_put(tw); | |
97 | refcnt--; | |
98 | } | |
e48c414e ACM |
99 | } |
100 | ||
4dbc8ef7 | 101 | static noinline void inet_twsk_free(struct inet_timewait_sock *tw) |
7054fb93 | 102 | { |
4dbc8ef7 ACM |
103 | struct module *owner = tw->tw_prot->owner; |
104 | twsk_destructor((struct sock *)tw); | |
7054fb93 | 105 | #ifdef SOCK_REFCNT_DEBUG |
4dbc8ef7 | 106 | pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw); |
7054fb93 | 107 | #endif |
4dbc8ef7 ACM |
108 | release_net(twsk_net(tw)); |
109 | kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw); | |
110 | module_put(owner); | |
111 | } | |
112 | ||
113 | void inet_twsk_put(struct inet_timewait_sock *tw) | |
114 | { | |
115 | if (atomic_dec_and_test(&tw->tw_refcnt)) | |
116 | inet_twsk_free(tw); | |
7054fb93 PE |
117 | } |
118 | EXPORT_SYMBOL_GPL(inet_twsk_put); | |
119 | ||
e48c414e ACM |
120 | /* |
121 | * Enter the time wait state. This is called with locally disabled BH. | |
122 | * Essentially we whip up a timewait bucket, copy the relevant info into it | |
123 | * from the SK, and mess with hash chains and list linkage. | |
124 | */ | |
125 | void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk, | |
126 | struct inet_hashinfo *hashinfo) | |
127 | { | |
128 | const struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 129 | const struct inet_connection_sock *icsk = inet_csk(sk); |
81c3d547 | 130 | struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash); |
9db66bdc | 131 | spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash); |
e48c414e ACM |
132 | struct inet_bind_hashbucket *bhead; |
133 | /* Step 1: Put TW into bind hash. Original socket stays there too. | |
134 | Note, that any socket with inet->num != 0 MUST be bound in | |
135 | binding cache, even if it is closed. | |
136 | */ | |
c720c7e8 | 137 | bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num, |
7f635ab7 | 138 | hashinfo->bhash_size)]; |
e48c414e | 139 | spin_lock(&bhead->lock); |
463c84b9 | 140 | tw->tw_tb = icsk->icsk_bind_hash; |
547b792c | 141 | WARN_ON(!icsk->icsk_bind_hash); |
e48c414e ACM |
142 | inet_twsk_add_bind_node(tw, &tw->tw_tb->owners); |
143 | spin_unlock(&bhead->lock); | |
144 | ||
9db66bdc | 145 | spin_lock(lock); |
e48c414e | 146 | |
3ab5aee7 ED |
147 | /* |
148 | * Step 2: Hash TW into TIMEWAIT chain. | |
149 | * Should be done before removing sk from established chain | |
150 | * because readers are lockless and search established first. | |
151 | */ | |
3ab5aee7 ED |
152 | inet_twsk_add_node_rcu(tw, &ehead->twchain); |
153 | ||
154 | /* Step 3: Remove SK from established hash. */ | |
155 | if (__sk_nulls_del_node_init_rcu(sk)) | |
156 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); | |
e48c414e | 157 | |
47e1c323 ED |
158 | /* |
159 | * Notes : | |
2a8875e7 | 160 | * - We initially set tw_refcnt to 0 in inet_twsk_alloc() |
47e1c323 ED |
161 | * - We add one reference for the bhash link |
162 | * - We add one reference for the ehash link | |
163 | * - We want this refcnt update done before allowing other | |
164 | * threads to find this tw in ehash chain. | |
165 | */ | |
166 | atomic_add(1 + 1 + 1, &tw->tw_refcnt); | |
167 | ||
9db66bdc | 168 | spin_unlock(lock); |
e48c414e | 169 | } |
696ab2d3 ACM |
170 | EXPORT_SYMBOL_GPL(__inet_twsk_hashdance); |
171 | ||
c676270b ACM |
172 | struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, const int state) |
173 | { | |
6d6ee43e ACM |
174 | struct inet_timewait_sock *tw = |
175 | kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab, | |
54e6ecb2 | 176 | GFP_ATOMIC); |
c676270b ACM |
177 | if (tw != NULL) { |
178 | const struct inet_sock *inet = inet_sk(sk); | |
179 | ||
9e337b0f VN |
180 | kmemcheck_annotate_bitfield(tw, flags); |
181 | ||
c676270b | 182 | /* Give us an identity. */ |
c720c7e8 ED |
183 | tw->tw_daddr = inet->inet_daddr; |
184 | tw->tw_rcv_saddr = inet->inet_rcv_saddr; | |
c676270b | 185 | tw->tw_bound_dev_if = sk->sk_bound_dev_if; |
c720c7e8 | 186 | tw->tw_num = inet->inet_num; |
c676270b ACM |
187 | tw->tw_state = TCP_TIME_WAIT; |
188 | tw->tw_substate = state; | |
c720c7e8 ED |
189 | tw->tw_sport = inet->inet_sport; |
190 | tw->tw_dport = inet->inet_dport; | |
c676270b ACM |
191 | tw->tw_family = sk->sk_family; |
192 | tw->tw_reuse = sk->sk_reuse; | |
81c3d547 | 193 | tw->tw_hash = sk->sk_hash; |
c676270b | 194 | tw->tw_ipv6only = 0; |
f5715aea | 195 | tw->tw_transparent = inet->transparent; |
c676270b | 196 | tw->tw_prot = sk->sk_prot_creator; |
cd5342d9 | 197 | twsk_net_set(tw, hold_net(sock_net(sk))); |
47e1c323 ED |
198 | /* |
199 | * Because we use RCU lookups, we should not set tw_refcnt | |
200 | * to a non null value before everything is setup for this | |
201 | * timewait socket. | |
202 | */ | |
203 | atomic_set(&tw->tw_refcnt, 0); | |
c676270b | 204 | inet_twsk_dead_node_init(tw); |
eeb2b856 | 205 | __module_get(tw->tw_prot->owner); |
c676270b ACM |
206 | } |
207 | ||
208 | return tw; | |
209 | } | |
696ab2d3 ACM |
210 | EXPORT_SYMBOL_GPL(inet_twsk_alloc); |
211 | ||
212 | /* Returns non-zero if quota exceeded. */ | |
213 | static int inet_twdr_do_twkill_work(struct inet_timewait_death_row *twdr, | |
214 | const int slot) | |
215 | { | |
216 | struct inet_timewait_sock *tw; | |
217 | struct hlist_node *node; | |
218 | unsigned int killed; | |
219 | int ret; | |
220 | ||
221 | /* NOTE: compare this to previous version where lock | |
222 | * was released after detaching chain. It was racy, | |
223 | * because tw buckets are scheduled in not serialized context | |
224 | * in 2.3 (with netfilter), and with softnet it is common, because | |
225 | * soft irqs are not sequenced. | |
226 | */ | |
227 | killed = 0; | |
228 | ret = 0; | |
229 | rescan: | |
230 | inet_twsk_for_each_inmate(tw, node, &twdr->cells[slot]) { | |
231 | __inet_twsk_del_dead_node(tw); | |
232 | spin_unlock(&twdr->death_lock); | |
233 | __inet_twsk_kill(tw, twdr->hashinfo); | |
f2bf415c PE |
234 | #ifdef CONFIG_NET_NS |
235 | NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITED); | |
236 | #endif | |
696ab2d3 ACM |
237 | inet_twsk_put(tw); |
238 | killed++; | |
239 | spin_lock(&twdr->death_lock); | |
240 | if (killed > INET_TWDR_TWKILL_QUOTA) { | |
241 | ret = 1; | |
242 | break; | |
243 | } | |
244 | ||
245 | /* While we dropped twdr->death_lock, another cpu may have | |
246 | * killed off the next TW bucket in the list, therefore | |
247 | * do a fresh re-read of the hlist head node with the | |
248 | * lock reacquired. We still use the hlist traversal | |
249 | * macro in order to get the prefetches. | |
250 | */ | |
251 | goto rescan; | |
252 | } | |
253 | ||
254 | twdr->tw_count -= killed; | |
f2bf415c PE |
255 | #ifndef CONFIG_NET_NS |
256 | NET_ADD_STATS_BH(&init_net, LINUX_MIB_TIMEWAITED, killed); | |
257 | #endif | |
696ab2d3 ACM |
258 | return ret; |
259 | } | |
260 | ||
261 | void inet_twdr_hangman(unsigned long data) | |
262 | { | |
263 | struct inet_timewait_death_row *twdr; | |
264 | int unsigned need_timer; | |
265 | ||
266 | twdr = (struct inet_timewait_death_row *)data; | |
267 | spin_lock(&twdr->death_lock); | |
268 | ||
269 | if (twdr->tw_count == 0) | |
270 | goto out; | |
271 | ||
272 | need_timer = 0; | |
273 | if (inet_twdr_do_twkill_work(twdr, twdr->slot)) { | |
274 | twdr->thread_slots |= (1 << twdr->slot); | |
696ab2d3 ACM |
275 | schedule_work(&twdr->twkill_work); |
276 | need_timer = 1; | |
277 | } else { | |
278 | /* We purged the entire slot, anything left? */ | |
279 | if (twdr->tw_count) | |
280 | need_timer = 1; | |
80a1096b | 281 | twdr->slot = ((twdr->slot + 1) & (INET_TWDR_TWKILL_SLOTS - 1)); |
696ab2d3 | 282 | } |
696ab2d3 ACM |
283 | if (need_timer) |
284 | mod_timer(&twdr->tw_timer, jiffies + twdr->period); | |
285 | out: | |
286 | spin_unlock(&twdr->death_lock); | |
287 | } | |
696ab2d3 ACM |
288 | EXPORT_SYMBOL_GPL(inet_twdr_hangman); |
289 | ||
65f27f38 | 290 | void inet_twdr_twkill_work(struct work_struct *work) |
696ab2d3 | 291 | { |
65f27f38 DH |
292 | struct inet_timewait_death_row *twdr = |
293 | container_of(work, struct inet_timewait_death_row, twkill_work); | |
696ab2d3 ACM |
294 | int i; |
295 | ||
95c9382a PE |
296 | BUILD_BUG_ON((INET_TWDR_TWKILL_SLOTS - 1) > |
297 | (sizeof(twdr->thread_slots) * 8)); | |
696ab2d3 ACM |
298 | |
299 | while (twdr->thread_slots) { | |
300 | spin_lock_bh(&twdr->death_lock); | |
301 | for (i = 0; i < INET_TWDR_TWKILL_SLOTS; i++) { | |
302 | if (!(twdr->thread_slots & (1 << i))) | |
303 | continue; | |
304 | ||
305 | while (inet_twdr_do_twkill_work(twdr, i) != 0) { | |
306 | if (need_resched()) { | |
307 | spin_unlock_bh(&twdr->death_lock); | |
308 | schedule(); | |
309 | spin_lock_bh(&twdr->death_lock); | |
310 | } | |
311 | } | |
312 | ||
313 | twdr->thread_slots &= ~(1 << i); | |
314 | } | |
315 | spin_unlock_bh(&twdr->death_lock); | |
316 | } | |
317 | } | |
696ab2d3 ACM |
318 | EXPORT_SYMBOL_GPL(inet_twdr_twkill_work); |
319 | ||
320 | /* These are always called from BH context. See callers in | |
321 | * tcp_input.c to verify this. | |
322 | */ | |
323 | ||
324 | /* This is for handling early-kills of TIME_WAIT sockets. */ | |
325 | void inet_twsk_deschedule(struct inet_timewait_sock *tw, | |
326 | struct inet_timewait_death_row *twdr) | |
327 | { | |
328 | spin_lock(&twdr->death_lock); | |
329 | if (inet_twsk_del_dead_node(tw)) { | |
330 | inet_twsk_put(tw); | |
331 | if (--twdr->tw_count == 0) | |
332 | del_timer(&twdr->tw_timer); | |
333 | } | |
334 | spin_unlock(&twdr->death_lock); | |
335 | __inet_twsk_kill(tw, twdr->hashinfo); | |
336 | } | |
696ab2d3 ACM |
337 | EXPORT_SYMBOL(inet_twsk_deschedule); |
338 | ||
339 | void inet_twsk_schedule(struct inet_timewait_sock *tw, | |
340 | struct inet_timewait_death_row *twdr, | |
341 | const int timeo, const int timewait_len) | |
342 | { | |
343 | struct hlist_head *list; | |
344 | int slot; | |
345 | ||
346 | /* timeout := RTO * 3.5 | |
347 | * | |
348 | * 3.5 = 1+2+0.5 to wait for two retransmits. | |
349 | * | |
350 | * RATIONALE: if FIN arrived and we entered TIME-WAIT state, | |
351 | * our ACK acking that FIN can be lost. If N subsequent retransmitted | |
352 | * FINs (or previous seqments) are lost (probability of such event | |
353 | * is p^(N+1), where p is probability to lose single packet and | |
354 | * time to detect the loss is about RTO*(2^N - 1) with exponential | |
355 | * backoff). Normal timewait length is calculated so, that we | |
356 | * waited at least for one retransmitted FIN (maximal RTO is 120sec). | |
357 | * [ BTW Linux. following BSD, violates this requirement waiting | |
358 | * only for 60sec, we should wait at least for 240 secs. | |
359 | * Well, 240 consumes too much of resources 8) | |
360 | * ] | |
361 | * This interval is not reduced to catch old duplicate and | |
362 | * responces to our wandering segments living for two MSLs. | |
363 | * However, if we use PAWS to detect | |
364 | * old duplicates, we can reduce the interval to bounds required | |
365 | * by RTO, rather than MSL. So, if peer understands PAWS, we | |
366 | * kill tw bucket after 3.5*RTO (it is important that this number | |
367 | * is greater than TS tick!) and detect old duplicates with help | |
368 | * of PAWS. | |
369 | */ | |
370 | slot = (timeo + (1 << INET_TWDR_RECYCLE_TICK) - 1) >> INET_TWDR_RECYCLE_TICK; | |
371 | ||
372 | spin_lock(&twdr->death_lock); | |
373 | ||
374 | /* Unlink it, if it was scheduled */ | |
375 | if (inet_twsk_del_dead_node(tw)) | |
376 | twdr->tw_count--; | |
377 | else | |
378 | atomic_inc(&tw->tw_refcnt); | |
379 | ||
380 | if (slot >= INET_TWDR_RECYCLE_SLOTS) { | |
381 | /* Schedule to slow timer */ | |
382 | if (timeo >= timewait_len) { | |
383 | slot = INET_TWDR_TWKILL_SLOTS - 1; | |
384 | } else { | |
172589cc | 385 | slot = DIV_ROUND_UP(timeo, twdr->period); |
696ab2d3 ACM |
386 | if (slot >= INET_TWDR_TWKILL_SLOTS) |
387 | slot = INET_TWDR_TWKILL_SLOTS - 1; | |
388 | } | |
389 | tw->tw_ttd = jiffies + timeo; | |
390 | slot = (twdr->slot + slot) & (INET_TWDR_TWKILL_SLOTS - 1); | |
391 | list = &twdr->cells[slot]; | |
392 | } else { | |
393 | tw->tw_ttd = jiffies + (slot << INET_TWDR_RECYCLE_TICK); | |
394 | ||
395 | if (twdr->twcal_hand < 0) { | |
396 | twdr->twcal_hand = 0; | |
397 | twdr->twcal_jiffie = jiffies; | |
398 | twdr->twcal_timer.expires = twdr->twcal_jiffie + | |
399 | (slot << INET_TWDR_RECYCLE_TICK); | |
400 | add_timer(&twdr->twcal_timer); | |
401 | } else { | |
402 | if (time_after(twdr->twcal_timer.expires, | |
403 | jiffies + (slot << INET_TWDR_RECYCLE_TICK))) | |
404 | mod_timer(&twdr->twcal_timer, | |
405 | jiffies + (slot << INET_TWDR_RECYCLE_TICK)); | |
406 | slot = (twdr->twcal_hand + slot) & (INET_TWDR_RECYCLE_SLOTS - 1); | |
407 | } | |
408 | list = &twdr->twcal_row[slot]; | |
409 | } | |
410 | ||
411 | hlist_add_head(&tw->tw_death_node, list); | |
412 | ||
413 | if (twdr->tw_count++ == 0) | |
414 | mod_timer(&twdr->tw_timer, jiffies + twdr->period); | |
415 | spin_unlock(&twdr->death_lock); | |
416 | } | |
696ab2d3 ACM |
417 | EXPORT_SYMBOL_GPL(inet_twsk_schedule); |
418 | ||
419 | void inet_twdr_twcal_tick(unsigned long data) | |
420 | { | |
421 | struct inet_timewait_death_row *twdr; | |
422 | int n, slot; | |
423 | unsigned long j; | |
424 | unsigned long now = jiffies; | |
425 | int killed = 0; | |
426 | int adv = 0; | |
427 | ||
428 | twdr = (struct inet_timewait_death_row *)data; | |
429 | ||
430 | spin_lock(&twdr->death_lock); | |
431 | if (twdr->twcal_hand < 0) | |
432 | goto out; | |
433 | ||
434 | slot = twdr->twcal_hand; | |
435 | j = twdr->twcal_jiffie; | |
436 | ||
437 | for (n = 0; n < INET_TWDR_RECYCLE_SLOTS; n++) { | |
438 | if (time_before_eq(j, now)) { | |
439 | struct hlist_node *node, *safe; | |
440 | struct inet_timewait_sock *tw; | |
441 | ||
442 | inet_twsk_for_each_inmate_safe(tw, node, safe, | |
443 | &twdr->twcal_row[slot]) { | |
444 | __inet_twsk_del_dead_node(tw); | |
445 | __inet_twsk_kill(tw, twdr->hashinfo); | |
f2bf415c PE |
446 | #ifdef CONFIG_NET_NS |
447 | NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED); | |
448 | #endif | |
696ab2d3 ACM |
449 | inet_twsk_put(tw); |
450 | killed++; | |
451 | } | |
452 | } else { | |
453 | if (!adv) { | |
454 | adv = 1; | |
455 | twdr->twcal_jiffie = j; | |
456 | twdr->twcal_hand = slot; | |
457 | } | |
458 | ||
459 | if (!hlist_empty(&twdr->twcal_row[slot])) { | |
460 | mod_timer(&twdr->twcal_timer, j); | |
461 | goto out; | |
462 | } | |
463 | } | |
464 | j += 1 << INET_TWDR_RECYCLE_TICK; | |
465 | slot = (slot + 1) & (INET_TWDR_RECYCLE_SLOTS - 1); | |
466 | } | |
467 | twdr->twcal_hand = -1; | |
468 | ||
469 | out: | |
470 | if ((twdr->tw_count -= killed) == 0) | |
471 | del_timer(&twdr->tw_timer); | |
f2bf415c PE |
472 | #ifndef CONFIG_NET_NS |
473 | NET_ADD_STATS_BH(&init_net, LINUX_MIB_TIMEWAITKILLED, killed); | |
474 | #endif | |
696ab2d3 ACM |
475 | spin_unlock(&twdr->death_lock); |
476 | } | |
696ab2d3 | 477 | EXPORT_SYMBOL_GPL(inet_twdr_twcal_tick); |
d315492b | 478 | |
b099ce26 | 479 | void inet_twsk_purge(struct inet_hashinfo *hashinfo, |
d315492b DL |
480 | struct inet_timewait_death_row *twdr, int family) |
481 | { | |
482 | struct inet_timewait_sock *tw; | |
483 | struct sock *sk; | |
3ab5aee7 | 484 | struct hlist_nulls_node *node; |
575f4cd5 | 485 | unsigned int slot; |
d315492b | 486 | |
575f4cd5 EB |
487 | for (slot = 0; slot <= hashinfo->ehash_mask; slot++) { |
488 | struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; | |
489 | restart_rcu: | |
490 | rcu_read_lock(); | |
d315492b | 491 | restart: |
575f4cd5 | 492 | sk_nulls_for_each_rcu(sk, node, &head->twchain) { |
d315492b | 493 | tw = inet_twsk(sk); |
b099ce26 EB |
494 | if ((tw->tw_family != family) || |
495 | atomic_read(&twsk_net(tw)->count)) | |
d315492b DL |
496 | continue; |
497 | ||
575f4cd5 EB |
498 | if (unlikely(!atomic_inc_not_zero(&tw->tw_refcnt))) |
499 | continue; | |
500 | ||
b099ce26 EB |
501 | if (unlikely((tw->tw_family != family) || |
502 | atomic_read(&twsk_net(tw)->count))) { | |
575f4cd5 EB |
503 | inet_twsk_put(tw); |
504 | goto restart; | |
505 | } | |
506 | ||
507 | rcu_read_unlock(); | |
d315492b DL |
508 | inet_twsk_deschedule(tw, twdr); |
509 | inet_twsk_put(tw); | |
575f4cd5 | 510 | goto restart_rcu; |
d315492b | 511 | } |
575f4cd5 EB |
512 | /* If the nulls value we got at the end of this lookup is |
513 | * not the expected one, we must restart lookup. | |
514 | * We probably met an item that was moved to another chain. | |
515 | */ | |
516 | if (get_nulls_value(node) != slot) | |
517 | goto restart; | |
518 | rcu_read_unlock(); | |
d315492b | 519 | } |
d315492b DL |
520 | } |
521 | EXPORT_SYMBOL_GPL(inet_twsk_purge); |