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> |
5a0e3ad6 | 12 | #include <linux/slab.h> |
3a9a231d | 13 | #include <linux/module.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_bind_unhash - unhash a timewait socket from bind hash | |
21 | * @tw: timewait socket | |
22 | * @hashinfo: hashinfo pointer | |
23 | * | |
24 | * unhash a timewait socket from bind hash, if hashed. | |
25 | * bind hash lock must be held by caller. | |
26 | * Returns 1 if caller should call inet_twsk_put() after lock release. | |
3cdaedae | 27 | */ |
fc01538f | 28 | void inet_twsk_bind_unhash(struct inet_timewait_sock *tw, |
3cdaedae ED |
29 | struct inet_hashinfo *hashinfo) |
30 | { | |
31 | struct inet_bind_bucket *tb = tw->tw_tb; | |
32 | ||
33 | if (!tb) | |
fc01538f | 34 | return; |
3cdaedae ED |
35 | |
36 | __hlist_del(&tw->tw_bind_node); | |
37 | tw->tw_tb = NULL; | |
38 | inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb); | |
fc01538f | 39 | __sock_put((struct sock *)tw); |
3cdaedae ED |
40 | } |
41 | ||
e48c414e | 42 | /* Must be called with locally disabled BHs. */ |
789f558c | 43 | static void inet_twsk_kill(struct inet_timewait_sock *tw) |
e48c414e | 44 | { |
789f558c | 45 | struct inet_hashinfo *hashinfo = tw->tw_dr->hashinfo; |
9db66bdc | 46 | spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash); |
fc01538f | 47 | struct inet_bind_hashbucket *bhead; |
e48c414e | 48 | |
9db66bdc | 49 | spin_lock(lock); |
fc01538f | 50 | sk_nulls_del_node_init_rcu((struct sock *)tw); |
9db66bdc | 51 | spin_unlock(lock); |
e48c414e ACM |
52 | |
53 | /* Disassociate with bind bucket. */ | |
7f635ab7 PE |
54 | bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num, |
55 | hashinfo->bhash_size)]; | |
3cdaedae | 56 | |
e48c414e | 57 | spin_lock(&bhead->lock); |
fc01538f | 58 | inet_twsk_bind_unhash(tw, hashinfo); |
e48c414e | 59 | spin_unlock(&bhead->lock); |
3cdaedae | 60 | |
789f558c ED |
61 | atomic_dec(&tw->tw_dr->tw_count); |
62 | inet_twsk_put(tw); | |
e48c414e ACM |
63 | } |
64 | ||
05dbc7b5 | 65 | void inet_twsk_free(struct inet_timewait_sock *tw) |
7054fb93 | 66 | { |
4dbc8ef7 ACM |
67 | struct module *owner = tw->tw_prot->owner; |
68 | twsk_destructor((struct sock *)tw); | |
7054fb93 | 69 | #ifdef SOCK_REFCNT_DEBUG |
4dbc8ef7 | 70 | pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw); |
7054fb93 | 71 | #endif |
4dbc8ef7 ACM |
72 | kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw); |
73 | module_put(owner); | |
74 | } | |
75 | ||
76 | void inet_twsk_put(struct inet_timewait_sock *tw) | |
77 | { | |
41c6d650 | 78 | if (refcount_dec_and_test(&tw->tw_refcnt)) |
4dbc8ef7 | 79 | inet_twsk_free(tw); |
7054fb93 PE |
80 | } |
81 | EXPORT_SYMBOL_GPL(inet_twsk_put); | |
82 | ||
05dbc7b5 ED |
83 | static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw, |
84 | struct hlist_nulls_head *list) | |
85 | { | |
86 | hlist_nulls_add_head_rcu(&tw->tw_node, list); | |
87 | } | |
88 | ||
89 | static void inet_twsk_add_bind_node(struct inet_timewait_sock *tw, | |
90 | struct hlist_head *list) | |
91 | { | |
92 | hlist_add_head(&tw->tw_bind_node, list); | |
93 | } | |
94 | ||
e48c414e | 95 | /* |
614bdd4d | 96 | * Enter the time wait state. |
e48c414e ACM |
97 | * Essentially we whip up a timewait bucket, copy the relevant info into it |
98 | * from the SK, and mess with hash chains and list linkage. | |
99 | */ | |
100 | void __inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk, | |
101 | struct inet_hashinfo *hashinfo) | |
102 | { | |
103 | const struct inet_sock *inet = inet_sk(sk); | |
463c84b9 | 104 | const struct inet_connection_sock *icsk = inet_csk(sk); |
81c3d547 | 105 | struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash); |
9db66bdc | 106 | spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash); |
e48c414e ACM |
107 | struct inet_bind_hashbucket *bhead; |
108 | /* Step 1: Put TW into bind hash. Original socket stays there too. | |
109 | Note, that any socket with inet->num != 0 MUST be bound in | |
110 | binding cache, even if it is closed. | |
111 | */ | |
c720c7e8 | 112 | bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num, |
7f635ab7 | 113 | hashinfo->bhash_size)]; |
614bdd4d | 114 | spin_lock_bh(&bhead->lock); |
463c84b9 | 115 | tw->tw_tb = icsk->icsk_bind_hash; |
547b792c | 116 | WARN_ON(!icsk->icsk_bind_hash); |
e48c414e ACM |
117 | inet_twsk_add_bind_node(tw, &tw->tw_tb->owners); |
118 | spin_unlock(&bhead->lock); | |
119 | ||
9db66bdc | 120 | spin_lock(lock); |
e48c414e | 121 | |
3ab5aee7 | 122 | /* |
05dbc7b5 ED |
123 | * Step 2: Hash TW into tcp ehash chain. |
124 | * Notes : | |
ed2e9239 | 125 | * - tw_refcnt is set to 4 because : |
05dbc7b5 ED |
126 | * - We have one reference from bhash chain. |
127 | * - We have one reference from ehash chain. | |
ed2e9239 ED |
128 | * - We have one reference from timer. |
129 | * - One reference for ourself (our caller will release it). | |
05dbc7b5 ED |
130 | * We can use atomic_set() because prior spin_lock()/spin_unlock() |
131 | * committed into memory all tw fields. | |
3ab5aee7 | 132 | */ |
41c6d650 | 133 | refcount_set(&tw->tw_refcnt, 4); |
05dbc7b5 | 134 | inet_twsk_add_node_rcu(tw, &ehead->chain); |
3ab5aee7 | 135 | |
05dbc7b5 | 136 | /* Step 3: Remove SK from hash chain */ |
3ab5aee7 ED |
137 | if (__sk_nulls_del_node_init_rcu(sk)) |
138 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); | |
e48c414e | 139 | |
614bdd4d | 140 | spin_unlock_bh(lock); |
e48c414e | 141 | } |
696ab2d3 ACM |
142 | EXPORT_SYMBOL_GPL(__inet_twsk_hashdance); |
143 | ||
1ab791dc | 144 | static void tw_timer_handler(struct timer_list *t) |
c676270b | 145 | { |
1ab791dc | 146 | struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer); |
789f558c ED |
147 | |
148 | if (tw->tw_kill) | |
02a1d6e7 | 149 | __NET_INC_STATS(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED); |
789f558c | 150 | else |
02a1d6e7 | 151 | __NET_INC_STATS(twsk_net(tw), LINUX_MIB_TIMEWAITED); |
789f558c ED |
152 | inet_twsk_kill(tw); |
153 | } | |
154 | ||
155 | struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, | |
156 | struct inet_timewait_death_row *dr, | |
157 | const int state) | |
158 | { | |
159 | struct inet_timewait_sock *tw; | |
160 | ||
161 | if (atomic_read(&dr->tw_count) >= dr->sysctl_max_tw_buckets) | |
162 | return NULL; | |
163 | ||
164 | tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab, | |
165 | GFP_ATOMIC); | |
00db4124 | 166 | if (tw) { |
c676270b ACM |
167 | const struct inet_sock *inet = inet_sk(sk); |
168 | ||
789f558c | 169 | tw->tw_dr = dr; |
c676270b | 170 | /* Give us an identity. */ |
c720c7e8 ED |
171 | tw->tw_daddr = inet->inet_daddr; |
172 | tw->tw_rcv_saddr = inet->inet_rcv_saddr; | |
c676270b | 173 | tw->tw_bound_dev_if = sk->sk_bound_dev_if; |
66b13d99 | 174 | tw->tw_tos = inet->tos; |
c720c7e8 | 175 | tw->tw_num = inet->inet_num; |
c676270b ACM |
176 | tw->tw_state = TCP_TIME_WAIT; |
177 | tw->tw_substate = state; | |
c720c7e8 ED |
178 | tw->tw_sport = inet->inet_sport; |
179 | tw->tw_dport = inet->inet_dport; | |
c676270b ACM |
180 | tw->tw_family = sk->sk_family; |
181 | tw->tw_reuse = sk->sk_reuse; | |
81c3d547 | 182 | tw->tw_hash = sk->sk_hash; |
c676270b | 183 | tw->tw_ipv6only = 0; |
f5715aea | 184 | tw->tw_transparent = inet->transparent; |
c676270b | 185 | tw->tw_prot = sk->sk_prot_creator; |
33cf7c90 | 186 | atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie)); |
efd7ef1c | 187 | twsk_net_set(tw, sock_net(sk)); |
1ab791dc | 188 | timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED); |
47e1c323 ED |
189 | /* |
190 | * Because we use RCU lookups, we should not set tw_refcnt | |
191 | * to a non null value before everything is setup for this | |
192 | * timewait socket. | |
193 | */ | |
41c6d650 | 194 | refcount_set(&tw->tw_refcnt, 0); |
789f558c | 195 | |
eeb2b856 | 196 | __module_get(tw->tw_prot->owner); |
c676270b ACM |
197 | } |
198 | ||
199 | return tw; | |
200 | } | |
696ab2d3 ACM |
201 | EXPORT_SYMBOL_GPL(inet_twsk_alloc); |
202 | ||
696ab2d3 ACM |
203 | /* These are always called from BH context. See callers in |
204 | * tcp_input.c to verify this. | |
205 | */ | |
206 | ||
dbe7faa4 ED |
207 | /* This is for handling early-kills of TIME_WAIT sockets. |
208 | * Warning : consume reference. | |
209 | * Caller should not access tw anymore. | |
210 | */ | |
211 | void inet_twsk_deschedule_put(struct inet_timewait_sock *tw) | |
696ab2d3 | 212 | { |
789f558c ED |
213 | if (del_timer_sync(&tw->tw_timer)) |
214 | inet_twsk_kill(tw); | |
dbe7faa4 | 215 | inet_twsk_put(tw); |
696ab2d3 | 216 | } |
dbe7faa4 | 217 | EXPORT_SYMBOL(inet_twsk_deschedule_put); |
696ab2d3 | 218 | |
ed2e9239 | 219 | void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm) |
696ab2d3 | 220 | { |
696ab2d3 ACM |
221 | /* timeout := RTO * 3.5 |
222 | * | |
223 | * 3.5 = 1+2+0.5 to wait for two retransmits. | |
224 | * | |
225 | * RATIONALE: if FIN arrived and we entered TIME-WAIT state, | |
226 | * our ACK acking that FIN can be lost. If N subsequent retransmitted | |
227 | * FINs (or previous seqments) are lost (probability of such event | |
228 | * is p^(N+1), where p is probability to lose single packet and | |
229 | * time to detect the loss is about RTO*(2^N - 1) with exponential | |
230 | * backoff). Normal timewait length is calculated so, that we | |
231 | * waited at least for one retransmitted FIN (maximal RTO is 120sec). | |
232 | * [ BTW Linux. following BSD, violates this requirement waiting | |
233 | * only for 60sec, we should wait at least for 240 secs. | |
234 | * Well, 240 consumes too much of resources 8) | |
235 | * ] | |
236 | * This interval is not reduced to catch old duplicate and | |
237 | * responces to our wandering segments living for two MSLs. | |
238 | * However, if we use PAWS to detect | |
239 | * old duplicates, we can reduce the interval to bounds required | |
240 | * by RTO, rather than MSL. So, if peer understands PAWS, we | |
241 | * kill tw bucket after 3.5*RTO (it is important that this number | |
242 | * is greater than TS tick!) and detect old duplicates with help | |
243 | * of PAWS. | |
244 | */ | |
696ab2d3 | 245 | |
789f558c | 246 | tw->tw_kill = timeo <= 4*HZ; |
ed2e9239 | 247 | if (!rearm) { |
f3438bc7 | 248 | BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo)); |
789f558c | 249 | atomic_inc(&tw->tw_dr->tw_count); |
ed2e9239 ED |
250 | } else { |
251 | mod_timer_pending(&tw->tw_timer, jiffies + timeo); | |
696ab2d3 | 252 | } |
696ab2d3 | 253 | } |
ed2e9239 | 254 | EXPORT_SYMBOL_GPL(__inet_twsk_schedule); |
696ab2d3 | 255 | |
1946e672 | 256 | void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family) |
d315492b DL |
257 | { |
258 | struct inet_timewait_sock *tw; | |
259 | struct sock *sk; | |
3ab5aee7 | 260 | struct hlist_nulls_node *node; |
575f4cd5 | 261 | unsigned int slot; |
d315492b | 262 | |
575f4cd5 EB |
263 | for (slot = 0; slot <= hashinfo->ehash_mask; slot++) { |
264 | struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; | |
265 | restart_rcu: | |
738e6d30 | 266 | cond_resched(); |
575f4cd5 | 267 | rcu_read_lock(); |
d315492b | 268 | restart: |
05dbc7b5 ED |
269 | sk_nulls_for_each_rcu(sk, node, &head->chain) { |
270 | if (sk->sk_state != TCP_TIME_WAIT) | |
271 | continue; | |
d315492b | 272 | tw = inet_twsk(sk); |
b099ce26 EB |
273 | if ((tw->tw_family != family) || |
274 | atomic_read(&twsk_net(tw)->count)) | |
d315492b DL |
275 | continue; |
276 | ||
41c6d650 | 277 | if (unlikely(!refcount_inc_not_zero(&tw->tw_refcnt))) |
575f4cd5 EB |
278 | continue; |
279 | ||
b099ce26 EB |
280 | if (unlikely((tw->tw_family != family) || |
281 | atomic_read(&twsk_net(tw)->count))) { | |
575f4cd5 EB |
282 | inet_twsk_put(tw); |
283 | goto restart; | |
284 | } | |
285 | ||
286 | rcu_read_unlock(); | |
91035f0b | 287 | local_bh_disable(); |
dbe7faa4 | 288 | inet_twsk_deschedule_put(tw); |
91035f0b | 289 | local_bh_enable(); |
575f4cd5 | 290 | goto restart_rcu; |
d315492b | 291 | } |
575f4cd5 EB |
292 | /* If the nulls value we got at the end of this lookup is |
293 | * not the expected one, we must restart lookup. | |
294 | * We probably met an item that was moved to another chain. | |
295 | */ | |
296 | if (get_nulls_value(node) != slot) | |
297 | goto restart; | |
298 | rcu_read_unlock(); | |
d315492b | 299 | } |
d315492b DL |
300 | } |
301 | EXPORT_SYMBOL_GPL(inet_twsk_purge); |