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
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/module.h>
14 #include <net/inet_hashtables.h>
15 #include <net/inet_timewait_sock.h>
16 #include <net/ip.h>
17
18
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.
27 */
inet_twsk_bind_unhash(struct inet_timewait_sock * tw,struct inet_hashinfo * hashinfo)28 void inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
29 struct inet_hashinfo *hashinfo)
30 {
31 struct inet_bind_bucket *tb = tw->tw_tb;
32
33 if (!tb)
34 return;
35
36 __hlist_del(&tw->tw_bind_node);
37 tw->tw_tb = NULL;
38 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
39 __sock_put((struct sock *)tw);
40 }
41
42 /* Must be called with locally disabled BHs. */
inet_twsk_kill(struct inet_timewait_sock * tw)43 static void inet_twsk_kill(struct inet_timewait_sock *tw)
44 {
45 struct inet_hashinfo *hashinfo = tw->tw_dr->hashinfo;
46 spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash);
47 struct inet_bind_hashbucket *bhead;
48
49 spin_lock(lock);
50 sk_nulls_del_node_init_rcu((struct sock *)tw);
51 spin_unlock(lock);
52
53 /* Disassociate with bind bucket. */
54 bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num,
55 hashinfo->bhash_size)];
56
57 spin_lock(&bhead->lock);
58 inet_twsk_bind_unhash(tw, hashinfo);
59 spin_unlock(&bhead->lock);
60
61 atomic_dec(&tw->tw_dr->tw_count);
62 inet_twsk_put(tw);
63 }
64
inet_twsk_free(struct inet_timewait_sock * tw)65 void inet_twsk_free(struct inet_timewait_sock *tw)
66 {
67 struct module *owner = tw->tw_prot->owner;
68 twsk_destructor((struct sock *)tw);
69 #ifdef SOCK_REFCNT_DEBUG
70 pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw);
71 #endif
72 kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
73 module_put(owner);
74 }
75
inet_twsk_put(struct inet_timewait_sock * tw)76 void inet_twsk_put(struct inet_timewait_sock *tw)
77 {
78 if (refcount_dec_and_test(&tw->tw_refcnt))
79 inet_twsk_free(tw);
80 }
81 EXPORT_SYMBOL_GPL(inet_twsk_put);
82
inet_twsk_add_node_rcu(struct inet_timewait_sock * tw,struct hlist_nulls_head * list)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
inet_twsk_add_bind_node(struct inet_timewait_sock * tw,struct hlist_head * list)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
95 /*
96 * Enter the time wait state.
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 */
__inet_twsk_hashdance(struct inet_timewait_sock * tw,struct sock * sk,struct inet_hashinfo * hashinfo)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);
104 const struct inet_connection_sock *icsk = inet_csk(sk);
105 struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
106 spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
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 */
112 bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num,
113 hashinfo->bhash_size)];
114 spin_lock_bh(&bhead->lock);
115 tw->tw_tb = icsk->icsk_bind_hash;
116 WARN_ON(!icsk->icsk_bind_hash);
117 inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
118 spin_unlock(&bhead->lock);
119
120 spin_lock(lock);
121
122 /*
123 * Step 2: Hash TW into tcp ehash chain.
124 * Notes :
125 * - tw_refcnt is set to 4 because :
126 * - We have one reference from bhash chain.
127 * - We have one reference from ehash chain.
128 * - We have one reference from timer.
129 * - One reference for ourself (our caller will release it).
130 * We can use atomic_set() because prior spin_lock()/spin_unlock()
131 * committed into memory all tw fields.
132 */
133 refcount_set(&tw->tw_refcnt, 4);
134 inet_twsk_add_node_rcu(tw, &ehead->chain);
135
136 /* Step 3: Remove SK from hash chain */
137 if (__sk_nulls_del_node_init_rcu(sk))
138 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
139
140 spin_unlock_bh(lock);
141 }
142 EXPORT_SYMBOL_GPL(__inet_twsk_hashdance);
143
tw_timer_handler(unsigned long data)144 static void tw_timer_handler(unsigned long data)
145 {
146 struct inet_timewait_sock *tw = (struct inet_timewait_sock *)data;
147
148 if (tw->tw_kill)
149 __NET_INC_STATS(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED);
150 else
151 __NET_INC_STATS(twsk_net(tw), LINUX_MIB_TIMEWAITED);
152 inet_twsk_kill(tw);
153 }
154
inet_twsk_alloc(const struct sock * sk,struct inet_timewait_death_row * dr,const int state)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);
166 if (tw) {
167 const struct inet_sock *inet = inet_sk(sk);
168
169 tw->tw_dr = dr;
170 /* Give us an identity. */
171 tw->tw_daddr = inet->inet_daddr;
172 tw->tw_rcv_saddr = inet->inet_rcv_saddr;
173 tw->tw_bound_dev_if = sk->sk_bound_dev_if;
174 tw->tw_tos = inet->tos;
175 tw->tw_num = inet->inet_num;
176 tw->tw_state = TCP_TIME_WAIT;
177 tw->tw_substate = state;
178 tw->tw_sport = inet->inet_sport;
179 tw->tw_dport = inet->inet_dport;
180 tw->tw_family = sk->sk_family;
181 tw->tw_reuse = sk->sk_reuse;
182 tw->tw_reuseport = sk->sk_reuseport;
183 tw->tw_hash = sk->sk_hash;
184 tw->tw_ipv6only = 0;
185 tw->tw_transparent = inet->transparent;
186 tw->tw_prot = sk->sk_prot_creator;
187 atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
188 twsk_net_set(tw, sock_net(sk));
189 setup_pinned_timer(&tw->tw_timer, tw_timer_handler,
190 (unsigned long)tw);
191 /*
192 * Because we use RCU lookups, we should not set tw_refcnt
193 * to a non null value before everything is setup for this
194 * timewait socket.
195 */
196 refcount_set(&tw->tw_refcnt, 0);
197
198 __module_get(tw->tw_prot->owner);
199 }
200
201 return tw;
202 }
203 EXPORT_SYMBOL_GPL(inet_twsk_alloc);
204
205 /* These are always called from BH context. See callers in
206 * tcp_input.c to verify this.
207 */
208
209 /* This is for handling early-kills of TIME_WAIT sockets.
210 * Warning : consume reference.
211 * Caller should not access tw anymore.
212 */
inet_twsk_deschedule_put(struct inet_timewait_sock * tw)213 void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
214 {
215 if (del_timer_sync(&tw->tw_timer))
216 inet_twsk_kill(tw);
217 inet_twsk_put(tw);
218 }
219 EXPORT_SYMBOL(inet_twsk_deschedule_put);
220
__inet_twsk_schedule(struct inet_timewait_sock * tw,int timeo,bool rearm)221 void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
222 {
223 /* timeout := RTO * 3.5
224 *
225 * 3.5 = 1+2+0.5 to wait for two retransmits.
226 *
227 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
228 * our ACK acking that FIN can be lost. If N subsequent retransmitted
229 * FINs (or previous seqments) are lost (probability of such event
230 * is p^(N+1), where p is probability to lose single packet and
231 * time to detect the loss is about RTO*(2^N - 1) with exponential
232 * backoff). Normal timewait length is calculated so, that we
233 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
234 * [ BTW Linux. following BSD, violates this requirement waiting
235 * only for 60sec, we should wait at least for 240 secs.
236 * Well, 240 consumes too much of resources 8)
237 * ]
238 * This interval is not reduced to catch old duplicate and
239 * responces to our wandering segments living for two MSLs.
240 * However, if we use PAWS to detect
241 * old duplicates, we can reduce the interval to bounds required
242 * by RTO, rather than MSL. So, if peer understands PAWS, we
243 * kill tw bucket after 3.5*RTO (it is important that this number
244 * is greater than TS tick!) and detect old duplicates with help
245 * of PAWS.
246 */
247
248 tw->tw_kill = timeo <= 4*HZ;
249 if (!rearm) {
250 BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
251 atomic_inc(&tw->tw_dr->tw_count);
252 } else {
253 mod_timer_pending(&tw->tw_timer, jiffies + timeo);
254 }
255 }
256 EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
257
inet_twsk_purge(struct inet_hashinfo * hashinfo,int family)258 void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
259 {
260 struct inet_timewait_sock *tw;
261 struct sock *sk;
262 struct hlist_nulls_node *node;
263 unsigned int slot;
264
265 for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
266 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
267 restart_rcu:
268 cond_resched();
269 rcu_read_lock();
270 restart:
271 sk_nulls_for_each_rcu(sk, node, &head->chain) {
272 if (sk->sk_state != TCP_TIME_WAIT)
273 continue;
274 tw = inet_twsk(sk);
275 if ((tw->tw_family != family) ||
276 atomic_read(&twsk_net(tw)->count))
277 continue;
278
279 if (unlikely(!refcount_inc_not_zero(&tw->tw_refcnt)))
280 continue;
281
282 if (unlikely((tw->tw_family != family) ||
283 atomic_read(&twsk_net(tw)->count))) {
284 inet_twsk_put(tw);
285 goto restart;
286 }
287
288 rcu_read_unlock();
289 local_bh_disable();
290 inet_twsk_deschedule_put(tw);
291 local_bh_enable();
292 goto restart_rcu;
293 }
294 /* If the nulls value we got at the end of this lookup is
295 * not the expected one, we must restart lookup.
296 * We probably met an item that was moved to another chain.
297 */
298 if (get_nulls_value(node) != slot)
299 goto restart;
300 rcu_read_unlock();
301 }
302 }
303 EXPORT_SYMBOL_GPL(inet_twsk_purge);
304