1 // SPDX-License-Identifier: GPL-2.0-only
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Generic TIME_WAIT sockets functions
9 * From code orinally in TCP
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
14 #include <linux/module.h>
15 #include <net/inet_hashtables.h>
16 #include <net/inet_timewait_sock.h>
21 * inet_twsk_bind_unhash - unhash a timewait socket from bind hash
22 * @tw: timewait socket
23 * @hashinfo: hashinfo pointer
25 * unhash a timewait socket from bind hash, if hashed.
26 * bind hash lock must be held by caller.
27 * Returns 1 if caller should call inet_twsk_put() after lock release.
29 void inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
30 struct inet_hashinfo *hashinfo)
32 struct inet_bind_bucket *tb = tw->tw_tb;
37 __hlist_del(&tw->tw_bind_node);
39 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
40 __sock_put((struct sock *)tw);
43 /* Must be called with locally disabled BHs. */
44 static void inet_twsk_kill(struct inet_timewait_sock *tw)
46 struct inet_hashinfo *hashinfo = tw->tw_dr->hashinfo;
47 spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash);
48 struct inet_bind_hashbucket *bhead;
51 sk_nulls_del_node_init_rcu((struct sock *)tw);
54 /* Disassociate with bind bucket. */
55 bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num,
56 hashinfo->bhash_size)];
58 spin_lock(&bhead->lock);
59 inet_twsk_bind_unhash(tw, hashinfo);
60 spin_unlock(&bhead->lock);
62 if (refcount_dec_and_test(&tw->tw_dr->tw_refcount))
68 void inet_twsk_free(struct inet_timewait_sock *tw)
70 struct module *owner = tw->tw_prot->owner;
71 twsk_destructor((struct sock *)tw);
72 #ifdef SOCK_REFCNT_DEBUG
73 pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw);
75 kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
79 void inet_twsk_put(struct inet_timewait_sock *tw)
81 if (refcount_dec_and_test(&tw->tw_refcnt))
84 EXPORT_SYMBOL_GPL(inet_twsk_put);
86 static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
87 struct hlist_nulls_head *list)
89 hlist_nulls_add_head_rcu(&tw->tw_node, list);
92 static void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
93 struct hlist_head *list)
95 hlist_add_head(&tw->tw_bind_node, list);
99 * Enter the time wait state. This is called with locally disabled BH.
100 * Essentially we whip up a timewait bucket, copy the relevant info into it
101 * from the SK, and mess with hash chains and list linkage.
103 void inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
104 struct inet_hashinfo *hashinfo)
106 const struct inet_sock *inet = inet_sk(sk);
107 const struct inet_connection_sock *icsk = inet_csk(sk);
108 struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
109 spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
110 struct inet_bind_hashbucket *bhead;
111 /* Step 1: Put TW into bind hash. Original socket stays there too.
112 Note, that any socket with inet->num != 0 MUST be bound in
113 binding cache, even if it is closed.
115 bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num,
116 hashinfo->bhash_size)];
117 spin_lock(&bhead->lock);
118 tw->tw_tb = icsk->icsk_bind_hash;
119 WARN_ON(!icsk->icsk_bind_hash);
120 inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
121 spin_unlock(&bhead->lock);
125 inet_twsk_add_node_rcu(tw, &ehead->chain);
127 /* Step 3: Remove SK from hash chain */
128 if (__sk_nulls_del_node_init_rcu(sk))
129 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
133 /* tw_refcnt is set to 3 because we have :
134 * - one reference for bhash chain.
135 * - one reference for ehash chain.
136 * - one reference for timer.
137 * We can use atomic_set() because prior spin_lock()/spin_unlock()
138 * committed into memory all tw fields.
139 * Also note that after this point, we lost our implicit reference
140 * so we are not allowed to use tw anymore.
142 refcount_set(&tw->tw_refcnt, 3);
144 EXPORT_SYMBOL_GPL(inet_twsk_hashdance);
146 static void tw_timer_handler(struct timer_list *t)
148 struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer);
153 struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
154 struct inet_timewait_death_row *dr,
157 struct inet_timewait_sock *tw;
159 if (refcount_read(&dr->tw_refcount) - 1 >= dr->sysctl_max_tw_buckets)
162 tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
165 const struct inet_sock *inet = inet_sk(sk);
168 /* Give us an identity. */
169 tw->tw_daddr = inet->inet_daddr;
170 tw->tw_rcv_saddr = inet->inet_rcv_saddr;
171 tw->tw_bound_dev_if = sk->sk_bound_dev_if;
172 tw->tw_tos = inet->tos;
173 tw->tw_num = inet->inet_num;
174 tw->tw_state = TCP_TIME_WAIT;
175 tw->tw_substate = state;
176 tw->tw_sport = inet->inet_sport;
177 tw->tw_dport = inet->inet_dport;
178 tw->tw_family = sk->sk_family;
179 tw->tw_reuse = sk->sk_reuse;
180 tw->tw_reuseport = sk->sk_reuseport;
181 tw->tw_hash = sk->sk_hash;
183 tw->tw_transparent = inet->transparent;
184 tw->tw_prot = sk->sk_prot_creator;
185 atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
186 twsk_net_set(tw, sock_net(sk));
187 timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED);
189 * Because we use RCU lookups, we should not set tw_refcnt
190 * to a non null value before everything is setup for this
193 refcount_set(&tw->tw_refcnt, 0);
195 __module_get(tw->tw_prot->owner);
200 EXPORT_SYMBOL_GPL(inet_twsk_alloc);
202 /* These are always called from BH context. See callers in
203 * tcp_input.c to verify this.
206 /* This is for handling early-kills of TIME_WAIT sockets.
207 * Warning : consume reference.
208 * Caller should not access tw anymore.
210 void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
212 if (del_timer_sync(&tw->tw_timer))
216 EXPORT_SYMBOL(inet_twsk_deschedule_put);
218 void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
220 /* timeout := RTO * 3.5
222 * 3.5 = 1+2+0.5 to wait for two retransmits.
224 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
225 * our ACK acking that FIN can be lost. If N subsequent retransmitted
226 * FINs (or previous seqments) are lost (probability of such event
227 * is p^(N+1), where p is probability to lose single packet and
228 * time to detect the loss is about RTO*(2^N - 1) with exponential
229 * backoff). Normal timewait length is calculated so, that we
230 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
231 * [ BTW Linux. following BSD, violates this requirement waiting
232 * only for 60sec, we should wait at least for 240 secs.
233 * Well, 240 consumes too much of resources 8)
235 * This interval is not reduced to catch old duplicate and
236 * responces to our wandering segments living for two MSLs.
237 * However, if we use PAWS to detect
238 * old duplicates, we can reduce the interval to bounds required
239 * by RTO, rather than MSL. So, if peer understands PAWS, we
240 * kill tw bucket after 3.5*RTO (it is important that this number
241 * is greater than TS tick!) and detect old duplicates with help
246 bool kill = timeo <= 4*HZ;
248 __NET_INC_STATS(twsk_net(tw), kill ? LINUX_MIB_TIMEWAITKILLED :
249 LINUX_MIB_TIMEWAITED);
250 BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
251 refcount_inc(&tw->tw_dr->tw_refcount);
253 mod_timer_pending(&tw->tw_timer, jiffies + timeo);
256 EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
258 void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
260 struct inet_timewait_sock *tw;
262 struct hlist_nulls_node *node;
265 for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
266 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
271 sk_nulls_for_each_rcu(sk, node, &head->chain) {
272 if (sk->sk_state != TCP_TIME_WAIT)
275 if ((tw->tw_family != family) ||
276 refcount_read(&twsk_net(tw)->ns.count))
279 if (unlikely(!refcount_inc_not_zero(&tw->tw_refcnt)))
282 if (unlikely((tw->tw_family != family) ||
283 refcount_read(&twsk_net(tw)->ns.count))) {
290 inet_twsk_deschedule_put(tw);
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.
298 if (get_nulls_value(node) != slot)
303 EXPORT_SYMBOL_GPL(inet_twsk_purge);