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_bind2_bucket *tb2 = tw->tw_tb2;
33 struct inet_bind_bucket *tb = tw->tw_tb;
38 __hlist_del(&tw->tw_bind_node);
40 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
42 __hlist_del(&tw->tw_bind2_node);
44 inet_bind2_bucket_destroy(hashinfo->bind2_bucket_cachep, tb2);
46 __sock_put((struct sock *)tw);
49 /* Must be called with locally disabled BHs. */
50 static void inet_twsk_kill(struct inet_timewait_sock *tw)
52 struct inet_hashinfo *hashinfo = tw->tw_dr->hashinfo;
53 spinlock_t *lock = inet_ehash_lockp(hashinfo, tw->tw_hash);
54 struct inet_bind_hashbucket *bhead, *bhead2;
57 sk_nulls_del_node_init_rcu((struct sock *)tw);
60 /* Disassociate with bind bucket. */
61 bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num,
62 hashinfo->bhash_size)];
63 bhead2 = inet_bhashfn_portaddr(hashinfo, (struct sock *)tw,
64 twsk_net(tw), tw->tw_num);
66 spin_lock(&bhead->lock);
67 spin_lock(&bhead2->lock);
68 inet_twsk_bind_unhash(tw, hashinfo);
69 spin_unlock(&bhead2->lock);
70 spin_unlock(&bhead->lock);
72 refcount_dec(&tw->tw_dr->tw_refcount);
76 void inet_twsk_free(struct inet_timewait_sock *tw)
78 struct module *owner = tw->tw_prot->owner;
79 twsk_destructor((struct sock *)tw);
80 #ifdef SOCK_REFCNT_DEBUG
81 pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw);
83 kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
87 void inet_twsk_put(struct inet_timewait_sock *tw)
89 if (refcount_dec_and_test(&tw->tw_refcnt))
92 EXPORT_SYMBOL_GPL(inet_twsk_put);
94 static void inet_twsk_add_node_tail_rcu(struct inet_timewait_sock *tw,
95 struct hlist_nulls_head *list)
97 hlist_nulls_add_tail_rcu(&tw->tw_node, list);
100 static void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
101 struct hlist_head *list)
103 hlist_add_head(&tw->tw_bind_node, list);
106 static void inet_twsk_add_bind2_node(struct inet_timewait_sock *tw,
107 struct hlist_head *list)
109 hlist_add_head(&tw->tw_bind2_node, list);
113 * Enter the time wait state. This is called with locally disabled BH.
114 * Essentially we whip up a timewait bucket, copy the relevant info into it
115 * from the SK, and mess with hash chains and list linkage.
117 void inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
118 struct inet_hashinfo *hashinfo)
120 const struct inet_sock *inet = inet_sk(sk);
121 const struct inet_connection_sock *icsk = inet_csk(sk);
122 struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
123 spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
124 struct inet_bind_hashbucket *bhead, *bhead2;
126 /* Step 1: Put TW into bind hash. Original socket stays there too.
127 Note, that any socket with inet->num != 0 MUST be bound in
128 binding cache, even if it is closed.
130 bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num,
131 hashinfo->bhash_size)];
132 bhead2 = inet_bhashfn_portaddr(hashinfo, sk, twsk_net(tw), inet->inet_num);
134 spin_lock(&bhead->lock);
135 spin_lock(&bhead2->lock);
137 tw->tw_tb = icsk->icsk_bind_hash;
138 WARN_ON(!icsk->icsk_bind_hash);
139 inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
141 tw->tw_tb2 = icsk->icsk_bind2_hash;
142 WARN_ON(!icsk->icsk_bind2_hash);
143 inet_twsk_add_bind2_node(tw, &tw->tw_tb2->deathrow);
145 spin_unlock(&bhead2->lock);
146 spin_unlock(&bhead->lock);
150 inet_twsk_add_node_tail_rcu(tw, &ehead->chain);
152 /* Step 3: Remove SK from hash chain */
153 if (__sk_nulls_del_node_init_rcu(sk))
154 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
158 /* tw_refcnt is set to 3 because we have :
159 * - one reference for bhash chain.
160 * - one reference for ehash chain.
161 * - one reference for timer.
162 * We can use atomic_set() because prior spin_lock()/spin_unlock()
163 * committed into memory all tw fields.
164 * Also note that after this point, we lost our implicit reference
165 * so we are not allowed to use tw anymore.
167 refcount_set(&tw->tw_refcnt, 3);
169 EXPORT_SYMBOL_GPL(inet_twsk_hashdance);
171 static void tw_timer_handler(struct timer_list *t)
173 struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer);
178 struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
179 struct inet_timewait_death_row *dr,
182 struct inet_timewait_sock *tw;
184 if (refcount_read(&dr->tw_refcount) - 1 >=
185 READ_ONCE(dr->sysctl_max_tw_buckets))
188 tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
191 const struct inet_sock *inet = inet_sk(sk);
194 /* Give us an identity. */
195 tw->tw_daddr = inet->inet_daddr;
196 tw->tw_rcv_saddr = inet->inet_rcv_saddr;
197 tw->tw_bound_dev_if = sk->sk_bound_dev_if;
198 tw->tw_tos = inet->tos;
199 tw->tw_num = inet->inet_num;
200 tw->tw_state = TCP_TIME_WAIT;
201 tw->tw_substate = state;
202 tw->tw_sport = inet->inet_sport;
203 tw->tw_dport = inet->inet_dport;
204 tw->tw_family = sk->sk_family;
205 tw->tw_reuse = sk->sk_reuse;
206 tw->tw_reuseport = sk->sk_reuseport;
207 tw->tw_hash = sk->sk_hash;
209 tw->tw_transparent = inet->transparent;
210 tw->tw_prot = sk->sk_prot_creator;
211 atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
212 twsk_net_set(tw, sock_net(sk));
213 timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED);
215 * Because we use RCU lookups, we should not set tw_refcnt
216 * to a non null value before everything is setup for this
219 refcount_set(&tw->tw_refcnt, 0);
221 __module_get(tw->tw_prot->owner);
226 EXPORT_SYMBOL_GPL(inet_twsk_alloc);
228 /* These are always called from BH context. See callers in
229 * tcp_input.c to verify this.
232 /* This is for handling early-kills of TIME_WAIT sockets.
233 * Warning : consume reference.
234 * Caller should not access tw anymore.
236 void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
238 if (del_timer_sync(&tw->tw_timer))
242 EXPORT_SYMBOL(inet_twsk_deschedule_put);
244 void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
246 /* timeout := RTO * 3.5
248 * 3.5 = 1+2+0.5 to wait for two retransmits.
250 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
251 * our ACK acking that FIN can be lost. If N subsequent retransmitted
252 * FINs (or previous seqments) are lost (probability of such event
253 * is p^(N+1), where p is probability to lose single packet and
254 * time to detect the loss is about RTO*(2^N - 1) with exponential
255 * backoff). Normal timewait length is calculated so, that we
256 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
257 * [ BTW Linux. following BSD, violates this requirement waiting
258 * only for 60sec, we should wait at least for 240 secs.
259 * Well, 240 consumes too much of resources 8)
261 * This interval is not reduced to catch old duplicate and
262 * responces to our wandering segments living for two MSLs.
263 * However, if we use PAWS to detect
264 * old duplicates, we can reduce the interval to bounds required
265 * by RTO, rather than MSL. So, if peer understands PAWS, we
266 * kill tw bucket after 3.5*RTO (it is important that this number
267 * is greater than TS tick!) and detect old duplicates with help
272 bool kill = timeo <= 4*HZ;
274 __NET_INC_STATS(twsk_net(tw), kill ? LINUX_MIB_TIMEWAITKILLED :
275 LINUX_MIB_TIMEWAITED);
276 BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
277 refcount_inc(&tw->tw_dr->tw_refcount);
279 mod_timer_pending(&tw->tw_timer, jiffies + timeo);
282 EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
284 void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
286 struct inet_timewait_sock *tw;
288 struct hlist_nulls_node *node;
291 for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
292 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
297 sk_nulls_for_each_rcu(sk, node, &head->chain) {
298 if (sk->sk_state != TCP_TIME_WAIT) {
299 /* A kernel listener socket might not hold refcnt for net,
300 * so reqsk_timer_handler() could be fired after net is
301 * freed. Userspace listener and reqsk never exist here.
303 if (unlikely(sk->sk_state == TCP_NEW_SYN_RECV &&
305 struct request_sock *req = inet_reqsk(sk);
307 inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, req);
314 if ((tw->tw_family != family) ||
315 refcount_read(&twsk_net(tw)->ns.count))
318 if (unlikely(!refcount_inc_not_zero(&tw->tw_refcnt)))
321 if (unlikely((tw->tw_family != family) ||
322 refcount_read(&twsk_net(tw)->ns.count))) {
329 inet_twsk_deschedule_put(tw);
333 /* If the nulls value we got at the end of this lookup is
334 * not the expected one, we must restart lookup.
335 * We probably met an item that was moved to another chain.
337 if (get_nulls_value(node) != slot)
342 EXPORT_SYMBOL_GPL(inet_twsk_purge);