xref: /linux/net/ipv4/inet_timewait_sock.c (revision 7fc7f25419f5a6b09199ba4b5026b94ef184fa79)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
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.
6  *
7  *		Generic TIME_WAIT sockets functions
8  *
9  *		From code orinally in TCP
10  */
11 
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>
17 #include <net/ip.h>
18 
19 
20 /**
21  *	inet_twsk_bind_unhash - unhash a timewait socket from bind hash
22  *	@tw: timewait socket
23  *	@hashinfo: hashinfo pointer
24  *
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.
28  */
29 void inet_twsk_bind_unhash(struct inet_timewait_sock *tw,
30 			  struct inet_hashinfo *hashinfo)
31 {
32 	struct inet_bind_bucket *tb = tw->tw_tb;
33 
34 	if (!tb)
35 		return;
36 
37 	__hlist_del(&tw->tw_bind_node);
38 	tw->tw_tb = NULL;
39 	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
40 	__sock_put((struct sock *)tw);
41 }
42 
43 /* Must be called with locally disabled BHs. */
44 static void inet_twsk_kill(struct inet_timewait_sock *tw)
45 {
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;
49 
50 	spin_lock(lock);
51 	sk_nulls_del_node_init_rcu((struct sock *)tw);
52 	spin_unlock(lock);
53 
54 	/* Disassociate with bind bucket. */
55 	bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), tw->tw_num,
56 			hashinfo->bhash_size)];
57 
58 	spin_lock(&bhead->lock);
59 	inet_twsk_bind_unhash(tw, hashinfo);
60 	spin_unlock(&bhead->lock);
61 
62 	refcount_dec(&tw->tw_dr->tw_refcount);
63 	inet_twsk_put(tw);
64 }
65 
66 void inet_twsk_free(struct inet_timewait_sock *tw)
67 {
68 	struct module *owner = tw->tw_prot->owner;
69 	twsk_destructor((struct sock *)tw);
70 #ifdef SOCK_REFCNT_DEBUG
71 	pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw);
72 #endif
73 	kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
74 	module_put(owner);
75 }
76 
77 void inet_twsk_put(struct inet_timewait_sock *tw)
78 {
79 	if (refcount_dec_and_test(&tw->tw_refcnt))
80 		inet_twsk_free(tw);
81 }
82 EXPORT_SYMBOL_GPL(inet_twsk_put);
83 
84 static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
85 				   struct hlist_nulls_head *list)
86 {
87 	hlist_nulls_add_head_rcu(&tw->tw_node, list);
88 }
89 
90 static void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
91 				    struct hlist_head *list)
92 {
93 	hlist_add_head(&tw->tw_bind_node, list);
94 }
95 
96 /*
97  * Enter the time wait state. This is called with locally disabled BH.
98  * Essentially we whip up a timewait bucket, copy the relevant info into it
99  * from the SK, and mess with hash chains and list linkage.
100  */
101 void inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
102 			   struct inet_hashinfo *hashinfo)
103 {
104 	const struct inet_sock *inet = inet_sk(sk);
105 	const struct inet_connection_sock *icsk = inet_csk(sk);
106 	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
107 	spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
108 	struct inet_bind_hashbucket *bhead;
109 	/* Step 1: Put TW into bind hash. Original socket stays there too.
110 	   Note, that any socket with inet->num != 0 MUST be bound in
111 	   binding cache, even if it is closed.
112 	 */
113 	bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num,
114 			hashinfo->bhash_size)];
115 	spin_lock(&bhead->lock);
116 	tw->tw_tb = icsk->icsk_bind_hash;
117 	WARN_ON(!icsk->icsk_bind_hash);
118 	inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
119 	spin_unlock(&bhead->lock);
120 
121 	spin_lock(lock);
122 
123 	inet_twsk_add_node_rcu(tw, &ehead->chain);
124 
125 	/* Step 3: Remove SK from hash chain */
126 	if (__sk_nulls_del_node_init_rcu(sk))
127 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
128 
129 	spin_unlock(lock);
130 
131 	/* tw_refcnt is set to 3 because we have :
132 	 * - one reference for bhash chain.
133 	 * - one reference for ehash chain.
134 	 * - one reference for timer.
135 	 * We can use atomic_set() because prior spin_lock()/spin_unlock()
136 	 * committed into memory all tw fields.
137 	 * Also note that after this point, we lost our implicit reference
138 	 * so we are not allowed to use tw anymore.
139 	 */
140 	refcount_set(&tw->tw_refcnt, 3);
141 }
142 EXPORT_SYMBOL_GPL(inet_twsk_hashdance);
143 
144 static void tw_timer_handler(struct timer_list *t)
145 {
146 	struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer);
147 
148 	inet_twsk_kill(tw);
149 }
150 
151 struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
152 					   struct inet_timewait_death_row *dr,
153 					   const int state)
154 {
155 	struct inet_timewait_sock *tw;
156 
157 	if (refcount_read(&dr->tw_refcount) - 1 >=
158 	    READ_ONCE(dr->sysctl_max_tw_buckets))
159 		return NULL;
160 
161 	tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
162 			      GFP_ATOMIC);
163 	if (tw) {
164 		const struct inet_sock *inet = inet_sk(sk);
165 
166 		tw->tw_dr	    = dr;
167 		/* Give us an identity. */
168 		tw->tw_daddr	    = inet->inet_daddr;
169 		tw->tw_rcv_saddr    = inet->inet_rcv_saddr;
170 		tw->tw_bound_dev_if = sk->sk_bound_dev_if;
171 		tw->tw_tos	    = inet->tos;
172 		tw->tw_num	    = inet->inet_num;
173 		tw->tw_state	    = TCP_TIME_WAIT;
174 		tw->tw_substate	    = state;
175 		tw->tw_sport	    = inet->inet_sport;
176 		tw->tw_dport	    = inet->inet_dport;
177 		tw->tw_family	    = sk->sk_family;
178 		tw->tw_reuse	    = sk->sk_reuse;
179 		tw->tw_reuseport    = sk->sk_reuseport;
180 		tw->tw_hash	    = sk->sk_hash;
181 		tw->tw_ipv6only	    = 0;
182 		tw->tw_transparent  = inet->transparent;
183 		tw->tw_prot	    = sk->sk_prot_creator;
184 		atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
185 		twsk_net_set(tw, sock_net(sk));
186 		timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED);
187 		/*
188 		 * Because we use RCU lookups, we should not set tw_refcnt
189 		 * to a non null value before everything is setup for this
190 		 * timewait socket.
191 		 */
192 		refcount_set(&tw->tw_refcnt, 0);
193 
194 		__module_get(tw->tw_prot->owner);
195 	}
196 
197 	return tw;
198 }
199 EXPORT_SYMBOL_GPL(inet_twsk_alloc);
200 
201 /* These are always called from BH context.  See callers in
202  * tcp_input.c to verify this.
203  */
204 
205 /* This is for handling early-kills of TIME_WAIT sockets.
206  * Warning : consume reference.
207  * Caller should not access tw anymore.
208  */
209 void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
210 {
211 	if (del_timer_sync(&tw->tw_timer))
212 		inet_twsk_kill(tw);
213 	inet_twsk_put(tw);
214 }
215 EXPORT_SYMBOL(inet_twsk_deschedule_put);
216 
217 void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
218 {
219 	/* timeout := RTO * 3.5
220 	 *
221 	 * 3.5 = 1+2+0.5 to wait for two retransmits.
222 	 *
223 	 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
224 	 * our ACK acking that FIN can be lost. If N subsequent retransmitted
225 	 * FINs (or previous seqments) are lost (probability of such event
226 	 * is p^(N+1), where p is probability to lose single packet and
227 	 * time to detect the loss is about RTO*(2^N - 1) with exponential
228 	 * backoff). Normal timewait length is calculated so, that we
229 	 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
230 	 * [ BTW Linux. following BSD, violates this requirement waiting
231 	 *   only for 60sec, we should wait at least for 240 secs.
232 	 *   Well, 240 consumes too much of resources 8)
233 	 * ]
234 	 * This interval is not reduced to catch old duplicate and
235 	 * responces to our wandering segments living for two MSLs.
236 	 * However, if we use PAWS to detect
237 	 * old duplicates, we can reduce the interval to bounds required
238 	 * by RTO, rather than MSL. So, if peer understands PAWS, we
239 	 * kill tw bucket after 3.5*RTO (it is important that this number
240 	 * is greater than TS tick!) and detect old duplicates with help
241 	 * of PAWS.
242 	 */
243 
244 	if (!rearm) {
245 		bool kill = timeo <= 4*HZ;
246 
247 		__NET_INC_STATS(twsk_net(tw), kill ? LINUX_MIB_TIMEWAITKILLED :
248 						     LINUX_MIB_TIMEWAITED);
249 		BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
250 		refcount_inc(&tw->tw_dr->tw_refcount);
251 	} else {
252 		mod_timer_pending(&tw->tw_timer, jiffies + timeo);
253 	}
254 }
255 EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
256 
257 void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
258 {
259 	struct inet_timewait_sock *tw;
260 	struct sock *sk;
261 	struct hlist_nulls_node *node;
262 	unsigned int slot;
263 
264 	for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
265 		struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
266 restart_rcu:
267 		cond_resched();
268 		rcu_read_lock();
269 restart:
270 		sk_nulls_for_each_rcu(sk, node, &head->chain) {
271 			if (sk->sk_state != TCP_TIME_WAIT) {
272 				/* A kernel listener socket might not hold refcnt for net,
273 				 * so reqsk_timer_handler() could be fired after net is
274 				 * freed.  Userspace listener and reqsk never exist here.
275 				 */
276 				if (unlikely(sk->sk_state == TCP_NEW_SYN_RECV &&
277 					     hashinfo->pernet)) {
278 					struct request_sock *req = inet_reqsk(sk);
279 
280 					inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, req);
281 				}
282 
283 				continue;
284 			}
285 
286 			tw = inet_twsk(sk);
287 			if ((tw->tw_family != family) ||
288 				refcount_read(&twsk_net(tw)->ns.count))
289 				continue;
290 
291 			if (unlikely(!refcount_inc_not_zero(&tw->tw_refcnt)))
292 				continue;
293 
294 			if (unlikely((tw->tw_family != family) ||
295 				     refcount_read(&twsk_net(tw)->ns.count))) {
296 				inet_twsk_put(tw);
297 				goto restart;
298 			}
299 
300 			rcu_read_unlock();
301 			local_bh_disable();
302 			inet_twsk_deschedule_put(tw);
303 			local_bh_enable();
304 			goto restart_rcu;
305 		}
306 		/* If the nulls value we got at the end of this lookup is
307 		 * not the expected one, we must restart lookup.
308 		 * We probably met an item that was moved to another chain.
309 		 */
310 		if (get_nulls_value(node) != slot)
311 			goto restart;
312 		rcu_read_unlock();
313 	}
314 }
315 EXPORT_SYMBOL_GPL(inet_twsk_purge);
316