xref: /openbmc/linux/net/ipv4/inet_timewait_sock.c (revision 2359ccdd)
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  */
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. */
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 
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 
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 
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 
95 /*
96  * Enter the time wait state. This is called with locally disabled BH.
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);
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(&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 	inet_twsk_add_node_rcu(tw, &ehead->chain);
123 
124 	/* Step 3: Remove SK from hash chain */
125 	if (__sk_nulls_del_node_init_rcu(sk))
126 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
127 
128 	spin_unlock(lock);
129 
130 	/* tw_refcnt is set to 3 because we have :
131 	 * - one reference for bhash chain.
132 	 * - one reference for ehash chain.
133 	 * - one reference for timer.
134 	 * We can use atomic_set() because prior spin_lock()/spin_unlock()
135 	 * committed into memory all tw fields.
136 	 * Also note that after this point, we lost our implicit reference
137 	 * so we are not allowed to use tw anymore.
138 	 */
139 	refcount_set(&tw->tw_refcnt, 3);
140 }
141 EXPORT_SYMBOL_GPL(inet_twsk_hashdance);
142 
143 static void tw_timer_handler(struct timer_list *t)
144 {
145 	struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer);
146 
147 	if (tw->tw_kill)
148 		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED);
149 	else
150 		__NET_INC_STATS(twsk_net(tw), LINUX_MIB_TIMEWAITED);
151 	inet_twsk_kill(tw);
152 }
153 
154 struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
155 					   struct inet_timewait_death_row *dr,
156 					   const int state)
157 {
158 	struct inet_timewait_sock *tw;
159 
160 	if (atomic_read(&dr->tw_count) >= dr->sysctl_max_tw_buckets)
161 		return NULL;
162 
163 	tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
164 			      GFP_ATOMIC);
165 	if (tw) {
166 		const struct inet_sock *inet = inet_sk(sk);
167 
168 		tw->tw_dr	    = dr;
169 		/* Give us an identity. */
170 		tw->tw_daddr	    = inet->inet_daddr;
171 		tw->tw_rcv_saddr    = inet->inet_rcv_saddr;
172 		tw->tw_bound_dev_if = sk->sk_bound_dev_if;
173 		tw->tw_tos	    = inet->tos;
174 		tw->tw_num	    = inet->inet_num;
175 		tw->tw_state	    = TCP_TIME_WAIT;
176 		tw->tw_substate	    = state;
177 		tw->tw_sport	    = inet->inet_sport;
178 		tw->tw_dport	    = inet->inet_dport;
179 		tw->tw_family	    = sk->sk_family;
180 		tw->tw_reuse	    = sk->sk_reuse;
181 		tw->tw_reuseport    = sk->sk_reuseport;
182 		tw->tw_hash	    = sk->sk_hash;
183 		tw->tw_ipv6only	    = 0;
184 		tw->tw_transparent  = inet->transparent;
185 		tw->tw_prot	    = sk->sk_prot_creator;
186 		atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
187 		twsk_net_set(tw, sock_net(sk));
188 		timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED);
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 		 */
194 		refcount_set(&tw->tw_refcnt, 0);
195 
196 		__module_get(tw->tw_prot->owner);
197 	}
198 
199 	return tw;
200 }
201 EXPORT_SYMBOL_GPL(inet_twsk_alloc);
202 
203 /* These are always called from BH context.  See callers in
204  * tcp_input.c to verify this.
205  */
206 
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)
212 {
213 	if (del_timer_sync(&tw->tw_timer))
214 		inet_twsk_kill(tw);
215 	inet_twsk_put(tw);
216 }
217 EXPORT_SYMBOL(inet_twsk_deschedule_put);
218 
219 void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
220 {
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 	 */
245 
246 	tw->tw_kill = timeo <= 4*HZ;
247 	if (!rearm) {
248 		BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
249 		atomic_inc(&tw->tw_dr->tw_count);
250 	} else {
251 		mod_timer_pending(&tw->tw_timer, jiffies + timeo);
252 	}
253 }
254 EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
255 
256 void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
257 {
258 	struct inet_timewait_sock *tw;
259 	struct sock *sk;
260 	struct hlist_nulls_node *node;
261 	unsigned int slot;
262 
263 	for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
264 		struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
265 restart_rcu:
266 		cond_resched();
267 		rcu_read_lock();
268 restart:
269 		sk_nulls_for_each_rcu(sk, node, &head->chain) {
270 			if (sk->sk_state != TCP_TIME_WAIT)
271 				continue;
272 			tw = inet_twsk(sk);
273 			if ((tw->tw_family != family) ||
274 				refcount_read(&twsk_net(tw)->count))
275 				continue;
276 
277 			if (unlikely(!refcount_inc_not_zero(&tw->tw_refcnt)))
278 				continue;
279 
280 			if (unlikely((tw->tw_family != family) ||
281 				     refcount_read(&twsk_net(tw)->count))) {
282 				inet_twsk_put(tw);
283 				goto restart;
284 			}
285 
286 			rcu_read_unlock();
287 			local_bh_disable();
288 			inet_twsk_deschedule_put(tw);
289 			local_bh_enable();
290 			goto restart_rcu;
291 		}
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();
299 	}
300 }
301 EXPORT_SYMBOL_GPL(inet_twsk_purge);
302