xref: /openbmc/linux/net/ipv4/inet_timewait_sock.c (revision 9cfc5c90)
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/kmemcheck.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 	atomic_dec(&tw->tw_dr->tw_count);
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 (atomic_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 	/*
124 	 * Step 2: Hash TW into tcp ehash chain.
125 	 * Notes :
126 	 * - tw_refcnt is set to 4 because :
127 	 * - We have one reference from bhash chain.
128 	 * - We have one reference from ehash chain.
129 	 * - We have one reference from timer.
130 	 * - One reference for ourself (our caller will release it).
131 	 * We can use atomic_set() because prior spin_lock()/spin_unlock()
132 	 * committed into memory all tw fields.
133 	 */
134 	atomic_set(&tw->tw_refcnt, 4);
135 	inet_twsk_add_node_rcu(tw, &ehead->chain);
136 
137 	/* Step 3: Remove SK from hash chain */
138 	if (__sk_nulls_del_node_init_rcu(sk))
139 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
140 
141 	spin_unlock(lock);
142 }
143 EXPORT_SYMBOL_GPL(__inet_twsk_hashdance);
144 
145 static void tw_timer_handler(unsigned long data)
146 {
147 	struct inet_timewait_sock *tw = (struct inet_timewait_sock *)data;
148 
149 	if (tw->tw_kill)
150 		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED);
151 	else
152 		NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITED);
153 	inet_twsk_kill(tw);
154 }
155 
156 struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
157 					   struct inet_timewait_death_row *dr,
158 					   const int state)
159 {
160 	struct inet_timewait_sock *tw;
161 
162 	if (atomic_read(&dr->tw_count) >= dr->sysctl_max_tw_buckets)
163 		return NULL;
164 
165 	tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
166 			      GFP_ATOMIC);
167 	if (tw) {
168 		const struct inet_sock *inet = inet_sk(sk);
169 
170 		kmemcheck_annotate_bitfield(tw, flags);
171 
172 		tw->tw_dr	    = dr;
173 		/* Give us an identity. */
174 		tw->tw_daddr	    = inet->inet_daddr;
175 		tw->tw_rcv_saddr    = inet->inet_rcv_saddr;
176 		tw->tw_bound_dev_if = sk->sk_bound_dev_if;
177 		tw->tw_tos	    = inet->tos;
178 		tw->tw_num	    = inet->inet_num;
179 		tw->tw_state	    = TCP_TIME_WAIT;
180 		tw->tw_substate	    = state;
181 		tw->tw_sport	    = inet->inet_sport;
182 		tw->tw_dport	    = inet->inet_dport;
183 		tw->tw_family	    = sk->sk_family;
184 		tw->tw_reuse	    = sk->sk_reuse;
185 		tw->tw_hash	    = sk->sk_hash;
186 		tw->tw_ipv6only	    = 0;
187 		tw->tw_transparent  = inet->transparent;
188 		tw->tw_prot	    = sk->sk_prot_creator;
189 		atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
190 		twsk_net_set(tw, sock_net(sk));
191 		setup_timer(&tw->tw_timer, tw_timer_handler, (unsigned long)tw);
192 		/*
193 		 * Because we use RCU lookups, we should not set tw_refcnt
194 		 * to a non null value before everything is setup for this
195 		 * timewait socket.
196 		 */
197 		atomic_set(&tw->tw_refcnt, 0);
198 
199 		__module_get(tw->tw_prot->owner);
200 	}
201 
202 	return tw;
203 }
204 EXPORT_SYMBOL_GPL(inet_twsk_alloc);
205 
206 /* These are always called from BH context.  See callers in
207  * tcp_input.c to verify this.
208  */
209 
210 /* This is for handling early-kills of TIME_WAIT sockets.
211  * Warning : consume reference.
212  * Caller should not access tw anymore.
213  */
214 void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
215 {
216 	if (del_timer_sync(&tw->tw_timer))
217 		inet_twsk_kill(tw);
218 	inet_twsk_put(tw);
219 }
220 EXPORT_SYMBOL(inet_twsk_deschedule_put);
221 
222 void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
223 {
224 	/* timeout := RTO * 3.5
225 	 *
226 	 * 3.5 = 1+2+0.5 to wait for two retransmits.
227 	 *
228 	 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
229 	 * our ACK acking that FIN can be lost. If N subsequent retransmitted
230 	 * FINs (or previous seqments) are lost (probability of such event
231 	 * is p^(N+1), where p is probability to lose single packet and
232 	 * time to detect the loss is about RTO*(2^N - 1) with exponential
233 	 * backoff). Normal timewait length is calculated so, that we
234 	 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
235 	 * [ BTW Linux. following BSD, violates this requirement waiting
236 	 *   only for 60sec, we should wait at least for 240 secs.
237 	 *   Well, 240 consumes too much of resources 8)
238 	 * ]
239 	 * This interval is not reduced to catch old duplicate and
240 	 * responces to our wandering segments living for two MSLs.
241 	 * However, if we use PAWS to detect
242 	 * old duplicates, we can reduce the interval to bounds required
243 	 * by RTO, rather than MSL. So, if peer understands PAWS, we
244 	 * kill tw bucket after 3.5*RTO (it is important that this number
245 	 * is greater than TS tick!) and detect old duplicates with help
246 	 * of PAWS.
247 	 */
248 
249 	tw->tw_kill = timeo <= 4*HZ;
250 	if (!rearm) {
251 		BUG_ON(mod_timer_pinned(&tw->tw_timer, jiffies + timeo));
252 		atomic_inc(&tw->tw_dr->tw_count);
253 	} else {
254 		mod_timer_pending(&tw->tw_timer, jiffies + timeo);
255 	}
256 }
257 EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
258 
259 void inet_twsk_purge(struct inet_hashinfo *hashinfo,
260 		     struct inet_timewait_death_row *twdr, int family)
261 {
262 	struct inet_timewait_sock *tw;
263 	struct sock *sk;
264 	struct hlist_nulls_node *node;
265 	unsigned int slot;
266 
267 	for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
268 		struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
269 restart_rcu:
270 		cond_resched();
271 		rcu_read_lock();
272 restart:
273 		sk_nulls_for_each_rcu(sk, node, &head->chain) {
274 			if (sk->sk_state != TCP_TIME_WAIT)
275 				continue;
276 			tw = inet_twsk(sk);
277 			if ((tw->tw_family != family) ||
278 				atomic_read(&twsk_net(tw)->count))
279 				continue;
280 
281 			if (unlikely(!atomic_inc_not_zero(&tw->tw_refcnt)))
282 				continue;
283 
284 			if (unlikely((tw->tw_family != family) ||
285 				     atomic_read(&twsk_net(tw)->count))) {
286 				inet_twsk_put(tw);
287 				goto restart;
288 			}
289 
290 			rcu_read_unlock();
291 			local_bh_disable();
292 			inet_twsk_deschedule_put(tw);
293 			local_bh_enable();
294 			goto restart_rcu;
295 		}
296 		/* If the nulls value we got at the end of this lookup is
297 		 * not the expected one, we must restart lookup.
298 		 * We probably met an item that was moved to another chain.
299 		 */
300 		if (get_nulls_value(node) != slot)
301 			goto restart;
302 		rcu_read_unlock();
303 	}
304 }
305 EXPORT_SYMBOL_GPL(inet_twsk_purge);
306