xref: /openbmc/linux/net/ipv4/inet_timewait_sock.c (revision 86e281fc)
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_bind2_bucket *tb2 = tw->tw_tb2;
33 	struct inet_bind_bucket *tb = tw->tw_tb;
34 
35 	if (!tb)
36 		return;
37 
38 	__hlist_del(&tw->tw_bind_node);
39 	tw->tw_tb = NULL;
40 	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
41 
42 	__hlist_del(&tw->tw_bind2_node);
43 	tw->tw_tb2 = NULL;
44 	inet_bind2_bucket_destroy(hashinfo->bind2_bucket_cachep, tb2);
45 
46 	__sock_put((struct sock *)tw);
47 }
48 
49 /* Must be called with locally disabled BHs. */
50 static void inet_twsk_kill(struct inet_timewait_sock *tw)
51 {
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;
55 
56 	spin_lock(lock);
57 	sk_nulls_del_node_init_rcu((struct sock *)tw);
58 	spin_unlock(lock);
59 
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);
65 
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);
71 
72 	refcount_dec(&tw->tw_dr->tw_refcount);
73 	inet_twsk_put(tw);
74 }
75 
76 void inet_twsk_free(struct inet_timewait_sock *tw)
77 {
78 	struct module *owner = tw->tw_prot->owner;
79 	twsk_destructor((struct sock *)tw);
80 	kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
81 	module_put(owner);
82 }
83 
84 void inet_twsk_put(struct inet_timewait_sock *tw)
85 {
86 	if (refcount_dec_and_test(&tw->tw_refcnt))
87 		inet_twsk_free(tw);
88 }
89 EXPORT_SYMBOL_GPL(inet_twsk_put);
90 
91 static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
92 				   struct hlist_nulls_head *list)
93 {
94 	hlist_nulls_add_head_rcu(&tw->tw_node, list);
95 }
96 
97 static void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
98 				    struct hlist_head *list)
99 {
100 	hlist_add_head(&tw->tw_bind_node, list);
101 }
102 
103 static void inet_twsk_add_bind2_node(struct inet_timewait_sock *tw,
104 				     struct hlist_head *list)
105 {
106 	hlist_add_head(&tw->tw_bind2_node, list);
107 }
108 
109 /*
110  * Enter the time wait state. This is called with locally disabled BH.
111  * Essentially we whip up a timewait bucket, copy the relevant info into it
112  * from the SK, and mess with hash chains and list linkage.
113  */
114 void inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
115 			   struct inet_hashinfo *hashinfo)
116 {
117 	const struct inet_sock *inet = inet_sk(sk);
118 	const struct inet_connection_sock *icsk = inet_csk(sk);
119 	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
120 	spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
121 	struct inet_bind_hashbucket *bhead, *bhead2;
122 
123 	/* Step 1: Put TW into bind hash. Original socket stays there too.
124 	   Note, that any socket with inet->num != 0 MUST be bound in
125 	   binding cache, even if it is closed.
126 	 */
127 	bhead = &hashinfo->bhash[inet_bhashfn(twsk_net(tw), inet->inet_num,
128 			hashinfo->bhash_size)];
129 	bhead2 = inet_bhashfn_portaddr(hashinfo, sk, twsk_net(tw), inet->inet_num);
130 
131 	spin_lock(&bhead->lock);
132 	spin_lock(&bhead2->lock);
133 
134 	tw->tw_tb = icsk->icsk_bind_hash;
135 	WARN_ON(!icsk->icsk_bind_hash);
136 	inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
137 
138 	tw->tw_tb2 = icsk->icsk_bind2_hash;
139 	WARN_ON(!icsk->icsk_bind2_hash);
140 	inet_twsk_add_bind2_node(tw, &tw->tw_tb2->deathrow);
141 
142 	spin_unlock(&bhead2->lock);
143 	spin_unlock(&bhead->lock);
144 
145 	spin_lock(lock);
146 
147 	inet_twsk_add_node_rcu(tw, &ehead->chain);
148 
149 	/* Step 3: Remove SK from hash chain */
150 	if (__sk_nulls_del_node_init_rcu(sk))
151 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
152 
153 	spin_unlock(lock);
154 
155 	/* tw_refcnt is set to 3 because we have :
156 	 * - one reference for bhash chain.
157 	 * - one reference for ehash chain.
158 	 * - one reference for timer.
159 	 * We can use atomic_set() because prior spin_lock()/spin_unlock()
160 	 * committed into memory all tw fields.
161 	 * Also note that after this point, we lost our implicit reference
162 	 * so we are not allowed to use tw anymore.
163 	 */
164 	refcount_set(&tw->tw_refcnt, 3);
165 }
166 EXPORT_SYMBOL_GPL(inet_twsk_hashdance);
167 
168 static void tw_timer_handler(struct timer_list *t)
169 {
170 	struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer);
171 
172 	inet_twsk_kill(tw);
173 }
174 
175 struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk,
176 					   struct inet_timewait_death_row *dr,
177 					   const int state)
178 {
179 	struct inet_timewait_sock *tw;
180 
181 	if (refcount_read(&dr->tw_refcount) - 1 >=
182 	    READ_ONCE(dr->sysctl_max_tw_buckets))
183 		return NULL;
184 
185 	tw = kmem_cache_alloc(sk->sk_prot_creator->twsk_prot->twsk_slab,
186 			      GFP_ATOMIC);
187 	if (tw) {
188 		const struct inet_sock *inet = inet_sk(sk);
189 
190 		tw->tw_dr	    = dr;
191 		/* Give us an identity. */
192 		tw->tw_daddr	    = inet->inet_daddr;
193 		tw->tw_rcv_saddr    = inet->inet_rcv_saddr;
194 		tw->tw_bound_dev_if = sk->sk_bound_dev_if;
195 		tw->tw_tos	    = inet->tos;
196 		tw->tw_num	    = inet->inet_num;
197 		tw->tw_state	    = TCP_TIME_WAIT;
198 		tw->tw_substate	    = state;
199 		tw->tw_sport	    = inet->inet_sport;
200 		tw->tw_dport	    = inet->inet_dport;
201 		tw->tw_family	    = sk->sk_family;
202 		tw->tw_reuse	    = sk->sk_reuse;
203 		tw->tw_reuseport    = sk->sk_reuseport;
204 		tw->tw_hash	    = sk->sk_hash;
205 		tw->tw_ipv6only	    = 0;
206 		tw->tw_transparent  = inet_test_bit(TRANSPARENT, sk);
207 		tw->tw_prot	    = sk->sk_prot_creator;
208 		atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
209 		twsk_net_set(tw, sock_net(sk));
210 		timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED);
211 		/*
212 		 * Because we use RCU lookups, we should not set tw_refcnt
213 		 * to a non null value before everything is setup for this
214 		 * timewait socket.
215 		 */
216 		refcount_set(&tw->tw_refcnt, 0);
217 
218 		__module_get(tw->tw_prot->owner);
219 	}
220 
221 	return tw;
222 }
223 EXPORT_SYMBOL_GPL(inet_twsk_alloc);
224 
225 /* These are always called from BH context.  See callers in
226  * tcp_input.c to verify this.
227  */
228 
229 /* This is for handling early-kills of TIME_WAIT sockets.
230  * Warning : consume reference.
231  * Caller should not access tw anymore.
232  */
233 void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
234 {
235 	if (del_timer_sync(&tw->tw_timer))
236 		inet_twsk_kill(tw);
237 	inet_twsk_put(tw);
238 }
239 EXPORT_SYMBOL(inet_twsk_deschedule_put);
240 
241 void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
242 {
243 	/* timeout := RTO * 3.5
244 	 *
245 	 * 3.5 = 1+2+0.5 to wait for two retransmits.
246 	 *
247 	 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
248 	 * our ACK acking that FIN can be lost. If N subsequent retransmitted
249 	 * FINs (or previous seqments) are lost (probability of such event
250 	 * is p^(N+1), where p is probability to lose single packet and
251 	 * time to detect the loss is about RTO*(2^N - 1) with exponential
252 	 * backoff). Normal timewait length is calculated so, that we
253 	 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
254 	 * [ BTW Linux. following BSD, violates this requirement waiting
255 	 *   only for 60sec, we should wait at least for 240 secs.
256 	 *   Well, 240 consumes too much of resources 8)
257 	 * ]
258 	 * This interval is not reduced to catch old duplicate and
259 	 * responces to our wandering segments living for two MSLs.
260 	 * However, if we use PAWS to detect
261 	 * old duplicates, we can reduce the interval to bounds required
262 	 * by RTO, rather than MSL. So, if peer understands PAWS, we
263 	 * kill tw bucket after 3.5*RTO (it is important that this number
264 	 * is greater than TS tick!) and detect old duplicates with help
265 	 * of PAWS.
266 	 */
267 
268 	if (!rearm) {
269 		bool kill = timeo <= 4*HZ;
270 
271 		__NET_INC_STATS(twsk_net(tw), kill ? LINUX_MIB_TIMEWAITKILLED :
272 						     LINUX_MIB_TIMEWAITED);
273 		BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
274 		refcount_inc(&tw->tw_dr->tw_refcount);
275 	} else {
276 		mod_timer_pending(&tw->tw_timer, jiffies + timeo);
277 	}
278 }
279 EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
280 
281 void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family)
282 {
283 	struct inet_timewait_sock *tw;
284 	struct sock *sk;
285 	struct hlist_nulls_node *node;
286 	unsigned int slot;
287 
288 	for (slot = 0; slot <= hashinfo->ehash_mask; slot++) {
289 		struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
290 restart_rcu:
291 		cond_resched();
292 		rcu_read_lock();
293 restart:
294 		sk_nulls_for_each_rcu(sk, node, &head->chain) {
295 			if (sk->sk_state != TCP_TIME_WAIT) {
296 				/* A kernel listener socket might not hold refcnt for net,
297 				 * so reqsk_timer_handler() could be fired after net is
298 				 * freed.  Userspace listener and reqsk never exist here.
299 				 */
300 				if (unlikely(sk->sk_state == TCP_NEW_SYN_RECV &&
301 					     hashinfo->pernet)) {
302 					struct request_sock *req = inet_reqsk(sk);
303 
304 					inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, req);
305 				}
306 
307 				continue;
308 			}
309 
310 			tw = inet_twsk(sk);
311 			if ((tw->tw_family != family) ||
312 				refcount_read(&twsk_net(tw)->ns.count))
313 				continue;
314 
315 			if (unlikely(!refcount_inc_not_zero(&tw->tw_refcnt)))
316 				continue;
317 
318 			if (unlikely((tw->tw_family != family) ||
319 				     refcount_read(&twsk_net(tw)->ns.count))) {
320 				inet_twsk_put(tw);
321 				goto restart;
322 			}
323 
324 			rcu_read_unlock();
325 			local_bh_disable();
326 			inet_twsk_deschedule_put(tw);
327 			local_bh_enable();
328 			goto restart_rcu;
329 		}
330 		/* If the nulls value we got at the end of this lookup is
331 		 * not the expected one, we must restart lookup.
332 		 * We probably met an item that was moved to another chain.
333 		 */
334 		if (get_nulls_value(node) != slot)
335 			goto restart;
336 		rcu_read_unlock();
337 	}
338 }
339 EXPORT_SYMBOL_GPL(inet_twsk_purge);
340