xref: /openbmc/linux/net/ipv4/inet_timewait_sock.c (revision 7a2f6f61)
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[tw->tw_bslot];
56 
57 	spin_lock(&bhead->lock);
58 	inet_twsk_bind_unhash(tw, hashinfo);
59 	spin_unlock(&bhead->lock);
60 
61 	if (refcount_dec_and_test(&tw->tw_dr->tw_refcount))
62 		kfree(tw->tw_dr);
63 
64 	inet_twsk_put(tw);
65 }
66 
67 void inet_twsk_free(struct inet_timewait_sock *tw)
68 {
69 	struct module *owner = tw->tw_prot->owner;
70 	twsk_destructor((struct sock *)tw);
71 #ifdef SOCK_REFCNT_DEBUG
72 	pr_debug("%s timewait_sock %p released\n", tw->tw_prot->name, tw);
73 #endif
74 	kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw);
75 	module_put(owner);
76 }
77 
78 void inet_twsk_put(struct inet_timewait_sock *tw)
79 {
80 	if (refcount_dec_and_test(&tw->tw_refcnt))
81 		inet_twsk_free(tw);
82 }
83 EXPORT_SYMBOL_GPL(inet_twsk_put);
84 
85 static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw,
86 				   struct hlist_nulls_head *list)
87 {
88 	hlist_nulls_add_head_rcu(&tw->tw_node, list);
89 }
90 
91 static void inet_twsk_add_bind_node(struct inet_timewait_sock *tw,
92 				    struct hlist_head *list)
93 {
94 	hlist_add_head(&tw->tw_bind_node, list);
95 }
96 
97 /*
98  * Enter the time wait state. This is called with locally disabled BH.
99  * Essentially we whip up a timewait bucket, copy the relevant info into it
100  * from the SK, and mess with hash chains and list linkage.
101  */
102 void inet_twsk_hashdance(struct inet_timewait_sock *tw, struct sock *sk,
103 			   struct inet_hashinfo *hashinfo)
104 {
105 	const struct inet_sock *inet = inet_sk(sk);
106 	const struct inet_connection_sock *icsk = inet_csk(sk);
107 	struct inet_ehash_bucket *ehead = inet_ehash_bucket(hashinfo, sk->sk_hash);
108 	spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
109 	struct inet_bind_hashbucket *bhead;
110 	/* Step 1: Put TW into bind hash. Original socket stays there too.
111 	   Note, that any socket with inet->num != 0 MUST be bound in
112 	   binding cache, even if it is closed.
113 	 */
114 	/* Cache inet_bhashfn(), because 'struct net' might be no longer
115 	 * available later in inet_twsk_kill().
116 	 */
117 	tw->tw_bslot = inet_bhashfn(twsk_net(tw), inet->inet_num,
118 				    hashinfo->bhash_size);
119 	bhead = &hashinfo->bhash[tw->tw_bslot];
120 	spin_lock(&bhead->lock);
121 	tw->tw_tb = icsk->icsk_bind_hash;
122 	WARN_ON(!icsk->icsk_bind_hash);
123 	inet_twsk_add_bind_node(tw, &tw->tw_tb->owners);
124 	spin_unlock(&bhead->lock);
125 
126 	spin_lock(lock);
127 
128 	inet_twsk_add_node_rcu(tw, &ehead->chain);
129 
130 	/* Step 3: Remove SK from hash chain */
131 	if (__sk_nulls_del_node_init_rcu(sk))
132 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
133 
134 	spin_unlock(lock);
135 
136 	/* tw_refcnt is set to 3 because we have :
137 	 * - one reference for bhash chain.
138 	 * - one reference for ehash chain.
139 	 * - one reference for timer.
140 	 * We can use atomic_set() because prior spin_lock()/spin_unlock()
141 	 * committed into memory all tw fields.
142 	 * Also note that after this point, we lost our implicit reference
143 	 * so we are not allowed to use tw anymore.
144 	 */
145 	refcount_set(&tw->tw_refcnt, 3);
146 }
147 EXPORT_SYMBOL_GPL(inet_twsk_hashdance);
148 
149 static void tw_timer_handler(struct timer_list *t)
150 {
151 	struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer);
152 
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 (refcount_read(&dr->tw_refcount) - 1 >= 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 		tw->tw_dr	    = dr;
171 		/* Give us an identity. */
172 		tw->tw_daddr	    = inet->inet_daddr;
173 		tw->tw_rcv_saddr    = inet->inet_rcv_saddr;
174 		tw->tw_bound_dev_if = sk->sk_bound_dev_if;
175 		tw->tw_tos	    = inet->tos;
176 		tw->tw_num	    = inet->inet_num;
177 		tw->tw_state	    = TCP_TIME_WAIT;
178 		tw->tw_substate	    = state;
179 		tw->tw_sport	    = inet->inet_sport;
180 		tw->tw_dport	    = inet->inet_dport;
181 		tw->tw_family	    = sk->sk_family;
182 		tw->tw_reuse	    = sk->sk_reuse;
183 		tw->tw_reuseport    = sk->sk_reuseport;
184 		tw->tw_hash	    = sk->sk_hash;
185 		tw->tw_ipv6only	    = 0;
186 		tw->tw_transparent  = inet->transparent;
187 		tw->tw_prot	    = sk->sk_prot_creator;
188 		atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie));
189 		twsk_net_set(tw, sock_net(sk));
190 		timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED);
191 		/*
192 		 * Because we use RCU lookups, we should not set tw_refcnt
193 		 * to a non null value before everything is setup for this
194 		 * timewait socket.
195 		 */
196 		refcount_set(&tw->tw_refcnt, 0);
197 
198 		__module_get(tw->tw_prot->owner);
199 	}
200 
201 	return tw;
202 }
203 EXPORT_SYMBOL_GPL(inet_twsk_alloc);
204 
205 /* These are always called from BH context.  See callers in
206  * tcp_input.c to verify this.
207  */
208 
209 /* This is for handling early-kills of TIME_WAIT sockets.
210  * Warning : consume reference.
211  * Caller should not access tw anymore.
212  */
213 void inet_twsk_deschedule_put(struct inet_timewait_sock *tw)
214 {
215 	if (del_timer_sync(&tw->tw_timer))
216 		inet_twsk_kill(tw);
217 	inet_twsk_put(tw);
218 }
219 EXPORT_SYMBOL(inet_twsk_deschedule_put);
220 
221 void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm)
222 {
223 	/* timeout := RTO * 3.5
224 	 *
225 	 * 3.5 = 1+2+0.5 to wait for two retransmits.
226 	 *
227 	 * RATIONALE: if FIN arrived and we entered TIME-WAIT state,
228 	 * our ACK acking that FIN can be lost. If N subsequent retransmitted
229 	 * FINs (or previous seqments) are lost (probability of such event
230 	 * is p^(N+1), where p is probability to lose single packet and
231 	 * time to detect the loss is about RTO*(2^N - 1) with exponential
232 	 * backoff). Normal timewait length is calculated so, that we
233 	 * waited at least for one retransmitted FIN (maximal RTO is 120sec).
234 	 * [ BTW Linux. following BSD, violates this requirement waiting
235 	 *   only for 60sec, we should wait at least for 240 secs.
236 	 *   Well, 240 consumes too much of resources 8)
237 	 * ]
238 	 * This interval is not reduced to catch old duplicate and
239 	 * responces to our wandering segments living for two MSLs.
240 	 * However, if we use PAWS to detect
241 	 * old duplicates, we can reduce the interval to bounds required
242 	 * by RTO, rather than MSL. So, if peer understands PAWS, we
243 	 * kill tw bucket after 3.5*RTO (it is important that this number
244 	 * is greater than TS tick!) and detect old duplicates with help
245 	 * of PAWS.
246 	 */
247 
248 	if (!rearm) {
249 		bool kill = timeo <= 4*HZ;
250 
251 		__NET_INC_STATS(twsk_net(tw), kill ? LINUX_MIB_TIMEWAITKILLED :
252 						     LINUX_MIB_TIMEWAITED);
253 		BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo));
254 		refcount_inc(&tw->tw_dr->tw_refcount);
255 	} else {
256 		mod_timer_pending(&tw->tw_timer, jiffies + timeo);
257 	}
258 }
259 EXPORT_SYMBOL_GPL(__inet_twsk_schedule);
260