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 | __sk_del_bind_node(sk: (struct sock *)tw); |
39 | tw->tw_tb = NULL; |
40 | tw->tw_tb2 = NULL; |
41 | inet_bind2_bucket_destroy(cachep: hashinfo->bind2_bucket_cachep, tb: tb2); |
42 | inet_bind_bucket_destroy(cachep: hashinfo->bind_bucket_cachep, tb); |
43 | |
44 | __sock_put(sk: (struct sock *)tw); |
45 | } |
46 | |
47 | /* Must be called with locally disabled BHs. */ |
48 | static void inet_twsk_kill(struct inet_timewait_sock *tw) |
49 | { |
50 | struct inet_hashinfo *hashinfo = tw->tw_dr->hashinfo; |
51 | spinlock_t *lock = inet_ehash_lockp(hashinfo, hash: tw->tw_hash); |
52 | struct inet_bind_hashbucket *bhead, *bhead2; |
53 | |
54 | spin_lock(lock); |
55 | sk_nulls_del_node_init_rcu(sk: (struct sock *)tw); |
56 | spin_unlock(lock); |
57 | |
58 | /* Disassociate with bind bucket. */ |
59 | bhead = &hashinfo->bhash[inet_bhashfn(net: twsk_net(twsk: tw), lport: tw->tw_num, |
60 | bhash_size: hashinfo->bhash_size)]; |
61 | bhead2 = inet_bhashfn_portaddr(hinfo: hashinfo, sk: (struct sock *)tw, |
62 | net: twsk_net(twsk: tw), port: tw->tw_num); |
63 | |
64 | spin_lock(lock: &bhead->lock); |
65 | spin_lock(lock: &bhead2->lock); |
66 | inet_twsk_bind_unhash(tw, hashinfo); |
67 | spin_unlock(lock: &bhead2->lock); |
68 | spin_unlock(lock: &bhead->lock); |
69 | |
70 | refcount_dec(r: &tw->tw_dr->tw_refcount); |
71 | inet_twsk_put(tw); |
72 | } |
73 | |
74 | void inet_twsk_free(struct inet_timewait_sock *tw) |
75 | { |
76 | struct module *owner = tw->tw_prot->owner; |
77 | twsk_destructor(sk: (struct sock *)tw); |
78 | kmem_cache_free(s: tw->tw_prot->twsk_prot->twsk_slab, objp: tw); |
79 | module_put(module: owner); |
80 | } |
81 | |
82 | void inet_twsk_put(struct inet_timewait_sock *tw) |
83 | { |
84 | if (refcount_dec_and_test(r: &tw->tw_refcnt)) |
85 | inet_twsk_free(tw); |
86 | } |
87 | EXPORT_SYMBOL_GPL(inet_twsk_put); |
88 | |
89 | static void inet_twsk_add_node_rcu(struct inet_timewait_sock *tw, |
90 | struct hlist_nulls_head *list) |
91 | { |
92 | hlist_nulls_add_head_rcu(n: &tw->tw_node, h: 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, hash: sk->sk_hash); |
106 | spinlock_t *lock = inet_ehash_lockp(hashinfo, hash: sk->sk_hash); |
107 | struct inet_bind_hashbucket *bhead, *bhead2; |
108 | |
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(net: twsk_net(twsk: tw), lport: inet->inet_num, |
114 | bhash_size: hashinfo->bhash_size)]; |
115 | bhead2 = inet_bhashfn_portaddr(hinfo: hashinfo, sk, net: twsk_net(twsk: tw), port: inet->inet_num); |
116 | |
117 | spin_lock(lock: &bhead->lock); |
118 | spin_lock(lock: &bhead2->lock); |
119 | |
120 | tw->tw_tb = icsk->icsk_bind_hash; |
121 | WARN_ON(!icsk->icsk_bind_hash); |
122 | |
123 | tw->tw_tb2 = icsk->icsk_bind2_hash; |
124 | WARN_ON(!icsk->icsk_bind2_hash); |
125 | sk_add_bind_node(sk: (struct sock *)tw, list: &tw->tw_tb2->owners); |
126 | |
127 | spin_unlock(lock: &bhead2->lock); |
128 | spin_unlock(lock: &bhead->lock); |
129 | |
130 | spin_lock(lock); |
131 | |
132 | inet_twsk_add_node_rcu(tw, list: &ehead->chain); |
133 | |
134 | /* Step 3: Remove SK from hash chain */ |
135 | if (__sk_nulls_del_node_init_rcu(sk)) |
136 | sock_prot_inuse_add(net: sock_net(sk), prot: sk->sk_prot, val: -1); |
137 | |
138 | spin_unlock(lock); |
139 | |
140 | /* tw_refcnt is set to 3 because we have : |
141 | * - one reference for bhash chain. |
142 | * - one reference for ehash chain. |
143 | * - one reference for timer. |
144 | * We can use atomic_set() because prior spin_lock()/spin_unlock() |
145 | * committed into memory all tw fields. |
146 | * Also note that after this point, we lost our implicit reference |
147 | * so we are not allowed to use tw anymore. |
148 | */ |
149 | refcount_set(r: &tw->tw_refcnt, n: 3); |
150 | } |
151 | EXPORT_SYMBOL_GPL(inet_twsk_hashdance); |
152 | |
153 | static void tw_timer_handler(struct timer_list *t) |
154 | { |
155 | struct inet_timewait_sock *tw = from_timer(tw, t, tw_timer); |
156 | |
157 | inet_twsk_kill(tw); |
158 | } |
159 | |
160 | struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, |
161 | struct inet_timewait_death_row *dr, |
162 | const int state) |
163 | { |
164 | struct inet_timewait_sock *tw; |
165 | |
166 | if (refcount_read(r: &dr->tw_refcount) - 1 >= |
167 | READ_ONCE(dr->sysctl_max_tw_buckets)) |
168 | return NULL; |
169 | |
170 | tw = kmem_cache_alloc(cachep: sk->sk_prot_creator->twsk_prot->twsk_slab, |
171 | GFP_ATOMIC); |
172 | if (tw) { |
173 | const struct inet_sock *inet = inet_sk(sk); |
174 | |
175 | tw->tw_dr = dr; |
176 | /* Give us an identity. */ |
177 | tw->tw_daddr = inet->inet_daddr; |
178 | tw->tw_rcv_saddr = inet->inet_rcv_saddr; |
179 | tw->tw_bound_dev_if = sk->sk_bound_dev_if; |
180 | tw->tw_tos = inet->tos; |
181 | tw->tw_num = inet->inet_num; |
182 | tw->tw_state = TCP_TIME_WAIT; |
183 | tw->tw_substate = state; |
184 | tw->tw_sport = inet->inet_sport; |
185 | tw->tw_dport = inet->inet_dport; |
186 | tw->tw_family = sk->sk_family; |
187 | tw->tw_reuse = sk->sk_reuse; |
188 | tw->tw_reuseport = sk->sk_reuseport; |
189 | tw->tw_hash = sk->sk_hash; |
190 | tw->tw_ipv6only = 0; |
191 | tw->tw_transparent = inet_test_bit(TRANSPARENT, sk); |
192 | tw->tw_prot = sk->sk_prot_creator; |
193 | atomic64_set(v: &tw->tw_cookie, i: atomic64_read(v: &sk->sk_cookie)); |
194 | twsk_net_set(twsk: tw, net: sock_net(sk)); |
195 | timer_setup(&tw->tw_timer, tw_timer_handler, TIMER_PINNED); |
196 | /* |
197 | * Because we use RCU lookups, we should not set tw_refcnt |
198 | * to a non null value before everything is setup for this |
199 | * timewait socket. |
200 | */ |
201 | refcount_set(r: &tw->tw_refcnt, n: 0); |
202 | |
203 | __module_get(module: tw->tw_prot->owner); |
204 | } |
205 | |
206 | return tw; |
207 | } |
208 | EXPORT_SYMBOL_GPL(inet_twsk_alloc); |
209 | |
210 | /* These are always called from BH context. See callers in |
211 | * tcp_input.c to verify this. |
212 | */ |
213 | |
214 | /* This is for handling early-kills of TIME_WAIT sockets. |
215 | * Warning : consume reference. |
216 | * Caller should not access tw anymore. |
217 | */ |
218 | void inet_twsk_deschedule_put(struct inet_timewait_sock *tw) |
219 | { |
220 | if (del_timer_sync(timer: &tw->tw_timer)) |
221 | inet_twsk_kill(tw); |
222 | inet_twsk_put(tw); |
223 | } |
224 | EXPORT_SYMBOL(inet_twsk_deschedule_put); |
225 | |
226 | void __inet_twsk_schedule(struct inet_timewait_sock *tw, int timeo, bool rearm) |
227 | { |
228 | /* timeout := RTO * 3.5 |
229 | * |
230 | * 3.5 = 1+2+0.5 to wait for two retransmits. |
231 | * |
232 | * RATIONALE: if FIN arrived and we entered TIME-WAIT state, |
233 | * our ACK acking that FIN can be lost. If N subsequent retransmitted |
234 | * FINs (or previous seqments) are lost (probability of such event |
235 | * is p^(N+1), where p is probability to lose single packet and |
236 | * time to detect the loss is about RTO*(2^N - 1) with exponential |
237 | * backoff). Normal timewait length is calculated so, that we |
238 | * waited at least for one retransmitted FIN (maximal RTO is 120sec). |
239 | * [ BTW Linux. following BSD, violates this requirement waiting |
240 | * only for 60sec, we should wait at least for 240 secs. |
241 | * Well, 240 consumes too much of resources 8) |
242 | * ] |
243 | * This interval is not reduced to catch old duplicate and |
244 | * responces to our wandering segments living for two MSLs. |
245 | * However, if we use PAWS to detect |
246 | * old duplicates, we can reduce the interval to bounds required |
247 | * by RTO, rather than MSL. So, if peer understands PAWS, we |
248 | * kill tw bucket after 3.5*RTO (it is important that this number |
249 | * is greater than TS tick!) and detect old duplicates with help |
250 | * of PAWS. |
251 | */ |
252 | |
253 | if (!rearm) { |
254 | bool kill = timeo <= 4*HZ; |
255 | |
256 | __NET_INC_STATS(twsk_net(tw), kill ? LINUX_MIB_TIMEWAITKILLED : |
257 | LINUX_MIB_TIMEWAITED); |
258 | BUG_ON(mod_timer(&tw->tw_timer, jiffies + timeo)); |
259 | refcount_inc(r: &tw->tw_dr->tw_refcount); |
260 | } else { |
261 | mod_timer_pending(timer: &tw->tw_timer, expires: jiffies + timeo); |
262 | } |
263 | } |
264 | EXPORT_SYMBOL_GPL(__inet_twsk_schedule); |
265 | |
266 | /* Remove all non full sockets (TIME_WAIT and NEW_SYN_RECV) for dead netns */ |
267 | void inet_twsk_purge(struct inet_hashinfo *hashinfo, int family) |
268 | { |
269 | struct hlist_nulls_node *node; |
270 | unsigned int slot; |
271 | struct sock *sk; |
272 | |
273 | for (slot = 0; slot <= hashinfo->ehash_mask; slot++) { |
274 | struct inet_ehash_bucket *head = &hashinfo->ehash[slot]; |
275 | restart_rcu: |
276 | cond_resched(); |
277 | rcu_read_lock(); |
278 | restart: |
279 | sk_nulls_for_each_rcu(sk, node, &head->chain) { |
280 | int state = inet_sk_state_load(sk); |
281 | |
282 | if ((1 << state) & ~(TCPF_TIME_WAIT | |
283 | TCPF_NEW_SYN_RECV)) |
284 | continue; |
285 | |
286 | if (sk->sk_family != family || |
287 | refcount_read(r: &sock_net(sk)->ns.count)) |
288 | continue; |
289 | |
290 | if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) |
291 | continue; |
292 | |
293 | if (unlikely(sk->sk_family != family || |
294 | refcount_read(&sock_net(sk)->ns.count))) { |
295 | sock_gen_put(sk); |
296 | goto restart; |
297 | } |
298 | |
299 | rcu_read_unlock(); |
300 | local_bh_disable(); |
301 | if (state == TCP_TIME_WAIT) { |
302 | inet_twsk_deschedule_put(inet_twsk(sk)); |
303 | } else { |
304 | struct request_sock *req = inet_reqsk(sk); |
305 | |
306 | inet_csk_reqsk_queue_drop_and_put(sk: req->rsk_listener, |
307 | req); |
308 | } |
309 | local_bh_enable(); |
310 | goto restart_rcu; |
311 | } |
312 | /* If the nulls value we got at the end of this lookup is |
313 | * not the expected one, we must restart lookup. |
314 | * We probably met an item that was moved to another chain. |
315 | */ |
316 | if (get_nulls_value(ptr: node) != slot) |
317 | goto restart; |
318 | rcu_read_unlock(); |
319 | } |
320 | } |
321 | EXPORT_SYMBOL_GPL(inet_twsk_purge); |
322 | |