| 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 3 because : |
| 127 | * - We have one reference from bhash chain. |
| 128 | * - We have one reference from ehash chain. |
| 129 | * We can use atomic_set() because prior spin_lock()/spin_unlock() |
| 130 | * committed into memory all tw fields. |
| 131 | */ |
| 132 | atomic_set(&tw->tw_refcnt, 1 + 1 + 1); |
| 133 | inet_twsk_add_node_rcu(tw, &ehead->chain); |
| 134 | |
| 135 | /* Step 3: Remove SK from hash chain */ |
| 136 | if (__sk_nulls_del_node_init_rcu(sk)) |
| 137 | sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); |
| 138 | |
| 139 | spin_unlock(lock); |
| 140 | } |
| 141 | EXPORT_SYMBOL_GPL(__inet_twsk_hashdance); |
| 142 | |
| 143 | static void tw_timer_handler(unsigned long data) |
| 144 | { |
| 145 | struct inet_timewait_sock *tw = (struct inet_timewait_sock *)data; |
| 146 | |
| 147 | if (tw->tw_kill) |
| 148 | NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_TIMEWAITKILLED); |
| 149 | else |
| 150 | NET_INC_STATS_BH(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 | kmemcheck_annotate_bitfield(tw, flags); |
| 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_hash = sk->sk_hash; |
| 184 | tw->tw_ipv6only = 0; |
| 185 | tw->tw_transparent = inet->transparent; |
| 186 | tw->tw_prot = sk->sk_prot_creator; |
| 187 | atomic64_set(&tw->tw_cookie, atomic64_read(&sk->sk_cookie)); |
| 188 | twsk_net_set(tw, sock_net(sk)); |
| 189 | setup_timer(&tw->tw_timer, tw_timer_handler, (unsigned long)tw); |
| 190 | /* |
| 191 | * Because we use RCU lookups, we should not set tw_refcnt |
| 192 | * to a non null value before everything is setup for this |
| 193 | * timewait socket. |
| 194 | */ |
| 195 | atomic_set(&tw->tw_refcnt, 0); |
| 196 | |
| 197 | __module_get(tw->tw_prot->owner); |
| 198 | } |
| 199 | |
| 200 | return tw; |
| 201 | } |
| 202 | EXPORT_SYMBOL_GPL(inet_twsk_alloc); |
| 203 | |
| 204 | /* These are always called from BH context. See callers in |
| 205 | * tcp_input.c to verify this. |
| 206 | */ |
| 207 | |
| 208 | /* This is for handling early-kills of TIME_WAIT sockets. |
| 209 | * Warning : consume reference. |
| 210 | * Caller should not access tw anymore. |
| 211 | */ |
| 212 | void inet_twsk_deschedule_put(struct inet_timewait_sock *tw) |
| 213 | { |
| 214 | if (del_timer_sync(&tw->tw_timer)) |
| 215 | inet_twsk_kill(tw); |
| 216 | inet_twsk_put(tw); |
| 217 | } |
| 218 | EXPORT_SYMBOL(inet_twsk_deschedule_put); |
| 219 | |
| 220 | void inet_twsk_schedule(struct inet_timewait_sock *tw, const int timeo) |
| 221 | { |
| 222 | /* timeout := RTO * 3.5 |
| 223 | * |
| 224 | * 3.5 = 1+2+0.5 to wait for two retransmits. |
| 225 | * |
| 226 | * RATIONALE: if FIN arrived and we entered TIME-WAIT state, |
| 227 | * our ACK acking that FIN can be lost. If N subsequent retransmitted |
| 228 | * FINs (or previous seqments) are lost (probability of such event |
| 229 | * is p^(N+1), where p is probability to lose single packet and |
| 230 | * time to detect the loss is about RTO*(2^N - 1) with exponential |
| 231 | * backoff). Normal timewait length is calculated so, that we |
| 232 | * waited at least for one retransmitted FIN (maximal RTO is 120sec). |
| 233 | * [ BTW Linux. following BSD, violates this requirement waiting |
| 234 | * only for 60sec, we should wait at least for 240 secs. |
| 235 | * Well, 240 consumes too much of resources 8) |
| 236 | * ] |
| 237 | * This interval is not reduced to catch old duplicate and |
| 238 | * responces to our wandering segments living for two MSLs. |
| 239 | * However, if we use PAWS to detect |
| 240 | * old duplicates, we can reduce the interval to bounds required |
| 241 | * by RTO, rather than MSL. So, if peer understands PAWS, we |
| 242 | * kill tw bucket after 3.5*RTO (it is important that this number |
| 243 | * is greater than TS tick!) and detect old duplicates with help |
| 244 | * of PAWS. |
| 245 | */ |
| 246 | |
| 247 | tw->tw_kill = timeo <= 4*HZ; |
| 248 | if (!mod_timer_pinned(&tw->tw_timer, jiffies + timeo)) { |
| 249 | atomic_inc(&tw->tw_refcnt); |
| 250 | atomic_inc(&tw->tw_dr->tw_count); |
| 251 | } |
| 252 | } |
| 253 | EXPORT_SYMBOL_GPL(inet_twsk_schedule); |
| 254 | |
| 255 | void inet_twsk_purge(struct inet_hashinfo *hashinfo, |
| 256 | struct inet_timewait_death_row *twdr, 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 | atomic_read(&twsk_net(tw)->count)) |
| 275 | continue; |
| 276 | |
| 277 | if (unlikely(!atomic_inc_not_zero(&tw->tw_refcnt))) |
| 278 | continue; |
| 279 | |
| 280 | if (unlikely((tw->tw_family != family) || |
| 281 | atomic_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); |