Merge branch 'for_paulus' of master.kernel.org:/pub/scm/linux/kernel/git/galak/powerpc
[pandora-kernel.git] / net / ipv4 / inet_hashtables.c
index e8d29fe..ee9b551 100644 (file)
 
 #include <linux/config.h>
 #include <linux/module.h>
+#include <linux/random.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/wait.h>
 
 #include <net/inet_connection_sock.h>
 #include <net/inet_hashtables.h>
+#include <net/ip.h>
 
 /*
  * Allocate and initialize a new local port bind bucket.
@@ -41,8 +43,6 @@ struct inet_bind_bucket *inet_bind_bucket_create(kmem_cache_t *cachep,
        return tb;
 }
 
-EXPORT_SYMBOL(inet_bind_bucket_create);
-
 /*
  * Caller must hold hashbucket lock for this tb with local BH disabled
  */
@@ -62,8 +62,6 @@ void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
        inet_csk(sk)->icsk_bind_hash = tb;
 }
 
-EXPORT_SYMBOL(inet_bind_hash);
-
 /*
  * Get rid of any references to a local port held by the given sock.
  */
@@ -163,3 +161,179 @@ struct sock *__inet_lookup_listener(const struct hlist_head *head, const u32 dad
 }
 
 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
+
+/* called with local bh disabled */
+static int __inet_check_established(struct inet_timewait_death_row *death_row,
+                                   struct sock *sk, __u16 lport,
+                                   struct inet_timewait_sock **twp)
+{
+       struct inet_hashinfo *hinfo = death_row->hashinfo;
+       struct inet_sock *inet = inet_sk(sk);
+       u32 daddr = inet->rcv_saddr;
+       u32 saddr = inet->daddr;
+       int dif = sk->sk_bound_dev_if;
+       INET_ADDR_COOKIE(acookie, saddr, daddr)
+       const __u32 ports = INET_COMBINED_PORTS(inet->dport, lport);
+       unsigned int hash = inet_ehashfn(daddr, lport, saddr, inet->dport);
+       struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
+       struct sock *sk2;
+       const struct hlist_node *node;
+       struct inet_timewait_sock *tw;
+
+       prefetch(head->chain.first);
+       write_lock(&head->lock);
+
+       /* Check TIME-WAIT sockets first. */
+       sk_for_each(sk2, node, &(head + hinfo->ehash_size)->chain) {
+               tw = inet_twsk(sk2);
+
+               if (INET_TW_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif)) {
+                       if (twsk_unique(sk, sk2, twp))
+                               goto unique;
+                       else
+                               goto not_unique;
+               }
+       }
+       tw = NULL;
+
+       /* And established part... */
+       sk_for_each(sk2, node, &head->chain) {
+               if (INET_MATCH(sk2, hash, acookie, saddr, daddr, ports, dif))
+                       goto not_unique;
+       }
+
+unique:
+       /* Must record num and sport now. Otherwise we will see
+        * in hash table socket with a funny identity. */
+       inet->num = lport;
+       inet->sport = htons(lport);
+       sk->sk_hash = hash;
+       BUG_TRAP(sk_unhashed(sk));
+       __sk_add_node(sk, &head->chain);
+       sock_prot_inc_use(sk->sk_prot);
+       write_unlock(&head->lock);
+
+       if (twp) {
+               *twp = tw;
+               NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
+       } else if (tw) {
+               /* Silly. Should hash-dance instead... */
+               inet_twsk_deschedule(tw, death_row);
+               NET_INC_STATS_BH(LINUX_MIB_TIMEWAITRECYCLED);
+
+               inet_twsk_put(tw);
+       }
+
+       return 0;
+
+not_unique:
+       write_unlock(&head->lock);
+       return -EADDRNOTAVAIL;
+}
+
+static inline u32 inet_sk_port_offset(const struct sock *sk)
+{
+       const struct inet_sock *inet = inet_sk(sk);
+       return secure_ipv4_port_ephemeral(inet->rcv_saddr, inet->daddr, 
+                                         inet->dport);
+}
+
+/*
+ * Bind a port for a connect operation and hash it.
+ */
+int inet_hash_connect(struct inet_timewait_death_row *death_row,
+                     struct sock *sk)
+{
+       struct inet_hashinfo *hinfo = death_row->hashinfo;
+       const unsigned short snum = inet_sk(sk)->num;
+       struct inet_bind_hashbucket *head;
+       struct inet_bind_bucket *tb;
+       int ret;
+
+       if (!snum) {
+               int low = sysctl_local_port_range[0];
+               int high = sysctl_local_port_range[1];
+               int range = high - low;
+               int i;
+               int port;
+               static u32 hint;
+               u32 offset = hint + inet_sk_port_offset(sk);
+               struct hlist_node *node;
+               struct inet_timewait_sock *tw = NULL;
+
+               local_bh_disable();
+               for (i = 1; i <= range; i++) {
+                       port = low + (i + offset) % range;
+                       head = &hinfo->bhash[inet_bhashfn(port, hinfo->bhash_size)];
+                       spin_lock(&head->lock);
+
+                       /* Does not bother with rcv_saddr checks,
+                        * because the established check is already
+                        * unique enough.
+                        */
+                       inet_bind_bucket_for_each(tb, node, &head->chain) {
+                               if (tb->port == port) {
+                                       BUG_TRAP(!hlist_empty(&tb->owners));
+                                       if (tb->fastreuse >= 0)
+                                               goto next_port;
+                                       if (!__inet_check_established(death_row,
+                                                                     sk, port,
+                                                                     &tw))
+                                               goto ok;
+                                       goto next_port;
+                               }
+                       }
+
+                       tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, head, port);
+                       if (!tb) {
+                               spin_unlock(&head->lock);
+                               break;
+                       }
+                       tb->fastreuse = -1;
+                       goto ok;
+
+               next_port:
+                       spin_unlock(&head->lock);
+               }
+               local_bh_enable();
+
+               return -EADDRNOTAVAIL;
+
+ok:
+               hint += i;
+
+               /* Head lock still held and bh's disabled */
+               inet_bind_hash(sk, tb, port);
+               if (sk_unhashed(sk)) {
+                       inet_sk(sk)->sport = htons(port);
+                       __inet_hash(hinfo, sk, 0);
+               }
+               spin_unlock(&head->lock);
+
+               if (tw) {
+                       inet_twsk_deschedule(tw, death_row);
+                       inet_twsk_put(tw);
+               }
+
+               ret = 0;
+               goto out;
+       }
+
+       head = &hinfo->bhash[inet_bhashfn(snum, hinfo->bhash_size)];
+       tb  = inet_csk(sk)->icsk_bind_hash;
+       spin_lock_bh(&head->lock);
+       if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
+               __inet_hash(hinfo, sk, 0);
+               spin_unlock_bh(&head->lock);
+               return 0;
+       } else {
+               spin_unlock(&head->lock);
+               /* No definite answer... Walk to established hash table */
+               ret = __inet_check_established(death_row, sk, snum, NULL);
+out:
+               local_bh_enable();
+               return ret;
+       }
+}
+
+EXPORT_SYMBOL_GPL(inet_hash_connect);