[NETFILTER]: nf_conntrack: early_drop improvement

When the maximum number of conntrack entries is reached and a new
one needs to be allocated, conntrack tries to drop an unassured
connection from the same hash bucket the new conntrack would hash
to. Since with a properly sized hash the average number of entries
per bucket is 1, the chances of actually finding one are not very
good. This patch makes it walk the hash until a minimum number of
8 entries are checked.

Based on patch by Vasily Averin <vvs@sw.ru>.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c
index d1fc019760a179d8a1dbebed20e11dc338612fd2..472396dac05c3d1e8ca4860ce60d2acee6e9d280 100644 (file)
--- a/net/netfilter/nf_conntrack_core.c
+++ b/net/netfilter/nf_conntrack_core.c
@@ -377,21 +377,30 @@ nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 
+#define NF_CT_EVICTION_RANGE   8
+
 /* There's a small race here where we may free a just-assured
    connection.  Too bad: we're in trouble anyway. */
-static int early_drop(struct hlist_head *chain)
+static int early_drop(unsigned int hash)
 {
        /* Use oldest entry, which is roughly LRU */
        struct nf_conntrack_tuple_hash *h;
        struct nf_conn *ct = NULL, *tmp;
        struct hlist_node *n;
+       unsigned int i, cnt = 0;
        int dropped = 0;
 
        read_lock_bh(&nf_conntrack_lock);
-       hlist_for_each_entry(h, n, chain, hnode) {
-               tmp = nf_ct_tuplehash_to_ctrack(h);
-               if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
-                       ct = tmp;
+       for (i = 0; i < nf_conntrack_htable_size; i++) {
+               hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
+                       tmp = nf_ct_tuplehash_to_ctrack(h);
+                       if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
+                               ct = tmp;
+                       cnt++;
+               }
+               if (ct || cnt >= NF_CT_EVICTION_RANGE)
+                       break;
+               hash = (hash + 1) % nf_conntrack_htable_size;
        }
        if (ct)
                atomic_inc(&ct->ct_general.use);
@@ -425,8 +434,7 @@ struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
        if (nf_conntrack_max
            && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
                unsigned int hash = hash_conntrack(orig);
-               /* Try dropping from this hash chain. */
-               if (!early_drop(&nf_conntrack_hash[hash])) {
+               if (!early_drop(hash)) {
                        atomic_dec(&nf_conntrack_count);
                        if (net_ratelimit())
                                printk(KERN_WARNING