net_sched: avoid generating same handle for u32 filters

When kernel generates a handle for a u32 filter, it tries to start
from the max in the bucket. So when we have a filter with the max (fff)
handle, it will cause kernel always generates the same handle for new
filters. This can be shown by the following command:

	tc qdisc add dev eth0 ingress
	tc filter add dev eth0 parent ffff: protocol ip pref 770 handle 800::fff u32 match ip protocol 1 0xff
	tc filter add dev eth0 parent ffff: protocol ip pref 770 u32 match ip protocol 1 0xff
	...

we will get some u32 filters with same handle:

 # tc filter show dev eth0 parent ffff:
filter protocol ip pref 770 u32
filter protocol ip pref 770 u32 fh 800: ht divisor 1
filter protocol ip pref 770 u32 fh 800::fff order 4095 key ht 800 bkt 0
  match 00010000/00ff0000 at 8
filter protocol ip pref 770 u32 fh 800::fff order 4095 key ht 800 bkt 0
  match 00010000/00ff0000 at 8
filter protocol ip pref 770 u32 fh 800::fff order 4095 key ht 800 bkt 0
  match 00010000/00ff0000 at 8
filter protocol ip pref 770 u32 fh 800::fff order 4095 key ht 800 bkt 0
  match 00010000/00ff0000 at 8

handles should be unique. This patch fixes it by looking up a bitmap,
so that can guarantee the handle is as unique as possible. For compatibility,
we still start from 0x800.

Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Cong Wang <cwang@twopensource.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/sched/cls_u32.c b/net/sched/cls_u32.c
index c39b583ace3229d4bae6a7b3774593e5eebd7141..70c0be8d0121db461c1e21793a1b68d844f37e8b 100644 (file)
--- a/net/sched/cls_u32.c
+++ b/net/sched/cls_u32.c
@@ -38,6 +38,7 @@
 #include <linux/errno.h>
 #include <linux/rtnetlink.h>
 #include <linux/skbuff.h>
+#include <linux/bitmap.h>
 #include <net/netlink.h>
 #include <net/act_api.h>
 #include <net/pkt_cls.h>
@@ -460,17 +461,25 @@ static int u32_delete(struct tcf_proto *tp, unsigned long arg)
        return 0;
 }
 
+#define NR_U32_NODE (1<<12)
 static u32 gen_new_kid(struct tc_u_hnode *ht, u32 handle)
 {
        struct tc_u_knode *n;
-       unsigned int i = 0x7FF;
+       unsigned long i;
+       unsigned long *bitmap = kzalloc(BITS_TO_LONGS(NR_U32_NODE) * sizeof(unsigned long),
+                                       GFP_KERNEL);
+       if (!bitmap)
+               return handle | 0xFFF;
 
        for (n = ht->ht[TC_U32_HASH(handle)]; n; n = n->next)
-               if (i < TC_U32_NODE(n->handle))
-                       i = TC_U32_NODE(n->handle);
-       i++;
+               set_bit(TC_U32_NODE(n->handle), bitmap);
 
-       return handle | (i > 0xFFF ? 0xFFF : i);
+       i = find_next_zero_bit(bitmap, NR_U32_NODE, 0x800);
+       if (i >= NR_U32_NODE)
+               i = find_next_zero_bit(bitmap, NR_U32_NODE, 1);
+
+       kfree(bitmap);
+       return handle | (i >= NR_U32_NODE ? 0xFFF : i);
 }
 
 static const struct nla_policy u32_policy[TCA_U32_MAX + 1] = {