netfilter: nf_tables: Fix nft limit burst handling

Current implementation treats the burst configuration the same as
rate configuration. This can cause the per packet cost to be lower
than configured. In effect, this bug causes the token bucket to be
refilled at a higher rate than what user has specified.

This patch changes the implementation so that the token bucket size
is controlled by "rate + burst", while maintain the token bucket
refill rate the same as user specified.

Fixes: 96518518cc41 ("netfilter: add nftables")
Signed-off-by: Andy Zhou <azhou@ovn.org>
Acked-by: Joe Stringer <joe@ovn.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

diff --git a/net/netfilter/nft_limit.c b/net/netfilter/nft_limit.c
index 18dd57a526513bd726944fa7ad7a9e41fcfb0251..14538b1d4d110e9b00445f477fb55377a5189f94 100644 (file)
--- a/net/netfilter/nft_limit.c
+++ b/net/netfilter/nft_limit.c
@@ -65,19 +65,23 @@ static int nft_limit_init(struct nft_limit *limit,
        limit->nsecs = unit * NSEC_PER_SEC;
        if (limit->rate == 0 || limit->nsecs < unit)
                return -EOVERFLOW;
-       limit->tokens = limit->tokens_max = limit->nsecs;
-
-       if (tb[NFTA_LIMIT_BURST]) {
-               u64 rate;
 
+       if (tb[NFTA_LIMIT_BURST])
                limit->burst = ntohl(nla_get_be32(tb[NFTA_LIMIT_BURST]));
+       else
+               limit->burst = 0;
+
+       if (limit->rate + limit->burst < limit->rate)
+               return -EOVERFLOW;
 
-               rate = limit->rate + limit->burst;
-               if (rate < limit->rate)
-                       return -EOVERFLOW;
+       /* The token bucket size limits the number of tokens can be
+        * accumulated. tokens_max specifies the bucket size.
+        * tokens_max = unit * (rate + burst) / rate.
+        */
+       limit->tokens = div_u64(limit->nsecs * (limit->rate + limit->burst),
+                               limit->rate);
+       limit->tokens_max = limit->tokens;
 
-               limit->rate = rate;
-       }
        if (tb[NFTA_LIMIT_FLAGS]) {
                u32 flags = ntohl(nla_get_be32(tb[NFTA_LIMIT_FLAGS]));
 
@@ -95,9 +99,8 @@ static int nft_limit_dump(struct sk_buff *skb, const struct nft_limit *limit,
 {
        u32 flags = limit->invert ? NFT_LIMIT_F_INV : 0;
        u64 secs = div_u64(limit->nsecs, NSEC_PER_SEC);
-       u64 rate = limit->rate - limit->burst;
 
-       if (nla_put_be64(skb, NFTA_LIMIT_RATE, cpu_to_be64(rate),
+       if (nla_put_be64(skb, NFTA_LIMIT_RATE, cpu_to_be64(limit->rate),
                         NFTA_LIMIT_PAD) ||
            nla_put_be64(skb, NFTA_LIMIT_UNIT, cpu_to_be64(secs),
                         NFTA_LIMIT_PAD) ||