]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
[TCP]: Fix truesize underflow
authorHerbert Xu <herbert@gondor.apana.org.au>
Tue, 18 Apr 2006 20:24:14 +0000 (13:24 -0700)
committerDavid S. Miller <davem@sunset.davemloft.net>
Tue, 18 Apr 2006 22:57:49 +0000 (15:57 -0700)
There is a problem with the TSO packet trimming code.  The cause of
this lies in the tcp_fragment() function.

When we allocate a fragment for a completely non-linear packet the
truesize is calculated for a payload length of zero.  This means that
truesize could in fact be less than the real payload length.

When that happens the TSO packet trimming can cause truesize to become
negative.  This in turn can cause sk_forward_alloc to be -n * PAGE_SIZE
which would trigger the warning.

I've copied the code DaveM used in tso_fragment which should work here.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_output.c

index b871db6adc55cd838c07b11b4461bc5e879f5377..44df1db726a37bc77c28dda33d5238b9f40a5a8e 100644 (file)
@@ -551,7 +551,9 @@ int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len, unsigned int mss
        buff = sk_stream_alloc_skb(sk, nsize, GFP_ATOMIC);
        if (buff == NULL)
                return -ENOMEM; /* We'll just try again later. */
-       sk_charge_skb(sk, buff);
+
+       buff->truesize = skb->len - len;
+       skb->truesize -= buff->truesize;
 
        /* Correct the sequence numbers. */
        TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len;