[PATCH] NFS: Check lengths more thoroughly in NFS4 readdir XDR decode

[Trond Myklebust]

On Thu, 2006-08-24 at 19:35 +0100, David Howells wrote:
> So, what you've done is:
> 
> -+		if (end - p < xlen)
> ++		if (end - p < xlen + 1)
>   			goto short_pkt;
>  +		dprintk("filename = %*s\n", len, (char *)p);
>  +		p += xlen;
>   		len = ntohl(*p++);	/* bitmap length */
>  -		p += len;
>  -		if (p + 1 > end)
> -+		if (end - p < len)
> ++		if (end - p < len + 1)
>   			goto short_pkt;
>  +		p += len;
>   		attrlen = XDR_QUADLEN(ntohl(*p++));
>  -		p += attrlen;		/* attributes */
>  -		if (p + 2 > end)
> -+		if (end - p < attrlen + 1)
> ++		if (end - p < attrlen + 2)
> 
> But is this equivalent:
> 
> -+		if (end - p < xlen)
> ++		if (end - p <= xlen)
>   			goto short_pkt;
>  +		dprintk("filename = %*s\n", len, (char *)p);
>  +		p += xlen;
>   		len = ntohl(*p++);	/* bitmap length */
>  -		p += len;
>  -		if (p + 1 > end)
> -+		if (end - p < len)
> ++		if (end - p <= len)
>   			goto short_pkt;
>  +		p += len;
>   		attrlen = XDR_QUADLEN(ntohl(*p++));
>  -		p += attrlen;		/* attributes */
>  -		if (p + 2 > end)
> -+		if (end - p < attrlen + 1)
> ++		if (end - p <= attrlen + 1)
                              ^^^^^^^^^^^^^^
> 
> Do you think?

No. I find that mixture of < and <= is much less easy to read. Besides,
the compiler should be able to optimise that for me.

Cheers,
  Trond