Re: [patch] perf: potential underflow in perf_sample_ustack_size()

[Jiri Olsa <jolsa@redhat.com>]

On Fri, Sep 20, 2013 at 02:37:36PM +0300, Dan Carpenter wrote:
> The code here is trying to ensure that we don't have a
> "header_size + stack_size" which is more than USHRT_MAX.  I changed
> the overflow check a little to make it more clear.
> 
> My concern here is that if "header_size + sizeof(u64)" is very large
> then we could end up with underflow doing the subtraction and end up
> with a "stack_size" larger than intended.
> 
> I don't know perf well enough to say if this is possible.
> 
> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index dd236b6..eec9e683 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -4217,11 +4217,13 @@ perf_sample_ustack_size(u16 stack_size, u16 header_size,
>  	header_size += 2 * sizeof(u64);
>  
>  	/* Do we fit in with the current stack dump size? */
> -	if ((u16) (header_size + stack_size) < header_size) {
> +	if (header_size > USHRT_MAX - stack_size) {

hum, the original check looks clear enough to me

>  		/*
>  		 * If we overflow the maximum size for the sample,
>  		 * we customize the stack dump size to fit in.
>  		 */
> +		if (header_size + sizeof(u64) > USHRT_MAX)
> +			return 0;

ok, I wonder if we could practically reach header size
this big, but it's safer to check

thanks,
jirka

>  		stack_size = USHRT_MAX - header_size - sizeof(u64);
>  		stack_size = round_up(stack_size, sizeof(u64));
>  	}