Re: [PATCH 1/2] Introduce check_readable_bytes()

[Josef Bacik <josef@redhat.com>]

On Thu, May 14, 2009 at 06:19:00PM +0200, Vitaly Mayatskikh wrote:
> This routine acts almost as fault_in_pages_readable(), but returns
> accessible amount of bytes instead of plain OK/FAIL, so callers
> may know how many data can be proceeded without GPF.
> 
> Signed-off-by: Vitaly Mayatskikh <v.mayatskih@gmail.com>
> ---
>  include/linux/pagemap.h |   35 +++++++++++++++++++++++++++++++++++
>  1 files changed, 35 insertions(+), 0 deletions(-)
> 
> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
> index 34da523..f931308 100644
> --- a/include/linux/pagemap.h
> +++ b/include/linux/pagemap.h
> @@ -439,6 +439,41 @@ static inline int fault_in_pages_readable(const char __user *uaddr, int size)
>  	return ret;
>  }
>  
> +/*
> + * check_readable_bytes - check if given amount of bytes can be read
> + * at given address.
> + * @uaddr: address to check
> + * @size: size to check
> + *
> + * Returns 0 when @size is 0, -EFAULT when @uaddr points to
> + * unaccessible region, or count of accessible bytes.
> + */
> +static inline int check_readable_bytes(const char __user *uaddr, int size)
> +{
> +	volatile char c;
> +	int ret = 0;
> +	long page_begin = (unsigned long)uaddr & PAGE_MASK;
> +	long page_end = ((unsigned long)uaddr + size) & PAGE_MASK;
> +	long ptr = page_begin;
> +

Firstly page_begin/page_end/ptr should all be unsigned long.  Secondly page_end
should be uaddr+size-1.

> +	if (unlikely(size == 0))
> +		goto out;
> +
> +	while (!ret && ptr <= page_end) {
> +		ret = __get_user(c, (const char __user*)ptr);
> +		if (!ret)
> +			ptr += PAGE_SIZE;
> +	}
> +
> +	if (likely(!ret))
> +		ret = size;
> +	else
> +		ret = (ptr == page_begin) ?
> +			-EFAULT : (const char __user *)ptr - uaddr;

This isn't quite right either, we do some pointer math, and return that and its
casted as an int?  That seems like a bad idea in general.  Thanks,

Josef