Re: [PATCH 4/8] pipe: fix limit checking in pipe_set_size()

[Vegard Nossum <vegard.nossum-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>]

On 08/19/2016 07:25 AM, Michael Kerrisk (man-pages) wrote:
> The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ))
> has the following problems:
[...]
> @@ -1030,6 +1030,7 @@ static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
>   {
>   	struct pipe_buffer *bufs;
>   	unsigned int size, nr_pages;
> +	long ret = 0;
>
>   	size = round_pipe_size(arg);
>   	nr_pages = size >> PAGE_SHIFT;
> @@ -1037,13 +1038,26 @@ static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
>   	if (!nr_pages)
>   		return -EINVAL;
>
> -	if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size)
> -		return -EPERM;
> +	account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
>
> -	if ((too_many_pipe_buffers_hard(pipe->user) ||
> -			too_many_pipe_buffers_soft(pipe->user)) &&
> -			!capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
> -		return -EPERM;
> +	/*
> +	 * If trying to increase the pipe capacity, check that an
> +	 * unprivileged user is not trying to exceed various limits.
> +	 * (Decreasing the pipe capacity is always permitted, even
> +	 * if the user is currently over a limit.)
> +	 */
> +	if (nr_pages > pipe->buffers) {
> +		if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
> +			ret = -EPERM;
> +			goto out_revert_acct;
> +		} else if ((too_many_pipe_buffers_hard(pipe->user) ||
> +				too_many_pipe_buffers_soft(pipe->user)) &&
> +				!capable(CAP_SYS_RESOURCE) &&
> +				!capable(CAP_SYS_ADMIN)) {
> +			ret = -EPERM;
> +			goto out_revert_acct;
> +		}
> +	}

I'm slightly worried about not checking arg/nr_pages before we pass it
on to account_pipe_buffers().

The potential problem happens if the user passes a very large number
which will overflow pipe->user->pipe_bufs.

On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX
then round_pipe_size() returns INT_MAX. Although it's true that the
accounting is done in terms of pages and not bytes, so you'd need on the
order of (1 << 13) = 8192 processes hitting the limit at the same time
in order to make it overflow, which seems a bit unlikely.

(See https://lkml.org/lkml/2016/8/12/215 for another discussion on the
limit checking)

Is there any reason why we couldn't do the (size > pipe_max_size) check
before calling account_pipe_buffers()?


Vegard