Re: [Qemu-devel] [PATCH 2/2] pci: rewrite devaddr parsing

[Michael S. Tsirkin]

On Fri, Feb 17, 2012 at 01:35:02PM +0100, Markus Armbruster wrote:
> "Michael S. Tsirkin" <address@hidden> writes:
> 
> > On Thu, Feb 16, 2012 at 01:39:02PM -0700, Eric Blake wrote:
> >> On 02/16/2012 12:23 PM, malc wrote:
> >> > On Thu, 16 Feb 2012, Michael S. Tsirkin wrote:
> >> > 
> >> >> Use scanf instead of manual string scanning.
> >> >>
> >> >> +
> >> >> +    /* Parse [[<domain>:]<bus>:]<slot> */
> >> >> +    sscanf(addr, "%x:%x:%x%n", &dom, &bus, &slot, &n);
> >> > 
> >> > sscanf can fail.
> >> 
> >> Worse, the *scanf family has undefined behavior on integer overflow.  If
> >> addr contains "100000000000000:0:0", there's no telling whether it will
> >> be diagnosed as a parse error, or silently accepted and truncated, in
> >> which case, there's no telling what dom will contain.
> >> 
> >> I cringe any time I see someone using scanf to parse numbers from
> >> arbitrary user input; I barely tolerate it for parsing things generated
> >> by the kernel, but even there, I won't ever use scanf myself.
> >> Same goes
> >> for atoi.  _Only_ strtol and friends can robustly parse arbitrary input
> >> into integers.
> >
> > Seems easy to fix: I'll just set maximum field width of 8.
> 
> Nope.  Functional change: "000000001.2" is no longer accepted.

OK, there's an easy way to fix with
%*[0]%8x

> > Any other issues?
> 
> Yes.
> 
> 1. More functional changes, e.g.
> 
>    * "1" is no longer rejected when funcp != NULL
> 
>    Probably more.  I'd be particularly wary of sscanf()'s appetite for
>    space.

BTW strtoul that we currently use skips whitespace
silently too. We probably should validate input with
%*[0-9a-f.:] beforehand to address this.

> 2. Diffstat: 1 files changed, 38 insertions(+), 43 deletions(-)
> 
> Why bother?