On Fri, Oct 26, 2018 at 09:51:06AM +0100, Ben Dooks wrote:
> I have been looking into testing the address_space() pointers in
> the current Linux kernel and came up with a case where I am not
> sure how this should work (the kernel's __user attribute).
>
> We tested the following cases, In this case sparse correctly
> identified the incorrect address space/no-dereference. The first
> was not identified:
>
> # define __user __attribute__((noderef, address_space(1)))
>
> void test(char __user *ptr)
> {
> printk("%s\n", ptr);
> printk("%c\n", ptr[0]);
> }
Hi,
Yes, but everything here is as it should:
* the 2nd line should trigger a message like
test.c:8:27: warning: dereference of noderef expression
since the pointer is dereferenced (before being passed to printk().
* no warning for the 1st line since there is no dereference and no
pointer assignment to a pointer of a different address space..
> I know that there is a gcc printf format attribute to allow
> specifying the positions of the format and start of variadic
> arguments.
>
> I have had a quick look at the code but have not tried to make
> any modifications yet, and would like some feedback or help
> before making any:
>
> - Should we automatically warn on possible de-ref (or address space)
> on passing to a variadic function (I think the simplest)
I think this would be wrong.
For example, there is no problems if the corresponding address is
simply displayed (via "%p" or "%lu") but it would if the pointer
is dereferenced (for example, via "%s").
> - Is adding support for __attribute__((format(printf,a,b))) a
> good idea (and if so is anyone looking into this already)
Yes, I think it's a good idea.
I don't think anyone is looking to add this to sparse already
(it's something I've added in my nice-to-have list but this list
is quite long ...).
> - If adding printf format a good idea, then should we extend it
> to adding a kernel printk style one too (modern kernels seem
> to love extending the printf formatting)
I really think so, yes.
> Any help would be appreciated.
You're welcome,
-- Luc Van Oostenryck
