On Tue, 10 Jan 2017, Kees Cook wrote:
> On Tue, Jan 10, 2017 at 1:14 PM, Julia Lawall <julia.law...@lip6.fr> wrote: > > OK, I have the impression that what you are looking for is the following, > > that currently does not seem to work well. Still maybe it gives an idea. > > > > The basic pattern is the following sequence: > > > > 1. copy_from_user > > 2. test on a field of the copied value > > 3. another copy_from_user > > 4. a use of the same field as tested in step 2 from the structure obtained > > by the second copy_from_user or a function call with the structure as an > > argument > > This looks pretty good! > > > In the case where the second copy_from_user stores the result in a > > pointer, then a return with no reference of the tested field is also a > > concern, unless, the pointer was already kfreed. > > I think sequence "2" above missing just looking at a direct value, > like if instead of a field it was a u32. Also, should binop include > "=="? I wasn't sure what to do with a direct value, because one wouldn't know what field it would correspond to. A solution could be to pull the first field out of the structure declaration. I'll add == and !=. > And we need to add back in get_user() too... hmmm OK. julia
