On Wed, 2017-11-29 at 11:39 -0800, Linus Torvalds wrote: > On Wed, Nov 29, 2017 at 11:22 AM, Linus Torvalds > <torva...@linux-foundation.org> wrote: > > > > What I didn't realize until after pulling this and testing, is that it > > completely breaks '%pK'. > > > > We've marked various sensitive pointers with %pK, but that is now > > _less_ secure than %p is, since it doesn't do the hashing because of > > how you refactored the %pK code out of 'pointer()' into its own > > function. > > > > So now %pK ends up using the plain "number()" function. Reading > > through the series I hadn't noticed that the refactoring ended up > > messing with that. > > > > I'll fix it up somehow. > > I ended up just doing this: > > case 'K': > + if (!kptr_restrict) > + break; > return restricted_pointer(buf, end, ptr, spec); > > which basically says that "if kptr_restrict isn't set, %pK is the same as %p". > > Now, I feel that we should probably get rid of 'restricted_pointer()' > entirely, since now the regular '%p' is arguably safer than '%pK' is, > but I also didn't want to mess with the case that I have never used > and that most distros don't seem to set. > > Alternatively, we might make the 'K' behavior of clearing the pointer > be in addition to the other flags, so that you could do '%pxK' and get > the old %pK behavior. But since I am not a huge fan of %pK to begin > with, I can't find it in myself to care too much. > > So I'll leave that for Kees & co to decide on. Comments?
I'd prefer a global sed of '%pK' to '%pxK' and remove '%pK' altogether