On Fri, Apr 24, 2020 at 12:21:14PM +0100, Will Deacon wrote: > Also, since you mentioned the lack of redzoning, isn't it a bit dodgy > allocating blindly out of the kmem_cache? It means we don't have a redzone > or a guard page, so if you can trigger something like a recursion bug then > could you scribble past the SCS before the main stack overflows? Would this > clobber somebody else's SCS? I agree that allocating from a kmem_cache isn't ideal for safety. It's a compromise to reduce memory overhead. > The vmap version that I asked Sami to drop > is at least better in this regard, although the guard page is at the wrong > end of the stack and we just hope that the allocation below us didn't pass > VM_NO_GUARD. Looks like the same story for vmap stack :/ SCS grows up and the guard page is after the allocation, so how is it at the wrong end? Am I missing something here? > If we split the pointer in two (base, offset) then we could leave the > base live in the thread_info, not require alignment of the stacks (which > may allow for unconditional redzoning?) and then just update the offset > value on context switch, which could be trivially checked as part of the > existing stack overflow checking on kernel entry. I sent out v13 with split pointers, but I'm not sure it's convenient to add an overflow check to kernel_ventry where the VMAP_STACK check is done. I suppose I could add a check to kernel_entry after we load x18 from tsk. Thoughts? Sami
