On Mon, May 04, 2020 at 05:52:28PM +0100, Will Deacon wrote: > On Mon, Apr 27, 2020 at 01:45:46PM -0700, Sami Tolvanen wrote: > > On Fri, Apr 24, 2020 at 12:21:14PM +0100, Will Deacon wrote: > > > 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? > > Sorry, I'd got the SCS upside-down in my head (hey, that second 'S' stands > for 'Stack'!). But I think I'm right about vmap stack, which feels a > little fragile even though it seems to work out today with the very limited > uses of VM_NO_GUARD. Yeah, when VMAP_STACK was originally being developed, IIRC, there was an effort made to eliminate all the users of VM_NO_GUARD, and it looks like it's mostly there. Really the only use left is arm64's kernel image mapping routines, and then it's not actually used in the traditional sense -- it's just a boolean for whether to toss in a guard page at the end of the data section, and the VMAs are built manually. I think that code could actually be refactored to drop it too and then the only user would be KASAN, which, IIUC, wants to build consecutive vmap areas. -- Kees Cook
