> On Jun 7, 2019, at 11:58 AM, Dave Hansen <dave.hansen@xxxxxxxxx> wrote: > > On 6/7/19 11:29 AM, Andy Lutomirski wrote: > ... >>> I think this new MSR probably needs to get included in oops output when >>> CET is enabled. >> >> This shouldn’t be able to OOPS because it only happens at CPL 3, >> right? We should put it into core dumps, though. > > Good point. > > Yu-cheng, can you just confirm that the bitmap can't be referenced in > ring-0, no matter what? We should also make sure that no funny business > happens if we put an address in the bitmap that faults, or is > non-canonical. Do we have any self-tests for that? > > Let's say userspace gets a fault on this. Do they have the > introspection capability to figure out why they faulted, say in their > signal handler? We need to stick the tracker state in the sigcontext somewhere. Did we end up defining a signal frame shadow stack token? > >>> Why don't we require that a VMA be in place for the entire bitmap? >>> Don't we need a "get" prctl function too in case something like a JIT is >>> running and needs to find the location of this bitmap to set bits itself? >>> >>> Or, do we just go whole-hog and have the kernel manage the bitmap >>> itself. Our interface here could be: >>> >>> prctl(PR_MARK_CODE_AS_LEGACY, start, size); >>> >>> and then have the kernel allocate and set the bitmap for those code >>> locations. >> >> Given that the format depends on the VA size, this might be a good >> idea. > > Yeah, making userspace know how large the address space is or could be > is rather nasty, especially if we ever get any fancy CPU features that > eat up address bits (a la ARM top-byte-ignore or SPARC ADI). That gets extra bad if we ever grow user code that uses it but is unaware. It could poke the wrong part of the bitmap. > >> Hmm. Can we be creative and skip populating it with zeros? The CPU > should only ever touch a page if we miss an ENDBR on it, so, in normal > operation, we don’t need anything to be there. We could try to prevent > anyone from *reading* it outside of ENDBR tracking if we want to avoid > people accidentally wasting lots of memory by forcing it to be fully > populated when the read it. > > Won't reads on a big, contiguous private mapping get the huge zero page > anyway? The zero pages may be free, but the page tables could be decently large. Does the core mm code use huge, immense, etc huge zero pages? Or can it synthesize them by reusing page table pages that map zeros? > >> The one downside is this forces it to be per-mm, but that seems like >> a generally reasonable model anyway. > > Yeah, practically, you could only make it shared if you shared the > layout of all code in the address space. I'm sure the big database(s) > do that cross-process, but I bet nobody else does. User ASLR > practically guarantees that nobody can do this. I meant per-mm instead of per-task.
