On Fri, 2023-02-24 at 13:20 +0100, Borislav Petkov wrote: > On Sat, Feb 18, 2023 at 01:14:20PM -0800, Rick Edgecombe wrote: > > diff --git a/arch/x86/include/uapi/asm/prctl.h > > b/arch/x86/include/uapi/asm/prctl.h > > index 500b96e71f18..b2b3b7200b2d 100644 > > --- a/arch/x86/include/uapi/asm/prctl.h > > +++ b/arch/x86/include/uapi/asm/prctl.h > > @@ -20,4 +20,10 @@ > > #define ARCH_MAP_VDSO_32 0x2002 > > #define ARCH_MAP_VDSO_64 0x2003 > > > > +/* Don't use 0x3001-0x3004 because of old glibcs */ > > So where is this all new interface to userspace programs documented? In the first patch: https://lore.kernel.org/lkml/20230218211433.26859-2-rick.p.edgecombe@xxxxxxxxx/ Then some more documentation is added about ARCH_SHSTK_UNLOCK and ARCH_SHSTK_STATUS (which are for CRIU) in those patches. > Do > we have an agreement with all the involved parties that this is how > we're going to support shadow stacks and that this is what userspace > should do? > > I'd like to avoid one more fiasco with glibc etc here... There are glibc patches prepared by HJ to use the new interface and it's my understanding that he has discussed the changes with the other glibc folks. Those glibc patches are used for testing these kernel patches, but will not get upstream until the kernel patches to avoid repeating the past problems. So I think it's as prepared as it can be. One future thing that might come up... Glibc has this mode called "permissive mode". When glibc is configured this way (compile time or env var), it is supposed to disable shadow stack when dlopen()ing any DSO that doesn't have the shadow stack elf header bit. The problem is, it doesn't really work as intended. It only turns off shadow stack for the calling thread, leaving the other threads free to call into the DSO with shadow stack enabled. It's not clear yet if they are going to be able to fix it in userspace. So this prompted some thinking of if there could be an additional kernel mode to help glibc in this scenario. But for non-permissive mode, glibc is queued up to use the interface as is.
