Re: [PATCH] x86/stackprotector: Work around strict Clang TLS symbol requirements

[Ard Biesheuvel]

On Sat, 5 Oct 2024 at 17:48, Brian Gerst <brgerst@xxxxxxxxx> wrote:
>
> On Wed, Oct 2, 2024 at 7:04 AM Ard Biesheuvel <ardb@xxxxxxxxxx> wrote:
> >
> > On Wed, 2 Oct 2024 at 11:25, Ard Biesheuvel <ardb+git@xxxxxxxxxx> wrote:
> > >
> > > From: Ard Biesheuvel <ardb@xxxxxxxxxx>
> > >
> > > GCC and Clang both implement stack protector support based on Thread
> > > Local Storage (TLS) variables, and this is used in the kernel to
> > > implement per-task stack cookies, by copying a task's stack cookie into
> > > a per-CPU variable every time it is scheduled in.
> > >
> > > Both now also implement -mstack-protector-guard-symbol=, which permits
> > > the TLS variable to be specified directly. This is useful because it
> > > will allow us to move away from using a fixed offset of 40 bytes into
> > > the per-CPU area on x86_64, which requires a lot of special handling in
> > > the per-CPU code and the runtime relocation code.
> > >
> > > However, while GCC is rather lax in its implementation of this command
> > > line option, Clang actually requires that the provided symbol name
> > > refers to a TLS variable (i.e., one declared with __thread), although it
> > > also permits the variable to be undeclared entirely, in which case it
> > > will use an implicit declaration of the right type.
> > >
> > > The upshot of this is that Clang will emit the correct references to the
> > > stack cookie variable in most cases, e.g.,
> > >
> > >    10d:       64 a1 00 00 00 00       mov    %fs:0x0,%eax
> > >                       10f: R_386_32   __stack_chk_guard
> > >
> > > However, if a non-TLS definition of the symbol in question is visible in
> > > the same compilation unit (which amounts to the whole of vmlinux if LTO
> > > is enabled), it will drop the per-CPU prefix and emit a load from a
> > > bogus address.
> > >
> > > Work around this by using a symbol name that never occurs in C code, and
> > > emit it as an alias in the linker script.
> > >
> > > Fixes: 3fb0fdb3bbe7 ("x86/stackprotector/32: Make the canary into a regular percpu variable")
> > > Cc: <stable@xxxxxxxxxxxxxxx>
> > > Cc: Fangrui Song <i@xxxxxxxxxx>
> > > Cc: Brian Gerst <brgerst@xxxxxxxxx>
> > > Cc: Uros Bizjak <ubizjak@xxxxxxxxx>
> > > Cc: Nathan Chancellor <nathan@xxxxxxxxxx>
> > > Cc: Andy Lutomirski <luto@xxxxxxxxxx>
> > > Link: https://github.com/ClangBuiltLinux/linux/issues/1854
> > > Signed-off-by: Ard Biesheuvel <ardb@xxxxxxxxxx>
> > > ---
> > >  arch/x86/Makefile             |  5 +++--
> > >  arch/x86/entry/entry.S        | 16 ++++++++++++++++
> > >  arch/x86/kernel/cpu/common.c  |  2 ++
> > >  arch/x86/kernel/vmlinux.lds.S |  3 +++
> > >  4 files changed, 24 insertions(+), 2 deletions(-)
> > >
> >
> > This needs the hunk below applied on top for CONFIG_MODVERSIONS:
> >
> > --- a/arch/x86/include/asm/asm-prototypes.h
> > +++ b/arch/x86/include/asm/asm-prototypes.h
> > @@ -20,3 +20,6 @@
> >  extern void cmpxchg8b_emu(void);
> >  #endif
> >
> > +#ifdef CONFIG_STACKPROTECTOR
> > +extern unsigned long __ref_stack_chk_guard;
> > +#endif
>
> Shouldn't this also be guarded by __GENKSYMS__, since the whole point
> of this is to hide the declaration from the compiler?
>

Yes, good point. Even though it does not matter in practice (the issue
is tickled only by a visible *definition*, not by a declaration), this
file is included into C code, which should be avoided.