Re: [PATCH] x86/stackprotector: fix build failure with CONFIG_STACKPROTECTOR=n

From: Ard Biesheuvel
Date: Fri Dec 06 2024 - 10:18:07 EST


On Fri, 6 Dec 2024 at 16:12, Brian Gerst <brgerst@xxxxxxxxx> wrote:
>
> On Fri, Dec 6, 2024 at 9:37 AM Ard Biesheuvel <ardb@xxxxxxxxxx> wrote:
> >
> > On Fri, 6 Dec 2024 at 15:22, Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
> > >
> > > On 12/06, Ard Biesheuvel wrote:
> > > >
> > > > On Fri, 6 Dec 2024 at 13:32, Oleg Nesterov <oleg@xxxxxxxxxx> wrote:
> > > > >
> > > > > +#ifdef CONFIG_STACKPROTECTOR
> > > > > /* needed for Clang - see arch/x86/entry/entry.S */
> > > > > PROVIDE(__ref_stack_chk_guard = __stack_chk_guard);
> > > > > +#endif
> > > > >
> > > > > #ifdef CONFIG_X86_64
> > > > > /*
> > > >
> > > > This shouldn't be necessary - PROVIDE() is only evaluated if a
> > > > reference exists to the symbol it defines.
> > > >
> > > > Also, I'm failing to reproduce this. Could you share your .config,
> > > > please, and the error that you get during the build?
> > >
> > > Please see the attached .config
> > >
> > > without the change above:
> > >
> > > $ make bzImage
> > > CALL scripts/checksyscalls.sh
> > > DESCEND objtool
> > > INSTALL libsubcmd_headers
> > > UPD include/generated/utsversion.h
> > > CC init/version-timestamp.o
> > > KSYMS .tmp_vmlinux0.kallsyms.S
> > > AS .tmp_vmlinux0.kallsyms.o
> > > LD .tmp_vmlinux1
> > > ./arch/x86/kernel/vmlinux.lds:154: undefined symbol `__stack_chk_guard' referenced in expression
> > > scripts/Makefile.vmlinux:77: recipe for target 'vmlinux' failed
> > > make[2]: *** [vmlinux] Error 1
> > > /home/oleg/tmp/LINUX/Makefile:1225: recipe for target 'vmlinux' failed
> > > make[1]: *** [vmlinux] Error 2
> > > Makefile:251: recipe for target '__sub-make' failed
> > > make: *** [__sub-make] Error 2
> > >
> > > perhaps this is because my toolchain is quite old,
> > >
> > > $ ld -v
> > > GNU ld version 2.25-17.fc23
> > >
> > > but according to Documentation/process/changes.rst
> > >
> > > binutils 2.25 ld -v
> > >
> > > it is still supported.
> > >
> >
> > We're about to bump the minimum toolchain requirements to GCC 8.1 (and
> > whichever version of binutils was current at the time), so you might
> > want to consider upgrading.
> >
> > However, you are right that these are still supported today, and so we
> > need this fix this, especially because this has been backported to
> > older stable kernels too.
> >
> > For the patch,
> >
> > Acked-by: Ard Biesheuvel <ardb@xxxxxxxxxx>
>
> Using PROVIDES() is now unnecessary.
>

At this point, the use of -mstack-protector-guard-symbol is still
limited to 32-bit x86. However, if we drop PROVIDE() here, the 64-bit
kernel will also gain a symbol `__ref_stack_chk_guard` in its symbol
table (and /proc/kallsyms, most likely).

Not sure whether that matters or not, but I'd rather keep the
PROVIDE() as it doesn't do any harm.