x86/shstk change for v6.10

Linus,

Please pull the latest x86/shstk Git tree from:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git x86-shstk-2024-05-13

# HEAD: 2883f01ec37dd8668e7222dfdb5980c86fdfe277 x86/shstk: Enable shadow stacks for x32

Enable shadow stacks for x32.

While we normally don't do such feature-enabling on 32-bit kernels anymore, this change is small, straightforward & tested on upstream glibc.

Thanks,

Ingo

------------------>
H.J. Lu (1):
x86/shstk: Enable shadow stacks for x32

 arch/x86/entry/syscalls/syscall_64.tbl | 2 +-
 arch/x86/kernel/shstk.c                | 4 ++--
 arch/x86/kernel/signal_64.c            | 6 ++++++
 3 files changed, 9 insertions(+), 3 deletions(-)

index 7e8d46f4147f..cc78226ffc35 100644
@@ -374,7 +374,7 @@
 450	common	set_mempolicy_home_node	sys_set_mempolicy_home_node
 451	common	cachestat		sys_cachestat
 452	common	fchmodat2		sys_fchmodat2
-453	64	map_shadow_stack	sys_map_shadow_stack
+453	common	map_shadow_stack	sys_map_shadow_stack
 454	common	futex_wake		sys_futex_wake
 455	common	futex_wait		sys_futex_wait
 456	common	futex_requeue		sys_futex_requeue

index 59e15dd8d0f8..6f1e9883f074 100644
@@ -163,8 +163,8 @@ static int shstk_setup(void)
 	if (features_enabled(ARCH_SHSTK_SHSTK))
 		return 0;
 
-	/* Also not supported for 32 bit and x32 */
-	if (!cpu_feature_enabled(X86_FEATURE_USER_SHSTK) || in_32bit_syscall())
+	/* Also not supported for 32 bit */
+	if (!cpu_feature_enabled(X86_FEATURE_USER_SHSTK) || in_ia32_syscall())
 		return -EOPNOTSUPP;
 
 	size = adjust_shstk_size(0);

index 23d8aaf8d9fd..8a94053c5444 100644
@@ -315,6 +315,9 @@ int x32_setup_rt_frame(struct ksignal *ksig, struct pt_regs *regs)
 
 	uc_flags = frame_uc_flags(regs);
 
+	if (setup_signal_shadow_stack(ksig))
+		return -EFAULT;
+
 	if (!user_access_begin(frame, sizeof(*frame)))
 		return -EFAULT;
 

@@ -377,6 +380,9 @@ COMPAT_SYSCALL_DEFINE0(x32_rt_sigreturn)
 	if (!restore_sigcontext(regs, &frame->uc.uc_mcontext, uc_flags))
 		goto badframe;
 
+	if (restore_signal_shadow_stack())
+		goto badframe;
+
 	if (compat_restore_altstack(&frame->uc.uc_stack))
 		goto badframe;
 

Color me confused.

"feature-enabling on 32-bit kernels"

This is not for 32-bit kernels, as far as I can tell. This is just the x32 user mode for x86-64 kernels.

Or am I missing something?

I've pulled this, but does anybody actually use x32? I feel like it was a failed experiment. No?

Linus

Well, I did ask that at the time:

https://lore.kernel.org/r/20240322164912.GAZf22iBdqdkIFcrsl@fat_crate.local

The answer on another subthread was: "Because it just works?"

Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

Brainfart: feature-enabling for 32-bit user-space ...

Yeah, so H.J. Lu suggested that shadow-stacks are a natural extension of our security facilities on OSs where x32 is already enabled:

https://lore.kernel.org/all/CAMe9rOo1ZONFgBkuN_Ni3REBRsedNwj3gNnXj1oxB0bQzuNipA@mail.gmail.com/

H.J: *which* are those OSs? I don't think any major Linux distro enables x32 anymore - here's Ubuntu and Fedora for example:

kepler:~/tip> grep X32 /boot/config-6.5.0-35-generic # CONFIG_X86_X32_ABI is not set

kepler:~/s/fedora> grep X32 lib/modules/6.9.0-64.fc41.x86_64/config # CONFIG_X86_X32_ABI is not set

Another feedback was that the observed lack of x32 kernel regressions upstream could be because 'it just works':

https://lore.kernel.org/all/CAMe9rOoEQ3jUUXy+Kai9Hg83b+79azmGfu8DBR=A3HSL05kj0A@mail.gmail.com/

... so at this point I think we should be permissive towards well-tested patches, barring contrary evidence.

'Contrary evidence' would be for example some x32 regression that wasn't fixed for a long time while nobody cared, at which point we'd remove x32 instead of fixing something that wasn't working for a long time. I'm not aware of such a regression yet, BYMMV.

Thanks,

Ingo

For what it's worth, Arch Linux just turned it back on at a user's request:

https://gitlab.archlinux.org/archlinux/packaging/packages/linux/-/commit/b0c9bf78acbdcdf90405c427cf6b751259baee74

https://gitlab.archlinux.org/archlinux/packaging/packages/linux/-/issues/51

Cheers,
Nathan