Since commit 4e57a4ddf6b0 ("ARM: 9107/1: syscall: always store
thread_info->abi_syscall"), the seccomp selftests "syscall_restart" has
been broken. This was caused by the restart syscall not being stored to
"abi_syscall" during restart setup before branching to the "local_restart"
label. Tracers would see the wrong syscall, and scno would get overwritten
while returning from the TIF_WORK path. Add the missing store.

Cc: Russell King <linux@armlinux.org.uk>
Cc: Arnd Bergmann <arnd@kernel.org>
Cc: Lecopzer Chen <lecopzer.chen@mediatek.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: linux-arm-kernel@lists.infradead.org
Fixes: 4e57a4ddf6b0 ("ARM: 9107/1: syscall: always store thread_info->abi_syscall")
Signed-off-by: Kees Cook <keescook@chromium.org>
---
 arch/arm/kernel/entry-common.S | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index bcc4c9ec3aa4..5c31e9de7a60 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S
@@ -90,6 +90,7 @@ slow_work_pending:
 	cmp	r0, #0
 	beq	no_work_pending
 	movlt	scno, #(__NR_restart_syscall - __NR_SYSCALL_BASE)
+	str	scno, [tsk, #TI_ABI_SYSCALL]	@ make sure tracers see update
 	ldmia	sp, {r0 - r6}			@ have to reload r0 - r6
 	b	local_restart			@ ... and off we go
 ENDPROC(ret_fast_syscall)
-- 
2.34.1

On Thu, Aug 10, 2023, at 21:54, Kees Cook wrote:
> Since commit 4e57a4ddf6b0 ("ARM: 9107/1: syscall: always store
> thread_info->abi_syscall"), the seccomp selftests "syscall_restart" has
> been broken. This was caused by the restart syscall not being stored to
> "abi_syscall" during restart setup before branching to the "local_restart"
> label. Tracers would see the wrong syscall, and scno would get overwritten
> while returning from the TIF_WORK path. Add the missing store.
>

> diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
> index bcc4c9ec3aa4..5c31e9de7a60 100644
> --- a/arch/arm/kernel/entry-common.S
> +++ b/arch/arm/kernel/entry-common.S
> @@ -90,6 +90,7 @@ slow_work_pending:
>  	cmp	r0, #0
>  	beq	no_work_pending
>  	movlt	scno, #(__NR_restart_syscall - __NR_SYSCALL_BASE)
> +	str	scno, [tsk, #TI_ABI_SYSCALL]	@ make sure tracers see update
>  	ldmia	sp, {r0 - r6}			@ have to reload r0 - r6
>  	b	local_restart			@ ... and off we go
>  ENDPROC(ret_fast_syscall)

Looks good to me, as in my previous reply this is safe for all cases
with the current restartable syscalls, though it would be wrong if
we had a restartblock version of epoll_wait on an OABI_COMPAT task,
which needs to look at the high bits of the number to tell which
structure layout to use.

I also looked at my original patch now and see what I did wrong
there.

Reviewed-by: Arnd Bergmann <arnd@arndb.de>