Re: [PATCH] powerpc: Fix sigset_t copy

[Christophe Leroy <christophe.leroy@csgroup.eu>]

Le 10/11/2021 à 00:47, Finn Thain a écrit :
> From: Christophe Leroy <christophe.leroy@csgroup.eu>
> 
> The conversion from __copy_from_user() to __get_user() introduced a
> regression in __get_user_sigset() in v5.13. The bug was subsequently
> copied and pasted in unsafe_get_user_sigset().
> 
> The regression was reported by users of the Xorg packages distributed in
> Debian/powerpc --
> 
>      "The symptoms are that the fb screen goes blank, with the backlight
>      remaining on and no errors logged in /var/log; wdm (or startx) run
>      with no effect (I tried logging in in the blind, with no effect).
>      And they are hard to kill, requiring 'kill -KILL ...'"
> 
> Fix the regression by casting the __get_user() assignment lvalue to u64
> so that the entire struct gets copied.
> 
> Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
> Cc: Christopher M. Riedl <cmr@bluescreens.de>
> Link: https://lore.kernel.org/linuxppc-dev/FEtBUOuFPMN4zJy4bIOqz6C4xoliCbTxS7VtMKD6UZkbvEbycUceRgGAd7e9-trRdwVN3hWAbQi0qrNx8Zgn8niTQf2KPVdw-W35czDIaeQ=@protonmail.com/
> Fixes: 887f3ceb51cd ("powerpc/signal32: Convert do_setcontext[_tm]() to user access block")
> Fixes: d3ccc9781560 ("powerpc/signal: Use __get_user() to copy sigset_t")
> Reported-and-tested-by: Stan Johnson <userm57@yahoo.com>
> Signed-off-by: Finn Thain <fthain@linux-m68k.org>
> ---

Hi Finn,

> Christophe, I hope this change is the one you wanted to see upstream (?).
> If it is acceptable please add your signed-off-by tag.

I'm on holidays, I was planing to handle this next week.

Only PPC64 uses __get_user_sigset() on mainline so I don't think it is 
worth modifying it. If we decide to modify it anyway in mainline, it 
should be another patch that can be backported without additional effort.

For unsafe_get_user_sigset(), as we don't have the KUAP overhead that we 
had with __get_user(), I'd prefer we simply perform two 32 bits 
unsafe_get_user(), one for sig[0] and one for sig[1], instead of all 
those casts to u64.

Thanks anyway for the detailed description of the problem.
Christophe

> ---
>   arch/powerpc/kernel/signal.h | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/powerpc/kernel/signal.h b/arch/powerpc/kernel/signal.h
> index 1f07317964e4..44e736b88e91 100644
> --- a/arch/powerpc/kernel/signal.h
> +++ b/arch/powerpc/kernel/signal.h
> @@ -23,10 +23,10 @@ static inline int __get_user_sigset(sigset_t *dst, const sigset_t __user *src)
>   {
>   	BUILD_BUG_ON(sizeof(sigset_t) != sizeof(u64));
>   
> -	return __get_user(dst->sig[0], (u64 __user *)&src->sig[0]);
> +	return __get_user(*(u64 *)&dst->sig[0], (u64 __user *)&src->sig[0]);
>   }
>   #define unsafe_get_user_sigset(dst, src, label) \
> -	unsafe_get_user((dst)->sig[0], (u64 __user *)&(src)->sig[0], label)
> +	unsafe_get_user(*(u64 *)&(dst)->sig[0], (u64 __user *)&(src)->sig[0], label)
>   
>   #ifdef CONFIG_VSX
>   extern unsigned long copy_vsx_to_user(void __user *to,
>