Re: [PATCH] fix infoleak in mm

[Catalin Marinas]

On Tue, May 03, 2016 at 04:36:08PM -0400, Kangjie Lu wrote:
> The stack object âsiâ has a total size of 128; however, only 20
> bytes are initialized. The remaining uninitialized bytes are sent
> to userland via send_signal
> 
> Signed-off-by: Kangjie Lu <kjlu@xxxxxxxxxx>
> ---
>  arch/arm64/mm/fault.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
> index 95df28b..f790eda 100644
> --- a/arch/arm64/mm/fault.c
> +++ b/arch/arm64/mm/fault.c
> @@ -117,6 +117,7 @@ static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
>  {
>  	struct siginfo si;
>  
> +	memset(&si, 0, sizeof(si));
>  	if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
>  		pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x\n",
>  			tsk->comm, task_pid_nr(tsk), fault_name(esr), sig,

I'm not convinced this is necessary. Have you actually seen such
information leak getting to user space? The actual writing of siginfo to
the user stack happens in copy_siginfo_to_user() (called from
setup_rt_frame) which should (at least in theory) only copy
pre-populated fields.

-- 
Catalin