Re: [PATCH] x86/fault: Fix wrong signal when vsyscall fails with pkey

[Dave Hansen]

On 7/28/21 4:44 AM, Jiashuo Liang wrote:
> The function __bad_area_nosemaphore calls kernelmode_fixup_or_oops with
> parameter "signal" being "pkey", which will send a signal numbered "pkey".

Yikes.

> When emulating vsyscall, the kernel may fail to access user-given memory
> pages that are protected by pkey. In such a case, the kernel should send a
> SIGSEGV signal with si_code=SEGV_PKUERR and si_pkey=pkey.

This could use a bit more context.

First of all this is for user address space faults in the
do_user_addr_fault() path.  Second, the buggy code is under a
!user_mode() check, so this must be a kernel fault in the user address
space.  Third, the only notice this problem when the page fault handler
ends up delivering a signal as a result of the fault.  Most cases will
simply return an error code to the faulting kernel code which will see
-EFAULT come back from copy_to/from_user() and friends.

The *only* condition in which we generate that signal from the fault
handler is when current->thread.sig_on_uaccess_err=1, and the only place
that gets used is in emulate_vsyscall().

This makes me want to add some code that tickles vsyscall emulation in
the pkey selftests, but I think I'll resist the urge for now. :)

Is that all correct?

> So a new parameter "pkey" is added to kernelmode_fixup_or_oops to fix it.

Yeah, I think that's the right fix.  You also need this:

Fixes: 5042d40a264c ("x86/fault: Bypass no_context() for implicit kernel
faults from usermode")

I believe that's where this issue originated.

How did you find this, by the way?

>  arch/x86/mm/fault.c | 23 +++++++++++++++--------
>  1 file changed, 15 insertions(+), 8 deletions(-)
> 
> diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
> index b2eefdefc108..883294282e1e 100644
> --- a/arch/x86/mm/fault.c
> +++ b/arch/x86/mm/fault.c
> @@ -710,7 +710,8 @@ page_fault_oops(struct pt_regs *regs, unsigned long error_code,
>  
>  static noinline void
>  kernelmode_fixup_or_oops(struct pt_regs *regs, unsigned long error_code,
> -			 unsigned long address, int signal, int si_code)
> +			 unsigned long address, int signal, int si_code,
> +			 u32 pkey)
>  {
>  	WARN_ON_ONCE(user_mode(regs));
>  
> @@ -735,8 +736,12 @@ kernelmode_fixup_or_oops(struct pt_regs *regs, unsigned long error_code,
>  
>  			set_signal_archinfo(address, error_code);
>  
> -			/* XXX: hwpoison faults will set the wrong code. */
> -			force_sig_fault(signal, si_code, (void __user *)address);
> +			if (si_code == SEGV_PKUERR) {
> +				force_sig_pkuerr((void __user *)address, pkey);
> +			} else {
> +				/* XXX: hwpoison faults will set the wrong code. */
> +				force_sig_fault(signal, si_code, (void __user *)address);
> +			}
>  		}
>  
>  		/*
> @@ -798,7 +803,8 @@ __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
>  	struct task_struct *tsk = current;
>  
>  	if (!user_mode(regs)) {
> -		kernelmode_fixup_or_oops(regs, error_code, address, pkey, si_code);
> +		kernelmode_fixup_or_oops(regs, error_code, address,
> +					 SIGSEGV, si_code, pkey);
>  		return;
>  	}
>  
> @@ -930,7 +936,8 @@ do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
>  {
>  	/* Kernel mode? Handle exceptions or die: */
>  	if (!user_mode(regs)) {
> -		kernelmode_fixup_or_oops(regs, error_code, address, SIGBUS, BUS_ADRERR);
> +		kernelmode_fixup_or_oops(regs, error_code, address,
> +					 SIGBUS, BUS_ADRERR, 0);
>  		return;
>  	}

Could we please use ARCH_DEFAULT_PKEY instead of 0's in all these call
sites?  I just detest seeing mystery functions with lots of 0's and 1's
as parameters.