Patchwork Fix msr check in compat_sys_swapcontext

login
register
mail settings
Submitter Andreas Schwab
Date Nov. 5, 2008, 12:30 p.m.
Message ID <jeprlabbi6.fsf@sykes.suse.de>
Download mbox | patch
Permalink /patch/7326/
State Superseded, archived
Headers show

Comments

Andreas Schwab - Nov. 5, 2008, 12:30 p.m.
The new context may not be 16-byte aligned, so the real address of the
mcontext structure should be read from the uc_regs pointer instead of
directly using the (unaligned) uc_mcontext field.

Signed-off-by: Andreas Schwab <schwab@suse.de>

---
Paul Mackerras - Nov. 6, 2008, 12:52 a.m.
Andreas Schwab writes:

> The new context may not be 16-byte aligned, so the real address of the
> mcontext structure should be read from the uc_regs pointer instead of
> directly using the (unaligned) uc_mcontext field.

Good catch, but...

> @@ -941,9 +941,17 @@ long sys_swapcontext(struct ucontext __user *old_ctx,
>  #ifdef CONFIG_PPC64
>  	unsigned long new_msr = 0;
>  
> -	if (new_ctx &&
> -	    get_user(new_msr, &new_ctx->uc_mcontext.mc_gregs[PT_MSR]))
> -		return -EFAULT;
> +	if (new_ctx) {
> +		struct mcontext __user *mcp;
> +		u32 cmcp;
> +
> +		/* Get pointer to the real mcontext. */
> +		if (__get_user(cmcp, &new_ctx->uc_regs))

we need to use get_user, not __get_user, since we haven't done an
access_ok() check on the address.

> +			return -EFAULT;
> +		mcp = (struct mcontext __user *)(u64)cmcp;
> +		if (__get_user(new_msr, &mcp->mc_gregs[PT_MSR]))

ditto here.

Paul.
Andreas Schwab - Nov. 6, 2008, 9:23 a.m.
Paul Mackerras <paulus@samba.org> writes:

> we need to use get_user, not __get_user, since we haven't done an
> access_ok() check on the address.

The address is always ok since its a compat pointer, see do_setcontext.

Andreas.
Paul Mackerras - Nov. 6, 2008, 10:36 a.m.
Andreas Schwab writes:

> Paul Mackerras <paulus@samba.org> writes:
> 
> > we need to use get_user, not __get_user, since we haven't done an
> > access_ok() check on the address.
> 
> The address is always ok since its a compat pointer, see do_setcontext.

OK, since it's inside a CONFIG_PPC64 block.  I'll add a paragraph to
the patch description pointing that out.

Paul.

Patch

diff --git a/arch/powerpc/kernel/signal_32.c b/arch/powerpc/kernel/signal_32.c
index a6a4310..dc10720 100644
--- a/arch/powerpc/kernel/signal_32.c
+++ b/arch/powerpc/kernel/signal_32.c
@@ -941,9 +941,17 @@  long sys_swapcontext(struct ucontext __user *old_ctx,
 #ifdef CONFIG_PPC64
 	unsigned long new_msr = 0;
 
-	if (new_ctx &&
-	    get_user(new_msr, &new_ctx->uc_mcontext.mc_gregs[PT_MSR]))
-		return -EFAULT;
+	if (new_ctx) {
+		struct mcontext __user *mcp;
+		u32 cmcp;
+
+		/* Get pointer to the real mcontext. */
+		if (__get_user(cmcp, &new_ctx->uc_regs))
+			return -EFAULT;
+		mcp = (struct mcontext __user *)(u64)cmcp;
+		if (__get_user(new_msr, &mcp->mc_gregs[PT_MSR]))
+			return -EFAULT;
+	}
 	/*
 	 * Check that the context is not smaller than the original
 	 * size (with VMX but without VSX)