The MMIO sign-extension logic in kvm_handle_mmio_return can
trigger an integer overflow or undefined behavior when len
is invalid (e.g., len == 0 or len exceeds the size of unsigned
long). Specifically, the expression (len * 8) - 1 may result
in an out-of-bounds shift in the computation of the mask.

This patch adds validation to ensure len is greater than
0 and less than the size of unsigned long before performing
the sign-extension logic. If len falls outside this range,
the problematic logic is skipped, preventing potential issues.

Signed-off-by: Steven Davis <goldside000@outlook.com>
---
 arch/arm64/kvm/mmio.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/kvm/mmio.c b/arch/arm64/kvm/mmio.c
index ab365e839..e4132dbc3 100644
--- a/arch/arm64/kvm/mmio.c
+++ b/arch/arm64/kvm/mmio.c
@@ -124,12 +124,15 @@ int kvm_handle_mmio_return(struct kvm_vcpu *vcpu)
 		len = kvm_vcpu_dabt_get_as(vcpu);
 		data = kvm_mmio_read_buf(run->mmio.data, len);
 
-		if (kvm_vcpu_dabt_issext(vcpu) &&
-		    len < sizeof(unsigned long)) {
-			mask = 1U << ((len * 8) - 1);
-			data = (data ^ mask) - mask;
+		if (kvm_vcpu_dabt_issext(vcpu)) {
+			if (len > 0 && len < sizeof(unsigned long)) {
+				mask = 1U << ((len * 8) - 1);
+				data = (data ^ mask) - mask;
+			}
 		}
 
+
+
 		if (!kvm_vcpu_dabt_issf(vcpu))
 			data = data & 0xffffffff;
 
-- 
2.39.5

On Sat, 28 Dec 2024 02:01:27 +0000,
Steven Davis <goldside000@outlook.com> wrote:
> 
> The MMIO sign-extension logic in kvm_handle_mmio_return can
> trigger an integer overflow or undefined behavior when len
> is invalid (e.g., len == 0 or len exceeds the size of unsigned
> long). Specifically, the expression (len * 8) - 1 may result
> in an out-of-bounds shift in the computation of the mask.
> 
> This patch adds validation to ensure len is greater than
> 0 and less than the size of unsigned long before performing
> the sign-extension logic. If len falls outside this range,
> the problematic logic is skipped, preventing potential issues.

I'd be curious to understand how you came to this conclusion, given
how len is computed. If anything, we could *remove* some of the
checks, rather than adding additional ones.

Also, "skipping" things is rarely an acceptable behaviour when
emulating hardware behaviour.

	M.

-- 
Without deviation from the norm, progress is not possible.

On Sat, Dec 28, 2024 at 02:01:27AM +0000, Steven Davis wrote:
> The MMIO sign-extension logic in kvm_handle_mmio_return can
> trigger an integer overflow or undefined behavior when len
> is invalid (e.g., len == 0 or len exceeds the size of unsigned
> long). Specifically, the expression (len * 8) - 1 may result
> in an out-of-bounds shift in the computation of the mask.

I don't believe we need this. len is known to be nonzero and at most 8,
which is already being tested for in the existing condition. See the
definition of kvm_vcpu_dabt_get_as() if you're curious why that is.

-- 
Thanks,
Oliver