Re: [PATCH] kvm: Fix warning in__kvm_gpc_refresh

[Sean Christopherson <seanjc@xxxxxxxxxx>]

On Tue, Jun 25, 2024, Pei Li wrote:
> Check for invalid hva address stored in data before calling
> kvm_gpc_activate_hva() instead of only compare with 0.
> 
> Reported-by: syzbot+fd555292a1da3180fc82@xxxxxxxxxxxxxxxxxxxxxxxxx
> Closes: https://syzkaller.appspot.com/bug?extid=fd555292a1da3180fc82
> Tested-by: syzbot+fd555292a1da3180fc82@xxxxxxxxxxxxxxxxxxxxxxxxx
> Signed-off-by: Pei Li <peili.dev@xxxxxxxxx>
> ---
> Syzbot reports a warning message in __kvm_gpc_refresh(). This warning
> requires at least one of gpa and uhva to be valid.
> WARNING: CPU: 0 PID: 5090 at arch/x86/kvm/../../../virt/kvm/pfncache.c:259 __kvm_gpc_refresh+0xf17/0x1090 arch/x86/kvm/../../../virt/kvm/pfncache.c:259
> 
> We are calling it from kvm_gpc_activate_hva(). This function always calls
> __kvm_gpc_activate() with INVALID_GPA. Thus, uhva must be valid to
> disable this warning.
> 
> This patch checks for invalid hva address as well instead of only
> comparing hva with 0 before calling kvm_gpc_activate_hva()
> 
> syzbot has tested the proposed patch and the reproducer did not trigger
> any issue.
> 
> Tested on:
> 
> commit:         55027e68 Merge tag 'input-for-v6.10-rc5' of git://git...
> git tree:       upstream
> console output: https://syzkaller.appspot.com/x/log.txt?x=16ea803a980000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=e40800950091403a
> dashboard link: https://syzkaller.appspot.com/bug?extid=fd555292a1da3180fc82
> compiler:       Debian clang version 15.0.6, GNU ld (GNU Binutils for Debian) 2.40
> patch:          https://syzkaller.appspot.com/x/patch.diff?x=16eeb53e980000
> 
> Note: testing is done by a robot and is best-effort only.
> ---
>  arch/x86/kvm/xen.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> index f65b35a05d91..de5f34492405 100644
> --- a/arch/x86/kvm/xen.c
> +++ b/arch/x86/kvm/xen.c
> @@ -881,7 +881,7 @@ int kvm_xen_vcpu_set_attr(struct kvm_vcpu *vcpu, struct kvm_xen_vcpu_attr *data)
>  			r = kvm_gpc_activate(&vcpu->arch.xen.vcpu_info_cache,
>  					     data->u.gpa, sizeof(struct vcpu_info));
>  		} else {
> -			if (data->u.hva == 0) {
> +			if (data->u.hva == 0 || kvm_is_error_hva(data->u.hva)) {
>  				kvm_gpc_deactivate(&vcpu->arch.xen.vcpu_info_cache);
>  				r = 0;
>  				break;

Hmm, I think what we want is to return -EINVAL in this case, not deactivate the
region.   I could have sworn KVM does that.  Gah, I caught
KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA during review, but missed this one.  So to fix
this immediate bug, and avoid similar issues in the future, this?

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 93814d3850eb..622fe24da910 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -741,7 +741,7 @@ int kvm_xen_hvm_set_attr(struct kvm *kvm, struct kvm_xen_hvm_attr *data)
                } else {
                        void __user * hva = u64_to_user_ptr(data->u.shared_info.hva);
 
-                       if (!PAGE_ALIGNED(hva) || !access_ok(hva, PAGE_SIZE)) {
+                       if (!PAGE_ALIGNED(hva)) {
                                r = -EINVAL;
                        } else if (!hva) {
                                kvm_gpc_deactivate(&kvm->arch.xen.shinfo_cache);
diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c
index 0ab90f45db37..728d2c1b488a 100644
--- a/virt/kvm/pfncache.c
+++ b/virt/kvm/pfncache.c
@@ -438,6 +438,9 @@ int kvm_gpc_activate(struct gfn_to_pfn_cache *gpc, gpa_t gpa, unsigned long len)
 
 int kvm_gpc_activate_hva(struct gfn_to_pfn_cache *gpc, unsigned long uhva, unsigned long len)
 {
+       if (!access_ok((void __user *)uhva, len))
+               return -EINVAL;
+
        return __kvm_gpc_activate(gpc, INVALID_GPA, uhva, len);
 }