Re: [PATCH v2 10/10] KVM: Pre-check uaccesses in KVM's APIs to read/write guest memory

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] kvm: bounds check bypass in kvm_is_guest_access_ok due to integer truncation and size ignore
--

commit 3b0fbcae79b36e79408bde507ed9fb555ab9f477
Author: Sean Christopherson <seanjc@google.com>

KVM: Pre-check uaccesses in KVM's APIs to read/write guest memory

This commit adds an early check for user access validity in KVM's guest
memory APIs. It introduces kvm_is_guest_access_ok() to validate access
bounds before performing memslot lookups, aiming to catch bugs during VM
teardown more noisily.

> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 770a2c3bd4458..80b19d47840b2 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
[ ... ]
> @@ -3211,6 +3211,11 @@ static int next_segment(unsigned long len, int offset)
>  		return len;
>  }
>  
> +static bool kvm_is_guest_access_ok(struct kvm *kvm, int offset, int len, int size)
> +{
> +	return kvm_can_do_uaccess(kvm) && !WARN_ON_ONCE(offset + len > PAGE_SIZE);
> +}

[Severity: Medium]
Does this helper ignore the provided size parameter? 

By hardcoding PAGE_SIZE in the WARN_ON_ONCE check, callers that pass a
smaller boundary (such as ghc->len) will have their custom bounds check
bypassed. If offset + len exceeds the custom size limit but remains under
PAGE_SIZE, it appears the access would be incorrectly permitted.

[ ... ]
> @@ -3294,7 +3299,7 @@ static int __kvm_read_guest_atomic(struct kvm *kvm, struct kvm_memory_slot *slot
>  	int r;
>  	unsigned long addr;
>  
> -	if (WARN_ON_ONCE(offset + len > PAGE_SIZE))
> +	if (!kvm_is_guest_access_ok(kvm, offset, len, PAGE_SIZE))

[Severity: Medium]
Does this silently truncate the len parameter?

In __kvm_read_guest_atomic(), len is defined as an unsigned long, but
kvm_is_guest_access_ok() takes len as a signed int.

>  		return -EFAULT;
>  
>  	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
[ ... ]
> @@ -3457,7 +3462,7 @@ int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
>  	int r;
>  	gpa_t gpa = ghc->gpa + offset;
>  
> -	if (WARN_ON_ONCE(len + offset > ghc->len))
> +	if (!kvm_is_guest_access_ok(kvm, offset, len, ghc->len))

[Severity: Medium]
Similar to the above, len is an unsigned long here, which gets truncated
when passed to the signed int argument of the helper. 

Additionally, ghc->len is passed as the size parameter, which is completely
ignored by kvm_is_guest_access_ok() in favor of PAGE_SIZE. Will this bypass
the intended cache length check?

>  		return -EINVAL;
>  
>  	if (slots->generation != ghc->generation) {

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261001202234.3794060-1-seanjc@google.com?part=10