Re: [PATCH v2 5/8] KVM: x86: Add WARNs to detect out-of-bounds register indices

[Vitaly Kuznetsov]

Sean Christopherson <sean.j.christopherson@xxxxxxxxx> writes:

> Add WARN_ON_ONCE() checks in kvm_register_{read,write}() to detect reg
> values that would cause KVM to overflow vcpu->arch.regs.  Change the reg
> param to an 'int' to make it clear that the reg index is unverified.
>

Hm, on multiple occasions I was thinking "an enum would do better here
but whatever" but maybe 'int' was there on purpose? Interesting... :-)

> Open code the RIP and RSP accessors so as to avoid pointless overhead of
> WARN_ON_ONCE().  Alternatively, lower-level helpers could be provided,
> but that opens the door for improper use of said helpers, and the
> ugliness of the open-coding will be slightly improved in future patches.
>
> Regarding the overhead of WARN_ON_ONCE(), now that all fixed GPR reads
> and writes use dedicated accessors, e.g. kvm_rax_read(), the overhead
> is limited to flows where the reg index is generated at runtime.  And
> there is at least one historical bug where KVM has generated an out-of-
> bounds access to arch.regs (see commit b68f3cc7d9789, "KVM: x86: Always
> use 32-bit SMRAM save state for 32-bit kernels").
>
> Adding the WARN_ON_ONCE() protection paves the way for additional
> cleanup related to kvm_reg and kvm_reg_ex.
>
> Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> ---
>  arch/x86/kvm/kvm_cache_regs.h | 30 ++++++++++++++++++++++--------
>  arch/x86/kvm/x86.h            |  6 ++----
>  2 files changed, 24 insertions(+), 12 deletions(-)
>
> diff --git a/arch/x86/kvm/kvm_cache_regs.h b/arch/x86/kvm/kvm_cache_regs.h
> index 1cc6c47dc77e..3972e1b65635 100644
> --- a/arch/x86/kvm/kvm_cache_regs.h
> +++ b/arch/x86/kvm/kvm_cache_regs.h
> @@ -37,19 +37,23 @@ BUILD_KVM_GPR_ACCESSORS(r14, R14)
>  BUILD_KVM_GPR_ACCESSORS(r15, R15)
>  #endif
>  
> -static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu,
> -					      enum kvm_reg reg)
> +static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg)
>  {
> +	if (WARN_ON_ONCE((unsigned int)reg >= NR_VCPU_REGS))
> +		return 0;
> +

(I'm just trying to think outside of the box) when this WARN fires it
means we have a bug in KVM but replacing this with BUG_ON() is probably
not justified (like other VMs on the host may be doing all
right). Propagating (and checking) errors from every such place is
probably too cumbersome so what if we introduce a flag "emit
KVM_INTERNAL_ERROR and kill the VM ASAP" and check it before launching
vCPU again? The goal is to not allow the VM to proceed because its state
is definitely invalid.

>  	if (!test_bit(reg, (unsigned long *)&vcpu->arch.regs_avail))
>  		kvm_x86_ops->cache_reg(vcpu, reg);
>  
>  	return vcpu->arch.regs[reg];
>  }
>  
> -static inline void kvm_register_write(struct kvm_vcpu *vcpu,
> -				      enum kvm_reg reg,
> +static inline void kvm_register_write(struct kvm_vcpu *vcpu, int reg,
>  				      unsigned long val)
>  {
> +	if (WARN_ON_ONCE((unsigned int)reg >= NR_VCPU_REGS))
> +		return;
> +
>  	vcpu->arch.regs[reg] = val;
>  	__set_bit(reg, (unsigned long *)&vcpu->arch.regs_dirty);
>  	__set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
> @@ -57,22 +61,32 @@ static inline void kvm_register_write(struct kvm_vcpu *vcpu,
>  
>  static inline unsigned long kvm_rip_read(struct kvm_vcpu *vcpu)
>  {
> -	return kvm_register_read(vcpu, VCPU_REGS_RIP);
> +	if (!test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_avail))
> +		kvm_x86_ops->cache_reg(vcpu, VCPU_REGS_RIP);
> +
> +	return vcpu->arch.regs[VCPU_REGS_RIP];
>  }
>  
>  static inline void kvm_rip_write(struct kvm_vcpu *vcpu, unsigned long val)
>  {
> -	kvm_register_write(vcpu, VCPU_REGS_RIP, val);
> +	vcpu->arch.regs[VCPU_REGS_RIP] = val;
> +	__set_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty);
> +	__set_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_avail);
>  }
>  
>  static inline unsigned long kvm_rsp_read(struct kvm_vcpu *vcpu)
>  {
> -	return kvm_register_read(vcpu, VCPU_REGS_RSP);
> +	if (!test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_avail))
> +		kvm_x86_ops->cache_reg(vcpu, VCPU_REGS_RSP);
> +
> +	return vcpu->arch.regs[VCPU_REGS_RSP];
>  }
>  
>  static inline void kvm_rsp_write(struct kvm_vcpu *vcpu, unsigned long val)
>  {
> -	kvm_register_write(vcpu, VCPU_REGS_RSP, val);
> +	vcpu->arch.regs[VCPU_REGS_RSP] = val;
> +	__set_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty);
> +	__set_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_avail);
>  }
>  
>  static inline u64 kvm_pdptr_read(struct kvm_vcpu *vcpu, int index)
> diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
> index dbf7442a822b..45d82b8277e5 100644
> --- a/arch/x86/kvm/x86.h
> +++ b/arch/x86/kvm/x86.h
> @@ -238,8 +238,7 @@ static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
>  	return false;
>  }
>  
> -static inline unsigned long kvm_register_readl(struct kvm_vcpu *vcpu,
> -					       enum kvm_reg reg)
> +static inline unsigned long kvm_register_readl(struct kvm_vcpu *vcpu, int reg)
>  {
>  	unsigned long val = kvm_register_read(vcpu, reg);
>  
> @@ -247,8 +246,7 @@ static inline unsigned long kvm_register_readl(struct kvm_vcpu *vcpu,
>  }
>  
>  static inline void kvm_register_writel(struct kvm_vcpu *vcpu,
> -				       enum kvm_reg reg,
> -				       unsigned long val)
> +				       int reg, unsigned long val)
>  {
>  	if (!is_64_bit_mode(vcpu))
>  		val = (u32)val;

Reviewed-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>

-- 
Vitaly