[PATCH 15/34] KVM: Add KVM_CREATE_GUEST_MEMFD ioctl() for guest-specific backing memory

[Xiaoyao Li]

On 11/11/2023 2:22 AM, Sean Christopherson wrote:
> On Fri, Nov 10, 2023, Xiaoyao Li wrote:
>> On 11/6/2023 12:30 AM, Paolo Bonzini wrote:
>>> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
>>> index 68a144cb7dbc..a6de526c0426 100644
>>> --- a/include/linux/kvm_host.h
>>> +++ b/include/linux/kvm_host.h
>>> @@ -589,8 +589,20 @@ struct kvm_memory_slot {
>>>    	u32 flags;
>>>    	short id;
>>>    	u16 as_id;
>>> +
>>> +#ifdef CONFIG_KVM_PRIVATE_MEM
>>> +	struct {
>>> +		struct file __rcu *file;
>>> +		pgoff_t pgoff;
>>> +	} gmem;
>>> +#endif
>>>    };
>>> +static inline bool kvm_slot_can_be_private(const struct kvm_memory_slot *slot)
>>> +{
>>> +	return slot && (slot->flags & KVM_MEM_GUEST_MEMFD);
>>> +}
>>> +
>>
>> maybe we can move this block and ...
>>
>> <snip>
>>
>>> @@ -2355,6 +2379,30 @@ bool kvm_arch_pre_set_memory_attributes(struct kvm *kvm,
>>>    					struct kvm_gfn_range *range);
>>>    bool kvm_arch_post_set_memory_attributes(struct kvm *kvm,
>>>    					 struct kvm_gfn_range *range);
>>> +
>>> +static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
>>> +{
>>> +	return IS_ENABLED(CONFIG_KVM_PRIVATE_MEM) &&
>>> +	       kvm_get_memory_attributes(kvm, gfn) & KVM_MEMORY_ATTRIBUTE_PRIVATE;
>>> +}
>>> +#else
>>> +static inline bool kvm_mem_is_private(struct kvm *kvm, gfn_t gfn)
>>> +{
>>> +	return false;
>>> +}
>>>    #endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
>>
>> this block to Patch 18?
> 
> It would work, but my vote is to keep them here to minimize the changes to common
> KVM code in the x86 enabling.  It's not a strong preference though.  Of course,
> at this point, fiddling with this sort of thing is probably a bad idea in terms
> of landing guest_memfd.

Indeed. It's OK then.

>>> @@ -4844,6 +4875,10 @@ static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
>>>    #ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
>>>    	case KVM_CAP_MEMORY_ATTRIBUTES:
>>>    		return kvm_supported_mem_attributes(kvm);
>>> +#endif
>>> +#ifdef CONFIG_KVM_PRIVATE_MEM
>>> +	case KVM_CAP_GUEST_MEMFD:
>>> +		return !kvm || kvm_arch_has_private_mem(kvm);
>>>    #endif
>>>    	default:
>>>    		break;
>>> @@ -5277,6 +5312,18 @@ static long kvm_vm_ioctl(struct file *filp,
>>>    	case KVM_GET_STATS_FD:
>>>    		r = kvm_vm_ioctl_get_stats_fd(kvm);
>>>    		break;
>>> +#ifdef CONFIG_KVM_PRIVATE_MEM
>>> +	case KVM_CREATE_GUEST_MEMFD: {
>>> +		struct kvm_create_guest_memfd guest_memfd;
>>
>> Do we need a guard of below?
>>
>> 		r = -EINVAL;
>> 		if (!kvm_arch_has_private_mem(kvm))
>> 			goto out;
> 
> Argh, yeah, that's weird since KVM_CAP_GUEST_MEMFD says "not supported" if the
> VM doesn't support private memory.
> 
> Enforcing that would break guest_memfd_test.c though.  And having to create a
> "special" VM just to test basic guest_memfd functionality would be quite
> annoying.
> 
> So my vote is to do:
> 
> 	case KVM_CAP_GUEST_MEMFD:
> 		return IS_ENABLED(CONFIG_KVM_PRIVATE_MEM);

I'm fine with it.

> There's no harm to KVM if userspace creates a file it can't use, and at some
> point KVM will hopefully support guest_memfd irrespective of private memory.