Re: [RFC PATCH] iommufd: Destroy vdevice on device unbind

[Aneesh Kumar K . V]

Jason Gunthorpe <jgg@xxxxxxxx> writes:

> On Wed, Jun 18, 2025 at 10:59:00AM +0530, Aneesh Kumar K.V wrote:
>> Jason Gunthorpe <jgg@xxxxxxxx> writes:
>> 
>> > On Tue, Jun 17, 2025 at 01:37:04PM +0530, Aneesh Kumar K.V wrote:
>> >  
>> >> How do we reclaim that object id for further reuse? 
>> >
>> > Maybe just don't? Userspace did something it shouldn't, it now leaked
>> > 8 bytes of kernel memory until the FD is closed.
>> >
>> 
>> Between the two sequences below, Sequence 1 is the correct one, since we
>> want the object ID to be released after calling ioctl(DESTROY,
>> vdevice_id), right?
>> 
>> Sequence 1 (Correct):
>> 
>> close(vfio_cdev)          → triggers vdevice destruction  
>> ioctl(DESTROY, vdevice_id) → reclaims vdevice object ID  
>> close(iommufd)  
>
> This is wrong, the vdevice has outlived the idevice
>
>> Sequence 2:
>> 
>> ioctl(DESTROY, vdevice_id) → returns EBUSY
>
> It should not return EBUSY, it should destry the vdevice.
>
> The full sequence I would expect a sane userspace to do is:
>
> open(vfio_cdev)
> ioctl(vfio_cdev, VFIO_DEVICE_BIND_IOMMUFD, iommufd)
> ioctl(iommufd, IOMMUFD_CMD_VIOMMU_ALLOC)
> ioctl(iommufd, IOMMUFD_CMD_VDEVICE_ALLOC)
> ioctl(iommufd, IOMMUFD_CMD_VDEVICE_DEALLOC)
> ioctl(iommufd, IOMMUFD_CMD_VIOMMU_DEALLOC)
> close(vfio_cdev);
>

And if the user does

open(vfio_cdev)
ioctl(vfio_cdev, VFIO_DEVICE_BIND_IOMMUFD, iommufd)
ioctl(iommufd, IOMMUFD_CMD_VIOMMU_ALLOC)
ioctl(iommufd, IOMMUFD_CMD_VDEVICE_ALLOC)
close(vfio_cdev);   -> this should call vdevice_destroy because idevice is getting destroyed here (we will put XA_ZERO_ENTRY here).
ioctl(iommufd, IOMMUFD_CMD_VDEVICE_DEALLOC) -> No error, we convert the XA_ZERO_ENTRY to NULL here?
ioctl(iommufd, IOMMUFD_CMD_VIOMMU_DEALLOC)

-aneesh