In the Linux kernel, the following vulnerability has been resolved: KVM: Don't null dereference ops->destroy A KVM device cleanup happens in either of two callbacks: 1) destroy() which is called when the VM is being destroyed; 2) release() which is called when a device fd is closed. Most KVM devices use 1) but Book3s's interrupt controller KVM devices (XICS, XIVE, XIVE-native) use 2) as they need to close and reopen during the machine execution. The error handling in kvm_ioctl_create_device() assumes destroy() is always defined which leads to NULL dereference as discovered by Syzkaller. This adds a checks for destroy!=NULL and adds a missing release(). This is not changing kvm_destroy_devices() as devices with defined release() should have been removed from the KVM devices list by then.
https://git.kernel.org/stable/c/e91665fbbf3ccb268b268a7d71a6513538d813ac
https://git.kernel.org/stable/c/e8bc2427018826e02add7b0ed0fc625a60390ae5
https://git.kernel.org/stable/c/d4a5a79b780891c5cbdfdc6124d46fdf8d13dba1
https://git.kernel.org/stable/c/3616776bc51cd3262bb1be60cc01c72e0a1959cf
https://git.kernel.org/stable/c/170465715a60cbb7876e6b961b21bd3225469da8