On Mon, Sep 25, 2023 at 10:05:06AM -0300, Jason Gunthorpe wrote: > On Thu, Sep 21, 2023 at 01:58:19PM -0700, Nicolin Chen wrote: > > > Perhaps we could put all min_len defines in uAPI header, like: > > include/uapi/linux/gfs2_ondisk.h:442:#define LH_V1_SIZE (offsetofend(struct gfs2_log_header, lh_hash)) > > In this way, drivers won't need to deal with that nor have risks > > of breaking ABI by changing a min_len. > > I don't think we need constants, just be sure that every call to > iommu_copy_user_data() has an offsetof() as the last parameter. > > Indeed perhaps you should put it in a macro and force this to happen eg: > > #define iommu_copy_user_data(user_data, from, min_size_member) \ > __iommu_copy_user_data(user_data, from, offsetofend(typeof(*from), min_size_member)) > > iommu_copy_user_data(user_data, &data, iotlb); OK. And always use sizeof(typeof(*from)) in the mcaro for the current data_len, I assume? Thanks Nic
