On Tue, Jan 29, 2019 at 7:45 AM Christoph Hellwig <hch@xxxxxx> wrote: > > On Mon, Jan 28, 2019 at 06:20:08PM -0700, Jens Axboe wrote: > > Sure, that would be straight forward. Is there a strong reason to do > > so outside of "that would be nice"? It's not like it's a huge amount > > of code. > > And it would be really painful for userspace. Because now you > can't pass struct iovec through from a higher level, but will instead > of to copy the iovec to a different type in the submission path. Agreed. However, if we decide to add the in_compat_syscall() check to set_user_sigmask()/set_compat_user_sigmask(), we probably want to do the same thing in import_iovec()/compat_import_iovec() and rw_copy_check_uvector()/compat_rw_copy_check_uvector(). Arnd