On 05/10/2017 04:24 PM, Christoph Hellwig wrote: > On Wed, May 10, 2017 at 03:20:41PM +0300, Boaz Harrosh wrote: >> Would you not want to call ->read_iter() in the NULL case >> and have all FSs supported as today? > > As IMA has particular requirements on the fs (e.g. that it can > read with i_rwsem held as seen in this patch, or useful i_version > which only the file systems converted in this patch do), having > an explicit opt-in seems much safer. This optional method is > a very easy way to provide this opt-in behavior. > >> + if (!file->f_op->integrity_read) >> + return -EBADF; > > Would you not want to call ->read_iter() in the NULL case > and have all FSs supported as today? > > Thanks > Boaz > >> >> - old_fs = get_fs(); >> - set_fs(get_ds()); >> - ret = __vfs_read(file, buf, count, &offset); >> - set_fs(old_fs); If you look here above, it used to call __vfs_read regardless of i_version and "opt-in" so it looks like a regression for all those FSs that used to work. So I did not understand, but I guess you are right opt-in seems much safer. Thanks Boaz >> + init_sync_kiocb(&kiocb, file); >> + kiocb.ki_pos = offset; >> + iov_iter_init(&iter, READ | ITER_KVEC, &iov, 1, count);
