From mboxrd@z Thu Jan 1 00:00:00 1970 From: Alex Gartrell Subject: Re: [PATCH net-next 1/2] socket: Allow external sockets to use socket syscalls Date: Fri, 26 Dec 2014 14:26:24 -0500 Message-ID: <549DB660.8040606@fb.com> References: <1419576624-8999-1-git-send-email-agartrell@fb.com> <1419576624-8999-2-git-send-email-agartrell@fb.com> <549D2E51.3040200@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset="windows-1252"; format=flowed Content-Transfer-Encoding: 7bit Cc: , , To: Jason Wang , , Return-path: Received: from mx0b-00082601.pphosted.com ([67.231.153.30]:18433 "EHLO mx0b-00082601.pphosted.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751127AbaLZT0g (ORCPT ); Fri, 26 Dec 2014 14:26:36 -0500 In-Reply-To: <549D2E51.3040200@redhat.com> Sender: netdev-owner@vger.kernel.org List-ID: Hello Jason, Thanks again for your comments. On 12/26/14 4:45 AM, Jason Wang wrote: >> @@ -388,6 +388,7 @@ struct file *sock_alloc_file(struct socket *sock, int flags, const char *dname) >> sock->file = file; >> file->f_flags = O_RDWR | (flags & O_NONBLOCK); >> file->private_data = sock; >> + file->private_data_is_socket = true; > > This is only safe if all user of sock_alloc_file() have full support for > each method in proto_ops. This doesn't change anything in the invocation of the syscalls, as every file allocated through this path already passes the "is_socket" test due to the (sadly missing here) file = alloc_file(..., &socket_file_ops) above, so this makes things no less safe than they were. Of course we also need to implement these proto_ops for the tun device in the subsequent patch to make it work. >> return file; >> } >> EXPORT_SYMBOL(sock_alloc_file); >> @@ -411,7 +412,7 @@ static int sock_map_fd(struct socket *sock, int flags) >> >> struct socket *sock_from_file(struct file *file, int *err) >> { >> - if (file->f_op == &socket_file_ops) >> + if (file->private_data_is_socket) >> return file->private_data; /* set in sock_map_fd */ >> >> *err = -ENOTSOCK; > > Not sure it's the best method, how about a dedicated f_op to do this? So like a get_socket operation? That would simplify this a lot, certainly (we wouldn't need to move private_data around in the tun driver). Thanks, -- Alex Gartrell