From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S934919AbaGXKYP (ORCPT ); Thu, 24 Jul 2014 06:24:15 -0400 Received: from mga11.intel.com ([192.55.52.93]:20975 "EHLO mga11.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755462AbaGXKYM (ORCPT ); Thu, 24 Jul 2014 06:24:12 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.01,723,1400050800"; d="scan'208";a="574728549" Message-ID: <53D0DEBE.2030502@intel.com> Date: Thu, 24 Jul 2014 18:23:58 +0800 From: "xinhui.pan" User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.6.0 MIME-Version: 1.0 To: Peter Hurley CC: Greg KH , mnipxh , jslaby@suse.cz, linux-kernel@vger.kernel.org, yanmin_zhang@linux.intel.com Subject: Re: [PATCH] tty/tty_io.c: make a check before reuse cdev References: <53CD0BD4.4050007@gmail.com> <20140721153840.GA6802@kroah.com> <53CE5097.2000502@intel.com> <53CE941B.1030102@hurleysoftware.com> <53CF7EAC.8010201@intel.com> <53CFDD05.4010104@hurleysoftware.com> In-Reply-To: <53CFDD05.4010104@hurleysoftware.com> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org hi, Peter 于 2014年07月24日 00:04, Peter Hurley 写道: > Hi Xinhui, > > On 07/23/2014 05:21 AM, xinhui.pan wrote: >> 于 2014年07月23日 00:40, Peter Hurley 写道: >>> On 07/22/2014 07:52 AM, xinhui.pan wrote: >>>> 于 2014年07月21日 23:38, Greg KH 写道: >>>>> On Mon, Jul 21, 2014 at 08:47:16PM +0800, pp wrote: > >>>>>> tty driver register its device and (D)init the cdevs again. >>>>> >>>>> What driver does this with an "old" device, it should have created a new >>>>> one, otherwise, as you have pointed out, it's a bug. >>>>> >>>> >>>> I can't agree more with you. we should not use "old" device. >>> >>> This is a gsm driver problem. The GSM driver is reusing device indexes >>> for still-open ttys. >>> >>> The GSM driver uses a global table, gsm_mux[], to allocate device indexes >>> but prematurely clears the table entry in gsm_mux_cleanup(). If instead, >>> clearing the gsm_mux table entry were deferred to gsm_mux_free(), then >>> device indexes would not be getting reused until after the last tty >>> associated with the last gsm attach was closed. >>> >> >> Very nice solution. We will check if this can cause any risk, both to kernel and user space. >> Using a new tty base to register with new cdevs may give us more chance to wait PROCESS quit/close. >> when total 256 tty used up, what we should do is still in discuss. > > I saw your patch for the use of gsm->num before gsm_activate_mux() has > allocated the table entry; thanks for fixing that. > > As for what to do if all the gsm_mux table entries are used: if the error > is infrequent, I suggest simply returning an error which is what the > driver does currently. Otherwise, a more dynamic allocation scheme may be required. > Yes, agree with you. This error is infrequent. Intel software will close the gsmtty after read/write hit errors. We don't need dynamic allocation. So keep returning an error here. :) > I did notice while reviewing the error handling that gsmld_open() will > leak the entire composite ldisc data allocated by gsm_alloc_mux() if > gsmld_attach_gsm() fails. > Thanks for your reviewing. Do you mean gsm = gsm_alloc_mux() will cause leak if gsmld_attach_gsm fails? As there is no gsm_free_mux()? If so, Yes, it is. In such scenario, gsmld_close is not called. and gsm_free_mux is not called, either. Thanks for your nice comments. You really help us fix several ugly issues. Let me have a deep think about it. thanks, xinhui > Regards, > Peter Hurley >