From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga04.intel.com ([192.55.52.120]:48090 "EHLO mga04.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S934537AbdBQPvj (ORCPT ); Fri, 17 Feb 2017 10:51:39 -0500 Date: Fri, 17 Feb 2017 08:41:57 -0700 From: Scott Bauer To: Keith Busch Cc: Christoph Hellwig , jonathan.derrick@intel.com, axboe@fb.com, linux-block@vger.kernel.org, linux-nvme@lists.infradead.org Subject: Re: [PATCH 4/4] nvme: re-check security protocol support after reset Message-ID: <20170217154157.GA2254@sbauer-Z170X-UD5> References: <20170217125941.14319-1-hch@lst.de> <20170217125941.14319-5-hch@lst.de> <20170217152651.GA18275@localhost.localdomain> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <20170217152651.GA18275@localhost.localdomain> Sender: linux-block-owner@vger.kernel.org List-Id: linux-block@vger.kernel.org On Fri, Feb 17, 2017 at 10:26:51AM -0500, Keith Busch wrote: > On Fri, Feb 17, 2017 at 01:59:41PM +0100, Christoph Hellwig wrote: > > @@ -1789,7 +1789,8 @@ static void nvme_reset_work(struct work_struct *work) > > if (result) > > goto out; > > > > - if ((dev->ctrl.oacs & NVME_CTRL_OACS_SEC_SUPP) && !dev->ctrl.opal_dev) { > > + kfree(dev->ctrl.opal_dev); > > + if (dev->ctrl.oacs & NVME_CTRL_OACS_SEC_SUPP) { > > dev->ctrl.opal_dev = > > init_opal_dev(&dev->ctrl, &nvme_sec_submit); > > } > > A couple things. > > This has a use-after-free in opal_unlock_from_suspend if the nvme > device had an opal_dev before, but no longer support the capability > after resume. So you'd want to set ctrl.opal_dev to NULL after the free. > > But we don't want to unconditionally free it anyway during resume > since opal_unlock_from_suspend requires the exisiting opal_dev state > information saved in the 'unlk_list'. > > Something like this instead: > > --- > diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c > index ddc51ad..8fa6be9 100644 > --- a/drivers/nvme/host/pci.c > +++ b/drivers/nvme/host/pci.c > @@ -1789,13 +1789,17 @@ static void nvme_reset_work(struct work_struct *work) > if (result) > goto out; > > - if ((dev->ctrl.oacs & NVME_CTRL_OACS_SEC_SUPP) && !dev->ctrl.opal_dev) { > - dev->ctrl.opal_dev = > - init_opal_dev(&dev->ctrl, &nvme_sec_submit); > + if (dev->ctrl.oacs & NVME_CTRL_OACS_SEC_SUPP) > + if (was_suspend && dev->ctrl.opal_dev) > + opal_unlock_from_suspend(dev->ctrl.opal_dev); > + else if (!dev->ctrl.opal_dev) > + dev->ctrl.opal_dev = > + init_opal_dev(&dev->ctrl, &nvme_sec_submit); > + } else { > + kfree(dev->ctrl.opal_dev); > + dev->ctrl.opal_dev = NULL; Keith's comments made me realize something even deeper as well. Assuming the firmware changed and we no longer support security commands we need to free the opal_dev structure like we're doing but there is a possiblity that there were saved ranges in the structure that we need to free as well. If the user had previously told the kernel to unlock 5 ranges coming out of a suspend there are 5 structures we need to free inside the opal_dev before we free the opal dev. We'll need to re-introduce free_opal_dev() in the opal code like we had a while back.