From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933159AbdKFT32 (ORCPT ); Mon, 6 Nov 2017 14:29:28 -0500 Received: from mail.kernel.org ([198.145.29.99]:34928 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932576AbdKFT3Z (ORCPT ); Mon, 6 Nov 2017 14:29:25 -0500 DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org ED55C218D0 Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=kernel.org Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=helgaas@kernel.org Date: Mon, 6 Nov 2017 13:29:23 -0600 From: Bjorn Helgaas To: Pankaj Dubey Cc: Kishon Vijay Abraham I , linux-pci@vger.kernel.org, linux-kernel@vger.kernel.org, Bjorn Helgaas Subject: Re: [PATCH] PCI: endpoint: handle probable NULL pointer access Message-ID: <20171106192923.GD31930@bhelgaas-glaptop.roam.corp.google.com> References: <1507780677-7983-1-git-send-email-pankaj.dubey@samsung.com> <20171024200200.GE21840@bhelgaas-glaptop.roam.corp.google.com> <6ee65418-8589-01c1-2156-67fbf881260a@ti.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sat, Oct 28, 2017 at 04:13:56PM +0530, Pankaj Dubey wrote: > On 25 October 2017 at 17:32, Kishon Vijay Abraham I wrote: > > Hi, > > > > On Wednesday 25 October 2017 01:32 AM, Bjorn Helgaas wrote: > >> On Thu, Oct 12, 2017 at 09:27:57AM +0530, Pankaj Dubey wrote: > >>> controller_group allocation in pci_ep_cfs_init function can fail > >>> so we should have a check while using it in pci_ep_cfs_add_epc_group > >>> for registering group, else we will hit NULL pointer access. > >>> > >>> This patch adds required check for the same and returns -EPROBE_DEFER, > >>> so that endpoint controller driver probe can be reattempted later > >>> in case controller_group is not allocated because pci_ep_cfs_init is > >>> not yet called. > >>> > >>> Signed-off-by: Pankaj Dubey > >> > >> Looking for Kishon's ack here. > >> > >>> --- > >>> drivers/pci/endpoint/pci-ep-cfs.c | 7 ++++++- > >>> drivers/pci/endpoint/pci-epc-core.c | 4 ++++ > >>> 2 files changed, 10 insertions(+), 1 deletion(-) > >>> > >>> diff --git a/drivers/pci/endpoint/pci-ep-cfs.c b/drivers/pci/endpoint/pci-ep-cfs.c > >>> index 424fdd6..3cac818 100644 > >>> --- a/drivers/pci/endpoint/pci-ep-cfs.c > >>> +++ b/drivers/pci/endpoint/pci-ep-cfs.c > >>> @@ -172,7 +172,12 @@ struct config_group *pci_ep_cfs_add_epc_group(const char *name) > >>> group = &epc_group->group; > >>> > >>> config_group_init_type_name(group, name, &pci_epc_type); > >>> - ret = configfs_register_group(controllers_group, group); > >>> + > >>> + if (controllers_group) > >>> + ret = configfs_register_group(controllers_group, group); > >>> + else > >>> + ret = -EPROBE_DEFER; > >>> + > > > > Do you ever face this issue? > > Yes, I was adding support for PCIe endpoint in Exynos driver and if we > see pci-exynos.c, > platform_driver_probe is called via subsys_initcall, which will happen > much before that module_init > and during endpoint probe sequence I got kernel panic for NULL pointer access. > > > Ideally controllers_group should always be > > initialized if the dependencies are modeled properly. > > Ideally Yes. > > But we can't ignore error cases. Even though dependencies are modeled properly, > this check is mandatory, if we see pci_ep_cfs_init function where > "controllers_group" is suppose > to be allocated via call to "configfs_register_default_group", is > prone to failure as allocated via > kzalloc. We are handling error condition in pci_ep_cfs_init if it > fails to allocate "controllers_group" > but during EP initialization sequence, there is no check on > "controllers_group" pointer in > "configfs_register_default_group" function. So I feel it should have a > check for valid pointer. It seems plausible to me to check whether controllers_group is NULL, but it'd be nicer to do it *first* in the function so there's no cleanup to do, e.g., if (!controllers_group) return -EPROBE_DEFER; epc_group = kzalloc(sizeof(*epc_group), GFP_KERNEL); ... > >>> if (ret) { > >>> pr_err("failed to register configfs group for %s\n", name); > >>> goto err_register_group; > >>> diff --git a/drivers/pci/endpoint/pci-epc-core.c b/drivers/pci/endpoint/pci-epc-core.c > >>> index 42c2a11..d327a2a 100644 > >>> --- a/drivers/pci/endpoint/pci-epc-core.c > >>> +++ b/drivers/pci/endpoint/pci-epc-core.c > >>> @@ -518,6 +518,10 @@ __pci_epc_create(struct device *dev, const struct pci_epc_ops *ops, > >>> goto put_dev; > >>> > >>> epc->group = pci_ep_cfs_add_epc_group(dev_name(dev)); > >>> + if (IS_ERR(epc->group)) { > >>> + ret = -EPROBE_DEFER; > > > > should use the return value of pci_ep_cfs_add_epc_group(). > > > > OK. Will modify in next version. > > > However I don't think this is required since drivers might not actually need cfs. > > Ok, we can avoid propagating error to the caller here, but in case if > ERR_PTR there should be > at least one warn message. What do you say? > > > Thanks, > Pankaj Dubey > > > > Thanks > > Kishon