From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from aserp1040.oracle.com ([141.146.126.69]:46151 "EHLO aserp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750967AbbJGOHu (ORCPT ); Wed, 7 Oct 2015 10:07:50 -0400 Message-ID: <56152725.10809@oracle.com> Date: Wed, 07 Oct 2015 10:07:33 -0400 From: Sasha Levin MIME-Version: 1.0 To: Prarit Bhargava , Bjorn Helgaas CC: bhelgaas@google.com, linux-pci@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH] PCI: prevent out of bounds access in numa_node override References: <1443995369-11399-1-git-send-email-sasha.levin@oracle.com> <20151006193627.GE29420@localhost> <561428CE.2040601@redhat.com> In-Reply-To: <561428CE.2040601@redhat.com> Content-Type: text/plain; charset=windows-1252 Sender: linux-pci-owner@vger.kernel.org List-ID: On 10/06/2015 04:02 PM, Prarit Bhargava wrote: > > > On 10/06/2015 03:36 PM, Bjorn Helgaas wrote: >> Hi Sasha, >> >> On Sun, Oct 04, 2015 at 05:49:29PM -0400, Sasha Levin wrote: >>> Commit 63692df1 ("PCI: Allow numa_node override via sysfs") didn't check that >>> the numa node provided by userspace is valid. Passing a node number too high >>> would attempt to access invalid memory and trigger a kernel panic. >>> >>> Fixes: 63692df1 ("PCI: Allow numa_node override via sysfs") >>> Signed-off-by: Sasha Levin >>> --- >>> drivers/pci/pci-sysfs.c | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c >>> index 312f23a..e9abca8 100644 >>> --- a/drivers/pci/pci-sysfs.c >>> +++ b/drivers/pci/pci-sysfs.c >>> @@ -216,7 +216,7 @@ static ssize_t numa_node_store(struct device *dev, >>> if (ret) >>> return ret; >>> >>> - if (!node_online(node)) >>> + if (node > MAX_NUMNODES || !node_online(node)) >> >> This needs to be "node >= MAX_NUMNODES", doesn't it? I'll fix it up if >> you agree. Yup, you're right. > > Not a strenuous objection, but I don't see much bound checking using > MAX_NUMNODES in the kernel outside of the core numa area. Is fixing > node_online() with bounds checking a better option here so that other callers > get the fix? I would have thought that calling node_online() with node > > MAX_NUMNODES should be safe to call. I don't know, this will add overhead to node_online(), and isn't really done in any other similar function. For example, cpu_online() isn't safe to call with cpu > NR_CPUS either. Thanks, Sasha