Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751586AbdHVRQK (ORCPT ); Tue, 22 Aug 2017 13:16:10 -0400 Received: from mx1.redhat.com ([209.132.183.28]:53606 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751061AbdHVRQI (ORCPT ); Tue, 22 Aug 2017 13:16:08 -0400 DMARC-Filter: OpenDMARC Filter v1.3.2 mx1.redhat.com 2573B61475 Authentication-Results: ext-mx10.extmail.prod.ext.phx2.redhat.com; dmarc=none (p=none dis=none) header.from=redhat.com Authentication-Results: ext-mx10.extmail.prod.ext.phx2.redhat.com; spf=fail smtp.mailfrom=alex.williamson@redhat.com Date: Tue, 22 Aug 2017 11:16:07 -0600 From: Alex Williamson To: Paolo Bonzini Cc: nixiaoming , kvm@vger.kernel.org, linux-api@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH] virt/lib avoids oops by adding parameter checking Message-ID: <20170822111607.0397a16d@w520.home> In-Reply-To: References: <20170822010753.102857-1-nixiaoming@huawei.com> MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.39]); Tue, 22 Aug 2017 17:16:08 +0000 (UTC) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 3319 Lines: 86 On Tue, 22 Aug 2017 15:39:04 +0200 Paolo Bonzini wrote: > On 22/08/2017 03:07, nixiaoming wrote: > > The error parameter passed through the external interface > > causes the system oops. So it is necessary to increase the > > parameter check for all EXPORT_SYMBOL_GPL > > > > example: > > int irq_bypass_register_producer(struct irq_bypass_producer *producer) > > { > > if (!producer->token) /* oops if producer == null */ > > return -einval; > > } > > EXPORT_SYMBOL_GPL(irq_bypass_register_producer); > > This is used like this: > > drivers/vfio/pci/vfio_pci_intrs.c: irq_bypass_unregister_producer(&vdev->ctx[vector].producer); > drivers/vfio/pci/vfio_pci_intrs.c: ret = irq_bypass_register_producer(&vdev->ctx[vector].producer); > virt/kvm/eventfd.c: irq_bypass_unregister_consumer(&irqfd->consumer); > virt/kvm/eventfd.c: ret = irq_bypass_register_consumer(&irqfd->consumer); > > So your check won't actually catch irqfd or vdev being NULL. Additionally, I don't know of a reference to how extensively we should harden these sorts of interfaces to misuse. Just because a symbol is exported doesn't necessarily mean that the function needs to be hardened against misuse as we might for an interface exposed to a user. It's called by kernel mode code which can already do anything it pleases to cause an oops. If such hardening makes the calling conventions or code flow easier, or results in a function that's easier to use correctly or harder to misuse, then I think it's justified. None of that has been proven here and certainly doesn't seem to be the case based on the current in-tree users. Can the patch submitter justify that this is "necessary" as claimed in the commit log? Thanks, Alex > > Signed-off-by: nixiaoming > > --- > > virt/lib/irqbypass.c | 8 ++++---- > > 1 file changed, 4 insertions(+), 4 deletions(-) > > > > diff --git a/virt/lib/irqbypass.c b/virt/lib/irqbypass.c > > index 6d2fcd6..2bb99e8 100644 > > --- a/virt/lib/irqbypass.c > > +++ b/virt/lib/irqbypass.c > > @@ -89,7 +89,7 @@ int irq_bypass_register_producer(struct irq_bypass_producer *producer) > > struct irq_bypass_producer *tmp; > > struct irq_bypass_consumer *consumer; > > > > - if (!producer->token) > > + if (!producer || !producer->token) > > return -EINVAL; > > > > might_sleep(); > > @@ -139,7 +139,7 @@ void irq_bypass_unregister_producer(struct irq_bypass_producer *producer) > > struct irq_bypass_producer *tmp; > > struct irq_bypass_consumer *consumer; > > > > - if (!producer->token) > > + if (!producer || !producer->token) > > return; > > > > might_sleep(); > > @@ -183,7 +183,7 @@ int irq_bypass_register_consumer(struct irq_bypass_consumer *consumer) > > struct irq_bypass_consumer *tmp; > > struct irq_bypass_producer *producer; > > > > - if (!consumer->token || > > + if (!consumer || !consumer->token || > > !consumer->add_producer || !consumer->del_producer) > > return -EINVAL; > > > > @@ -234,7 +234,7 @@ void irq_bypass_unregister_consumer(struct irq_bypass_consumer *consumer) > > struct irq_bypass_consumer *tmp; > > struct irq_bypass_producer *producer; > > > > - if (!consumer->token) > > + if (!consumer || !consumer->token) > > return; > > > > might_sleep(); > > >