Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <71774d67-6c7f-ea42-2911-a3eb1955777d@arm.com>
Date:   Mon, 1 Aug 2022 12:06:56 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; rv:91.0) Gecko/20100101
 Thunderbird/91.11.0
Subject: Re: [POSSIBLE BUG] iommu/io-pgtable-arm: possible dereferencing of
 NULL pointer
Content-Language: en-GB
To:     Will Deacon <will@kernel.org>,
        Subkhankulov Rustam <subkhankulov@ispras.ru>
Cc:     Joerg Roedel <joro@8bytes.org>,
        linux-arm-kernel@lists.infradead.org, iommu@lists.linux.dev,
        linux-kernel@vger.kernel.org,
        Alexey Khoroshilov <khoroshilov@ispras.ru>,
        ldv-project@linuxtesting.org
References: <28df50012344fb1c925a7ceaf55ae400152ffb48.camel@ispras.ru>
 <20220719173610.GA14526@willie-the-truck>
From:   Robin Murphy <robin.murphy@arm.com>
In-Reply-To: <20220719173610.GA14526@willie-the-truck>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2022-07-19 18:36, Will Deacon wrote:
> On Mon, Jul 18, 2022 at 12:20:06PM +0300, Subkhankulov Rustam wrote:
>> Version: 5-19-rc6
>>
>> In function '__arm_lpae_alloc_pages' pointer 'dev' is compared with
>> NULL at [drivers/iommu/io-pgtable-arm.c: 203]. This means that the
>> pointer can be NULL.
>>
>> -----------------------------------------------------------------------
>> 203 	p = alloc_pages_node(dev ? dev_to_node(dev) : NUMA_NO_NODE,
>> 204 			     gfp | __GFP_ZERO, order);
>> -----------------------------------------------------------------------
>>
>> Then, if cfg->coherent_walk == 0 at [drivers/iommu/io-pgtable-arm.c:
>> 209], function 'dma_map_single', which is defined as
>> 'dma_map_single_attrs', is called and pointer dev is passed as
>> first parameter.
>>
>> -----------------------------------------------------------------------
>> 209 	if (!cfg->coherent_walk) {
>> 208 		dma = dma_map_single(dev, pages, size, DMA_TO_DEVICE);
>> -----------------------------------------------------------------------
>>
>> Therefore, pointer 'dev' passed to function 'dev_driver_string'
>> in macro 'dev_WARN_ONCE' at [include/linux/dma-mapping.h: 326],
>> where it is dereferenced at [drivers/base/core.c: 2091].
>>
>> -----------------------------------------------------------------------
>> 2083	const char *dev_driver_string(const struct device *dev)
>> 2084	{
>> 2085		struct device_driver *drv;
>> 2086
>> ---
>> 2091		drv = READ_ONCE(dev->driver);
>> -----------------------------------------------------------------------
>>
>> Thus, if it is possible that 'dev' is null at the same time
>> that flag 'coherent_walk' is 0, then NULL pointer will be
>> dereferenced.
>>
>> Should we somehow avoid NULL pointer dereference or is this
>> situation impossible and we should remove comparison with NULL?
> 
> I think 'dev' is only null in the case of the selftest initcall
> (see arm_lpae_do_selftests()), and 'coherent_walk' is always true there.

Indeed, the intent is that cfg->iommu_dev == NULL is a special case for 
the selftest, which must always claim coherency as well for this reason. 
I suppose we could add an explicit assertion along those lines in 
alloc_pgtable if anyone really thinks it matters.

Cheers,
Robin.