[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <781ea2fd-637f-b896-aad4-d70f43ad245c@quicinc.com>
Date: Thu, 7 Nov 2024 19:05:15 +0530
From: Vikash Garodia <quic_vgarodia@...cinc.com>
To: Dmitry Baryshkov <dmitry.baryshkov@...aro.org>
CC: Bryan O'Donoghue <bryan.odonoghue@...aro.org>,
Stanimir Varbanov
<stanimir.k.varbanov@...il.com>,
Mauro Carvalho Chehab <mchehab@...nel.org>,
<linux-media@...r.kernel.org>, <linux-arm-msm@...r.kernel.org>,
<linux-kernel@...r.kernel.org>, <stable@...r.kernel.org>
Subject: Re: [PATCH 1/4] media: venus: hfi_parser: add check to avoid out of
bound access
On 11/7/2024 6:52 PM, Dmitry Baryshkov wrote:
> On Thu, Nov 07, 2024 at 06:32:33PM +0530, Vikash Garodia wrote:
>>
>> On 11/7/2024 5:37 PM, Bryan O'Donoghue wrote:
>>> On 07/11/2024 10:41, Dmitry Baryshkov wrote:
>>>>> init_codecs() parses the payload received from firmware and . I don't think we
>>>>> can control this part when we have something like this from a malicious firmware
>>>>> payload
>>>>> HFI_PROPERTY_PARAM_CODEC_SUPPORTED
>>>>> HFI_PROPERTY_PARAM_CODEC_SUPPORTED
>>>>> HFI_PROPERTY_PARAM_CODEC_SUPPORTED
>>>>> ...
>>>>> Limiting it to second iteration would restrict the functionality when property
>>>>> HFI_PROPERTY_PARAM_CODEC_SUPPORTED is sent for supported number of codecs.
>>>> If you can have a malicious firmware (which is owned and signed by
>>>> Qualcomm / OEM), then you have to be careful and skip duplicates. So
>>>> instead of just adding new cap to core->caps, you have to go through
>>>> that array, check that you are not adding a duplicate (and report a
>>>> [Firmware Bug] for duplicates), check that there is an empty slot, etc.
>>>>
>>>> Just ignoring the "extra" entries is not enough.
>> Thinking of something like this
>>
>> for_each_set_bit(bit, &core->dec_codecs, MAX_CODEC_NUM) {
>> if (core->codecs_count >= MAX_CODEC_NUM)
>> return;
>> cap = &caps[core->codecs_count++];
>> if (cap->codec == BIT(bit)) --> each code would have unique bitfield
>> return;
>
> This won't work and it's pretty obvious why.
Could you please elaborate what would break in above logic ?
>
>>> +1
>>>
>>> This is a more rational argument. If you get a second message, you should surely
>>> reinit the whole array i.e. update the array with the new list, as opposed to
>>> throwing away the second message because it over-indexes your local storage..
>> That would be incorrect to overwrite the array with new list, whenever new
>> payload is received.
>
> I'd say, don't overwrite the array. Instead the driver should extend it
> with the new information.
That is exactly the existing patch is currently doing.
Regards,
Vikash
>
>>
>> Regards,
>> Vikash
>
Powered by blists - more mailing lists