Re: [PATCH v1] ptp: ocp: fix NULL deref in _signal_summary_show

[Vadim Fedorenko <vadim.fedorenko@linux.dev>]

On 16/04/2025 14:59, Sagi Maimon wrote:
> On Wed, Apr 16, 2025 at 1:35 PM Vadim Fedorenko
> <vadim.fedorenko@linux.dev> wrote:
>>
>> On 16/04/2025 07:33, Sagi Maimon wrote:
>>> On Mon, Apr 14, 2025 at 4:55 PM Vadim Fedorenko
>>> <vadim.fedorenko@linux.dev> wrote:
>>>>
>>>> On 14/04/2025 14:43, Sagi Maimon wrote:
>>>>> On Mon, Apr 14, 2025 at 4:01 PM Vadim Fedorenko
>>>>> <vadim.fedorenko@linux.dev> wrote:
>>>>>>
>>>>>> On 14/04/2025 12:38, Sagi Maimon wrote:
>>>>>>> On Mon, Apr 14, 2025 at 2:09 PM Vadim Fedorenko
>>>>>>> <vadim.fedorenko@linux.dev> wrote:
>>>>>>>>
>>>>>>>> On 14/04/2025 11:56, Sagi Maimon wrote:
>>>>>>>>> On Mon, Apr 14, 2025 at 12:37 PM Vadim Fedorenko
>>>>>>>>> <vadim.fedorenko@linux.dev> wrote:
>>>>>>>>>>
>>>>>>>>>> On 14/04/2025 09:54, Sagi Maimon wrote:
>>>>>>>>>>> Sysfs signal show operations can invoke _signal_summary_show before
>>>>>>>>>>> signal_out array elements are initialized, causing a NULL pointer
>>>>>>>>>>> dereference. Add NULL checks for signal_out elements to prevent kernel
>>>>>>>>>>> crashes.
>>>>>>>>>>>
>>>>>>>>>>> Fixes: b325af3cfab9 ("ptp: ocp: Add signal generators and update sysfs nodes")
>>>>>>>>>>> Signed-off-by: Sagi Maimon <maimon.sagi@gmail.com>
>>>>>>>>>>> ---
>>>>>>>>>>>        drivers/ptp/ptp_ocp.c | 3 +++
>>>>>>>>>>>        1 file changed, 3 insertions(+)
>>>>>>>>>>>
>>>>>>>>>>> diff --git a/drivers/ptp/ptp_ocp.c b/drivers/ptp/ptp_ocp.c
>>>>>>>>>>> index 7945c6be1f7c..4c7893539cec 100644
>>>>>>>>>>> --- a/drivers/ptp/ptp_ocp.c
>>>>>>>>>>> +++ b/drivers/ptp/ptp_ocp.c
>>>>>>>>>>> @@ -3963,6 +3963,9 @@ _signal_summary_show(struct seq_file *s, struct ptp_ocp *bp, int nr)
>>>>>>>>>>>            bool on;
>>>>>>>>>>>            u32 val;
>>>>>>>>>>>
>>>>>>>>>>> +     if (!bp->signal_out[nr])
>>>>>>>>>>> +             return;
>>>>>>>>>>> +
>>>>>>>>>>>            on = signal->running;
>>>>>>>>>>>            sprintf(label, "GEN%d", nr + 1);
>>>>>>>>>>>            seq_printf(s, "%7s: %s, period:%llu duty:%d%% phase:%llu pol:%d",
>>>>>>>>>>
>>>>>>>>>> That's not correct, the dereference of bp->signal_out[nr] happens before
>>>>>>>>>> the check. But I just wonder how can that even happen?
>>>>>>>>>>
>>>>>>>>> The scenario (our case): on ptp_ocp_adva_board_init we
>>>>>>>>> initiate only signals 0 and 1 so 2 and 3 are NULL.
>>>>>>>>> Later ptp_ocp_summary_show runs on all 4 signals and calls _signal_summary_show
>>>>>>>>> when calling signal 2 or 3  the dereference occurs.
>>>>>>>>> can you please explain: " the dereference of bp->signal_out[nr] happens before
>>>>>>>>> the check", where exactly? do you mean in those lines:
>>>>>>>>> struct signal_reg __iomem *reg = bp->signal_out[nr]->mem;
>>>>>>>>         ^^^
>>>>>>>> yes, this is the line which dereferences the pointer.
>>>>>>>>
>>>>>>>> but in case you have only 2 pins to configure, why the driver exposes 4
>>>>>>>> SMAs? You can simply adjust the attributes (adva_timecard_attrs).
>>>>>>>>
>>>>>>> I can (and will) expose only 2 sma in adva_timecard_attrs, but still
>>>>>>> ptp_ocp_summary_show runs
>>>>>>> on all 4 signals and not only on the on that exposed, is it not a bug?
>>>>>>
>>>>>> Yeah, it's a bug, but different one, and we have to fix it other way.
>>>>>>
>>>>> Do you want to instruct me how to fix it , or will you fix it?
>>>>
>>>> well, the original device structure was not designed to have the amount
>>>> of SMAs less than 4. We have to introduce another field to store actual
>>>> amount of SMAs to work with, and adjust the code to check the value. The
>>>> best solution would be to keep maximum amount of 4 SMAs in the structure
>>>> but create a helper which will init new field and will have
>>>> BUILD_BUG_ON() to prevent having more SMAs than fixed size array for
>>>> them. That will solve your problem, but I will need to check it on the
>>>> HW we run.
>>>>
>>> just to be clear you will write the fix and test it on your HW, so you
>>> don't want me to write the fix?
>>
>> Well, it would be great if you can write the code which will make SMA
>> functions flexible to the amount of pin the HW has. All our HW has fixed
>> amount of 4 pins that's why the driver was coded with constants. Now
>> your hardware has slightly different amount of pins, so it needs
>> adjustments to the driver to work properly. I just want to be sure that
>> any adjustments will not break my HW - that's what I meant saying I'll
>> test it.
>>
> Just to be clear (correct me please if I am wrong):
> I will write the code, then create a patch and upstream to the vanilla
> you will test my change on your HW and only then approve the patch

Yes, that's correct

>>>>>>>>> struct ptp_ocp_signal *signal = &bp->signal[nr];
>>>>>>>>>> I believe the proper fix is to move ptp_ocp_attr_group_add() closer to
>>>>>>>>>> the end of ptp_ocp_adva_board_init() like it's done for other boards.
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>>> pw-bot: cr
>>>>>>>>
>>>>>>
>>>>
>>