Re: [PATCH][next] hwmon: (aspeed-pwm-tacho) Fix -Wstringop-overflow warning in aspeed_create_fan_tach_channel()

[Gustavo A. R. Silva]

On 11/14/23 08:52, Guenter Roeck wrote:
> On Mon, Nov 13, 2023 at 01:38:12PM -0600, Gustavo A. R. Silva wrote:
> > Based on the documentation below, the maximum number of Fan tach
> > channels is 16:
> >
> > Documentation/devicetree/bindings/hwmon/aspeed-pwm-tacho.txt:45:
> >   45 - aspeed,fan-tach-ch : should specify the Fan tach input channel.
> >   46                 integer value in the range 0 through 15, with 0 indicating
> >   47                 Fan tach channel 0 and 15 indicating Fan tach channel 15.
> >   48                 At least one Fan tach input channel is required.
> >
> > However, the compiler doesn't know that, and legitimaly warns about a potential
> > overwrite in array `u8 fan_tach_ch_source[16]` in `struct aspeed_pwm_tacho_data`,
> > in case `index` takes a value outside the boundaries of the array:
>
> All that doesn't warrant introducing checkpatch warnings.

Do you mean this?

WARNING: Prefer a maximum 75 chars per line (possible unwrapped commit description?)
#17:
 46                 integer value in the range 0 through 15, with 0 indicating

I honestly didn't consider that relevant, and I didn't want to alter the format of
the Doc text.

However, if you want me to split any offending line, that's not a problem. Just let
me know. :)

> > drivers/hwmon/aspeed-pwm-tacho.c:
> > 179 struct aspeed_pwm_tacho_data {
> > ...
> > 184         bool fan_tach_present[16];
> > ...
> > 193         u8 fan_tach_ch_source[16];
> > ...
> > 196 };
> >
> > In function ‘aspeed_create_fan_tach_channel’,
> >      inlined from ‘aspeed_create_fan’ at drivers/hwmon/aspeed-pwm-tacho.c:877:2,
> >      inlined from ‘aspeed_pwm_tacho_probe’ at drivers/hwmon/aspeed-pwm-tacho.c:936:9:
> > drivers/hwmon/aspeed-pwm-tacho.c:751:49: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
> >    751 |                 priv->fan_tach_ch_source[index] = pwm_source;
> >        |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~
> > drivers/hwmon/aspeed-pwm-tacho.c: In function ‘aspeed_pwm_tacho_probe’:
> > drivers/hwmon/aspeed-pwm-tacho.c:193:12: note: at offset [48, 255] into destination object ‘fan_tach_ch_source’ of size 16
> >    193 |         u8 fan_tach_ch_source[16];
> >        |            ^~~~~~~~~~~~~~~~~~
> >
> > Fix this by sanity checking `index` before using it to index arrays of
> > size 16 elements in `struct aspeed_pwm_tacho_data`. Also, and just for
> > completeness, add a `pr_err()` message to display in the unlikely case
> > `0 > index >= 16`.
> >
> > This is probably the last remaining -Wstringop-overflow issue in the
> > kernel, and this patch helps with the ongoing efforts to enable such
> > compiler option globally.
>
> I am sorry, but this description almost completely misses the point.
> The real issue is that the values in aspeed,fan-tach-ch are not range
> checked, which can cause real problems if is elements are set to values
> larger than 15. This is a real problem and has nothing to do with string
> overflows.

Yeah, the above paragraph was extra, and I removed it in v2[1]. The rest
of the changelog text describes the issue in the code.

> This should use dev_err() (and, yes, that means dev needs to be passed
> as argument), and the function should return -EINVAL if this is
> encountered. Also, error handling should come first.
>
> 		if (index >= MAX_ASPEED_FAN_TACH_CHANNELS) {
> 			dev_err(dev, "Invalid Fan Tach input channel %u\n.", index);
> 			return -EINVAL;
> 		}

Done in v2.

Thanks a lot for the feedback.
--
Gustavo

[1] https://lore.kernel.org/linux-hardening/ZVPQJIP26dIzRAr6@work/