On 12/9/2020 3:30 PM, Pierre-Louis Bossart wrote:
Thanks Dan for the bug report.
On 12/9/20 12:53 AM, Dan Carpenter wrote:
[ This bug is way older than your patch but you might know the answer. -dan ]
Hello Pierre-Louis Bossart,
The patch 5aebe7c7f9c2: "ASoC: topology: fix endianness issues" from Apr 4, 2019, leads to the following static checker warning:
sound/soc/soc-topology.c:903 soc_tplg_denum_create_values() warn: potential pointer math issue ('se->dobj.control.dvalues' is a 64 bit pointer)
sound/soc/soc-topology.c 887 static int soc_tplg_denum_create_values(struct soc_tplg *tplg, struct soc_enum *se, 888 struct snd_soc_tplg_enum_control *ec) 889 { 890 int i; 891 892 if (le32_to_cpu(ec->items) > sizeof(*ec->values))
The warning is from this line where Smatch starts to think that "ec->items" is the number of bytes, but later in the function we treat it was the number of elements.
I do think probably this should be if:
if (le32_to_cpu(ec->items) > ARRAY_SIZE(ec->values)) return -EINVAL;
The ec->values[] is an array of u32:
__le32 values[SND_SOC_TPLG_NUM_TEXTS * SNDRV_CTL_ELEM_ID_NAME_MAXLEN / 4];
so this code will return -EINVAL for anything larger than 4. Changing it to ARRAY_SIZE() would raise the limit to 176.
I agree with your analysis, even in the initial code the pattern was
if (ec->items > sizeof(*ec->values))
and that's indeed a clear confusion between number of elements and total number of bytes.
Ranjani and Amadeusz are more familiar than me with the topology code, let's see if they concur?
Yes, seems about right, we can also replace devm_kzalloc below with devm_kcalloc to make it even more clear that we are dealing with array.
On related note looking at the UAPI header... I noticed:
struct snd_soc_tplg_enum_control { (...) char texts[SND_SOC_TPLG_NUM_TEXTS][SNDRV_CTL_ELEM_ID_NAME_MAXLEN]; __le32 values[SND_SOC_TPLG_NUM_TEXTS * SNDRV_CTL_ELEM_ID_NAME_MAXLEN / 4];
which means that there is 16 names and then there is an array of values, which seems somehow oversized compared to names... not sure why we multiply it by maximum length of name and then divide...
I guess we could change it to something like: __le32 values[SND_SOC_TPLG_NUM_TEXTS]; __le32 reserved[160];
I'm not sure if we want to change it, but I guess it is something to keep in the back of the head. Quite frankly looking at it bit more I'm starting to think that those values from topology are not used anywhere after being assigned in the function from which this investigation started, simple grep for "dvalues" reports nothing more than soc-topology.c file, but I might have missed something.
Another thing which I started wondering about, why we even keep those values in "se->dobj.control", can't we just use "se" directly, as it seems to have required fields.
I guess I will look at it again tomorrow when I have a bit more fresh head, but I guess for now ARRAY_SIZE and devm_kcalloc change seems good to me, as the rest will have less to do with "array" type and more with general logic of parsing things.
893 return -EINVAL; 894 895 se->dobj.control.dvalues = devm_kzalloc(tplg->dev, le32_to_cpu(ec->items) * 896 sizeof(u32), 897 GFP_KERNEL); 898 if (!se->dobj.control.dvalues) 899 return -ENOMEM; 900 901 /* convert from little-endian */ 902 for (i = 0; i < le32_to_cpu(ec->items); i++) { 903 se->dobj.control.dvalues[i] = le32_to_cpu(ec->values[i]); 904 } 905 906 return 0; 907 }
regards, dan carpenter