On 4/1/22 11:40, Larry Finger wrote:
> On 4/1/22 10:12, Guenter Roeck wrote:
>> In __nat25_add_pppoe_tag(), the tag length is read from the tag data
>> structure. The value is kept in network format, but read as raw value.
>> With -Warray-bounds, this results in the following gcc error/warning
>> when building the driver on a big endian system such as alpha.
>>
>> In function '__nat25_add_pppoe_tag',
>> inlined from 'nat25_db_handle' at
>> drivers/staging/r8188eu/core/rtw_br_ext.c:479:11:
>> arch/alpha/include/asm/string.h:22:16: error:
>> '__builtin_memcpy' forming offset [40, 2051] is out of the bounds
>> [0, 40] of object 'tag_buf' with type 'unsigned char[40]'
>>
>> Add the missing ntohs().
>>
>> Fixes: 15865124feed ("staging: r8188eu: introduce new core dir for RTL8188eu driver")
>> Cc: Phillip Potter <[email protected]>
>> Signed-off-by: Guenter Roeck <[email protected]>
>> ---
>> drivers/staging/r8188eu/core/rtw_br_ext.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/staging/r8188eu/core/rtw_br_ext.c b/drivers/staging/r8188eu/core/rtw_br_ext.c
>> index d68611ef22f8..31bcd495ec04 100644
>> --- a/drivers/staging/r8188eu/core/rtw_br_ext.c
>> +++ b/drivers/staging/r8188eu/core/rtw_br_ext.c
>> @@ -70,7 +70,7 @@ static int __nat25_add_pppoe_tag(struct sk_buff *skb, struct pppoe_tag *tag)
>> struct pppoe_hdr *ph = (struct pppoe_hdr *)(skb->data + ETH_HLEN);
>> int data_len;
>> - data_len = tag->tag_len + TAG_HDR_LEN;
>> + data_len = htons(tag->tag_len) + TAG_HDR_LEN;
Wonder where my brain was. That should have been ntohs().
>> if (skb_tailroom(skb) < data_len)
>> return -1;
>
> Strange that a BE compiler would generate a warning for what is actually an execution error on LE platforms.
>
> I prefer be16_to_cpu() to htons() as the former makes it clearer what is happening, but I suppose that is a matter of choice.
>
The rest of the code uses htons/ntohs, so I prefer to follow that lead.
> Reviewed_by: Larry Finger <[email protected]>
>
> Incidentally, Sparse shows 3 more __be16 problems in this routine. I leave their fixing to you.
>
I'll need to resend anyway, so I'll track those down and fix them as well.
Thanks,
Guenter
On 4/1/22 14:25, Guenter Roeck wrote:
> The rest of the code uses htons/ntohs, so I prefer to follow that lead.
You just proved my point. It is hard to get be16_to_cpu() wrong. Sparse will
flag the error when you use cpu_to_be16() instead. I expect that your
htons/ntohs problem would also have shown up with Sparse.
Larry