Re: [PATCH] ipvs: initialize 'ret' variable in do_ip_vs_set_ctl()

[liqiong <liqiong@xxxxxxxxxxxx>]

在 2022年12月02日 19:26, Julian Anastasov 写道:
>       Hello,
>
> On Fri, 2 Dec 2022, Dan Carpenter wrote:
>
>> On Fri, Dec 02, 2022 at 06:18:37PM +0800, liqiong wrote:
>>>
>>> 在 2022年12月02日 18:07, Dan Carpenter 写道:
>>>> On Fri, Dec 02, 2022 at 11:25:11AM +0800, Li Qiong wrote:
>>>>> The 'ret' should need to be initialized to 0, in case
>>>>> return a uninitialized value because no default process
>>>>> for "switch (cmd)".
>>>>>
>>>>> Signed-off-by: Li Qiong <liqiong@xxxxxxxxxxxx>
>>>> If this is a real bug, then it needs a fixes tag.  The fixes tag helps
>>>> us know whether to back port or not and it also helps in reviewing the
>>>> patch.  Also get_maintainer.pl will CC the person who introduced the
>>>> bug so they can review it.  They are normally the best person to review
>>>> their own code.
>>>>
>>>> Here it would be:
>>>> Fixes: c5a8a8498eed ("ipvs: Fix uninit-value in do_ip_vs_set_ctl()")
>>>>
>>>> Which is strange...  Also it suggest that the correct value is -EINVAL
>>>> and not 0.
>>>>
>>>> The thing about uninitialized variable bugs is that Smatch and Clang
>>>> both warn about them so they tend to get reported pretty quick.
>>>> Apparently neither Nathan nor I sent forwarded this static checker
>>>> warning.  :/
>>>>
>>>> regards,
>>>> dan carpenter
>>> It is not a real bug,   I  use tool (eg: smatch, sparse) to audit the
>>> code,  got this warning and check it, found may be a real problem.
>> Yeah.  If it is a false positive just ignore it, do not bother to
>> silence wrong static checker warnings.
>>
>> The code in question here is:
>>
>>      if (len != set_arglen[CMDID(cmd)]) {
>>
>> The only time that condition can be true is for the cases in the switch
>> statement.  So Peilin's patch is correct.
>>
>> Smatch is bad at understanding arrays so Smatch cannot parse the if
>> statement above as a human reader can.
>       Yes, no bug in current code. But it is better to return the 
> default switch case with -EINVAL (not 0), in case new commands are added.
> Such patch should target net-next, it is just for compilers/tools
> that do not look into set_arglen[].
>
> Regards
>
> --
> Julian Anastasov <ja@xxxxxx>
Thanks, I will send a v2 patch.