Message-ID: <ff4be1ef-ef02-471e-b8f0-8e9cdb312794@linux.dev>
Date: Wed, 26 Nov 2025 17:15:12 +0800
From: Tao Chen <chen.dylane@...ux.dev>
To: Andrii Nakryiko <andrii.nakryiko@...il.com>,
 Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc: Alexei Starovoitov <ast@...nel.org>,
 Daniel Borkmann <daniel@...earbox.net>, Andrii Nakryiko <andrii@...nel.org>,
 Martin KaFai Lau <martin.lau@...ux.dev>, Eduard <eddyz87@...il.com>,
 Song Liu <song@...nel.org>, Yonghong Song <yonghong.song@...ux.dev>,
 John Fastabend <john.fastabend@...il.com>, KP Singh <kpsingh@...nel.org>,
 Stanislav Fomichev <sdf@...ichev.me>, Hao Luo <haoluo@...gle.com>,
 Jiri Olsa <jolsa@...nel.org>, bpf <bpf@...r.kernel.org>,
 LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH bpf-next 1/2] bpf: Add bpf_get_task_cmdline kfunc

在 2025/11/26 07:32, Andrii Nakryiko 写道:
> On Fri, Nov 21, 2025 at 5:17 PM Alexei Starovoitov
> <alexei.starovoitov@...il.com> wrote:
>>
>> On Tue, Nov 18, 2025 at 4:58 AM Tao Chen <chen.dylane@...ux.dev> wrote:
>>>
>>> Add the bpf_get_task_cmdline kfunc. One use case is as follows: In
>>> production environments, there are often short-lived script tasks executed,
>>> and sometimes these tasks may cause stability issues. It is desirable to
>>> detect these script tasks via eBPF. The common approach is to check
>>> the process name, but it can be difficult to distinguish specific
>>> tasks in some cases. Take the shell as an example: some tasks are
>>> started via bash xxx.sh – their process name is bash, but the script
>>> name of the task can be obtained through the cmdline. Additionally,
>>> myabe this is helpful for security auditing purposes.
>>
>> maybe
>>
>>>
>>> Signed-off-by: Tao Chen <chen.dylane@...ux.dev>
>>> ---
>>>   kernel/bpf/helpers.c | 22 ++++++++++++++++++++++
>>>   1 file changed, 22 insertions(+)
>>>
>>> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
>>> index 865b0dae38d..7cac17d58d5 100644
>>> --- a/kernel/bpf/helpers.c
>>> +++ b/kernel/bpf/helpers.c
>>> @@ -2685,6 +2685,27 @@ __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)
>>>          return p;
>>>   }
>>>
>>> +/*
>>> + * bpf_get_task_cmdline - Get the cmdline to a buffer
>>> + *
>>> + * @task: The task whose cmdline to get.
>>> + * @buffer: The buffer to save cmdline info.
>>> + * @len: The length of the buffer.
>>> + *
>>> + * Return: the size of the cmdline field copied. Note that the copy does
>>> + * not guarantee an ending NULL byte. A negative error code on failure.
>>> + */
>>> +__bpf_kfunc int bpf_get_task_cmdline(struct task_struct *task, char *buffer, size_t len)
>>
>> 'size_t len' doesn't make the verifier track the size of the buffer.
>> while 'char *buffer' tells the verifier to check that _one_ byte is available.
>> So this is buggy.
>>
>> In general the kfunc seems useful, but selftest in patch 2 is just bad
>>
> 
> Besides that mm->arg_lock spinlock (which I don't think matters all
> that much for BPF programs), is there anything special in
> get_cmdline() that BPF program cannot just implemented? Ultimately,
> it's just copying mm->arg_start and mm->env_start zero-separated
> strings, no? We have bpf_copy_from_user_task_str() and also
> dynptr-based equivalent of it for even more variable-length
> flexibility. That should be all one needs, no?
> 

 From a quick look at how both are implemented, it seems that way.
Hold off on this patch for now. I will move forward if we find something 
new.

>> + ret = bpf_get_task_cmdline(task, buf, sizeof(buf));
>> + if (ret < 0)
>> +    err = 1;
>> +
>> + return 0;
>> +}
>>
>> it's not testing much.
>>
>> Also you must explain the true motivation for the kfunc.
>> "maybe helpful for security" is too vague.
>> Do you have a proprietary bpf-lsm that needs it?
>> What is the exact use case?
>>
>> pw-bot: cr


-- 
Best Regards
Tao Chen

Powered by blists - more mailing lists