> On Sep 4, 2019, at 8:15 PM, Alexei Starovoitov <alexei.starovoi...@gmail.com>
> wrote:
>
> On Thu, Sep 05, 2019 at 02:51:51AM +0000, Song Liu wrote:
>>
>>
>>> On Sep 4, 2019, at 6:32 PM, Alexei Starovoitov
>>> <alexei.starovoi...@gmail.com> wrote:
>>>
>>> On Thu, Sep 05, 2019 at 12:34:36AM +0000, Song Liu wrote:
>>>>
>>>>
>>>>> On Sep 4, 2019, at 11:43 AM, Alexei Starovoitov <a...@kernel.org> wrote:
>>>>>
>>>>> Implement permissions as stated in uapi/linux/capability.h
>>>>>
>>>>> Signed-off-by: Alexei Starovoitov <a...@kernel.org>
>>>>>
>>>>
>>>> [...]
>>>>
>>>>> @@ -1648,11 +1648,11 @@ static int bpf_prog_load(union bpf_attr *attr,
>>>>> union bpf_attr __user *uattr)
>>>>> is_gpl = license_is_gpl_compatible(license);
>>>>>
>>>>> if (attr->insn_cnt == 0 ||
>>>>> - attr->insn_cnt > (capable(CAP_SYS_ADMIN) ?
>>>>> BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
>>>>> + attr->insn_cnt > (capable_bpf() ? BPF_COMPLEXITY_LIMIT_INSNS :
>>>>> BPF_MAXINSNS))
>>>>> return -E2BIG;
>>>>> if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
>>>>> type != BPF_PROG_TYPE_CGROUP_SKB &&
>>>>> - !capable(CAP_SYS_ADMIN))
>>>>> + !capable_bpf())
>>>>> return -EPERM;
>>>>
>>>> Do we allow load BPF_PROG_TYPE_SOCKET_FILTER and BPF_PROG_TYPE_CGROUP_SKB
>>>> without CAP_BPF? If so, maybe highlight in the header?
>>>
>>> of course. there is no change in behavior.
>>> 'highlight in the header'?
>>> you mean in commit log?
>>> I think it's a bit weird to describe things in commit that patch
>>> is _not_ changing vs things that patch does actually change.
>>> This type of comment would be great in a doc though.
>>> The doc will be coming separately in the follow up assuming
>>> the whole thing lands. I'll remember to note that bit.
>>
>> I meant capability.h:
>>
>> + * CAP_BPF allows the following BPF operations:
>> + * - Loading all types of BPF programs
>>
>> But CAP_BPF is not required to load all types of programs.
>
> yes, but above statement is still correct, right?
>
> And right below it says:
> * CAP_BPF allows the following BPF operations:
> * - Loading all types of BPF programs
> * - Creating all types of BPF maps except:
> * - stackmap that needs CAP_TRACING
> * - devmap that needs CAP_NET_ADMIN
> * - cpumap that needs CAP_SYS_ADMIN
> which is also correct, but CAP_BPF is not required
> for array, hash, prog_array, percpu, map-in-map ...
> except their lru variants...
> and except if they contain bpf_spin_lock...
> and if they need BTF it currently can be loaded with cap_sys_admin only...
>
> If we say something about socket_filter, cg_skb progs in capability.h
> we should clarify maps as well, but then it will become too big for .h
> The comments in capability.h already look too long to me.
> All that info and a lot more belongs in the doc.
Agreed. We cannot put all these details in capability.h. Doc/wikipages
would be better fit for these information.
Thanks,
Song