[PATCH bpf-next 0/4] Make inode storage available to tracing prog

[Casey Schaufler]

On 11/12/2024 5:37 PM, Song Liu wrote:
>
>> On Nov 12, 2024, at 5:10 PM, Casey Schaufler <casey at schaufler-ca.com> wrote:
>>
>> On 11/12/2024 10:44 AM, Song Liu wrote:
>>> Hi Casey, 
>>>
>>> Thanks for your input. 
>>>
>>>> On Nov 12, 2024, at 10:09 AM, Casey Schaufler <casey at schaufler-ca.com> wrote:
>>>>
>>>> On 11/12/2024 12:25 AM, Song Liu wrote:
>>>>> bpf inode local storage can be useful beyond LSM programs. For example,
>>>>> bcc/libbpf-tools file* can use inode local storage to simplify the logic.
>>>>> This set makes inode local storage available to tracing program.
>>>> Mixing the storage and scope of LSM data and tracing data leaves all sorts
>>>> of opportunities for abuse. Add inode data for tracing if you can get the
>>>> patch accepted, but do not move the LSM data out of i_security. Moving
>>>> the LSM data would break the integrity (such that there is) of the LSM
>>>> model.
>>> I honestly don't see how this would cause any issues. Each bpf inode 
>>> storage maps are independent of each other, and the bpf local storage is 
>>> designed to handle multiple inode storage maps properly. Therefore, if
>>> the user decide to stick with only LSM hooks, there isn't any behavior 
>>> change. OTOH, if the user decides some tracing hooks (on tracepoints, 
>>> etc.) are needed, making a inode storage map available for both tracing 
>>> programs and LSM programs would help simplify the logic. (Alternatively,
>>> the tracing programs need to store per inode data in a hash map, and 
>>> the LSM program would read that instead of the inode storage map.)
>>>
>>> Does this answer the question and address the concerns?
>> First off, I had no question. No, this does not address my concern.
>> LSM data should be kept in and managed by the LSMs. We're making an
>> effort to make the LSM infrastructure more consistent.
> Could you provide more information on the definition of "more 
> consistent" LSM infrastructure?

We're doing several things. The management of security blobs
(e.g. inode->i_security) has been moved out of the individual
modules and into the infrastructure. The use of a u32 secid is
being replaced with a more general lsm_prop structure, except
where networking code won't allow it. A good deal of work has
gone into making the return values of LSM hooks consistent.

Some of this was done as part of the direct call change, and some
in support of LSM stacking. There are also some hardening changes
that aren't ready for prime-time, but that are in the works.
There have been concerns about the potential expoitability of the
LSM infrastructure, and we're serious about addressing those.

>
> BPF LSM programs have full access to regular BPF maps (hash map, 
> array, etc.). There was never a separation of LSM data vs. other 
> data. 
>
> AFAICT, other LSMs also use kzalloc and similar APIs for memory 
> allocation. So data separation is not a goal for any LSM, right?
>
> Thanks,
> Song
>
>> Moving some of
>> the LSM data into an LSM specific field in the inode structure goes
>> against this. What you're proposing is a one-off clever optimization
>> hack. We have too many of those already.
>
>