Am 26.05.21 um 16:38 schrieb Paul Moore:
> On Wed, May 26, 2021 at 6:19 AM Pavel Begunkov <asml.silence@xxxxxxxxx> wrote:
>> On 5/26/21 3:04 AM, Paul Moore wrote:
>>> On Tue, May 25, 2021 at 9:11 PM Jens Axboe <axboe@xxxxxxxxx> wrote:
>>>> On 5/24/21 1:59 PM, Paul Moore wrote:
>>>>> That said, audit is not for everyone, and we have build time and
>>>>> runtime options to help make life easier. Beyond simply disabling
>>>>> audit at compile time a number of Linux distributions effectively
>>>>> shortcut audit at runtime by adding a "never" rule to the audit
>>>>> filter, for example:
>>>>>
>>>>> % auditctl -a task,never
>>>>
>>>> As has been brought up, the issue we're facing is that distros have
>>>> CONFIG_AUDIT=y and hence the above is the best real world case outside
>>>> of people doing custom kernels. My question would then be how much
>>>> overhead the above will add, considering it's an entry/exit call per op.
>>>> If auditctl is turned off, what is the expectation in turns of overhead?
>>>
>>> I commented on that case in my last email to Pavel, but I'll try to go
>>> over it again in a little more detail.
>>>
>>> As we discussed earlier in this thread, we can skip the req->opcode
>>> check before both the _entry and _exit calls, so we are left with just
>>> the bare audit calls in the io_uring code. As the _entry and _exit
>>> functions are small, I've copied them and their supporting functions
>>> below and I'll try to explain what would happen in CONFIG_AUDIT=y,
>>> "task,never" case.
>>>
>>> + static inline struct audit_context *audit_context(void)
>>> + {
>>> + return current->audit_context;
>>> + }
>>>
>>> + static inline bool audit_dummy_context(void)
>>> + {
>>> + void *p = audit_context();
>>> + return !p || *(int *)p;
>>> + }
>>>
>>> + static inline void audit_uring_entry(u8 op)
>>> + {
>>> + if (unlikely(audit_enabled && audit_context()))
>>> + __audit_uring_entry(op);
>>> + }
>>
>> I'd rather agree that it's my cycle-picking. The case I care about
>> is CONFIG_AUDIT=y (because everybody enable it), and io_uring
>> tracing _not_ enabled at runtime. If enabled let them suffer
>> the overhead, it will probably dip down the performance
>>
>> So, for the case I care about it's two of
>>
>> if (unlikely(audit_enabled && current->audit_context))
>>
>> in the hot path. load-test-jump + current, so it will
>> be around 7x2 instructions. We can throw away audit_enabled
>> as you say systemd already enables it, that will give
>> 4x2 instructions including 2 conditional jumps.
>
> We've basically got it down to the equivalent of two
> "current->audit_context != NULL" checks in the case where audit is
> built into the kernel but disabled at runtime, e.g. CONFIG_AUDIT=y and
> "task,never". I'm at a loss for how we can lower the overhead any
> further, but I'm open to suggestions.
>
>> That's not great at all. And that's why I brought up
>> the question about need of pre and post hooks and whether
>> can be combined. Would be just 4 instructions and that is
>> ok (ish).
>
> As discussed previously in this thread that isn't really an option
> from an audit perspective.
>
>>> We would need to check with the current security requirements (there
>>> are distro people on the linux-audit list that keep track of that
>>> stuff), but looking at the opcodes right now my gut feeling is that
>>> most of the opcodes would be considered "security relevant" so
>>> selective auditing might not be that useful in practice. It would
>>> definitely clutter the code and increase the chances that new opcodes
>>> would not be properly audited when they are merged.
>>
>> I'm curious, why it's enabled by many distros by default? Are there
>> use cases they use?
>
> We've already talked about certain users and environments where audit
> is an important requirement, e.g. public sector, health care,
> financial institutions, etc.; without audit Linux wouldn't be an
> option for these users, at least not without heavy modification,
> out-of-tree/ISV patches, etc. I currently don't have any direct ties
> to any distros, "Enterprise" or otherwise, but in the past it has been
> my experience that distros much prefer to have a single kernel build
> to address the needs of all their users. In the few cases I have seen
> where a second kernel build is supported it is usually for hardware
> enablement. I'm sure there are other cases too, I just haven't seen
> them personally; the big distros definitely seem to have a strong
> desire to limit the number of supported kernel configs/builds.
>
>> Tempting to add AUDIT_IOURING=default N, but won't work I guess
>
> One of the nice things about audit is that it can give you a history
> of what a user did on a system, which is very important for a number
> of use cases. If we selectively disable audit for certain subsystems
> we create a blind spot in the audit log, and in the case of io_uring
> this can be a very serious blind spot. I fear that if we can't come
> to some agreement here we will need to make io_uring and audit
> mutually exclusive at build time which would be awful; forcing many
> distros to either make a hard choice or carry out-of-tree patches.
I'm wondering why it's not enough to have the native auditing just to happen.
E.g. all (I have checked RECVMSG,SENDMSG,SEND and CONNECT) socket related io_uring opcodes
already go via security_socket_{recvmsg,sendmsg,connect}()
IORING_OP_OPENAT* goes via do_filp_open() which is in common with the open[at[2]]() syscalls
and should also trigger audit_inode() and security_file_open().
So why is there anything special needed for io_uring (now that the native worker threads are used)?
Is there really any io_uring opcode that bypasses the security checks the corresponding native syscall
would do? If so, I think that should just be fixed...
Additional LSM based restrictions could be hooked into the io_check_restriction() path
and setup at io_uring_setup() or early io_uring_register() time.
What do you think?
metze
