On Wednesday, May 10, 2023 9:30:19 AM EDT Tetsuo Handa wrote:
> On 2023/05/10 21:12, Rinat Gadelshin wrote:
> >> Please try to find who is calling audit_send_reply_thread for many
> >> times.
> >
> > I've rebuilt the kernel with 'dump stack()'.
>
> Oops, I thought dump_stack() shows pid and comm name, but
> it is dump_stack_print_info() that shows pid and comm name.
>
> > As far as I can see, it's the exit of `sendto` syscall.
> > It seems that the kernel just creates a new kthreadd for each sendto
> > syscall. But I think that I'm wrong and just missing something.
>
> Yes, sendto() on netlink socket calls netlink_sendmsg().
> For some reason, audit_send_reply() is called for many times.
> audit_send_reply() is called by audit_receive_msg() for the following
> types.
>
> AUDIT_GET
> AUDIT_SIGNAL_INFO
> AUDIT_TTY_GET
> AUDIT_GET_FEATURE
The audit userspace always adds NLM_F_ACK to any netlink communication to
make sure it did not get discarded before it arrived. It has done this since
before I started working on audit code.
-Steve
> Would you re-caputure with
>
> - dump_stack();
> + pr_info("%s %s:%d type=%d\n", __func__, current->comm, current->pid,
> type);
>
> ?
>
> Regardless of the result of re-caputure, it seems there is no switch that
> can prevent audit_send_reply() from calling
> kthread_run(audit_send_reply_thread).
>
> But since kthreadd runs with PID=2 and PPID=0, you might be able to use
> PID=2 and/or PPID=0 in your rules in order to let kernel audit subsystem
> ignore kthreadd. (I can't test because I haven't found how to reproduce
> audit_receive_msg() in my environment...)
>
> # cat /proc/2/status
> Name: kthreadd
> Umask: 0000
> State: S (sleeping)
> Tgid: 2
> Ngid: 0
> Pid: 2
> PPid: 0
>
> --
> Linux-audit mailing list
> Linux-audit@redhat.com
> https://listman.redhat.com/mailman/listinfo/linux-audit
--
Linux-audit mailing list
Linux-audit@redhat.com
https://listman.redhat.com/mailman/listinfo/linux-audit
