Re: [PATCH v2.1] audit: Only use the syscall slowpath when syscall audit rules exist

On 02/03, Andy Lutomirski wrote: Confused... read_lock(tasklist) doesn't need to disable irqs. (ftrace does this for no reason too, perhaps I should resend the patch) probably this can be done outside of read_lock? for_each_process_thread ;) do_each_thread will die, I hope. > +void audit_dec_n_rules() > +{ > + struct task_struct *p, *g; > + unsigned long flags; > + > + read_lock_irqsave(&tasklist_lock, flags); > + > + --audit_n_rules; > + BUG_ON(audit_n_rules < 0); > + > + if (audit_n_rules == 0) { > + clear_tsk_thread_flag(p, TIF_SYSCALL_AUDIT); > + } while_each_thread(g, p); > + } The same, and... On a second thought it seems that audit_dec_n_rules() has a problem. Note the BUG_ON(context->in_syscall) in __audit_syscall_entry(). Suppose that audit_dec_n_rules() clears TIF_SYSCALL_AUDIT when a task runs a syscall. In this case (afaics) __audit_syscall_exit() won't be called. The next audit_inc_n_rules() can set TIF_SYSCALL_AUDIT and trigger another __audit_syscall_entry() which will hit this BUG_ON(). And in general it doesn't look safe although I know almost nothing about audit. I mean, currently __audit_syscall_entry() or __audit_log_bprm_fcaps() assume that __audit_syscall_exit() or __audit_free() will "cleanup" ->audit_context, perhaps we should not break the rules? Once again, I do not pretend I understand this code, this is the question, not the comment. But if I am right, then TIF_SYSCALL_AUDIT should be cleared in __audit_syscall_exit() as you suggested before. Oleg.