On 06/08, Oleg Nesterov wrote:
OK. Thanks a lot Eric for your explanations.
Yes. but may I ask another one?
Shouldn't copy_process()->audit_alloc(tsk) path do
clear_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT) if it doesn't
set tsk->audit_context?
I can be easily wrong, but afaics otherwise the child can run
with TIF_SYSCALL_AUDIT bit copied from parent's thread_info by
dup_task_struct()->setup_thread_stack() and without ->audit_context,
right? For what?
Any other reason why audit_syscall_entry() checks context != NULL?
IOW. Any reason the patch below is wrong?
I am just curious, thanks.
Oleg.
--- x/kernel/auditsc.c
+++ x/kernel/auditsc.c
@@ -885,6 +885,8 @@ int audit_alloc(struct task_struct *tsk)
if (likely(!audit_ever_enabled))
return 0; /* Return if not auditing. */
+ clear_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
+
state = audit_filter_task(tsk, &key);
if (likely(state == AUDIT_DISABLED))
return 0;
@@ -1591,9 +1593,7 @@ void audit_syscall_entry(int arch, int m
struct audit_context *context = tsk->audit_context;
enum audit_state state;
- if (unlikely(!context))
- return;
-
+ BUG_ON(!context);
/*
* This happens only on certain architectures that make system
* calls in kernel_thread via the entry.S interface, instead of