Re: [PATCH] mm: fix lockdep build in rcu-protected get_mm_exe_file()

From: Oleg Nesterov
Date: Tue Mar 24 2015 - 14:12:28 EST


On 03/24, Konstantin Khlebnikov wrote:
>
> On 23.03.2015 22:10, Oleg Nesterov wrote:
>> On 03/23, Davidlohr Bueso wrote:
>>>
>>> void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
>>> {
>>> struct file *old_exe_file = rcu_dereference_protected(mm->exe_file,
>>> - !atomic_read(&mm->mm_users) || current->in_execve ||
>>> - lock_is_held(&mm->mmap_sem));
>>> + !atomic_read(&mm->mm_users) || current->in_execve);
>>
>> Thanks, looks correct at first glance...
>>
>> But can't we remove the ->in_execve check above? and check
>>
>> atomic_read(&mm->mm_users) <= 1
>>
>> instead. OK, this is subjective, I won't insist. Just current->in_execve
>> looks a bit confusing, it means "I swear, the caller is flush_old_exec()
>> and this mm is actualy bprm->mm".
>>
>> "atomic_read(&mm->mm_users) <= 1" looks a bit more "safe". But again,
>> I won't insist.
>
> Not so safe: this will race with get_task_mm().

How?

If set_mm_exe_file() can race with get_task_mm() then we have a bug.
And it will be reported ;)

> A lot of proc files grab temporary reference to task mm.
> But this just a debug -- we can place here "true".

Yeees, probably rcu_dereference_raw() would be even better. set_mm_exe_file()
must be called only if nobody but us can access this mm.

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/