Re: [RFC PATCH] tracing: Atomically get refcounts of event_call and trace_array

[Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>]

(2013/07/05 9:32), Oleg Nesterov wrote:
> On 07/04, Masami Hiramatsu wrote:
>>
>> Currently ftrace_open_generic_file gets an event_file from
>> inode->i_private, and then locks event_mutex and gets refcount.
>> However, this can cause a race as below scenario;
>>
>> CPU0                              CPU1
>> open(kprobe_events)
>>   trace_remove_event_call()    open(enable)
>>     lock event_mutex             get event_file from inode->i_private
>>       event_remove()             wait for unlock event_mutex
>>         ...
>>         free event_file
>>     unlock event_mutex
>>                                  lock event_mutex
>>                                  add refcount of event_file->call (*)
>>
>> So, at (*) point, the event_file is already freed and we
>> may access the corrupted object.
> 
> Yes, but the same can happen with event_call, so it seems that
> this patch is not enough too.

Oops, right!

> Say, open(id) can take event_mutex when the caller of
> trace_remove_event_call() has already freed ftrace_event_call.
> 
> Or I missed something...
> 
> Perhaps we can rely on d_unlinked() ? IOW, the caller of
> __ftrace_event_call_get() should take event_mutex, check
> d_unhashed(f_dentry) and only then do _get().

It sounds a good idea! :)

Thank you!

> 
> Nasty, I agree.
> 
> Oleg.
> 
> 


-- 
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com