Re: [PATCH 1/2] tracing: fix bad use of igrab in trace_uprobe.c

[Song Liu <songliubraving@fb.com>]

> On Apr 18, 2018, at 7:25 AM, Steven Rostedt <rostedt@goodmis.org> wrote:
> 
> On Wed, 18 Apr 2018 16:03:42 +0200
> Miklos Szeredi <miklos@szeredi.hu> wrote:
> 
>>> @@ -937,7 +928,8 @@ probe_event_enable(struct trace_uprobe *tu, struct trace_event_file *file,
>>>                goto err_flags;
>>> 
>>>        tu->consumer.filter = filter;
>>> -       ret = uprobe_register(tu->inode, tu->offset, &tu->consumer);
>>> +       ret = uprobe_register(d_inode(tu->path.dentry), tu->offset,
>>> +                             &tu->consumer);  
>> 
>> It is not entirely clear how the lifetime of uprobe relates to the
>> lifetime of trace_uprobe.  Is the uprobe object never going to survive
>> its creator trace_uprobe object?
> 
> Not exactly sure what you mean here.
> 
> The trace_uprobe (the probe event) is created, it doesn't do anything
> until it is enabled. This function is called when it is enabled. The
> trace_uprobe (probe event) can not be deleted while it is enabled
> (EBUSY).
> 
> Are you asking what happens if the file is deleted while it has probe?
> That I don't know about (haven't tried it out). But I would hope that
> it keeps a reference to the inode, isn't that what the igrab is for?
> And is now being replaced by a reference on the path, or is that the
> problem?
> 
> -- Steve
> 

Just as Miklos pointed out, I run tests with the uprobe and confirmed 
that igrab() is not sufficient to prevent umount. When we change it to 
path_get()/path_put(), umount will abort because of the trace_uprobe. 

Song


>> 
>> If that's the case, it warrants a comment.  If that's not the case,
>> then the path would need to be passed to uprobe_resister() which would
>> need to obtain its own reference.
>> 
>>>        if (ret)
>>>                goto err_buffer;
>>>