Re: [PATCH 00/16 v3] tracing: Add new file system tracefs

From: Steven Rostedt
Date: Mon Jan 26 2015 - 19:19:54 EST


On Mon, 26 Jan 2015 21:46:58 +0000
Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:

> On Mon, Jan 26, 2015 at 03:42:59PM -0500, Steven Rostedt wrote:
> > On Mon, 26 Jan 2015 19:30:49 +0000
> > Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> >
> > > You are still fighting an inconvenient API, but now it's not debugfs one -
> > > it's your copy thereof. Why not give your instances/ an inode_operations
> > > of its own? One with ->mkdir() and ->rmdir(), leaving all other directories
> > > as-is. That way you don't need the secondary methods at all. And sure,
> > > debugfs_create_dir() grabs ->i_mutex on parent, making you drop that in
> > > your ->mkdir() if you want to call it. But now you are not talking to it -
> > > just to your own code, where you are free to change the calling conventions,
> > > making it caller's responsibility to get that ->i_mutex. The same goes for
> > > the rmdir side...
> >
> > The vfs layer grabs the i_mutex, which needs to be dropped.
>
> What for? Just keep it through your instance_mkdir/instance_rmdir and be
> done with that. Sure, it means that you need variants of file/directory
> creation/removal primitives that would assume that parent is already locked
> by caller (and would leave unlocking to the caller as well). So add them...

I originally had this, but the issue isn't with i_mutex only. The
instance_mkdir and instance_rmdir need to grab trace_types_lock. This
is also held when calling into tracefs.

That is, we can not hold i_mutex and take trace_types_lock.

trace_types_lock needs to be held with the creation or destruction of
events, which is what mkdir an rmdir do.


-- Steve
--
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/