Re: [PATCH net 3/5] netdevsim: avoid debugfs warning message when module is remove

[Taehee Yoo <ap420073@gmail.com>]

On Sun, 12 Jan 2020 at 23:45, Jakub Kicinski <kuba@kernel.org> wrote:
>

Hi Jakub,
Thank you for catching the problem!

> On Sat, 11 Jan 2020 16:37:23 +0000, Taehee Yoo wrote:
> > When module is being removed, it couldn't be held by try_module_get().
> > debugfs's open function internally tries to hold file_operation->owner
> > if .owner is set.
> > If holding owner operation is failed, it prints a warning message.
>
> > [  412.227709][ T1720] debugfs file owner did not clean up at exit: ipsec
>
> > In order to avoid the warning message, this patch makes netdevsim module
> > does not set .owner. Unsetting .owner is safe because these are protected
> > by inode_lock().
>
> So inode_lock will protect from the code getting unloaded/disappearing?
> At a quick glance at debugs code it doesn't seem that inode_lock would
> do that. Could you explain a little more to a non-fs developer like
> myself? :)
>
> Alternatively should we perhaps hold a module reference for each device
> created and force user space to clean up the devices? That may require
> some fixes to the test which use netdevsim.
>

Sorry, I misunderstood the debugfs logic.
inode_lock() is called by debugfs_remove() and debugfs_create_file().
It doesn't protect read and write operations.

Currently, I have been taking look at debugfs_file_{get/put}() function,
which increases and decreases the reference counter.
In the __debugfs_file_removed(), this reference counter is used for
waiting read and write operations. Unfortunately, the
__debugfs_file_removed() isn't used because of "dentry->d_flags" value.
So, I'm looking for a way to use these functions.

Thanks a lot!
Taehee Yoo