Re: [PATCH RFD] alternative kobject release wait mechanism

[Alan Stern <stern@rowland.harvard.edu>]

On Mon, 23 Apr 2007, Cornelia Huck wrote:

> On Sun, 22 Apr 2007 10:40:51 -0700,
> Greg KH <greg@kroah.com> wrote:
> 
> > > Looking some more, kobject_get_path() is used for kobject renaming,
> > > uevent handling, and a little bit in the input core.  None of these things 
> > > should try to access a kobject after it has been del()ed.  After all, it's 
> > > no longer present in the filesystem so it doesn't _have_ a path.
> > 
> > But we _have_ to have a full path at that time to tell userspace what
> > just went away.  That is the main reason we enforce this (there were
> > tons of issues with scsi devices and this in the past which is what
> > caused us to enforce this.)
> 
> What we need to ensure is that the parent device is kept at least until
> all children, grandchildren and so on are done with their uevent needs.
> This would imply it needed to stay as long as those children,
> grandchildren, ... are still registered. Would it be save to suggest
> that a ->remove callback would always need to unregister the children?
> Then putting the parent reference at the end of kobject_del() (which is
> after kobject_uevent() in kobject_unregister()) should be safe.

Yes, this is the weak spot.

For some reason I had assumed that it was illegal to unregister a device
while it had registered children (just as it is illegal to rmdir a
non-empty directory).  If it isn't illegal, then perhaps we should arrange
things so that device_del() will recursively call itself for all the
device's children.

> Question: What now?
> 
> 1. Make it mandatory that all children must be unregistered when
> device_del() returns.
> 
> 2. Don't demand an empty directory in sysfs_drop_dentry().

I'm in favor of (1).  But instead of making it mandatory, simply force it
to be true by having device_del() call itself for all remaining children.

For this to be safe, we also have to allow device_del() to be called 
multiple times (since the device's owner might not be aware that the core 
had already unregistered it).  That's no problem; just add:

	if (!device_is_registered(dev))
		return;

to the start of the routine.

Alan Stern

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