Received-SPF: pass (google.com: domain of linux-kernel+bounces-145686-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Date: Mon, 15 Apr 2024 14:02:33 -0400
From: Kent Overstreet <kent.overstreet@linux.dev>
To: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Johannes Berg <johannes.berg@intel.com>, 
	linux-bcachefs@vger.kernel.org, linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: debugfs revoke broken?
Message-ID: <ra3ig25pwua6wm6v7vqexjvwybfpk23i5qqfeihth4f3zygjb6@7ykqlkyu25tb>
References: <nxucitm2agdzdodrkm5rjyuwnnf6keivjiqlp5rn6poxkpkye6@yor2lprsxh7x>
 <2024041557-stereo-wafer-1551@gregkh>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <2024041557-stereo-wafer-1551@gregkh>

On Mon, Apr 15, 2024 at 10:47:56AM +0200, Greg Kroah-Hartman wrote:
> So this file was open when debugfs_remove() was called?

Seems likely.

> Any chance you can bisect?  We just fixed some issues here in
> 952c3fce297f ("debugfs: fix wait/cancellation handling during remove")
> that I thought should have handled this.  If you revert that commit,
> does things work again?  And/or what about commit 8c88a474357e
> ("debugfs: add API to allow debugfs operations cancellation")?  Maybe we
> need to go back to not having completions at all in the debugfs remove
> path and just live with waiting for the files to be removed before
> continuing (which should at the least, resolve the issue you see here,
> while slowing things down a bit.)

No, this doesn't occur often enough to bisect, unfortunately. All my
tests bang on sysfs/debugfs in the background to torture test those
codepaths, and I've seen 2 occurrences out of thousands of test runs -
it's rare.