Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
Date:   Wed, 1 Nov 2023 18:54:07 +0100
From:   Greg KH <gregkh@linuxfoundation.org>
To:     =?iso-8859-1?Q?Jos=E9?= Pekkarinen <jose.pekkarinen@foxhound.fi>
Cc:     rafael@kernel.org, len.brown@intel.com, pavel@ucw.cz,
        skhan@linuxfoundation.org, linux-pm@vger.kernel.org,
        linux-kernel@vger.kernel.org,
        linux-kernel-mentees@lists.linuxfoundation.org,
        syzbot+95f2e2439b97575ec3c0@syzkaller.appspotmail.com
Subject: Re: [PATCH] drivers core: lookup sysfs power group before removal
Message-ID: <2023110139-dupe-snipping-5700@gregkh>
References: <20231101173627.2658-1-jose.pekkarinen@foxhound.fi>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20231101173627.2658-1-jose.pekkarinen@foxhound.fi>
Precedence: bulk

On Wed, Nov 01, 2023 at 07:36:27PM +0200, Jos? Pekkarinen wrote:
> Hinted by syzboot, there is a few cases where the sysfs power group may
> not be there, like the failure while adding it, or adding its runtime
> group, or when the sysfs firmware loader fallback fail to populate. In
> the last case, the device_del function will be called leading to attempt
> to remove the sysfs group. This patch will lookup for it in advance to
> grant that it is effectively there before cleaning it up.
> 
> Reported-by: syzbot+95f2e2439b97575ec3c0@syzkaller.appspotmail.com
> 
> Signed-off-by: Jos? Pekkarinen <jose.pekkarinen@foxhound.fi>
> ---
>  drivers/base/power/sysfs.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/base/power/sysfs.c b/drivers/base/power/sysfs.c
> index a1474fb67db9..6601729c4698 100644
> --- a/drivers/base/power/sysfs.c
> +++ b/drivers/base/power/sysfs.c
> @@ -834,5 +834,7 @@ void dpm_sysfs_remove(struct device *dev)
>  	dev_pm_qos_constraints_destroy(dev);
>  	rpm_sysfs_remove(dev);
>  	sysfs_unmerge_group(&dev->kobj, &pm_wakeup_attr_group);
> -	sysfs_remove_group(&dev->kobj, &pm_attr_group);
> +
> +	if (kernfs_find_and_get((&dev->kobj)->sd, pm_attr_group.name))
> +		sysfs_remove_group(&dev->kobj, &pm_attr_group);

What's to keep it from going away right after finding it?

In other words, what is wrong with removing a group that is not there?
What error happens?  It should be fine, or are you seeing real code
failures somewhere?

Also, I think you just leaked a reference count here, how was this
tested?

thanks,

greg k-h