Re: [PATCH v2 1/2] EDAC/device_sysfs: Fix calling kobject_put() with ->state_initialized unset

[Harshit Mogalapalli]

On 25/01/24 6:39 pm, Borislav Petkov wrote:
> On Tue, Nov 28, 2023 at 10:30:35AM -0800, Harshit Mogalapalli wrote:
> > In edac_device_register_sysfs_main_kobj(), when dev_root is NULL,
Thanks for checking this.

> When is dev_root NULL?
>
> A real use case or this is just from code staring?
This is based on static analysis, not real testing.

> > diff --git a/drivers/edac/edac_device_sysfs.c b/drivers/edac/edac_device_sysfs.c
> > index 010c26be5846..4cac14cbdb60 100644
> > --- a/drivers/edac/edac_device_sysfs.c
> > +++ b/drivers/edac/edac_device_sysfs.c
> > @@ -253,11 +253,13 @@ int edac_device_register_sysfs_main_kobj(struct edac_device_ctl_info *edac_dev)
> >   
> >   	/* register */
> >   	dev_root = bus_get_dev_root(edac_subsys);
> > -	if (dev_root) {
> > -		err = kobject_init_and_add(&edac_dev->kobj, &ktype_device_ctrl,
> > -					   &dev_root->kobj, "%s", edac_dev->name);
>
> What cb4a0bec0bb9 did looks wrong. That if (err) check should be inside
> the if (dev_root).
>
> IOW, that function's error checking needs balancing. Something like
> this below.
>
> And looking at the other one, it has the same issue...
>
> ---
> diff --git a/drivers/edac/edac_device_sysfs.c b/drivers/edac/edac_device_sysfs.c
> index 237a542e045a..3476ef13e681 100644
> --- a/drivers/edac/edac_device_sysfs.c
> +++ b/drivers/edac/edac_device_sysfs.c
> @@ -228,8 +228,8 @@ static struct kobj_type ktype_device_ctrl = {
>    */
>   int edac_device_register_sysfs_main_kobj(struct edac_device_ctl_info *edac_dev)
>   {
> -	struct device *dev_root;
>   	const struct bus_type *edac_subsys;
> +	struct device *dev_root;
>   	int err = -ENODEV;
>   
>   	edac_dbg(1, "\n");
> @@ -243,26 +243,26 @@ int edac_device_register_sysfs_main_kobj(struct edac_device_ctl_info *edac_dev)
>   	/* Init the devices's kobject */
>   	memset(&edac_dev->kobj, 0, sizeof(struct kobject));
>   
> -	/* Record which module 'owns' this control structure
> -	 * and bump the ref count of the module
> +	/*
> +	 * Record which module 'owns' this control structure and bump
> +	 * the ref count of the module
>   	 */
>   	edac_dev->owner = THIS_MODULE;
>   
>   	if (!try_module_get(edac_dev->owner))
>   		goto err_out;
>   
> -	/* register */
>   	dev_root = bus_get_dev_root(edac_subsys);
> -	if (dev_root) {
> -		err = kobject_init_and_add(&edac_dev->kobj, &ktype_device_ctrl,
> -					   &dev_root->kobj, "%s", edac_dev->name);
> -		put_device(dev_root);
> -	}
> +	if (!dev_root)
> +		goto err_module;
> +
> +	err = kobject_init_and_add(&edac_dev->kobj, &ktype_device_ctrl,
> +				   &dev_root->kobj, "%s", edac_dev->name);
>   	if (err) {
> -		edac_dbg(1, "Failed to register '.../edac/%s'\n",
> -			 edac_dev->name);
> +		edac_dbg(1, "Failed to register '.../edac/%s'\n", edac_dev->name);
>   		goto err_kobj_reg;
>   	}
> +
>   	kobject_uevent(&edac_dev->kobj, KOBJ_ADD);
>   
>   	/* At this point, to 'free' the control struct,
> @@ -273,9 +273,11 @@ int edac_device_register_sysfs_main_kobj(struct edac_device_ctl_info *edac_dev)
>   
>   	return 0;
>   
> -	/* Error exit stack */
>   err_kobj_reg:
>   	kobject_put(&edac_dev->kobj);
> +	put_device(dev_root);

In this diff, put_device(dev_root) is not called on the success path. I 
checked couple of other callers of bus_get_dev_root() and they call 
put_device(dev_root) after using dev_root. I think we need to have 
put_device() on the success path as well.

> +
> +err_module:
>   	module_put(edac_dev->owner);

An alternate diff could be:

diff --git a/drivers/edac/edac_device_sysfs.c 
b/drivers/edac/edac_device_sysfs.c
index 237a542e045a..0c8e66e40af8 100644
--- a/drivers/edac/edac_device_sysfs.c
+++ b/drivers/edac/edac_device_sysfs.c
@@ -275,7 +275,8 @@ int edac_device_register_sysfs_main_kobj(struct 
edac_device_ctl_info *edac_dev)

        /* Error exit stack */
 err_kobj_reg:
-       kobject_put(&edac_dev->kobj);
+       if(dev_root)
+               kobject_put(&edac_dev->kobj);
        module_put(edac_dev->owner);

 err_out:

Thanks,
Harshit


>   err_out: