Re: [PATCHv2] bonding: wait for sysfs kobject destruction before freeing struct slave

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Tue, Nov 17, 2020 at 12:34:01PM -0800, Jakub Kicinski wrote:
> On Fri, 13 Nov 2020 17:12:44 +0000 Jamie Iles wrote:
> > syzkaller found that with CONFIG_DEBUG_KOBJECT_RELEASE=y, releasing a
> > struct slave device could result in the following splat:
> 
> > This is a potential use-after-free if the sysfs nodes are being accessed
> > whilst removing the struct slave, so wait for the object destruction to
> > complete before freeing the struct slave itself.
> > 
> > Fixes: 07699f9a7c8d ("bonding: add sysfs /slave dir for bond slave devices.")
> > Fixes: a068aab42258 ("bonding: Fix reference count leak in bond_sysfs_slave_add.")
> 
> This code looks surprising, although admittedly my kobj understanding
> is cursory at best. So CCing Greg to keep me honest.
> 
> kobj itself is a refcounting mechanism. Adding another refcount and
> then releasing the reference from .release method of kobject looks like
> a pointless duplication.

Yes, that's wrong.

> Just free the object from the .release method. Why not?

Correct, that is what should be happening.  Otherwise it seems that
something else has a pointer to this object that forgot to increment it?

> 
> > diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
> > index 84ecbc6fa0ff..66e56642e6c2 100644
> > --- a/drivers/net/bonding/bond_main.c
> > +++ b/drivers/net/bonding/bond_main.c
> > @@ -1478,11 +1478,14 @@ static struct slave *bond_alloc_slave(struct bonding *bond)
> >  	}
> >  	INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
> >  
> > +	kref_init(&slave->ref);
> > +
> >  	return slave;
> >  }
> >  
> > -static void bond_free_slave(struct slave *slave)
> > +static void __bond_free_slave(struct kref *ref)
> >  {
> > +	struct slave *slave = container_of(ref, struct slave, ref);
> >  	struct bonding *bond = bond_get_bond_by_slave(slave);
> >  
> >  	cancel_delayed_work_sync(&slave->notify_work);
> > @@ -1492,6 +1495,16 @@ static void bond_free_slave(struct slave *slave)
> >  	kfree(slave);
> >  }
> >  
> > +void bond_slave_put_ref(struct slave *slave)
> > +{
> > +	kref_put(&slave->ref, __bond_free_slave);
> > +}
> > +
> > +void bond_slave_get_ref(struct slave *slave)
> > +{
> > +	kref_get(&slave->ref);
> > +}
> > +
> >  static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
> >  {
> >  	info->bond_mode = BOND_MODE(bond);
> > @@ -2007,7 +2020,7 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
> >  	dev_set_mtu(slave_dev, new_slave->original_mtu);
> >  
> >  err_free:
> > -	bond_free_slave(new_slave);
> > +	bond_slave_put_ref(new_slave);
> >  
> >  err_undo_flags:
> >  	/* Enslave of first slave has failed and we need to fix master's mac */
> > @@ -2187,7 +2200,7 @@ static int __bond_release_one(struct net_device *bond_dev,
> >  	if (!netif_is_bond_master(slave_dev))
> >  		slave_dev->priv_flags &= ~IFF_BONDING;
> >  
> > -	bond_free_slave(slave);
> > +	bond_slave_put_ref(slave);
> >  
> >  	return 0;
> >  }
> > diff --git a/drivers/net/bonding/bond_sysfs_slave.c b/drivers/net/bonding/bond_sysfs_slave.c
> > index 9b8346638f69..5f8aac715ee8 100644
> > --- a/drivers/net/bonding/bond_sysfs_slave.c
> > +++ b/drivers/net/bonding/bond_sysfs_slave.c
> > @@ -136,7 +136,15 @@ static const struct sysfs_ops slave_sysfs_ops = {
> >  	.show = slave_show,
> >  };
> >  
> > +static void slave_release(struct kobject *kobj)
> > +{
> > +	struct slave *slave = to_slave(kobj);
> > +
> > +	bond_slave_put_ref(slave);
> > +}
> > +
> >  static struct kobj_type slave_ktype = {
> > +	.release = slave_release,
> >  #ifdef CONFIG_SYSFS
> >  	.sysfs_ops = &slave_sysfs_ops,
> >  #endif
> > @@ -147,22 +155,26 @@ int bond_sysfs_slave_add(struct slave *slave)
> >  	const struct slave_attribute **a;
> >  	int err;
> >  
> > +	bond_slave_get_ref(slave);
> > +
> >  	err = kobject_init_and_add(&slave->kobj, &slave_ktype,
> >  				   &(slave->dev->dev.kobj), "bonding_slave");
> > -	if (err) {
> > -		kobject_put(&slave->kobj);
> > -		return err;
> > -	}
> > +	if (err)
> > +		goto out_put_slave;
> >  
> >  	for (a = slave_attrs; *a; ++a) {
> >  		err = sysfs_create_file(&slave->kobj, &((*a)->attr));
> > -		if (err) {
> > -			kobject_put(&slave->kobj);
> > -			return err;
> > -		}
> > +		if (err)
> > +			goto out_put_slave;
> >  	}
> >  
> >  	return 0;
> > +
> > +out_put_slave:
> > +	kobject_put(&slave->kobj);
> > +	bond_slave_put_ref(slave);
> > +
> > +	return err;
> >  }
> >  
> >  void bond_sysfs_slave_del(struct slave *slave)
> > diff --git a/include/net/bonding.h b/include/net/bonding.h
> > index 7d132cc1e584..e286ff4e0882 100644
> > --- a/include/net/bonding.h
> > +++ b/include/net/bonding.h
> > @@ -25,6 +25,7 @@
> >  #include <linux/etherdevice.h>
> >  #include <linux/reciprocal_div.h>
> >  #include <linux/if_link.h>
> > +#include <linux/kref.h>
> >  
> >  #include <net/bond_3ad.h>
> >  #include <net/bond_alb.h>
> > @@ -157,6 +158,7 @@ struct bond_parm_tbl {
> >  struct slave {
> >  	struct net_device *dev; /* first - useful for panic debug */
> >  	struct bonding *bond; /* our master */
> > +	struct kref ref;

Now you have 2 different reference counts for the same structure, a
guaranteed way to cause total confusion and mistakes about lifetime
rules.

This is not the correct way to fix it, use the reference count in the
kobject, that is what it is there for.

thanks,

greg k-h