batman-adv: handle hash_add() return value in the TT code

  hash_add() can fail and for this reason its return value has to be
correctly interpreted to avoid memory leaks.

Signed-off-by: Antonio Quartulli <ordex@autistici.org>
---

This patch needs
"[PATCH] batman-adv: hash_add() has to discriminate on the return value"
to work

 translation-table.c |   27 +++++++++++++++++++++------
 1 files changed, 21 insertions(+), 6 deletions(-)
  

On Sunday, July 10, 2011 01:50:57 Antonio Quartulli wrote:
> @@ -220,8 +221,14 @@ void tt_local_add(struct net_device *soft_iface, const
> uint8_t *addr) * (consistency check) */
>  	tt_local_entry->flags |= TT_CLIENT_NEW;
> 
> -	hash_add(bat_priv->tt_local_hash, compare_ltt, choose_orig,
> -		 tt_local_entry, &tt_local_entry->hash_entry);
> +	hash_added = hash_add(bat_priv->tt_local_hash, compare_ltt,
> +			      choose_orig, tt_local_entry,
> +			      &tt_local_entry->hash_entry);
> +	if (hash_added != 0)
> +		tt_local_entry_free_ref(tt_local_entry);
> +
> +	if (hash_added < 0)
> +		goto out;

No matter whether we add a new element to the tt hash or not we always 
generate a log message and throw a tt_local_event(). Is that what we want ?


> @@ -519,10 +527,17 @@ int tt_global_add(struct bat_priv *bat_priv, struct
> orig_node *orig_node, tt_global_entry->roam_at = 0;
>  		atomic_set(&tt_global_entry->refcount, 2);
> 
> -		hash_add(bat_priv->tt_global_hash, compare_gtt,
> -			 choose_orig, tt_global_entry,
> -			 &tt_global_entry->hash_entry);
> -		atomic_inc(&orig_node->tt_size);
> +		hash_added = hash_add(bat_priv->tt_global_hash, compare_gtt,
> +				      choose_orig, tt_global_entry,
> +				      &tt_global_entry->hash_entry);
> +		if (hash_added != 0)
> +			tt_global_entry_free_ref(tt_global_entry);
> +
> +		if (hash_added < 0)
> +			goto out;
> +
> +		if (!hash_added)
> +			atomic_inc(&orig_node->tt_size);

Here we only generate a log message when the element has been added or already 
exists. Do we want that too ?

Regards,
Marek
  

On Sat, Jul 23, 2011 at 11:17:22 +0200, Marek Lindner wrote:
> On Sunday, July 10, 2011 01:50:57 Antonio Quartulli wrote:
> > @@ -220,8 +221,14 @@ void tt_local_add(struct net_device *soft_iface, const
> > uint8_t *addr) * (consistency check) */
> >  	tt_local_entry->flags |= TT_CLIENT_NEW;
> > 
> > -	hash_add(bat_priv->tt_local_hash, compare_ltt, choose_orig,
> > -		 tt_local_entry, &tt_local_entry->hash_entry);
> > +	hash_added = hash_add(bat_priv->tt_local_hash, compare_ltt,
> > +			      choose_orig, tt_local_entry,
> > +			      &tt_local_entry->hash_entry);
> > +	if (hash_added != 0)
> > +		tt_local_entry_free_ref(tt_local_entry);
> > +
> > +	if (hash_added < 0)
> > +		goto out;
> 
> No matter whether we add a new element to the tt hash or not we always 
> generate a log message and throw a tt_local_event(). Is that what we want ?

What about going to 'out' in case of hash_added != 0? In both cases (<0 or >0) 
we cannot go ahead with the function as the entry has already been added or it
is not possible to add it.

> 
> 
> > @@ -519,10 +527,17 @@ int tt_global_add(struct bat_priv *bat_priv, struct
> > orig_node *orig_node, tt_global_entry->roam_at = 0;
> >  		atomic_set(&tt_global_entry->refcount, 2);
> > 
> > -		hash_add(bat_priv->tt_global_hash, compare_gtt,
> > -			 choose_orig, tt_global_entry,
> > -			 &tt_global_entry->hash_entry);
> > -		atomic_inc(&orig_node->tt_size);
> > +		hash_added = hash_add(bat_priv->tt_global_hash, compare_gtt,
> > +				      choose_orig, tt_global_entry,
> > +				      &tt_global_entry->hash_entry);
> > +		if (hash_added != 0)
> > +			tt_global_entry_free_ref(tt_global_entry);
> > +
> > +		if (hash_added < 0)
> > +			goto out;
> > +
> > +		if (!hash_added)
> > +			atomic_inc(&orig_node->tt_size);
> 
> Here we only generate a log message when the element has been added or already 
> exists. Do we want that too ?
> 

The same as before.

Cheers,
Antonio
  

On Sunday, July 31, 2011 23:53:19 Antonio Quartulli wrote:
> > > +	hash_added = hash_add(bat_priv->tt_local_hash, compare_ltt,
> > > +			      choose_orig, tt_local_entry,
> > > +			      &tt_local_entry->hash_entry);
> > > +	if (hash_added != 0)
> > > +		tt_local_entry_free_ref(tt_local_entry);
> > > +
> > > +	if (hash_added < 0)
> > > +		goto out;
> > 
> > No matter whether we add a new element to the tt hash or not we always
> > generate a log message and throw a tt_local_event(). Is that what we want
> > ?
> 
> What about going to 'out' in case of hash_added != 0? In both cases (<0 or
> >0) we cannot go ahead with the function as the entry has already been
> added or it is not possible to add it.

If I am not mistaken we should keep the additional "tt_local_entry_free_ref()" 
because we initialize the newly allocated struct with 2.
Still, the original problem remains: We generate a log message claiming to 
have added a new entry even though it might fail later. You have to move the 
log message to the end of the function.


> > > +		hash_added = hash_add(bat_priv->tt_global_hash, compare_gtt,
> > > +				      choose_orig, tt_global_entry,
> > > +				      &tt_global_entry->hash_entry);
> > > +		if (hash_added != 0)
> > > +			tt_global_entry_free_ref(tt_global_entry);
> > > +
> > > +		if (hash_added < 0)
> > > +			goto out;
> > > +
> > > +		if (!hash_added)
> > > +			atomic_inc(&orig_node->tt_size);
> > 
> > Here we only generate a log message when the element has been added or
> > already exists. Do we want that too ?
> 
> The same as before.

Yeah, same problem as above.  :-)

Cheers,
Marek