Re: [PATCH 2/2] net/netlabel: Avoid call to genlmsg_cancel

[Paul Moore <paul.moore@hp.com>]

On Fri, 2011-01-28 at 15:58 +0100, Julia Lawall wrote:
> On Fri, 28 Jan 2011, Paul Moore wrote:
> 
> > On Fri, 2011-01-28 at 15:17 +0100, Julia Lawall wrote:
> > > genlmsg_cancel subtracts some constants from its second argument before
> > > calling nlmsg_cancel.  nlmsg_cancel then calls nlmsg_trim on the same
> > > arguments.  nlmsg_trim tests for NULL before doing any computation, but a
> > > NULL second argument to genlmsg_cancel is no longer NULL due to the initial
> > > subtraction.  Nothing else happens in this execution, so the call to
> > > genlmsg_cancel is simply unnecessary in this case.
> > > 
> > > The semantic match that finds this problem is as follows:
> > > (http://coccinelle.lip6.fr/)
> > > 
> > > // <smpl>
> > > @@
> > > expression data;
> > > @@
> > > 
> > > if (data = NULL) { ...
> > > * genlmsg_cancel(..., data);
> > >   ...
> > >   return ...;
> > > }
> > > // </smpl>
> > > 
> > > Signed-off-by: Julia Lawall <julia@diku.dk>
> > 
> > In all of the cases below, these functions are called multiple times to
> > generate data chunks (additional netlink attributes) which are appended
> > to an existing skbuff.  I believe that the calls to genlmsg_cancel() are
> > still needed to help cleanup in the case where the functions fail on the
> > Nth call.
> > 
> > If I'm wrong, feel free to enlighten me.
> 
> Perhaps something is needed, but I don't see how the current code can 
> work.  The call is genlmsg_cancel(cb_arg->skb, NULL) in each case.

Ah yes, you're right.  You will have to forgive me as it has been quite
a while since I have looked at NetLabel's netlink code.

You also might consider putting a NULL check in genlmsg_cancel() similar
to the check nlmsg_trim(); that seems like a worthwhile addition.

> The definition of genlmsg_cancel is:
> 
> static inline void genlmsg_cancel(struct sk_buff *skb, void *hdr)
> {
> 	nlmsg_cancel(skb, hdr - GENL_HDRLEN - NLMSG_HDRLEN);
> }
> 
> Now the second argument to nlmsg_cancel is essentially a negative integer 
> (or a very large pointer).
> 
> nlmsg_cancel will call nlmsg_trim, which is defined as follows:
> 
> static inline void nlmsg_trim(struct sk_buff *skb, const void *mark)
> {
> 	if (mark)
> 		skb_trim(skb, (unsigned char *) mark - skb->data);
> }
> 
> I guess that the subtraction is going to result in an even larger negative 
> number.  The whole process is likely to end in doing nothing in the 
> definition of skb_trim, which is as follows:
> 
> void skb_trim(struct sk_buff *skb, unsigned int len)
> {
> 	if (skb->len > len)
> 		__skb_trim(skb, len);
> }
> 
> since the result of casting a negative number to unsigned is likely to be 
> larger than skb->len.
> 
> 
> > > ---
> > >  net/netlabel/netlabel_cipso_v4.c  |    2 +-
> > >  net/netlabel/netlabel_mgmt.c      |    4 ++--
> > >  net/netlabel/netlabel_unlabeled.c |    2 +-
> > >  3 files changed, 4 insertions(+), 4 deletions(-)
> > > 
> > > diff --git a/net/netlabel/netlabel_cipso_v4.c b/net/netlabel/netlabel_cipso_v4.c
> > > index 5f14c84..0a1f77b 100644
> > > --- a/net/netlabel/netlabel_cipso_v4.c
> > > +++ b/net/netlabel/netlabel_cipso_v4.c
> > > @@ -635,7 +635,7 @@ static int netlbl_cipsov4_listall_cb(struct cipso_v4_doi *doi_def, void *arg)
> > >  			   cb_arg->seq, &netlbl_cipsov4_gnl_family,
> > >  			   NLM_F_MULTI, NLBL_CIPSOV4_C_LISTALL);
> > >  	if (data = NULL)
> > > -		goto listall_cb_failure;
> > > +		return ret_val;
> > >  
> > >  	ret_val = nla_put_u32(cb_arg->skb, NLBL_CIPSOV4_A_DOI, doi_def->doi);
> > >  	if (ret_val != 0)
> > > diff --git a/net/netlabel/netlabel_mgmt.c b/net/netlabel/netlabel_mgmt.c
> > > index 998e85e..daaa01d 100644
> > > --- a/net/netlabel/netlabel_mgmt.c
> > > +++ b/net/netlabel/netlabel_mgmt.c
> > > @@ -452,7 +452,7 @@ static int netlbl_mgmt_listall_cb(struct netlbl_dom_map *entry, void *arg)
> > >  			   cb_arg->seq, &netlbl_mgmt_gnl_family,
> > >  			   NLM_F_MULTI, NLBL_MGMT_C_LISTALL);
> > >  	if (data = NULL)
> > > -		goto listall_cb_failure;
> > > +		return ret_val;
> > >  
> > >  	ret_val = netlbl_mgmt_listentry(cb_arg->skb, entry);
> > >  	if (ret_val != 0)
> > > @@ -617,7 +617,7 @@ static int netlbl_mgmt_protocols_cb(struct sk_buff *skb,
> > >  			   &netlbl_mgmt_gnl_family, NLM_F_MULTI,
> > >  			   NLBL_MGMT_C_PROTOCOLS);
> > >  	if (data = NULL)
> > > -		goto protocols_cb_failure;
> > > +		return ret_val;
> > >  
> > >  	ret_val = nla_put_u32(skb, NLBL_MGMT_A_PROTOCOL, protocol);
> > >  	if (ret_val != 0)
> > > diff --git a/net/netlabel/netlabel_unlabeled.c b/net/netlabel/netlabel_unlabeled.c
> > > index e2b0a68..b5d3945 100644
> > > --- a/net/netlabel/netlabel_unlabeled.c
> > > +++ b/net/netlabel/netlabel_unlabeled.c
> > > @@ -1141,7 +1141,7 @@ static int netlbl_unlabel_staticlist_gen(u32 cmd,
> > >  			   cb_arg->seq, &netlbl_unlabel_gnl_family,
> > >  			   NLM_F_MULTI, cmd);
> > >  	if (data = NULL)
> > > -		goto list_cb_failure;
> > > +		return ret_val;
> > >  
> > >  	if (iface->ifindex > 0) {
> > >  		dev = dev_get_by_index(&init_net, iface->ifindex);
> > > 
> > 
> > -- 
> > paul moore
> > linux @ hp
> > 
> > 
> > --
> > To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > 

-- 
paul moore
linux @ hp