[NEIGH]: Fix race between neigh_parms_release and neightbl_fill_parms
authorPavel Emelyanov <xemul@openvz.org>
Thu, 10 Jan 2008 11:48:38 +0000 (03:48 -0800)
committerDavid S. Miller <davem@davemloft.net>
Thu, 10 Jan 2008 11:48:38 +0000 (03:48 -0800)
The neightbl_fill_parms() is called under the write-locked tbl->lock
and accesses the parms->dev. The negh_parm_release() calls the
dev_put(parms->dev) without this lock. This creates a tiny race window
on which the parms contains potentially stale dev pointer.

To fix this race it's enough to move the dev_put() upper under the
tbl->lock, but note, that the parms are held by neighbors and thus can
live after the neigh_parms_release() is called, so we still can have a
parm with bad dev pointer.

I didn't find where the neigh->parms->dev is accessed, but still think
that putting the dev is to be done in a place, where the parms are
really freed. Am I right with that?

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/core/neighbour.c

index 29b8ee4e35d6dc53f32064c1c067c7287ce80924..cc8a2f190acfa75dd9cd0fef1149e82689fcfccb 100644 (file)
@@ -1316,8 +1316,6 @@ void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
                        *p = parms->next;
                        parms->dead = 1;
                        write_unlock_bh(&tbl->lock);
-                       if (parms->dev)
-                               dev_put(parms->dev);
                        call_rcu(&parms->rcu_head, neigh_rcu_free_parms);
                        return;
                }
@@ -1328,6 +1326,8 @@ void neigh_parms_release(struct neigh_table *tbl, struct neigh_parms *parms)
 
 void neigh_parms_destroy(struct neigh_parms *parms)
 {
+       if (parms->dev)
+               dev_put(parms->dev);
        kfree(parms);
 }