Re: [PATCH] SELinux fixups needed for preemptable RCU from -rt

From: Paul E. McKenney
Date: Mon Apr 21 2008 - 21:12:50 EST


On Mon, Apr 21, 2008 at 08:54:51PM -0400, Steven Rostedt wrote:
>
> On Mon, 21 Apr 2008, Paul E. McKenney wrote:
> > @@ -807,8 +810,14 @@ int avc_ss_reset(u32 seqno)
> >
> > for (i = 0; i < AVC_CACHE_SLOTS; i++) {
> > spin_lock_irqsave(&avc_cache.slots_lock[i], flag);
> > + /*
> > + * On -rt the outer spinlock does not prevent RCU
> > + * from being performed:
>
> I would suggest to change this comment to "With preemptible RCU" from
> "On -rt".

Good point... How about the following?

> -- Steve
>
> > + */
> > + rcu_read_lock();
> > list_for_each_entry(node, &avc_cache.slots[i], list)
> > avc_node_delete(node);
> > + rcu_read_unlock();
> > spin_unlock_irqrestore(&avc_cache.slots_lock[i], flag);
> > }


Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx> (comment change)
---
security/selinux/avc.c | 9 +++++++++
security/selinux/netif.c | 2 ++
2 files changed, 11 insertions(+)

Index: linux-2.6.24.4-rt4/security/selinux/avc.c
===================================================================
--- linux-2.6.24.4-rt4.orig/security/selinux/avc.c 2008-03-24 19:05:09.000000000 -0400
+++ linux-2.6.24.4-rt4/security/selinux/avc.c 2008-03-24 19:06:41.000000000 -0400
@@ -312,6 +312,7 @@ static inline int avc_reclaim_node(void)
if (!spin_trylock_irqsave(&avc_cache.slots_lock[hvalue], flags))
continue;

+ rcu_read_lock();
list_for_each_entry(node, &avc_cache.slots[hvalue], list) {
if (atomic_dec_and_test(&node->ae.used)) {
/* Recently Unused */
@@ -319,11 +320,13 @@ static inline int avc_reclaim_node(void)
avc_cache_stats_incr(reclaims);
ecx++;
if (ecx >= AVC_CACHE_RECLAIM) {
+ rcu_read_unlock();
spin_unlock_irqrestore(&avc_cache.slots_lock[hvalue], flags);
goto out;
}
}
}
+ rcu_read_unlock();
spin_unlock_irqrestore(&avc_cache.slots_lock[hvalue], flags);
}
out:
@@ -807,8 +810,14 @@ int avc_ss_reset(u32 seqno)

for (i = 0; i < AVC_CACHE_SLOTS; i++) {
spin_lock_irqsave(&avc_cache.slots_lock[i], flag);
+ /*
+ * With preemptable RCU, the outer spinlock does not
+ * prevent RCU grace periods from ending.
+ */
+ rcu_read_lock();
list_for_each_entry(node, &avc_cache.slots[i], list)
avc_node_delete(node);
+ rcu_read_unlock();
spin_unlock_irqrestore(&avc_cache.slots_lock[i], flag);
}

Index: linux-2.6.24.4-rt4/security/selinux/netif.c
===================================================================
--- linux-2.6.24.4-rt4.orig/security/selinux/netif.c 2008-03-24 19:05:09.000000000 -0400
+++ linux-2.6.24.4-rt4/security/selinux/netif.c 2008-03-24 19:06:41.000000000 -0400
@@ -210,6 +210,7 @@ static void sel_netif_flush(void)
{
int idx;

+ rcu_read_lock();
spin_lock_bh(&sel_netif_lock);
for (idx = 0; idx < SEL_NETIF_HASH_SIZE; idx++) {
struct sel_netif *netif;
@@ -218,6 +219,7 @@ static void sel_netif_flush(void)
sel_netif_destroy(netif);
}
spin_unlock_bh(&sel_netif_lock);
+ rcu_read_unlock();
}

static int sel_netif_avc_callback(u32 event, u32 ssid, u32 tsid,
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/