netfilter: ecache: document extension area access rules

Once ct->ext gets free'd via kfree() rather than kfree_rcu we can't
access the extension area anymore without owning the conntrack.

This is a special case:

The worker is walking the pcpu dying list while holding dying list lock:
Neither ct nor ct->ext can be free'd until after the walk has completed.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

1 files changed, 15 insertions, 2 deletions

diff --git a/net/netfilter/nf_conntrack_ecache.c b/net/netfilter/nf_conntrack_ecache.c
index 6fba74b5aaf7..0d83c159671c 100644
--- a/net/netfilter/nf_conntrack_ecache.c
+++ b/net/netfilter/nf_conntrack_ecache.c
@@ -30,6 +30,7 @@
 static DEFINE_MUTEX(nf_ct_ecache_mutex);
 
 #define ECACHE_RETRY_WAIT (HZ/10)
+#define ECACHE_STACK_ALLOC (256 / sizeof(void *))
 
 enum retry_state {
 	STATE_CONGESTED,
@@ -39,11 +40,11 @@ enum retry_state {
 
 static enum retry_state ecache_work_evict_list(struct ct_pcpu *pcpu)
 {
-	struct nf_conn *refs[16];
+	struct nf_conn *refs[ECACHE_STACK_ALLOC];
+	enum retry_state ret = STATE_DONE;
 	struct nf_conntrack_tuple_hash *h;
 	struct hlist_nulls_node *n;
 	unsigned int evicted = 0;
-	enum retry_state ret = STATE_DONE;
 
 	spin_lock(&pcpu->lock);
 
@@ -54,10 +55,22 @@ static enum retry_state ecache_work_evict_list(struct ct_pcpu *pcpu)
 		if (!nf_ct_is_confirmed(ct))
 			continue;
 
+		/* This ecache access is safe because the ct is on the
+		 * pcpu dying list and we hold the spinlock -- the entry
+		 * cannot be free'd until after the lock is released.
+		 *
+		 * This is true even if ct has a refcount of 0: the
+		 * cpu that is about to free the entry must remove it
+		 * from the dying list and needs the lock to do so.
+		 */
 		e = nf_ct_ecache_find(ct);
 		if (!e || e->state != NFCT_ECACHE_DESTROY_FAIL)
 			continue;
 
+		/* ct is in NFCT_ECACHE_DESTROY_FAIL state, this means
+		 * the worker owns this entry: the ct will remain valid
+		 * until the worker puts its ct reference.
+		 */
 		if (nf_conntrack_event(IPCT_DESTROY, ct)) {
 			ret = STATE_CONGESTED;
 			break;