netfilter: nf_tables: adapt set backend to use GC transaction API

Use the GC transaction API to replace the old and buggy gc API and the
busy mark approach.

No set elements are removed from async garbage collection anymore,
instead the _DEAD bit is set on so the set element is not visible from
lookup path anymore. Async GC enqueues transaction work that might be
aborted and retried later.

rbtree and pipapo set backends does not set on the _DEAD bit from the
sync GC path since this runs in control plane path where mutex is held.
In this case, set elements are deactivated, removed and then released
via RCU callback, sync GC never fails.

Fixes: 3c4287f62044 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Fixes: 8d8540c4f5e0 ("netfilter: nft_set_rbtree: add timeout support")
Fixes: 9d0982927e79 ("netfilter: nft_hash: add support for timeouts")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

1 files changed, 48 insertions, 29 deletions

diff --git a/net/netfilter/nft_set_hash.c b/net/netfilter/nft_set_hash.c
index 24caa31fa231..2f067e4596b0 100644
--- a/net/netfilter/nft_set_hash.c
+++ b/net/netfilter/nft_set_hash.c
@@ -59,6 +59,8 @@ static inline int nft_rhash_cmp(struct rhashtable_compare_arg *arg,
 
 	if (memcmp(nft_set_ext_key(&he->ext), x->key, x->set->klen))
 		return 1;
+	if (nft_set_elem_is_dead(&he->ext))
+		return 1;
 	if (nft_set_elem_expired(&he->ext))
 		return 1;
 	if (!nft_set_elem_active(&he->ext, x->genmask))
@@ -188,7 +190,6 @@ static void nft_rhash_activate(const struct net *net, const struct nft_set *set,
 	struct nft_rhash_elem *he = elem->priv;
 
 	nft_set_elem_change_active(net, set, &he->ext);
-	nft_set_elem_clear_busy(&he->ext);
 }
 
 static bool nft_rhash_flush(const struct net *net,
@@ -196,12 +197,9 @@ static bool nft_rhash_flush(const struct net *net,
 {
 	struct nft_rhash_elem *he = priv;
 
-	if (!nft_set_elem_mark_busy(&he->ext) ||
-	    !nft_is_active(net, &he->ext)) {
-		nft_set_elem_change_active(net, set, &he->ext);
-		return true;
-	}
-	return false;
+	nft_set_elem_change_active(net, set, &he->ext);
+
+	return true;
 }
 
 static void *nft_rhash_deactivate(const struct net *net,
@@ -218,9 +216,8 @@ static void *nft_rhash_deactivate(const struct net *net,
 
 	rcu_read_lock();
 	he = rhashtable_lookup(&priv->ht, &arg, nft_rhash_params);
-	if (he != NULL &&
-	    !nft_rhash_flush(net, set, he))
-		he = NULL;
+	if (he)
+		nft_set_elem_change_active(net, set, &he->ext);
 
 	rcu_read_unlock();
 
@@ -312,25 +309,48 @@ static bool nft_rhash_expr_needs_gc_run(const struct nft_set *set,
 
 static void nft_rhash_gc(struct work_struct *work)
 {
+	struct nftables_pernet *nft_net;
 	struct nft_set *set;
 	struct nft_rhash_elem *he;
 	struct nft_rhash *priv;
-	struct nft_set_gc_batch *gcb = NULL;
 	struct rhashtable_iter hti;
+	struct nft_trans_gc *gc;
+	struct net *net;
+	u32 gc_seq;
 
 	priv = container_of(work, struct nft_rhash, gc_work.work);
 	set  = nft_set_container_of(priv);
+	net  = read_pnet(&set->net);
+	nft_net = nft_pernet(net);
+	gc_seq = READ_ONCE(nft_net->gc_seq);
+
+	gc = nft_trans_gc_alloc(set, gc_seq, GFP_KERNEL);
+	if (!gc)
+		goto done;
 
 	rhashtable_walk_enter(&priv->ht, &hti);
 	rhashtable_walk_start(&hti);
 
 	while ((he = rhashtable_walk_next(&hti))) {
 		if (IS_ERR(he)) {
-			if (PTR_ERR(he) != -EAGAIN)
-				break;
+			if (PTR_ERR(he) != -EAGAIN) {
+				nft_trans_gc_destroy(gc);
+				gc = NULL;
+				goto try_later;
+			}
 			continue;
 		}
 
+		/* Ruleset has been updated, try later. */
+		if (READ_ONCE(nft_net->gc_seq) != gc_seq) {
+			nft_trans_gc_destroy(gc);
+			gc = NULL;
+			goto try_later;
+		}
+
+		if (nft_set_elem_is_dead(&he->ext))
+			goto dead_elem;
+
 		if (nft_set_ext_exists(&he->ext, NFT_SET_EXT_EXPRESSIONS) &&
 		    nft_rhash_expr_needs_gc_run(set, &he->ext))
 			goto needs_gc_run;
@@ -338,26 +358,26 @@ static void nft_rhash_gc(struct work_struct *work)
 		if (!nft_set_elem_expired(&he->ext))
 			continue;
 needs_gc_run:
-		if (nft_set_elem_mark_busy(&he->ext))
-			continue;
+		nft_set_elem_dead(&he->ext);
+dead_elem:
+		gc = nft_trans_gc_queue_async(gc, gc_seq, GFP_ATOMIC);
+		if (!gc)
+			goto try_later;
 
-		gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
-		if (gcb == NULL)
-			break;
-		rhashtable_remove_fast(&priv->ht, &he->node, nft_rhash_params);
-		atomic_dec(&set->nelems);
-		nft_set_gc_batch_add(gcb, he);
+		nft_trans_gc_elem_add(gc, he);
 	}
+
+	gc = nft_trans_gc_catchall(gc, gc_seq);
+
+try_later:
+	/* catchall list iteration requires rcu read side lock. */
 	rhashtable_walk_stop(&hti);
 	rhashtable_walk_exit(&hti);
 
-	he = nft_set_catchall_gc(set);
-	if (he) {
-		gcb = nft_set_gc_batch_check(set, gcb, GFP_ATOMIC);
-		if (gcb)
-			nft_set_gc_batch_add(gcb, he);
-	}
-	nft_set_gc_batch_complete(gcb);
+	if (gc)
+		nft_trans_gc_queue_async_done(gc);
+
+done:
 	queue_delayed_work(system_power_efficient_wq, &priv->gc_work,
 			   nft_set_gc_interval(set));
 }
@@ -420,7 +440,6 @@ static void nft_rhash_destroy(const struct nft_ctx *ctx,
 	};
 
 	cancel_delayed_work_sync(&priv->gc_work);
-	rcu_barrier();
 	rhashtable_free_and_destroy(&priv->ht, nft_rhash_elem_destroy,
 				    (void *)&rhash_ctx);
 }