[PATCH nf-next v7 6/6] netfilter: ipset: re-add forceadd support

[Florian Westphal <fw@strlen.de>]

The rhashtable conversion removed the SET_WITH_FORCEADD eviction logic.

Add mtype_remove_random() helper to look up a random key slot: a group
for IP_SET_HASH_WITH_MULTI sets, or a plain element otherwise. If there
is one, try to evict it and allow add of the new element just like
before the rhashtable conversion.

Signed-off-by: Florian Westphal <fw@strlen.de>
---
 v7: use u32 k[] in mtype_remove_random (sashiko)

 net/netfilter/ipset/ip_set_hash_gen.h | 92 ++++++++++++++++++++++++++-
 1 file changed, 91 insertions(+), 1 deletion(-)

diff --git a/net/netfilter/ipset/ip_set_hash_gen.h b/net/netfilter/ipset/ip_set_hash_gen.h
index e3d0f83e6d9d..7c74fcc73725 100644
--- a/net/netfilter/ipset/ip_set_hash_gen.h
+++ b/net/netfilter/ipset/ip_set_hash_gen.h
@@ -165,6 +165,8 @@ static const union nf_inet_addr zeromask = {};
 #undef mtype_kadt
 #undef mtype_uadt
 
+#undef mtype_remove_random
+#undef mtype_remove_cmpfn
 #undef mtype_new_elem
 #undef mtype_add_replace_old
 #undef mtype_add
@@ -234,6 +236,8 @@ static const union nf_inet_addr zeromask = {};
 #define mtype_kadt		IPSET_TOKEN(MTYPE, _kadt)
 #define mtype_uadt		IPSET_TOKEN(MTYPE, _uadt)
 
+#define mtype_remove_random	IPSET_TOKEN(MTYPE, _remove_random)
+#define mtype_remove_cmpfn	IPSET_TOKEN(MTYPE, _remove_cmpfn)
 #define mtype_new_elem		IPSET_TOKEN(MTYPE, _new_elem)
 #define mtype_add_replace_old	IPSET_TOKEN(MTYPE, _add_replace_old)
 #define mtype_add		IPSET_TOKEN(MTYPE, _add)
@@ -974,6 +978,92 @@ mtype_rht_size(struct ip_set *set, u32 *elements, size_t *ext_size)
 #endif
 }
 
+static int mtype_remove_cmpfn(struct rhashtable_compare_arg *arg, const void *obj)
+{
+	return 0; /* always match */
+}
+
+/**
+ * mtype_remove_random() - Remove a random element from the set (forceadd)
+ * @set: Pointer to the ip_set
+ *
+ * When sets created with forceadd option become full the next addition
+ * to the set may succeed and evict a random entry from the set.
+ * Best-effort: no linear scan; 'return false' is fine.
+ *
+ * Return: true if an element was evicted, false otherwise.
+ */
+static bool
+mtype_remove_random(struct ip_set *set)
+{
+	u32 k[HKEY_DATALEN / sizeof(u32)];
+	struct htype *h = set->data;
+
+	/* make sure k[] doesn't eat too much stack space */
+	BUILD_BUG_ON(sizeof(k) > 64);
+
+	/* The real hashfn is needed instead of a "get_random_u32" stub,
+	 * because rhashtable_lookup() might fail to constant-fold the
+	 * params below, in which case it would fall back to h->params
+	 * hashfn.
+	 */
+	get_random_bytes(k, sizeof(k));
+
+#ifdef IP_SET_HASH_WITH_MULTI
+	{
+		static const struct rhashtable_params ip_set_hash_grp_rnd_params = {
+			.head_offset	= offsetof(struct mtype_group, node),
+			.key_offset	= offsetof(struct mtype_group, key),
+			.hashfn		= mtype_rht_hashfn,
+			.obj_hashfn	= mtype_group_obj_hashfn,
+			.obj_cmpfn	= mtype_remove_cmpfn,
+			.key_len	= HKEY_DATALEN,
+		};
+		struct mtype_group *grp;
+		struct mtype_rht_elem *e;
+
+		grp = rhashtable_lookup(&h->ht, k, ip_set_hash_grp_rnd_params);
+		if (!grp)
+			return false;
+
+		/* mtype_add() can call us while already holding this exact
+		 * group's lock (maxelem reached while inserting into it);
+		 * trylock to avoid self-deadlock. Best-effort: no linear
+		 * scan, 'return false' is fine either way.
+		 */
+		if (!spin_trylock_bh(&grp->lock))
+			return false;
+		e = hlist_entry_safe(grp->iface_list.first,
+				     struct mtype_rht_elem, node);
+		if (!e) {
+			spin_unlock_bh(&grp->lock);
+			return false;
+		}
+		mtype_group_remove_elem(set, h, grp, e);
+		return true;
+	}
+#else
+	{
+		static const struct rhashtable_params ip_set_hash_rnd_params = {
+			.head_offset	= offsetof(struct mtype_rht_elem, node),
+			.key_offset	= offsetof(struct mtype_rht_elem, elem),
+			.hashfn		= mtype_rht_hashfn,
+			.obj_hashfn	= mtype_rht_obj_hashfn,
+			.obj_cmpfn	= mtype_remove_cmpfn,
+			.key_len	= HKEY_DATALEN,
+		};
+		struct mtype_rht_elem *e;
+
+		e = rhashtable_lookup(&h->ht, &k, ip_set_hash_rnd_params);
+		if (!e || rhashtable_remove_fast(&h->ht, &e->node, mtype_rht_params))
+			return false;
+
+		ipset_hash_elem_destroy_free(set, e);
+		return true;
+	}
+#endif
+}
+
 /**
  * mtype_new_elem() - allocate and initialize a replacement/new element
  * @set: Pointer to the ip_set
@@ -1017,7 +1107,7 @@ mtype_new_elem(struct ip_set *set, struct htype *h, const struct mtype_elem *d,
 		 * its pretty much pointless, userspace can just retry once
 		 * gc has fired, resp. decrease the gc interval.
 		 */
-		if (1) {
+		if (!SET_WITH_FORCEADD(set) || !mtype_remove_random(set)) {
 			if (net_ratelimit())
 				pr_warn("Set %s is full, maxelem %u reached\n",
 					set->name, h->maxelem);
-- 
2.56.0