[PATCH v1 1/3] bpf: hash: use atomic count

[Ming Lei]

Preparing for removing global per-hashtable lock, so
the counter need to be defined as aotmic_t first.

Acked-by: Daniel Borkmann <dan...@iogearbox.net>
Signed-off-by: Ming Lei <tom.leim...@gmail.com>
---
 kernel/bpf/hashtab.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c
index 34777b3..2615388 100644
--- a/kernel/bpf/hashtab.c
+++ b/kernel/bpf/hashtab.c
@@ -18,7 +18,7 @@ struct bpf_htab {
        struct bpf_map map;
        struct hlist_head *buckets;
        raw_spinlock_t lock;
-       u32 count;      /* number of elements in this hashtable */
+       atomic_t count; /* number of elements in this hashtable */
        u32 n_buckets;  /* number of hash buckets */
        u32 elem_size;  /* size of each element in bytes */
 };
@@ -106,7 +106,7 @@ static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
                INIT_HLIST_HEAD(&htab->buckets[i]);
 
        raw_spin_lock_init(&htab->lock);
-       htab->count = 0;
+       atomic_set(&htab->count, 0);
 
        return &htab->map;
 
@@ -256,7 +256,7 @@ static int htab_map_update_elem(struct bpf_map *map, void 
*key, void *value,
 
        l_old = lookup_elem_raw(head, l_new->hash, key, key_size);
 
-       if (!l_old && unlikely(htab->count >= map->max_entries)) {
+       if (!l_old && unlikely(atomic_read(&htab->count) >= map->max_entries)) {
                /* if elem with this 'key' doesn't exist and we've reached
                 * max_entries limit, fail insertion of new elem
                 */
@@ -284,7 +284,7 @@ static int htab_map_update_elem(struct bpf_map *map, void 
*key, void *value,
                hlist_del_rcu(&l_old->hash_node);
                kfree_rcu(l_old, rcu);
        } else {
-               htab->count++;
+               atomic_inc(&htab->count);
        }
        raw_spin_unlock_irqrestore(&htab->lock, flags);
 
@@ -319,7 +319,7 @@ static int htab_map_delete_elem(struct bpf_map *map, void 
*key)
 
        if (l) {
                hlist_del_rcu(&l->hash_node);
-               htab->count--;
+               atomic_dec(&htab->count);
                kfree_rcu(l, rcu);
                ret = 0;
        }
@@ -339,7 +339,7 @@ static void delete_all_elements(struct bpf_htab *htab)
 
                hlist_for_each_entry_safe(l, n, head, hash_node) {
                        hlist_del_rcu(&l->hash_node);
-                       htab->count--;
+                       atomic_dec(&htab->count);
                        kfree(l);
                }
        }
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html