bpf, sockhash: Synchronize_rcu before free'ing map
authorJakub Sitnicki <jakub@cloudflare.com>
Thu, 6 Feb 2020 11:16:51 +0000 (12:16 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 14 Feb 2020 21:34:09 +0000 (16:34 -0500)
commit 0b2dc83906cf1e694e48003eae5df8fa63f76fd9 upstream.

We need to have a synchronize_rcu before free'ing the sockhash because any
outstanding psock references will have a pointer to the map and when they
use it, this could trigger a use after free.

This is a sister fix for sockhash, following commit 2bb90e5cc90e ("bpf:
sockmap, synchronize_rcu before free'ing map") which addressed sockmap,
which comes from a manual audit.

Fixes: 604326b41a6fb ("bpf, sockmap: convert to generic sk_msg interface")
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20200206111652.694507-3-jakub@cloudflare.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/core/sock_map.c

index fd8b426..f36e13e 100644 (file)
@@ -250,6 +250,7 @@ static void sock_map_free(struct bpf_map *map)
        }
        raw_spin_unlock_bh(&stab->lock);
 
+       /* wait for psock readers accessing its map link */
        synchronize_rcu();
 
        bpf_map_area_free(stab->sks);
@@ -873,6 +874,9 @@ static void sock_hash_free(struct bpf_map *map)
                raw_spin_unlock_bh(&bucket->lock);
        }
 
+       /* wait for psock readers accessing its map link */
+       synchronize_rcu();
+
        bpf_map_area_free(htab->buckets);
        kfree(htab);
 }