src: delete the old cache when dumping is interrupted

When the dumping operation is interrupted, we will restart the
cache_init(), but unfortunatly, we forget to delete the old cache.
So in extreme case, we will leak a huge amount of memory.

Running the following commands can simulate the extreme case:
  # nft add table t
  # nft add set t s {type inet_service \;}
  # for i in $(seq 65000); do
  nft add element t s {$i}
  done &
  # while : ; do
  time nft list ruleset -nn
  done

After a while, oom killer will be triggered:
  [ 2808.243537] Out of memory: Kill process 16975 (nft) score 649 or
                 sacrifice child
  [ 2808.255372] Killed process 16975 (nft) total-vm:1955348kB,
                 anon-rss:1952120kB, file-rss:0kB, shmem-rss:0kB
  [ 2858.353729] nft invoked oom-killer: gfp_mask=0x14201ca(GFP_HIGHUSER_
                 MOVABLE|__GFP_COLD), nodemask=(null),  order=0,
                 oom_score_adj=0
  [ 2858.374521] nft cpuset=/ mems_allowed=0
  ...

Signed-off-by: Liping Zhang <zlpnobody@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

diff --git a/src/rule.c b/src/rule.c
index 209cf2d7d9f6097368bbd822ce1094e23fa1693c..5923bf64101ecda48aefbc056b653d1d7ff0eba3 100644 (file)
--- a/src/rule.c
+++ b/src/rule.c
@@ -156,11 +156,11 @@ replay:
        netlink_genid_get();
        ret = cache_init(cmd, msgs);
        if (ret < 0) {
+               cache_release();
                if (errno == EINTR) {
                        netlink_restart();
                        goto replay;
                }
-               cache_release();
                return -1;
        }
        cache_initialized = true;