Fix soft lockup caused by deadlock on htable access

This appeared on kernels where xtables target is called though nftables.

Lockdep message: WARNING: inconsistent lock state

  [  292.286849] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
  [  292.287776] kworker/0:0/5 [HC0[0]:SC1[1]:HE1:SE0] takes:
  [  292.288618] ffffffffc0de6a98 (&(&htable_stripes[i].lock)->rlock){+.?.}, at: netflow_target+0x604/0x4150 [ipt_NETFLOW]
  [  292.290323] {SOFTIRQ-ON-W} state was registered at:
  [  292.291095]   lock_acquire+0x14f/0x3b0
  [  292.291679]   _raw_spin_lock+0x30/0x70
  [  292.292276]   netflow_target+0x604/0x4150 [ipt_NETFLOW]
  [  292.293125]   nft_target_eval_xt+0x11a/0x220 [nft_compat]
  [  292.294027]   nft_do_chain+0x25a/0x10e0 [nf_tables]
  [  292.294813]   nft_do_chain_ipv4+0x17e/0x200 [nf_tables]
  [  292.295622]   nf_hook_slow+0xb1/0x180

Non-debug kernel error message:

  watchdog: BUG: soft lockup - CPU#6 stuck for 22s!

Fixes #141 and #145.

Author: aabc

ipt_NETFLOW.c

@@ -5238,7 +5238,7 @@ static unsigned int netflow_target(
 #else /* !SAMPLING_HASH */
 	hash = hash_netflow(&tuple);
 #endif
-	read_lock(&htable_rwlock);
+	read_lock_bh(&htable_rwlock);
 	stripe = &htable_stripes[hash & LOCK_COUNT_MASK];
 	spin_lock(&stripe->lock);
 	/* record */
@@ -5391,7 +5391,7 @@ static unsigned int netflow_target(
 
 unlock_return:
 	spin_unlock(&stripe->lock);
-	read_unlock(&htable_rwlock);
+	read_unlock_bh(&htable_rwlock);
 
 	return IPT_CONTINUE;
 }