pf: remove pd_refs from pfsync

[kp]

Summary

It only served to complicate cleanup, and added no value.

While here drop packets in pfsync_defer_tmo() if we don't have a syncif,
rather than just leaving them on the queue.

Sponsored by: Rubicon Communications, LLC (Netgate)

[zlei]

sys/netpfil/pf/if_pfsync.c

414

On clone destroy event, is it possible b is hold locked by other thread ?
Do we need lock for sync, or at least a memory fence ?

416

Do we need lock for sync, or at least a memory fence ?

So that TAILQ_FIRST will be safe.

1832

I think PF_STATE_LOCK() is needed.

[kp]

sys/netpfil/pf/if_pfsync.c

416

Good catch. Mark also pointed that out in D39223. I'll fix it there.

1832

There are a number of locking issues around state accesses in pfsync. I don't intend to fix them in this commit.

[kp]

• rebase