net_sched: act_bpf: remove spinlock in fast path

Similar to act_gact/act_mirred, act_bpf can be lockless in packet processing
with extra care taken to free bpf programs after rcu grace period.
Replacement of existing act_bpf (very rare) is done with synchronize_rcu()
and final destruction is done from tc_action_ops->cleanup() callback that is
called from tcf_exts_destroy()->tcf_action_destroy()->__tcf_hash_release() when
bind and refcnt reach zero which is only possible when classifier is destroyed.
Previous two patches fixed the last two classifiers (tcindex and rsvp) to
call tcf_exts_destroy() from rcu callback.

Similar to gact/mirred there is a race between prog->filter and
prog->tcf_action. Meaning that the program being replaced may use
previous default action if it happened to return TC_ACT_UNSPEC.
act_mirred race betwen tcf_action and tcfm_dev is similar.
In all cases the race is harmless.
Long term we may want to improve the situation by replacing the whole
tc_action->priv as single pointer instead of updating inner fields one by one.

Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/tc_act/tc_bpf.h

@@ -15,7 +15,7 @@
 
 struct tcf_bpf {
 	struct tcf_common	common;
-	struct bpf_prog		*filter;
+	struct bpf_prog __rcu	*filter;
 	union {
 		u32		bpf_fd;
 		u16		bpf_num_ops;