On 3/24/23 4:46 AM, Martin KaFai Lau wrote:
On 3/9/23 3:49 AM, D. Wythe wrote:
--- /dev/null
+++ b/net/smc/bpf_smc_struct_ops.c
@@ -0,0 +1,146 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/kernel.h>
+#include <linux/bpf_verifier.h>
+#include <linux/btf_ids.h>
+#include <linux/bpf.h>
+#include <linux/btf.h>
+#include <net/sock.h>
+#include <net/smc.h>
+
+extern struct bpf_struct_ops smc_sock_negotiator_ops;
+
+DEFINE_RWLOCK(smc_sock_negotiator_ops_rwlock);
+struct smc_sock_negotiator_ops *negotiator;
Is it sure one global negotiator (policy) will work for all
smc_sock? or each sk should have its own negotiator and the
negotiator is selected by setsockopt.
This is really a good question, we can really consider adding an
independent negotiator for each sock.
But just like the TCP congestion control , the global negotiator can
be used for sock without
special requirements.
It is different from TCP congestion control (CC). TCP CC has a global
default but each sk can select what tcp-cc to use and there can be
multiple tcp-cc registered under different names.
It sounds like smc using tcp_sock should be able to have different
negotiator also (eg. based on dst IP or some other tcp connection
characteristic). The tcp-cc registration, per-sock selection and the
rcu_read_lock+refcnt are well understood and there are other bpf
infrastructure to support the per sock tcp-cc selection (like
bpf_setsockopt).
For the network stack, there is little reason other af_* should not
follow at the beginning considering the infrastructure has already
been built. The one single global negotiator and reader/writer lock in
this patch reads like an effort wanted to give it a try and see if it
will be useful before implementing the whole thing. It is better to
keep it off the tree for now until it is more ready.
Hi Martin,
Thank you very much for your comments. I have indeed removed global
negotiator from my latest implementation.
The latest design is that users can register a negotiator implementation
indexed by name, smc_sock can use bpf_setsockopt to specify
whether a specific negotiation implementation is required via name. If
there are no settings, there will be no negotiators.
What do you think?
In addition, I am very sorry that I have not issued my implementation
for such a long time, and I have encountered some problems with the
implementation because
the SMC needs to be built as kernel module, I have struggled with the
bpf_setsockopt implementation, and there are some new self-testes that
need to be written.
However, I believe that I can send a new version as soon as possible.
Best wishes
D. Wythe
