Re: [PATCH bpf-next v2 2/2] bpf/selftests: add selftest for SMC bpf capability

[Martin KaFai Lau <martin.lau@linux.dev>]

On 2/21/23 4:18 AM, D. Wythe wrote:
> From: "D. Wythe" <alibuda@linux.alibaba.com>
> 
> This PATCH adds a tiny selftest for SMC bpf capability,
> making decisions on whether to use SMC by collecting
> certain information from kernel smc sock.
> 
> Follow the steps below to run this test.
> 
> make -C tools/testing/selftests/bpf
> cd tools/testing/selftests/bpf
> sudo ./test_progs -t bpf_smc
> 
> Results shows:
> 18      bpf_smc:OK
> Summary: 1/0 PASSED, 0 SKIPPED, 0 FAILED
> 
> Signed-off-by: D. Wythe <alibuda@linux.alibaba.com>
> ---
>   tools/testing/selftests/bpf/prog_tests/bpf_smc.c |  39 +++
>   tools/testing/selftests/bpf/progs/bpf_smc.c      | 315 +++++++++++++++++++++++
>   2 files changed, 354 insertions(+)
>   create mode 100644 tools/testing/selftests/bpf/prog_tests/bpf_smc.c
>   create mode 100644 tools/testing/selftests/bpf/progs/bpf_smc.c
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/bpf_smc.c b/tools/testing/selftests/bpf/prog_tests/bpf_smc.c
> new file mode 100644
> index 0000000..b143932
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/prog_tests/bpf_smc.c
> @@ -0,0 +1,39 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/* Copyright (c) 2019 Facebook */

copy-and-paste left-over...

> diff --git a/tools/testing/selftests/bpf/progs/bpf_smc.c b/tools/testing/selftests/bpf/progs/bpf_smc.c
> new file mode 100644
> index 0000000..78c7976
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/progs/bpf_smc.c
> @@ -0,0 +1,315 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +
> +#include <linux/bpf.h>
> +#include <linux/stddef.h>
> +#include <linux/smc.h>
> +#include <stdbool.h>
> +#include <linux/types.h>
> +#include <bpf/bpf_helpers.h>
> +#include <bpf/bpf_core_read.h>
> +#include <bpf/bpf_tracing.h>
> +
> +#define BPF_STRUCT_OPS(name, args...) \
> +	SEC("struct_ops/"#name) \
> +	BPF_PROG(name, args)
> +
> +#define SMC_LISTEN		(10)
> +#define SMC_SOCK_CLOSED_TIMING	(0)
> +extern unsigned long CONFIG_HZ __kconfig;
> +#define HZ CONFIG_HZ
> +
> +char _license[] SEC("license") = "GPL";
> +#define max(a, b) ((a) > (b) ? (a) : (b))
> +
> +struct sock_common {
> +	unsigned char	skc_state;
> +	__u16	skc_num;
> +} __attribute__((preserve_access_index));
> +
> +struct sock {
> +	struct sock_common	__sk_common;
> +	int	sk_sndbuf;
> +} __attribute__((preserve_access_index));
> +
> +struct inet_sock {
> +	struct sock	sk;
> +} __attribute__((preserve_access_index));
> +
> +struct inet_connection_sock {
> +	struct inet_sock	icsk_inet;
> +} __attribute__((preserve_access_index));
> +
> +struct tcp_sock {
> +	struct inet_connection_sock	inet_conn;
> +	__u32	rcv_nxt;
> +	__u32	snd_nxt;
> +	__u32	snd_una;
> +	__u32	delivered;
> +	__u8	syn_data:1,	/* SYN includes data */
> +		syn_fastopen:1,	/* SYN includes Fast Open option */
> +		syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
> +		syn_fastopen_ch:1, /* Active TFO re-enabling probe */
> +		syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
> +		save_syn:1,	/* Save headers of SYN packet */
> +		is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */
> +		syn_smc:1;	/* SYN includes SMC */
> +} __attribute__((preserve_access_index));
> +
> +struct socket {
> +	struct sock *sk;
> +} __attribute__((preserve_access_index));

All these tcp_sock, socket, inet_sock definitions can go away if it includes 
"vmlinux.h". tcp_ca_write_sk_pacing.c is a better example to follow. Try to 
define the "common" (eg. tcp, tc...etc) missing macros in bpf_tracing_net.h. The 
smc specific macros can stay in this file.

> +static inline struct smc_prediction *smc_prediction_get(const struct smc_sock *smc,
> +							const struct tcp_sock *tp, __u64 tstamp)
> +{
> +	struct smc_prediction zero = {}, *smc_predictor;
> +	__u16 key;
> +	__u32 gap;
> +	int err;
> +
> +	err = bpf_core_read(&key, sizeof(__u16), &tp->inet_conn.icsk_inet.sk.__sk_common.skc_num);
> +	if (err)
> +		return NULL;
> +
> +	/* BAD key */
> +	if (key == 0)
> +		return NULL;
> +
> +	smc_predictor = bpf_map_lookup_elem(&negotiator_map, &key);
> +	if (!smc_predictor) {
> +		zero.start_tstamp = bpf_jiffies64();
> +		zero.pacing_delta = SMC_PREDICTION_MIN_PACING_DELTA;
> +		bpf_map_update_elem(&negotiator_map, &key, &zero, 0);
> +		smc_predictor =  bpf_map_lookup_elem(&negotiator_map, &key);
> +		if (!smc_predictor)
> +			return NULL;
> +	}
> +
> +	if (tstamp) {
> +		bpf_spin_lock(&smc_predictor->lock);
> +		gap = (tstamp - smc_predictor->start_tstamp) / smc_predictor->pacing_delta;
> +		/* new splice */
> +		if (gap > 0) {
> +			smc_predictor->start_tstamp = tstamp;
> +			smc_predictor->last_rate_of_lcc =
> +				(smc_prediction_calt_rate(smc_predictor) * 7) >> (2 + gap);
> +			smc_predictor->closed_long_cc = 0;
> +			smc_predictor->closed_total_cc = 0;
> +			smc_predictor->incoming_long_cc = 0;
> +		}
> +		bpf_spin_unlock(&smc_predictor->lock);
> +	}
> +	return smc_predictor;
> +}
> +
> +/* BPF struct ops for smc protocol negotiator */
> +struct smc_sock_negotiator_ops {
> +	/* ret for negotiate */
> +	int (*negotiate)(struct smc_sock *smc);
> +
> +	/* info gathering timing */
> +	void (*collect_info)(struct smc_sock *smc, int timing);
> +};
> +
> +int BPF_STRUCT_OPS(bpf_smc_negotiate, struct smc_sock *smc)
> +{
> +	struct smc_prediction *smc_predictor;
> +	struct tcp_sock *tp;
> +	struct sock *clcsk;
> +	int ret = SK_DROP;
> +	__u32 rate = 0;
> +
> +	/* Only make decison during listen */
> +	if (smc->sk.__sk_common.skc_state != SMC_LISTEN)
> +		return SK_PASS;
> +
> +	clcsk = BPF_CORE_READ(smc, clcsock, sk);

Instead of using bpf_core_read here, why not directly gets the clcsk like the 
'smc->sk.__sk_common.skc_state' above.

> +	if (!clcsk)
> +		goto error;
> +
> +	tp = tcp_sk(clcsk);

There is a bpf_skc_to_tcp_sock(). Give it a try after changing the above 
BPF_CORE_READ.

> +	if (!tp)
> +		goto error;
> +
> +	smc_predictor = smc_prediction_get(smc, tp, bpf_jiffies64());
> +	if (!smc_predictor)
> +		return SK_PASS;
> +
> +	bpf_spin_lock(&smc_predictor->lock);
> +
> +	if (smc_predictor->incoming_long_cc == 0)
> +		goto out_locked_pass;
> +
> +	if (smc_predictor->incoming_long_cc > SMC_PREDICTION_MAX_LONGCC_PER_SPLICE) {
> +		ret = 100;
> +		goto out_locked_drop;
> +	}
> +
> +	rate = smc_prediction_calt_rate(smc_predictor);
> +	if (rate < SMC_PREDICTION_LONGCC_RATE_THRESHOLD) {
> +		ret = 200;
> +		goto out_locked_drop;
> +	}
> +out_locked_pass:
> +	smc_predictor->incoming_long_cc++;
> +	bpf_spin_unlock(&smc_predictor->lock);
> +	return SK_PASS;
> +out_locked_drop:
> +	bpf_spin_unlock(&smc_predictor->lock);
> +error:
> +	return SK_DROP;
> +}
> +
> +void BPF_STRUCT_OPS(bpf_smc_collect_info, struct smc_sock *smc, int timing)

Try to stay with SEC("struct_ops/...") void BPF_PROG(....)