[PATCH mptcp-next v11 0/3] BPF round-robin scheduler

[PATCH mptcp-next v11 0/3] BPF round-robin scheduler

[Geliang Tang]

v11:
 - rename array to contexts.
 - drop number of subflows in mptcp_sched_data.
 - set unused array elements to NULL.
 - add MPTCP_SUBFLOWS_MAX check in mptcp_sched_data_init.

v10:
 - init subflows array in mptcp_sched_data_init.
 - for (int i = 0; i < data->subflows; i++) is not allowed in BPF, using
   this instead:
	for (int i = 0; i < MPTCP_SUBFLOWS_MAX && i < data->subflows; i++)
 - deponds on: "BPF packet scheduler" series v18.

v9:
 - add subflows array in mptcp_sched_data
 - deponds on: "BPF packet scheduler" series v17 +
               Squash to "mptcp: add struct mptcp_sched_ops v17".

v8:
 - use struct mptcp_sched_data.
 - deponds on: "BPF packet scheduler" series v14.

v7:
 - rename retrans to reinject.
 - drop last_snd setting.
 - deponds on: "BPF packet scheduler" series v13.

v6:
 - set call_me_again flag.
 - deponds on: "BPF packet scheduler" series v12.

v5:
 - update patch 2, use temporary storage instead.
 - update patch 3, use new helpers.
 - deponds on: "BPF packet scheduler" series v11.

v4:
 - add retrans argment for get_subflow()

v3:
 - add last_snd write access.
 - keep msk->last_snd setting in get_subflow().
 - deponds on: "BPF packet scheduler" series v10.

v2:
 - merge the squash-to patch.
 - implement bpf_mptcp_get_subflows helper, instead of
   bpf_mptcp_get_next_subflow.
 - deponds on: "BPF packet scheduler v9".

This patchset implements round-robin scheduler using BPF. Address to
some commends for the RFC version:

https://patchwork.kernel.org/project/mptcp/cover/cover.1631011068.git.geliangtang@xiaomi.com/

Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/75

Geliang Tang (3):
  mptcp: add subflows array in sched data
  selftests: bpf: add bpf_rr scheduler
  selftests: bpf: add bpf_rr test

 include/net/mptcp.h                           |  2 +
 net/mptcp/sched.c                             | 14 ++++++
 .../testing/selftests/bpf/bpf_mptcp_helpers.h |  8 ++++
 .../testing/selftests/bpf/prog_tests/mptcp.c  | 37 ++++++++++++++
 .../selftests/bpf/progs/mptcp_bpf_rr.c        | 48 +++++++++++++++++++
 5 files changed, 109 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c

-- 
2.34.1

[PATCH mptcp-next v11 1/3] mptcp: add subflows array in sched data

[Geliang Tang]

This patch adds a subflow pointers array in struct mptcp_sched_data. Set
the array before invoking get_subflow(), then get it in get_subflow() in
the BPF contexts.

Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 include/net/mptcp.h                             |  2 ++
 net/mptcp/sched.c                               | 14 ++++++++++++++
 tools/testing/selftests/bpf/bpf_mptcp_helpers.h |  2 ++
 3 files changed, 18 insertions(+)

diff --git a/include/net/mptcp.h b/include/net/mptcp.h
index b596ba7a8494..345f27a68eaa 100644
--- a/include/net/mptcp.h
+++ b/include/net/mptcp.h
@@ -96,10 +96,12 @@ struct mptcp_out_options {
 };
 
 #define MPTCP_SCHED_NAME_MAX	16
+#define MPTCP_SUBFLOWS_MAX	8
 
 struct mptcp_sched_data {
 	struct sock	*sock;
 	bool		call_again;
+	struct mptcp_subflow_context *contexts[MPTCP_SUBFLOWS_MAX];
 };
 
 struct mptcp_sched_ops {
diff --git a/net/mptcp/sched.c b/net/mptcp/sched.c
index 3ceb721e6489..9ee2d30a6f19 100644
--- a/net/mptcp/sched.c
+++ b/net/mptcp/sched.c
@@ -91,9 +91,23 @@ void mptcp_release_sched(struct mptcp_sock *msk)
 static int mptcp_sched_data_init(struct mptcp_sock *msk,
 				 struct mptcp_sched_data *data)
 {
+	struct mptcp_subflow_context *subflow;
+	int i = 0;
+
 	data->sock = NULL;
 	data->call_again = 0;
 
+	mptcp_for_each_subflow(msk, subflow) {
+		if (i == MPTCP_SUBFLOWS_MAX) {
+			pr_warn_once("too many subflows");
+			break;
+		}
+		data->contexts[i++] = subflow;
+	}
+
+	for (; i < MPTCP_SUBFLOWS_MAX; i++)
+		data->contexts[i++] = NULL;
+
 	return 0;
 }
 
diff --git a/tools/testing/selftests/bpf/bpf_mptcp_helpers.h b/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
index 4b7812c335fc..7ecd0b14666a 100644
--- a/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
+++ b/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
@@ -7,10 +7,12 @@
 #include "bpf_tcp_helpers.h"
 
 #define MPTCP_SCHED_NAME_MAX	16
+#define MPTCP_SUBFLOWS_MAX	8
 
 struct mptcp_sched_data {
 	struct sock	*sock;
 	bool		call_again;
+	struct mptcp_subflow_context *contexts[MPTCP_SUBFLOWS_MAX];
 };
 
 struct mptcp_sched_ops {
-- 
2.34.1

[PATCH mptcp-next v11 2/3] selftests: bpf: add bpf_rr scheduler

[Geliang Tang]

This patch implements the round-robin BPF MPTCP scheduler, named bpf_rr,
which always picks the next available subflow to send data. If no such
next subflow available, picks the first one.

Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 .../testing/selftests/bpf/bpf_mptcp_helpers.h |  6 +++
 .../selftests/bpf/progs/mptcp_bpf_rr.c        | 48 +++++++++++++++++++
 2 files changed, 54 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c

diff --git a/tools/testing/selftests/bpf/bpf_mptcp_helpers.h b/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
index 7ecd0b14666a..2d5109f459b4 100644
--- a/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
+++ b/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
@@ -29,10 +29,16 @@ struct mptcp_sched_ops {
 struct mptcp_sock {
 	struct inet_connection_sock	sk;
 
+	struct sock	*last_snd;
 	__u32		token;
 	struct sock	*first;
 	struct mptcp_sched_ops *sched;
 	char		ca_name[TCP_CA_NAME_MAX];
 } __attribute__((preserve_access_index));
 
+struct mptcp_subflow_context {
+	__u32	token;
+	struct	sock *tcp_sock;	    /* tcp sk backpointer */
+} __attribute__((preserve_access_index));
+
 #endif
diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c b/tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c
new file mode 100644
index 000000000000..ab0bb2a38891
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c
@@ -0,0 +1,48 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2022, SUSE. */
+
+#include <linux/bpf.h>
+#include <linux/stddef.h>
+#include <linux/tcp.h>
+#include "bpf_mptcp_helpers.h"
+
+char _license[] SEC("license") = "GPL";
+
+SEC("struct_ops/mptcp_sched_rr_init")
+void BPF_PROG(mptcp_sched_rr_init, const struct mptcp_sock *msk)
+{
+}
+
+SEC("struct_ops/mptcp_sched_rr_release")
+void BPF_PROG(mptcp_sched_rr_release, const struct mptcp_sock *msk)
+{
+}
+
+void BPF_STRUCT_OPS(bpf_rr_get_subflow, const struct mptcp_sock *msk,
+		    bool reinject, struct mptcp_sched_data *data)
+{
+	struct mptcp_subflow_context *subflow;
+	struct sock *ssk = msk->first;
+
+	for (int i = 0; i < MPTCP_SUBFLOWS_MAX; i++) {
+		subflow = data->contexts[i];
+		if (!subflow)
+			break;
+
+		if (subflow->tcp_sock != msk->last_snd) {
+			ssk = subflow->tcp_sock;
+			break;
+		}
+	}
+
+	data->sock = ssk;
+	data->call_again = 0;
+}
+
+SEC(".struct_ops")
+struct mptcp_sched_ops rr = {
+	.init		= (void *)mptcp_sched_rr_init,
+	.release	= (void *)mptcp_sched_rr_release,
+	.get_subflow	= (void *)bpf_rr_get_subflow,
+	.name		= "bpf_rr",
+};
-- 
2.34.1

Re: [PATCH mptcp-next v11 2/3] selftests: bpf: add bpf_rr scheduler

[Mat Martineau]

On Tue, 3 May 2022, Geliang Tang wrote:

> This patch implements the round-robin BPF MPTCP scheduler, named bpf_rr,
> which always picks the next available subflow to send data. If no such
> next subflow available, picks the first one.
>
> Signed-off-by: Geliang Tang <geliang.tang@suse.com>
> ---
> .../testing/selftests/bpf/bpf_mptcp_helpers.h |  6 +++
> .../selftests/bpf/progs/mptcp_bpf_rr.c        | 48 +++++++++++++++++++
> 2 files changed, 54 insertions(+)
> create mode 100644 tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c
>
> diff --git a/tools/testing/selftests/bpf/bpf_mptcp_helpers.h b/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
> index 7ecd0b14666a..2d5109f459b4 100644
> --- a/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
> +++ b/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
> @@ -29,10 +29,16 @@ struct mptcp_sched_ops {
> struct mptcp_sock {
> 	struct inet_connection_sock	sk;
>
> +	struct sock	*last_snd;
> 	__u32		token;
> 	struct sock	*first;
> 	struct mptcp_sched_ops *sched;
> 	char		ca_name[TCP_CA_NAME_MAX];
> } __attribute__((preserve_access_index));
>
> +struct mptcp_subflow_context {
> +	__u32	token;
> +	struct	sock *tcp_sock;	    /* tcp sk backpointer */
> +} __attribute__((preserve_access_index));
> +
> #endif
> diff --git a/tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c b/tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c
> new file mode 100644
> index 000000000000..ab0bb2a38891
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c
> @@ -0,0 +1,48 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/* Copyright (c) 2022, SUSE. */
> +
> +#include <linux/bpf.h>
> +#include <linux/stddef.h>
> +#include <linux/tcp.h>
> +#include "bpf_mptcp_helpers.h"
> +
> +char _license[] SEC("license") = "GPL";
> +
> +SEC("struct_ops/mptcp_sched_rr_init")
> +void BPF_PROG(mptcp_sched_rr_init, const struct mptcp_sock *msk)
> +{
> +}
> +
> +SEC("struct_ops/mptcp_sched_rr_release")
> +void BPF_PROG(mptcp_sched_rr_release, const struct mptcp_sock *msk)
> +{
> +}
> +
> +void BPF_STRUCT_OPS(bpf_rr_get_subflow, const struct mptcp_sock *msk,
> +		    bool reinject, struct mptcp_sched_data *data)
> +{
> +	struct mptcp_subflow_context *subflow;
> +	struct sock *ssk = msk->first;

Hi Geliang -

I suggest initializing this from data->contexts[0] to make it more obvious 
what is happening in the single-subflow case (or if last_ssk isn't found).

Now that we pass in contexts[] every time, is access to msk->first still 
needed?

> +
> +	for (int i = 0; i < MPTCP_SUBFLOWS_MAX; i++) {
> +		subflow = data->contexts[i];
> +		if (!subflow)
> +			break;
> +
> +		if (subflow->tcp_sock != msk->last_snd) {
> +			ssk = subflow->tcp_sock;
> +			break;

This will only alternate between the first two subflows in the list.

last_ssk could be any one of the values in contexts[], and round-robin 
should cycle through all the subflows (including wrapping around).

- Mat

> +		}
> +	}
> +
> +	data->sock = ssk;
> +	data->call_again = 0;
> +}
> +
> +SEC(".struct_ops")
> +struct mptcp_sched_ops rr = {
> +	.init		= (void *)mptcp_sched_rr_init,
> +	.release	= (void *)mptcp_sched_rr_release,
> +	.get_subflow	= (void *)bpf_rr_get_subflow,
> +	.name		= "bpf_rr",
> +};
> -- 
> 2.34.1
>
>
>

--
Mat Martineau
Intel

[PATCH mptcp-next v11 3/3] selftests: bpf: add bpf_rr test

[Geliang Tang]

This patch adds the round-robin BPF MPTCP scheduler test. Use sysctl to
set net.mptcp.scheduler to use this sched. Add a veth net device to
simulate the multiple addresses case. Use 'ip mptcp endpoint' command to
add this new endpoint to PM netlink.

Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 .../testing/selftests/bpf/prog_tests/mptcp.c  | 37 +++++++++++++++++++
 1 file changed, 37 insertions(+)

diff --git a/tools/testing/selftests/bpf/prog_tests/mptcp.c b/tools/testing/selftests/bpf/prog_tests/mptcp.c
index 21462c29683c..c400be612f72 100644
--- a/tools/testing/selftests/bpf/prog_tests/mptcp.c
+++ b/tools/testing/selftests/bpf/prog_tests/mptcp.c
@@ -5,6 +5,7 @@
 #include "cgroup_helpers.h"
 #include "network_helpers.h"
 #include "mptcp_bpf_first.skel.h"
+#include "mptcp_bpf_rr.skel.h"
 
 #ifndef TCP_CA_NAME_MAX
 #define TCP_CA_NAME_MAX	16
@@ -375,10 +376,46 @@ static void test_first(void)
 	mptcp_bpf_first__destroy(first_skel);
 }
 
+static void test_rr(void)
+{
+	struct mptcp_bpf_rr *rr_skel;
+	int server_fd, client_fd;
+	struct bpf_link *link;
+
+	rr_skel = mptcp_bpf_rr__open_and_load();
+	if (CHECK(!rr_skel, "bpf_rr__open_and_load", "failed\n"))
+		return;
+
+	link = bpf_map__attach_struct_ops(rr_skel->maps.rr);
+	if (!ASSERT_OK_PTR(link, "bpf_map__attach_struct_ops")) {
+		mptcp_bpf_rr__destroy(rr_skel);
+		return;
+	}
+
+	system("ip link add veth1 type veth; \
+		ip addr add 10.0.1.1/24 dev veth1; \
+		ip link set veth1 up");
+	system("ip mptcp endpoint add 10.0.1.1 subflow");
+	system("sysctl -q net.mptcp.scheduler=bpf_rr");
+	server_fd = start_mptcp_server(AF_INET, NULL, 0, 0);
+	client_fd = connect_to_mptcp_fd(server_fd, 0);
+
+	send_data(server_fd, client_fd);
+
+	close(client_fd);
+	close(server_fd);
+	system("ip mptcp endpoint flush");
+	system("ip link del veth1");
+	bpf_link__destroy(link);
+	mptcp_bpf_rr__destroy(rr_skel);
+}
+
 void test_mptcp(void)
 {
 	if (test__start_subtest("base"))
 		test_base();
 	if (test__start_subtest("first"))
 		test_first();
+	if (test__start_subtest("rr"))
+		test_rr();
 }
-- 
2.34.1