[PATCH mptcp-next v8 0/3] BPF round-robin scheduler

[PATCH mptcp-next v8 0/3] BPF round-robin scheduler

[Geliang Tang]

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 bpf get_subflows helper
  selftests: bpf: add bpf_rr scheduler
  selftests: bpf: add bpf_rr test

 net/mptcp/bpf.c                               | 47 ++++++++++++++++
 net/mptcp/protocol.h                          |  7 +++
 .../testing/selftests/bpf/bpf_mptcp_helpers.h | 13 +++++
 .../testing/selftests/bpf/prog_tests/mptcp.c  | 37 +++++++++++++
 .../selftests/bpf/progs/mptcp_bpf_rr.c        | 55 +++++++++++++++++++
 5 files changed, 159 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c

-- 
2.34.1

[PATCH mptcp-next v8 1/3] mptcp: add bpf get_subflows helper

[Geliang Tang]

This patch implements a new helper bpf_mptcp_get_subflows() to get all the
subflows of the given mptcp_sock, it returns the number of suflows. Add
a new member subflows in struct mptcp_sock as a pointers array of all the
subflows.

Register this helper in bpf_mptcp_sched_kfunc_init() to make sure it can
be accessed from the BPF context.

Signed-off-by: Geliang Tang <geliang.tang@suse.com>
---
 net/mptcp/bpf.c      | 47 ++++++++++++++++++++++++++++++++++++++++++++
 net/mptcp/protocol.h |  7 +++++++
 2 files changed, 54 insertions(+)

diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c
index 6c01f6b959a3..3367541b353c 100644
--- a/net/mptcp/bpf.c
+++ b/net/mptcp/bpf.c
@@ -160,6 +160,23 @@ struct bpf_struct_ops bpf_mptcp_sched_ops = {
 	.name		= "mptcp_sched_ops",
 };
 
+BTF_SET_START(bpf_mptcp_sched_kfunc_ids)
+BTF_ID(func, bpf_mptcp_get_subflows_array)
+BTF_ID(func, bpf_mptcp_put_subflows_array)
+BTF_SET_END(bpf_mptcp_sched_kfunc_ids)
+
+static const struct btf_kfunc_id_set bpf_mptcp_sched_kfunc_set = {
+	.owner		= THIS_MODULE,
+	.check_set	= &bpf_mptcp_sched_kfunc_ids,
+};
+
+static int __init bpf_mptcp_sched_kfunc_init(void)
+{
+	return register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS,
+					 &bpf_mptcp_sched_kfunc_set);
+}
+late_initcall(bpf_mptcp_sched_kfunc_init);
+
 struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk)
 {
 	if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk))
@@ -168,3 +185,33 @@ struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk)
 	return NULL;
 }
 EXPORT_SYMBOL(bpf_mptcp_sock_from_subflow);
+
+struct mptcp_subflows_array *bpf_mptcp_get_subflows_array(struct mptcp_sock *msk)
+{
+	struct mptcp_subflow_context *subflow;
+	struct mptcp_subflows_array *array;
+
+	array = kzalloc(sizeof(*array), GFP_KERNEL);
+	if (!array)
+		return array;
+
+	mptcp_for_each_subflow(msk, subflow)
+		array->subflows[array->nr++] = subflow;
+
+	return array;
+}
+EXPORT_SYMBOL(bpf_mptcp_get_subflows_array);
+
+void bpf_mptcp_put_subflows_array(struct mptcp_subflows_array *array)
+{
+	int i;
+
+	if (!array)
+		return;
+
+	for (i = 0; i < MPTCP_SUBFLOWS_MAX; i++)
+		array->subflows[i] = NULL;
+	array->nr = 0;
+	kfree(array);
+}
+EXPORT_SYMBOL(bpf_mptcp_put_subflows_array);
diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
index 006914cb78de..c42fb54298ef 100644
--- a/net/mptcp/protocol.h
+++ b/net/mptcp/protocol.h
@@ -500,6 +500,13 @@ struct mptcp_subflow_context {
 	struct	rcu_head rcu;
 };
 
+#define MPTCP_SUBFLOWS_MAX	8
+
+struct mptcp_subflows_array {
+	u8 nr;
+	struct mptcp_subflow_context *subflows[MPTCP_SUBFLOWS_MAX];
+};
+
 static inline struct mptcp_subflow_context *
 mptcp_subflow_ctx(const struct sock *sk)
 {
-- 
2.34.1

[PATCH mptcp-next v8 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 | 13 +++++
 .../selftests/bpf/progs/mptcp_bpf_rr.c        | 55 +++++++++++++++++++
 2 files changed, 68 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 a0b83fbe8133..f0e120dfe48d 100644
--- a/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
+++ b/tools/testing/selftests/bpf/bpf_mptcp_helpers.h
@@ -27,10 +27,23 @@ 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));
+
+#define MPTCP_SUBFLOWS_MAX	8
+
+struct mptcp_subflows_array {
+	__u8 nr;
+	struct mptcp_subflow_context *subflows[MPTCP_SUBFLOWS_MAX];
+};
+
 #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..e68e3753fb4a
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/mptcp_bpf_rr.c
@@ -0,0 +1,55 @@
+// 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";
+
+extern struct mptcp_subflows_array *
+bpf_mptcp_get_subflows_array(struct mptcp_sock *msk) __ksym;
+extern void bpf_mptcp_put_subflows_array(struct mptcp_subflows_array *array) __ksym;
+
+SEC("struct_ops/mptcp_sched_rr_init")
+void BPF_PROG(mptcp_sched_rr_init, struct mptcp_sock *msk)
+{
+}
+
+SEC("struct_ops/mptcp_sched_rr_release")
+void BPF_PROG(mptcp_sched_rr_release, struct mptcp_sock *msk)
+{
+}
+
+void BPF_STRUCT_OPS(bpf_rr_get_subflow, struct mptcp_sock *msk,
+		    bool reinject, struct mptcp_sched_data *data)
+{
+	struct mptcp_subflow_context *subflow;
+	struct mptcp_subflows_array *array;
+	struct sock *ssk = msk->first;
+
+	array = bpf_mptcp_get_subflows_array(msk);
+	for (int i = 0; i < MPTCP_SUBFLOWS_MAX; i++) {
+		if (i >= array->nr)
+			break;
+
+		subflow = array->subflows[i];
+		if (subflow->tcp_sock != msk->last_snd) {
+			ssk = subflow->tcp_sock;
+			break;
+		}
+	}
+	bpf_mptcp_put_subflows_array(array);
+
+	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

[PATCH mptcp-next v8 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 377ebc0fbcbe..7052c520c0f7 100644
--- a/tools/testing/selftests/bpf/prog_tests/mptcp.c
+++ b/tools/testing/selftests/bpf/prog_tests/mptcp.c
@@ -6,6 +6,7 @@
 #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
@@ -362,10 +363,46 @@ 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

Re: [PATCH mptcp-next v8 1/3] mptcp: add bpf get_subflows helper

[Mat Martineau]

On Fri, 22 Apr 2022, Geliang Tang wrote:

> This patch implements a new helper bpf_mptcp_get_subflows() to get all the
> subflows of the given mptcp_sock, it returns the number of suflows. Add
> a new member subflows in struct mptcp_sock as a pointers array of all the
> subflows.
>
> Register this helper in bpf_mptcp_sched_kfunc_init() to make sure it can
> be accessed from the BPF context.
>
> Signed-off-by: Geliang Tang <geliang.tang@suse.com>
> ---
> net/mptcp/bpf.c      | 47 ++++++++++++++++++++++++++++++++++++++++++++
> net/mptcp/protocol.h |  7 +++++++
> 2 files changed, 54 insertions(+)
>
> diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c
> index 6c01f6b959a3..3367541b353c 100644
> --- a/net/mptcp/bpf.c
> +++ b/net/mptcp/bpf.c
> @@ -160,6 +160,23 @@ struct bpf_struct_ops bpf_mptcp_sched_ops = {
> 	.name		= "mptcp_sched_ops",
> };
>
> +BTF_SET_START(bpf_mptcp_sched_kfunc_ids)
> +BTF_ID(func, bpf_mptcp_get_subflows_array)
> +BTF_ID(func, bpf_mptcp_put_subflows_array)
> +BTF_SET_END(bpf_mptcp_sched_kfunc_ids)
> +
> +static const struct btf_kfunc_id_set bpf_mptcp_sched_kfunc_set = {
> +	.owner		= THIS_MODULE,
> +	.check_set	= &bpf_mptcp_sched_kfunc_ids,
> +};
> +
> +static int __init bpf_mptcp_sched_kfunc_init(void)
> +{
> +	return register_btf_kfunc_id_set(BPF_PROG_TYPE_STRUCT_OPS,
> +					 &bpf_mptcp_sched_kfunc_set);
> +}
> +late_initcall(bpf_mptcp_sched_kfunc_init);
> +
> struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk)
> {
> 	if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && sk_is_mptcp(sk))
> @@ -168,3 +185,33 @@ struct mptcp_sock *bpf_mptcp_sock_from_subflow(struct sock *sk)
> 	return NULL;
> }
> EXPORT_SYMBOL(bpf_mptcp_sock_from_subflow);
> +
> +struct mptcp_subflows_array *bpf_mptcp_get_subflows_array(struct mptcp_sock *msk)
> +{
> +	struct mptcp_subflow_context *subflow;
> +	struct mptcp_subflows_array *array;
> +
> +	array = kzalloc(sizeof(*array), GFP_KERNEL);
> +	if (!array)
> +		return array;
> +
> +	mptcp_for_each_subflow(msk, subflow)
> +		array->subflows[array->nr++] = subflow;
> +
> +	return array;
> +}
> +EXPORT_SYMBOL(bpf_mptcp_get_subflows_array);
> +
> +void bpf_mptcp_put_subflows_array(struct mptcp_subflows_array *array)
> +{
> +	int i;
> +
> +	if (!array)
> +		return;
> +
> +	for (i = 0; i < MPTCP_SUBFLOWS_MAX; i++)
> +		array->subflows[i] = NULL;
> +	array->nr = 0;
> +	kfree(array);

We can't trust the caller to always call this function, since we don't 
want to allow userspace to leak this array memory either accidentally or 
intentionally.

Can BPF code call a helper with a pointer to a struct on the BPF stack?

> +}
> +EXPORT_SYMBOL(bpf_mptcp_put_subflows_array);
> diff --git a/net/mptcp/protocol.h b/net/mptcp/protocol.h
> index 006914cb78de..c42fb54298ef 100644
> --- a/net/mptcp/protocol.h
> +++ b/net/mptcp/protocol.h
> @@ -500,6 +500,13 @@ struct mptcp_subflow_context {
> 	struct	rcu_head rcu;
> };
>
> +#define MPTCP_SUBFLOWS_MAX	8

I noticed that this value is separately defined here and in the next 
commit (tools/testing/selftests/bpf/bpf_mptcp_helpers.h). That seems 
fragile to me, there's no way for bpf_mptcp_sched_btf_struct_access() to 
enforce access limits correctly.

Do you have some other ideas for how to share this data between kernel and 
BPF contexts? Maybe mptcp_sched_data could be expanded to pass the subflow 
list/array in to the scheduler function.

> +
> +struct mptcp_subflows_array {
> +	u8 nr;
> +	struct mptcp_subflow_context *subflows[MPTCP_SUBFLOWS_MAX];
> +};
> +
> static inline struct mptcp_subflow_context *
> mptcp_subflow_ctx(const struct sock *sk)
> {
> -- 
> 2.34.1
>
>
>

--
Mat Martineau
Intel