Re: [PATCH 2/2] BCC: python: add a useful tool for mptcp

[Matthieu Baerts <matttbe@kernel.org>]

Hi Gang,

On 07/04/2025 04:37, Gang Yan wrote:
> Multipath TCP (MPTCP) is an extension of the standard TCP protocol
> that allows a single transport connection to use multiple network
> interfaces. MPTCP is useful for applications like bandwidth
> aggregation, failover, and more resilient connections.
> 
> Linux kernel starts to support MPTCP since v5.6, this patch provides
> a method which can easily force applications use MPTCP socket without
> modifing its code.
> 
> Signed-off-by: Gang Yan <yangang@kylinos.cn>
> ---
>  tools/mptcpboost.py  | 82 ++++++++++++++++++++++++++++++++++++++++++++
>  tools/mptcpboost.txt | 22 ++++++++++++

Small detail: what about re-using the "mptcpify" name

https://elixir.bootlin.com/linux/latest/source/tools/testing/selftests/bpf/progs/mptcpify.c
https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/commit/?id=ddba122428a7

Once applied on BCC side, we could update our website to point to this
program:

  https://www.mptcp.dev/setup.html#force-applications-to-use-mptcp

>  2 files changed, 104 insertions(+)
>  create mode 100644 tools/mptcpboost.py
>  create mode 100644 tools/mptcpboost.txt
> 
> diff --git a/tools/mptcpboost.py b/tools/mptcpboost.py
> new file mode 100644
> index 00000000..0ddd731a
> --- /dev/null
> +++ b/tools/mptcpboost.py
> @@ -0,0 +1,82 @@
> +#!/usr/bin/env python
> +#
> +# mptcpboost Make the applications to use MPTCP.
> +#           For Linux, uses BCC, eBPF. Embedded C.
> +#
> +# USAGE: mptcpboost -t
> +#
> +# Copyright 2025 Kylin Software, Inc.
> +# Licensed under the Apache License, Version 2.0 (the "License")
> +#
> +# 05-Apr-2025   Gang Yan   Created this.
> +
> +import ctypes as ct
> +import argparse
> +import signal
> +import time
> +
> +from bcc import BPF
> +
> +#arguments
> +parser = argparse.ArgumentParser(
> +         description="mptcpboost try to force applications use MPTCP")

detail: 'to' missing: try to force applications *to* use MPTCP

Maybe clearer if we add 'instead of TCP'?

> +parser.add_argument("-t", "--targets", required=True, type=str,
> +                    help="use ',' for multi targets, eg: 'iperf3,rsync'")

Could it be eventually optional to force all apps from the same CGroup
to use MPTCP instead of TCP?


Another idea is to restrict this transformation to the app passed in
argument, e.g.

  mptcpify -- iperf3 -s

(but can be done later if this makes sense)

> +
> +args_str = parser.parse_args()
> +args_list = [t.strip() for t in args_str.targets.split(',')]
> +
> +if (not BPF.support_fmod_ret()):
> +    print("Your kernel version is too old,"
> +          " fmod_ret method is only support kernel v5.6 and later.")

v5.7?

> +    exit()
> +
> +TASK_COMM_LEN = 18
> +
> +class app_name(ct.Structure):
> +    _fields_ = [("str", ct.c_char * TASK_COMM_LEN)]
> +
> +# define BPF program
> +prog = """
> +#include <linux/net.h>
> +#include <uapi/linux/in.h>
> +#include <linux/string.h>
> +
> +struct app_name {
> +    char name[TASK_COMM_LEN];
> +};
> +
> +BPF_HASH(support_apps, struct app_name);
> +
> +KMOD_RET(update_socket_protocol, int family, int type, int protocol, int ret)
> +{
> +    struct app_name target = {};
> +    bpf_get_current_comm(&target.name, TASK_COMM_LEN);
> +
> +    if ((family == AF_INET || family == AF_INET6) &&
> +        type == SOCK_STREAM &&
> +        (!protocol || protocol ==IPPROTO_TCP) &&

detail: space missing after '=='

> +        support_apps.lookup(&target))
> +        return IPPROTO_MPTCP;
> +
> +    return protocol;
> +
> +}
> +
> +"""
> +
> +b = BPF(text=prog)
> +b.attach_fmod_ret("update_socket_protocol")
> +
> +support_apps = b.get_table("support_apps")
> +for i in args_list:
> +    app = i.encode()
> +    name = app_name()
> +    name.str = app[:TASK_COMM_LEN-1].ljust(TASK_COMM_LEN, b'\0')
> +    support_apps[name] = ct.c_uint32(1)
> +
> +print("MPTCP is been applied in ", args_list);

detail: MPTCP is being forced for:

> +signal.pause()
> +#while(1):
> +#   time.sleep(200)

I guess you can remove the comments.

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.