Tcp拥塞控制算法设置相关

April 18, 2022 | 2 Minute Read

运行时修改系统默认tcp拥塞控制算法

/proc/sys/net/ipv4/tcp_available_congestion_control
echo bbr >  /proc/sys/net/ipv4/tcp_congestion_control

内核里面创建tcp socket时初始化默认的拥塞控制算法

tcp_init_sock
  tcp_assign_congestion_control(sk);

内核修改某个socket的拥塞控制算法的接口

do_tcp_setsockopt(TCP_CONGESTION)
    tcp_set_congestion_control()

c语言设置tcp socket的拥塞控制算法

https://stackoverflow.com/questions/59265004/how-to-change-tcp-congestion-control-algorithm-using-setsockopt-call-from-c

#include <sys/socket.h>
#include <netinet/in.h>
#include <netinet/tcp.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>

int main(int argc, char **argv)
{
    char buf[256];
    socklen_t len;
    int sock = socket(AF_INET, SOCK_STREAM, 0);

    if (sock == -1)
    {
        perror("socket");
        return -1;
    }

    len = sizeof(buf);

    if (getsockopt(sock, IPPROTO_TCP, TCP_CONGESTION, buf, &len) != 0)
    {
        perror("getsockopt");
        return -1;
    }

    printf("Current: %s\n", buf);

    strcpy(buf, "reno");

    len = strlen(buf);

    if (setsockopt(sock, IPPROTO_TCP, TCP_CONGESTION, buf, len) != 0)
    {
        perror("setsockopt");
        return -1;
    }

    len = sizeof(buf);

    if (getsockopt(sock, IPPROTO_TCP, TCP_CONGESTION, buf, &len) != 0)
    {
        perror("getsockopt");
        return -1;
    }

    printf("New: %s\n", buf);

    close(sock);
    return 0;
}

专门的补丁允许bpf里面修改tcp socket的 拥塞控制算法

bpf: Allow bpf tcp iter to do bpf_setsockopt https://lwn.net/Articles/861140/

BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level, int, optname, char *, optval, int, optlen) { if (level == SOL_TCP && optname == TCP_CONGESTION) { if (optlen >= sizeof(“cdg”) - 1 && !strncmp(“cdg”, optval, optlen)) return -ENOTSUPP; }

return _bpf_setsockopt(sk, level, optname, optval, optlen); }

tcp里面预先定义很多事件或者状态的变化bpf函数回调钩子

tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_CONNECT_CB, 0, NULL);
https://elixir.bootlin.com/linux/latest/source/include/uapi/linux/bpf.h

这里是一个遍历查找某个tcp连接 修改它的“拥塞控制算法”的bpf程序的例子？ https://github.com/torvalds/linux/blob/master/tools/testing/selftests/bpf/prog_tests/bpf_iter_setsockopt.c

Customize TCP initial RTO (retransmission timeout) with BPF

http://arthurchiao.art/blog/customize-tcp-initial-rto-with-bpf/

Making the Linux TCP stack more extensible with eBPF

https://inl.info.ucl.ac.be/system/files/Netdev+0x13+-+Making+the+Linux+TCP+stack+more+extensible+with+eBPF+(1).pdf