tcp westwood源代码分析

/*
 * TCP Westwood+: end-to-end bandwidth estimation for TCP
 *
 *      Angelo Dell‘Aera: author of the first version of TCP Westwood+ in Linux 2.4
 *
 * Support at http://c3lab.poliba.it/index.php/Westwood
 * Main references in literature:
 *
 * - Mascolo S, Casetti, M. Gerla et al.
 *   "TCP Westwood: bandwidth estimation for TCP" Proc. ACM Mobicom 2001
 *
 * - A. Grieco, s. Mascolo
 *   "Performance evaluation of New Reno, Vegas, Westwood+ TCP" ACM Computer
 *     Comm. Review, 2004
 *
 * - A. Dell‘Aera, L. Grieco, S. Mascolo.
 *   "Linux 2.4 Implementation of Westwood+ TCP with Rate-Halving :
 *    A Performance Evaluation Over the Internet" (ICC 2004), Paris, June 2004
 *
 * Westwood+ employs end-to-end bandwidth measurement to set cwnd and
 * ssthresh after packet loss. The probing phase is as the original Reno.
 */

#include <linux/mm.h>
#include <linux/module.h>
#include <linux/skbuff.h>
#include <linux/inet_diag.h>
#include <net/tcp.h>

/* TCP Westwood structure */
struct westwood {
    u32    bw_ns_est;        /*中间变量，经过一次平滑后的带宽值  first bandwidth estimation..not too smoothed 8) */
    u32    bw_est;           /*最终估计的带宽值  bandwidth estimate */
    u32    rtt_win_sx;       /*采样周期的起始点，一个RTT后结束 here starts a new evaluation... */
    u32    bk;               //在某个时间段delta内确认的字节数
    u32    snd_una;          /*snd_una历史记录，用于计算bk    used for evaluating the number of acked bytes */
    u32    cumul_ack;        //在慢速路径下，一个ACK确认的数据量  cumulative ack
    u32    accounted;        //在慢速路径下，收到的重复数据包个数
    u32    rtt;              //当前RTT值，以毫秒为单位，每收到一个ACK都更新
    u32    rtt_min;          /*最小RTT值，以毫秒为单位，inimum observed RTT */
    u8     first_ack;        /*是否第一个ack包   flag which infers that this is the first ack */
    u8     reset_rtt_min;    /*是否重置采样周期  Reset RTT min to next RTT sample*/
};

/* TCP Westwood functions and constants */
#define TCP_WESTWOOD_RTT_MIN   (HZ/20)    /* 50ms */
#define TCP_WESTWOOD_INIT_RTT  (20*HZ)    /* 20000ms 太保守???  maybe too conservative?! */

/*
westwood在丢包率较高的无线网络中表现较好。
Reno对随机丢包和拥塞丢包都较为敏感，随机丢包会导致Reno不必要的降低拥塞窗口和慢启动阈值。
在westwood算法中，需要强调的一点是：丢包对带宽的影响不大。
每个RTT采样一次带宽值，而这次样本只占bw_est的1/64。丢包后进入快速恢复阶段，尽管在快速恢复阶段
中得到的几个采样值较小，但是整体的bw_est却没有太大的减小。
来看一下为什么westwood对随机丢包不敏感。
（1）随机丢包
丢包前：cwnd = bw_est * RTT
丢包后：cwnd = bw_est * RTTmin
因为是随机丢包，所以丢包前的RTT只是比RTTmin略大。丢包后的bw_est也只是微略减小。
所以丢包后的cwnd只是微略的减小。
（2）拥塞丢包
丢包前：cwnd = bw_est * RTTmax => BDP + Queue
丢包后：cwnd = bw_est * RTTmin => BDP
因为是拥塞丢包，所以丢包前的bw_est已经是连接的最大带宽，并且时延也达到了最大值。
这是丢包后就达到了完全利用BDP，同时使Queue为空的效果。
可以看到，westwood对随机丢包和拥塞丢包采取同样的算法来处理，却能达到不同的效果。
但是，westwood不能主动的区分随机丢包和拥塞丢包。
*/

/*
 * @tcp_westwood_create
 * This function initializes fields used in TCP Westwood+,
 * it is called after the initial SYN, so the sequence numbers
 * are correct but new passive connections we have no
 * information about RTTmin at this time so we simply set it to
 * TCP_WESTWOOD_INIT_RTT. This value was chosen to be too conservative
 * since in this way we‘re sure it will be updated in a consistent
 * way as soon as possible. It will reasonably happen within the first
 * RTT period of the connection lifetime.
 */
static void tcp_westwood_init(struct sock *sk)
{ //初始化westwood拥塞算法的参数
    struct westwood *w = inet_csk_ca(sk);

    w->bk = 0;
    w->bw_ns_est = 0;
    w->bw_est = 0;
    w->accounted = 0;
    w->cumul_ack = 0;
    w->reset_rtt_min = 1;
    w->rtt_min = w->rtt = TCP_WESTWOOD_INIT_RTT;
    w->rtt_win_sx = tcp_time_stamp;//jiffies毫秒
    w->snd_una = tcp_sk(sk)->snd_una;
    w->first_ack = 1;
}

/*
 * @westwood_do_filter
 * Low-pass filter. Implemented using constant coefficients.
 */
static inline u32 westwood_do_filter(u32 a, u32 b)
{
    return (((7 * a) + b) >> 3); //返回7a/8与1b/8之和
}

static void westwood_filter(struct westwood *w, u32 delta)
{ //带宽过滤器
    /* If the filter is empty fill it with the first sample of bandwidth  */
    if (w->bw_ns_est == 0 && w->bw_est == 0) {
        //如果是第一次得到带宽测量样本
        w->bw_ns_est = w->bk / delta;
        w->bw_est = w->bw_ns_est;
    } else {//如果已经有收到过测量样本了
        w->bw_ns_est = westwood_do_filter(w->bw_ns_est, w->bk / delta);
        w->bw_est = westwood_do_filter(w->bw_est, w->bw_ns_est);
    }
}

/*
 * @westwood_pkts_acked
 * Called after processing group of packets.
 * but all westwood needs is the last sample of srtt.
 */
static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt, s32 rtt)
{ //每收到一个ACK时，会更新当前的RTT（w->rtt）
    struct westwood *w = inet_csk_ca(sk);

    if (rtt > 0) //这个参数rtt是以微秒为单位的，rtt为-1则不会执行转换
        w->rtt = usecs_to_jiffies(rtt); //w->rtt是以毫秒为单位的
}

/*
 * @westwood_update_window
 * It updates RTT evaluation window if it is the right moment to do
 * it. If so it calls filter for evaluating bandwidth.
 */
static void westwood_update_window(struct sock *sk)
{ //每经过一个RTT后，采集一个新的测量样本，更新带宽估计值。
    struct westwood *w = inet_csk_ca(sk);
    s32 delta = tcp_time_stamp - w->rtt_win_sx; //计算时间差

    /* Initialize w->snd_una with the first acked sequence number in order
     * to fix mismatch between tp->snd_una and w->snd_una for the first
     * bandwidth sample
     */
    if (w->first_ack) {//如果是第一个ack包
        w->snd_una = tcp_sk(sk)->snd_una;
        w->first_ack = 0;
    }

    /*
     * See if a RTT-window has passed.
     * Be careful since if RTT is less than
     * 50ms we don‘t filter but we continue ‘building the sample‘.
     * This minimum limit was chosen since an estimation on small
     * time intervals is better to avoid...
     * Obviously on a LAN we reasonably will always have
     * right_bound = left_bound + WESTWOOD_RTT_MIN
     */
     //如果当前的rtt大于TCP_WESTWOOD_RTT_MIN，也就是超时了
    if (w->rtt && delta > max_t(u32, w->rtt, TCP_WESTWOOD_RTT_MIN)) {
        westwood_filter(w, delta);// 更新带宽估计值

        w->bk = 0;//清零确认字节数
        w->rtt_win_sx = tcp_time_stamp;//重设取样周期开始时间
    }
}

static inline void update_rtt_min(struct westwood *w)
{//更新最小RTT
    if (w->reset_rtt_min) { //如果非0
        //当发生超时后，最小RTT可能不再准确，需要更新
        w->rtt_min = w->rtt;
        w->reset_rtt_min = 0;
    } else  //如果为0
        w->rtt_min = min(w->rtt, w->rtt_min);//更新最小RTT
}

/*
 * @westwood_fast_bw
 * It is called when we are in fast path. In particular it is called when
 * header prediction is successful. In such case in fact update is
 * straight forward and doesn‘t need any particular care.
 */
//快速路径时的带宽估计值更新
//处于快速路径时调用，说明此时收到的数据包是顺序的，此时应该处于Open状态。
//这种状态下，收到新的ACK会使tp->snd_una前进。
//所以，tp->snd_una - w->snd_una能代表此ACK确认的数据量。
static inline void westwood_fast_bw(struct sock *sk)
{
    const struct tcp_sock *tp = tcp_sk(sk);
    struct westwood *w = inet_csk_ca(sk);

    westwood_update_window(sk); //更新带宽估计值

    w->bk += tp->snd_una - w->snd_una; //累计确认的字节数
    w->snd_una = tp->snd_una;//记录当前snd_una
    update_rtt_min(w); //更新最小RTT
}

/*
 * @westwood_acked_count
 * This function evaluates cumul_ack for evaluating bk in case of
 * delayed or partial acks.
 */
//慢速路径时，计算所收到ACK确认的数据量。
//这时候的ACK可能是delayed ACK、partial ACK、duplicate ACK、
//cumulative ACK following a retransmission event.

static inline u32 westwood_acked_count(struct sock *sk)
{
    const struct tcp_sock *tp = tcp_sk(sk);
    struct westwood *w = inet_csk_ca(sk);

    //计算此ACK确认的字节数
    w->cumul_ack = tp->snd_una - w->snd_una;

    /* If cumul_ack is 0 this is a dupack since it‘s not moving
     * tp->snd_una.
     */
    //如果cumul_ack=0，那么此ACK是dupack，
    //代表接收端收到一个数据包。
    if (!w->cumul_ack) {
        w->accounted += tp->mss_cache;//接收端保存的乱序数据包加一
        w->cumul_ack = tp->mss_cache;//代表传输了一个数据包
    }

    //如果cumul_ack > 1，则有可能是多种情况。
    if (w->cumul_ack > tp->mss_cache) {
        /* Partial or delayed ack   表示此ACK为partial ACK */
        if (w->accounted >= w->cumul_ack) {
            w->accounted -= w->cumul_ack;
            //表示只确认了一个包，其它包已经被dupack确认过了
            w->cumul_ack = tp->mss_cache;
        } else {
        /* delayed ack or cumulative ack，
         * 表示被延迟的确认，或者结束Recovery的累积确认
         */
            w->cumul_ack -= w->accounted;
            w->accounted = 0;
        }
    }

    w->snd_una = tp->snd_una; //记录当前的snd_una

    return w->cumul_ack;//返回此ACK确认的字节数
}

/*
 * TCP Westwood
 * Here limit is evaluated as Bw estimation*RTTmin (for obtaining it
 * in packets we use mss_cache). Rttmin is guaranteed to be >= 2
 * so avoids ever returning 0.
 */
static u32 tcp_westwood_bw_rttmin(const struct sock *sk)
{//此函数在丢包后调用，根据带宽来设置拥塞窗口和慢启动阈值。
    const struct tcp_sock *tp = tcp_sk(sk);
    const struct westwood *w = inet_csk_ca(sk);
    return max_t(u32, (w->bw_est * w->rtt_min) / tp->mss_cache, 2);
}

static void tcp_westwood_event(struct sock *sk, enum tcp_ca_event event)
{//westwood的"入口函数"，其他函数都是通过这个函数调用的，
 //不同与其他的TCP拥塞控制算法。
    struct tcp_sock *tp = tcp_sk(sk);
    struct westwood *w = inet_csk_ca(sk);

    switch (event) {
        /* 处于快速路径时，用此函数更新w->bw_est和w->rtt_min */
    case CA_EVENT_FAST_ACK:
        westwood_fast_bw(sk);
        break;

        /* 退出Recovery或CWR状态时，进行拥塞窗口和慢启动阈值设置*/
    case CA_EVENT_COMPLETE_CWR:
        tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
        break;

        //当RTO超时时，会先进行FRTO检测，这时可设置慢启动阈值，
        //而拥塞窗口则设置为1.
    case CA_EVENT_FRTO:
        tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
        /* Update RTT_min when next ack arrives */
        //如果超时了，那么最小RTT可能不准确，需要重新设置
        w->reset_rtt_min = 1;
        break;

        //处于慢速路径时，这时候的拥塞状态可能是CWR、Recovery、Loss等。
        //必须采用westwood_acked_count()来统计此ACK确认的数据量。
        //同时也进行w->bw_est和w->rtt_min的更新。
    case CA_EVENT_SLOW_ACK:
        westwood_update_window(sk);//更新带宽估计值
        w->bk += westwood_acked_count(sk);//计算所收到ACK确认的数据量
        update_rtt_min(w);//更新最小rtt
        break;

    default:
        /* don‘t care 对其它的事件则不做响应 */
        break;
    }
}

/* Extract info for Tcp socket info provided via netlink. */
//通过netlink提供信息给tcp_socket
static void tcp_westwood_info(struct sock *sk, u32 ext,
                  struct sk_buff *skb)
{
    const struct westwood *ca = inet_csk_ca(sk);
    if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
        struct tcpvegas_info info = {
            .tcpv_enabled = 1,
            .tcpv_rtt = jiffies_to_usecs(ca->rtt),
            .tcpv_minrtt = jiffies_to_usecs(ca->rtt_min),
        };

        nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info);
    }
}

static struct tcp_congestion_ops tcp_westwood = {
    .init        = tcp_westwood_init,
    .ssthresh    = tcp_reno_ssthresh,
    .cong_avoid    = tcp_reno_cong_avoid,
    .min_cwnd    = tcp_westwood_bw_rttmin,
    .cwnd_event    = tcp_westwood_event,
    .get_info    = tcp_westwood_info,
    .pkts_acked    = tcp_westwood_pkts_acked,

    .owner        = THIS_MODULE,
    .name        = "westwood"
};

static int __init tcp_westwood_register(void)
{ //注册westwood算法
    BUILD_BUG_ON(sizeof(struct westwood) > ICSK_CA_PRIV_SIZE);
    return tcp_register_congestion_control(&tcp_westwood);
}

static void __exit tcp_westwood_unregister(void)
{ //注销westwood算法
    tcp_unregister_congestion_control(&tcp_westwood);
}

module_init(tcp_westwood_register);//westwood模块的入口函数
module_exit(tcp_westwood_unregister);//westwood模块的出口函数

MODULE_AUTHOR("Stephen Hemminger, Angelo Dell‘Aera");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("TCP Westwood+");