net/ipv4/tcp_veno.c

   1 /*
   2  * TCP Veno congestion control
   3  *
   4  * This is based on the congestion detection/avoidance scheme described in
   5  *    C. P. Fu, S. C. Liew.
   6  *    "TCP Veno: TCP Enhancement for Transmission over Wireless Access Networks."
   7  *    IEEE Journal on Selected Areas in Communication,
   8  *    Feb. 2003.
   9  *      See http://www.ntu.edu.sg/home5/ZHOU0022/papers/CPFu03a.pdf
  10  */
  11
  12 #include <linux/mm.h>
  13 #include <linux/module.h>
  14 #include <linux/skbuff.h>
  15 #include <linux/inet_diag.h>
  16
  17 #include <net/tcp.h>
  18
  19 /* Default values of the Veno variables, in fixed-point representation
  20  * with V_PARAM_SHIFT bits to the right of the binary point.
  21  */
  22 #define V_PARAM_SHIFT 1
  23 static const int beta = 3 << V_PARAM_SHIFT;
  24
  25 /* Veno variables */
  26 struct veno {
  27         u8 doing_veno_now;      /* if true, do veno for this rtt */
  28         u16 cntrtt;             /* # of rtts measured within last rtt */
  29         u32 minrtt;             /* min of rtts measured within last rtt (in usec) */
  30         u32 basertt;            /* the min of all Veno rtt measurements seen (in usec) */
  31         u32 inc;                /* decide whether to increase cwnd */
  32         u32 diff;               /* calculate the diff rate */
  33 };
  34
  35 /* There are several situations when we must "re-start" Veno:
  36  *
  37  *  o when a connection is established
  38  *  o after an RTO
  39  *  o after fast recovery
  40  *  o when we send a packet and there is no outstanding
  41  *    unacknowledged data (restarting an idle connection)
  42  *
  43  */
  44 static inline void veno_enable(struct sock *sk)
  45 {
  46         struct veno *veno = inet_csk_ca(sk);
  47
  48         /* turn on Veno */
  49         veno->doing_veno_now = 1;
  50
  51         veno->minrtt = 0x7fffffff;
  52 }
  53
  54 static inline void veno_disable(struct sock *sk)
  55 {
  56         struct veno *veno = inet_csk_ca(sk);
  57
  58         /* turn off Veno */
  59         veno->doing_veno_now = 0;
  60 }
  61
  62 static void tcp_veno_init(struct sock *sk)
  63 {
  64         struct veno *veno = inet_csk_ca(sk);
  65
  66         veno->basertt = 0x7fffffff;
  67         veno->inc = 1;
  68         veno_enable(sk);
  69 }
  70
  71 /* Do rtt sampling needed for Veno. */
  72 static void tcp_veno_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
  73 {
  74         struct veno *veno = inet_csk_ca(sk);
  75         u32 vrtt;
  76
  77         /* Never allow zero rtt or baseRTT */
  78         vrtt = ktime_to_us(net_timedelta(last)) + 1;
  79
  80         /* Filter to find propagation delay: */
  81         if (vrtt < veno->basertt)
  82                 veno->basertt = vrtt;
  83
  84         /* Find the min rtt during the last rtt to find
  85          * the current prop. delay + queuing delay:
  86          */
  87         veno->minrtt = min(veno->minrtt, vrtt);
  88         veno->cntrtt++;
  89 }
  90
  91 static void tcp_veno_state(struct sock *sk, u8 ca_state)
  92 {
  93         if (ca_state == TCP_CA_Open)
  94                 veno_enable(sk);
  95         else
  96                 veno_disable(sk);
  97 }
  98
  99 /*
 100  * If the connection is idle and we are restarting,
 101  * then we don't want to do any Veno calculations
 102  * until we get fresh rtt samples.  So when we
 103  * restart, we reset our Veno state to a clean
 104  * state. After we get acks for this flight of
 105  * packets, _then_ we can make Veno calculations
 106  * again.
 107  */
 108 static void tcp_veno_cwnd_event(struct sock *sk, enum tcp_ca_event event)
 109 {
 110         if (event == CA_EVENT_CWND_RESTART || event == CA_EVENT_TX_START)
 111                 tcp_veno_init(sk);
 112 }
 113
 114 static void tcp_veno_cong_avoid(struct sock *sk, u32 ack,
 115                                 u32 seq_rtt, u32 in_flight, int flag)
 116 {
 117         struct tcp_sock *tp = tcp_sk(sk);
 118         struct veno *veno = inet_csk_ca(sk);
 119
 120         if (!veno->doing_veno_now)
 121                 return tcp_reno_cong_avoid(sk, ack, seq_rtt, in_flight, flag);
 122
 123         /* limited by applications */
 124         if (!tcp_is_cwnd_limited(sk, in_flight))
 125                 return;
 126
 127         /* We do the Veno calculations only if we got enough rtt samples */
 128         if (veno->cntrtt <= 2) {
 129                 /* We don't have enough rtt samples to do the Veno
 130                  * calculation, so we'll behave like Reno.
 131                  */
 132                 tcp_reno_cong_avoid(sk, ack, seq_rtt, in_flight, flag);
 133         } else {
 134                 u32 rtt, target_cwnd;
 135
 136                 /* We have enough rtt samples, so, using the Veno
 137                  * algorithm, we determine the state of the network.
 138                  */
 139
 140                 rtt = veno->minrtt;
 141
 142                 target_cwnd = ((tp->snd_cwnd * veno->basertt)
 143                                << V_PARAM_SHIFT) / rtt;
 144
 145                 veno->diff = (tp->snd_cwnd << V_PARAM_SHIFT) - target_cwnd;
 146
 147                 if (tp->snd_cwnd <= tp->snd_ssthresh) {
 148                         /* Slow start.  */
 149                         tcp_slow_start(tp);
 150                 } else {
 151                         /* Congestion avoidance. */
 152                         if (veno->diff < beta) {
 153                                 /* In the "non-congestive state", increase cwnd
 154                                  *  every rtt.
 155                                  */
 156                                 if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
 157                                         if (tp->snd_cwnd < tp->snd_cwnd_clamp)
 158                                                 tp->snd_cwnd++;
 159                                         tp->snd_cwnd_cnt = 0;
 160                                 } else
 161                                         tp->snd_cwnd_cnt++;
 162                         } else {
 163                                 /* In the "congestive state", increase cwnd
 164                                  * every other rtt.
 165                                  */
 166                                 if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
 167                                         if (veno->inc
 168                                             && tp->snd_cwnd <
 169                                             tp->snd_cwnd_clamp) {
 170                                                 tp->snd_cwnd++;
 171                                                 veno->inc = 0;
 172                                         } else
 173                                                 veno->inc = 1;
 174                                         tp->snd_cwnd_cnt = 0;
 175                                 } else
 176                                         tp->snd_cwnd_cnt++;
 177                         }
 178
 179                 }
 180                 if (tp->snd_cwnd < 2)
 181                         tp->snd_cwnd = 2;
 182                 else if (tp->snd_cwnd > tp->snd_cwnd_clamp)
 183                         tp->snd_cwnd = tp->snd_cwnd_clamp;
 184         }
 185         /* Wipe the slate clean for the next rtt. */
 186         /* veno->cntrtt = 0; */
 187         veno->minrtt = 0x7fffffff;
 188 }
 189
 190 /* Veno MD phase */
 191 static u32 tcp_veno_ssthresh(struct sock *sk)
 192 {
 193         const struct tcp_sock *tp = tcp_sk(sk);
 194         struct veno *veno = inet_csk_ca(sk);
 195
 196         if (veno->diff < beta)
 197                 /* in "non-congestive state", cut cwnd by 1/5 */
 198                 return max(tp->snd_cwnd * 4 / 5, 2U);
 199         else
 200                 /* in "congestive state", cut cwnd by 1/2 */
 201                 return max(tp->snd_cwnd >> 1U, 2U);
 202 }
 203
 204 static struct tcp_congestion_ops tcp_veno = {
 205         .flags          = TCP_CONG_RTT_STAMP,
 206         .init           = tcp_veno_init,
 207         .ssthresh       = tcp_veno_ssthresh,
 208         .cong_avoid     = tcp_veno_cong_avoid,
 209         .pkts_acked     = tcp_veno_pkts_acked,
 210         .set_state      = tcp_veno_state,
 211         .cwnd_event     = tcp_veno_cwnd_event,
 212
 213         .owner          = THIS_MODULE,
 214         .name           = "veno",
 215 };
 216
 217 static int __init tcp_veno_register(void)
 218 {
 219         BUILD_BUG_ON(sizeof(struct veno) > ICSK_CA_PRIV_SIZE);
 220         tcp_register_congestion_control(&tcp_veno);
 221         return 0;
 222 }
 223
 224 static void __exit tcp_veno_unregister(void)
 225 {
 226         tcp_unregister_congestion_control(&tcp_veno);
 227 }
 228
 229 module_init(tcp_veno_register);
 230 module_exit(tcp_veno_unregister);
 231
 232 MODULE_AUTHOR("Bin Zhou, Cheng Peng Fu");
 233 MODULE_LICENSE("GPL");
 234 MODULE_DESCRIPTION("TCP Veno");