* assumed senders never went idle.
*/
-#include <linux/config.h>
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/skbuff.h>
#include <net/tcp.h>
-/* Default values of the Vegas variables, in fixed-point representation
- * with V_PARAM_SHIFT bits to the right of the binary point.
- */
-#define V_PARAM_SHIFT 1
-static int alpha = 1<<V_PARAM_SHIFT;
-static int beta = 3<<V_PARAM_SHIFT;
-static int gamma = 1<<V_PARAM_SHIFT;
+#include "tcp_vegas.h"
+
+static int alpha = 2;
+static int beta = 4;
+static int gamma = 1;
module_param(alpha, int, 0644);
-MODULE_PARM_DESC(alpha, "lower bound of packets in network (scale by 2)");
+MODULE_PARM_DESC(alpha, "lower bound of packets in network");
module_param(beta, int, 0644);
-MODULE_PARM_DESC(beta, "upper bound of packets in network (scale by 2)");
+MODULE_PARM_DESC(beta, "upper bound of packets in network");
module_param(gamma, int, 0644);
MODULE_PARM_DESC(gamma, "limit on increase (scale by 2)");
-/* Vegas variables */
-struct vegas {
- u32 beg_snd_nxt; /* right edge during last RTT */
- u32 beg_snd_una; /* left edge during last RTT */
- u32 beg_snd_cwnd; /* saves the size of the cwnd */
- u8 doing_vegas_now;/* if true, do vegas for this RTT */
- u16 cntRTT; /* # of RTTs measured within last RTT */
- u32 minRTT; /* min of RTTs measured within last RTT (in usec) */
- u32 baseRTT; /* the min of all Vegas RTT measurements seen (in usec) */
-};
-
/* There are several situations when we must "re-start" Vegas:
*
* o when a connection is established
* Instead we must wait until the completion of an RTT during
* which we actually receive ACKs.
*/
-static inline void vegas_enable(struct sock *sk)
+static void vegas_enable(struct sock *sk)
{
const struct tcp_sock *tp = tcp_sk(sk);
struct vegas *vegas = inet_csk_ca(sk);
vegas->doing_vegas_now = 0;
}
-static void tcp_vegas_init(struct sock *sk)
+void tcp_vegas_init(struct sock *sk)
{
struct vegas *vegas = inet_csk_ca(sk);
vegas->baseRTT = 0x7fffffff;
vegas_enable(sk);
}
+EXPORT_SYMBOL_GPL(tcp_vegas_init);
/* Do RTT sampling needed for Vegas.
* Basically we:
* o min-filter RTT samples from a much longer window (forever for now)
* to find the propagation delay (baseRTT)
*/
-static void tcp_vegas_rtt_calc(struct sock *sk, u32 usrtt)
+void tcp_vegas_pkts_acked(struct sock *sk, u32 cnt, s32 rtt_us)
{
struct vegas *vegas = inet_csk_ca(sk);
- u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */
+ u32 vrtt;
+
+ if (rtt_us < 0)
+ return;
+
+ /* Never allow zero rtt or baseRTT */
+ vrtt = rtt_us + 1;
/* Filter to find propagation delay: */
if (vrtt < vegas->baseRTT)
vegas->minRTT = min(vegas->minRTT, vrtt);
vegas->cntRTT++;
}
+EXPORT_SYMBOL_GPL(tcp_vegas_pkts_acked);
-static void tcp_vegas_state(struct sock *sk, u8 ca_state)
+void tcp_vegas_state(struct sock *sk, u8 ca_state)
{
if (ca_state == TCP_CA_Open)
else
vegas_disable(sk);
}
+EXPORT_SYMBOL_GPL(tcp_vegas_state);
/*
* If the connection is idle and we are restarting,
* packets, _then_ we can make Vegas calculations
* again.
*/
-static void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event)
+void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event)
{
if (event == CA_EVENT_CWND_RESTART ||
event == CA_EVENT_TX_START)
tcp_vegas_init(sk);
}
+EXPORT_SYMBOL_GPL(tcp_vegas_cwnd_event);
-static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack,
- u32 seq_rtt, u32 in_flight, int flag)
+static inline u32 tcp_vegas_ssthresh(struct tcp_sock *tp)
+{
+ return min(tp->snd_ssthresh, tp->snd_cwnd-1);
+}
+
+static void tcp_vegas_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
{
struct tcp_sock *tp = tcp_sk(sk);
struct vegas *vegas = inet_csk_ca(sk);
- if (!vegas->doing_vegas_now)
- return tcp_reno_cong_avoid(sk, ack, seq_rtt, in_flight, flag);
-
- /* The key players are v_beg_snd_una and v_beg_snd_nxt.
- *
- * These are so named because they represent the approximate values
- * of snd_una and snd_nxt at the beginning of the current RTT. More
- * precisely, they represent the amount of data sent during the RTT.
- * At the end of the RTT, when we receive an ACK for v_beg_snd_nxt,
- * we will calculate that (v_beg_snd_nxt - v_beg_snd_una) outstanding
- * bytes of data have been ACKed during the course of the RTT, giving
- * an "actual" rate of:
- *
- * (v_beg_snd_nxt - v_beg_snd_una) / (rtt duration)
- *
- * Unfortunately, v_beg_snd_una is not exactly equal to snd_una,
- * because delayed ACKs can cover more than one segment, so they
- * don't line up nicely with the boundaries of RTTs.
- *
- * Another unfortunate fact of life is that delayed ACKs delay the
- * advance of the left edge of our send window, so that the number
- * of bytes we send in an RTT is often less than our cwnd will allow.
- * So we keep track of our cwnd separately, in v_beg_snd_cwnd.
- */
+ if (!vegas->doing_vegas_now) {
+ tcp_reno_cong_avoid(sk, ack, in_flight);
+ return;
+ }
if (after(ack, vegas->beg_snd_nxt)) {
/* Do the Vegas once-per-RTT cwnd adjustment. */
- u32 old_wnd, old_snd_cwnd;
-
-
- /* Here old_wnd is essentially the window of data that was
- * sent during the previous RTT, and has all
- * been acknowledged in the course of the RTT that ended
- * with the ACK we just received. Likewise, old_snd_cwnd
- * is the cwnd during the previous RTT.
- */
- old_wnd = (vegas->beg_snd_nxt - vegas->beg_snd_una) /
- tp->mss_cache;
- old_snd_cwnd = vegas->beg_snd_cwnd;
/* Save the extent of the current window so we can use this
* at the end of the next RTT.
*/
- vegas->beg_snd_una = vegas->beg_snd_nxt;
vegas->beg_snd_nxt = tp->snd_nxt;
- vegas->beg_snd_cwnd = tp->snd_cwnd;
-
- /* Take into account the current RTT sample too, to
- * decrease the impact of delayed acks. This double counts
- * this sample since we count it for the next window as well,
- * but that's not too awful, since we're taking the min,
- * rather than averaging.
- */
- tcp_vegas_rtt_calc(sk, seq_rtt * 1000);
/* We do the Vegas calculations only if we got enough RTT
* samples that we can be reasonably sure that we got
/* We don't have enough RTT samples to do the Vegas
* calculation, so we'll behave like Reno.
*/
- if (tp->snd_cwnd > tp->snd_ssthresh)
- tp->snd_cwnd++;
+ tcp_reno_cong_avoid(sk, ack, in_flight);
} else {
- u32 rtt, target_cwnd, diff;
+ u32 rtt, diff;
+ u64 target_cwnd;
/* We have enough RTT samples, so, using the Vegas
* algorithm, we determine if we should increase or
*
* This is:
* (actual rate in segments) * baseRTT
- * We keep it as a fixed point number with
- * V_PARAM_SHIFT bits to the right of the binary point.
*/
- target_cwnd = ((old_wnd * vegas->baseRTT)
- << V_PARAM_SHIFT) / rtt;
+ target_cwnd = tp->snd_cwnd * vegas->baseRTT / rtt;
/* Calculate the difference between the window we had,
* and the window we would like to have. This quantity
* is the "Diff" from the Arizona Vegas papers.
- *
- * Again, this is a fixed point number with
- * V_PARAM_SHIFT bits to the right of the binary
- * point.
*/
- diff = (old_wnd << V_PARAM_SHIFT) - target_cwnd;
+ diff = tp->snd_cwnd * (rtt-vegas->baseRTT) / vegas->baseRTT;
- if (tp->snd_cwnd < tp->snd_ssthresh) {
- /* Slow start. */
- if (diff > gamma) {
- /* Going too fast. Time to slow down
- * and switch to congestion avoidance.
- */
- tp->snd_ssthresh = 2;
-
- /* Set cwnd to match the actual rate
- * exactly:
- * cwnd = (actual rate) * baseRTT
- * Then we add 1 because the integer
- * truncation robs us of full link
- * utilization.
- */
- tp->snd_cwnd = min(tp->snd_cwnd,
- (target_cwnd >>
- V_PARAM_SHIFT)+1);
+ if (diff > gamma && tp->snd_cwnd <= tp->snd_ssthresh) {
+ /* Going too fast. Time to slow down
+ * and switch to congestion avoidance.
+ */
- }
+ /* Set cwnd to match the actual rate
+ * exactly:
+ * cwnd = (actual rate) * baseRTT
+ * Then we add 1 because the integer
+ * truncation robs us of full link
+ * utilization.
+ */
+ tp->snd_cwnd = min(tp->snd_cwnd, (u32)target_cwnd+1);
+ tp->snd_ssthresh = tcp_vegas_ssthresh(tp);
+
+ } else if (tp->snd_cwnd <= tp->snd_ssthresh) {
+ /* Slow start. */
+ tcp_slow_start(tp);
} else {
/* Congestion avoidance. */
- u32 next_snd_cwnd;
/* Figure out where we would like cwnd
* to be.
/* The old window was too fast, so
* we slow down.
*/
- next_snd_cwnd = old_snd_cwnd - 1;
+ tp->snd_cwnd--;
+ tp->snd_ssthresh
+ = tcp_vegas_ssthresh(tp);
} else if (diff < alpha) {
/* We don't have enough extra packets
* in the network, so speed up.
*/
- next_snd_cwnd = old_snd_cwnd + 1;
+ tp->snd_cwnd++;
} else {
/* Sending just as fast as we
* should be.
*/
- next_snd_cwnd = old_snd_cwnd;
}
-
- /* Adjust cwnd upward or downward, toward the
- * desired value.
- */
- if (next_snd_cwnd > tp->snd_cwnd)
- tp->snd_cwnd++;
- else if (next_snd_cwnd < tp->snd_cwnd)
- tp->snd_cwnd--;
}
+
+ if (tp->snd_cwnd < 2)
+ tp->snd_cwnd = 2;
+ else if (tp->snd_cwnd > tp->snd_cwnd_clamp)
+ tp->snd_cwnd = tp->snd_cwnd_clamp;
+
+ tp->snd_ssthresh = tcp_current_ssthresh(sk);
}
/* Wipe the slate clean for the next RTT. */
vegas->cntRTT = 0;
vegas->minRTT = 0x7fffffff;
}
+ /* Use normal slow start */
+ else if (tp->snd_cwnd <= tp->snd_ssthresh)
+ tcp_slow_start(tp);
- /* The following code is executed for every ack we receive,
- * except for conditions checked in should_advance_cwnd()
- * before the call to tcp_cong_avoid(). Mainly this means that
- * we only execute this code if the ack actually acked some
- * data.
- */
-
- /* If we are in slow start, increase our cwnd in response to this ACK.
- * (If we are not in slow start then we are in congestion avoidance,
- * and adjust our congestion window only once per RTT. See the code
- * above.)
- */
- if (tp->snd_cwnd <= tp->snd_ssthresh)
- tp->snd_cwnd++;
-
- /* to keep cwnd from growing without bound */
- tp->snd_cwnd = min_t(u32, tp->snd_cwnd, tp->snd_cwnd_clamp);
-
- /* Make sure that we are never so timid as to reduce our cwnd below
- * 2 MSS.
- *
- * Going below 2 MSS would risk huge delayed ACKs from our receiver.
- */
- tp->snd_cwnd = max(tp->snd_cwnd, 2U);
}
/* Extract info for Tcp socket info provided via netlink. */
-static void tcp_vegas_get_info(struct sock *sk, u32 ext,
- struct sk_buff *skb)
+void tcp_vegas_get_info(struct sock *sk, u32 ext, struct sk_buff *skb)
{
const struct vegas *ca = inet_csk_ca(sk);
if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
- struct tcpvegas_info *info;
-
- info = RTA_DATA(__RTA_PUT(skb, INET_DIAG_VEGASINFO,
- sizeof(*info)));
-
- info->tcpv_enabled = ca->doing_vegas_now;
- info->tcpv_rttcnt = ca->cntRTT;
- info->tcpv_rtt = ca->baseRTT;
- info->tcpv_minrtt = ca->minRTT;
- rtattr_failure: ;
+ struct tcpvegas_info info = {
+ .tcpv_enabled = ca->doing_vegas_now,
+ .tcpv_rttcnt = ca->cntRTT,
+ .tcpv_rtt = ca->baseRTT,
+ .tcpv_minrtt = ca->minRTT,
+ };
+
+ nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info);
}
}
+EXPORT_SYMBOL_GPL(tcp_vegas_get_info);
static struct tcp_congestion_ops tcp_vegas = {
+ .flags = TCP_CONG_RTT_STAMP,
.init = tcp_vegas_init,
.ssthresh = tcp_reno_ssthresh,
.cong_avoid = tcp_vegas_cong_avoid,
.min_cwnd = tcp_reno_min_cwnd,
- .rtt_sample = tcp_vegas_rtt_calc,
+ .pkts_acked = tcp_vegas_pkts_acked,
.set_state = tcp_vegas_state,
.cwnd_event = tcp_vegas_cwnd_event,
.get_info = tcp_vegas_get_info,
static int __init tcp_vegas_register(void)
{
- BUG_ON(sizeof(struct vegas) > ICSK_CA_PRIV_SIZE);
+ BUILD_BUG_ON(sizeof(struct vegas) > ICSK_CA_PRIV_SIZE);
tcp_register_congestion_control(&tcp_vegas);
return 0;
}