2005-04-16 16:20:36 -06:00
|
|
|
/*
|
|
|
|
* net/sched/sch_red.c Random Early Detection queue.
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU General Public License
|
|
|
|
* as published by the Free Software Foundation; either version
|
|
|
|
* 2 of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
|
|
|
|
*
|
|
|
|
* Changes:
|
2005-11-05 13:14:08 -07:00
|
|
|
* J Hadi Salim 980914: computation fixes
|
2005-04-16 16:20:36 -06:00
|
|
|
* Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
|
2005-11-05 13:14:08 -07:00
|
|
|
* J Hadi Salim 980816: ECN support
|
2005-04-16 16:20:36 -06:00
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/types.h>
|
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/skbuff.h>
|
|
|
|
#include <net/pkt_sched.h>
|
2017-11-05 23:23:41 -07:00
|
|
|
#include <net/pkt_cls.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
#include <net/inet_ecn.h>
|
2005-11-05 13:14:05 -07:00
|
|
|
#include <net/red.h>
|
2005-04-16 16:20:36 -06:00
|
|
|
|
|
|
|
|
2005-11-05 13:14:05 -07:00
|
|
|
/* Parameters, settable by user:
|
2005-04-16 16:20:36 -06:00
|
|
|
-----------------------------
|
|
|
|
|
|
|
|
limit - bytes (must be > qth_max + burst)
|
|
|
|
|
|
|
|
Hard limit on queue length, should be chosen >qth_max
|
|
|
|
to allow packet bursts. This parameter does not
|
|
|
|
affect the algorithms behaviour and can be chosen
|
|
|
|
arbitrarily high (well, less than ram size)
|
|
|
|
Really, this limit will never be reached
|
|
|
|
if RED works correctly.
|
|
|
|
*/
|
|
|
|
|
2011-01-19 12:26:56 -07:00
|
|
|
struct red_sched_data {
|
2005-11-05 13:14:05 -07:00
|
|
|
u32 limit; /* HARD maximal queue length */
|
|
|
|
unsigned char flags;
|
sch_red: Adaptative RED AQM
Adaptative RED AQM for linux, based on paper from Sally FLoyd,
Ramakrishna Gummadi, and Scott Shenker, August 2001 :
http://icir.org/floyd/papers/adaptiveRed.pdf
Goal of Adaptative RED is to make max_p a dynamic value between 1% and
50% to reach the target average queue : (max_th - min_th) / 2
Every 500 ms:
if (avg > target and max_p <= 0.5)
increase max_p : max_p += alpha;
else if (avg < target and max_p >= 0.01)
decrease max_p : max_p *= beta;
target :[min_th + 0.4*(min_th - max_th),
min_th + 0.6*(min_th - max_th)].
alpha : min(0.01, max_p / 4)
beta : 0.9
max_P is a Q0.32 fixed point number (unsigned, with 32 bits mantissa)
Changes against our RED implementation are :
max_p is no longer a negative power of two (1/(2^Plog)), but a Q0.32
fixed point number, to allow full range described in Adatative paper.
To deliver a random number, we now use a reciprocal divide (thats really
a multiply), but this operation is done once per marked/droped packet
when in RED_BETWEEN_TRESH window, so added cost (compared to previous
AND operation) is near zero.
dump operation gives current max_p value in a new TCA_RED_MAX_P
attribute.
Example on a 10Mbit link :
tc qdisc add dev $DEV parent 1:1 handle 10: est 1sec 8sec red \
limit 400000 min 30000 max 90000 avpkt 1000 \
burst 55 ecn adaptative bandwidth 10Mbit
# tc -s -d qdisc show dev eth3
...
qdisc red 10: parent 1:1 limit 400000b min 30000b max 90000b ecn
adaptative ewma 5 max_p=0.113335 Scell_log 15
Sent 50414282 bytes 34504 pkt (dropped 35, overlimits 1392 requeues 0)
rate 9749Kbit 831pps backlog 72056b 16p requeues 0
marked 1357 early 35 pdrop 0 other 0
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-12-07 23:06:03 -07:00
|
|
|
struct timer_list adapt_timer;
|
2017-10-16 18:29:17 -06:00
|
|
|
struct Qdisc *sch;
|
2005-11-05 13:14:05 -07:00
|
|
|
struct red_parms parms;
|
2012-01-04 19:25:16 -07:00
|
|
|
struct red_vars vars;
|
2005-11-05 13:14:05 -07:00
|
|
|
struct red_stats stats;
|
2006-03-20 20:20:44 -07:00
|
|
|
struct Qdisc *qdisc;
|
2005-04-16 16:20:36 -06:00
|
|
|
};
|
|
|
|
|
2005-11-05 13:14:05 -07:00
|
|
|
static inline int red_use_ecn(struct red_sched_data *q)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2005-11-05 13:14:05 -07:00
|
|
|
return q->flags & TC_RED_ECN;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2005-11-05 13:14:28 -07:00
|
|
|
static inline int red_use_harddrop(struct red_sched_data *q)
|
|
|
|
{
|
|
|
|
return q->flags & TC_RED_HARDDROP;
|
|
|
|
}
|
|
|
|
|
2016-06-22 00:16:49 -06:00
|
|
|
static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch,
|
|
|
|
struct sk_buff **to_free)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
2006-03-20 20:20:44 -07:00
|
|
|
struct Qdisc *child = q->qdisc;
|
|
|
|
int ret;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2012-01-04 19:25:16 -07:00
|
|
|
q->vars.qavg = red_calc_qavg(&q->parms,
|
|
|
|
&q->vars,
|
|
|
|
child->qstats.backlog);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2012-01-04 19:25:16 -07:00
|
|
|
if (red_is_idling(&q->vars))
|
|
|
|
red_end_of_idle_period(&q->vars);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2012-01-04 19:25:16 -07:00
|
|
|
switch (red_action(&q->parms, &q->vars, q->vars.qavg)) {
|
2011-01-19 12:26:56 -07:00
|
|
|
case RED_DONT_MARK:
|
|
|
|
break;
|
|
|
|
|
|
|
|
case RED_PROB_MARK:
|
2014-09-28 12:53:29 -06:00
|
|
|
qdisc_qstats_overlimit(sch);
|
2011-01-19 12:26:56 -07:00
|
|
|
if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) {
|
|
|
|
q->stats.prob_drop++;
|
|
|
|
goto congestion_drop;
|
|
|
|
}
|
|
|
|
|
|
|
|
q->stats.prob_mark++;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case RED_HARD_MARK:
|
2014-09-28 12:53:29 -06:00
|
|
|
qdisc_qstats_overlimit(sch);
|
2011-01-19 12:26:56 -07:00
|
|
|
if (red_use_harddrop(q) || !red_use_ecn(q) ||
|
|
|
|
!INET_ECN_set_ce(skb)) {
|
|
|
|
q->stats.forced_drop++;
|
|
|
|
goto congestion_drop;
|
|
|
|
}
|
|
|
|
|
|
|
|
q->stats.forced_mark++;
|
|
|
|
break;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2016-06-22 00:16:49 -06:00
|
|
|
ret = qdisc_enqueue(skb, child, to_free);
|
2006-03-20 20:20:44 -07:00
|
|
|
if (likely(ret == NET_XMIT_SUCCESS)) {
|
2016-06-01 17:15:18 -06:00
|
|
|
qdisc_qstats_backlog_inc(sch, skb);
|
2006-03-20 20:20:44 -07:00
|
|
|
sch->q.qlen++;
|
2008-08-04 23:31:03 -06:00
|
|
|
} else if (net_xmit_drop_count(ret)) {
|
2006-03-20 20:20:44 -07:00
|
|
|
q->stats.pdrop++;
|
2014-09-28 12:53:29 -06:00
|
|
|
qdisc_qstats_drop(sch);
|
2006-03-20 20:20:44 -07:00
|
|
|
}
|
|
|
|
return ret;
|
2005-11-05 13:14:05 -07:00
|
|
|
|
|
|
|
congestion_drop:
|
2016-06-22 00:16:49 -06:00
|
|
|
qdisc_drop(skb, sch, to_free);
|
2005-04-16 16:20:36 -06:00
|
|
|
return NET_XMIT_CN;
|
|
|
|
}
|
|
|
|
|
2011-01-19 12:26:56 -07:00
|
|
|
static struct sk_buff *red_dequeue(struct Qdisc *sch)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
struct sk_buff *skb;
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
2006-03-20 20:20:44 -07:00
|
|
|
struct Qdisc *child = q->qdisc;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-03-20 20:20:44 -07:00
|
|
|
skb = child->dequeue(child);
|
2011-01-21 00:31:33 -07:00
|
|
|
if (skb) {
|
|
|
|
qdisc_bstats_update(sch, skb);
|
2016-06-01 17:15:18 -06:00
|
|
|
qdisc_qstats_backlog_dec(sch, skb);
|
2006-03-20 20:20:44 -07:00
|
|
|
sch->q.qlen--;
|
2011-01-21 00:31:33 -07:00
|
|
|
} else {
|
2012-01-04 19:25:16 -07:00
|
|
|
if (!red_is_idling(&q->vars))
|
|
|
|
red_start_of_idle_period(&q->vars);
|
2011-01-21 00:31:33 -07:00
|
|
|
}
|
2005-11-05 13:14:06 -07:00
|
|
|
return skb;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2011-01-19 12:26:56 -07:00
|
|
|
static struct sk_buff *red_peek(struct Qdisc *sch)
|
2008-10-31 01:45:55 -06:00
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
|
|
|
struct Qdisc *child = q->qdisc;
|
|
|
|
|
|
|
|
return child->ops->peek(child);
|
|
|
|
}
|
|
|
|
|
2011-01-19 12:26:56 -07:00
|
|
|
static void red_reset(struct Qdisc *sch)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
|
|
|
|
2006-03-20 20:20:44 -07:00
|
|
|
qdisc_reset(q->qdisc);
|
2016-06-01 17:15:18 -06:00
|
|
|
sch->qstats.backlog = 0;
|
2006-03-20 20:20:44 -07:00
|
|
|
sch->q.qlen = 0;
|
2012-01-04 19:25:16 -07:00
|
|
|
red_restart(&q->vars);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2017-11-05 23:23:41 -07:00
|
|
|
static int red_offload(struct Qdisc *sch, bool enable)
|
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
|
|
|
struct net_device *dev = qdisc_dev(sch);
|
|
|
|
struct tc_red_qopt_offload opt = {
|
|
|
|
.handle = sch->handle,
|
|
|
|
.parent = sch->parent,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
|
|
|
if (enable) {
|
|
|
|
opt.command = TC_RED_REPLACE;
|
|
|
|
opt.set.min = q->parms.qth_min >> q->parms.Wlog;
|
|
|
|
opt.set.max = q->parms.qth_max >> q->parms.Wlog;
|
|
|
|
opt.set.probability = q->parms.max_P;
|
|
|
|
opt.set.is_ecn = red_use_ecn(q);
|
2018-01-14 21:01:26 -07:00
|
|
|
opt.set.qstats = &sch->qstats;
|
2017-11-05 23:23:41 -07:00
|
|
|
} else {
|
|
|
|
opt.command = TC_RED_DESTROY;
|
|
|
|
}
|
|
|
|
|
2017-12-25 01:51:41 -07:00
|
|
|
return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt);
|
2017-11-05 23:23:41 -07:00
|
|
|
}
|
|
|
|
|
2006-03-20 20:20:44 -07:00
|
|
|
static void red_destroy(struct Qdisc *sch)
|
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
sch_red: Adaptative RED AQM
Adaptative RED AQM for linux, based on paper from Sally FLoyd,
Ramakrishna Gummadi, and Scott Shenker, August 2001 :
http://icir.org/floyd/papers/adaptiveRed.pdf
Goal of Adaptative RED is to make max_p a dynamic value between 1% and
50% to reach the target average queue : (max_th - min_th) / 2
Every 500 ms:
if (avg > target and max_p <= 0.5)
increase max_p : max_p += alpha;
else if (avg < target and max_p >= 0.01)
decrease max_p : max_p *= beta;
target :[min_th + 0.4*(min_th - max_th),
min_th + 0.6*(min_th - max_th)].
alpha : min(0.01, max_p / 4)
beta : 0.9
max_P is a Q0.32 fixed point number (unsigned, with 32 bits mantissa)
Changes against our RED implementation are :
max_p is no longer a negative power of two (1/(2^Plog)), but a Q0.32
fixed point number, to allow full range described in Adatative paper.
To deliver a random number, we now use a reciprocal divide (thats really
a multiply), but this operation is done once per marked/droped packet
when in RED_BETWEEN_TRESH window, so added cost (compared to previous
AND operation) is near zero.
dump operation gives current max_p value in a new TCA_RED_MAX_P
attribute.
Example on a 10Mbit link :
tc qdisc add dev $DEV parent 1:1 handle 10: est 1sec 8sec red \
limit 400000 min 30000 max 90000 avpkt 1000 \
burst 55 ecn adaptative bandwidth 10Mbit
# tc -s -d qdisc show dev eth3
...
qdisc red 10: parent 1:1 limit 400000b min 30000b max 90000b ecn
adaptative ewma 5 max_p=0.113335 Scell_log 15
Sent 50414282 bytes 34504 pkt (dropped 35, overlimits 1392 requeues 0)
rate 9749Kbit 831pps backlog 72056b 16p requeues 0
marked 1357 early 35 pdrop 0 other 0
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-12-07 23:06:03 -07:00
|
|
|
|
|
|
|
del_timer_sync(&q->adapt_timer);
|
2017-11-05 23:23:41 -07:00
|
|
|
red_offload(sch, false);
|
2006-03-20 20:20:44 -07:00
|
|
|
qdisc_destroy(q->qdisc);
|
|
|
|
}
|
|
|
|
|
2008-01-23 21:35:39 -07:00
|
|
|
static const struct nla_policy red_policy[TCA_RED_MAX + 1] = {
|
|
|
|
[TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) },
|
|
|
|
[TCA_RED_STAB] = { .len = RED_STAB_SIZE },
|
2011-12-08 19:46:45 -07:00
|
|
|
[TCA_RED_MAX_P] = { .type = NLA_U32 },
|
2008-01-23 21:35:39 -07:00
|
|
|
};
|
|
|
|
|
2017-12-20 10:35:14 -07:00
|
|
|
static int red_change(struct Qdisc *sch, struct nlattr *opt,
|
|
|
|
struct netlink_ext_ack *extack)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
2008-01-22 23:11:17 -07:00
|
|
|
struct nlattr *tb[TCA_RED_MAX + 1];
|
2005-04-16 16:20:36 -06:00
|
|
|
struct tc_red_qopt *ctl;
|
2006-03-20 20:20:44 -07:00
|
|
|
struct Qdisc *child = NULL;
|
2008-01-23 21:33:32 -07:00
|
|
|
int err;
|
2011-12-08 19:46:45 -07:00
|
|
|
u32 max_P;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-01-23 21:33:32 -07:00
|
|
|
if (opt == NULL)
|
2005-11-05 13:14:08 -07:00
|
|
|
return -EINVAL;
|
|
|
|
|
2017-04-12 06:34:07 -06:00
|
|
|
err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL);
|
2008-01-23 21:33:32 -07:00
|
|
|
if (err < 0)
|
|
|
|
return err;
|
|
|
|
|
2008-01-22 23:11:17 -07:00
|
|
|
if (tb[TCA_RED_PARMS] == NULL ||
|
2008-01-23 21:35:39 -07:00
|
|
|
tb[TCA_RED_STAB] == NULL)
|
2005-04-16 16:20:36 -06:00
|
|
|
return -EINVAL;
|
|
|
|
|
2011-12-08 19:46:45 -07:00
|
|
|
max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0;
|
|
|
|
|
2008-01-22 23:11:17 -07:00
|
|
|
ctl = nla_data(tb[TCA_RED_PARMS]);
|
2017-12-04 04:31:11 -07:00
|
|
|
if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog))
|
|
|
|
return -EINVAL;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2006-03-20 20:20:44 -07:00
|
|
|
if (ctl->limit > 0) {
|
2017-12-20 10:35:21 -07:00
|
|
|
child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit,
|
|
|
|
extack);
|
2008-07-06 00:40:21 -06:00
|
|
|
if (IS_ERR(child))
|
|
|
|
return PTR_ERR(child);
|
2006-03-20 20:20:44 -07:00
|
|
|
|
2018-05-18 06:51:44 -06:00
|
|
|
/* child is fifo, no need to check for noop_qdisc */
|
2017-03-08 08:03:32 -07:00
|
|
|
qdisc_hash_add(child, true);
|
2018-05-18 06:51:44 -06:00
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
sch_tree_lock(sch);
|
|
|
|
q->flags = ctl->flags;
|
|
|
|
q->limit = ctl->limit;
|
2006-11-29 18:36:20 -07:00
|
|
|
if (child) {
|
2016-02-25 15:55:01 -07:00
|
|
|
qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen,
|
|
|
|
q->qdisc->qstats.backlog);
|
2008-11-20 05:11:36 -07:00
|
|
|
qdisc_destroy(q->qdisc);
|
|
|
|
q->qdisc = child;
|
2006-11-29 18:36:20 -07:00
|
|
|
}
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2012-01-04 19:25:16 -07:00
|
|
|
red_set_parms(&q->parms,
|
|
|
|
ctl->qth_min, ctl->qth_max, ctl->Wlog,
|
2011-12-08 19:46:45 -07:00
|
|
|
ctl->Plog, ctl->Scell_log,
|
|
|
|
nla_data(tb[TCA_RED_STAB]),
|
|
|
|
max_P);
|
2012-01-04 19:25:16 -07:00
|
|
|
red_set_vars(&q->vars);
|
2005-11-05 13:14:05 -07:00
|
|
|
|
sch_red: Adaptative RED AQM
Adaptative RED AQM for linux, based on paper from Sally FLoyd,
Ramakrishna Gummadi, and Scott Shenker, August 2001 :
http://icir.org/floyd/papers/adaptiveRed.pdf
Goal of Adaptative RED is to make max_p a dynamic value between 1% and
50% to reach the target average queue : (max_th - min_th) / 2
Every 500 ms:
if (avg > target and max_p <= 0.5)
increase max_p : max_p += alpha;
else if (avg < target and max_p >= 0.01)
decrease max_p : max_p *= beta;
target :[min_th + 0.4*(min_th - max_th),
min_th + 0.6*(min_th - max_th)].
alpha : min(0.01, max_p / 4)
beta : 0.9
max_P is a Q0.32 fixed point number (unsigned, with 32 bits mantissa)
Changes against our RED implementation are :
max_p is no longer a negative power of two (1/(2^Plog)), but a Q0.32
fixed point number, to allow full range described in Adatative paper.
To deliver a random number, we now use a reciprocal divide (thats really
a multiply), but this operation is done once per marked/droped packet
when in RED_BETWEEN_TRESH window, so added cost (compared to previous
AND operation) is near zero.
dump operation gives current max_p value in a new TCA_RED_MAX_P
attribute.
Example on a 10Mbit link :
tc qdisc add dev $DEV parent 1:1 handle 10: est 1sec 8sec red \
limit 400000 min 30000 max 90000 avpkt 1000 \
burst 55 ecn adaptative bandwidth 10Mbit
# tc -s -d qdisc show dev eth3
...
qdisc red 10: parent 1:1 limit 400000b min 30000b max 90000b ecn
adaptative ewma 5 max_p=0.113335 Scell_log 15
Sent 50414282 bytes 34504 pkt (dropped 35, overlimits 1392 requeues 0)
rate 9749Kbit 831pps backlog 72056b 16p requeues 0
marked 1357 early 35 pdrop 0 other 0
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-12-07 23:06:03 -07:00
|
|
|
del_timer(&q->adapt_timer);
|
|
|
|
if (ctl->flags & TC_RED_ADAPTATIVE)
|
|
|
|
mod_timer(&q->adapt_timer, jiffies + HZ/2);
|
|
|
|
|
sch_red: fix red_change
Le mercredi 30 novembre 2011 à 14:36 -0800, Stephen Hemminger a écrit :
> (Almost) nobody uses RED because they can't figure it out.
> According to Wikipedia, VJ says that:
> "there are not one, but two bugs in classic RED."
RED is useful for high throughput routers, I doubt many linux machines
act as such devices.
I was considering adding Adaptative RED (Sally Floyd, Ramakrishna
Gummadi, Scott Shender), August 2001
In this version, maxp is dynamic (from 1% to 50%), and user only have to
setup min_th (target average queue size)
(max_th and wq (burst in linux RED) are automatically setup)
By the way it seems we have a small bug in red_change()
if (skb_queue_empty(&sch->q))
red_end_of_idle_period(&q->parms);
First, if queue is empty, we should call
red_start_of_idle_period(&q->parms);
Second, since we dont use anymore sch->q, but q->qdisc, the test is
meaningless.
Oh well...
[PATCH] sch_red: fix red_change()
Now RED is classful, we must check q->qdisc->q.qlen, and if queue is empty,
we start an idle period, not end it.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-12-01 04:06:34 -07:00
|
|
|
if (!q->qdisc->q.qlen)
|
2012-01-04 19:25:16 -07:00
|
|
|
red_start_of_idle_period(&q->vars);
|
2005-11-05 13:14:08 -07:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
sch_tree_unlock(sch);
|
2017-11-05 23:23:41 -07:00
|
|
|
red_offload(sch, true);
|
2005-04-16 16:20:36 -06:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2017-10-16 18:29:17 -06:00
|
|
|
static inline void red_adaptative_timer(struct timer_list *t)
|
sch_red: Adaptative RED AQM
Adaptative RED AQM for linux, based on paper from Sally FLoyd,
Ramakrishna Gummadi, and Scott Shenker, August 2001 :
http://icir.org/floyd/papers/adaptiveRed.pdf
Goal of Adaptative RED is to make max_p a dynamic value between 1% and
50% to reach the target average queue : (max_th - min_th) / 2
Every 500 ms:
if (avg > target and max_p <= 0.5)
increase max_p : max_p += alpha;
else if (avg < target and max_p >= 0.01)
decrease max_p : max_p *= beta;
target :[min_th + 0.4*(min_th - max_th),
min_th + 0.6*(min_th - max_th)].
alpha : min(0.01, max_p / 4)
beta : 0.9
max_P is a Q0.32 fixed point number (unsigned, with 32 bits mantissa)
Changes against our RED implementation are :
max_p is no longer a negative power of two (1/(2^Plog)), but a Q0.32
fixed point number, to allow full range described in Adatative paper.
To deliver a random number, we now use a reciprocal divide (thats really
a multiply), but this operation is done once per marked/droped packet
when in RED_BETWEEN_TRESH window, so added cost (compared to previous
AND operation) is near zero.
dump operation gives current max_p value in a new TCA_RED_MAX_P
attribute.
Example on a 10Mbit link :
tc qdisc add dev $DEV parent 1:1 handle 10: est 1sec 8sec red \
limit 400000 min 30000 max 90000 avpkt 1000 \
burst 55 ecn adaptative bandwidth 10Mbit
# tc -s -d qdisc show dev eth3
...
qdisc red 10: parent 1:1 limit 400000b min 30000b max 90000b ecn
adaptative ewma 5 max_p=0.113335 Scell_log 15
Sent 50414282 bytes 34504 pkt (dropped 35, overlimits 1392 requeues 0)
rate 9749Kbit 831pps backlog 72056b 16p requeues 0
marked 1357 early 35 pdrop 0 other 0
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-12-07 23:06:03 -07:00
|
|
|
{
|
2017-10-16 18:29:17 -06:00
|
|
|
struct red_sched_data *q = from_timer(q, t, adapt_timer);
|
|
|
|
struct Qdisc *sch = q->sch;
|
sch_red: Adaptative RED AQM
Adaptative RED AQM for linux, based on paper from Sally FLoyd,
Ramakrishna Gummadi, and Scott Shenker, August 2001 :
http://icir.org/floyd/papers/adaptiveRed.pdf
Goal of Adaptative RED is to make max_p a dynamic value between 1% and
50% to reach the target average queue : (max_th - min_th) / 2
Every 500 ms:
if (avg > target and max_p <= 0.5)
increase max_p : max_p += alpha;
else if (avg < target and max_p >= 0.01)
decrease max_p : max_p *= beta;
target :[min_th + 0.4*(min_th - max_th),
min_th + 0.6*(min_th - max_th)].
alpha : min(0.01, max_p / 4)
beta : 0.9
max_P is a Q0.32 fixed point number (unsigned, with 32 bits mantissa)
Changes against our RED implementation are :
max_p is no longer a negative power of two (1/(2^Plog)), but a Q0.32
fixed point number, to allow full range described in Adatative paper.
To deliver a random number, we now use a reciprocal divide (thats really
a multiply), but this operation is done once per marked/droped packet
when in RED_BETWEEN_TRESH window, so added cost (compared to previous
AND operation) is near zero.
dump operation gives current max_p value in a new TCA_RED_MAX_P
attribute.
Example on a 10Mbit link :
tc qdisc add dev $DEV parent 1:1 handle 10: est 1sec 8sec red \
limit 400000 min 30000 max 90000 avpkt 1000 \
burst 55 ecn adaptative bandwidth 10Mbit
# tc -s -d qdisc show dev eth3
...
qdisc red 10: parent 1:1 limit 400000b min 30000b max 90000b ecn
adaptative ewma 5 max_p=0.113335 Scell_log 15
Sent 50414282 bytes 34504 pkt (dropped 35, overlimits 1392 requeues 0)
rate 9749Kbit 831pps backlog 72056b 16p requeues 0
marked 1357 early 35 pdrop 0 other 0
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-12-07 23:06:03 -07:00
|
|
|
spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
|
|
|
|
|
|
|
|
spin_lock(root_lock);
|
2012-01-04 19:25:16 -07:00
|
|
|
red_adaptative_algo(&q->parms, &q->vars);
|
sch_red: Adaptative RED AQM
Adaptative RED AQM for linux, based on paper from Sally FLoyd,
Ramakrishna Gummadi, and Scott Shenker, August 2001 :
http://icir.org/floyd/papers/adaptiveRed.pdf
Goal of Adaptative RED is to make max_p a dynamic value between 1% and
50% to reach the target average queue : (max_th - min_th) / 2
Every 500 ms:
if (avg > target and max_p <= 0.5)
increase max_p : max_p += alpha;
else if (avg < target and max_p >= 0.01)
decrease max_p : max_p *= beta;
target :[min_th + 0.4*(min_th - max_th),
min_th + 0.6*(min_th - max_th)].
alpha : min(0.01, max_p / 4)
beta : 0.9
max_P is a Q0.32 fixed point number (unsigned, with 32 bits mantissa)
Changes against our RED implementation are :
max_p is no longer a negative power of two (1/(2^Plog)), but a Q0.32
fixed point number, to allow full range described in Adatative paper.
To deliver a random number, we now use a reciprocal divide (thats really
a multiply), but this operation is done once per marked/droped packet
when in RED_BETWEEN_TRESH window, so added cost (compared to previous
AND operation) is near zero.
dump operation gives current max_p value in a new TCA_RED_MAX_P
attribute.
Example on a 10Mbit link :
tc qdisc add dev $DEV parent 1:1 handle 10: est 1sec 8sec red \
limit 400000 min 30000 max 90000 avpkt 1000 \
burst 55 ecn adaptative bandwidth 10Mbit
# tc -s -d qdisc show dev eth3
...
qdisc red 10: parent 1:1 limit 400000b min 30000b max 90000b ecn
adaptative ewma 5 max_p=0.113335 Scell_log 15
Sent 50414282 bytes 34504 pkt (dropped 35, overlimits 1392 requeues 0)
rate 9749Kbit 831pps backlog 72056b 16p requeues 0
marked 1357 early 35 pdrop 0 other 0
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-12-07 23:06:03 -07:00
|
|
|
mod_timer(&q->adapt_timer, jiffies + HZ/2);
|
|
|
|
spin_unlock(root_lock);
|
|
|
|
}
|
|
|
|
|
2017-12-20 10:35:13 -07:00
|
|
|
static int red_init(struct Qdisc *sch, struct nlattr *opt,
|
|
|
|
struct netlink_ext_ack *extack)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
2006-03-20 20:20:44 -07:00
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
|
|
|
|
|
|
|
q->qdisc = &noop_qdisc;
|
2017-10-16 18:29:17 -06:00
|
|
|
q->sch = sch;
|
|
|
|
timer_setup(&q->adapt_timer, red_adaptative_timer, 0);
|
2017-12-20 10:35:14 -07:00
|
|
|
return red_change(sch, opt, extack);
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2017-12-14 06:54:30 -07:00
|
|
|
static int red_dump_offload_stats(struct Qdisc *sch, struct tc_red_qopt *opt)
|
2017-11-05 23:23:41 -07:00
|
|
|
{
|
|
|
|
struct net_device *dev = qdisc_dev(sch);
|
|
|
|
struct tc_red_qopt_offload hw_stats = {
|
2017-11-10 16:09:53 -07:00
|
|
|
.command = TC_RED_STATS,
|
2017-11-05 23:23:41 -07:00
|
|
|
.handle = sch->handle,
|
|
|
|
.parent = sch->parent,
|
2017-11-10 16:09:53 -07:00
|
|
|
{
|
|
|
|
.stats.bstats = &sch->bstats,
|
|
|
|
.stats.qstats = &sch->qstats,
|
|
|
|
},
|
2017-11-05 23:23:41 -07:00
|
|
|
};
|
2017-12-25 01:51:41 -07:00
|
|
|
int err;
|
|
|
|
|
|
|
|
sch->flags &= ~TCQ_F_OFFLOADED;
|
2017-11-05 23:23:41 -07:00
|
|
|
|
2017-12-25 01:51:41 -07:00
|
|
|
if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED,
|
|
|
|
&hw_stats);
|
|
|
|
if (err == -EOPNOTSUPP)
|
2017-11-05 23:23:41 -07:00
|
|
|
return 0;
|
|
|
|
|
2017-12-25 01:51:41 -07:00
|
|
|
if (!err)
|
|
|
|
sch->flags |= TCQ_F_OFFLOADED;
|
|
|
|
|
|
|
|
return err;
|
2017-11-05 23:23:41 -07:00
|
|
|
}
|
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
2008-01-22 23:11:17 -07:00
|
|
|
struct nlattr *opts = NULL;
|
2005-11-05 13:14:05 -07:00
|
|
|
struct tc_red_qopt opt = {
|
|
|
|
.limit = q->limit,
|
|
|
|
.flags = q->flags,
|
|
|
|
.qth_min = q->parms.qth_min >> q->parms.Wlog,
|
|
|
|
.qth_max = q->parms.qth_max >> q->parms.Wlog,
|
|
|
|
.Wlog = q->parms.Wlog,
|
|
|
|
.Plog = q->parms.Plog,
|
|
|
|
.Scell_log = q->parms.Scell_log,
|
|
|
|
};
|
2017-11-05 23:23:41 -07:00
|
|
|
int err;
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2017-12-14 06:54:30 -07:00
|
|
|
err = red_dump_offload_stats(sch, &opt);
|
2017-11-05 23:23:41 -07:00
|
|
|
if (err)
|
|
|
|
goto nla_put_failure;
|
|
|
|
|
2008-01-22 23:11:17 -07:00
|
|
|
opts = nla_nest_start(skb, TCA_OPTIONS);
|
|
|
|
if (opts == NULL)
|
|
|
|
goto nla_put_failure;
|
2012-03-29 03:11:39 -06:00
|
|
|
if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) ||
|
|
|
|
nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P))
|
|
|
|
goto nla_put_failure;
|
2008-01-22 23:11:17 -07:00
|
|
|
return nla_nest_end(skb, opts);
|
2005-04-16 16:20:36 -06:00
|
|
|
|
2008-01-22 23:11:17 -07:00
|
|
|
nla_put_failure:
|
2008-06-03 17:36:54 -06:00
|
|
|
nla_nest_cancel(skb, opts);
|
|
|
|
return -EMSGSIZE;
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
|
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
2017-11-05 23:23:41 -07:00
|
|
|
struct net_device *dev = qdisc_dev(sch);
|
2018-01-10 06:59:59 -07:00
|
|
|
struct tc_red_xstats st = {0};
|
2005-11-05 13:14:05 -07:00
|
|
|
|
2017-12-14 06:54:30 -07:00
|
|
|
if (sch->flags & TCQ_F_OFFLOADED) {
|
2017-11-05 23:23:41 -07:00
|
|
|
struct tc_red_qopt_offload hw_stats_request = {
|
2017-11-10 16:09:53 -07:00
|
|
|
.command = TC_RED_XSTATS,
|
2017-11-05 23:23:41 -07:00
|
|
|
.handle = sch->handle,
|
|
|
|
.parent = sch->parent,
|
2017-11-10 16:09:53 -07:00
|
|
|
{
|
2018-01-10 06:59:59 -07:00
|
|
|
.xstats = &q->stats,
|
2017-11-10 16:09:53 -07:00
|
|
|
},
|
2017-11-05 23:23:41 -07:00
|
|
|
};
|
2018-01-10 06:59:59 -07:00
|
|
|
dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED,
|
|
|
|
&hw_stats_request);
|
2017-11-05 23:23:41 -07:00
|
|
|
}
|
2018-01-10 06:59:59 -07:00
|
|
|
st.early = q->stats.prob_drop + q->stats.forced_drop;
|
|
|
|
st.pdrop = q->stats.pdrop;
|
|
|
|
st.other = q->stats.other;
|
|
|
|
st.marked = q->stats.prob_mark + q->stats.forced_mark;
|
2017-11-05 23:23:41 -07:00
|
|
|
|
2005-11-05 13:14:05 -07:00
|
|
|
return gnet_stats_copy_app(d, &st, sizeof(st));
|
2005-04-16 16:20:36 -06:00
|
|
|
}
|
|
|
|
|
2006-03-20 20:20:44 -07:00
|
|
|
static int red_dump_class(struct Qdisc *sch, unsigned long cl,
|
|
|
|
struct sk_buff *skb, struct tcmsg *tcm)
|
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
|
|
|
|
|
|
|
tcm->tcm_handle |= TC_H_MIN(1);
|
|
|
|
tcm->tcm_info = q->qdisc->handle;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
|
2017-12-20 10:35:17 -07:00
|
|
|
struct Qdisc **old, struct netlink_ext_ack *extack)
|
2006-03-20 20:20:44 -07:00
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
|
|
|
|
|
|
|
if (new == NULL)
|
|
|
|
new = &noop_qdisc;
|
|
|
|
|
2016-02-25 15:55:00 -07:00
|
|
|
*old = qdisc_replace(sch, new, &q->qdisc);
|
2006-03-20 20:20:44 -07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
|
|
|
|
{
|
|
|
|
struct red_sched_data *q = qdisc_priv(sch);
|
|
|
|
return q->qdisc;
|
|
|
|
}
|
|
|
|
|
net_sched: remove tc class reference counting
For TC classes, their ->get() and ->put() are always paired, and the
reference counting is completely useless, because:
1) For class modification and dumping paths, we already hold RTNL lock,
so all of these ->get(),->change(),->put() are atomic.
2) For filter bindiing/unbinding, we use other reference counter than
this one, and they should have RTNL lock too.
3) For ->qlen_notify(), it is special because it is called on ->enqueue()
path, but we already hold qdisc tree lock there, and we hold this
tree lock when graft or delete the class too, so it should not be gone
or changed until we release the tree lock.
Therefore, this patch removes ->get() and ->put(), but:
1) Adds a new ->find() to find the pointer to a class by classid, no
refcnt.
2) Move the original class destroy upon the last refcnt into ->delete(),
right after releasing tree lock. This is fine because the class is
already removed from hash when holding the lock.
For those who also use ->put() as ->unbind(), just rename them to reflect
this change.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-24 17:51:29 -06:00
|
|
|
static unsigned long red_find(struct Qdisc *sch, u32 classid)
|
2006-03-20 20:20:44 -07:00
|
|
|
{
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker)
|
|
|
|
{
|
|
|
|
if (!walker->stop) {
|
|
|
|
if (walker->count >= walker->skip)
|
|
|
|
if (walker->fn(sch, 1, walker) < 0) {
|
|
|
|
walker->stop = 1;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
walker->count++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-11-14 02:44:41 -07:00
|
|
|
static const struct Qdisc_class_ops red_class_ops = {
|
2006-03-20 20:20:44 -07:00
|
|
|
.graft = red_graft,
|
|
|
|
.leaf = red_leaf,
|
net_sched: remove tc class reference counting
For TC classes, their ->get() and ->put() are always paired, and the
reference counting is completely useless, because:
1) For class modification and dumping paths, we already hold RTNL lock,
so all of these ->get(),->change(),->put() are atomic.
2) For filter bindiing/unbinding, we use other reference counter than
this one, and they should have RTNL lock too.
3) For ->qlen_notify(), it is special because it is called on ->enqueue()
path, but we already hold qdisc tree lock there, and we hold this
tree lock when graft or delete the class too, so it should not be gone
or changed until we release the tree lock.
Therefore, this patch removes ->get() and ->put(), but:
1) Adds a new ->find() to find the pointer to a class by classid, no
refcnt.
2) Move the original class destroy upon the last refcnt into ->delete(),
right after releasing tree lock. This is fine because the class is
already removed from hash when holding the lock.
For those who also use ->put() as ->unbind(), just rename them to reflect
this change.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-24 17:51:29 -06:00
|
|
|
.find = red_find,
|
2006-03-20 20:20:44 -07:00
|
|
|
.walk = red_walk,
|
|
|
|
.dump = red_dump_class,
|
|
|
|
};
|
|
|
|
|
2007-11-14 02:44:41 -07:00
|
|
|
static struct Qdisc_ops red_qdisc_ops __read_mostly = {
|
2005-04-16 16:20:36 -06:00
|
|
|
.id = "red",
|
|
|
|
.priv_size = sizeof(struct red_sched_data),
|
2006-03-20 20:20:44 -07:00
|
|
|
.cl_ops = &red_class_ops,
|
2005-04-16 16:20:36 -06:00
|
|
|
.enqueue = red_enqueue,
|
|
|
|
.dequeue = red_dequeue,
|
2008-10-31 01:45:55 -06:00
|
|
|
.peek = red_peek,
|
2005-04-16 16:20:36 -06:00
|
|
|
.init = red_init,
|
|
|
|
.reset = red_reset,
|
2006-03-20 20:20:44 -07:00
|
|
|
.destroy = red_destroy,
|
2005-04-16 16:20:36 -06:00
|
|
|
.change = red_change,
|
|
|
|
.dump = red_dump,
|
|
|
|
.dump_stats = red_dump_stats,
|
|
|
|
.owner = THIS_MODULE,
|
|
|
|
};
|
|
|
|
|
|
|
|
static int __init red_module_init(void)
|
|
|
|
{
|
|
|
|
return register_qdisc(&red_qdisc_ops);
|
|
|
|
}
|
2005-11-05 13:14:08 -07:00
|
|
|
|
|
|
|
static void __exit red_module_exit(void)
|
2005-04-16 16:20:36 -06:00
|
|
|
{
|
|
|
|
unregister_qdisc(&red_qdisc_ops);
|
|
|
|
}
|
2005-11-05 13:14:08 -07:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
module_init(red_module_init)
|
|
|
|
module_exit(red_module_exit)
|
2005-11-05 13:14:08 -07:00
|
|
|
|
2005-04-16 16:20:36 -06:00
|
|
|
MODULE_LICENSE("GPL");
|