linux/net/sched/sch_red.c

461 lines
11 KiB
C
Raw Normal View History

/*
* 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:
* J Hadi Salim 980914: computation fixes
* Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
* J Hadi Salim 980816: ECN support
*/
#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/skbuff.h>
#include <net/pkt_sched.h>
#include <net/pkt_cls.h>
#include <net/inet_ecn.h>
#include <net/red.h>
/* Parameters, settable by user:
-----------------------------
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.
*/
struct red_sched_data {
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>
2011-12-08 14:06:03 +08:00
struct timer_list adapt_timer;
struct Qdisc *sch;
struct red_parms parms;
struct red_vars vars;
struct red_stats stats;
struct Qdisc *qdisc;
};
static inline int red_use_ecn(struct red_sched_data *q)
{
return q->flags & TC_RED_ECN;
}
static inline int red_use_harddrop(struct red_sched_data *q)
{
return q->flags & TC_RED_HARDDROP;
}
static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch,
struct sk_buff **to_free)
{
struct red_sched_data *q = qdisc_priv(sch);
struct Qdisc *child = q->qdisc;
int ret;
q->vars.qavg = red_calc_qavg(&q->parms,
&q->vars,
child->qstats.backlog);
if (red_is_idling(&q->vars))
red_end_of_idle_period(&q->vars);
switch (red_action(&q->parms, &q->vars, q->vars.qavg)) {
case RED_DONT_MARK:
break;
case RED_PROB_MARK:
qdisc_qstats_overlimit(sch);
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:
qdisc_qstats_overlimit(sch);
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;
}
ret = qdisc_enqueue(skb, child, to_free);
if (likely(ret == NET_XMIT_SUCCESS)) {
qdisc_qstats_backlog_inc(sch, skb);
sch->q.qlen++;
} else if (net_xmit_drop_count(ret)) {
q->stats.pdrop++;
qdisc_qstats_drop(sch);
}
return ret;
congestion_drop:
qdisc_drop(skb, sch, to_free);
return NET_XMIT_CN;
}
static struct sk_buff *red_dequeue(struct Qdisc *sch)
{
struct sk_buff *skb;
struct red_sched_data *q = qdisc_priv(sch);
struct Qdisc *child = q->qdisc;
skb = child->dequeue(child);
if (skb) {
qdisc_bstats_update(sch, skb);
qdisc_qstats_backlog_dec(sch, skb);
sch->q.qlen--;
} else {
if (!red_is_idling(&q->vars))
red_start_of_idle_period(&q->vars);
}
return skb;
}
static struct sk_buff *red_peek(struct Qdisc *sch)
{
struct red_sched_data *q = qdisc_priv(sch);
struct Qdisc *child = q->qdisc;
return child->ops->peek(child);
}
static void red_reset(struct Qdisc *sch)
{
struct red_sched_data *q = qdisc_priv(sch);
qdisc_reset(q->qdisc);
sch->qstats.backlog = 0;
sch->q.qlen = 0;
red_restart(&q->vars);
}
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.limit = q->limit;
opt.set.is_ecn = red_use_ecn(q);
opt.set.is_harddrop = red_use_harddrop(q);
opt.set.qstats = &sch->qstats;
} else {
opt.command = TC_RED_DESTROY;
}
return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt);
}
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>
2011-12-08 14:06:03 +08:00
del_timer_sync(&q->adapt_timer);
red_offload(sch, false);
qdisc_put(q->qdisc);
}
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 },
[TCA_RED_MAX_P] = { .type = NLA_U32 },
};
static int red_change(struct Qdisc *sch, struct nlattr *opt,
struct netlink_ext_ack *extack)
{
struct Qdisc *old_child = NULL, *child = NULL;
struct red_sched_data *q = qdisc_priv(sch);
struct nlattr *tb[TCA_RED_MAX + 1];
struct tc_red_qopt *ctl;
int err;
u32 max_P;
if (opt == NULL)
return -EINVAL;
err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL);
if (err < 0)
return err;
if (tb[TCA_RED_PARMS] == NULL ||
tb[TCA_RED_STAB] == NULL)
return -EINVAL;
max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0;
ctl = nla_data(tb[TCA_RED_PARMS]);
if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog))
return -EINVAL;
if (ctl->limit > 0) {
child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit,
extack);
if (IS_ERR(child))
return PTR_ERR(child);
net: sched: red: avoid hashing NULL child Hangbin reported an Oops triggered by the syzkaller qdisc rules: kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN PTI Modules linked in: sch_red CPU: 0 PID: 28699 Comm: syz-executor5 Not tainted 4.17.0-rc4.kcov #1 Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 RIP: 0010:qdisc_hash_add+0x26/0xa0 RSP: 0018:ffff8800589cf470 EFLAGS: 00010203 RAX: dffffc0000000000 RBX: 0000000000000000 RCX: ffffffff824ad971 RDX: 0000000000000007 RSI: ffffc9000ce9f000 RDI: 000000000000003c RBP: 0000000000000001 R08: ffffed000b139ea2 R09: ffff8800589cf4f0 R10: ffff8800589cf50f R11: ffffed000b139ea2 R12: ffff880054019fc0 R13: ffff880054019fb4 R14: ffff88005c0af600 R15: ffff880054019fb0 FS: 00007fa6edcb1700(0000) GS:ffff88005ce00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000020000740 CR3: 000000000fc16000 CR4: 00000000000006f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: red_change+0x2d2/0xed0 [sch_red] qdisc_create+0x57e/0xef0 tc_modify_qdisc+0x47f/0x14e0 rtnetlink_rcv_msg+0x6a8/0x920 netlink_rcv_skb+0x2a2/0x3c0 netlink_unicast+0x511/0x740 netlink_sendmsg+0x825/0xc30 sock_sendmsg+0xc5/0x100 ___sys_sendmsg+0x778/0x8e0 __sys_sendmsg+0xf5/0x1b0 do_syscall_64+0xbd/0x3b0 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x450869 RSP: 002b:00007fa6edcb0c48 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: ffffffffffffffda RBX: 00007fa6edcb16b4 RCX: 0000000000450869 RDX: 0000000000000000 RSI: 00000000200000c0 RDI: 0000000000000013 RBP: 000000000072bea0 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff R13: 0000000000008778 R14: 0000000000702838 R15: 00007fa6edcb1700 Code: e9 0b fe ff ff 0f 1f 44 00 00 55 53 48 89 fb 89 f5 e8 3f 07 f3 fe 48 8d 7b 3c 48 b8 00 00 00 00 00 fc ff df 48 89 fa 48 c1 ea 03 <0f> b6 14 02 48 89 f8 83 e0 07 83 c0 03 38 d0 7c 04 84 d2 75 51 RIP: qdisc_hash_add+0x26/0xa0 RSP: ffff8800589cf470 When a red qdisc is updated with a 0 limit, the child qdisc is left unmodified, no additional scheduler is created in red_change(), the 'child' local variable is rightfully NULL and must not add it to the hash table. This change addresses the above issue moving qdisc_hash_add() right after the child qdisc creation. It additionally removes unneeded checks for noop_qdisc. Reported-by: Hangbin Liu <liuhangbin@gmail.com> Fixes: 49b499718fa1 ("net: sched: make default fifo qdiscs appear in the dump") Signed-off-by: Paolo Abeni <pabeni@redhat.com> Acked-by: Jiri Kosina <jkosina@suse.cz> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-05-18 20:51:44 +08:00
/* child is fifo, no need to check for noop_qdisc */
qdisc_hash_add(child, true);
net: sched: red: avoid hashing NULL child Hangbin reported an Oops triggered by the syzkaller qdisc rules: kasan: GPF could be caused by NULL-ptr deref or user memory access general protection fault: 0000 [#1] SMP KASAN PTI Modules linked in: sch_red CPU: 0 PID: 28699 Comm: syz-executor5 Not tainted 4.17.0-rc4.kcov #1 Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011 RIP: 0010:qdisc_hash_add+0x26/0xa0 RSP: 0018:ffff8800589cf470 EFLAGS: 00010203 RAX: dffffc0000000000 RBX: 0000000000000000 RCX: ffffffff824ad971 RDX: 0000000000000007 RSI: ffffc9000ce9f000 RDI: 000000000000003c RBP: 0000000000000001 R08: ffffed000b139ea2 R09: ffff8800589cf4f0 R10: ffff8800589cf50f R11: ffffed000b139ea2 R12: ffff880054019fc0 R13: ffff880054019fb4 R14: ffff88005c0af600 R15: ffff880054019fb0 FS: 00007fa6edcb1700(0000) GS:ffff88005ce00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000020000740 CR3: 000000000fc16000 CR4: 00000000000006f0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 Call Trace: red_change+0x2d2/0xed0 [sch_red] qdisc_create+0x57e/0xef0 tc_modify_qdisc+0x47f/0x14e0 rtnetlink_rcv_msg+0x6a8/0x920 netlink_rcv_skb+0x2a2/0x3c0 netlink_unicast+0x511/0x740 netlink_sendmsg+0x825/0xc30 sock_sendmsg+0xc5/0x100 ___sys_sendmsg+0x778/0x8e0 __sys_sendmsg+0xf5/0x1b0 do_syscall_64+0xbd/0x3b0 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x450869 RSP: 002b:00007fa6edcb0c48 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: ffffffffffffffda RBX: 00007fa6edcb16b4 RCX: 0000000000450869 RDX: 0000000000000000 RSI: 00000000200000c0 RDI: 0000000000000013 RBP: 000000000072bea0 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff R13: 0000000000008778 R14: 0000000000702838 R15: 00007fa6edcb1700 Code: e9 0b fe ff ff 0f 1f 44 00 00 55 53 48 89 fb 89 f5 e8 3f 07 f3 fe 48 8d 7b 3c 48 b8 00 00 00 00 00 fc ff df 48 89 fa 48 c1 ea 03 <0f> b6 14 02 48 89 f8 83 e0 07 83 c0 03 38 d0 7c 04 84 d2 75 51 RIP: qdisc_hash_add+0x26/0xa0 RSP: ffff8800589cf470 When a red qdisc is updated with a 0 limit, the child qdisc is left unmodified, no additional scheduler is created in red_change(), the 'child' local variable is rightfully NULL and must not add it to the hash table. This change addresses the above issue moving qdisc_hash_add() right after the child qdisc creation. It additionally removes unneeded checks for noop_qdisc. Reported-by: Hangbin Liu <liuhangbin@gmail.com> Fixes: 49b499718fa1 ("net: sched: make default fifo qdiscs appear in the dump") Signed-off-by: Paolo Abeni <pabeni@redhat.com> Acked-by: Jiri Kosina <jkosina@suse.cz> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-05-18 20:51:44 +08:00
}
sch_tree_lock(sch);
q->flags = ctl->flags;
q->limit = ctl->limit;
if (child) {
qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen,
q->qdisc->qstats.backlog);
old_child = q->qdisc;
q->qdisc = child;
}
red_set_parms(&q->parms,
ctl->qth_min, ctl->qth_max, ctl->Wlog,
ctl->Plog, ctl->Scell_log,
nla_data(tb[TCA_RED_STAB]),
max_P);
red_set_vars(&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>
2011-12-08 14:06:03 +08:00
del_timer(&q->adapt_timer);
if (ctl->flags & TC_RED_ADAPTATIVE)
mod_timer(&q->adapt_timer, jiffies + HZ/2);
if (!q->qdisc->q.qlen)
red_start_of_idle_period(&q->vars);
sch_tree_unlock(sch);
red_offload(sch, true);
if (old_child)
qdisc_put(old_child);
return 0;
}
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>
2011-12-08 14:06:03 +08:00
{
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>
2011-12-08 14:06:03 +08:00
spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
spin_lock(root_lock);
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>
2011-12-08 14:06:03 +08:00
mod_timer(&q->adapt_timer, jiffies + HZ/2);
spin_unlock(root_lock);
}
static int red_init(struct Qdisc *sch, struct nlattr *opt,
struct netlink_ext_ack *extack)
{
struct red_sched_data *q = qdisc_priv(sch);
q->qdisc = &noop_qdisc;
q->sch = sch;
timer_setup(&q->adapt_timer, red_adaptative_timer, 0);
return red_change(sch, opt, extack);
}
static int red_dump_offload_stats(struct Qdisc *sch)
{
struct tc_red_qopt_offload hw_stats = {
.command = TC_RED_STATS,
.handle = sch->handle,
.parent = sch->parent,
{
.stats.bstats = &sch->bstats,
.stats.qstats = &sch->qstats,
},
};
return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_RED, &hw_stats);
}
static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
{
struct red_sched_data *q = qdisc_priv(sch);
struct nlattr *opts = NULL;
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,
};
int err;
err = red_dump_offload_stats(sch);
if (err)
goto nla_put_failure;
opts = nla_nest_start(skb, TCA_OPTIONS);
if (opts == NULL)
goto nla_put_failure;
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;
return nla_nest_end(skb, opts);
nla_put_failure:
nla_nest_cancel(skb, opts);
return -EMSGSIZE;
}
static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
{
struct red_sched_data *q = qdisc_priv(sch);
struct net_device *dev = qdisc_dev(sch);
struct tc_red_xstats st = {0};
if (sch->flags & TCQ_F_OFFLOADED) {
struct tc_red_qopt_offload hw_stats_request = {
.command = TC_RED_XSTATS,
.handle = sch->handle,
.parent = sch->parent,
{
.xstats = &q->stats,
},
};
dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED,
&hw_stats_request);
}
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;
return gnet_stats_copy_app(d, &st, sizeof(st));
}
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 void red_graft_offload(struct Qdisc *sch,
struct Qdisc *new, struct Qdisc *old,
struct netlink_ext_ack *extack)
{
struct tc_red_qopt_offload graft_offload = {
.handle = sch->handle,
.parent = sch->parent,
.child_handle = new->handle,
.command = TC_RED_GRAFT,
};
qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, old,
TC_SETUP_QDISC_RED, &graft_offload, extack);
}
static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
struct Qdisc **old, struct netlink_ext_ack *extack)
{
struct red_sched_data *q = qdisc_priv(sch);
if (new == NULL)
new = &noop_qdisc;
*old = qdisc_replace(sch, new, &q->qdisc);
red_graft_offload(sch, new, *old, extack);
return 0;
}
static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
{
struct red_sched_data *q = qdisc_priv(sch);
return q->qdisc;
}
static unsigned long red_find(struct Qdisc *sch, u32 classid)
{
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++;
}
}
static const struct Qdisc_class_ops red_class_ops = {
.graft = red_graft,
.leaf = red_leaf,
.find = red_find,
.walk = red_walk,
.dump = red_dump_class,
};
static struct Qdisc_ops red_qdisc_ops __read_mostly = {
.id = "red",
.priv_size = sizeof(struct red_sched_data),
.cl_ops = &red_class_ops,
.enqueue = red_enqueue,
.dequeue = red_dequeue,
.peek = red_peek,
.init = red_init,
.reset = red_reset,
.destroy = red_destroy,
.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);
}
static void __exit red_module_exit(void)
{
unregister_qdisc(&red_qdisc_ops);
}
module_init(red_module_init)
module_exit(red_module_exit)
MODULE_LICENSE("GPL");