* Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
*
* Changes:
- * J Hadi Salim <hadi@nortel.com> 980914: computation fixes
+ * J Hadi Salim 980914: computation fixes
* Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
- * J Hadi Salim <hadi@nortelnetworks.com> 980816: ECN support
+ * J Hadi Salim 980816: ECN support
*/
-#include <linux/config.h>
#include <linux/module.h>
-#include <asm/uaccess.h>
-#include <asm/system.h>
-#include <linux/bitops.h>
#include <linux/types.h>
#include <linux/kernel.h>
-#include <linux/sched.h>
-#include <linux/string.h>
-#include <linux/mm.h>
-#include <linux/socket.h>
-#include <linux/sockios.h>
-#include <linux/in.h>
-#include <linux/errno.h>
-#include <linux/interrupt.h>
-#include <linux/if_ether.h>
-#include <linux/inet.h>
#include <linux/netdevice.h>
-#include <linux/etherdevice.h>
-#include <linux/notifier.h>
-#include <net/ip.h>
-#include <net/route.h>
#include <linux/skbuff.h>
-#include <net/sock.h>
#include <net/pkt_sched.h>
#include <net/inet_ecn.h>
-#include <net/dsfield.h>
+#include <net/red.h>
-/* Random Early Detection (RED) algorithm.
- =======================================
-
- Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways
- for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking.
-
- This file codes a "divisionless" version of RED algorithm
- as written down in Fig.17 of the paper.
-
-Short description.
-------------------
-
- When a new packet arrives we calculate the average queue length:
-
- avg = (1-W)*avg + W*current_queue_len,
-
- W is the filter time constant (chosen as 2^(-Wlog)), it controls
- the inertia of the algorithm. To allow larger bursts, W should be
- decreased.
-
- if (avg > th_max) -> packet marked (dropped).
- if (avg < th_min) -> packet passes.
- if (th_min < avg < th_max) we calculate probability:
-
- Pb = max_P * (avg - th_min)/(th_max-th_min)
-
- and mark (drop) packet with this probability.
- Pb changes from 0 (at avg==th_min) to max_P (avg==th_max).
- max_P should be small (not 1), usually 0.01..0.02 is good value.
-
- max_P is chosen as a number, so that max_P/(th_max-th_min)
- is a negative power of two in order arithmetics to contain
- only shifts.
-
-
- Parameters, settable by user:
+/* Parameters, settable by user:
-----------------------------
limit - bytes (must be > qth_max + burst)
arbitrarily high (well, less than ram size)
Really, this limit will never be reached
if RED works correctly.
-
- qth_min - bytes (should be < qth_max/2)
- qth_max - bytes (should be at least 2*qth_min and less limit)
- Wlog - bits (<32) log(1/W).
- Plog - bits (<32)
-
- Plog is related to max_P by formula:
-
- max_P = (qth_max-qth_min)/2^Plog;
-
- F.e. if qth_max=128K and qth_min=32K, then Plog=22
- corresponds to max_P=0.02
-
- Scell_log
- Stab
-
- Lookup table for log((1-W)^(t/t_ave).
-
-
-NOTES:
-
-Upper bound on W.
------------------
-
- If you want to allow bursts of L packets of size S,
- you should choose W:
-
- L + 1 - th_min/S < (1-(1-W)^L)/W
-
- th_min/S = 32 th_min/S = 4
-
- log(W) L
- -1 33
- -2 35
- -3 39
- -4 46
- -5 57
- -6 75
- -7 101
- -8 135
- -9 190
- etc.
*/
struct red_sched_data
{
-/* Parameters */
- u32 limit; /* HARD maximal queue length */
- u32 qth_min; /* Min average length threshold: A scaled */
- u32 qth_max; /* Max average length threshold: A scaled */
- u32 Rmask;
- u32 Scell_max;
- unsigned char flags;
- char Wlog; /* log(W) */
- char Plog; /* random number bits */
- char Scell_log;
- u8 Stab[256];
-
-/* Variables */
- unsigned long qave; /* Average queue length: A scaled */
- int qcount; /* Packets since last random number generation */
- u32 qR; /* Cached random number */
-
- psched_time_t qidlestart; /* Start of idle period */
- struct tc_red_xstats st;
+ u32 limit; /* HARD maximal queue length */
+ unsigned char flags;
+ struct red_parms parms;
+ struct red_stats stats;
+ struct Qdisc *qdisc;
};
-static int red_ecn_mark(struct sk_buff *skb)
+static inline int red_use_ecn(struct red_sched_data *q)
{
- if (skb->nh.raw + 20 > skb->tail)
- return 0;
-
- switch (skb->protocol) {
- case __constant_htons(ETH_P_IP):
- if (INET_ECN_is_not_ect(skb->nh.iph->tos))
- return 0;
- IP_ECN_set_ce(skb->nh.iph);
- return 1;
- case __constant_htons(ETH_P_IPV6):
- if (INET_ECN_is_not_ect(ipv6_get_dsfield(skb->nh.ipv6h)))
- return 0;
- IP6_ECN_set_ce(skb->nh.ipv6h);
- return 1;
- default:
- return 0;
- }
+ return q->flags & TC_RED_ECN;
}
-static int
-red_enqueue(struct sk_buff *skb, struct Qdisc* sch)
+static inline int red_use_harddrop(struct red_sched_data *q)
{
- struct red_sched_data *q = qdisc_priv(sch);
-
- psched_time_t now;
-
- if (!PSCHED_IS_PASTPERFECT(q->qidlestart)) {
- long us_idle;
- int shift;
-
- PSCHED_GET_TIME(now);
- us_idle = PSCHED_TDIFF_SAFE(now, q->qidlestart, q->Scell_max);
- PSCHED_SET_PASTPERFECT(q->qidlestart);
+ return q->flags & TC_RED_HARDDROP;
+}
-/*
- The problem: ideally, average length queue recalcultion should
- be done over constant clock intervals. This is too expensive, so that
- the calculation is driven by outgoing packets.
- When the queue is idle we have to model this clock by hand.
-
- SF+VJ proposed to "generate" m = idletime/(average_pkt_size/bandwidth)
- dummy packets as a burst after idle time, i.e.
-
- q->qave *= (1-W)^m
-
- This is an apparently overcomplicated solution (f.e. we have to precompute
- a table to make this calculation in reasonable time)
- I believe that a simpler model may be used here,
- but it is field for experiments.
-*/
- shift = q->Stab[us_idle>>q->Scell_log];
-
- if (shift) {
- q->qave >>= shift;
- } else {
- /* Approximate initial part of exponent
- with linear function:
- (1-W)^m ~= 1-mW + ...
-
- Seems, it is the best solution to
- problem of too coarce exponent tabulation.
- */
-
- us_idle = (q->qave * us_idle)>>q->Scell_log;
- if (us_idle < q->qave/2)
- q->qave -= us_idle;
- else
- q->qave >>= 1;
- }
- } else {
- q->qave += sch->qstats.backlog - (q->qave >> q->Wlog);
- /* NOTE:
- q->qave is fixed point number with point at Wlog.
- The formulae above is equvalent to floating point
- version:
-
- qave = qave*(1-W) + sch->qstats.backlog*W;
- --ANK (980924)
- */
+static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch)
+{
+ struct red_sched_data *q = qdisc_priv(sch);
+ struct Qdisc *child = q->qdisc;
+ int ret;
+
+ q->parms.qavg = red_calc_qavg(&q->parms, child->qstats.backlog);
+
+ if (red_is_idling(&q->parms))
+ red_end_of_idle_period(&q->parms);
+
+ switch (red_action(&q->parms, q->parms.qavg)) {
+ case RED_DONT_MARK:
+ break;
+
+ case RED_PROB_MARK:
+ sch->qstats.overlimits++;
+ 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:
+ sch->qstats.overlimits++;
+ 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;
}
- if (q->qave < q->qth_min) {
- q->qcount = -1;
-enqueue:
- if (sch->qstats.backlog + skb->len <= q->limit) {
- __skb_queue_tail(&sch->q, skb);
- sch->qstats.backlog += skb->len;
- sch->bstats.bytes += skb->len;
- sch->bstats.packets++;
- return NET_XMIT_SUCCESS;
- } else {
- q->st.pdrop++;
- }
- kfree_skb(skb);
+ ret = child->enqueue(skb, child);
+ if (likely(ret == NET_XMIT_SUCCESS)) {
+ sch->bstats.bytes += skb->len;
+ sch->bstats.packets++;
+ sch->q.qlen++;
+ } else {
+ q->stats.pdrop++;
sch->qstats.drops++;
- return NET_XMIT_DROP;
- }
- if (q->qave >= q->qth_max) {
- q->qcount = -1;
- sch->qstats.overlimits++;
-mark:
- if (!(q->flags&TC_RED_ECN) || !red_ecn_mark(skb)) {
- q->st.early++;
- goto drop;
- }
- q->st.marked++;
- goto enqueue;
}
+ return ret;
- if (++q->qcount) {
- /* The formula used below causes questions.
-
- OK. qR is random number in the interval 0..Rmask
- i.e. 0..(2^Plog). If we used floating point
- arithmetics, it would be: (2^Plog)*rnd_num,
- where rnd_num is less 1.
-
- Taking into account, that qave have fixed
- point at Wlog, and Plog is related to max_P by
- max_P = (qth_max-qth_min)/2^Plog; two lines
- below have the following floating point equivalent:
-
- max_P*(qave - qth_min)/(qth_max-qth_min) < rnd/qcount
-
- Any questions? --ANK (980924)
- */
- if (((q->qave - q->qth_min)>>q->Wlog)*q->qcount < q->qR)
- goto enqueue;
- q->qcount = 0;
- q->qR = net_random()&q->Rmask;
- sch->qstats.overlimits++;
- goto mark;
- }
- q->qR = net_random()&q->Rmask;
- goto enqueue;
-
-drop:
- kfree_skb(skb);
- sch->qstats.drops++;
+congestion_drop:
+ qdisc_drop(skb, sch);
return NET_XMIT_CN;
}
-static int
-red_requeue(struct sk_buff *skb, struct Qdisc* sch)
+static int red_requeue(struct sk_buff *skb, struct Qdisc* sch)
{
struct red_sched_data *q = qdisc_priv(sch);
+ struct Qdisc *child = q->qdisc;
+ int ret;
- PSCHED_SET_PASTPERFECT(q->qidlestart);
+ if (red_is_idling(&q->parms))
+ red_end_of_idle_period(&q->parms);
- __skb_queue_head(&sch->q, skb);
- sch->qstats.backlog += skb->len;
- sch->qstats.requeues++;
- return 0;
+ ret = child->ops->requeue(skb, child);
+ if (likely(ret == NET_XMIT_SUCCESS)) {
+ sch->qstats.requeues++;
+ sch->q.qlen++;
+ }
+ return ret;
}
-static struct sk_buff *
-red_dequeue(struct Qdisc* sch)
+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 = __skb_dequeue(&sch->q);
- if (skb) {
- sch->qstats.backlog -= skb->len;
- return skb;
- }
- PSCHED_GET_TIME(q->qidlestart);
- return NULL;
+ skb = child->dequeue(child);
+ if (skb)
+ sch->q.qlen--;
+ else if (!red_is_idling(&q->parms))
+ red_start_of_idle_period(&q->parms);
+
+ return skb;
}
static unsigned int red_drop(struct Qdisc* sch)
{
- struct sk_buff *skb;
struct red_sched_data *q = qdisc_priv(sch);
+ struct Qdisc *child = q->qdisc;
+ unsigned int len;
- skb = __skb_dequeue_tail(&sch->q);
- if (skb) {
- unsigned int len = skb->len;
- sch->qstats.backlog -= len;
+ if (child->ops->drop && (len = child->ops->drop(child)) > 0) {
+ q->stats.other++;
sch->qstats.drops++;
- q->st.other++;
- kfree_skb(skb);
+ sch->q.qlen--;
return len;
}
- PSCHED_GET_TIME(q->qidlestart);
+
+ if (!red_is_idling(&q->parms))
+ red_start_of_idle_period(&q->parms);
+
return 0;
}
{
struct red_sched_data *q = qdisc_priv(sch);
- __skb_queue_purge(&sch->q);
- sch->qstats.backlog = 0;
- PSCHED_SET_PASTPERFECT(q->qidlestart);
- q->qave = 0;
- q->qcount = -1;
+ qdisc_reset(q->qdisc);
+ sch->q.qlen = 0;
+ red_restart(&q->parms);
+}
+
+static void red_destroy(struct Qdisc *sch)
+{
+ struct red_sched_data *q = qdisc_priv(sch);
+ qdisc_destroy(q->qdisc);
+}
+
+static struct Qdisc *red_create_dflt(struct Qdisc *sch, u32 limit)
+{
+ struct Qdisc *q;
+ struct rtattr *rta;
+ int ret;
+
+ q = qdisc_create_dflt(sch->dev, &bfifo_qdisc_ops,
+ TC_H_MAKE(sch->handle, 1));
+ if (q) {
+ rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)),
+ GFP_KERNEL);
+ if (rta) {
+ rta->rta_type = RTM_NEWQDISC;
+ rta->rta_len = RTA_LENGTH(sizeof(struct tc_fifo_qopt));
+ ((struct tc_fifo_qopt *)RTA_DATA(rta))->limit = limit;
+
+ ret = q->ops->change(q, rta);
+ kfree(rta);
+
+ if (ret == 0)
+ return q;
+ }
+ qdisc_destroy(q);
+ }
+ return NULL;
}
static int red_change(struct Qdisc *sch, struct rtattr *opt)
{
struct red_sched_data *q = qdisc_priv(sch);
- struct rtattr *tb[TCA_RED_STAB];
+ struct rtattr *tb[TCA_RED_MAX];
struct tc_red_qopt *ctl;
+ struct Qdisc *child = NULL;
- if (opt == NULL ||
- rtattr_parse_nested(tb, TCA_RED_STAB, opt) ||
- tb[TCA_RED_PARMS-1] == 0 || tb[TCA_RED_STAB-1] == 0 ||
+ if (opt == NULL || rtattr_parse_nested(tb, TCA_RED_MAX, opt))
+ return -EINVAL;
+
+ if (tb[TCA_RED_PARMS-1] == NULL ||
RTA_PAYLOAD(tb[TCA_RED_PARMS-1]) < sizeof(*ctl) ||
- RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < 256)
+ tb[TCA_RED_STAB-1] == NULL ||
+ RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < RED_STAB_SIZE)
return -EINVAL;
ctl = RTA_DATA(tb[TCA_RED_PARMS-1]);
+ if (ctl->limit > 0) {
+ child = red_create_dflt(sch, ctl->limit);
+ if (child == NULL)
+ return -ENOMEM;
+ }
+
sch_tree_lock(sch);
q->flags = ctl->flags;
- q->Wlog = ctl->Wlog;
- q->Plog = ctl->Plog;
- q->Rmask = ctl->Plog < 32 ? ((1<<ctl->Plog) - 1) : ~0UL;
- q->Scell_log = ctl->Scell_log;
- q->Scell_max = (255<<q->Scell_log);
- q->qth_min = ctl->qth_min<<ctl->Wlog;
- q->qth_max = ctl->qth_max<<ctl->Wlog;
q->limit = ctl->limit;
- memcpy(q->Stab, RTA_DATA(tb[TCA_RED_STAB-1]), 256);
+ if (child)
+ qdisc_destroy(xchg(&q->qdisc, child));
+
+ red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
+ ctl->Plog, ctl->Scell_log,
+ RTA_DATA(tb[TCA_RED_STAB-1]));
+
+ if (skb_queue_empty(&sch->q))
+ red_end_of_idle_period(&q->parms);
- q->qcount = -1;
- if (skb_queue_len(&sch->q) == 0)
- PSCHED_SET_PASTPERFECT(q->qidlestart);
sch_tree_unlock(sch);
return 0;
}
static int red_init(struct Qdisc* sch, struct rtattr *opt)
{
+ struct red_sched_data *q = qdisc_priv(sch);
+
+ q->qdisc = &noop_qdisc;
return red_change(sch, opt);
}
static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
{
struct red_sched_data *q = qdisc_priv(sch);
- unsigned char *b = skb->tail;
- struct rtattr *rta;
- struct tc_red_qopt opt;
-
- rta = (struct rtattr*)b;
- RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
- opt.limit = q->limit;
- opt.qth_min = q->qth_min>>q->Wlog;
- opt.qth_max = q->qth_max>>q->Wlog;
- opt.Wlog = q->Wlog;
- opt.Plog = q->Plog;
- opt.Scell_log = q->Scell_log;
- opt.flags = q->flags;
+ struct rtattr *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,
+ };
+
+ opts = RTA_NEST(skb, TCA_OPTIONS);
RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt);
- rta->rta_len = skb->tail - b;
-
- return skb->len;
+ return RTA_NEST_END(skb, opts);
rtattr_failure:
- skb_trim(skb, b - skb->data);
- return -1;
+ return RTA_NEST_CANCEL(skb, opts);
}
static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
{
struct red_sched_data *q = qdisc_priv(sch);
+ struct tc_red_xstats st = {
+ .early = q->stats.prob_drop + q->stats.forced_drop,
+ .pdrop = q->stats.pdrop,
+ .other = q->stats.other,
+ .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);
+
+ if (cl != 1)
+ return -ENOENT;
+ 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,
+ struct Qdisc **old)
+{
+ struct red_sched_data *q = qdisc_priv(sch);
+
+ if (new == NULL)
+ new = &noop_qdisc;
+
+ sch_tree_lock(sch);
+ *old = xchg(&q->qdisc, new);
+ qdisc_reset(*old);
+ sch->q.qlen = 0;
+ sch_tree_unlock(sch);
+ 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_get(struct Qdisc *sch, u32 classid)
+{
+ return 1;
+}
+
+static void red_put(struct Qdisc *sch, unsigned long arg)
+{
+ return;
+}
+
+static int red_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
+ struct rtattr **tca, unsigned long *arg)
+{
+ return -ENOSYS;
+}
- return gnet_stats_copy_app(d, &q->st, sizeof(q->st));
+static int red_delete(struct Qdisc *sch, unsigned long cl)
+{
+ return -ENOSYS;
+}
+
+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 struct tcf_proto **red_find_tcf(struct Qdisc *sch, unsigned long cl)
+{
+ return NULL;
}
+static struct Qdisc_class_ops red_class_ops = {
+ .graft = red_graft,
+ .leaf = red_leaf,
+ .get = red_get,
+ .put = red_put,
+ .change = red_change_class,
+ .delete = red_delete,
+ .walk = red_walk,
+ .tcf_chain = red_find_tcf,
+ .dump = red_dump_class,
+};
+
static struct Qdisc_ops red_qdisc_ops = {
- .next = NULL,
- .cl_ops = NULL,
.id = "red",
.priv_size = sizeof(struct red_sched_data),
+ .cl_ops = &red_class_ops,
.enqueue = red_enqueue,
.dequeue = red_dequeue,
.requeue = red_requeue,
.drop = red_drop,
.init = red_init,
.reset = red_reset,
+ .destroy = red_destroy,
.change = red_change,
.dump = red_dump,
.dump_stats = red_dump_stats,
{
return register_qdisc(&red_qdisc_ops);
}
-static void __exit red_module_exit(void)
+
+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");