[PATCH 0/5] PKT_SCHED: Convert Qdiscs to use generic network statistics/estimator

[PATCH 0/5] PKT_SCHED: Convert Qdiscs to use generic network statistics/estimator

[Thomas Graf]

Dave,

This patchset converts Qdiscs to use the generic network
statistics/estimator. A new TLV type TCA_STATS2 is introduced
carrying the new statistics while the old TLV type TCA_STATS
is still provided for backward compatibility.

Patch 1: Replace tc_stats with new gnet_stats in struct Qdisc
Patch 2: Use gnet_stats API to copy statistics into netlink message
Patch 3: Introduce gen_replace_estimator
Patch 4: Use generic rate estimator
Patch 5: Qdisc are not supposed to dump TCA_STATS themselves

I tested these patches for a few days and didn't encounter
any problems. iproute2 is able to read out statistics via
backward compatibility. A slightly patched iproute2 can
read the new statistics.

[PATCH 1/5] PKT_SCHED: Replace tc_stats with new gnet_stats in struct Qdisc

[Thomas Graf]

Replaces tc_stats with gnet_stats replacements in struct
Qdisc and adapts all qdiscs to use them.

Signed-off-by: Thomas Graf <tgraf@suug.ch>


diff -Nru linux-2.6.9-rc3-bk6.orig/include/net/pkt_sched.h linux-2.6.9-rc3-bk6/include/net/pkt_sched.h
--- linux-2.6.9-rc3-bk6.orig/include/net/pkt_sched.h	2004-10-06 14:19:13.000000000 +0200
+++ linux-2.6.9-rc3-bk6/include/net/pkt_sched.h	2004-10-06 20:13:56.000000000 +0200
@@ -9,6 +9,7 @@
 #include <net/pkt_cls.h>
 #include <linux/module.h>
 #include <linux/rtnetlink.h>
+#include <net/gen_stats.h>
 
 struct rtattr;
 struct Qdisc;
@@ -86,7 +87,9 @@
 	struct net_device	*dev;
 	struct list_head	list;
 
-	struct tc_stats		stats;
+	struct gnet_stats_basic	bstats;
+	struct gnet_stats_queue	qstats;
+	struct gnet_stats_rate_est	rate_est;
 	spinlock_t		*stats_lock;
 	struct rcu_head 	q_rcu;
 	int			(*reshape_fail)(struct sk_buff *skb, struct Qdisc *q);
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_cbq.c linux-2.6.9-rc3-bk6/net/sched/sch_cbq.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_cbq.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_cbq.c	2004-10-06 20:35:09.000000000 +0200
@@ -433,8 +433,8 @@
 #endif
 		if ((ret = cl->q->enqueue(skb, cl->q)) == NET_XMIT_SUCCESS) {
 			sch->q.qlen++;
-			sch->stats.packets++;
-			sch->stats.bytes+=len;
+			sch->bstats.packets++;
+			sch->bstats.bytes+=len;
 			cbq_mark_toplevel(q, cl);
 			if (!cl->next_alive)
 				cbq_activate_class(cl);
@@ -443,7 +443,7 @@
 	}
 
 #ifndef CONFIG_NET_CLS_ACT
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	if (cl == NULL)
 		kfree_skb(skb);
 	else {
@@ -452,7 +452,7 @@
 	}
 #else
 	if ( NET_XMIT_DROP == ret) {
-		sch->stats.drops++;
+		sch->qstats.drops++;
 	}
 
 	if (cl != NULL) {
@@ -472,7 +472,7 @@
 
 	if ((cl = q->tx_class) == NULL) {
 		kfree_skb(skb);
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return NET_XMIT_CN;
 	}
 	q->tx_class = NULL;
@@ -489,7 +489,7 @@
 			cbq_activate_class(cl);
 		return 0;
 	}
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	cl->stats.drops++;
 	return ret;
 }
@@ -729,17 +729,17 @@
 
 		if (cl->q->enqueue(skb, cl->q) == 0) {
 			sch->q.qlen++;
-			sch->stats.packets++;
-			sch->stats.bytes+=len;
+			sch->bstats.packets++;
+			sch->bstats.bytes+=len;
 			if (!cl->next_alive)
 				cbq_activate_class(cl);
 			return 0;
 		}
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return 0;
 	}
 
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	return -1;
 }
 #endif
@@ -1090,7 +1090,7 @@
 	   Sigh... start watchdog timer in the last case. */
 
 	if (sch->q.qlen) {
-		sch->stats.overlimits++;
+		sch->qstats.overlimits++;
 		if (q->wd_expires) {
 			long delay = PSCHED_US2JIFFIE(q->wd_expires);
 			if (delay <= 0)
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_dsmark.c linux-2.6.9-rc3-bk6/net/sched/sch_dsmark.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_dsmark.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_dsmark.c	2004-10-06 20:45:26.000000000 +0200
@@ -241,11 +241,11 @@
 #endif
 
 	    ((ret = p->q->enqueue(skb,p->q)) != 0)) {
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return ret;
 	}
-	sch->stats.bytes += skb->len;
-	sch->stats.packets++;
+	sch->bstats.bytes += skb->len;
+	sch->bstats.packets++;
 	sch->q.qlen++;
 	return ret;
 }
@@ -299,7 +299,7 @@
 		sch->q.qlen++;
 		return 0;
 	}
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	return ret;
 }
 
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_fifo.c linux-2.6.9-rc3-bk6/net/sched/sch_fifo.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_fifo.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_fifo.c	2004-10-06 20:29:31.000000000 +0200
@@ -47,14 +47,14 @@
 {
 	struct fifo_sched_data *q = qdisc_priv(sch);
 
-	if (sch->stats.backlog + skb->len <= q->limit) {
+	if (sch->qstats.backlog + skb->len <= q->limit) {
 		__skb_queue_tail(&sch->q, skb);
-		sch->stats.backlog += skb->len;
-		sch->stats.bytes += skb->len;
-		sch->stats.packets++;
+		sch->qstats.backlog += skb->len;
+		sch->bstats.bytes += skb->len;
+		sch->bstats.packets++;
 		return 0;
 	}
-	sch->stats.drops++;
+	sch->qstats.drops++;
 #ifdef CONFIG_NET_CLS_POLICE
 	if (sch->reshape_fail==NULL || sch->reshape_fail(skb, sch))
 #endif
@@ -66,7 +66,7 @@
 bfifo_requeue(struct sk_buff *skb, struct Qdisc* sch)
 {
 	__skb_queue_head(&sch->q, skb);
-	sch->stats.backlog += skb->len;
+	sch->qstats.backlog += skb->len;
 	return 0;
 }
 
@@ -77,7 +77,7 @@
 
 	skb = __skb_dequeue(&sch->q);
 	if (skb)
-		sch->stats.backlog -= skb->len;
+		sch->qstats.backlog -= skb->len;
 	return skb;
 }
 
@@ -89,7 +89,7 @@
 	skb = __skb_dequeue_tail(&sch->q);
 	if (skb) {
 		unsigned int len = skb->len;
-		sch->stats.backlog -= len;
+		sch->qstats.backlog -= len;
 		kfree_skb(skb);
 		return len;
 	}
@@ -100,7 +100,7 @@
 fifo_reset(struct Qdisc* sch)
 {
 	skb_queue_purge(&sch->q);
-	sch->stats.backlog = 0;
+	sch->qstats.backlog = 0;
 }
 
 static int
@@ -110,11 +110,11 @@
 
 	if (sch->q.qlen < q->limit) {
 		__skb_queue_tail(&sch->q, skb);
-		sch->stats.bytes += skb->len;
-		sch->stats.packets++;
+		sch->bstats.bytes += skb->len;
+		sch->bstats.packets++;
 		return 0;
 	}
-	sch->stats.drops++;
+	sch->qstats.drops++;
 #ifdef CONFIG_NET_CLS_POLICE
 	if (sch->reshape_fail==NULL || sch->reshape_fail(skb, sch))
 #endif
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_generic.c linux-2.6.9-rc3-bk6/net/sched/sch_generic.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_generic.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_generic.c	2004-10-07 00:17:21.000000000 +0200
@@ -318,11 +318,11 @@
 	if (list->qlen < qdisc->dev->tx_queue_len) {
 		__skb_queue_tail(list, skb);
 		qdisc->q.qlen++;
-		qdisc->stats.bytes += skb->len;
-		qdisc->stats.packets++;
+		qdisc->bstats.bytes += skb->len;
+		qdisc->bstats.packets++;
 		return 0;
 	}
-	qdisc->stats.drops++;
+	qdisc->qstats.drops++;
 	kfree_skb(skb);
 	return NET_XMIT_DROP;
 }
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_gred.c linux-2.6.9-rc3-bk6/net/sched/sch_gred.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_gred.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_gred.c	2004-10-06 20:44:10.000000000 +0200
@@ -130,7 +130,7 @@
 
 	D2PRINTK("gred_enqueue virtualQ 0x%x classid %x backlog %d "
 	    "general backlog %d\n",skb->tc_index&0xf,sch->handle,q->backlog,
-	    sch->stats.backlog);
+	    sch->qstats.backlog);
 	/* sum up all the qaves of prios <= to ours to get the new qave*/
 	if (!t->eqp && t->grio) {
 		for (i=0;i<t->DPs;i++) {
@@ -161,7 +161,7 @@
 		q->qave >>= q->Stab[(us_idle>>q->Scell_log)&0xFF];
 	} else {
 		if (t->eqp) {
-			q->qave += sch->stats.backlog - (q->qave >> q->Wlog);
+			q->qave += sch->qstats.backlog - (q->qave >> q->Wlog);
 		} else {
 			q->qave += q->backlog - (q->qave >> q->Wlog);
 		}
@@ -179,9 +179,9 @@
 			q->backlog += skb->len;
 do_enqueue:
 			__skb_queue_tail(&sch->q, skb);
-			sch->stats.backlog += skb->len;
-			sch->stats.bytes += skb->len;
-			sch->stats.packets++;
+			sch->qstats.backlog += skb->len;
+			sch->bstats.bytes += skb->len;
+			sch->bstats.packets++;
 			return 0;
 		} else {
 			q->pdrop++;
@@ -189,12 +189,12 @@
 
 drop:
 		kfree_skb(skb);
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return NET_XMIT_DROP;
 	}
 	if ((q->qave+qave) >= q->qth_max) {
 		q->qcount = -1;
-		sch->stats.overlimits++;
+		sch->qstats.overlimits++;
 		q->forced++;
 		goto drop;
 	}
@@ -203,7 +203,7 @@
 			goto enqueue;
 		q->qcount = 0;
 		q->qR = net_random()&q->Rmask;
-		sch->stats.overlimits++;
+		sch->qstats.overlimits++;
 		q->early++;
 		goto drop;
 	}
@@ -221,7 +221,7 @@
 	PSCHED_SET_PASTPERFECT(q->qidlestart);
 
 	__skb_queue_head(&sch->q, skb);
-	sch->stats.backlog += skb->len;
+	sch->qstats.backlog += skb->len;
 	q->backlog += skb->len;
 	return 0;
 }
@@ -235,7 +235,7 @@
 
 	skb = __skb_dequeue(&sch->q);
 	if (skb) {
-		sch->stats.backlog -= skb->len;
+		sch->qstats.backlog -= skb->len;
 		q= t->tab[(skb->tc_index&0xf)];
 		if (q) {
 			q->backlog -= skb->len;
@@ -269,8 +269,8 @@
 	skb = __skb_dequeue_tail(&sch->q);
 	if (skb) {
 		unsigned int len = skb->len;
-		sch->stats.backlog -= len;
-		sch->stats.drops++;
+		sch->qstats.backlog -= len;
+		sch->qstats.drops++;
 		q= t->tab[(skb->tc_index&0xf)];
 		if (q) {
 			q->backlog -= len;
@@ -304,7 +304,7 @@
 
 	__skb_queue_purge(&sch->q);
 
-	sch->stats.backlog = 0;
+	sch->qstats.backlog = 0;
 
         for (i=0;i<t->DPs;i++) {
 	        q= t->tab[i];
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_hfsc.c linux-2.6.9-rc3-bk6/net/sched/sch_hfsc.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_hfsc.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_hfsc.c	2004-10-07 00:18:23.000000000 +0200
@@ -1677,14 +1677,14 @@
 #ifdef CONFIG_NET_CLS_ACT
 	if (cl == NULL) {
 		if (NET_XMIT_DROP == ret) {
-			sch->stats.drops++;
+			sch->qstats.drops++;
 		}
 		return ret;
 	}
 #else
 	if (cl == NULL) {
 		kfree_skb(skb);
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return NET_XMIT_DROP;
 	}
 #endif
@@ -1692,7 +1692,7 @@
 	err = cl->qdisc->enqueue(skb, cl->qdisc);
 	if (unlikely(err != NET_XMIT_SUCCESS)) {
 		cl->stats.drops++;
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return err;
 	}
 
@@ -1701,8 +1701,8 @@
 
 	cl->stats.packets++;
 	cl->stats.bytes += len;
-	sch->stats.packets++;
-	sch->stats.bytes += len;
+	sch->bstats.packets++;
+	sch->bstats.bytes += len;
 	sch->q.qlen++;
 
 	return NET_XMIT_SUCCESS;
@@ -1739,7 +1739,7 @@
 		 */
 		cl = vttree_get_minvt(&q->root, cur_time);
 		if (cl == NULL) {
-			sch->stats.overlimits++;
+			sch->qstats.overlimits++;
 			hfsc_schedule_watchdog(sch, cur_time);
 			return NULL;
 		}
@@ -1804,7 +1804,7 @@
 				list_move_tail(&cl->dlist, &q->droplist);
 			}
 			cl->stats.drops++;
-			sch->stats.drops++;
+			sch->qstats.drops++;
 			sch->q.qlen--;
 			return len;
 		}
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_htb.c linux-2.6.9-rc3-bk6/net/sched/sch_htb.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_htb.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_htb.c	2004-10-07 00:19:38.000000000 +0200
@@ -735,7 +735,7 @@
 	}
     } else if (!cl) {
 	    if (NET_XMIT_DROP == ret) {
-		    sch->stats.drops++;
+		    sch->qstats.drops++;
 	    }
 	    return ret;
     }
@@ -747,13 +747,13 @@
 	    q->direct_pkts++;
 	} else {
 	    kfree_skb (skb);
-	    sch->stats.drops++;
+	    sch->qstats.drops++;
 	    return NET_XMIT_DROP;
 	}
     }
 #endif
     else if (cl->un.leaf.q->enqueue(skb, cl->un.leaf.q) != NET_XMIT_SUCCESS) {
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	cl->stats.drops++;
 	return NET_XMIT_DROP;
     } else {
@@ -762,7 +762,7 @@
     }
 
     sch->q.qlen++;
-    sch->stats.packets++; sch->stats.bytes += skb->len;
+    sch->bstats.packets++; sch->bstats.bytes += skb->len;
     HTB_DBG(1,1,"htb_enq_ok cl=%X skb=%p\n",(cl && cl != HTB_DIRECT)?cl->classid:0,skb);
     return NET_XMIT_SUCCESS;
 }
@@ -783,11 +783,11 @@
             __skb_queue_head(&q->direct_queue, skb);
             tskb = __skb_dequeue_tail(&q->direct_queue);
             kfree_skb (tskb);
-            sch->stats.drops++;
+            sch->qstats.drops++;
             return NET_XMIT_CN;	
 	}
     } else if (cl->un.leaf.q->ops->requeue(skb, cl->un.leaf.q) != NET_XMIT_SUCCESS) {
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	cl->stats.drops++;
 	return NET_XMIT_DROP;
     } else 
@@ -1117,7 +1117,7 @@
 	/* why don't use jiffies here ? because expires can be in past */
 	mod_timer(&q->timer, q->jiffies + delay);
 	sch->flags |= TCQ_F_THROTTLED;
-	sch->stats.overlimits++;
+	sch->qstats.overlimits++;
 	HTB_DBG(3,1,"htb_deq t_delay=%ld\n",delay);
 }
 
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_ingress.c linux-2.6.9-rc3-bk6/net/sched/sch_ingress.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_ingress.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_ingress.c	2004-10-06 20:45:04.000000000 +0200
@@ -151,12 +151,12 @@
 	 * firewall FW_* code.
 	 */
 #ifdef CONFIG_NET_CLS_ACT
-	sch->stats.packets++;
-	sch->stats.bytes += skb->len;
+	sch->bstats.packets++;
+	sch->bstats.bytes += skb->len;
 	switch (result) {
 		case TC_ACT_SHOT:
 			result = TC_ACT_SHOT;
-			sch->stats.drops++;
+			sch->qstats.drops++;
 			break;
 		case TC_ACT_STOLEN:
 		case TC_ACT_QUEUED:
@@ -176,14 +176,14 @@
 	switch (result) {
 		case TC_POLICE_SHOT:
 		result = NF_DROP;
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		break;
 		case TC_POLICE_RECLASSIFY: /* DSCP remarking here ? */
 		case TC_POLICE_OK:
 		case TC_POLICE_UNSPEC:
 		default:
-		sch->stats.packets++;
-		sch->stats.bytes += skb->len;
+		sch->bstats.packets++;
+		sch->bstats.bytes += skb->len;
 		result = NF_ACCEPT;
 		break;
 	};
@@ -191,8 +191,8 @@
 #else
 	D2PRINTK("Overriding result to ACCEPT\n");
 	result = NF_ACCEPT;
-	sch->stats.packets++;
-	sch->stats.bytes += skb->len;
+	sch->bstats.packets++;
+	sch->bstats.bytes += skb->len;
 #endif
 #endif
 
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_netem.c linux-2.6.9-rc3-bk6/net/sched/sch_netem.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_netem.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_netem.c	2004-10-06 21:02:46.000000000 +0200
@@ -153,12 +153,12 @@
 	if (likely(q->delayed.qlen < q->limit)) {
 		__skb_queue_tail(&q->delayed, skb);
 		sch->q.qlen++;
-		sch->stats.bytes += skb->len;
-		sch->stats.packets++;
+		sch->bstats.bytes += skb->len;
+		sch->bstats.packets++;
 		return NET_XMIT_SUCCESS;
 	}
 
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	kfree_skb(skb);
 	return NET_XMIT_DROP;
 }
@@ -172,7 +172,7 @@
 	/* Random packet drop 0 => none, ~0 => all */
 	if (q->loss && q->loss >= get_crandom(&q->loss_cor)) {
 		pr_debug("netem_enqueue: random loss\n");
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return 0;	/* lie about loss so TCP doesn't know */
 	}
 
@@ -196,7 +196,7 @@
 		++q->counter;
 		ret = q->qdisc->enqueue(skb, q->qdisc);
 		if (ret)
-			sch->stats.drops++;
+			sch->qstats.drops++;
 		return ret;
 	}
 	
@@ -224,7 +224,7 @@
 
 	if ((len = q->qdisc->ops->drop(q->qdisc)) != 0) {
 		sch->q.qlen--;
-		sch->stats.drops++;
+		sch->qstats.drops++;
 	}
 	return len;
 }
@@ -256,7 +256,7 @@
 		__skb_unlink(skb, &q->delayed);
 
 		if (q->qdisc->enqueue(skb, q->qdisc))
-			sch->stats.drops++;
+			sch->qstats.drops++;
 	}
 
 	skb = q->qdisc->dequeue(q->qdisc);
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_prio.c linux-2.6.9-rc3-bk6/net/sched/sch_prio.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_prio.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_prio.c	2004-10-06 21:02:03.000000000 +0200
@@ -107,8 +107,8 @@
 		goto dropped;
 
 	if ((ret = qdisc->enqueue(skb, qdisc)) == NET_XMIT_SUCCESS) {
-		sch->stats.bytes += skb->len;
-		sch->stats.packets++;
+		sch->bstats.bytes += skb->len;
+		sch->bstats.packets++;
 		sch->q.qlen++;
 		return NET_XMIT_SUCCESS;
 	}
@@ -117,10 +117,10 @@
 #ifdef CONFIG_NET_CLS_ACT
 	if (NET_XMIT_DROP == ret) {
 #endif
-		sch->stats.drops++;
+		sch->qstats.drops++;
 #ifdef CONFIG_NET_CLS_ACT
 	} else {
-		sch->stats.overlimits++; /* abuse, but noone uses it */
+		sch->qstats.overlimits++; /* abuse, but noone uses it */
 	}
 #endif
 	return ret; 
@@ -142,7 +142,7 @@
 		return 0;
 	}
 dropped:
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	return NET_XMIT_DROP;
 }
 
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_red.c linux-2.6.9-rc3-bk6/net/sched/sch_red.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_red.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_red.c	2004-10-06 20:41:48.000000000 +0200
@@ -228,13 +228,13 @@
 				q->qave >>= 1;
 		}
 	} else {
-		q->qave += sch->stats.backlog - (q->qave >> q->Wlog);
+		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->stats.backlog*W;
+		   qave = qave*(1-W) + sch->qstats.backlog*W;
 		                                           --ANK (980924)
 		 */
 	}
@@ -242,22 +242,22 @@
 	if (q->qave < q->qth_min) {
 		q->qcount = -1;
 enqueue:
-		if (sch->stats.backlog + skb->len <= q->limit) {
+		if (sch->qstats.backlog + skb->len <= q->limit) {
 			__skb_queue_tail(&sch->q, skb);
-			sch->stats.backlog += skb->len;
-			sch->stats.bytes += skb->len;
-			sch->stats.packets++;
+			sch->qstats.backlog += skb->len;
+			sch->bstats.bytes += skb->len;
+			sch->bstats.packets++;
 			return NET_XMIT_SUCCESS;
 		} else {
 			q->st.pdrop++;
 		}
 		kfree_skb(skb);
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return NET_XMIT_DROP;
 	}
 	if (q->qave >= q->qth_max) {
 		q->qcount = -1;
-		sch->stats.overlimits++;
+		sch->qstats.overlimits++;
 mark:
 		if  (!(q->flags&TC_RED_ECN) || !red_ecn_mark(skb)) {
 			q->st.early++;
@@ -288,7 +288,7 @@
 			goto enqueue;
 		q->qcount = 0;
 		q->qR = net_random()&q->Rmask;
-		sch->stats.overlimits++;
+		sch->qstats.overlimits++;
 		goto mark;
 	}
 	q->qR = net_random()&q->Rmask;
@@ -296,7 +296,7 @@
 
 drop:
 	kfree_skb(skb);
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	return NET_XMIT_CN;
 }
 
@@ -308,7 +308,7 @@
 	PSCHED_SET_PASTPERFECT(q->qidlestart);
 
 	__skb_queue_head(&sch->q, skb);
-	sch->stats.backlog += skb->len;
+	sch->qstats.backlog += skb->len;
 	return 0;
 }
 
@@ -320,7 +320,7 @@
 
 	skb = __skb_dequeue(&sch->q);
 	if (skb) {
-		sch->stats.backlog -= skb->len;
+		sch->qstats.backlog -= skb->len;
 		return skb;
 	}
 	PSCHED_GET_TIME(q->qidlestart);
@@ -335,8 +335,8 @@
 	skb = __skb_dequeue_tail(&sch->q);
 	if (skb) {
 		unsigned int len = skb->len;
-		sch->stats.backlog -= len;
-		sch->stats.drops++;
+		sch->qstats.backlog -= len;
+		sch->qstats.drops++;
 		q->st.other++;
 		kfree_skb(skb);
 		return len;
@@ -350,7 +350,7 @@
 	struct red_sched_data *q = qdisc_priv(sch);
 
 	__skb_queue_purge(&sch->q);
-	sch->stats.backlog = 0;
+	sch->qstats.backlog = 0;
 	PSCHED_SET_PASTPERFECT(q->qidlestart);
 	q->qave = 0;
 	q->qcount = -1;
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_sfq.c linux-2.6.9-rc3-bk6/net/sched/sch_sfq.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_sfq.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_sfq.c	2004-10-06 20:46:11.000000000 +0200
@@ -227,7 +227,7 @@
 		kfree_skb(skb);
 		sfq_dec(q, x);
 		sch->q.qlen--;
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return len;
 	}
 
@@ -243,7 +243,7 @@
 		sfq_dec(q, d);
 		sch->q.qlen--;
 		q->ht[q->hash[d]] = SFQ_DEPTH;
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return len;
 	}
 
@@ -276,8 +276,8 @@
 		}
 	}
 	if (++sch->q.qlen < q->limit-1) {
-		sch->stats.bytes += skb->len;
-		sch->stats.packets++;
+		sch->bstats.bytes += skb->len;
+		sch->bstats.packets++;
 		return 0;
 	}
 
@@ -313,7 +313,7 @@
 	if (++sch->q.qlen < q->limit - 1)
 		return 0;
 
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	sfq_drop(sch);
 	return NET_XMIT_CN;
 }
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_tbf.c linux-2.6.9-rc3-bk6/net/sched/sch_tbf.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_tbf.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_tbf.c	2004-10-06 21:00:24.000000000 +0200
@@ -141,7 +141,7 @@
 	int ret;
 
 	if (skb->len > q->max_size) {
-		sch->stats.drops++;
+		sch->qstats.drops++;
 #ifdef CONFIG_NET_CLS_POLICE
 		if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
 #endif
@@ -151,13 +151,13 @@
 	}
 
 	if ((ret = q->qdisc->enqueue(skb, q->qdisc)) != 0) {
-		sch->stats.drops++;
+		sch->qstats.drops++;
 		return ret;
 	}
 
 	sch->q.qlen++;
-	sch->stats.bytes += skb->len;
-	sch->stats.packets++;
+	sch->bstats.bytes += skb->len;
+	sch->bstats.packets++;
 	return 0;
 }
 
@@ -179,7 +179,7 @@
 
 	if ((len = q->qdisc->ops->drop(q->qdisc)) != 0) {
 		sch->q.qlen--;
-		sch->stats.drops++;
+		sch->qstats.drops++;
 	}
 	return len;
 }
@@ -250,11 +250,11 @@
 		if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) {
 			/* When requeue fails skb is dropped */
 			sch->q.qlen--;
-			sch->stats.drops++;
+			sch->qstats.drops++;
 		}
 
 		sch->flags |= TCQ_F_THROTTLED;
-		sch->stats.overlimits++;
+		sch->qstats.overlimits++;
 	}
 	return NULL;
 }
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_teql.c linux-2.6.9-rc3-bk6/net/sched/sch_teql.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_teql.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_teql.c	2004-10-06 21:01:33.000000000 +0200
@@ -96,14 +96,14 @@
 
 	__skb_queue_tail(&q->q, skb);
 	if (q->q.qlen <= dev->tx_queue_len) {
-		sch->stats.bytes += skb->len;
-		sch->stats.packets++;
+		sch->bstats.bytes += skb->len;
+		sch->bstats.packets++;
 		return 0;
 	}
 
 	__skb_unlink(skb, &q->q);
 	kfree_skb(skb);
-	sch->stats.drops++;
+	sch->qstats.drops++;
 	return NET_XMIT_DROP;
 }
 

[PATCH 2/5] PKT_SCHED: Use gnet_stats API to copy statistics into netlink message

[Thomas Graf]

Adapts qdisc API to use new gnet_stats functions to copy
statistics into netlink message.

Signed-off-by: Thomas Graf <tgraf@suug.ch>

diff -Nru linux-2.6.9-rc3-bk6.orig/include/linux/rtnetlink.h linux-2.6.9-rc3-bk6/include/linux/rtnetlink.h
--- linux-2.6.9-rc3-bk6.orig/include/linux/rtnetlink.h	2004-10-06 14:19:13.000000000 +0200
+++ linux-2.6.9-rc3-bk6/include/linux/rtnetlink.h	2004-10-06 20:24:32.000000000 +0200
@@ -698,6 +698,7 @@
 	TCA_XSTATS,
 	TCA_RATE,
 	TCA_FCNT,
+	TCA_STATS2,
 	__TCA_MAX
 };
 
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_api.c linux-2.6.9-rc3-bk6/net/sched/sch_api.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_api.c	2004-10-06 14:19:49.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_api.c	2004-10-07 00:39:04.000000000 +0200
@@ -750,6 +750,7 @@
 	struct tcmsg *tcm;
 	struct nlmsghdr  *nlh;
 	unsigned char	 *b = skb->tail;
+	struct gnet_dump d;
 
 	nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*tcm));
 	nlh->nlmsg_flags = flags;
@@ -762,9 +763,22 @@
 	RTA_PUT(skb, TCA_KIND, IFNAMSIZ, q->ops->id);
 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
 		goto rtattr_failure;
-	q->stats.qlen = q->q.qlen;
-	if (qdisc_copy_stats(skb, &q->stats, q->stats_lock))
+	q->qstats.qlen = q->q.qlen;
+
+	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
+			TCA_XSTATS, q->stats_lock, &d) < 0)
+		goto rtattr_failure;
+
+	if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
+#ifdef CONFIG_NET_ESTIMATOR
+	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
+#endif
+	    gnet_stats_copy_queue(&d, &q->qstats) < 0)
+		goto rtattr_failure;
+	
+	if (gnet_stats_finish_copy(&d) < 0)
 		goto rtattr_failure;
+	
 	nlh->nlmsg_len = skb->tail - b;
 	return skb->len;
 

[PATCH 3/5] PKT_SCHED: Introduce gen_replace_estimator

[Thomas Graf]

Introduces gen_replace_estimator.

Signed-off-by: Thomas Graf <tgraf@suug.ch>

diff -Nru linux-2.6.9-rc3-bk6.orig/include/net/gen_stats.h linux-2.6.9-rc3-bk6/include/net/gen_stats.h
--- linux-2.6.9-rc3-bk6.orig/include/net/gen_stats.h	2004-10-06 14:19:13.000000000 +0200
+++ linux-2.6.9-rc3-bk6/include/net/gen_stats.h	2004-10-06 20:34:13.000000000 +0200
@@ -41,5 +41,8 @@
 			     spinlock_t *stats_lock, struct rtattr *opt);
 extern void gen_kill_estimator(struct gnet_stats_basic *bstats,
 			       struct gnet_stats_rate_est *rate_est);
+extern int gen_replace_estimator(struct gnet_stats_basic *bstats,
+				 struct gnet_stats_rate_est *rate_est,
+				 spinlock_t *stats_lock, struct rtattr *opt);
 
 #endif
diff -Nru linux-2.6.9-rc3-bk6.orig/net/core/gen_estimator.c linux-2.6.9-rc3-bk6/net/core/gen_estimator.c
--- linux-2.6.9-rc3-bk6.orig/net/core/gen_estimator.c	2004-10-06 14:19:46.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/core/gen_estimator.c	2004-10-06 20:33:43.000000000 +0200
@@ -200,5 +200,16 @@
 	}
 }
 
+int
+gen_replace_estimator(struct gnet_stats_basic *bstats,
+	struct gnet_stats_rate_est *rate_est, spinlock_t *stats_lock,
+	struct rtattr *opt)
+{
+    gen_kill_estimator(bstats, rate_est);
+    return gen_new_estimator(bstats, rate_est, stats_lock, opt);
+}
+    
+
 EXPORT_SYMBOL(gen_kill_estimator);
 EXPORT_SYMBOL(gen_new_estimator);
+EXPORT_SYMBOL(gen_replace_estimator);

[PATCH 4/5] PKT_SCHED: Use generic rate estimator

[Thomas Graf]

Adapts qdiscs to use generic estimator.

Signed-off-by: Thomas Graf <tgraf@suug.ch>

diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_api.c linux-2.6.9-rc3-bk6/net/sched/sch_api.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_api.c	2004-10-07 00:48:10.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_api.c	2004-10-07 00:46:34.000000000 +0200
@@ -461,8 +461,8 @@
 
 #ifdef CONFIG_NET_ESTIMATOR
 		if (tca[TCA_RATE-1])
-			qdisc_new_estimator(&sch->stats, sch->stats_lock,
-					    tca[TCA_RATE-1]);
+			gen_new_estimator(&sch->bstats, &sch->rate_est,
+				sch->stats_lock, tca[TCA_RATE-1]);
 #endif
 		return sch;
 	}
@@ -489,11 +489,9 @@
 			return err;
 	}
 #ifdef CONFIG_NET_ESTIMATOR
-	if (tca[TCA_RATE-1]) {
-		qdisc_kill_estimator(&sch->stats);
-		qdisc_new_estimator(&sch->stats, sch->stats_lock,
-				    tca[TCA_RATE-1]);
-	}
+	if (tca[TCA_RATE-1])
+		gen_replace_estimator(&sch->bstats, &sch->rate_est,
+			sch->stats_lock, tca[TCA_RATE-1]);
 #endif
 	return 0;
 }
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_generic.c linux-2.6.9-rc3-bk6/net/sched/sch_generic.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_generic.c	2004-10-07 00:32:25.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_generic.c	2004-10-07 00:46:57.000000000 +0200
@@ -465,7 +465,7 @@
 	struct Qdisc_ops  *ops = qdisc->ops;
 
 #ifdef CONFIG_NET_ESTIMATOR
-	qdisc_kill_estimator(&qdisc->stats);
+	gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
 #endif
 	write_lock(&qdisc_tree_lock);
 	if (ops->reset)

[PATCH 5/5] PKT_SCHED: Qdisc are not supposed to dump TCA_STATS themselves

[Thomas Graf]

hfsc and htb qdisc are not supposed to copy TCA_STATS
on their own and queue length statistic is already
updated in generic code part.

Signed-off-by: Thomas Graf <tgraf@suug.ch>

diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_hfsc.c linux-2.6.9-rc3-bk6/net/sched/sch_hfsc.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_hfsc.c	2004-10-07 00:32:25.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_hfsc.c	2004-10-07 00:55:35.000000000 +0200
@@ -1653,11 +1653,6 @@
 
 	qopt.defcls = q->defcls;
 	RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
-
-	sch->stats.qlen = sch->q.qlen;
-	if (qdisc_copy_stats(skb, &sch->stats, sch->stats_lock) < 0)
-		goto rtattr_failure;
-
 	return skb->len;
 
  rtattr_failure:
diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_htb.c linux-2.6.9-rc3-bk6/net/sched/sch_htb.c
--- linux-2.6.9-rc3-bk6.orig/net/sched/sch_htb.c	2004-10-07 00:32:25.000000000 +0200
+++ linux-2.6.9-rc3-bk6/net/sched/sch_htb.c	2004-10-07 00:55:22.000000000 +0200
@@ -1332,8 +1332,6 @@
 	RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
 	RTA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt);
 	rta->rta_len = skb->tail - b;
-	sch->stats.qlen = sch->q.qlen;
-	RTA_PUT(skb, TCA_STATS, sizeof(sch->stats), &sch->stats);
 	HTB_QUNLOCK(sch);
 	return skb->len;
 rtattr_failure:

Re: [PATCH 5/5] PKT_SCHED: Qdisc are not supposed to dump TCA_STATS themselves

[jamal]

1-4 look good. 

Caveat to note Dave: both old TC_STAT as well as new TC_STAT2 will be
generated. Newer software can ignore TC_STAT and older s/ware can ignore
TC_STAT2.

patch 5 is a fix thats independent to 1-4 and should be applied
regardless of the status of 1-4

cheers,
jamal, semi-alive



On Wed, 2004-10-06 at 21:08, Thomas Graf wrote:
> hfsc and htb qdisc are not supposed to copy TCA_STATS
> on their own and queue length statistic is already
> updated in generic code part.
> 
> Signed-off-by: Thomas Graf <tgraf@suug.ch>
> 
> diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_hfsc.c linux-2.6.9-rc3-bk6/net/sched/sch_hfsc.c
> --- linux-2.6.9-rc3-bk6.orig/net/sched/sch_hfsc.c	2004-10-07 00:32:25.000000000 +0200
> +++ linux-2.6.9-rc3-bk6/net/sched/sch_hfsc.c	2004-10-07 00:55:35.000000000 +0200
> @@ -1653,11 +1653,6 @@
>  
>  	qopt.defcls = q->defcls;
>  	RTA_PUT(skb, TCA_OPTIONS, sizeof(qopt), &qopt);
> -
> -	sch->stats.qlen = sch->q.qlen;
> -	if (qdisc_copy_stats(skb, &sch->stats, sch->stats_lock) < 0)
> -		goto rtattr_failure;
> -
>  	return skb->len;
>  
>   rtattr_failure:
> diff -Nru linux-2.6.9-rc3-bk6.orig/net/sched/sch_htb.c linux-2.6.9-rc3-bk6/net/sched/sch_htb.c
> --- linux-2.6.9-rc3-bk6.orig/net/sched/sch_htb.c	2004-10-07 00:32:25.000000000 +0200
> +++ linux-2.6.9-rc3-bk6/net/sched/sch_htb.c	2004-10-07 00:55:22.000000000 +0200
> @@ -1332,8 +1332,6 @@
>  	RTA_PUT(skb, TCA_OPTIONS, 0, NULL);
>  	RTA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt);
>  	rta->rta_len = skb->tail - b;
> -	sch->stats.qlen = sch->q.qlen;
> -	RTA_PUT(skb, TCA_STATS, sizeof(sch->stats), &sch->stats);
>  	HTB_QUNLOCK(sch);
>  	return skb->len;
>  rtattr_failure:
> 
> 

Re: [PATCH 5/5] PKT_SCHED: Qdisc are not supposed to dump TCA_STATS themselves

[Thomas Graf]

> 1-4 look good. 
> 
> Caveat to note Dave: both old TC_STAT as well as new TC_STAT2 will be
> generated. Newer software can ignore TC_STAT and older s/ware can ignore
> TC_STAT2.

Correct, I tested all these cases and it works perfectly fine.

I also completed porting TCA_XSTATS to the new gnet_stats by
introducing dump_stats to Qdisc_ops. I ported your requeues patch
and added a missing update in SFQ qdisc. I will send those changes
once these have been approved. I will do the same for classes
and classifiers, although we will have to talk about it a little
as soon as you feel better again. I guess the classifier part will
involve changing the action code.

> patch 5 is a fix thats independent to 1-4 and should be applied
> regardless of the status of 1-4

Right, same for the cbq fix I sent a bit later which I have no clue how it is
possible that such a major bug can live for so long, I've been triggering
this bug for over 3 years and didn't notice anything until now :->

Re: [PATCH 5/5] PKT_SCHED: Qdisc are not supposed to dump TCA_STATS themselves

[jamal]

On Sat, 2004-10-09 at 12:26, Thomas Graf wrote:
> > 1-4 look good. 
> > 
> > Caveat to note Dave: both old TC_STAT as well as new TC_STAT2 will be
> > generated. Newer software can ignore TC_STAT and older s/ware can ignore
> > TC_STAT2.
> 
> Correct, I tested all these cases and it works perfectly fine.
> 
> I also completed porting TCA_XSTATS to the new gnet_stats by
> introducing dump_stats to Qdisc_ops. I ported your requeues patch
> and added a missing update in SFQ qdisc. I will send those changes
> once these have been approved. 

Much appreciated.

> I will do the same for classes
> and classifiers, although we will have to talk about it a little
> as soon as you feel better again. I guess the classifier part will
> involve changing the action code.

The action code change is trivial since the dump_stats exists. Located
in cls_api.c in call to copy_stats.
My recomendation is to to not even bother using old API since this is
new code. Go ahead and rip it out and just send TC_STAT2 _only_
Sorry, dont have the energy to do it right now.
 
> 
> > patch 5 is a fix thats independent to 1-4 and should be applied
> > regardless of the status of 1-4
> 
> Right, same for the cbq fix I sent a bit later which I have no clue how it is
> possible that such a major bug can live for so long, I've been triggering
> this bug for over 3 years and didn't notice anything until now :->

Youve heard of the TheLinuxWAY(tm) I hope ;-> Otherwise know as
cutnpaste. One has bug others follow ;->

BTW, you would be submitting the iproute2 patches as well, correct ?
i.e you are not just using your tool to do the testing?

cheers,
jamal

Re: [PATCH 5/5] PKT_SCHED: Qdisc are not supposed to dump TCA_STATS themselves

[Thomas Graf]

> The action code change is trivial since the dump_stats exists. Located
> in cls_api.c in call to copy_stats.
> My recomendation is to to not even bother using old API since this is
> new code. Go ahead and rip it out and just send TC_STAT2 _only_
> Sorry, dont have the energy to do it right now.

OK, will do, doesn't take too long I guess.

> Youve heard of the TheLinuxWAY(tm) I hope ;-> Otherwise know as
> cutnpaste. One has bug others follow ;->

I know, I was more amazed by how long a slab corruption can exist
without anyone noticing ;)

> BTW, you would be submitting the iproute2 patches as well, correct ?
> i.e you are not just using your tool to do the testing?

Sure, once the kernel part is done. I use iproute2 to check if the
backward compatibility is working and my tool in debug mode printing
out the hierarchy of the TLVs to see if the new stuff is all right.
Adding support to iproute will not take longer than 1-2 hours.

Re: [PATCH 0/5] PKT_SCHED: Convert Qdiscs to use generic network statistics/estimator

[David S. Miller]

On Thu, 7 Oct 2004 03:01:46 +0200
Thomas Graf <tgraf@suug.ch> wrote:

> This patchset converts Qdiscs to use the generic network
> statistics/estimator. A new TLV type TCA_STATS2 is introduced
> carrying the new statistics while the old TLV type TCA_STATS
> is still provided for backward compatibility.
> 
> Patch 1: Replace tc_stats with new gnet_stats in struct Qdisc
> Patch 2: Use gnet_stats API to copy statistics into netlink message
> Patch 3: Introduce gen_replace_estimator
> Patch 4: Use generic rate estimator
> Patch 5: Qdisc are not supposed to dump TCA_STATS themselves

All 5 patches applied, thanks Thomas.