[LARTC] Some work to extend SFQ to be more configurable

[Alexander Atanasov <alex@ssi.bg>]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1141 bytes --]

	Hi there!

On Tue, 4 Jun 2002, Martin Devera wrote:

> This is often discussed and is on "TODO" for someone ;)
> 
> >
> > SFQ is connection oriented. right?
> > Would be a good idea to make the queues per ip rather than per tcp flow?
> > So there would be per host fairnes.

	I've done some in this direction , probably needs more work, and
it's poorly tested - expect b00ms ;)
	
	This adds a new qdisc for now - esfq which is a 100% clone of
original sfq.
	- You can set all sfq parameters: hash table size, queue depths,
queue limits.
	- You can choose from 3 hash types: original(classic), dst ip, src
ip.
	Things to consider: perturbation with dst and src hashes is not
good IMHO, you can try with perturb 0 if it couses trouble.

	Please, see the attached files.

	Plaing with it gives interesting results:  
	higher depth -> makes flows equal slower
	small depth  -> makes flows equal faster
	limit kills big delays when set at about 75-85% of depth.

	Needs testings and mesurements - that's why i made it
separate qdisc and not a patch over sfq, i wanted to compare both.

	Any feedback good or bad is welcome. 

-- 
have fun,
alex

[-- Attachment #2: Type: TEXT/PLAIN, Size: 1838 bytes --]

	You need:
		iproute2-2.2.4-now-ss001007.tar.gz
		linux-2.4.18.tar.gz
	
	This may work with next versions too. 

	Example Step by Step install
	------------------------------------------------------------------
	Install kernel:
	tar zxvf linux-2.4.18.tar.gz
	cd linux
	cat linux-2.4.18-esfq.diff | patch -p1
	make menuconfig

	Now you have it in:
	Networking options  ---> QoS and/or fair queueing ---> ESFQ queue

	Configure and install kernel
	make dep clean bzImage modules_install
	cp System.map /boot/
	cp arch/i386/boot/bzImage /boot/bzImage
	
	Edit lilo.conf and add your new kernel.

	------------------------------------------------------------------
	Install iproute:
	tar zxvf iproute2-2.2.4-now-ss001007.tar.gz
	cd iproute2
	cat iproute2-2.2.4-now-ss001007-esfq.diff | patch -p1
	make
/* Note: If you get an error in lib/ll_proto.c comment _PF(ECHO,echo) */
	cp tc/tc your_favourite_iproute_path/tc
	cp ip/ip your_favourite_iproute_path/ip

	------------------------------------------------------------------
	
	Usage: ... esfq [ perturb SECS ] [ quantum BYTES ] [ depth FLOWS ]
        [ divisor HASHBITS ] [ limit PKTS ] [ hash HASHTYPE]

	Where:
	HASHTYPE := { classic | src | dst }

	Examples:
	tc qdisc add dev eth0 root esfq limit 128 depth 128 divisor 10 \
	 hash classic perturb 15

	Setups a classic SFQ.

        tc qdisc add dev eth0 root esfq limit 64 depth 64 divisor 11 \
         hash dst

	Setups a dst SFQ with limit and depth of 64 packets and
	11bits (2048 rows) hash table. 1:1 with sch_sfq.

	You can experiment with the values as you like to find
	the best which sfq can do for you.

	More can be found in:
	linux/net/sched/sch_sfq.c
	linux/net/sched/sch_esfq.c

	Limits: 
		- limit must be less than depth
		- divisor must be less than 15

[-- Attachment #3: Type: TEXT/PLAIN, Size: 4716 bytes --]

diff -urN iproute2.orig/tc/Makefile iproute2/tc/Makefile
--- iproute2.orig/tc/Makefile	Sun Apr 16 20:42:53 2000
+++ iproute2/tc/Makefile	Tue May 14 23:04:10 2002
@@ -5,6 +5,7 @@
 TCMODULES :=
 TCMODULES += q_fifo.o
 TCMODULES += q_sfq.o
+TCMODULES += q_esfq.o
 TCMODULES += q_red.o
 TCMODULES += q_prio.o
 TCMODULES += q_tbf.o
diff -urN iproute2.orig/tc/q_esfq.c iproute2/tc/q_esfq.c
--- iproute2.orig/tc/q_esfq.c	Thu Jan  1 02:00:00 1970
+++ iproute2/tc/q_esfq.c	Thu May 16 02:13:30 2002
@@ -0,0 +1,169 @@
+/*
+ * q_esfq.c		ESFQ.
+ *
+ *		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:	Alexander Atanasov, <alex@ssi.bg>
+ *		Added depth,limit,divisor,hash_kind options.
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <syslog.h>
+#include <fcntl.h>
+#include <math.h> 
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <arpa/inet.h>
+#include <string.h>
+
+#include "utils.h"
+#include "tc_util.h"
+
+static void explain(void)
+{
+	fprintf(stderr, "Usage: ... esfq [ perturb SECS ] [ quantum BYTES ] [ depth FLOWS ]\n\t[ divisor HASHBITS ] [ limit PKTS ] [ hash HASHTYPE]\n");
+	fprintf(stderr,"Where: \n");
+	fprintf(stderr,"HASHTYPE := { classic | src | dst }\n");
+}
+
+#define usage() return(-1)
+
+static int esfq_parse_opt(struct qdisc_util *qu, int argc, char **argv, struct nlmsghdr *n)
+{
+	int ok=0;
+	struct tc_sfq_qopt opt;
+
+	memset(&opt, 0, sizeof(opt));
+
+	opt.hash_kind= TCA_SFQ_HASH_CLASSIC;
+	
+	while (argc > 0) {
+		if (strcmp(*argv, "quantum") == 0) {
+			NEXT_ARG();
+			if (get_size(&opt.quantum, *argv)) {
+				fprintf(stderr, "Illegal \"quantum\"\n");
+				return -1;
+			}
+			ok++;
+		} else if (strcmp(*argv, "perturb") == 0) {
+			NEXT_ARG();
+			if (get_integer(&opt.perturb_period, *argv, 0)) {
+				fprintf(stderr, "Illegal \"perturb\"\n");
+				return -1;
+			}
+			ok++;
+		} else if (strcmp(*argv, "depth") == 0) {
+			NEXT_ARG();
+			if (get_integer(&opt.flows, *argv, 0)) {
+				fprintf(stderr, "Illegal \"depth\"\n");
+				return -1;
+			}
+			ok++;
+		} else if (strcmp(*argv, "divisor") == 0) {
+			NEXT_ARG();
+			if (get_integer(&opt.divisor, *argv, 0)) {
+				fprintf(stderr, "Illegal \"divisor\"\n");
+				return -1;
+			}
+			if(opt.divisor >= 15) {
+				fprintf(stderr, "Illegal \"divisor\" must be < 15\n");
+				return -1;
+			}
+			opt.divisor=pow(2,opt.divisor);
+			ok++;
+		} else if (strcmp(*argv, "limit") == 0) {
+			NEXT_ARG();
+			if (get_integer(&opt.limit, *argv, 0)) {
+				fprintf(stderr, "Illegal \"limit\"\n");
+				return -1;
+			}
+			ok++;
+		} else if (strcmp(*argv, "hash") == 0) {
+			NEXT_ARG();
+			if(strcmp(*argv,"classic") == 0) {
+				opt.hash_kind= TCA_SFQ_HASH_CLASSIC;
+			} else 
+			if(strcmp(*argv,"dst") == 0) {
+				opt.hash_kind= TCA_SFQ_HASH_DST;
+			} else
+			if(strcmp(*argv,"src") == 0) {
+				opt.hash_kind= TCA_SFQ_HASH_SRC;
+			} else {
+				fprintf(stderr, "Illegal \"hash\"\n");
+				explain();
+				return -1;
+			}
+			ok++;
+		} else if (strcmp(*argv, "help") == 0) {
+			explain();
+			return -1;
+		} else {
+			fprintf(stderr, "What is \"%s\"?\n", *argv);
+			explain();
+			return -1;
+		}
+		argc--; argv++;
+	}
+
+	if (ok)
+		addattr_l(n, 1024, TCA_OPTIONS, &opt, sizeof(opt));
+	return 0;
+}
+
+static int esfq_print_opt(struct qdisc_util *qu, FILE *f, struct rtattr *opt)
+{
+	struct tc_sfq_qopt *qopt;
+	SPRINT_BUF(b1);
+
+	if (opt == NULL)
+		return 0;
+
+	if (RTA_PAYLOAD(opt)  < sizeof(*qopt))
+		return -1;
+	qopt = RTA_DATA(opt);
+	fprintf(f, "quantum %s ", sprint_size(qopt->quantum, b1));
+	if (show_details) {
+		fprintf(f, "limit %up flows %u/%u ",
+			qopt->limit, qopt->flows, qopt->divisor);
+	}
+	if (qopt->perturb_period)
+		fprintf(f, "perturb %dsec ", qopt->perturb_period);
+
+		fprintf(f,"hash: ");
+	switch(qopt->hash_kind)
+	{
+	case TCA_SFQ_HASH_CLASSIC:
+		fprintf(f,"classic");
+		break;
+	case TCA_SFQ_HASH_DST:
+		fprintf(f,"dst");
+		break;
+	case TCA_SFQ_HASH_SRC:
+		fprintf(f,"src");
+		break;
+	default:
+		fprintf(f,"Unknown");
+	}
+	return 0;
+}
+
+static int esfq_print_xstats(struct qdisc_util *qu, FILE *f, struct rtattr *xstats)
+{
+	return 0;
+}
+
+
+struct qdisc_util esfq_util = {
+	NULL,
+	"esfq",
+	esfq_parse_opt,
+	esfq_print_opt,
+	esfq_print_xstats,
+};

[-- Attachment #4: Type: TEXT/PLAIN, Size: 17783 bytes --]

--- linux-2.4.18/include/linux/pkt_sched.h.orig	Tue May 14 23:25:13 2002
+++ linux-2.4.18/include/linux/pkt_sched.h	Tue May 14 23:34:57 2002
@@ -157,6 +157,13 @@
 
 /* SFQ section */
 
+enum
+{
+	TCA_SFQ_HASH_CLASSIC,
+	TCA_SFQ_HASH_DST,
+	TCA_SFQ_HASH_SRC,
+};
+
 struct tc_sfq_qopt
 {
 	unsigned	quantum;	/* Bytes per round allocated to flow */
@@ -164,6 +171,7 @@
 	__u32		limit;		/* Maximal packets in queue */
 	unsigned	divisor;	/* Hash divisor  */
 	unsigned	flows;		/* Maximal number of flows  */
+	unsigned	hash_kind;	/* Hash function to use for flow identification */
 };
 
 /*
@@ -173,6 +181,8 @@
  *
  *	The only reason for this is efficiency, it is possible
  *	to change these parameters in compile time.
+ *	
+ *	If you need to play with this values use esfq.
  */
 
 /* RED section */
--- linux-2.4.18/net/ipv4/netfilter/ipchains_core.c.orig	Fri May 24 19:27:01 2002
+++ linux-2.4.18/net/ipv4/netfilter/ipchains_core.c	Fri May 24 19:31:24 2002
@@ -723,6 +723,7 @@
 						      src_port, dst_port,
 						      count, tcpsyn)) {
 					ret = FW_BLOCK;
+					cleanup(chain, 0, slot);
 					goto out;
 				}
 				break;
--- linux-2.4.18/net/sched/Makefile.orig	Tue May 14 23:06:55 2002
+++ linux-2.4.18/net/sched/Makefile	Tue May 14 23:07:08 2002
@@ -17,6 +17,7 @@
 obj-$(CONFIG_NET_SCH_HPFQ)	+= sch_hpfq.o
 obj-$(CONFIG_NET_SCH_HFSC)	+= sch_hfsc.o
 obj-$(CONFIG_NET_SCH_SFQ)	+= sch_sfq.o
+obj-$(CONFIG_NET_SCH_ESFQ)	+= sch_esfq.o
 obj-$(CONFIG_NET_SCH_RED)	+= sch_red.o
 obj-$(CONFIG_NET_SCH_TBF)	+= sch_tbf.o
 obj-$(CONFIG_NET_SCH_PRIO)	+= sch_prio.o
--- linux-2.4.18/net/sched/Config.in.orig	Tue May 14 23:07:15 2002
+++ linux-2.4.18/net/sched/Config.in	Tue May 14 23:09:03 2002
@@ -11,6 +11,7 @@
 tristate '  The simplest PRIO pseudoscheduler' CONFIG_NET_SCH_PRIO
 tristate '  RED queue' CONFIG_NET_SCH_RED
 tristate '  SFQ queue' CONFIG_NET_SCH_SFQ
+tristate '  ESFQ queue' CONFIG_NET_SCH_ESFQ
 tristate '  TEQL queue' CONFIG_NET_SCH_TEQL
 tristate '  TBF queue' CONFIG_NET_SCH_TBF
 tristate '  GRED queue' CONFIG_NET_SCH_GRED
--- linux-2.4.18/Documentation/Configure.help.orig	Thu May 16 01:37:22 2002
+++ linux-2.4.18/Documentation/Configure.help	Mon May 27 01:09:03 2002
@@ -9433,6 +9433,24 @@
   whenever you want). If you want to compile it as a module, say M
   here and read <file:Documentation/modules.txt>.
 
+ESFQ queue
+CONFIG_NET_SCH_ESFQ
+  Say Y here if you want to use the Stochastic Fairness Queueing (SFQ)
+  packet scheduling algorithm for some of your network devices or as a
+  leaf discipline for the CBQ scheduling algorithm (see the top of
+  <file:net/sched/sch_esfq.c> for details and references about the SFQ
+  algorithm).
+
+  This is an enchanced SFQ version which allows you to control the
+  hardcoded values in the SFQ scheduler: queue depth, hash table size,
+  queues limit. Also adds control to the hash function used to identify
+  packet flows. Hash by src or dst ip and original sfq hash.
+
+  This code is also available as a module called sch_esfq.o ( = code
+  which can be inserted in and removed from the running kernel
+  whenever you want). If you want to compile it as a module, say M
+  here and read <file:Documentation/modules.txt>.
+
 TEQL queue
 CONFIG_NET_SCH_TEQL
   Say Y here if you want to use the True Link Equalizer (TLE) packet
--- /dev/null	Mon Jul 18 02:46:18 1994
+++ linux-2.4.18/net/sched/sch_esfq.c	Mon May 27 01:49:19 2002
@@ -0,0 +1,591 @@
+/*
+ * net/sched/sch_esfq.c	Extended Stochastic Fairness Queueing discipline.
+ *
+ *		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:	Alexander Atanasov, <alex@ssi.bg>
+ *		Added dynamic depth,limit,divisor,hash_kind options.
+ *		Added dst and src hashes.
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <asm/uaccess.h>
+#include <asm/system.h>
+#include <asm/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 <linux/init.h>
+#include <net/ip.h>
+#include <linux/ipv6.h>
+#include <net/route.h>
+#include <linux/skbuff.h>
+#include <net/sock.h>
+#include <net/pkt_sched.h>
+
+
+/*	Stochastic Fairness Queuing algorithm.
+	For more comments look at sch_sfq.c.
+	The difference is that you can change limit, depth,
+	hash table size and choose 3 hash types.
+	
+	classic:	same as in sch_sfq.c
+	dst:		destination IP address
+	src:		source IP address
+	
+	TODO: 
+		make sfq_change work.
+*/
+
+
+/* This type should contain at least SFQ_DEPTH*2 values */
+typedef unsigned int esfq_index;
+
+struct esfq_head
+{
+	esfq_index	next;
+	esfq_index	prev;
+};
+
+struct esfq_sched_data
+{
+/* Parameters */
+	int		perturb_period;
+	unsigned	quantum;	/* Allotment per round: MUST BE >= MTU */
+	int		limit;
+	unsigned	depth;
+	unsigned	hash_divisor;
+	unsigned	hash_kind;
+/* Variables */
+	struct timer_list perturb_timer;
+	int		perturbation;
+	esfq_index	tail;		/* Index of current slot in round */
+	esfq_index	max_depth;	/* Maximal depth */
+
+	esfq_index	*ht;			/* Hash table */
+	esfq_index	*next;			/* Active slots link */
+	short		*allot;			/* Current allotment per slot */
+	unsigned short	*hash;			/* Hash value indexed by slots */
+	struct sk_buff_head	*qs;		/* Slot queue */
+	struct esfq_head	*dep;		/* Linked list of slots, indexed by depth */
+};
+
+static __inline__ unsigned esfq_hash_u32(struct esfq_sched_data *q,u32 h)
+{
+	int pert = q->perturbation;
+	
+	if (pert)
+		h = (h<<pert) ^ (h>>(0x1F - pert));
+	h = ntohl(h) * 2654435761UL;
+	return h & (q->hash_divisor-1);
+}
+
+static __inline__ unsigned esfq_fold_hash_classic(struct esfq_sched_data *q, u32 h, u32 h1)
+{
+	int pert = q->perturbation;
+
+	/* Have we any rotation primitives? If not, WHY? */
+	h ^= (h1<<pert) ^ (h1>>(0x1F - pert));
+	h ^= h>>10;
+	return h & (q->hash_divisor-1);
+}
+
+#ifndef IPPROTO_ESP
+#define IPPROTO_ESP 50
+#endif
+
+static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb)
+{
+	u32 h, h2;
+	u32 hs;
+	
+	switch (skb->protocol) {
+	case __constant_htons(ETH_P_IP):
+	{
+		struct iphdr *iph = skb->nh.iph;
+		h = iph->daddr;
+		hs = iph->saddr;
+		h2 = hs^iph->protocol;
+		if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
+		    (iph->protocol == IPPROTO_TCP ||
+		     iph->protocol == IPPROTO_UDP ||
+		     iph->protocol == IPPROTO_ESP))
+			h2 ^= *(((u32*)iph) + iph->ihl);
+		break;
+	}
+	case __constant_htons(ETH_P_IPV6):
+	{
+		struct ipv6hdr *iph = skb->nh.ipv6h;
+		h = iph->daddr.s6_addr32[3];
+		hs = iph->saddr.s6_addr32[3];
+		h2 = hs^iph->nexthdr;
+		if (iph->nexthdr == IPPROTO_TCP ||
+		    iph->nexthdr == IPPROTO_UDP ||
+		    iph->nexthdr == IPPROTO_ESP)
+			h2 ^= *(u32*)&iph[1];
+		break;
+	}
+	default:
+		h = (u32)(unsigned long)skb->dst;
+		hs = (u32)(unsigned long)skb->sk;
+		h2 = hs^skb->protocol;
+	}
+	switch(q->hash_kind)
+	{
+	case TCA_SFQ_HASH_CLASSIC:
+		return esfq_fold_hash_classic(q, h, h2);
+	case TCA_SFQ_HASH_DST:
+		return esfq_hash_u32(q,h);
+	case TCA_SFQ_HASH_SRC:
+		return esfq_hash_u32(q,hs);
+	default:
+		if (net_ratelimit())
+			printk(KERN_DEBUG "esfq unknown hash method, fallback to classic\n");
+	}
+	return esfq_fold_hash_classic(q, h, h2);
+}
+
+extern __inline__ void esfq_link(struct esfq_sched_data *q, esfq_index x)
+{
+	esfq_index p, n;
+	int d = q->qs[x].qlen + q->depth;
+
+	p = d;
+	n = q->dep[d].next;
+	q->dep[x].next = n;
+	q->dep[x].prev = p;
+	q->dep[p].next = q->dep[n].prev = x;
+}
+
+extern __inline__ void esfq_dec(struct esfq_sched_data *q, esfq_index x)
+{
+	esfq_index p, n;
+
+	n = q->dep[x].next;
+	p = q->dep[x].prev;
+	q->dep[p].next = n;
+	q->dep[n].prev = p;
+
+	if (n == p && q->max_depth == q->qs[x].qlen + 1)
+		q->max_depth--;
+
+	esfq_link(q, x);
+}
+
+extern __inline__ void esfq_inc(struct esfq_sched_data *q, esfq_index x)
+{
+	esfq_index p, n;
+	int d;
+
+	n = q->dep[x].next;
+	p = q->dep[x].prev;
+	q->dep[p].next = n;
+	q->dep[n].prev = p;
+	d = q->qs[x].qlen;
+	if (q->max_depth < d)
+		q->max_depth = d;
+
+	esfq_link(q, x);
+}
+
+static int esfq_drop(struct Qdisc *sch)
+{
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+	esfq_index d = q->max_depth;
+	struct sk_buff *skb;
+
+	/* Queue is full! Find the longest slot and
+	   drop a packet from it */
+
+	if (d > 1) {
+		esfq_index x = q->dep[d+q->depth].next;
+		skb = q->qs[x].prev;
+		__skb_unlink(skb, &q->qs[x]);
+		kfree_skb(skb);
+		esfq_dec(q, x);
+		sch->q.qlen--;
+		sch->stats.drops++;
+		return 1;
+	}
+
+	if (d == 1) {
+		/* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
+		d = q->next[q->tail];
+		q->next[q->tail] = q->next[d];
+		q->allot[q->next[d]] += q->quantum;
+		skb = q->qs[d].prev;
+		__skb_unlink(skb, &q->qs[d]);
+		kfree_skb(skb);
+		esfq_dec(q, d);
+		sch->q.qlen--;
+		q->ht[q->hash[d]] = q->depth;
+		sch->stats.drops++;
+		return 1;
+	}
+
+	return 0;
+}
+
+static int
+esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
+{
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+	unsigned hash = esfq_hash(q, skb);
+	unsigned depth = q->depth;
+	esfq_index x;
+
+	x = q->ht[hash];
+	if (x == depth) {
+		q->ht[hash] = x = q->dep[depth].next;
+		q->hash[x] = hash;
+	}
+	__skb_queue_tail(&q->qs[x], skb);
+	esfq_inc(q, x);
+	if (q->qs[x].qlen == 1) {		/* The flow is new */
+		if (q->tail == depth) {	/* It is the first flow */
+			q->tail = x;
+			q->next[x] = x;
+			q->allot[x] = q->quantum;
+		} else {
+			q->next[x] = q->next[q->tail];
+			q->next[q->tail] = x;
+			q->tail = x;
+		}
+	}
+	if (++sch->q.qlen < q->limit-1) {
+		sch->stats.bytes += skb->len;
+		sch->stats.packets++;
+		return 0;
+	}
+
+	esfq_drop(sch);
+	return NET_XMIT_CN;
+}
+
+static int
+esfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
+{
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+	unsigned hash = esfq_hash(q, skb);
+	unsigned depth = q->depth;
+	esfq_index x;
+
+	x = q->ht[hash];
+	if (x == depth) {
+		q->ht[hash] = x = q->dep[depth].next;
+		q->hash[x] = hash;
+	}
+	__skb_queue_head(&q->qs[x], skb);
+	esfq_inc(q, x);
+	if (q->qs[x].qlen == 1) {		/* The flow is new */
+		if (q->tail == depth) {	/* It is the first flow */
+			q->tail = x;
+			q->next[x] = x;
+			q->allot[x] = q->quantum;
+		} else {
+			q->next[x] = q->next[q->tail];
+			q->next[q->tail] = x;
+			q->tail = x;
+		}
+	}
+	if (++sch->q.qlen < q->limit - 1)
+		return 0;
+
+	sch->stats.drops++;
+	esfq_drop(sch);
+	return NET_XMIT_CN;
+}
+
+
+
+
+static struct sk_buff *
+esfq_dequeue(struct Qdisc* sch)
+{
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+	struct sk_buff *skb;
+	unsigned depth = q->depth;
+	esfq_index a, old_a;
+
+	/* No active slots */
+	if (q->tail == depth)
+		return NULL;
+	
+	a = old_a = q->next[q->tail];
+	
+	/* Grab packet */
+	skb = __skb_dequeue(&q->qs[a]);
+	esfq_dec(q, a);
+	sch->q.qlen--;
+	
+	/* Is the slot empty? */
+	if (q->qs[a].qlen == 0) {
+		a = q->next[a];
+		if (a == old_a) {
+			q->tail = depth;
+			return skb;
+		}
+		q->next[q->tail] = a;
+		q->allot[a] += q->quantum;
+	} else if ((q->allot[a] -= skb->len) <= 0) {
+		q->tail = a;
+		a = q->next[a];
+		q->allot[a] += q->quantum;
+	}
+	
+	return skb;
+}
+
+static void
+esfq_reset(struct Qdisc* sch)
+{
+	struct sk_buff *skb;
+
+	while ((skb = esfq_dequeue(sch)) != NULL)
+		kfree_skb(skb);
+}
+
+static void esfq_perturbation(unsigned long arg)
+{
+	struct Qdisc *sch = (struct Qdisc*)arg;
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+
+	q->perturbation = net_random()&0x1F;
+	q->perturb_timer.expires = jiffies + q->perturb_period;
+
+	if (q->perturb_period) {
+		q->perturb_timer.expires = jiffies + q->perturb_period;
+		add_timer(&q->perturb_timer);
+	}
+}
+
+static int esfq_change(struct Qdisc *sch, struct rtattr *opt)
+{
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+	struct tc_sfq_qopt *ctl = RTA_DATA(opt);
+	int old_perturb = q->perturb_period;
+	
+	if (opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
+		return -EINVAL;
+	
+	sch_tree_lock(sch);
+	q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
+	q->perturb_period = ctl->perturb_period*HZ;
+//	q->hash_divisor = ctl->divisor;
+//	q->tail = q->limit = q->depth = ctl->flows;
+	
+	if (ctl->limit)
+		q->limit = min_t(u32, ctl->limit, q->depth);
+	
+	if (ctl->hash_kind) {
+		q->hash_kind = ctl->hash_kind;
+		if (q->hash_kind !=  TCA_SFQ_HASH_CLASSIC)
+			q->perturb_period = 0;
+	}
+	
+	// is sch_tree_lock enough to do this ?
+	while (sch->q.qlen >= q->limit-1)
+		esfq_drop(sch);
+	
+	if (old_perturb)
+		del_timer(&q->perturb_timer);
+	if (q->perturb_period) {
+		q->perturb_timer.expires = jiffies + q->perturb_period;
+		add_timer(&q->perturb_timer);
+	} else {
+		q->perturbation = 0;
+	}
+	sch_tree_unlock(sch);
+	return 0;
+}
+
+static int esfq_init(struct Qdisc *sch, struct rtattr *opt)
+{
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+	struct tc_sfq_qopt *ctl;
+	esfq_index p = ~0UL/2;
+	int i;
+	
+	if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
+		return -EINVAL;
+	q->perturb_timer.data = (unsigned long)sch;
+	q->perturb_timer.function = esfq_perturbation;
+	init_timer(&q->perturb_timer);
+	q->perturbation = 0;
+	q->hash_kind = TCA_SFQ_HASH_CLASSIC;
+	q->max_depth = 0;
+	if (opt == NULL) {
+		q->quantum = psched_mtu(sch->dev);
+		q->perturb_period = 0;
+		q->hash_divisor = 1024;
+		q->tail = q->limit = q->depth = 128;
+		
+	} else {
+		ctl = RTA_DATA(opt);
+		q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
+		q->perturb_period = ctl->perturb_period*HZ;
+		q->hash_divisor = ctl->divisor;
+		q->tail = q->limit = q->depth = ctl->flows;
+		
+		if ( q->depth > p - 1 )
+			return -EINVAL;
+		
+		if (ctl->limit)
+			q->limit = min_t(u32, ctl->limit, q->depth);
+		
+		if (ctl->hash_kind) {
+			q->hash_kind = ctl->hash_kind;
+			if (q->hash_kind !=  TCA_SFQ_HASH_CLASSIC)
+				q->perturb_period = 0;
+		}
+		
+		while (sch->q.qlen >= q->limit-1)
+			esfq_drop(sch);
+		
+		if (q->perturb_period) {
+			q->perturb_timer.expires = jiffies + q->perturb_period;
+			add_timer(&q->perturb_timer);
+		}
+	}
+	
+	q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL);
+	if (!q->ht)
+		goto err_case;
+		
+	q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL);
+	if (!q->dep)
+		goto err_case;
+	q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL);
+	if (!q->next)
+		goto err_case;
+	
+	q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL);
+	if (!q->allot)
+		goto err_case;
+	q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL);
+	if (!q->hash)
+		goto err_case;
+	q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL);
+	if (!q->qs)
+		goto err_case;
+	
+	for (i=0; i< q->hash_divisor; i++)
+		q->ht[i] = q->depth;
+	for (i=0; i<q->depth; i++) {
+		skb_queue_head_init(&q->qs[i]);
+		q->dep[i+q->depth].next = i+q->depth;
+		q->dep[i+q->depth].prev = i+q->depth;
+	}
+	
+	for (i=0; i<q->depth; i++)
+		esfq_link(q, i);
+	MOD_INC_USE_COUNT;
+	return 0;
+err_case:
+	if (q->ht)
+		kfree(q->ht);
+	if (q->dep)
+		kfree(q->dep);
+	if (q->next)
+		kfree(q->next);
+	if (q->allot)
+		kfree(q->allot);
+	if (q->hash)
+		kfree(q->hash);
+	if (q->qs)
+		kfree(q->qs);
+	return -ENOBUFS;
+}
+
+static void esfq_destroy(struct Qdisc *sch)
+{
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+	del_timer(&q->perturb_timer);
+	if(q->ht)
+		kfree(q->ht);
+	if(q->dep)
+		kfree(q->dep);
+	if(q->next)
+		kfree(q->next);
+	if(q->allot)
+		kfree(q->allot);
+	if(q->hash)
+		kfree(q->hash);
+	if(q->qs)
+		kfree(q->qs);
+	MOD_DEC_USE_COUNT;
+}
+
+static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb)
+{
+	struct esfq_sched_data *q = (struct esfq_sched_data *)sch->data;
+	unsigned char	 *b = skb->tail;
+	struct tc_sfq_qopt opt;
+
+	opt.quantum = q->quantum;
+	opt.perturb_period = q->perturb_period/HZ;
+
+	opt.limit = q->limit;
+	opt.divisor = q->hash_divisor;
+	opt.flows = q->depth;
+	opt.hash_kind = q->hash_kind;
+
+	RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
+
+	return skb->len;
+
+rtattr_failure:
+	skb_trim(skb, b - skb->data);
+	return -1;
+}
+
+struct Qdisc_ops esfq_qdisc_ops =
+{
+	NULL,
+	NULL,
+	"esfq",
+	sizeof(struct esfq_sched_data),
+
+	esfq_enqueue,
+	esfq_dequeue,
+	esfq_requeue,
+	esfq_drop,
+
+	esfq_init,
+	esfq_reset,
+	esfq_destroy,
+	NULL, /* esfq_change - needs more work */
+
+	esfq_dump,
+};
+
+#ifdef MODULE
+int init_module(void)
+{
+	return register_qdisc(&esfq_qdisc_ops);
+}
+
+void cleanup_module(void) 
+{
+	unregister_qdisc(&esfq_qdisc_ops);
+}
+#endif
+MODULE_LICENSE("GPL");