target/linux/generic-2.4/patches/621-tc_esfq.patch

   1 Index: linux-2.4.35.4/Documentation/Configure.help
   2 ===================================================================
   3 --- linux-2.4.35.4.orig/Documentation/Configure.help    2007-12-15 05:20:12.260538930 +0100
   4 +++ linux-2.4.35.4/Documentation/Configure.help 2007-12-15 05:20:12.604558535 +0100
   5 @@ -11153,6 +11153,24 @@
   6    whenever you want).  If you want to compile it as a module, say M
   7    here and read <file:Documentation/modules.txt>.
   8
   9 +ESFQ queue
  10 +CONFIG_NET_SCH_ESFQ
  11 +  Say Y here if you want to use the Stochastic Fairness Queueing (SFQ)
  12 +  packet scheduling algorithm for some of your network devices or as a
  13 +  leaf discipline for the CBQ scheduling algorithm (see the top of
  14 +  <file:net/sched/sch_esfq.c> for details and references about the SFQ
  15 +  algorithm).
  16 +
  17 +  This is an enchanced SFQ version which allows you to control the
  18 +  hardcoded values in the SFQ scheduler: queue depth, hash table size,
  19 +  queues limit. Also adds control to the hash function used to identify
  20 +  packet flows. Hash by src or dst ip and original sfq hash.
  21 +
  22 +  This code is also available as a module called sch_esfq.o ( = code
  23 +  which can be inserted in and removed from the running kernel
  24 +  whenever you want). If you want to compile it as a module, say M
  25 +  here and read <file:Documentation/modules.txt>.
  26 +
  27  CSZ packet scheduler
  28  CONFIG_NET_SCH_CSZ
  29    Say Y here if you want to use the Clark-Shenker-Zhang (CSZ) packet
  30 Index: linux-2.4.35.4/include/linux/pkt_sched.h
  31 ===================================================================
  32 --- linux-2.4.35.4.orig/include/linux/pkt_sched.h       2007-12-15 05:19:36.750515331 +0100
  33 +++ linux-2.4.35.4/include/linux/pkt_sched.h    2007-12-15 05:20:12.608558764 +0100
  34 @@ -173,8 +173,36 @@
  35   *
  36   *     The only reason for this is efficiency, it is possible
  37   *     to change these parameters in compile time.
  38 + *
  39 + *     If you need to play with these values use esfq instead.
  40   */
  41
  42 +/* ESFQ section */
  43 +
  44 +enum
  45 +{
  46 +       /* traditional */
  47 +       TCA_SFQ_HASH_CLASSIC,
  48 +       TCA_SFQ_HASH_DST,
  49 +       TCA_SFQ_HASH_SRC,
  50 +       /* conntrack */
  51 +       TCA_SFQ_HASH_CTORIGDST,
  52 +       TCA_SFQ_HASH_CTORIGSRC,
  53 +       TCA_SFQ_HASH_CTREPLDST,
  54 +       TCA_SFQ_HASH_CTREPLSRC,
  55 +       TCA_SFQ_HASH_CTNATCHG,
  56 +};
  57 +
  58 +struct tc_esfq_qopt
  59 +{
  60 +       unsigned        quantum;        /* Bytes per round allocated to flow */
  61 +       int             perturb_period; /* Period of hash perturbation */
  62 +       __u32           limit;          /* Maximal packets in queue */
  63 +       unsigned        divisor;        /* Hash divisor  */
  64 +       unsigned        flows;          /* Maximal number of flows  */
  65 +       unsigned        hash_kind;      /* Hash function to use for flow identification */
  66 +};
  67 +
  68  /* RED section */
  69
  70  enum
  71 Index: linux-2.4.35.4/net/sched/Config.in
  72 ===================================================================
  73 --- linux-2.4.35.4.orig/net/sched/Config.in     2007-12-15 05:19:36.758515786 +0100
  74 +++ linux-2.4.35.4/net/sched/Config.in  2007-12-15 05:20:12.608558764 +0100
  75 @@ -12,6 +12,7 @@
  76  tristate '  The simplest PRIO pseudoscheduler' CONFIG_NET_SCH_PRIO
  77  tristate '  RED queue' CONFIG_NET_SCH_RED
  78  tristate '  SFQ queue' CONFIG_NET_SCH_SFQ
  79 +tristate '  ESFQ queue' CONFIG_NET_SCH_ESFQ
  80  tristate '  TEQL queue' CONFIG_NET_SCH_TEQL
  81  tristate '  TBF queue' CONFIG_NET_SCH_TBF
  82  tristate '  GRED queue' CONFIG_NET_SCH_GRED
  83 Index: linux-2.4.35.4/net/sched/Makefile
  84 ===================================================================
  85 --- linux-2.4.35.4.orig/net/sched/Makefile      2007-12-15 05:19:36.766516242 +0100
  86 +++ linux-2.4.35.4/net/sched/Makefile   2007-12-15 05:20:12.608558764 +0100
  87 @@ -19,6 +19,7 @@
  88  obj-$(CONFIG_NET_SCH_HFSC)     += sch_hfsc.o
  89  obj-$(CONFIG_NET_SCH_HTB)      += sch_htb.o
  90  obj-$(CONFIG_NET_SCH_SFQ)      += sch_sfq.o
  91 +obj-$(CONFIG_NET_SCH_ESFQ)     += sch_esfq.o
  92  obj-$(CONFIG_NET_SCH_RED)      += sch_red.o
  93  obj-$(CONFIG_NET_SCH_TBF)      += sch_tbf.o
  94  obj-$(CONFIG_NET_SCH_PRIO)     += sch_prio.o
  95 Index: linux-2.4.35.4/net/sched/sch_esfq.c
  96 ===================================================================
  97 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
  98 +++ linux-2.4.35.4/net/sched/sch_esfq.c 2007-12-15 05:20:12.608558764 +0100
  99 @@ -0,0 +1,649 @@
 100 +/*
 101 + * net/sched/sch_esfq.c        Extended Stochastic Fairness Queueing discipline.
 102 + *
 103 + *             This program is free software; you can redistribute it and/or
 104 + *             modify it under the terms of the GNU General Public License
 105 + *             as published by the Free Software Foundation; either version
 106 + *             2 of the License, or (at your option) any later version.
 107 + *
 108 + * Authors:    Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 109 + *
 110 + * Changes:    Alexander Atanasov, <alex@ssi.bg>
 111 + *             Added dynamic depth,limit,divisor,hash_kind options.
 112 + *             Added dst and src hashes.
 113 + *
 114 + *             Alexander Clouter, <alex@digriz.org.uk>
 115 + *             Ported ESFQ to Linux 2.6.
 116 + *
 117 + *             Corey Hickey, <bugfood-c@fatooh.org>
 118 + *             Maintenance of the Linux 2.6 port.
 119 + *             Added fwmark hash (thanks to Robert Kurjata).
 120 + *             Added usage of jhash.
 121 + *
 122 + */
 123 +
 124 +#include <linux/config.h>
 125 +#include <linux/module.h>
 126 +#include <asm/uaccess.h>
 127 +#include <asm/system.h>
 128 +#include <linux/bitops.h>
 129 +#include <linux/types.h>
 130 +#include <linux/kernel.h>
 131 +#include <linux/sched.h>
 132 +#include <linux/string.h>
 133 +#include <linux/mm.h>
 134 +#include <linux/socket.h>
 135 +#include <linux/sockios.h>
 136 +#include <linux/in.h>
 137 +#include <linux/errno.h>
 138 +#include <linux/interrupt.h>
 139 +#include <linux/if_ether.h>
 140 +#include <linux/inet.h>
 141 +#include <linux/netdevice.h>
 142 +#include <linux/etherdevice.h>
 143 +#include <linux/notifier.h>
 144 +#include <linux/init.h>
 145 +#include <net/ip.h>
 146 +#include <net/route.h>
 147 +#include <linux/skbuff.h>
 148 +#include <net/sock.h>
 149 +#include <net/pkt_sched.h>
 150 +#include <linux/jhash.h>
 151 +
 152 +#define IPPROTO_DCCP 33
 153 +#define qdisc_priv(q)   ((void *)(q->data))
 154 +
 155 +#ifdef CONFIG_IP_NF_CONNTRACK
 156 +/* #include <net/netfilter/nf_conntrack.h> */
 157 +#include <linux/netfilter_ipv4/ip_conntrack.h>
 158 +#endif
 159 +
 160 +/*     Stochastic Fairness Queuing algorithm.
 161 +       For more comments look at sch_sfq.c.
 162 +       The difference is that you can change limit, depth,
 163 +       hash table size and choose alternate hash types.
 164 +
 165 +       classic:        same as in sch_sfq.c
 166 +       dst:            destination IP address
 167 +       src:            source IP address
 168 +       ctorigdst:      original destination IP address
 169 +       ctorigsrc:      original source IP address
 170 +       ctrepldst:      reply destination IP address
 171 +       ctreplsrc:      reply source IP
 172 +       ctnatchg:       use the address which changed via nat
 173 +
 174 +*/
 175 +
 176 +
 177 +/* This type should contain at least SFQ_DEPTH*2 values */
 178 +typedef unsigned int esfq_index;
 179 +
 180 +struct esfq_head
 181 +{
 182 +       esfq_index      next;
 183 +       esfq_index      prev;
 184 +};
 185 +
 186 +struct esfq_sched_data
 187 +{
 188 +/* Parameters */
 189 +       int             perturb_period;
 190 +       unsigned        quantum;        /* Allotment per round: MUST BE >= MTU */
 191 +       int             limit;
 192 +       unsigned        depth;
 193 +       unsigned        hash_divisor;
 194 +       unsigned        hash_kind;
 195 +/* Variables */
 196 +       struct timer_list perturb_timer;
 197 +       int             perturbation;
 198 +       esfq_index      tail;           /* Index of current slot in round */
 199 +       esfq_index      max_depth;      /* Maximal depth */
 200 +
 201 +       esfq_index      *ht;                    /* Hash table */
 202 +       esfq_index      *next;                  /* Active slots link */
 203 +       short           *allot;                 /* Current allotment per slot */
 204 +       unsigned short  *hash;                  /* Hash value indexed by slots */
 205 +       struct sk_buff_head     *qs;            /* Slot queue */
 206 +       struct esfq_head        *dep;           /* Linked list of slots, indexed by depth */
 207 +       unsigned        dyn_min;        /* For dynamic divisor adjustment; minimum value seen */
 208 +       unsigned        dyn_max;        /*                                 maximum value seen */
 209 +       unsigned        dyn_range;      /*                                 saved range */
 210 +};
 211 +
 212 +/* This contains the info we will hash. */
 213 +struct esfq_packet_info
 214 +{
 215 +       u32     proto;          /* protocol or port */
 216 +       u32     src;            /* source from packet header */
 217 +       u32     dst;            /* destination from packet header */
 218 +       u32     ctorigsrc;      /* original source from conntrack */
 219 +       u32     ctorigdst;      /* original destination from conntrack */
 220 +       u32     ctreplsrc;      /* reply source from conntrack */
 221 +       u32     ctrepldst;      /* reply destination from conntrack */
 222 +};
 223 +
 224 +static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a)
 225 +{
 226 +       return jhash_1word(a, q->perturbation) & (q->hash_divisor-1);
 227 +}
 228 +
 229 +static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b)
 230 +{
 231 +       return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1);
 232 +}
 233 +
 234 +static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c)
 235 +{
 236 +       return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1);
 237 +}
 238 +
 239 +
 240 +static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb)
 241 +{
 242 +       struct esfq_packet_info info;
 243 +#ifdef CONFIG_IP_NF_CONNTRACK
 244 +       enum ip_conntrack_info ctinfo;
 245 +       struct ip_conntrack *ct = ip_conntrack_get(skb, &ctinfo);
 246 +#endif
 247 +
 248 +       switch (skb->protocol) {
 249 +       case __constant_htons(ETH_P_IP):
 250 +       {
 251 +               struct iphdr *iph = skb->nh.iph;
 252 +               info.dst = iph->daddr;
 253 +               info.src = iph->saddr;
 254 +               if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
 255 +                   (iph->protocol == IPPROTO_TCP ||
 256 +                    iph->protocol == IPPROTO_UDP ||
 257 +                    iph->protocol == IPPROTO_SCTP ||
 258 +                    iph->protocol == IPPROTO_DCCP ||
 259 +                    iph->protocol == IPPROTO_ESP))
 260 +                       info.proto = *(((u32*)iph) + iph->ihl);
 261 +               else
 262 +                       info.proto = iph->protocol;
 263 +               break;
 264 +       }
 265 +       default:
 266 +               info.dst   = (u32)(unsigned long)skb->dst;
 267 +               info.src   = (u32)(unsigned long)skb->sk;
 268 +               info.proto = skb->protocol;
 269 +       }
 270 +
 271 +#ifdef CONFIG_IP_NF_CONNTRACK
 272 +       /* defaults if there is no conntrack info */
 273 +       info.ctorigsrc = info.src;
 274 +       info.ctorigdst = info.dst;
 275 +       info.ctreplsrc = info.dst;
 276 +       info.ctrepldst = info.src;
 277 +       /* collect conntrack info */
 278 +       IP_NF_ASSERT(ct);
 279 +       if (ct) {
 280 +               if (skb->protocol == __constant_htons(ETH_P_IP)) {
 281 +                       info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.ip;
 282 +                       info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.ip;
 283 +                       info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.ip;
 284 +                       info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.ip;
 285 +               }
 286 +       }
 287 +#endif
 288 +
 289 +       switch(q->hash_kind)
 290 +       {
 291 +       case TCA_SFQ_HASH_CLASSIC:
 292 +               return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 293 +       case TCA_SFQ_HASH_DST:
 294 +               return esfq_jhash_1word(q, info.dst);
 295 +       case TCA_SFQ_HASH_SRC:
 296 +               return esfq_jhash_1word(q, info.src);
 297 +#ifdef CONFIG_IP_NF_CONNTRACK
 298 +       case TCA_SFQ_HASH_CTORIGDST:
 299 +               return esfq_jhash_1word(q, info.ctorigdst);
 300 +       case TCA_SFQ_HASH_CTORIGSRC:
 301 +               return esfq_jhash_1word(q, info.ctorigsrc);
 302 +       case TCA_SFQ_HASH_CTREPLDST:
 303 +               return esfq_jhash_1word(q, info.ctrepldst);
 304 +       case TCA_SFQ_HASH_CTREPLSRC:
 305 +               return esfq_jhash_1word(q, info.ctreplsrc);
 306 +       case TCA_SFQ_HASH_CTNATCHG:
 307 +       {
 308 +               if (info.ctorigdst == info.ctreplsrc)
 309 +                       return esfq_jhash_1word(q, info.ctorigsrc);
 310 +               else
 311 +                       return esfq_jhash_1word(q, info.ctreplsrc);
 312 +       }
 313 +#endif
 314 +       default:
 315 +               if (net_ratelimit())
 316 +                       printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n");
 317 +       }
 318 +       return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 319 +}
 320 +
 321 +static inline void esfq_link(struct esfq_sched_data *q, esfq_index x)
 322 +{
 323 +       esfq_index p, n;
 324 +       int d = q->qs[x].qlen + q->depth;
 325 +
 326 +       p = d;
 327 +       n = q->dep[d].next;
 328 +       q->dep[x].next = n;
 329 +       q->dep[x].prev = p;
 330 +       q->dep[p].next = q->dep[n].prev = x;
 331 +}
 332 +
 333 +static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x)
 334 +{
 335 +       esfq_index p, n;
 336 +
 337 +       n = q->dep[x].next;
 338 +       p = q->dep[x].prev;
 339 +       q->dep[p].next = n;
 340 +       q->dep[n].prev = p;
 341 +
 342 +       if (n == p && q->max_depth == q->qs[x].qlen + 1)
 343 +               q->max_depth--;
 344 +
 345 +       esfq_link(q, x);
 346 +}
 347 +
 348 +static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x)
 349 +{
 350 +       esfq_index p, n;
 351 +       int d;
 352 +
 353 +       n = q->dep[x].next;
 354 +       p = q->dep[x].prev;
 355 +       q->dep[p].next = n;
 356 +       q->dep[n].prev = p;
 357 +       d = q->qs[x].qlen;
 358 +       if (q->max_depth < d)
 359 +               q->max_depth = d;
 360 +
 361 +       esfq_link(q, x);
 362 +}
 363 +
 364 +static unsigned int esfq_drop(struct Qdisc *sch)
 365 +{
 366 +       struct esfq_sched_data *q = qdisc_priv(sch);
 367 +       esfq_index d = q->max_depth;
 368 +       struct sk_buff *skb;
 369 +       unsigned int len;
 370 +
 371 +       /* Queue is full! Find the longest slot and
 372 +          drop a packet from it */
 373 +
 374 +       if (d > 1) {
 375 +               esfq_index x = q->dep[d+q->depth].next;
 376 +               skb = q->qs[x].prev;
 377 +               len = skb->len;
 378 +               __skb_unlink(skb, &q->qs[x]);
 379 +               kfree_skb(skb);
 380 +               esfq_dec(q, x);
 381 +               sch->q.qlen--;
 382 +               sch->stats.drops++;
 383 +               sch->stats.backlog -= len;
 384 +               return len;
 385 +       }
 386 +
 387 +       if (d == 1) {
 388 +               /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
 389 +               d = q->next[q->tail];
 390 +               q->next[q->tail] = q->next[d];
 391 +               q->allot[q->next[d]] += q->quantum;
 392 +               skb = q->qs[d].prev;
 393 +               len = skb->len;
 394 +               __skb_unlink(skb, &q->qs[d]);
 395 +               kfree_skb(skb);
 396 +               esfq_dec(q, d);
 397 +               sch->q.qlen--;
 398 +               q->ht[q->hash[d]] = q->depth;
 399 +               sch->stats.drops++;
 400 +               sch->stats.backlog -= len;
 401 +               return len;
 402 +       }
 403 +
 404 +       return 0;
 405 +}
 406 +
 407 +static int
 408 +esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 409 +{
 410 +       struct esfq_sched_data *q = qdisc_priv(sch);
 411 +       unsigned hash = esfq_hash(q, skb);
 412 +       unsigned depth = q->depth;
 413 +       esfq_index x;
 414 +
 415 +       x = q->ht[hash];
 416 +       if (x == depth) {
 417 +               q->ht[hash] = x = q->dep[depth].next;
 418 +               q->hash[x] = hash;
 419 +       }
 420 +       sch->stats.backlog += skb->len;
 421 +       __skb_queue_tail(&q->qs[x], skb);
 422 +       esfq_inc(q, x);
 423 +       if (q->qs[x].qlen == 1) {               /* The flow is new */
 424 +               if (q->tail == depth) { /* It is the first flow */
 425 +                       q->tail = x;
 426 +                       q->next[x] = x;
 427 +                       q->allot[x] = q->quantum;
 428 +               } else {
 429 +                       q->next[x] = q->next[q->tail];
 430 +                       q->next[q->tail] = x;
 431 +                       q->tail = x;
 432 +               }
 433 +       }
 434 +       if (++sch->q.qlen < q->limit-1) {
 435 +               sch->stats.bytes += skb->len;
 436 +               sch->stats.packets++;
 437 +               return 0;
 438 +       }
 439 +
 440 +       esfq_drop(sch);
 441 +       return NET_XMIT_CN;
 442 +}
 443 +
 444 +static int
 445 +esfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
 446 +{
 447 +       struct esfq_sched_data *q = qdisc_priv(sch);
 448 +       unsigned hash = esfq_hash(q, skb);
 449 +       unsigned depth = q->depth;
 450 +       esfq_index x;
 451 +
 452 +       x = q->ht[hash];
 453 +       if (x == depth) {
 454 +               q->ht[hash] = x = q->dep[depth].next;
 455 +               q->hash[x] = hash;
 456 +       }
 457 +       sch->stats.backlog += skb->len;
 458 +       __skb_queue_head(&q->qs[x], skb);
 459 +       esfq_inc(q, x);
 460 +       if (q->qs[x].qlen == 1) {               /* The flow is new */
 461 +               if (q->tail == depth) { /* It is the first flow */
 462 +                       q->tail = x;
 463 +                       q->next[x] = x;
 464 +                       q->allot[x] = q->quantum;
 465 +               } else {
 466 +                       q->next[x] = q->next[q->tail];
 467 +                       q->next[q->tail] = x;
 468 +                       q->tail = x;
 469 +               }
 470 +       }
 471 +       if (++sch->q.qlen < q->limit - 1) {
 472 +               return 0;
 473 +       }
 474 +
 475 +       sch->stats.drops++;
 476 +       esfq_drop(sch);
 477 +       return NET_XMIT_CN;
 478 +}
 479 +
 480 +
 481 +
 482 +
 483 +static struct sk_buff *
 484 +esfq_dequeue(struct Qdisc* sch)
 485 +{
 486 +       struct esfq_sched_data *q = qdisc_priv(sch);
 487 +       struct sk_buff *skb;
 488 +       unsigned depth = q->depth;
 489 +       esfq_index a, old_a;
 490 +
 491 +       /* No active slots */
 492 +       if (q->tail == depth)
 493 +               return NULL;
 494 +
 495 +       a = old_a = q->next[q->tail];
 496 +
 497 +       /* Grab packet */
 498 +       skb = __skb_dequeue(&q->qs[a]);
 499 +       esfq_dec(q, a);
 500 +       sch->q.qlen--;
 501 +       sch->stats.backlog -= skb->len;
 502 +
 503 +       /* Is the slot empty? */
 504 +       if (q->qs[a].qlen == 0) {
 505 +               q->ht[q->hash[a]] = depth;
 506 +               a = q->next[a];
 507 +               if (a == old_a) {
 508 +                       q->tail = depth;
 509 +                       return skb;
 510 +               }
 511 +               q->next[q->tail] = a;
 512 +               q->allot[a] += q->quantum;
 513 +       } else if ((q->allot[a] -= skb->len) <= 0) {
 514 +               q->tail = a;
 515 +               a = q->next[a];
 516 +               q->allot[a] += q->quantum;
 517 +       }
 518 +
 519 +       return skb;
 520 +}
 521 +
 522 +static void
 523 +esfq_reset(struct Qdisc* sch)
 524 +{
 525 +       struct sk_buff *skb;
 526 +
 527 +       while ((skb = esfq_dequeue(sch)) != NULL)
 528 +               kfree_skb(skb);
 529 +}
 530 +
 531 +static void esfq_perturbation(unsigned long arg)
 532 +{
 533 +       struct Qdisc *sch = (struct Qdisc*)arg;
 534 +       struct esfq_sched_data *q = qdisc_priv(sch);
 535 +
 536 +       q->perturbation = net_random()&0x1F;
 537 +
 538 +       if (q->perturb_period) {
 539 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 540 +               add_timer(&q->perturb_timer);
 541 +       }
 542 +}
 543 +
 544 +static int esfq_change(struct Qdisc *sch, struct rtattr *opt)
 545 +{
 546 +       struct esfq_sched_data *q = qdisc_priv(sch);
 547 +       struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 548 +       int old_perturb = q->perturb_period;
 549 +
 550 +       if (opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 551 +               return -EINVAL;
 552 +
 553 +       sch_tree_lock(sch);
 554 +       q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
 555 +       q->perturb_period = ctl->perturb_period*HZ;
 556 +//     q->hash_divisor = ctl->divisor;
 557 +//     q->tail = q->limit = q->depth = ctl->flows;
 558 +
 559 +       if (ctl->limit)
 560 +               q->limit = min_t(u32, ctl->limit, q->depth);
 561 +
 562 +       if (ctl->hash_kind) {
 563 +               q->hash_kind = ctl->hash_kind;
 564 +               if (q->hash_kind != TCA_SFQ_HASH_CLASSIC)
 565 +                       q->perturb_period = 0;
 566 +       }
 567 +
 568 +       // is sch_tree_lock enough to do this ?
 569 +       while (sch->q.qlen >= q->limit-1)
 570 +               esfq_drop(sch);
 571 +
 572 +       if (old_perturb)
 573 +               del_timer(&q->perturb_timer);
 574 +       if (q->perturb_period) {
 575 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 576 +               add_timer(&q->perturb_timer);
 577 +       } else {
 578 +               q->perturbation = 0;
 579 +       }
 580 +       sch_tree_unlock(sch);
 581 +       return 0;
 582 +}
 583 +
 584 +static int esfq_init(struct Qdisc *sch, struct rtattr *opt)
 585 +{
 586 +       struct esfq_sched_data *q = qdisc_priv(sch);
 587 +       struct tc_esfq_qopt *ctl;
 588 +       esfq_index p = ~0U/2;
 589 +       int i;
 590 +
 591 +       if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 592 +               return -EINVAL;
 593 +
 594 +       init_timer(&q->perturb_timer);
 595 +       q->perturb_timer.data = (unsigned long)sch;
 596 +       q->perturb_timer.function = esfq_perturbation;
 597 +       q->perturbation = 0;
 598 +       q->hash_kind = TCA_SFQ_HASH_CLASSIC;
 599 +       q->max_depth = 0;
 600 +       q->dyn_min = ~0U; /* maximum value for this type */
 601 +       q->dyn_max = 0;  /* dyn_min/dyn_max will be set properly upon first packet */
 602 +       if (opt == NULL) {
 603 +               q->quantum = psched_mtu(sch->dev);
 604 +               q->perturb_period = 0;
 605 +               q->hash_divisor = 1024;
 606 +               q->tail = q->limit = q->depth = 128;
 607 +
 608 +       } else {
 609 +               ctl = RTA_DATA(opt);
 610 +               q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
 611 +               q->perturb_period = ctl->perturb_period*HZ;
 612 +               q->hash_divisor = ctl->divisor ? : 1024;
 613 +               q->tail = q->limit = q->depth = ctl->flows ? : 128;
 614 +
 615 +               if ( q->depth > p - 1 )
 616 +                       return -EINVAL;
 617 +
 618 +               if (ctl->limit)
 619 +                       q->limit = min_t(u32, ctl->limit, q->depth);
 620 +
 621 +               if (ctl->hash_kind) {
 622 +                       q->hash_kind = ctl->hash_kind;
 623 +               }
 624 +
 625 +               if (q->perturb_period) {
 626 +                       q->perturb_timer.expires = jiffies + q->perturb_period;
 627 +                       add_timer(&q->perturb_timer);
 628 +               }
 629 +       }
 630 +
 631 +       q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL);
 632 +       if (!q->ht)
 633 +               goto err_case;
 634 +
 635 +       q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL);
 636 +       if (!q->dep)
 637 +               goto err_case;
 638 +       q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL);
 639 +       if (!q->next)
 640 +               goto err_case;
 641 +
 642 +       q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL);
 643 +       if (!q->allot)
 644 +               goto err_case;
 645 +       q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL);
 646 +       if (!q->hash)
 647 +               goto err_case;
 648 +       q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL);
 649 +       if (!q->qs)
 650 +               goto err_case;
 651 +
 652 +       for (i=0; i< q->hash_divisor; i++)
 653 +               q->ht[i] = q->depth;
 654 +       for (i=0; i<q->depth; i++) {
 655 +               skb_queue_head_init(&q->qs[i]);
 656 +               q->dep[i+q->depth].next = i+q->depth;
 657 +               q->dep[i+q->depth].prev = i+q->depth;
 658 +       }
 659 +
 660 +       for (i=0; i<q->depth; i++)
 661 +               esfq_link(q, i);
 662 +       return 0;
 663 +err_case:
 664 +       del_timer(&q->perturb_timer);
 665 +       if (q->ht)
 666 +               kfree(q->ht);
 667 +       if (q->dep)
 668 +               kfree(q->dep);
 669 +       if (q->next)
 670 +               kfree(q->next);
 671 +       if (q->allot)
 672 +               kfree(q->allot);
 673 +       if (q->hash)
 674 +               kfree(q->hash);
 675 +       if (q->qs)
 676 +               kfree(q->qs);
 677 +       return -ENOBUFS;
 678 +}
 679 +
 680 +static void esfq_destroy(struct Qdisc *sch)
 681 +{
 682 +       struct esfq_sched_data *q = qdisc_priv(sch);
 683 +       del_timer(&q->perturb_timer);
 684 +       if(q->ht)
 685 +               kfree(q->ht);
 686 +       if(q->dep)
 687 +               kfree(q->dep);
 688 +       if(q->next)
 689 +               kfree(q->next);
 690 +       if(q->allot)
 691 +               kfree(q->allot);
 692 +       if(q->hash)
 693 +               kfree(q->hash);
 694 +       if(q->qs)
 695 +               kfree(q->qs);
 696 +}
 697 +
 698 +static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb)
 699 +{
 700 +       struct esfq_sched_data *q = qdisc_priv(sch);
 701 +       unsigned char    *b = skb->tail;
 702 +       struct tc_esfq_qopt opt;
 703 +
 704 +       opt.quantum = q->quantum;
 705 +       opt.perturb_period = q->perturb_period/HZ;
 706 +
 707 +       opt.limit = q->limit;
 708 +       opt.divisor = q->hash_divisor;
 709 +       opt.flows = q->depth;
 710 +       opt.hash_kind = q->hash_kind;
 711 +
 712 +       RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
 713 +
 714 +       return skb->len;
 715 +
 716 +rtattr_failure:
 717 +       skb_trim(skb, b - skb->data);
 718 +       return -1;
 719 +}
 720 +
 721 +static struct Qdisc_ops esfq_qdisc_ops =
 722 +{
 723 +       .next           =       NULL,
 724 +       .cl_ops         =       NULL,
 725 +       .id             =       "esfq",
 726 +       .priv_size      =       sizeof(struct esfq_sched_data),
 727 +       .enqueue        =       esfq_enqueue,
 728 +       .dequeue        =       esfq_dequeue,
 729 +       .requeue        =       esfq_requeue,
 730 +       .drop           =       esfq_drop,
 731 +       .init           =       esfq_init,
 732 +       .reset          =       esfq_reset,
 733 +       .destroy        =       esfq_destroy,
 734 +       .change         =       NULL, /* esfq_change - needs more work */
 735 +       .dump           =       esfq_dump,
 736 +};
 737 +
 738 +static int __init esfq_module_init(void)
 739 +{
 740 +       return register_qdisc(&esfq_qdisc_ops);
 741 +}
 742 +static void __exit esfq_module_exit(void)
 743 +{
 744 +       unregister_qdisc(&esfq_qdisc_ops);
 745 +}
 746 +module_init(esfq_module_init)
 747 +module_exit(esfq_module_exit)
 748 +MODULE_LICENSE("GPL");