target/linux/generic-2.4/patches/620-tc_esfq.patch

   1 diff -urN target.old/linux/generic-2.4/patches/232-esfq_kmod.patch target/linux/generic-2.4/patches/232-esfq_kmod.patch
   2 --- target.old/linux/generic-2.4/patches/232-esfq_kmod.patch    1969-12-31 19:00:00.000000000 -0500
   3 +++ target/linux/generic-2.4/patches/232-esfq_kmod.patch        2007-06-08 02:35:06.000000000 -0400
   4 @@ -0,0 +1,743 @@
   5 +diff -urN linux-2.4.34/Documentation/Configure.help linux-2.4.34/Documentation/Configure.help
   6 +--- linux-2.4.34/Documentation/Configure.help  2007-05-10 19:37:42.000000000 -0400
   7 ++++ linux-2.4.34/Documentation/Configure.help  2007-05-10 19:49:49.000000000 -0400
   8 +@@ -11127,6 +11127,24 @@
   9 +   whenever you want).  If you want to compile it as a module, say M
  10 +   here and read <file:Documentation/modules.txt>.
  11 +
  12 ++ESFQ queue
  13 ++CONFIG_NET_SCH_ESFQ
  14 ++  Say Y here if you want to use the Stochastic Fairness Queueing (SFQ)
  15 ++  packet scheduling algorithm for some of your network devices or as a
  16 ++  leaf discipline for the CBQ scheduling algorithm (see the top of
  17 ++  <file:net/sched/sch_esfq.c> for details and references about the SFQ
  18 ++  algorithm).
  19 ++
  20 ++  This is an enchanced SFQ version which allows you to control the
  21 ++  hardcoded values in the SFQ scheduler: queue depth, hash table size,
  22 ++  queues limit. Also adds control to the hash function used to identify
  23 ++  packet flows. Hash by src or dst ip and original sfq hash.
  24 ++
  25 ++  This code is also available as a module called sch_esfq.o ( = code
  26 ++  which can be inserted in and removed from the running kernel
  27 ++  whenever you want). If you want to compile it as a module, say M
  28 ++  here and read <file:Documentation/modules.txt>.
  29 ++
  30 + CSZ packet scheduler
  31 + CONFIG_NET_SCH_CSZ
  32 +   Say Y here if you want to use the Clark-Shenker-Zhang (CSZ) packet
  33 +diff -urN linux-2.4.34/include/linux/pkt_sched.h linux-2.4.34/include/linux/pkt_sched.h
  34 +--- linux-2.4.34/include/linux/pkt_sched.h     2007-05-10 19:38:19.000000000 -0400
  35 ++++ linux-2.4.34/include/linux/pkt_sched.h     2007-05-10 19:53:59.000000000 -0400
  36 +@@ -173,8 +173,36 @@
  37 +  *
  38 +  *    The only reason for this is efficiency, it is possible
  39 +  *    to change these parameters in compile time.
  40 ++ *
  41 ++ *    If you need to play with these values use esfq instead.
  42 +  */
  43 +
  44 ++/* ESFQ section */
  45 ++
  46 ++enum
  47 ++{
  48 ++      /* traditional */
  49 ++      TCA_SFQ_HASH_CLASSIC,
  50 ++      TCA_SFQ_HASH_DST,
  51 ++      TCA_SFQ_HASH_SRC,
  52 ++      /* conntrack */
  53 ++      TCA_SFQ_HASH_CTORIGDST,
  54 ++      TCA_SFQ_HASH_CTORIGSRC,
  55 ++      TCA_SFQ_HASH_CTREPLDST,
  56 ++      TCA_SFQ_HASH_CTREPLSRC,
  57 ++      TCA_SFQ_HASH_CTNATCHG,
  58 ++};
  59 ++
  60 ++struct tc_esfq_qopt
  61 ++{
  62 ++      unsigned        quantum;        /* Bytes per round allocated to flow */
  63 ++      int             perturb_period; /* Period of hash perturbation */
  64 ++      __u32           limit;          /* Maximal packets in queue */
  65 ++      unsigned        divisor;        /* Hash divisor  */
  66 ++      unsigned        flows;          /* Maximal number of flows  */
  67 ++      unsigned        hash_kind;      /* Hash function to use for flow identification */
  68 ++};
  69 ++
  70 + /* RED section */
  71 +
  72 + enum
  73 +diff -urN linux-2.4.34/net/sched/Config.in linux-2.4.34/net/sched/Config.in
  74 +--- linux-2.4.34/net/sched/Config.in   2007-05-10 19:38:31.000000000 -0400
  75 ++++ linux-2.4.34/net/sched/Config.in   2007-05-10 19:54:45.000000000 -0400
  76 +@@ -12,6 +12,7 @@
  77 + tristate '  The simplest PRIO pseudoscheduler' CONFIG_NET_SCH_PRIO
  78 + tristate '  RED queue' CONFIG_NET_SCH_RED
  79 + tristate '  SFQ queue' CONFIG_NET_SCH_SFQ
  80 ++tristate '  ESFQ queue' CONFIG_NET_SCH_ESFQ
  81 + tristate '  TEQL queue' CONFIG_NET_SCH_TEQL
  82 + tristate '  TBF queue' CONFIG_NET_SCH_TBF
  83 + tristate '  GRED queue' CONFIG_NET_SCH_GRED
  84 +diff -urN linux-2.4.34/net/sched/Makefile linux-2.4.34/net/sched/Makefile
  85 +--- linux-2.4.34/net/sched/Makefile    2007-05-10 19:38:31.000000000 -0400
  86 ++++ linux-2.4.34/net/sched/Makefile    2007-05-10 19:55:13.000000000 -0400
  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 +diff -urN linux-2.4.34/net/sched/sch_esfq.c linux-2.4.34/net/sched/sch_esfq.c
  96 +--- linux-2.4.34/net/sched/sch_esfq.c  1969-12-31 19:00:00.000000000 -0500
  97 ++++ linux-2.4.34/net/sched/sch_esfq.c  2007-05-10 19:57:15.000000000 -0400
  98 +@@ -0,0 +1,649 @@
  99 ++/*
 100 ++ * net/sched/sch_esfq.c       Extended Stochastic Fairness Queueing discipline.
 101 ++ *
 102 ++ *            This program is free software; you can redistribute it and/or
 103 ++ *            modify it under the terms of the GNU General Public License
 104 ++ *            as published by the Free Software Foundation; either version
 105 ++ *            2 of the License, or (at your option) any later version.
 106 ++ *
 107 ++ * Authors:   Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 108 ++ *
 109 ++ * Changes:   Alexander Atanasov, <alex@ssi.bg>
 110 ++ *            Added dynamic depth,limit,divisor,hash_kind options.
 111 ++ *            Added dst and src hashes.
 112 ++ *
 113 ++ *            Alexander Clouter, <alex@digriz.org.uk>
 114 ++ *            Ported ESFQ to Linux 2.6.
 115 ++ *
 116 ++ *            Corey Hickey, <bugfood-c@fatooh.org>
 117 ++ *            Maintenance of the Linux 2.6 port.
 118 ++ *            Added fwmark hash (thanks to Robert Kurjata).
 119 ++ *            Added usage of jhash.
 120 ++ *
 121 ++ */
 122 ++
 123 ++#include <linux/config.h>
 124 ++#include <linux/module.h>
 125 ++#include <asm/uaccess.h>
 126 ++#include <asm/system.h>
 127 ++#include <linux/bitops.h>
 128 ++#include <linux/types.h>
 129 ++#include <linux/kernel.h>
 130 ++#include <linux/sched.h>
 131 ++#include <linux/string.h>
 132 ++#include <linux/mm.h>
 133 ++#include <linux/socket.h>
 134 ++#include <linux/sockios.h>
 135 ++#include <linux/in.h>
 136 ++#include <linux/errno.h>
 137 ++#include <linux/interrupt.h>
 138 ++#include <linux/if_ether.h>
 139 ++#include <linux/inet.h>
 140 ++#include <linux/netdevice.h>
 141 ++#include <linux/etherdevice.h>
 142 ++#include <linux/notifier.h>
 143 ++#include <linux/init.h>
 144 ++#include <net/ip.h>
 145 ++#include <net/route.h>
 146 ++#include <linux/skbuff.h>
 147 ++#include <net/sock.h>
 148 ++#include <net/pkt_sched.h>
 149 ++#include <linux/jhash.h>
 150 ++
 151 ++#define IPPROTO_DCCP 33
 152 ++#define qdisc_priv(q)   ((void *)(q->data))
 153 ++
 154 ++#ifdef CONFIG_IP_NF_CONNTRACK
 155 ++/* #include <net/netfilter/nf_conntrack.h> */
 156 ++#include <linux/netfilter_ipv4/ip_conntrack.h>
 157 ++#endif
 158 ++
 159 ++/*    Stochastic Fairness Queuing algorithm.
 160 ++      For more comments look at sch_sfq.c.
 161 ++      The difference is that you can change limit, depth,
 162 ++      hash table size and choose alternate hash types.
 163 ++
 164 ++      classic:        same as in sch_sfq.c
 165 ++      dst:            destination IP address
 166 ++      src:            source IP address
 167 ++      ctorigdst:      original destination IP address
 168 ++      ctorigsrc:      original source IP address
 169 ++      ctrepldst:      reply destination IP address
 170 ++      ctreplsrc:      reply source IP
 171 ++      ctnatchg:       use the address which changed via nat
 172 ++
 173 ++*/
 174 ++
 175 ++
 176 ++/* This type should contain at least SFQ_DEPTH*2 values */
 177 ++typedef unsigned int esfq_index;
 178 ++
 179 ++struct esfq_head
 180 ++{
 181 ++      esfq_index      next;
 182 ++      esfq_index      prev;
 183 ++};
 184 ++
 185 ++struct esfq_sched_data
 186 ++{
 187 ++/* Parameters */
 188 ++      int             perturb_period;
 189 ++      unsigned        quantum;        /* Allotment per round: MUST BE >= MTU */
 190 ++      int             limit;
 191 ++      unsigned        depth;
 192 ++      unsigned        hash_divisor;
 193 ++      unsigned        hash_kind;
 194 ++/* Variables */
 195 ++      struct timer_list perturb_timer;
 196 ++      int             perturbation;
 197 ++      esfq_index      tail;           /* Index of current slot in round */
 198 ++      esfq_index      max_depth;      /* Maximal depth */
 199 ++
 200 ++      esfq_index      *ht;                    /* Hash table */
 201 ++      esfq_index      *next;                  /* Active slots link */
 202 ++      short           *allot;                 /* Current allotment per slot */
 203 ++      unsigned short  *hash;                  /* Hash value indexed by slots */
 204 ++      struct sk_buff_head     *qs;            /* Slot queue */
 205 ++      struct esfq_head        *dep;           /* Linked list of slots, indexed by depth */
 206 ++      unsigned        dyn_min;        /* For dynamic divisor adjustment; minimum value seen */
 207 ++      unsigned        dyn_max;        /*                                 maximum value seen */
 208 ++      unsigned        dyn_range;      /*                                 saved range */
 209 ++};
 210 ++
 211 ++/* This contains the info we will hash. */
 212 ++struct esfq_packet_info
 213 ++{
 214 ++      u32     proto;          /* protocol or port */
 215 ++      u32     src;            /* source from packet header */
 216 ++      u32     dst;            /* destination from packet header */
 217 ++      u32     ctorigsrc;      /* original source from conntrack */
 218 ++      u32     ctorigdst;      /* original destination from conntrack */
 219 ++      u32     ctreplsrc;      /* reply source from conntrack */
 220 ++      u32     ctrepldst;      /* reply destination from conntrack */
 221 ++};
 222 ++
 223 ++static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a)
 224 ++{
 225 ++      return jhash_1word(a, q->perturbation) & (q->hash_divisor-1);
 226 ++}
 227 ++
 228 ++static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b)
 229 ++{
 230 ++      return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1);
 231 ++}
 232 ++
 233 ++static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c)
 234 ++{
 235 ++      return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1);
 236 ++}
 237 ++
 238 ++
 239 ++static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb)
 240 ++{
 241 ++      struct esfq_packet_info info;
 242 ++#ifdef CONFIG_IP_NF_CONNTRACK
 243 ++      enum ip_conntrack_info ctinfo;
 244 ++      struct ip_conntrack *ct = ip_conntrack_get(skb, &ctinfo);
 245 ++#endif
 246 ++
 247 ++      switch (skb->protocol) {
 248 ++      case __constant_htons(ETH_P_IP):
 249 ++      {
 250 ++              struct iphdr *iph = skb->nh.iph;
 251 ++              info.dst = iph->daddr;
 252 ++              info.src = iph->saddr;
 253 ++              if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
 254 ++                  (iph->protocol == IPPROTO_TCP ||
 255 ++                   iph->protocol == IPPROTO_UDP ||
 256 ++                   iph->protocol == IPPROTO_SCTP ||
 257 ++                   iph->protocol == IPPROTO_DCCP ||
 258 ++                   iph->protocol == IPPROTO_ESP))
 259 ++                      info.proto = *(((u32*)iph) + iph->ihl);
 260 ++              else
 261 ++                      info.proto = iph->protocol;
 262 ++              break;
 263 ++      }
 264 ++      default:
 265 ++              info.dst   = (u32)(unsigned long)skb->dst;
 266 ++              info.src   = (u32)(unsigned long)skb->sk;
 267 ++              info.proto = skb->protocol;
 268 ++      }
 269 ++
 270 ++#ifdef CONFIG_IP_NF_CONNTRACK
 271 ++      /* defaults if there is no conntrack info */
 272 ++      info.ctorigsrc = info.src;
 273 ++      info.ctorigdst = info.dst;
 274 ++      info.ctreplsrc = info.dst;
 275 ++      info.ctrepldst = info.src;
 276 ++      /* collect conntrack info */
 277 ++      IP_NF_ASSERT(ct);
 278 ++      if (ct) {
 279 ++              if (skb->protocol == __constant_htons(ETH_P_IP)) {
 280 ++                      info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.ip;
 281 ++                      info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.ip;
 282 ++                      info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.ip;
 283 ++                      info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.ip;
 284 ++              }
 285 ++      }
 286 ++#endif
 287 ++
 288 ++      switch(q->hash_kind)
 289 ++      {
 290 ++      case TCA_SFQ_HASH_CLASSIC:
 291 ++              return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 292 ++      case TCA_SFQ_HASH_DST:
 293 ++              return esfq_jhash_1word(q, info.dst);
 294 ++      case TCA_SFQ_HASH_SRC:
 295 ++              return esfq_jhash_1word(q, info.src);
 296 ++#ifdef CONFIG_IP_NF_CONNTRACK
 297 ++      case TCA_SFQ_HASH_CTORIGDST:
 298 ++              return esfq_jhash_1word(q, info.ctorigdst);
 299 ++      case TCA_SFQ_HASH_CTORIGSRC:
 300 ++              return esfq_jhash_1word(q, info.ctorigsrc);
 301 ++      case TCA_SFQ_HASH_CTREPLDST:
 302 ++              return esfq_jhash_1word(q, info.ctrepldst);
 303 ++      case TCA_SFQ_HASH_CTREPLSRC:
 304 ++              return esfq_jhash_1word(q, info.ctreplsrc);
 305 ++      case TCA_SFQ_HASH_CTNATCHG:
 306 ++      {
 307 ++              if (info.ctorigdst == info.ctreplsrc)
 308 ++                      return esfq_jhash_1word(q, info.ctorigsrc);
 309 ++              else
 310 ++                      return esfq_jhash_1word(q, info.ctreplsrc);
 311 ++      }
 312 ++#endif
 313 ++      default:
 314 ++              if (net_ratelimit())
 315 ++                      printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n");
 316 ++      }
 317 ++      return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 318 ++}
 319 ++
 320 ++static inline void esfq_link(struct esfq_sched_data *q, esfq_index x)
 321 ++{
 322 ++      esfq_index p, n;
 323 ++      int d = q->qs[x].qlen + q->depth;
 324 ++
 325 ++      p = d;
 326 ++      n = q->dep[d].next;
 327 ++      q->dep[x].next = n;
 328 ++      q->dep[x].prev = p;
 329 ++      q->dep[p].next = q->dep[n].prev = x;
 330 ++}
 331 ++
 332 ++static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x)
 333 ++{
 334 ++      esfq_index p, n;
 335 ++
 336 ++      n = q->dep[x].next;
 337 ++      p = q->dep[x].prev;
 338 ++      q->dep[p].next = n;
 339 ++      q->dep[n].prev = p;
 340 ++
 341 ++      if (n == p && q->max_depth == q->qs[x].qlen + 1)
 342 ++              q->max_depth--;
 343 ++
 344 ++      esfq_link(q, x);
 345 ++}
 346 ++
 347 ++static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x)
 348 ++{
 349 ++      esfq_index p, n;
 350 ++      int d;
 351 ++
 352 ++      n = q->dep[x].next;
 353 ++      p = q->dep[x].prev;
 354 ++      q->dep[p].next = n;
 355 ++      q->dep[n].prev = p;
 356 ++      d = q->qs[x].qlen;
 357 ++      if (q->max_depth < d)
 358 ++              q->max_depth = d;
 359 ++
 360 ++      esfq_link(q, x);
 361 ++}
 362 ++
 363 ++static unsigned int esfq_drop(struct Qdisc *sch)
 364 ++{
 365 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 366 ++      esfq_index d = q->max_depth;
 367 ++      struct sk_buff *skb;
 368 ++      unsigned int len;
 369 ++
 370 ++      /* Queue is full! Find the longest slot and
 371 ++         drop a packet from it */
 372 ++
 373 ++      if (d > 1) {
 374 ++              esfq_index x = q->dep[d+q->depth].next;
 375 ++              skb = q->qs[x].prev;
 376 ++              len = skb->len;
 377 ++              __skb_unlink(skb, &q->qs[x]);
 378 ++              kfree_skb(skb);
 379 ++              esfq_dec(q, x);
 380 ++              sch->q.qlen--;
 381 ++              sch->stats.drops++;
 382 ++              sch->stats.backlog -= len;
 383 ++              return len;
 384 ++      }
 385 ++
 386 ++      if (d == 1) {
 387 ++              /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
 388 ++              d = q->next[q->tail];
 389 ++              q->next[q->tail] = q->next[d];
 390 ++              q->allot[q->next[d]] += q->quantum;
 391 ++              skb = q->qs[d].prev;
 392 ++              len = skb->len;
 393 ++              __skb_unlink(skb, &q->qs[d]);
 394 ++              kfree_skb(skb);
 395 ++              esfq_dec(q, d);
 396 ++              sch->q.qlen--;
 397 ++              q->ht[q->hash[d]] = q->depth;
 398 ++              sch->stats.drops++;
 399 ++              sch->stats.backlog -= len;
 400 ++              return len;
 401 ++      }
 402 ++
 403 ++      return 0;
 404 ++}
 405 ++
 406 ++static int
 407 ++esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 408 ++{
 409 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 410 ++      unsigned hash = esfq_hash(q, skb);
 411 ++      unsigned depth = q->depth;
 412 ++      esfq_index x;
 413 ++
 414 ++      x = q->ht[hash];
 415 ++      if (x == depth) {
 416 ++              q->ht[hash] = x = q->dep[depth].next;
 417 ++              q->hash[x] = hash;
 418 ++      }
 419 ++      sch->stats.backlog += skb->len;
 420 ++      __skb_queue_tail(&q->qs[x], skb);
 421 ++      esfq_inc(q, x);
 422 ++      if (q->qs[x].qlen == 1) {               /* The flow is new */
 423 ++              if (q->tail == depth) { /* It is the first flow */
 424 ++                      q->tail = x;
 425 ++                      q->next[x] = x;
 426 ++                      q->allot[x] = q->quantum;
 427 ++              } else {
 428 ++                      q->next[x] = q->next[q->tail];
 429 ++                      q->next[q->tail] = x;
 430 ++                      q->tail = x;
 431 ++              }
 432 ++      }
 433 ++      if (++sch->q.qlen < q->limit-1) {
 434 ++              sch->stats.bytes += skb->len;
 435 ++              sch->stats.packets++;
 436 ++              return 0;
 437 ++      }
 438 ++
 439 ++      esfq_drop(sch);
 440 ++      return NET_XMIT_CN;
 441 ++}
 442 ++
 443 ++static int
 444 ++esfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
 445 ++{
 446 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 447 ++      unsigned hash = esfq_hash(q, skb);
 448 ++      unsigned depth = q->depth;
 449 ++      esfq_index x;
 450 ++
 451 ++      x = q->ht[hash];
 452 ++      if (x == depth) {
 453 ++              q->ht[hash] = x = q->dep[depth].next;
 454 ++              q->hash[x] = hash;
 455 ++      }
 456 ++      sch->stats.backlog += skb->len;
 457 ++      __skb_queue_head(&q->qs[x], skb);
 458 ++      esfq_inc(q, x);
 459 ++      if (q->qs[x].qlen == 1) {               /* The flow is new */
 460 ++              if (q->tail == depth) { /* It is the first flow */
 461 ++                      q->tail = x;
 462 ++                      q->next[x] = x;
 463 ++                      q->allot[x] = q->quantum;
 464 ++              } else {
 465 ++                      q->next[x] = q->next[q->tail];
 466 ++                      q->next[q->tail] = x;
 467 ++                      q->tail = x;
 468 ++              }
 469 ++      }
 470 ++      if (++sch->q.qlen < q->limit - 1) {
 471 ++              return 0;
 472 ++      }
 473 ++
 474 ++      sch->stats.drops++;
 475 ++      esfq_drop(sch);
 476 ++      return NET_XMIT_CN;
 477 ++}
 478 ++
 479 ++
 480 ++
 481 ++
 482 ++static struct sk_buff *
 483 ++esfq_dequeue(struct Qdisc* sch)
 484 ++{
 485 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 486 ++      struct sk_buff *skb;
 487 ++      unsigned depth = q->depth;
 488 ++      esfq_index a, old_a;
 489 ++
 490 ++      /* No active slots */
 491 ++      if (q->tail == depth)
 492 ++              return NULL;
 493 ++
 494 ++      a = old_a = q->next[q->tail];
 495 ++
 496 ++      /* Grab packet */
 497 ++      skb = __skb_dequeue(&q->qs[a]);
 498 ++      esfq_dec(q, a);
 499 ++      sch->q.qlen--;
 500 ++      sch->stats.backlog -= skb->len;
 501 ++
 502 ++      /* Is the slot empty? */
 503 ++      if (q->qs[a].qlen == 0) {
 504 ++              q->ht[q->hash[a]] = depth;
 505 ++              a = q->next[a];
 506 ++              if (a == old_a) {
 507 ++                      q->tail = depth;
 508 ++                      return skb;
 509 ++              }
 510 ++              q->next[q->tail] = a;
 511 ++              q->allot[a] += q->quantum;
 512 ++      } else if ((q->allot[a] -= skb->len) <= 0) {
 513 ++              q->tail = a;
 514 ++              a = q->next[a];
 515 ++              q->allot[a] += q->quantum;
 516 ++      }
 517 ++
 518 ++      return skb;
 519 ++}
 520 ++
 521 ++static void
 522 ++esfq_reset(struct Qdisc* sch)
 523 ++{
 524 ++      struct sk_buff *skb;
 525 ++
 526 ++      while ((skb = esfq_dequeue(sch)) != NULL)
 527 ++              kfree_skb(skb);
 528 ++}
 529 ++
 530 ++static void esfq_perturbation(unsigned long arg)
 531 ++{
 532 ++      struct Qdisc *sch = (struct Qdisc*)arg;
 533 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 534 ++
 535 ++      q->perturbation = net_random()&0x1F;
 536 ++
 537 ++      if (q->perturb_period) {
 538 ++              q->perturb_timer.expires = jiffies + q->perturb_period;
 539 ++              add_timer(&q->perturb_timer);
 540 ++      }
 541 ++}
 542 ++
 543 ++static int esfq_change(struct Qdisc *sch, struct rtattr *opt)
 544 ++{
 545 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 546 ++      struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 547 ++      int old_perturb = q->perturb_period;
 548 ++
 549 ++      if (opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 550 ++              return -EINVAL;
 551 ++
 552 ++      sch_tree_lock(sch);
 553 ++      q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
 554 ++      q->perturb_period = ctl->perturb_period*HZ;
 555 ++//    q->hash_divisor = ctl->divisor;
 556 ++//    q->tail = q->limit = q->depth = ctl->flows;
 557 ++
 558 ++      if (ctl->limit)
 559 ++              q->limit = min_t(u32, ctl->limit, q->depth);
 560 ++
 561 ++      if (ctl->hash_kind) {
 562 ++              q->hash_kind = ctl->hash_kind;
 563 ++              if (q->hash_kind != TCA_SFQ_HASH_CLASSIC)
 564 ++                      q->perturb_period = 0;
 565 ++      }
 566 ++
 567 ++      // is sch_tree_lock enough to do this ?
 568 ++      while (sch->q.qlen >= q->limit-1)
 569 ++              esfq_drop(sch);
 570 ++
 571 ++      if (old_perturb)
 572 ++              del_timer(&q->perturb_timer);
 573 ++      if (q->perturb_period) {
 574 ++              q->perturb_timer.expires = jiffies + q->perturb_period;
 575 ++              add_timer(&q->perturb_timer);
 576 ++      } else {
 577 ++              q->perturbation = 0;
 578 ++      }
 579 ++      sch_tree_unlock(sch);
 580 ++      return 0;
 581 ++}
 582 ++
 583 ++static int esfq_init(struct Qdisc *sch, struct rtattr *opt)
 584 ++{
 585 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 586 ++      struct tc_esfq_qopt *ctl;
 587 ++      esfq_index p = ~0U/2;
 588 ++      int i;
 589 ++
 590 ++      if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 591 ++              return -EINVAL;
 592 ++
 593 ++      init_timer(&q->perturb_timer);
 594 ++      q->perturb_timer.data = (unsigned long)sch;
 595 ++      q->perturb_timer.function = esfq_perturbation;
 596 ++      q->perturbation = 0;
 597 ++      q->hash_kind = TCA_SFQ_HASH_CLASSIC;
 598 ++      q->max_depth = 0;
 599 ++      q->dyn_min = ~0U; /* maximum value for this type */
 600 ++      q->dyn_max = 0;  /* dyn_min/dyn_max will be set properly upon first packet */
 601 ++      if (opt == NULL) {
 602 ++              q->quantum = psched_mtu(sch->dev);
 603 ++              q->perturb_period = 0;
 604 ++              q->hash_divisor = 1024;
 605 ++              q->tail = q->limit = q->depth = 128;
 606 ++
 607 ++      } else {
 608 ++              ctl = RTA_DATA(opt);
 609 ++              q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
 610 ++              q->perturb_period = ctl->perturb_period*HZ;
 611 ++              q->hash_divisor = ctl->divisor ? : 1024;
 612 ++              q->tail = q->limit = q->depth = ctl->flows ? : 128;
 613 ++
 614 ++              if ( q->depth > p - 1 )
 615 ++                      return -EINVAL;
 616 ++
 617 ++              if (ctl->limit)
 618 ++                      q->limit = min_t(u32, ctl->limit, q->depth);
 619 ++
 620 ++              if (ctl->hash_kind) {
 621 ++                      q->hash_kind = ctl->hash_kind;
 622 ++              }
 623 ++
 624 ++              if (q->perturb_period) {
 625 ++                      q->perturb_timer.expires = jiffies + q->perturb_period;
 626 ++                      add_timer(&q->perturb_timer);
 627 ++              }
 628 ++      }
 629 ++
 630 ++      q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL);
 631 ++      if (!q->ht)
 632 ++              goto err_case;
 633 ++
 634 ++      q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL);
 635 ++      if (!q->dep)
 636 ++              goto err_case;
 637 ++      q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL);
 638 ++      if (!q->next)
 639 ++              goto err_case;
 640 ++
 641 ++      q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL);
 642 ++      if (!q->allot)
 643 ++              goto err_case;
 644 ++      q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL);
 645 ++      if (!q->hash)
 646 ++              goto err_case;
 647 ++      q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL);
 648 ++      if (!q->qs)
 649 ++              goto err_case;
 650 ++
 651 ++      for (i=0; i< q->hash_divisor; i++)
 652 ++              q->ht[i] = q->depth;
 653 ++      for (i=0; i<q->depth; i++) {
 654 ++              skb_queue_head_init(&q->qs[i]);
 655 ++              q->dep[i+q->depth].next = i+q->depth;
 656 ++              q->dep[i+q->depth].prev = i+q->depth;
 657 ++      }
 658 ++
 659 ++      for (i=0; i<q->depth; i++)
 660 ++              esfq_link(q, i);
 661 ++      return 0;
 662 ++err_case:
 663 ++      del_timer(&q->perturb_timer);
 664 ++      if (q->ht)
 665 ++              kfree(q->ht);
 666 ++      if (q->dep)
 667 ++              kfree(q->dep);
 668 ++      if (q->next)
 669 ++              kfree(q->next);
 670 ++      if (q->allot)
 671 ++              kfree(q->allot);
 672 ++      if (q->hash)
 673 ++              kfree(q->hash);
 674 ++      if (q->qs)
 675 ++              kfree(q->qs);
 676 ++      return -ENOBUFS;
 677 ++}
 678 ++
 679 ++static void esfq_destroy(struct Qdisc *sch)
 680 ++{
 681 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 682 ++      del_timer(&q->perturb_timer);
 683 ++      if(q->ht)
 684 ++              kfree(q->ht);
 685 ++      if(q->dep)
 686 ++              kfree(q->dep);
 687 ++      if(q->next)
 688 ++              kfree(q->next);
 689 ++      if(q->allot)
 690 ++              kfree(q->allot);
 691 ++      if(q->hash)
 692 ++              kfree(q->hash);
 693 ++      if(q->qs)
 694 ++              kfree(q->qs);
 695 ++}
 696 ++
 697 ++static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb)
 698 ++{
 699 ++      struct esfq_sched_data *q = qdisc_priv(sch);
 700 ++      unsigned char    *b = skb->tail;
 701 ++      struct tc_esfq_qopt opt;
 702 ++
 703 ++      opt.quantum = q->quantum;
 704 ++      opt.perturb_period = q->perturb_period/HZ;
 705 ++
 706 ++      opt.limit = q->limit;
 707 ++      opt.divisor = q->hash_divisor;
 708 ++      opt.flows = q->depth;
 709 ++      opt.hash_kind = q->hash_kind;
 710 ++
 711 ++      RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
 712 ++
 713 ++      return skb->len;
 714 ++
 715 ++rtattr_failure:
 716 ++      skb_trim(skb, b - skb->data);
 717 ++      return -1;
 718 ++}
 719 ++
 720 ++static struct Qdisc_ops esfq_qdisc_ops =
 721 ++{
 722 ++      .next           =       NULL,
 723 ++      .cl_ops         =       NULL,
 724 ++      .id             =       "esfq",
 725 ++      .priv_size      =       sizeof(struct esfq_sched_data),
 726 ++      .enqueue        =       esfq_enqueue,
 727 ++      .dequeue        =       esfq_dequeue,
 728 ++      .requeue        =       esfq_requeue,
 729 ++      .drop           =       esfq_drop,
 730 ++      .init           =       esfq_init,
 731 ++      .reset          =       esfq_reset,
 732 ++      .destroy        =       esfq_destroy,
 733 ++      .change         =       NULL, /* esfq_change - needs more work */
 734 ++      .dump           =       esfq_dump,
 735 ++};
 736 ++
 737 ++static int __init esfq_module_init(void)
 738 ++{
 739 ++      return register_qdisc(&esfq_qdisc_ops);
 740 ++}
 741 ++static void __exit esfq_module_exit(void)
 742 ++{
 743 ++      unregister_qdisc(&esfq_qdisc_ops);
 744 ++}
 745 ++module_init(esfq_module_init)
 746 ++module_exit(esfq_module_exit)
 747 ++MODULE_LICENSE("GPL");