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

   1 diff -urN linux-2.4.34/Documentation/Configure.help linux-2.4.34/Documentation/Configure.help
   2 --- linux-2.4.34/Documentation/Configure.help   2007-05-10 19:37:42.000000000 -0400
   3 +++ linux-2.4.34/Documentation/Configure.help   2007-05-10 19:49:49.000000000 -0400
   4 @@ -11127,6 +11127,24 @@
   5    whenever you want).  If you want to compile it as a module, say M
   6    here and read <file:Documentation/modules.txt>.
   7
   8 +ESFQ queue
   9 +CONFIG_NET_SCH_ESFQ
  10 +  Say Y here if you want to use the Stochastic Fairness Queueing (SFQ)
  11 +  packet scheduling algorithm for some of your network devices or as a
  12 +  leaf discipline for the CBQ scheduling algorithm (see the top of
  13 +  <file:net/sched/sch_esfq.c> for details and references about the SFQ
  14 +  algorithm).
  15 +
  16 +  This is an enchanced SFQ version which allows you to control the
  17 +  hardcoded values in the SFQ scheduler: queue depth, hash table size,
  18 +  queues limit. Also adds control to the hash function used to identify
  19 +  packet flows. Hash by src or dst ip and original sfq hash.
  20 +
  21 +  This code is also available as a module called sch_esfq.o ( = code
  22 +  which can be inserted in and removed from the running kernel
  23 +  whenever you want). If you want to compile it as a module, say M
  24 +  here and read <file:Documentation/modules.txt>.
  25 +
  26  CSZ packet scheduler
  27  CONFIG_NET_SCH_CSZ
  28    Say Y here if you want to use the Clark-Shenker-Zhang (CSZ) packet
  29 diff -urN linux-2.4.34/include/linux/pkt_sched.h linux-2.4.34/include/linux/pkt_sched.h
  30 --- linux-2.4.34/include/linux/pkt_sched.h      2007-05-10 19:38:19.000000000 -0400
  31 +++ linux-2.4.34/include/linux/pkt_sched.h      2007-05-10 19:53:59.000000000 -0400
  32 @@ -173,8 +173,36 @@
  33   *
  34   *     The only reason for this is efficiency, it is possible
  35   *     to change these parameters in compile time.
  36 + *
  37 + *     If you need to play with these values use esfq instead.
  38   */
  39
  40 +/* ESFQ section */
  41 +
  42 +enum
  43 +{
  44 +       /* traditional */
  45 +       TCA_SFQ_HASH_CLASSIC,
  46 +       TCA_SFQ_HASH_DST,
  47 +       TCA_SFQ_HASH_SRC,
  48 +       /* conntrack */
  49 +       TCA_SFQ_HASH_CTORIGDST,
  50 +       TCA_SFQ_HASH_CTORIGSRC,
  51 +       TCA_SFQ_HASH_CTREPLDST,
  52 +       TCA_SFQ_HASH_CTREPLSRC,
  53 +       TCA_SFQ_HASH_CTNATCHG,
  54 +};
  55 +
  56 +struct tc_esfq_qopt
  57 +{
  58 +       unsigned        quantum;        /* Bytes per round allocated to flow */
  59 +       int             perturb_period; /* Period of hash perturbation */
  60 +       __u32           limit;          /* Maximal packets in queue */
  61 +       unsigned        divisor;        /* Hash divisor  */
  62 +       unsigned        flows;          /* Maximal number of flows  */
  63 +       unsigned        hash_kind;      /* Hash function to use for flow identification */
  64 +};
  65 +
  66  /* RED section */
  67
  68  enum
  69 diff -urN linux-2.4.34/net/sched/Config.in linux-2.4.34/net/sched/Config.in
  70 --- linux-2.4.34/net/sched/Config.in    2007-05-10 19:38:31.000000000 -0400
  71 +++ linux-2.4.34/net/sched/Config.in    2007-05-10 19:54:45.000000000 -0400
  72 @@ -12,6 +12,7 @@
  73  tristate '  The simplest PRIO pseudoscheduler' CONFIG_NET_SCH_PRIO
  74  tristate '  RED queue' CONFIG_NET_SCH_RED
  75  tristate '  SFQ queue' CONFIG_NET_SCH_SFQ
  76 +tristate '  ESFQ queue' CONFIG_NET_SCH_ESFQ
  77  tristate '  TEQL queue' CONFIG_NET_SCH_TEQL
  78  tristate '  TBF queue' CONFIG_NET_SCH_TBF
  79  tristate '  GRED queue' CONFIG_NET_SCH_GRED
  80 diff -urN linux-2.4.34/net/sched/Makefile linux-2.4.34/net/sched/Makefile
  81 --- linux-2.4.34/net/sched/Makefile     2007-05-10 19:38:31.000000000 -0400
  82 +++ linux-2.4.34/net/sched/Makefile     2007-05-10 19:55:13.000000000 -0400
  83 @@ -19,6 +19,7 @@
  84  obj-$(CONFIG_NET_SCH_HFSC)     += sch_hfsc.o
  85  obj-$(CONFIG_NET_SCH_HTB)      += sch_htb.o
  86  obj-$(CONFIG_NET_SCH_SFQ)      += sch_sfq.o
  87 +obj-$(CONFIG_NET_SCH_ESFQ)     += sch_esfq.o
  88  obj-$(CONFIG_NET_SCH_RED)      += sch_red.o
  89  obj-$(CONFIG_NET_SCH_TBF)      += sch_tbf.o
  90  obj-$(CONFIG_NET_SCH_PRIO)     += sch_prio.o
  91 diff -urN linux-2.4.34/net/sched/sch_esfq.c linux-2.4.34/net/sched/sch_esfq.c
  92 --- linux-2.4.34/net/sched/sch_esfq.c   1969-12-31 19:00:00.000000000 -0500
  93 +++ linux-2.4.34/net/sched/sch_esfq.c   2007-05-10 19:57:15.000000000 -0400
  94 @@ -0,0 +1,649 @@
  95 +/*
  96 + * net/sched/sch_esfq.c        Extended Stochastic Fairness Queueing discipline.
  97 + *
  98 + *             This program is free software; you can redistribute it and/or
  99 + *             modify it under the terms of the GNU General Public License
 100 + *             as published by the Free Software Foundation; either version
 101 + *             2 of the License, or (at your option) any later version.
 102 + *
 103 + * Authors:    Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 104 + *
 105 + * Changes:    Alexander Atanasov, <alex@ssi.bg>
 106 + *             Added dynamic depth,limit,divisor,hash_kind options.
 107 + *             Added dst and src hashes.
 108 + *
 109 + *             Alexander Clouter, <alex@digriz.org.uk>
 110 + *             Ported ESFQ to Linux 2.6.
 111 + *
 112 + *             Corey Hickey, <bugfood-c@fatooh.org>
 113 + *             Maintenance of the Linux 2.6 port.
 114 + *             Added fwmark hash (thanks to Robert Kurjata).
 115 + *             Added usage of jhash.
 116 + *
 117 + */
 118 +
 119 +#include <linux/config.h>
 120 +#include <linux/module.h>
 121 +#include <asm/uaccess.h>
 122 +#include <asm/system.h>
 123 +#include <linux/bitops.h>
 124 +#include <linux/types.h>
 125 +#include <linux/kernel.h>
 126 +#include <linux/sched.h>
 127 +#include <linux/string.h>
 128 +#include <linux/mm.h>
 129 +#include <linux/socket.h>
 130 +#include <linux/sockios.h>
 131 +#include <linux/in.h>
 132 +#include <linux/errno.h>
 133 +#include <linux/interrupt.h>
 134 +#include <linux/if_ether.h>
 135 +#include <linux/inet.h>
 136 +#include <linux/netdevice.h>
 137 +#include <linux/etherdevice.h>
 138 +#include <linux/notifier.h>
 139 +#include <linux/init.h>
 140 +#include <net/ip.h>
 141 +#include <net/route.h>
 142 +#include <linux/skbuff.h>
 143 +#include <net/sock.h>
 144 +#include <net/pkt_sched.h>
 145 +#include <linux/jhash.h>
 146 +
 147 +#define IPPROTO_DCCP 33
 148 +#define qdisc_priv(q)   ((void *)(q->data))
 149 +
 150 +#ifdef CONFIG_IP_NF_CONNTRACK
 151 +/* #include <net/netfilter/nf_conntrack.h> */
 152 +#include <linux/netfilter_ipv4/ip_conntrack.h>
 153 +#endif
 154 +
 155 +/*     Stochastic Fairness Queuing algorithm.
 156 +       For more comments look at sch_sfq.c.
 157 +       The difference is that you can change limit, depth,
 158 +       hash table size and choose alternate hash types.
 159 +
 160 +       classic:        same as in sch_sfq.c
 161 +       dst:            destination IP address
 162 +       src:            source IP address
 163 +       ctorigdst:      original destination IP address
 164 +       ctorigsrc:      original source IP address
 165 +       ctrepldst:      reply destination IP address
 166 +       ctreplsrc:      reply source IP
 167 +       ctnatchg:       use the address which changed via nat
 168 +
 169 +*/
 170 +
 171 +
 172 +/* This type should contain at least SFQ_DEPTH*2 values */
 173 +typedef unsigned int esfq_index;
 174 +
 175 +struct esfq_head
 176 +{
 177 +       esfq_index      next;
 178 +       esfq_index      prev;
 179 +};
 180 +
 181 +struct esfq_sched_data
 182 +{
 183 +/* Parameters */
 184 +       int             perturb_period;
 185 +       unsigned        quantum;        /* Allotment per round: MUST BE >= MTU */
 186 +       int             limit;
 187 +       unsigned        depth;
 188 +       unsigned        hash_divisor;
 189 +       unsigned        hash_kind;
 190 +/* Variables */
 191 +       struct timer_list perturb_timer;
 192 +       int             perturbation;
 193 +       esfq_index      tail;           /* Index of current slot in round */
 194 +       esfq_index      max_depth;      /* Maximal depth */
 195 +
 196 +       esfq_index      *ht;                    /* Hash table */
 197 +       esfq_index      *next;                  /* Active slots link */
 198 +       short           *allot;                 /* Current allotment per slot */
 199 +       unsigned short  *hash;                  /* Hash value indexed by slots */
 200 +       struct sk_buff_head     *qs;            /* Slot queue */
 201 +       struct esfq_head        *dep;           /* Linked list of slots, indexed by depth */
 202 +       unsigned        dyn_min;        /* For dynamic divisor adjustment; minimum value seen */
 203 +       unsigned        dyn_max;        /*                                 maximum value seen */
 204 +       unsigned        dyn_range;      /*                                 saved range */
 205 +};
 206 +
 207 +/* This contains the info we will hash. */
 208 +struct esfq_packet_info
 209 +{
 210 +       u32     proto;          /* protocol or port */
 211 +       u32     src;            /* source from packet header */
 212 +       u32     dst;            /* destination from packet header */
 213 +       u32     ctorigsrc;      /* original source from conntrack */
 214 +       u32     ctorigdst;      /* original destination from conntrack */
 215 +       u32     ctreplsrc;      /* reply source from conntrack */
 216 +       u32     ctrepldst;      /* reply destination from conntrack */
 217 +};
 218 +
 219 +static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a)
 220 +{
 221 +       return jhash_1word(a, q->perturbation) & (q->hash_divisor-1);
 222 +}
 223 +
 224 +static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b)
 225 +{
 226 +       return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1);
 227 +}
 228 +
 229 +static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c)
 230 +{
 231 +       return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1);
 232 +}
 233 +
 234 +
 235 +static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb)
 236 +{
 237 +       struct esfq_packet_info info;
 238 +#ifdef CONFIG_IP_NF_CONNTRACK
 239 +       enum ip_conntrack_info ctinfo;
 240 +       struct ip_conntrack *ct = ip_conntrack_get(skb, &ctinfo);
 241 +#endif
 242 +
 243 +       switch (skb->protocol) {
 244 +       case __constant_htons(ETH_P_IP):
 245 +       {
 246 +               struct iphdr *iph = skb->nh.iph;
 247 +               info.dst = iph->daddr;
 248 +               info.src = iph->saddr;
 249 +               if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
 250 +                   (iph->protocol == IPPROTO_TCP ||
 251 +                    iph->protocol == IPPROTO_UDP ||
 252 +                    iph->protocol == IPPROTO_SCTP ||
 253 +                    iph->protocol == IPPROTO_DCCP ||
 254 +                    iph->protocol == IPPROTO_ESP))
 255 +                       info.proto = *(((u32*)iph) + iph->ihl);
 256 +               else
 257 +                       info.proto = iph->protocol;
 258 +               break;
 259 +       }
 260 +       default:
 261 +               info.dst   = (u32)(unsigned long)skb->dst;
 262 +               info.src   = (u32)(unsigned long)skb->sk;
 263 +               info.proto = skb->protocol;
 264 +       }
 265 +
 266 +#ifdef CONFIG_IP_NF_CONNTRACK
 267 +       /* defaults if there is no conntrack info */
 268 +       info.ctorigsrc = info.src;
 269 +       info.ctorigdst = info.dst;
 270 +       info.ctreplsrc = info.dst;
 271 +       info.ctrepldst = info.src;
 272 +       /* collect conntrack info */
 273 +       IP_NF_ASSERT(ct);
 274 +       if (ct) {
 275 +               if (skb->protocol == __constant_htons(ETH_P_IP)) {
 276 +                       info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.ip;
 277 +                       info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.ip;
 278 +                       info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.ip;
 279 +                       info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.ip;
 280 +               }
 281 +       }
 282 +#endif
 283 +
 284 +       switch(q->hash_kind)
 285 +       {
 286 +       case TCA_SFQ_HASH_CLASSIC:
 287 +               return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 288 +       case TCA_SFQ_HASH_DST:
 289 +               return esfq_jhash_1word(q, info.dst);
 290 +       case TCA_SFQ_HASH_SRC:
 291 +               return esfq_jhash_1word(q, info.src);
 292 +#ifdef CONFIG_IP_NF_CONNTRACK
 293 +       case TCA_SFQ_HASH_CTORIGDST:
 294 +               return esfq_jhash_1word(q, info.ctorigdst);
 295 +       case TCA_SFQ_HASH_CTORIGSRC:
 296 +               return esfq_jhash_1word(q, info.ctorigsrc);
 297 +       case TCA_SFQ_HASH_CTREPLDST:
 298 +               return esfq_jhash_1word(q, info.ctrepldst);
 299 +       case TCA_SFQ_HASH_CTREPLSRC:
 300 +               return esfq_jhash_1word(q, info.ctreplsrc);
 301 +       case TCA_SFQ_HASH_CTNATCHG:
 302 +       {
 303 +               if (info.ctorigdst == info.ctreplsrc)
 304 +                       return esfq_jhash_1word(q, info.ctorigsrc);
 305 +               else
 306 +                       return esfq_jhash_1word(q, info.ctreplsrc);
 307 +       }
 308 +#endif
 309 +       default:
 310 +               if (net_ratelimit())
 311 +                       printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n");
 312 +       }
 313 +       return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 314 +}
 315 +
 316 +static inline void esfq_link(struct esfq_sched_data *q, esfq_index x)
 317 +{
 318 +       esfq_index p, n;
 319 +       int d = q->qs[x].qlen + q->depth;
 320 +
 321 +       p = d;
 322 +       n = q->dep[d].next;
 323 +       q->dep[x].next = n;
 324 +       q->dep[x].prev = p;
 325 +       q->dep[p].next = q->dep[n].prev = x;
 326 +}
 327 +
 328 +static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x)
 329 +{
 330 +       esfq_index p, n;
 331 +
 332 +       n = q->dep[x].next;
 333 +       p = q->dep[x].prev;
 334 +       q->dep[p].next = n;
 335 +       q->dep[n].prev = p;
 336 +
 337 +       if (n == p && q->max_depth == q->qs[x].qlen + 1)
 338 +               q->max_depth--;
 339 +
 340 +       esfq_link(q, x);
 341 +}
 342 +
 343 +static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x)
 344 +{
 345 +       esfq_index p, n;
 346 +       int d;
 347 +
 348 +       n = q->dep[x].next;
 349 +       p = q->dep[x].prev;
 350 +       q->dep[p].next = n;
 351 +       q->dep[n].prev = p;
 352 +       d = q->qs[x].qlen;
 353 +       if (q->max_depth < d)
 354 +               q->max_depth = d;
 355 +
 356 +       esfq_link(q, x);
 357 +}
 358 +
 359 +static unsigned int esfq_drop(struct Qdisc *sch)
 360 +{
 361 +       struct esfq_sched_data *q = qdisc_priv(sch);
 362 +       esfq_index d = q->max_depth;
 363 +       struct sk_buff *skb;
 364 +       unsigned int len;
 365 +
 366 +       /* Queue is full! Find the longest slot and
 367 +          drop a packet from it */
 368 +
 369 +       if (d > 1) {
 370 +               esfq_index x = q->dep[d+q->depth].next;
 371 +               skb = q->qs[x].prev;
 372 +               len = skb->len;
 373 +               __skb_unlink(skb, &q->qs[x]);
 374 +               kfree_skb(skb);
 375 +               esfq_dec(q, x);
 376 +               sch->q.qlen--;
 377 +               sch->stats.drops++;
 378 +               sch->stats.backlog -= len;
 379 +               return len;
 380 +       }
 381 +
 382 +       if (d == 1) {
 383 +               /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
 384 +               d = q->next[q->tail];
 385 +               q->next[q->tail] = q->next[d];
 386 +               q->allot[q->next[d]] += q->quantum;
 387 +               skb = q->qs[d].prev;
 388 +               len = skb->len;
 389 +               __skb_unlink(skb, &q->qs[d]);
 390 +               kfree_skb(skb);
 391 +               esfq_dec(q, d);
 392 +               sch->q.qlen--;
 393 +               q->ht[q->hash[d]] = q->depth;
 394 +               sch->stats.drops++;
 395 +               sch->stats.backlog -= len;
 396 +               return len;
 397 +       }
 398 +
 399 +       return 0;
 400 +}
 401 +
 402 +static int
 403 +esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 404 +{
 405 +       struct esfq_sched_data *q = qdisc_priv(sch);
 406 +       unsigned hash = esfq_hash(q, skb);
 407 +       unsigned depth = q->depth;
 408 +       esfq_index x;
 409 +
 410 +       x = q->ht[hash];
 411 +       if (x == depth) {
 412 +               q->ht[hash] = x = q->dep[depth].next;
 413 +               q->hash[x] = hash;
 414 +       }
 415 +       sch->stats.backlog += skb->len;
 416 +       __skb_queue_tail(&q->qs[x], skb);
 417 +       esfq_inc(q, x);
 418 +       if (q->qs[x].qlen == 1) {               /* The flow is new */
 419 +               if (q->tail == depth) { /* It is the first flow */
 420 +                       q->tail = x;
 421 +                       q->next[x] = x;
 422 +                       q->allot[x] = q->quantum;
 423 +               } else {
 424 +                       q->next[x] = q->next[q->tail];
 425 +                       q->next[q->tail] = x;
 426 +                       q->tail = x;
 427 +               }
 428 +       }
 429 +       if (++sch->q.qlen < q->limit-1) {
 430 +               sch->stats.bytes += skb->len;
 431 +               sch->stats.packets++;
 432 +               return 0;
 433 +       }
 434 +
 435 +       esfq_drop(sch);
 436 +       return NET_XMIT_CN;
 437 +}
 438 +
 439 +static int
 440 +esfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
 441 +{
 442 +       struct esfq_sched_data *q = qdisc_priv(sch);
 443 +       unsigned hash = esfq_hash(q, skb);
 444 +       unsigned depth = q->depth;
 445 +       esfq_index x;
 446 +
 447 +       x = q->ht[hash];
 448 +       if (x == depth) {
 449 +               q->ht[hash] = x = q->dep[depth].next;
 450 +               q->hash[x] = hash;
 451 +       }
 452 +       sch->stats.backlog += skb->len;
 453 +       __skb_queue_head(&q->qs[x], skb);
 454 +       esfq_inc(q, x);
 455 +       if (q->qs[x].qlen == 1) {               /* The flow is new */
 456 +               if (q->tail == depth) { /* It is the first flow */
 457 +                       q->tail = x;
 458 +                       q->next[x] = x;
 459 +                       q->allot[x] = q->quantum;
 460 +               } else {
 461 +                       q->next[x] = q->next[q->tail];
 462 +                       q->next[q->tail] = x;
 463 +                       q->tail = x;
 464 +               }
 465 +       }
 466 +       if (++sch->q.qlen < q->limit - 1) {
 467 +               return 0;
 468 +       }
 469 +
 470 +       sch->stats.drops++;
 471 +       esfq_drop(sch);
 472 +       return NET_XMIT_CN;
 473 +}
 474 +
 475 +
 476 +
 477 +
 478 +static struct sk_buff *
 479 +esfq_dequeue(struct Qdisc* sch)
 480 +{
 481 +       struct esfq_sched_data *q = qdisc_priv(sch);
 482 +       struct sk_buff *skb;
 483 +       unsigned depth = q->depth;
 484 +       esfq_index a, old_a;
 485 +
 486 +       /* No active slots */
 487 +       if (q->tail == depth)
 488 +               return NULL;
 489 +
 490 +       a = old_a = q->next[q->tail];
 491 +
 492 +       /* Grab packet */
 493 +       skb = __skb_dequeue(&q->qs[a]);
 494 +       esfq_dec(q, a);
 495 +       sch->q.qlen--;
 496 +       sch->stats.backlog -= skb->len;
 497 +
 498 +       /* Is the slot empty? */
 499 +       if (q->qs[a].qlen == 0) {
 500 +               q->ht[q->hash[a]] = depth;
 501 +               a = q->next[a];
 502 +               if (a == old_a) {
 503 +                       q->tail = depth;
 504 +                       return skb;
 505 +               }
 506 +               q->next[q->tail] = a;
 507 +               q->allot[a] += q->quantum;
 508 +       } else if ((q->allot[a] -= skb->len) <= 0) {
 509 +               q->tail = a;
 510 +               a = q->next[a];
 511 +               q->allot[a] += q->quantum;
 512 +       }
 513 +
 514 +       return skb;
 515 +}
 516 +
 517 +static void
 518 +esfq_reset(struct Qdisc* sch)
 519 +{
 520 +       struct sk_buff *skb;
 521 +
 522 +       while ((skb = esfq_dequeue(sch)) != NULL)
 523 +               kfree_skb(skb);
 524 +}
 525 +
 526 +static void esfq_perturbation(unsigned long arg)
 527 +{
 528 +       struct Qdisc *sch = (struct Qdisc*)arg;
 529 +       struct esfq_sched_data *q = qdisc_priv(sch);
 530 +
 531 +       q->perturbation = net_random()&0x1F;
 532 +
 533 +       if (q->perturb_period) {
 534 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 535 +               add_timer(&q->perturb_timer);
 536 +       }
 537 +}
 538 +
 539 +static int esfq_change(struct Qdisc *sch, struct rtattr *opt)
 540 +{
 541 +       struct esfq_sched_data *q = qdisc_priv(sch);
 542 +       struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 543 +       int old_perturb = q->perturb_period;
 544 +
 545 +       if (opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 546 +               return -EINVAL;
 547 +
 548 +       sch_tree_lock(sch);
 549 +       q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
 550 +       q->perturb_period = ctl->perturb_period*HZ;
 551 +//     q->hash_divisor = ctl->divisor;
 552 +//     q->tail = q->limit = q->depth = ctl->flows;
 553 +
 554 +       if (ctl->limit)
 555 +               q->limit = min_t(u32, ctl->limit, q->depth);
 556 +
 557 +       if (ctl->hash_kind) {
 558 +               q->hash_kind = ctl->hash_kind;
 559 +               if (q->hash_kind != TCA_SFQ_HASH_CLASSIC)
 560 +                       q->perturb_period = 0;
 561 +       }
 562 +
 563 +       // is sch_tree_lock enough to do this ?
 564 +       while (sch->q.qlen >= q->limit-1)
 565 +               esfq_drop(sch);
 566 +
 567 +       if (old_perturb)
 568 +               del_timer(&q->perturb_timer);
 569 +       if (q->perturb_period) {
 570 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 571 +               add_timer(&q->perturb_timer);
 572 +       } else {
 573 +               q->perturbation = 0;
 574 +       }
 575 +       sch_tree_unlock(sch);
 576 +       return 0;
 577 +}
 578 +
 579 +static int esfq_init(struct Qdisc *sch, struct rtattr *opt)
 580 +{
 581 +       struct esfq_sched_data *q = qdisc_priv(sch);
 582 +       struct tc_esfq_qopt *ctl;
 583 +       esfq_index p = ~0U/2;
 584 +       int i;
 585 +
 586 +       if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 587 +               return -EINVAL;
 588 +
 589 +       init_timer(&q->perturb_timer);
 590 +       q->perturb_timer.data = (unsigned long)sch;
 591 +       q->perturb_timer.function = esfq_perturbation;
 592 +       q->perturbation = 0;
 593 +       q->hash_kind = TCA_SFQ_HASH_CLASSIC;
 594 +       q->max_depth = 0;
 595 +       q->dyn_min = ~0U; /* maximum value for this type */
 596 +       q->dyn_max = 0;  /* dyn_min/dyn_max will be set properly upon first packet */
 597 +       if (opt == NULL) {
 598 +               q->quantum = psched_mtu(sch->dev);
 599 +               q->perturb_period = 0;
 600 +               q->hash_divisor = 1024;
 601 +               q->tail = q->limit = q->depth = 128;
 602 +
 603 +       } else {
 604 +               ctl = RTA_DATA(opt);
 605 +               q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
 606 +               q->perturb_period = ctl->perturb_period*HZ;
 607 +               q->hash_divisor = ctl->divisor ? : 1024;
 608 +               q->tail = q->limit = q->depth = ctl->flows ? : 128;
 609 +
 610 +               if ( q->depth > p - 1 )
 611 +                       return -EINVAL;
 612 +
 613 +               if (ctl->limit)
 614 +                       q->limit = min_t(u32, ctl->limit, q->depth);
 615 +
 616 +               if (ctl->hash_kind) {
 617 +                       q->hash_kind = ctl->hash_kind;
 618 +               }
 619 +
 620 +               if (q->perturb_period) {
 621 +                       q->perturb_timer.expires = jiffies + q->perturb_period;
 622 +                       add_timer(&q->perturb_timer);
 623 +               }
 624 +       }
 625 +
 626 +       q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL);
 627 +       if (!q->ht)
 628 +               goto err_case;
 629 +
 630 +       q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL);
 631 +       if (!q->dep)
 632 +               goto err_case;
 633 +       q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL);
 634 +       if (!q->next)
 635 +               goto err_case;
 636 +
 637 +       q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL);
 638 +       if (!q->allot)
 639 +               goto err_case;
 640 +       q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL);
 641 +       if (!q->hash)
 642 +               goto err_case;
 643 +       q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL);
 644 +       if (!q->qs)
 645 +               goto err_case;
 646 +
 647 +       for (i=0; i< q->hash_divisor; i++)
 648 +               q->ht[i] = q->depth;
 649 +       for (i=0; i<q->depth; i++) {
 650 +               skb_queue_head_init(&q->qs[i]);
 651 +               q->dep[i+q->depth].next = i+q->depth;
 652 +               q->dep[i+q->depth].prev = i+q->depth;
 653 +       }
 654 +
 655 +       for (i=0; i<q->depth; i++)
 656 +               esfq_link(q, i);
 657 +       return 0;
 658 +err_case:
 659 +       del_timer(&q->perturb_timer);
 660 +       if (q->ht)
 661 +               kfree(q->ht);
 662 +       if (q->dep)
 663 +               kfree(q->dep);
 664 +       if (q->next)
 665 +               kfree(q->next);
 666 +       if (q->allot)
 667 +               kfree(q->allot);
 668 +       if (q->hash)
 669 +               kfree(q->hash);
 670 +       if (q->qs)
 671 +               kfree(q->qs);
 672 +       return -ENOBUFS;
 673 +}
 674 +
 675 +static void esfq_destroy(struct Qdisc *sch)
 676 +{
 677 +       struct esfq_sched_data *q = qdisc_priv(sch);
 678 +       del_timer(&q->perturb_timer);
 679 +       if(q->ht)
 680 +               kfree(q->ht);
 681 +       if(q->dep)
 682 +               kfree(q->dep);
 683 +       if(q->next)
 684 +               kfree(q->next);
 685 +       if(q->allot)
 686 +               kfree(q->allot);
 687 +       if(q->hash)
 688 +               kfree(q->hash);
 689 +       if(q->qs)
 690 +               kfree(q->qs);
 691 +}
 692 +
 693 +static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb)
 694 +{
 695 +       struct esfq_sched_data *q = qdisc_priv(sch);
 696 +       unsigned char    *b = skb->tail;
 697 +       struct tc_esfq_qopt opt;
 698 +
 699 +       opt.quantum = q->quantum;
 700 +       opt.perturb_period = q->perturb_period/HZ;
 701 +
 702 +       opt.limit = q->limit;
 703 +       opt.divisor = q->hash_divisor;
 704 +       opt.flows = q->depth;
 705 +       opt.hash_kind = q->hash_kind;
 706 +
 707 +       RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
 708 +
 709 +       return skb->len;
 710 +
 711 +rtattr_failure:
 712 +       skb_trim(skb, b - skb->data);
 713 +       return -1;
 714 +}
 715 +
 716 +static struct Qdisc_ops esfq_qdisc_ops =
 717 +{
 718 +       .next           =       NULL,
 719 +       .cl_ops         =       NULL,
 720 +       .id             =       "esfq",
 721 +       .priv_size      =       sizeof(struct esfq_sched_data),
 722 +       .enqueue        =       esfq_enqueue,
 723 +       .dequeue        =       esfq_dequeue,
 724 +       .requeue        =       esfq_requeue,
 725 +       .drop           =       esfq_drop,
 726 +       .init           =       esfq_init,
 727 +       .reset          =       esfq_reset,
 728 +       .destroy        =       esfq_destroy,
 729 +       .change         =       NULL, /* esfq_change - needs more work */
 730 +       .dump           =       esfq_dump,
 731 +};
 732 +
 733 +static int __init esfq_module_init(void)
 734 +{
 735 +       return register_qdisc(&esfq_qdisc_ops);
 736 +}
 737 +static void __exit esfq_module_exit(void)
 738 +{
 739 +       unregister_qdisc(&esfq_qdisc_ops);
 740 +}
 741 +module_init(esfq_module_init)
 742 +module_exit(esfq_module_exit)
 743 +MODULE_LICENSE("GPL");