kernel-esfq.patch

   1 diff -Naur linux-2.6.24.orig/include/linux/pkt_sched.h linux-2.6.24/include/linux/pkt_sched.h
   2 --- linux-2.6.24.orig/include/linux/pkt_sched.h 2008-01-24 14:58:37.000000000 -0800
   3 +++ linux-2.6.24/include/linux/pkt_sched.h      2008-01-28 00:27:12.000000000 -0800
   4 @@ -157,6 +157,33 @@
   5   *     to change these parameters in compile time.
   6   */
   7
   8 +/* ESFQ section */
   9 +
  10 +enum
  11 +{
  12 +        /* traditional */
  13 +       TCA_ESFQ_HASH_CLASSIC,
  14 +       TCA_ESFQ_HASH_DST,
  15 +       TCA_ESFQ_HASH_SRC,
  16 +       TCA_ESFQ_HASH_FWMARK,
  17 +       /* conntrack */
  18 +       TCA_ESFQ_HASH_CTORIGDST,
  19 +       TCA_ESFQ_HASH_CTORIGSRC,
  20 +       TCA_ESFQ_HASH_CTREPLDST,
  21 +       TCA_ESFQ_HASH_CTREPLSRC,
  22 +       TCA_ESFQ_HASH_CTNATCHG,
  23 +};
  24 +
  25 +struct tc_esfq_qopt
  26 +{
  27 +       unsigned        quantum;        /* Bytes per round allocated to flow */
  28 +       int             perturb_period; /* Period of hash perturbation */
  29 +       __u32           limit;          /* Maximal packets in queue */
  30 +       unsigned        divisor;        /* Hash divisor  */
  31 +       unsigned        flows;          /* Maximal number of flows  */
  32 +       unsigned        hash_kind;      /* Hash function to use for flow identification */
  33 +};
  34 +
  35  /* RED section */
  36
  37  enum
  38 diff -Naur linux-2.6.24.orig/net/sched/Kconfig linux-2.6.24/net/sched/Kconfig
  39 --- linux-2.6.24.orig/net/sched/Kconfig 2008-01-24 14:58:37.000000000 -0800
  40 +++ linux-2.6.24/net/sched/Kconfig      2008-01-28 00:27:12.000000000 -0800
  41 @@ -139,6 +139,37 @@
  42           To compile this code as a module, choose M here: the
  43           module will be called sch_sfq.
  44
  45 +config NET_SCH_ESFQ
  46 +       tristate "Enhanced Stochastic Fairness Queueing (ESFQ)"
  47 +       ---help---
  48 +         Say Y here if you want to use the Enhanced Stochastic Fairness
  49 +         Queueing (ESFQ) packet scheduling algorithm for some of your network
  50 +         devices or as a leaf discipline for a classful qdisc such as HTB or
  51 +         CBQ (see the top of <file:net/sched/sch_esfq.c> for details and
  52 +         references to the SFQ algorithm).
  53 +
  54 +         This is an enchanced SFQ version which allows you to control some
  55 +         hardcoded values in the SFQ scheduler.
  56 +
  57 +         ESFQ also adds control of the hash function used to identify packet
  58 +         flows. The original SFQ discipline hashes by connection; ESFQ add
  59 +         several other hashing methods, such as by src IP or by dst IP, which
  60 +         can be more fair to users in some networking situations.
  61 +
  62 +         To compile this code as a module, choose M here: the
  63 +         module will be called sch_esfq.
  64 +
  65 +config NET_SCH_ESFQ_NFCT
  66 +       bool "Connection Tracking Hash Types"
  67 +       depends on NET_SCH_ESFQ && NF_CONNTRACK_ENABLED
  68 +       ---help---
  69 +         Say Y here to enable support for hashing based on netfilter connection
  70 +         tracking information. This is useful for a router that is also using
  71 +         NAT to connect privately-addressed hosts to the Internet. If you want
  72 +         to provide fair distribution of upstream bandwidth, ESFQ must use
  73 +         connection tracking information, since all outgoing packets will share
  74 +         the same source address.
  75 +
  76  config NET_SCH_TEQL
  77         tristate "True Link Equalizer (TEQL)"
  78         ---help---
  79 diff -Naur linux-2.6.24.orig/net/sched/Makefile linux-2.6.24/net/sched/Makefile
  80 --- linux-2.6.24.orig/net/sched/Makefile        2008-01-24 14:58:37.000000000 -0800
  81 +++ linux-2.6.24/net/sched/Makefile     2008-01-28 00:27:12.000000000 -0800
  82 @@ -23,6 +23,7 @@
  83  obj-$(CONFIG_NET_SCH_INGRESS)  += sch_ingress.o
  84  obj-$(CONFIG_NET_SCH_DSMARK)   += sch_dsmark.o
  85  obj-$(CONFIG_NET_SCH_SFQ)      += sch_sfq.o
  86 +obj-$(CONFIG_NET_SCH_ESFQ)     += sch_esfq.o
  87  obj-$(CONFIG_NET_SCH_TBF)      += sch_tbf.o
  88  obj-$(CONFIG_NET_SCH_TEQL)     += sch_teql.o
  89  obj-$(CONFIG_NET_SCH_PRIO)     += sch_prio.o
  90 diff -Naur linux-2.6.24.orig/net/sched/sch_esfq.c linux-2.6.24/net/sched/sch_esfq.c
  91 --- linux-2.6.24.orig/net/sched/sch_esfq.c      1969-12-31 16:00:00.000000000 -0800
  92 +++ linux-2.6.24/net/sched/sch_esfq.c   2008-01-28 00:27:22.000000000 -0800
  93 @@ -0,0 +1,703 @@
  94 +/*
  95 + * net/sched/sch_esfq.c        Extended Stochastic Fairness Queueing discipline.
  96 + *
  97 + *             This program is free software; you can redistribute it and/or
  98 + *             modify it under the terms of the GNU General Public License
  99 + *             as published by the Free Software Foundation; either version
 100 + *             2 of the License, or (at your option) any later version.
 101 + *
 102 + * Authors:    Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 103 + *
 104 + * Changes:    Alexander Atanasov, <alex@ssi.bg>
 105 + *             Added dynamic depth,limit,divisor,hash_kind options.
 106 + *             Added dst and src hashes.
 107 + *
 108 + *             Alexander Clouter, <alex@digriz.org.uk>
 109 + *             Ported ESFQ to Linux 2.6.
 110 + *
 111 + *             Corey Hickey, <bugfood-c@fatooh.org>
 112 + *             Maintenance of the Linux 2.6 port.
 113 + *             Added fwmark hash (thanks to Robert Kurjata).
 114 + *             Added usage of jhash.
 115 + *             Added conntrack support.
 116 + *             Added ctnatchg hash (thanks to Ben Pfountz).
 117 + */
 118 +
 119 +#include <linux/module.h>
 120 +#include <asm/uaccess.h>
 121 +#include <asm/system.h>
 122 +#include <linux/bitops.h>
 123 +#include <linux/types.h>
 124 +#include <linux/kernel.h>
 125 +#include <linux/jiffies.h>
 126 +#include <linux/string.h>
 127 +#include <linux/mm.h>
 128 +#include <linux/socket.h>
 129 +#include <linux/sockios.h>
 130 +#include <linux/in.h>
 131 +#include <linux/errno.h>
 132 +#include <linux/interrupt.h>
 133 +#include <linux/if_ether.h>
 134 +#include <linux/inet.h>
 135 +#include <linux/netdevice.h>
 136 +#include <linux/etherdevice.h>
 137 +#include <linux/notifier.h>
 138 +#include <linux/init.h>
 139 +#include <net/ip.h>
 140 +#include <net/netlink.h>
 141 +#include <linux/ipv6.h>
 142 +#include <net/route.h>
 143 +#include <linux/skbuff.h>
 144 +#include <net/sock.h>
 145 +#include <net/pkt_sched.h>
 146 +#include <linux/jhash.h>
 147 +#include <net/netfilter/nf_conntrack.h>
 148 +
 149 +/*     Stochastic Fairness Queuing algorithm.
 150 +       For more comments look at sch_sfq.c.
 151 +       The difference is that you can change limit, depth,
 152 +       hash table size and choose alternate hash types.
 153 +
 154 +       classic:        same as in sch_sfq.c
 155 +       dst:            destination IP address
 156 +       src:            source IP address
 157 +       fwmark:         netfilter mark value
 158 +       ctorigdst:      original destination IP address
 159 +       ctorigsrc:      original source IP address
 160 +       ctrepldst:      reply destination IP address
 161 +       ctreplsrc:      reply source IP
 162 +
 163 +*/
 164 +
 165 +#define ESFQ_HEAD 0
 166 +#define ESFQ_TAIL 1
 167 +
 168 +/* This type should contain at least SFQ_DEPTH*2 values */
 169 +typedef unsigned int esfq_index;
 170 +
 171 +struct esfq_head
 172 +{
 173 +       esfq_index      next;
 174 +       esfq_index      prev;
 175 +};
 176 +
 177 +struct esfq_sched_data
 178 +{
 179 +/* Parameters */
 180 +       int             perturb_period;
 181 +       unsigned        quantum;        /* Allotment per round: MUST BE >= MTU */
 182 +       int             limit;
 183 +       unsigned        depth;
 184 +       unsigned        hash_divisor;
 185 +       unsigned        hash_kind;
 186 +/* Variables */
 187 +       struct timer_list perturb_timer;
 188 +       int             perturbation;
 189 +       esfq_index      tail;           /* Index of current slot in round */
 190 +       esfq_index      max_depth;      /* Maximal depth */
 191 +
 192 +       esfq_index      *ht;                    /* Hash table */
 193 +       esfq_index      *next;                  /* Active slots link */
 194 +       short           *allot;                 /* Current allotment per slot */
 195 +       unsigned short  *hash;                  /* Hash value indexed by slots */
 196 +       struct sk_buff_head     *qs;            /* Slot queue */
 197 +       struct esfq_head        *dep;           /* Linked list of slots, indexed by depth */
 198 +};
 199 +
 200 +/* This contains the info we will hash. */
 201 +struct esfq_packet_info
 202 +{
 203 +       u32     proto;          /* protocol or port */
 204 +       u32     src;            /* source from packet header */
 205 +       u32     dst;            /* destination from packet header */
 206 +       u32     ctorigsrc;      /* original source from conntrack */
 207 +       u32     ctorigdst;      /* original destination from conntrack */
 208 +       u32     ctreplsrc;      /* reply source from conntrack */
 209 +       u32     ctrepldst;      /* reply destination from conntrack */
 210 +       u32     mark;           /* netfilter mark (fwmark) */
 211 +};
 212 +
 213 +static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a)
 214 +{
 215 +       return jhash_1word(a, q->perturbation) & (q->hash_divisor-1);
 216 +}
 217 +
 218 +static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b)
 219 +{
 220 +       return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1);
 221 +}
 222 +
 223 +static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c)
 224 +{
 225 +       return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1);
 226 +}
 227 +
 228 +static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb)
 229 +{
 230 +       struct esfq_packet_info info;
 231 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 232 +       enum ip_conntrack_info ctinfo;
 233 +       struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 234 +#endif
 235 +
 236 +       switch (skb->protocol) {
 237 +       case __constant_htons(ETH_P_IP):
 238 +       {
 239 +               struct iphdr *iph = ip_hdr(skb);
 240 +               info.dst = iph->daddr;
 241 +               info.src = iph->saddr;
 242 +               if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
 243 +                   (iph->protocol == IPPROTO_TCP ||
 244 +                    iph->protocol == IPPROTO_UDP ||
 245 +                    iph->protocol == IPPROTO_SCTP ||
 246 +                    iph->protocol == IPPROTO_DCCP ||
 247 +                    iph->protocol == IPPROTO_ESP))
 248 +                       info.proto = *(((u32*)iph) + iph->ihl);
 249 +               else
 250 +                       info.proto = iph->protocol;
 251 +               break;
 252 +       }
 253 +       case __constant_htons(ETH_P_IPV6):
 254 +       {
 255 +               struct ipv6hdr *iph = ipv6_hdr(skb);
 256 +               /* Hash ipv6 addresses into a u32. This isn't ideal,
 257 +                * but the code is simple. */
 258 +               info.dst = jhash2(iph->daddr.s6_addr32, 4, q->perturbation);
 259 +               info.src = jhash2(iph->saddr.s6_addr32, 4, q->perturbation);
 260 +               if (iph->nexthdr == IPPROTO_TCP ||
 261 +                   iph->nexthdr == IPPROTO_UDP ||
 262 +                   iph->nexthdr == IPPROTO_SCTP ||
 263 +                   iph->nexthdr == IPPROTO_DCCP ||
 264 +                   iph->nexthdr == IPPROTO_ESP)
 265 +                       info.proto = *(u32*)&iph[1];
 266 +               else
 267 +                       info.proto = iph->nexthdr;
 268 +               break;
 269 +       }
 270 +       default:
 271 +               info.dst   = (u32)(unsigned long)skb->dst;
 272 +               info.src   = (u32)(unsigned long)skb->sk;
 273 +               info.proto = skb->protocol;
 274 +       }
 275 +
 276 +       info.mark = skb->mark;
 277 +
 278 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 279 +       /* defaults if there is no conntrack info */
 280 +       info.ctorigsrc = info.src;
 281 +       info.ctorigdst = info.dst;
 282 +       info.ctreplsrc = info.dst;
 283 +       info.ctrepldst = info.src;
 284 +       /* collect conntrack info */
 285 +       if (ct && ct != &nf_conntrack_untracked) {
 286 +               if (skb->protocol == __constant_htons(ETH_P_IP)) {
 287 +                       info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
 288 +                       info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.ip;
 289 +                       info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip;
 290 +                       info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip;
 291 +               }
 292 +               else if (skb->protocol == __constant_htons(ETH_P_IPV6)) {
 293 +                       /* Again, hash ipv6 addresses into a single u32. */
 294 +                       info.ctorigsrc = jhash2(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6, 4, q->perturbation);
 295 +                       info.ctorigdst = jhash2(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.ip6, 4, q->perturbation);
 296 +                       info.ctreplsrc = jhash2(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip6, 4, q->perturbation);
 297 +                       info.ctrepldst = jhash2(ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip6, 4, q->perturbation);
 298 +               }
 299 +
 300 +       }
 301 +#endif
 302 +
 303 +       switch(q->hash_kind) {
 304 +       case TCA_ESFQ_HASH_CLASSIC:
 305 +               return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 306 +       case TCA_ESFQ_HASH_DST:
 307 +               return esfq_jhash_1word(q, info.dst);
 308 +       case TCA_ESFQ_HASH_SRC:
 309 +               return esfq_jhash_1word(q, info.src);
 310 +       case TCA_ESFQ_HASH_FWMARK:
 311 +               return esfq_jhash_1word(q, info.mark);
 312 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 313 +       case TCA_ESFQ_HASH_CTORIGDST:
 314 +               return esfq_jhash_1word(q, info.ctorigdst);
 315 +       case TCA_ESFQ_HASH_CTORIGSRC:
 316 +               return esfq_jhash_1word(q, info.ctorigsrc);
 317 +       case TCA_ESFQ_HASH_CTREPLDST:
 318 +               return esfq_jhash_1word(q, info.ctrepldst);
 319 +       case TCA_ESFQ_HASH_CTREPLSRC:
 320 +               return esfq_jhash_1word(q, info.ctreplsrc);
 321 +       case TCA_ESFQ_HASH_CTNATCHG:
 322 +       {
 323 +               if (info.ctorigdst == info.ctreplsrc)
 324 +                       return esfq_jhash_1word(q, info.ctorigsrc);
 325 +               return esfq_jhash_1word(q, info.ctreplsrc);
 326 +       }
 327 +#endif
 328 +       default:
 329 +               if (net_ratelimit())
 330 +                       printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n");
 331 +       }
 332 +       return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 333 +}
 334 +
 335 +static inline void esfq_link(struct esfq_sched_data *q, esfq_index x)
 336 +{
 337 +       esfq_index p, n;
 338 +       int d = q->qs[x].qlen + q->depth;
 339 +
 340 +       p = d;
 341 +       n = q->dep[d].next;
 342 +       q->dep[x].next = n;
 343 +       q->dep[x].prev = p;
 344 +       q->dep[p].next = q->dep[n].prev = x;
 345 +}
 346 +
 347 +static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x)
 348 +{
 349 +       esfq_index p, n;
 350 +
 351 +       n = q->dep[x].next;
 352 +       p = q->dep[x].prev;
 353 +       q->dep[p].next = n;
 354 +       q->dep[n].prev = p;
 355 +
 356 +       if (n == p && q->max_depth == q->qs[x].qlen + 1)
 357 +               q->max_depth--;
 358 +
 359 +       esfq_link(q, x);
 360 +}
 361 +
 362 +static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x)
 363 +{
 364 +       esfq_index p, n;
 365 +       int d;
 366 +
 367 +       n = q->dep[x].next;
 368 +       p = q->dep[x].prev;
 369 +       q->dep[p].next = n;
 370 +       q->dep[n].prev = p;
 371 +       d = q->qs[x].qlen;
 372 +       if (q->max_depth < d)
 373 +               q->max_depth = d;
 374 +
 375 +       esfq_link(q, x);
 376 +}
 377 +
 378 +static unsigned int esfq_drop(struct Qdisc *sch)
 379 +{
 380 +       struct esfq_sched_data *q = qdisc_priv(sch);
 381 +       esfq_index d = q->max_depth;
 382 +       struct sk_buff *skb;
 383 +       unsigned int len;
 384 +
 385 +       /* Queue is full! Find the longest slot and
 386 +          drop a packet from it */
 387 +
 388 +       if (d > 1) {
 389 +               esfq_index x = q->dep[d+q->depth].next;
 390 +               skb = q->qs[x].prev;
 391 +               len = skb->len;
 392 +               __skb_unlink(skb, &q->qs[x]);
 393 +               kfree_skb(skb);
 394 +               esfq_dec(q, x);
 395 +               sch->q.qlen--;
 396 +               sch->qstats.drops++;
 397 +               sch->qstats.backlog -= len;
 398 +               return len;
 399 +       }
 400 +
 401 +       if (d == 1) {
 402 +               /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
 403 +               d = q->next[q->tail];
 404 +               q->next[q->tail] = q->next[d];
 405 +               q->allot[q->next[d]] += q->quantum;
 406 +               skb = q->qs[d].prev;
 407 +               len = skb->len;
 408 +               __skb_unlink(skb, &q->qs[d]);
 409 +               kfree_skb(skb);
 410 +               esfq_dec(q, d);
 411 +               sch->q.qlen--;
 412 +               q->ht[q->hash[d]] = q->depth;
 413 +               sch->qstats.drops++;
 414 +               sch->qstats.backlog -= len;
 415 +               return len;
 416 +       }
 417 +
 418 +       return 0;
 419 +}
 420 +
 421 +static void esfq_q_enqueue(struct sk_buff *skb, struct esfq_sched_data *q, unsigned int end)
 422 +{
 423 +       unsigned hash = esfq_hash(q, skb);
 424 +       unsigned depth = q->depth;
 425 +       esfq_index x;
 426 +
 427 +       x = q->ht[hash];
 428 +       if (x == depth) {
 429 +               q->ht[hash] = x = q->dep[depth].next;
 430 +               q->hash[x] = hash;
 431 +       }
 432 +
 433 +       if (end == ESFQ_TAIL)
 434 +               __skb_queue_tail(&q->qs[x], skb);
 435 +       else
 436 +               __skb_queue_head(&q->qs[x], skb);
 437 +
 438 +       esfq_inc(q, x);
 439 +       if (q->qs[x].qlen == 1) {               /* The flow is new */
 440 +               if (q->tail == depth) { /* It is the first flow */
 441 +                       q->tail = x;
 442 +                       q->next[x] = x;
 443 +                       q->allot[x] = q->quantum;
 444 +               } else {
 445 +                       q->next[x] = q->next[q->tail];
 446 +                       q->next[q->tail] = x;
 447 +                       q->tail = x;
 448 +               }
 449 +       }
 450 +}
 451 +
 452 +static int esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 453 +{
 454 +       struct esfq_sched_data *q = qdisc_priv(sch);
 455 +       esfq_q_enqueue(skb, q, ESFQ_TAIL);
 456 +       sch->qstats.backlog += skb->len;
 457 +       if (++sch->q.qlen < q->limit-1) {
 458 +               sch->bstats.bytes += skb->len;
 459 +               sch->bstats.packets++;
 460 +               return 0;
 461 +       }
 462 +
 463 +       sch->qstats.drops++;
 464 +       esfq_drop(sch);
 465 +       return NET_XMIT_CN;
 466 +}
 467 +
 468 +
 469 +static int esfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
 470 +{
 471 +       struct esfq_sched_data *q = qdisc_priv(sch);
 472 +       esfq_q_enqueue(skb, q, ESFQ_HEAD);
 473 +       sch->qstats.backlog += skb->len;
 474 +       if (++sch->q.qlen < q->limit - 1) {
 475 +               sch->qstats.requeues++;
 476 +               return 0;
 477 +       }
 478 +
 479 +       sch->qstats.drops++;
 480 +       esfq_drop(sch);
 481 +       return NET_XMIT_CN;
 482 +}
 483 +
 484 +static struct sk_buff *esfq_q_dequeue(struct esfq_sched_data *q)
 485 +{
 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 +
 500 +       /* Is the slot empty? */
 501 +       if (q->qs[a].qlen == 0) {
 502 +               q->ht[q->hash[a]] = depth;
 503 +               a = q->next[a];
 504 +               if (a == old_a) {
 505 +                       q->tail = depth;
 506 +                       return skb;
 507 +               }
 508 +               q->next[q->tail] = a;
 509 +               q->allot[a] += q->quantum;
 510 +       } else if ((q->allot[a] -= skb->len) <= 0) {
 511 +               q->tail = a;
 512 +               a = q->next[a];
 513 +               q->allot[a] += q->quantum;
 514 +       }
 515 +
 516 +       return skb;
 517 +}
 518 +
 519 +static struct sk_buff *esfq_dequeue(struct Qdisc* sch)
 520 +{
 521 +       struct esfq_sched_data *q = qdisc_priv(sch);
 522 +       struct sk_buff *skb;
 523 +
 524 +       skb = esfq_q_dequeue(q);
 525 +       if (skb == NULL)
 526 +               return NULL;
 527 +       sch->q.qlen--;
 528 +       sch->qstats.backlog -= skb->len;
 529 +       return skb;
 530 +}
 531 +
 532 +static void esfq_q_destroy(struct esfq_sched_data *q)
 533 +{
 534 +       del_timer(&q->perturb_timer);
 535 +       if(q->ht)
 536 +               kfree(q->ht);
 537 +       if(q->dep)
 538 +               kfree(q->dep);
 539 +       if(q->next)
 540 +               kfree(q->next);
 541 +       if(q->allot)
 542 +               kfree(q->allot);
 543 +       if(q->hash)
 544 +               kfree(q->hash);
 545 +       if(q->qs)
 546 +               kfree(q->qs);
 547 +}
 548 +
 549 +static void esfq_destroy(struct Qdisc *sch)
 550 +{
 551 +       struct esfq_sched_data *q = qdisc_priv(sch);
 552 +       esfq_q_destroy(q);
 553 +}
 554 +
 555 +
 556 +static void esfq_reset(struct Qdisc* sch)
 557 +{
 558 +       struct sk_buff *skb;
 559 +
 560 +       while ((skb = esfq_dequeue(sch)) != NULL)
 561 +               kfree_skb(skb);
 562 +}
 563 +
 564 +static void esfq_perturbation(unsigned long arg)
 565 +{
 566 +       struct Qdisc *sch = (struct Qdisc*)arg;
 567 +       struct esfq_sched_data *q = qdisc_priv(sch);
 568 +
 569 +       q->perturbation = net_random()&0x1F;
 570 +
 571 +       if (q->perturb_period) {
 572 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 573 +               add_timer(&q->perturb_timer);
 574 +       }
 575 +}
 576 +
 577 +static unsigned int esfq_check_hash(unsigned int kind)
 578 +{
 579 +       switch (kind) {
 580 +       case TCA_ESFQ_HASH_CTORIGDST:
 581 +       case TCA_ESFQ_HASH_CTORIGSRC:
 582 +       case TCA_ESFQ_HASH_CTREPLDST:
 583 +       case TCA_ESFQ_HASH_CTREPLSRC:
 584 +       case TCA_ESFQ_HASH_CTNATCHG:
 585 +#ifndef CONFIG_NET_SCH_ESFQ_NFCT
 586 +       {
 587 +               if (net_ratelimit())
 588 +                       printk(KERN_WARNING "ESFQ: Conntrack hash types disabled in kernel config. Falling back to classic.\n");
 589 +               return TCA_ESFQ_HASH_CLASSIC;
 590 +       }
 591 +#endif
 592 +       case TCA_ESFQ_HASH_CLASSIC:
 593 +       case TCA_ESFQ_HASH_DST:
 594 +       case TCA_ESFQ_HASH_SRC:
 595 +       case TCA_ESFQ_HASH_FWMARK:
 596 +               return kind;
 597 +       default:
 598 +       {
 599 +               if (net_ratelimit())
 600 +                       printk(KERN_WARNING "ESFQ: Unknown hash type. Falling back to classic.\n");
 601 +               return TCA_ESFQ_HASH_CLASSIC;
 602 +       }
 603 +       }
 604 +}
 605 +
 606 +static int esfq_q_init(struct esfq_sched_data *q, struct rtattr *opt)
 607 +{
 608 +       struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 609 +       esfq_index p = ~0U/2;
 610 +       int i;
 611 +
 612 +       if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 613 +               return -EINVAL;
 614 +
 615 +       q->perturbation = 0;
 616 +       q->hash_kind = TCA_ESFQ_HASH_CLASSIC;
 617 +       q->max_depth = 0;
 618 +       if (opt == NULL) {
 619 +               q->perturb_period = 0;
 620 +               q->hash_divisor = 1024;
 621 +               q->tail = q->limit = q->depth = 128;
 622 +
 623 +       } else {
 624 +               struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 625 +               if (ctl->quantum)
 626 +                       q->quantum = ctl->quantum;
 627 +               q->perturb_period = ctl->perturb_period*HZ;
 628 +               q->hash_divisor = ctl->divisor ? : 1024;
 629 +               q->tail = q->limit = q->depth = ctl->flows ? : 128;
 630 +
 631 +               if ( q->depth > p - 1 )
 632 +                       return -EINVAL;
 633 +
 634 +               if (ctl->limit)
 635 +                       q->limit = min_t(u32, ctl->limit, q->depth);
 636 +
 637 +               if (ctl->hash_kind) {
 638 +                       q->hash_kind = esfq_check_hash(ctl->hash_kind);
 639 +               }
 640 +       }
 641 +
 642 +       q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL);
 643 +       if (!q->ht)
 644 +               goto err_case;
 645 +       q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL);
 646 +       if (!q->dep)
 647 +               goto err_case;
 648 +       q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL);
 649 +       if (!q->next)
 650 +               goto err_case;
 651 +       q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL);
 652 +       if (!q->allot)
 653 +               goto err_case;
 654 +       q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL);
 655 +       if (!q->hash)
 656 +               goto err_case;
 657 +       q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL);
 658 +       if (!q->qs)
 659 +               goto err_case;
 660 +
 661 +       for (i=0; i< q->hash_divisor; i++)
 662 +               q->ht[i] = q->depth;
 663 +       for (i=0; i<q->depth; i++) {
 664 +               skb_queue_head_init(&q->qs[i]);
 665 +               q->dep[i+q->depth].next = i+q->depth;
 666 +               q->dep[i+q->depth].prev = i+q->depth;
 667 +       }
 668 +
 669 +       for (i=0; i<q->depth; i++)
 670 +               esfq_link(q, i);
 671 +       return 0;
 672 +err_case:
 673 +       esfq_q_destroy(q);
 674 +       return -ENOBUFS;
 675 +}
 676 +
 677 +static int esfq_init(struct Qdisc *sch, struct rtattr *opt)
 678 +{
 679 +       struct esfq_sched_data *q = qdisc_priv(sch);
 680 +       int err;
 681 +
 682 +       q->quantum = psched_mtu(sch->dev); /* default */
 683 +       if ((err = esfq_q_init(q, opt)))
 684 +               return err;
 685 +
 686 +       init_timer(&q->perturb_timer);
 687 +       q->perturb_timer.data = (unsigned long)sch;
 688 +       q->perturb_timer.function = esfq_perturbation;
 689 +       if (q->perturb_period) {
 690 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 691 +               add_timer(&q->perturb_timer);
 692 +       }
 693 +
 694 +       return 0;
 695 +}
 696 +
 697 +static int esfq_change(struct Qdisc *sch, struct rtattr *opt)
 698 +{
 699 +       struct esfq_sched_data *q = qdisc_priv(sch);
 700 +       struct esfq_sched_data new;
 701 +       struct sk_buff *skb;
 702 +       int err;
 703 +
 704 +       /* set up new queue */
 705 +       memset(&new, 0, sizeof(struct esfq_sched_data));
 706 +       new.quantum = psched_mtu(sch->dev); /* default */
 707 +       if ((err = esfq_q_init(&new, opt)))
 708 +               return err;
 709 +
 710 +       /* copy all packets from the old queue to the new queue */
 711 +       sch_tree_lock(sch);
 712 +       while ((skb = esfq_q_dequeue(q)) != NULL)
 713 +               esfq_q_enqueue(skb, &new, ESFQ_TAIL);
 714 +
 715 +       /* clean up the old queue */
 716 +       esfq_q_destroy(q);
 717 +
 718 +       /* copy elements of the new queue into the old queue */
 719 +       q->perturb_period = new.perturb_period;
 720 +       q->quantum        = new.quantum;
 721 +       q->limit          = new.limit;
 722 +       q->depth          = new.depth;
 723 +       q->hash_divisor   = new.hash_divisor;
 724 +       q->hash_kind      = new.hash_kind;
 725 +       q->tail           = new.tail;
 726 +       q->max_depth      = new.max_depth;
 727 +       q->ht    = new.ht;
 728 +       q->dep   = new.dep;
 729 +       q->next  = new.next;
 730 +       q->allot = new.allot;
 731 +       q->hash  = new.hash;
 732 +       q->qs    = new.qs;
 733 +
 734 +       /* finish up */
 735 +       if (q->perturb_period) {
 736 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 737 +               add_timer(&q->perturb_timer);
 738 +       } else {
 739 +               q->perturbation = 0;
 740 +       }
 741 +       sch_tree_unlock(sch);
 742 +       return 0;
 743 +}
 744 +
 745 +static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb)
 746 +{
 747 +       struct esfq_sched_data *q = qdisc_priv(sch);
 748 +       unsigned char *b = skb_tail_pointer(skb);
 749 +       struct tc_esfq_qopt opt;
 750 +
 751 +       opt.quantum = q->quantum;
 752 +       opt.perturb_period = q->perturb_period/HZ;
 753 +
 754 +       opt.limit = q->limit;
 755 +       opt.divisor = q->hash_divisor;
 756 +       opt.flows = q->depth;
 757 +       opt.hash_kind = q->hash_kind;
 758 +
 759 +       RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
 760 +
 761 +       return skb->len;
 762 +
 763 +rtattr_failure:
 764 +       nlmsg_trim(skb, b);
 765 +       return -1;
 766 +}
 767 +
 768 +static struct Qdisc_ops esfq_qdisc_ops =
 769 +{
 770 +       .next           =       NULL,
 771 +       .cl_ops         =       NULL,
 772 +       .id             =       "esfq",
 773 +       .priv_size      =       sizeof(struct esfq_sched_data),
 774 +       .enqueue        =       esfq_enqueue,
 775 +       .dequeue        =       esfq_dequeue,
 776 +       .requeue        =       esfq_requeue,
 777 +       .drop           =       esfq_drop,
 778 +       .init           =       esfq_init,
 779 +       .reset          =       esfq_reset,
 780 +       .destroy        =       esfq_destroy,
 781 +       .change         =       esfq_change,
 782 +       .dump           =       esfq_dump,
 783 +       .owner          =       THIS_MODULE,
 784 +};
 785 +
 786 +static int __init esfq_module_init(void)
 787 +{
 788 +       return register_qdisc(&esfq_qdisc_ops);
 789 +}
 790 +static void __exit esfq_module_exit(void)
 791 +{
 792 +       unregister_qdisc(&esfq_qdisc_ops);
 793 +}
 794 +module_init(esfq_module_init)
 795 +module_exit(esfq_module_exit)
 796 +MODULE_LICENSE("GPL");