target/linux/generic-2.6/patches-2.6.25/200-sched_esfq.patch

   1 Index: linux-2.6.25.4/include/linux/pkt_sched.h
   2 ===================================================================
   3 --- linux-2.6.25.4.orig/include/linux/pkt_sched.h
   4 +++ linux-2.6.25.4/include/linux/pkt_sched.h
   5 @@ -162,8 +162,37 @@ struct tc_sfq_xstats
   6   *
   7   *     The only reason for this is efficiency, it is possible
   8   *     to change these parameters in compile time.
   9 + *
  10 + *     If you need to play with these values, use esfq instead.
  11   */
  12
  13 +/* ESFQ section */
  14 +
  15 +enum
  16 +{
  17 +        /* traditional */
  18 +       TCA_SFQ_HASH_CLASSIC,
  19 +       TCA_SFQ_HASH_DST,
  20 +       TCA_SFQ_HASH_SRC,
  21 +       TCA_SFQ_HASH_FWMARK,
  22 +       /* conntrack */
  23 +       TCA_SFQ_HASH_CTORIGDST,
  24 +       TCA_SFQ_HASH_CTORIGSRC,
  25 +       TCA_SFQ_HASH_CTREPLDST,
  26 +       TCA_SFQ_HASH_CTREPLSRC,
  27 +       TCA_SFQ_HASH_CTNATCHG,
  28 +};
  29 +
  30 +struct tc_esfq_qopt
  31 +{
  32 +       unsigned        quantum;        /* Bytes per round allocated to flow */
  33 +       int             perturb_period; /* Period of hash perturbation */
  34 +       __u32           limit;          /* Maximal packets in queue */
  35 +       unsigned        divisor;        /* Hash divisor  */
  36 +       unsigned        flows;          /* Maximal number of flows  */
  37 +       unsigned        hash_kind;      /* Hash function to use for flow identification */
  38 +};
  39 +
  40  /* RED section */
  41
  42  enum
  43 Index: linux-2.6.25.4/net/sched/Kconfig
  44 ===================================================================
  45 --- linux-2.6.25.4.orig/net/sched/Kconfig
  46 +++ linux-2.6.25.4/net/sched/Kconfig
  47 @@ -139,6 +139,37 @@ config NET_SCH_SFQ
  48           To compile this code as a module, choose M here: the
  49           module will be called sch_sfq.
  50
  51 +config NET_SCH_ESFQ
  52 +       tristate "Enhanced Stochastic Fairness Queueing (ESFQ)"
  53 +       ---help---
  54 +         Say Y here if you want to use the Enhanced Stochastic Fairness
  55 +         Queueing (ESFQ) packet scheduling algorithm for some of your network
  56 +         devices or as a leaf discipline for a classful qdisc such as HTB or
  57 +         CBQ (see the top of <file:net/sched/sch_esfq.c> for details and
  58 +         references to the SFQ algorithm).
  59 +
  60 +         This is an enchanced SFQ version which allows you to control some
  61 +         hardcoded values in the SFQ scheduler.
  62 +
  63 +         ESFQ also adds control of the hash function used to identify packet
  64 +         flows. The original SFQ discipline hashes by connection; ESFQ add
  65 +         several other hashing methods, such as by src IP or by dst IP, which
  66 +         can be more fair to users in some networking situations.
  67 +
  68 +         To compile this code as a module, choose M here: the
  69 +         module will be called sch_esfq.
  70 +
  71 +config NET_SCH_ESFQ_NFCT
  72 +       bool "Connection Tracking Hash Types"
  73 +       depends on NET_SCH_ESFQ && NF_CONNTRACK
  74 +       ---help---
  75 +         Say Y here to enable support for hashing based on netfilter connection
  76 +         tracking information. This is useful for a router that is also using
  77 +         NAT to connect privately-addressed hosts to the Internet. If you want
  78 +         to provide fair distribution of upstream bandwidth, ESFQ must use
  79 +         connection tracking information, since all outgoing packets will share
  80 +         the same source address.
  81 +
  82  config NET_SCH_TEQL
  83         tristate "True Link Equalizer (TEQL)"
  84         ---help---
  85 Index: linux-2.6.25.4/net/sched/Makefile
  86 ===================================================================
  87 --- linux-2.6.25.4.orig/net/sched/Makefile
  88 +++ linux-2.6.25.4/net/sched/Makefile
  89 @@ -23,6 +23,7 @@ obj-$(CONFIG_NET_SCH_GRED)    += sch_gred.o
  90  obj-$(CONFIG_NET_SCH_INGRESS)  += sch_ingress.o
  91  obj-$(CONFIG_NET_SCH_DSMARK)   += sch_dsmark.o
  92  obj-$(CONFIG_NET_SCH_SFQ)      += sch_sfq.o
  93 +obj-$(CONFIG_NET_SCH_ESFQ)     += sch_esfq.o
  94  obj-$(CONFIG_NET_SCH_TBF)      += sch_tbf.o
  95  obj-$(CONFIG_NET_SCH_TEQL)     += sch_teql.o
  96  obj-$(CONFIG_NET_SCH_PRIO)     += sch_prio.o
  97 Index: linux-2.6.25.4/net/sched/sch_esfq.c
  98 ===================================================================
  99 --- /dev/null
 100 +++ linux-2.6.25.4/net/sched/sch_esfq.c
 101 @@ -0,0 +1,702 @@
 102 +/*
 103 + * net/sched/sch_esfq.c        Extended Stochastic Fairness Queueing discipline.
 104 + *
 105 + *             This program is free software; you can redistribute it and/or
 106 + *             modify it under the terms of the GNU General Public License
 107 + *             as published by the Free Software Foundation; either version
 108 + *             2 of the License, or (at your option) any later version.
 109 + *
 110 + * Authors:    Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 111 + *
 112 + * Changes:    Alexander Atanasov, <alex@ssi.bg>
 113 + *             Added dynamic depth,limit,divisor,hash_kind options.
 114 + *             Added dst and src hashes.
 115 + *
 116 + *             Alexander Clouter, <alex@digriz.org.uk>
 117 + *             Ported ESFQ to Linux 2.6.
 118 + *
 119 + *             Corey Hickey, <bugfood-c@fatooh.org>
 120 + *             Maintenance of the Linux 2.6 port.
 121 + *             Added fwmark hash (thanks to Robert Kurjata).
 122 + *             Added usage of jhash.
 123 + *             Added conntrack support.
 124 + *             Added ctnatchg hash (thanks to Ben Pfountz).
 125 + */
 126 +
 127 +#include <linux/module.h>
 128 +#include <asm/uaccess.h>
 129 +#include <asm/system.h>
 130 +#include <linux/bitops.h>
 131 +#include <linux/types.h>
 132 +#include <linux/kernel.h>
 133 +#include <linux/jiffies.h>
 134 +#include <linux/string.h>
 135 +#include <linux/mm.h>
 136 +#include <linux/socket.h>
 137 +#include <linux/sockios.h>
 138 +#include <linux/in.h>
 139 +#include <linux/errno.h>
 140 +#include <linux/interrupt.h>
 141 +#include <linux/if_ether.h>
 142 +#include <linux/inet.h>
 143 +#include <linux/netdevice.h>
 144 +#include <linux/etherdevice.h>
 145 +#include <linux/notifier.h>
 146 +#include <linux/init.h>
 147 +#include <net/ip.h>
 148 +#include <linux/ipv6.h>
 149 +#include <net/route.h>
 150 +#include <linux/skbuff.h>
 151 +#include <net/sock.h>
 152 +#include <net/pkt_sched.h>
 153 +#include <linux/jhash.h>
 154 +#include <net/netfilter/nf_conntrack.h>
 155 +
 156 +/*     Stochastic Fairness Queuing algorithm.
 157 +       For more comments look at sch_sfq.c.
 158 +       The difference is that you can change limit, depth,
 159 +       hash table size and choose alternate hash types.
 160 +
 161 +       classic:        same as in sch_sfq.c
 162 +       dst:            destination IP address
 163 +       src:            source IP address
 164 +       fwmark:         netfilter mark value
 165 +       ctorigdst:      original destination IP address
 166 +       ctorigsrc:      original source IP address
 167 +       ctrepldst:      reply destination IP address
 168 +       ctreplsrc:      reply source IP
 169 +
 170 +*/
 171 +
 172 +#define ESFQ_HEAD 0
 173 +#define ESFQ_TAIL 1
 174 +
 175 +/* This type should contain at least SFQ_DEPTH*2 values */
 176 +typedef unsigned int esfq_index;
 177 +
 178 +struct esfq_head
 179 +{
 180 +       esfq_index      next;
 181 +       esfq_index      prev;
 182 +};
 183 +
 184 +struct esfq_sched_data
 185 +{
 186 +/* Parameters */
 187 +       int             perturb_period;
 188 +       unsigned        quantum;        /* Allotment per round: MUST BE >= MTU */
 189 +       int             limit;
 190 +       unsigned        depth;
 191 +       unsigned        hash_divisor;
 192 +       unsigned        hash_kind;
 193 +/* Variables */
 194 +       struct timer_list perturb_timer;
 195 +       int             perturbation;
 196 +       esfq_index      tail;           /* Index of current slot in round */
 197 +       esfq_index      max_depth;      /* Maximal depth */
 198 +
 199 +       esfq_index      *ht;                    /* Hash table */
 200 +       esfq_index      *next;                  /* Active slots link */
 201 +       short           *allot;                 /* Current allotment per slot */
 202 +       unsigned short  *hash;                  /* Hash value indexed by slots */
 203 +       struct sk_buff_head     *qs;            /* Slot queue */
 204 +       struct esfq_head        *dep;           /* Linked list of slots, indexed by depth */
 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 +       u32     mark;           /* netfilter mark (fwmark) */
 218 +};
 219 +
 220 +static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a)
 221 +{
 222 +       return jhash_1word(a, q->perturbation) & (q->hash_divisor-1);
 223 +}
 224 +
 225 +static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b)
 226 +{
 227 +       return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1);
 228 +}
 229 +
 230 +static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c)
 231 +{
 232 +       return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1);
 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_NET_SCH_ESFQ_NFCT
 239 +       enum ip_conntrack_info ctinfo;
 240 +       struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 241 +#endif
 242 +
 243 +       switch (skb->protocol) {
 244 +       case __constant_htons(ETH_P_IP):
 245 +       {
 246 +               struct iphdr *iph = ip_hdr(skb);
 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 +       case __constant_htons(ETH_P_IPV6):
 261 +       {
 262 +               struct ipv6hdr *iph = ipv6_hdr(skb);
 263 +               /* Hash ipv6 addresses into a u32. This isn't ideal,
 264 +                * but the code is simple. */
 265 +               info.dst = jhash2(iph->daddr.s6_addr32, 4, q->perturbation);
 266 +               info.src = jhash2(iph->saddr.s6_addr32, 4, q->perturbation);
 267 +               if (iph->nexthdr == IPPROTO_TCP ||
 268 +                   iph->nexthdr == IPPROTO_UDP ||
 269 +                   iph->nexthdr == IPPROTO_SCTP ||
 270 +                   iph->nexthdr == IPPROTO_DCCP ||
 271 +                   iph->nexthdr == IPPROTO_ESP)
 272 +                       info.proto = *(u32*)&iph[1];
 273 +               else
 274 +                       info.proto = iph->nexthdr;
 275 +               break;
 276 +       }
 277 +       default:
 278 +               info.dst   = (u32)(unsigned long)skb->dst;
 279 +               info.src   = (u32)(unsigned long)skb->sk;
 280 +               info.proto = skb->protocol;
 281 +       }
 282 +
 283 +       info.mark = skb->mark;
 284 +
 285 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 286 +       /* defaults if there is no conntrack info */
 287 +       info.ctorigsrc = info.src;
 288 +       info.ctorigdst = info.dst;
 289 +       info.ctreplsrc = info.dst;
 290 +       info.ctrepldst = info.src;
 291 +       /* collect conntrack info */
 292 +       if (ct && ct != &nf_conntrack_untracked) {
 293 +               if (skb->protocol == __constant_htons(ETH_P_IP)) {
 294 +                       info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
 295 +                       info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.ip;
 296 +                       info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip;
 297 +                       info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip;
 298 +               }
 299 +               else if (skb->protocol == __constant_htons(ETH_P_IPV6)) {
 300 +                       /* Again, hash ipv6 addresses into a single u32. */
 301 +                       info.ctorigsrc = jhash2(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6, 4, q->perturbation);
 302 +                       info.ctorigdst = jhash2(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.ip6, 4, q->perturbation);
 303 +                       info.ctreplsrc = jhash2(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip6, 4, q->perturbation);
 304 +                       info.ctrepldst = jhash2(ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip6, 4, q->perturbation);
 305 +               }
 306 +
 307 +       }
 308 +#endif
 309 +
 310 +       switch(q->hash_kind) {
 311 +       case TCA_SFQ_HASH_CLASSIC:
 312 +               return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 313 +       case TCA_SFQ_HASH_DST:
 314 +               return esfq_jhash_1word(q, info.dst);
 315 +       case TCA_SFQ_HASH_SRC:
 316 +               return esfq_jhash_1word(q, info.src);
 317 +       case TCA_SFQ_HASH_FWMARK:
 318 +               return esfq_jhash_1word(q, info.mark);
 319 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 320 +       case TCA_SFQ_HASH_CTORIGDST:
 321 +               return esfq_jhash_1word(q, info.ctorigdst);
 322 +       case TCA_SFQ_HASH_CTORIGSRC:
 323 +               return esfq_jhash_1word(q, info.ctorigsrc);
 324 +       case TCA_SFQ_HASH_CTREPLDST:
 325 +               return esfq_jhash_1word(q, info.ctrepldst);
 326 +       case TCA_SFQ_HASH_CTREPLSRC:
 327 +               return esfq_jhash_1word(q, info.ctreplsrc);
 328 +       case TCA_SFQ_HASH_CTNATCHG:
 329 +       {
 330 +               if (info.ctorigdst == info.ctreplsrc)
 331 +                       return esfq_jhash_1word(q, info.ctorigsrc);
 332 +               return esfq_jhash_1word(q, info.ctreplsrc);
 333 +       }
 334 +#endif
 335 +       default:
 336 +               if (net_ratelimit())
 337 +                       printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n");
 338 +       }
 339 +       return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 340 +}
 341 +
 342 +static inline void esfq_link(struct esfq_sched_data *q, esfq_index x)
 343 +{
 344 +       esfq_index p, n;
 345 +       int d = q->qs[x].qlen + q->depth;
 346 +
 347 +       p = d;
 348 +       n = q->dep[d].next;
 349 +       q->dep[x].next = n;
 350 +       q->dep[x].prev = p;
 351 +       q->dep[p].next = q->dep[n].prev = x;
 352 +}
 353 +
 354 +static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x)
 355 +{
 356 +       esfq_index p, n;
 357 +
 358 +       n = q->dep[x].next;
 359 +       p = q->dep[x].prev;
 360 +       q->dep[p].next = n;
 361 +       q->dep[n].prev = p;
 362 +
 363 +       if (n == p && q->max_depth == q->qs[x].qlen + 1)
 364 +               q->max_depth--;
 365 +
 366 +       esfq_link(q, x);
 367 +}
 368 +
 369 +static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x)
 370 +{
 371 +       esfq_index p, n;
 372 +       int d;
 373 +
 374 +       n = q->dep[x].next;
 375 +       p = q->dep[x].prev;
 376 +       q->dep[p].next = n;
 377 +       q->dep[n].prev = p;
 378 +       d = q->qs[x].qlen;
 379 +       if (q->max_depth < d)
 380 +               q->max_depth = d;
 381 +
 382 +       esfq_link(q, x);
 383 +}
 384 +
 385 +static unsigned int esfq_drop(struct Qdisc *sch)
 386 +{
 387 +       struct esfq_sched_data *q = qdisc_priv(sch);
 388 +       esfq_index d = q->max_depth;
 389 +       struct sk_buff *skb;
 390 +       unsigned int len;
 391 +
 392 +       /* Queue is full! Find the longest slot and
 393 +          drop a packet from it */
 394 +
 395 +       if (d > 1) {
 396 +               esfq_index x = q->dep[d+q->depth].next;
 397 +               skb = q->qs[x].prev;
 398 +               len = skb->len;
 399 +               __skb_unlink(skb, &q->qs[x]);
 400 +               kfree_skb(skb);
 401 +               esfq_dec(q, x);
 402 +               sch->q.qlen--;
 403 +               sch->qstats.drops++;
 404 +               sch->qstats.backlog -= len;
 405 +               return len;
 406 +       }
 407 +
 408 +       if (d == 1) {
 409 +               /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
 410 +               d = q->next[q->tail];
 411 +               q->next[q->tail] = q->next[d];
 412 +               q->allot[q->next[d]] += q->quantum;
 413 +               skb = q->qs[d].prev;
 414 +               len = skb->len;
 415 +               __skb_unlink(skb, &q->qs[d]);
 416 +               kfree_skb(skb);
 417 +               esfq_dec(q, d);
 418 +               sch->q.qlen--;
 419 +               q->ht[q->hash[d]] = q->depth;
 420 +               sch->qstats.drops++;
 421 +               sch->qstats.backlog -= len;
 422 +               return len;
 423 +       }
 424 +
 425 +       return 0;
 426 +}
 427 +
 428 +static void esfq_q_enqueue(struct sk_buff *skb, struct esfq_sched_data *q, unsigned int end)
 429 +{
 430 +       unsigned hash = esfq_hash(q, skb);
 431 +       unsigned depth = q->depth;
 432 +       esfq_index x;
 433 +
 434 +       x = q->ht[hash];
 435 +       if (x == depth) {
 436 +               q->ht[hash] = x = q->dep[depth].next;
 437 +               q->hash[x] = hash;
 438 +       }
 439 +
 440 +       if (end == ESFQ_TAIL)
 441 +               __skb_queue_tail(&q->qs[x], skb);
 442 +       else
 443 +               __skb_queue_head(&q->qs[x], skb);
 444 +
 445 +       esfq_inc(q, x);
 446 +       if (q->qs[x].qlen == 1) {               /* The flow is new */
 447 +               if (q->tail == depth) { /* It is the first flow */
 448 +                       q->tail = x;
 449 +                       q->next[x] = x;
 450 +                       q->allot[x] = q->quantum;
 451 +               } else {
 452 +                       q->next[x] = q->next[q->tail];
 453 +                       q->next[q->tail] = x;
 454 +                       q->tail = x;
 455 +               }
 456 +       }
 457 +}
 458 +
 459 +static int esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 460 +{
 461 +       struct esfq_sched_data *q = qdisc_priv(sch);
 462 +       esfq_q_enqueue(skb, q, ESFQ_TAIL);
 463 +       sch->qstats.backlog += skb->len;
 464 +       if (++sch->q.qlen < q->limit-1) {
 465 +               sch->bstats.bytes += skb->len;
 466 +               sch->bstats.packets++;
 467 +               return 0;
 468 +       }
 469 +
 470 +       sch->qstats.drops++;
 471 +       esfq_drop(sch);
 472 +       return NET_XMIT_CN;
 473 +}
 474 +
 475 +
 476 +static int esfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
 477 +{
 478 +       struct esfq_sched_data *q = qdisc_priv(sch);
 479 +       esfq_q_enqueue(skb, q, ESFQ_HEAD);
 480 +       sch->qstats.backlog += skb->len;
 481 +       if (++sch->q.qlen < q->limit - 1) {
 482 +               sch->qstats.requeues++;
 483 +               return 0;
 484 +       }
 485 +
 486 +       sch->qstats.drops++;
 487 +       esfq_drop(sch);
 488 +       return NET_XMIT_CN;
 489 +}
 490 +
 491 +static struct sk_buff *esfq_q_dequeue(struct esfq_sched_data *q)
 492 +{
 493 +       struct sk_buff *skb;
 494 +       unsigned depth = q->depth;
 495 +       esfq_index a, old_a;
 496 +
 497 +       /* No active slots */
 498 +       if (q->tail == depth)
 499 +               return NULL;
 500 +
 501 +       a = old_a = q->next[q->tail];
 502 +
 503 +       /* Grab packet */
 504 +       skb = __skb_dequeue(&q->qs[a]);
 505 +       esfq_dec(q, a);
 506 +
 507 +       /* Is the slot empty? */
 508 +       if (q->qs[a].qlen == 0) {
 509 +               q->ht[q->hash[a]] = depth;
 510 +               a = q->next[a];
 511 +               if (a == old_a) {
 512 +                       q->tail = depth;
 513 +                       return skb;
 514 +               }
 515 +               q->next[q->tail] = a;
 516 +               q->allot[a] += q->quantum;
 517 +       } else if ((q->allot[a] -= skb->len) <= 0) {
 518 +               q->tail = a;
 519 +               a = q->next[a];
 520 +               q->allot[a] += q->quantum;
 521 +       }
 522 +
 523 +       return skb;
 524 +}
 525 +
 526 +static struct sk_buff *esfq_dequeue(struct Qdisc* sch)
 527 +{
 528 +       struct esfq_sched_data *q = qdisc_priv(sch);
 529 +       struct sk_buff *skb;
 530 +
 531 +       skb = esfq_q_dequeue(q);
 532 +       if (skb == NULL)
 533 +               return NULL;
 534 +       sch->q.qlen--;
 535 +       sch->qstats.backlog -= skb->len;
 536 +       return skb;
 537 +}
 538 +
 539 +static void esfq_q_destroy(struct esfq_sched_data *q)
 540 +{
 541 +       del_timer(&q->perturb_timer);
 542 +       if(q->ht)
 543 +               kfree(q->ht);
 544 +       if(q->dep)
 545 +               kfree(q->dep);
 546 +       if(q->next)
 547 +               kfree(q->next);
 548 +       if(q->allot)
 549 +               kfree(q->allot);
 550 +       if(q->hash)
 551 +               kfree(q->hash);
 552 +       if(q->qs)
 553 +               kfree(q->qs);
 554 +}
 555 +
 556 +static void esfq_destroy(struct Qdisc *sch)
 557 +{
 558 +       struct esfq_sched_data *q = qdisc_priv(sch);
 559 +       esfq_q_destroy(q);
 560 +}
 561 +
 562 +
 563 +static void esfq_reset(struct Qdisc* sch)
 564 +{
 565 +       struct sk_buff *skb;
 566 +
 567 +       while ((skb = esfq_dequeue(sch)) != NULL)
 568 +               kfree_skb(skb);
 569 +}
 570 +
 571 +static void esfq_perturbation(unsigned long arg)
 572 +{
 573 +       struct Qdisc *sch = (struct Qdisc*)arg;
 574 +       struct esfq_sched_data *q = qdisc_priv(sch);
 575 +
 576 +       q->perturbation = net_random()&0x1F;
 577 +
 578 +       if (q->perturb_period) {
 579 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 580 +               add_timer(&q->perturb_timer);
 581 +       }
 582 +}
 583 +
 584 +static unsigned int esfq_check_hash(unsigned int kind)
 585 +{
 586 +       switch (kind) {
 587 +       case TCA_SFQ_HASH_CTORIGDST:
 588 +       case TCA_SFQ_HASH_CTORIGSRC:
 589 +       case TCA_SFQ_HASH_CTREPLDST:
 590 +       case TCA_SFQ_HASH_CTREPLSRC:
 591 +       case TCA_SFQ_HASH_CTNATCHG:
 592 +#ifndef CONFIG_NET_SCH_ESFQ_NFCT
 593 +       {
 594 +               if (net_ratelimit())
 595 +                       printk(KERN_WARNING "ESFQ: Conntrack hash types disabled in kernel config. Falling back to classic.\n");
 596 +               return TCA_SFQ_HASH_CLASSIC;
 597 +       }
 598 +#endif
 599 +       case TCA_SFQ_HASH_CLASSIC:
 600 +       case TCA_SFQ_HASH_DST:
 601 +       case TCA_SFQ_HASH_SRC:
 602 +       case TCA_SFQ_HASH_FWMARK:
 603 +               return kind;
 604 +       default:
 605 +       {
 606 +               if (net_ratelimit())
 607 +                       printk(KERN_WARNING "ESFQ: Unknown hash type. Falling back to classic.\n");
 608 +               return TCA_SFQ_HASH_CLASSIC;
 609 +       }
 610 +       }
 611 +}
 612 +
 613 +static int esfq_q_init(struct esfq_sched_data *q, struct rtattr *opt)
 614 +{
 615 +       struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 616 +       esfq_index p = ~0U/2;
 617 +       int i;
 618 +
 619 +       if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 620 +               return -EINVAL;
 621 +
 622 +       q->perturbation = 0;
 623 +       q->hash_kind = TCA_SFQ_HASH_CLASSIC;
 624 +       q->max_depth = 0;
 625 +       if (opt == NULL) {
 626 +               q->perturb_period = 0;
 627 +               q->hash_divisor = 1024;
 628 +               q->tail = q->limit = q->depth = 128;
 629 +
 630 +       } else {
 631 +               struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 632 +               if (ctl->quantum)
 633 +                       q->quantum = ctl->quantum;
 634 +               q->perturb_period = ctl->perturb_period*HZ;
 635 +               q->hash_divisor = ctl->divisor ? : 1024;
 636 +               q->tail = q->limit = q->depth = ctl->flows ? : 128;
 637 +
 638 +               if ( q->depth > p - 1 )
 639 +                       return -EINVAL;
 640 +
 641 +               if (ctl->limit)
 642 +                       q->limit = min_t(u32, ctl->limit, q->depth);
 643 +
 644 +               if (ctl->hash_kind) {
 645 +                       q->hash_kind = esfq_check_hash(ctl->hash_kind);
 646 +               }
 647 +       }
 648 +
 649 +       q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL);
 650 +       if (!q->ht)
 651 +               goto err_case;
 652 +       q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL);
 653 +       if (!q->dep)
 654 +               goto err_case;
 655 +       q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL);
 656 +       if (!q->next)
 657 +               goto err_case;
 658 +       q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL);
 659 +       if (!q->allot)
 660 +               goto err_case;
 661 +       q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL);
 662 +       if (!q->hash)
 663 +               goto err_case;
 664 +       q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL);
 665 +       if (!q->qs)
 666 +               goto err_case;
 667 +
 668 +       for (i=0; i< q->hash_divisor; i++)
 669 +               q->ht[i] = q->depth;
 670 +       for (i=0; i<q->depth; i++) {
 671 +               skb_queue_head_init(&q->qs[i]);
 672 +               q->dep[i+q->depth].next = i+q->depth;
 673 +               q->dep[i+q->depth].prev = i+q->depth;
 674 +       }
 675 +
 676 +       for (i=0; i<q->depth; i++)
 677 +               esfq_link(q, i);
 678 +       return 0;
 679 +err_case:
 680 +       esfq_q_destroy(q);
 681 +       return -ENOBUFS;
 682 +}
 683 +
 684 +static int esfq_init(struct Qdisc *sch, struct rtattr *opt)
 685 +{
 686 +       struct esfq_sched_data *q = qdisc_priv(sch);
 687 +       int err;
 688 +
 689 +       q->quantum = psched_mtu(sch->dev); /* default */
 690 +       if ((err = esfq_q_init(q, opt)))
 691 +               return err;
 692 +
 693 +       init_timer(&q->perturb_timer);
 694 +       q->perturb_timer.data = (unsigned long)sch;
 695 +       q->perturb_timer.function = esfq_perturbation;
 696 +       if (q->perturb_period) {
 697 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 698 +               add_timer(&q->perturb_timer);
 699 +       }
 700 +
 701 +       return 0;
 702 +}
 703 +
 704 +static int esfq_change(struct Qdisc *sch, struct rtattr *opt)
 705 +{
 706 +       struct esfq_sched_data *q = qdisc_priv(sch);
 707 +       struct esfq_sched_data new;
 708 +       struct sk_buff *skb;
 709 +       int err;
 710 +
 711 +       /* set up new queue */
 712 +       memset(&new, 0, sizeof(struct esfq_sched_data));
 713 +       new.quantum = psched_mtu(sch->dev); /* default */
 714 +       if ((err = esfq_q_init(&new, opt)))
 715 +               return err;
 716 +
 717 +       /* copy all packets from the old queue to the new queue */
 718 +       sch_tree_lock(sch);
 719 +       while ((skb = esfq_q_dequeue(q)) != NULL)
 720 +               esfq_q_enqueue(skb, &new, ESFQ_TAIL);
 721 +
 722 +       /* clean up the old queue */
 723 +       esfq_q_destroy(q);
 724 +
 725 +       /* copy elements of the new queue into the old queue */
 726 +       q->perturb_period = new.perturb_period;
 727 +       q->quantum        = new.quantum;
 728 +       q->limit          = new.limit;
 729 +       q->depth          = new.depth;
 730 +       q->hash_divisor   = new.hash_divisor;
 731 +       q->hash_kind      = new.hash_kind;
 732 +       q->tail           = new.tail;
 733 +       q->max_depth      = new.max_depth;
 734 +       q->ht    = new.ht;
 735 +       q->dep   = new.dep;
 736 +       q->next  = new.next;
 737 +       q->allot = new.allot;
 738 +       q->hash  = new.hash;
 739 +       q->qs    = new.qs;
 740 +
 741 +       /* finish up */
 742 +       if (q->perturb_period) {
 743 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 744 +               add_timer(&q->perturb_timer);
 745 +       } else {
 746 +               q->perturbation = 0;
 747 +       }
 748 +       sch_tree_unlock(sch);
 749 +       return 0;
 750 +}
 751 +
 752 +static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb)
 753 +{
 754 +       struct esfq_sched_data *q = qdisc_priv(sch);
 755 +       unsigned char *b = skb->tail;
 756 +       struct tc_esfq_qopt opt;
 757 +
 758 +       opt.quantum = q->quantum;
 759 +       opt.perturb_period = q->perturb_period/HZ;
 760 +
 761 +       opt.limit = q->limit;
 762 +       opt.divisor = q->hash_divisor;
 763 +       opt.flows = q->depth;
 764 +       opt.hash_kind = q->hash_kind;
 765 +
 766 +       RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
 767 +
 768 +       return skb->len;
 769 +
 770 +rtattr_failure:
 771 +       skb_trim(skb, b - skb->data);
 772 +       return -1;
 773 +}
 774 +
 775 +static struct Qdisc_ops esfq_qdisc_ops =
 776 +{
 777 +       .next           =       NULL,
 778 +       .cl_ops         =       NULL,
 779 +       .id             =       "esfq",
 780 +       .priv_size      =       sizeof(struct esfq_sched_data),
 781 +       .enqueue        =       esfq_enqueue,
 782 +       .dequeue        =       esfq_dequeue,
 783 +       .requeue        =       esfq_requeue,
 784 +       .drop           =       esfq_drop,
 785 +       .init           =       esfq_init,
 786 +       .reset          =       esfq_reset,
 787 +       .destroy        =       esfq_destroy,
 788 +       .change         =       esfq_change,
 789 +       .dump           =       esfq_dump,
 790 +       .owner          =       THIS_MODULE,
 791 +};
 792 +
 793 +static int __init esfq_module_init(void)
 794 +{
 795 +       return register_qdisc(&esfq_qdisc_ops);
 796 +}
 797 +static void __exit esfq_module_exit(void)
 798 +{
 799 +       unregister_qdisc(&esfq_qdisc_ops);
 800 +}
 801 +module_init(esfq_module_init)
 802 +module_exit(esfq_module_exit)
 803 +MODULE_LICENSE("GPL");