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

   1 --- a/include/linux/pkt_sched.h
   2 +++ b/include/linux/pkt_sched.h
   3 @@ -162,8 +162,37 @@
   4   *
   5   *     The only reason for this is efficiency, it is possible
   6   *     to change these parameters in compile time.
   7 + *
   8 + *     If you need to play with these values, use esfq instead.
   9   */
  10
  11 +/* ESFQ section */
  12 +
  13 +enum
  14 +{
  15 +        /* traditional */
  16 +       TCA_SFQ_HASH_CLASSIC,
  17 +       TCA_SFQ_HASH_DST,
  18 +       TCA_SFQ_HASH_SRC,
  19 +       TCA_SFQ_HASH_FWMARK,
  20 +       /* conntrack */
  21 +       TCA_SFQ_HASH_CTORIGDST,
  22 +       TCA_SFQ_HASH_CTORIGSRC,
  23 +       TCA_SFQ_HASH_CTREPLDST,
  24 +       TCA_SFQ_HASH_CTREPLSRC,
  25 +       TCA_SFQ_HASH_CTNATCHG,
  26 +};
  27 +
  28 +struct tc_esfq_qopt
  29 +{
  30 +       unsigned        quantum;        /* Bytes per round allocated to flow */
  31 +       int             perturb_period; /* Period of hash perturbation */
  32 +       __u32           limit;          /* Maximal packets in queue */
  33 +       unsigned        divisor;        /* Hash divisor  */
  34 +       unsigned        flows;          /* Maximal number of flows  */
  35 +       unsigned        hash_kind;      /* Hash function to use for flow identification */
  36 +};
  37 +
  38  /* RED section */
  39
  40  enum
  41 --- a/net/sched/Kconfig
  42 +++ b/net/sched/Kconfig
  43 @@ -139,6 +139,37 @@
  44           To compile this code as a module, choose M here: the
  45           module will be called sch_sfq.
  46
  47 +config NET_SCH_ESFQ
  48 +       tristate "Enhanced Stochastic Fairness Queueing (ESFQ)"
  49 +       ---help---
  50 +         Say Y here if you want to use the Enhanced Stochastic Fairness
  51 +         Queueing (ESFQ) packet scheduling algorithm for some of your network
  52 +         devices or as a leaf discipline for a classful qdisc such as HTB or
  53 +         CBQ (see the top of <file:net/sched/sch_esfq.c> for details and
  54 +         references to the SFQ algorithm).
  55 +
  56 +         This is an enchanced SFQ version which allows you to control some
  57 +         hardcoded values in the SFQ scheduler.
  58 +
  59 +         ESFQ also adds control of the hash function used to identify packet
  60 +         flows. The original SFQ discipline hashes by connection; ESFQ add
  61 +         several other hashing methods, such as by src IP or by dst IP, which
  62 +         can be more fair to users in some networking situations.
  63 +
  64 +         To compile this code as a module, choose M here: the
  65 +         module will be called sch_esfq.
  66 +
  67 +config NET_SCH_ESFQ_NFCT
  68 +       bool "Connection Tracking Hash Types"
  69 +       depends on NET_SCH_ESFQ && NF_CONNTRACK
  70 +       ---help---
  71 +         Say Y here to enable support for hashing based on netfilter connection
  72 +         tracking information. This is useful for a router that is also using
  73 +         NAT to connect privately-addressed hosts to the Internet. If you want
  74 +         to provide fair distribution of upstream bandwidth, ESFQ must use
  75 +         connection tracking information, since all outgoing packets will share
  76 +         the same source address.
  77 +
  78  config NET_SCH_TEQL
  79         tristate "True Link Equalizer (TEQL)"
  80         ---help---
  81 --- a/net/sched/Makefile
  82 +++ b/net/sched/Makefile
  83 @@ -23,6 +23,7 @@
  84  obj-$(CONFIG_NET_SCH_INGRESS)  += sch_ingress.o
  85  obj-$(CONFIG_NET_SCH_DSMARK)   += sch_dsmark.o
  86  obj-$(CONFIG_NET_SCH_SFQ)      += sch_sfq.o
  87 +obj-$(CONFIG_NET_SCH_ESFQ)     += sch_esfq.o
  88  obj-$(CONFIG_NET_SCH_TBF)      += sch_tbf.o
  89  obj-$(CONFIG_NET_SCH_TEQL)     += sch_teql.o
  90  obj-$(CONFIG_NET_SCH_PRIO)     += sch_prio.o
  91 --- /dev/null
  92 +++ b/net/sched/sch_esfq.c
  93 @@ -0,0 +1,702 @@
  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 <linux/ipv6.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 +#include <net/netfilter/nf_conntrack.h>
 147 +
 148 +/*     Stochastic Fairness Queuing algorithm.
 149 +       For more comments look at sch_sfq.c.
 150 +       The difference is that you can change limit, depth,
 151 +       hash table size and choose alternate hash types.
 152 +
 153 +       classic:        same as in sch_sfq.c
 154 +       dst:            destination IP address
 155 +       src:            source IP address
 156 +       fwmark:         netfilter mark value
 157 +       ctorigdst:      original destination IP address
 158 +       ctorigsrc:      original source IP address
 159 +       ctrepldst:      reply destination IP address
 160 +       ctreplsrc:      reply source IP
 161 +
 162 +*/
 163 +
 164 +#define ESFQ_HEAD 0
 165 +#define ESFQ_TAIL 1
 166 +
 167 +/* This type should contain at least SFQ_DEPTH*2 values */
 168 +typedef unsigned int esfq_index;
 169 +
 170 +struct esfq_head
 171 +{
 172 +       esfq_index      next;
 173 +       esfq_index      prev;
 174 +};
 175 +
 176 +struct esfq_sched_data
 177 +{
 178 +/* Parameters */
 179 +       int             perturb_period;
 180 +       unsigned        quantum;        /* Allotment per round: MUST BE >= MTU */
 181 +       int             limit;
 182 +       unsigned        depth;
 183 +       unsigned        hash_divisor;
 184 +       unsigned        hash_kind;
 185 +/* Variables */
 186 +       struct timer_list perturb_timer;
 187 +       int             perturbation;
 188 +       esfq_index      tail;           /* Index of current slot in round */
 189 +       esfq_index      max_depth;      /* Maximal depth */
 190 +
 191 +       esfq_index      *ht;                    /* Hash table */
 192 +       esfq_index      *next;                  /* Active slots link */
 193 +       short           *allot;                 /* Current allotment per slot */
 194 +       unsigned short  *hash;                  /* Hash value indexed by slots */
 195 +       struct sk_buff_head     *qs;            /* Slot queue */
 196 +       struct esfq_head        *dep;           /* Linked list of slots, indexed by depth */
 197 +};
 198 +
 199 +/* This contains the info we will hash. */
 200 +struct esfq_packet_info
 201 +{
 202 +       u32     proto;          /* protocol or port */
 203 +       u32     src;            /* source from packet header */
 204 +       u32     dst;            /* destination from packet header */
 205 +       u32     ctorigsrc;      /* original source from conntrack */
 206 +       u32     ctorigdst;      /* original destination from conntrack */
 207 +       u32     ctreplsrc;      /* reply source from conntrack */
 208 +       u32     ctrepldst;      /* reply destination from conntrack */
 209 +       u32     mark;           /* netfilter mark (fwmark) */
 210 +};
 211 +
 212 +static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a)
 213 +{
 214 +       return jhash_1word(a, q->perturbation) & (q->hash_divisor-1);
 215 +}
 216 +
 217 +static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b)
 218 +{
 219 +       return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1);
 220 +}
 221 +
 222 +static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c)
 223 +{
 224 +       return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1);
 225 +}
 226 +
 227 +static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb)
 228 +{
 229 +       struct esfq_packet_info info;
 230 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 231 +       enum ip_conntrack_info ctinfo;
 232 +       struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 233 +#endif
 234 +
 235 +       switch (skb->protocol) {
 236 +       case __constant_htons(ETH_P_IP):
 237 +       {
 238 +               struct iphdr *iph = ip_hdr(skb);
 239 +               info.dst = iph->daddr;
 240 +               info.src = iph->saddr;
 241 +               if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
 242 +                   (iph->protocol == IPPROTO_TCP ||
 243 +                    iph->protocol == IPPROTO_UDP ||
 244 +                    iph->protocol == IPPROTO_SCTP ||
 245 +                    iph->protocol == IPPROTO_DCCP ||
 246 +                    iph->protocol == IPPROTO_ESP))
 247 +                       info.proto = *(((u32*)iph) + iph->ihl);
 248 +               else
 249 +                       info.proto = iph->protocol;
 250 +               break;
 251 +       }
 252 +       case __constant_htons(ETH_P_IPV6):
 253 +       {
 254 +               struct ipv6hdr *iph = ipv6_hdr(skb);
 255 +               /* Hash ipv6 addresses into a u32. This isn't ideal,
 256 +                * but the code is simple. */
 257 +               info.dst = jhash2(iph->daddr.s6_addr32, 4, q->perturbation);
 258 +               info.src = jhash2(iph->saddr.s6_addr32, 4, q->perturbation);
 259 +               if (iph->nexthdr == IPPROTO_TCP ||
 260 +                   iph->nexthdr == IPPROTO_UDP ||
 261 +                   iph->nexthdr == IPPROTO_SCTP ||
 262 +                   iph->nexthdr == IPPROTO_DCCP ||
 263 +                   iph->nexthdr == IPPROTO_ESP)
 264 +                       info.proto = *(u32*)&iph[1];
 265 +               else
 266 +                       info.proto = iph->nexthdr;
 267 +               break;
 268 +       }
 269 +       default:
 270 +               info.dst   = (u32)(unsigned long)skb->dst;
 271 +               info.src   = (u32)(unsigned long)skb->sk;
 272 +               info.proto = skb->protocol;
 273 +       }
 274 +
 275 +       info.mark = skb->mark;
 276 +
 277 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 278 +       /* defaults if there is no conntrack info */
 279 +       info.ctorigsrc = info.src;
 280 +       info.ctorigdst = info.dst;
 281 +       info.ctreplsrc = info.dst;
 282 +       info.ctrepldst = info.src;
 283 +       /* collect conntrack info */
 284 +       if (ct && ct != &nf_conntrack_untracked) {
 285 +               if (skb->protocol == __constant_htons(ETH_P_IP)) {
 286 +                       info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
 287 +                       info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.ip;
 288 +                       info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip;
 289 +                       info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip;
 290 +               }
 291 +               else if (skb->protocol == __constant_htons(ETH_P_IPV6)) {
 292 +                       /* Again, hash ipv6 addresses into a single u32. */
 293 +                       info.ctorigsrc = jhash2(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6, 4, q->perturbation);
 294 +                       info.ctorigdst = jhash2(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.ip6, 4, q->perturbation);
 295 +                       info.ctreplsrc = jhash2(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip6, 4, q->perturbation);
 296 +                       info.ctrepldst = jhash2(ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip6, 4, q->perturbation);
 297 +               }
 298 +
 299 +       }
 300 +#endif
 301 +
 302 +       switch(q->hash_kind) {
 303 +       case TCA_SFQ_HASH_CLASSIC:
 304 +               return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 305 +       case TCA_SFQ_HASH_DST:
 306 +               return esfq_jhash_1word(q, info.dst);
 307 +       case TCA_SFQ_HASH_SRC:
 308 +               return esfq_jhash_1word(q, info.src);
 309 +       case TCA_SFQ_HASH_FWMARK:
 310 +               return esfq_jhash_1word(q, info.mark);
 311 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 312 +       case TCA_SFQ_HASH_CTORIGDST:
 313 +               return esfq_jhash_1word(q, info.ctorigdst);
 314 +       case TCA_SFQ_HASH_CTORIGSRC:
 315 +               return esfq_jhash_1word(q, info.ctorigsrc);
 316 +       case TCA_SFQ_HASH_CTREPLDST:
 317 +               return esfq_jhash_1word(q, info.ctrepldst);
 318 +       case TCA_SFQ_HASH_CTREPLSRC:
 319 +               return esfq_jhash_1word(q, info.ctreplsrc);
 320 +       case TCA_SFQ_HASH_CTNATCHG:
 321 +       {
 322 +               if (info.ctorigdst == info.ctreplsrc)
 323 +                       return esfq_jhash_1word(q, info.ctorigsrc);
 324 +               return esfq_jhash_1word(q, info.ctreplsrc);
 325 +       }
 326 +#endif
 327 +       default:
 328 +               if (net_ratelimit())
 329 +                       printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n");
 330 +       }
 331 +       return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 332 +}
 333 +
 334 +static inline void esfq_link(struct esfq_sched_data *q, esfq_index x)
 335 +{
 336 +       esfq_index p, n;
 337 +       int d = q->qs[x].qlen + q->depth;
 338 +
 339 +       p = d;
 340 +       n = q->dep[d].next;
 341 +       q->dep[x].next = n;
 342 +       q->dep[x].prev = p;
 343 +       q->dep[p].next = q->dep[n].prev = x;
 344 +}
 345 +
 346 +static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x)
 347 +{
 348 +       esfq_index p, n;
 349 +
 350 +       n = q->dep[x].next;
 351 +       p = q->dep[x].prev;
 352 +       q->dep[p].next = n;
 353 +       q->dep[n].prev = p;
 354 +
 355 +       if (n == p && q->max_depth == q->qs[x].qlen + 1)
 356 +               q->max_depth--;
 357 +
 358 +       esfq_link(q, x);
 359 +}
 360 +
 361 +static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x)
 362 +{
 363 +       esfq_index p, n;
 364 +       int d;
 365 +
 366 +       n = q->dep[x].next;
 367 +       p = q->dep[x].prev;
 368 +       q->dep[p].next = n;
 369 +       q->dep[n].prev = p;
 370 +       d = q->qs[x].qlen;
 371 +       if (q->max_depth < d)
 372 +               q->max_depth = d;
 373 +
 374 +       esfq_link(q, x);
 375 +}
 376 +
 377 +static unsigned int esfq_drop(struct Qdisc *sch)
 378 +{
 379 +       struct esfq_sched_data *q = qdisc_priv(sch);
 380 +       esfq_index d = q->max_depth;
 381 +       struct sk_buff *skb;
 382 +       unsigned int len;
 383 +
 384 +       /* Queue is full! Find the longest slot and
 385 +          drop a packet from it */
 386 +
 387 +       if (d > 1) {
 388 +               esfq_index x = q->dep[d+q->depth].next;
 389 +               skb = q->qs[x].prev;
 390 +               len = skb->len;
 391 +               __skb_unlink(skb, &q->qs[x]);
 392 +               kfree_skb(skb);
 393 +               esfq_dec(q, x);
 394 +               sch->q.qlen--;
 395 +               sch->qstats.drops++;
 396 +               sch->qstats.backlog -= len;
 397 +               return len;
 398 +       }
 399 +
 400 +       if (d == 1) {
 401 +               /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
 402 +               d = q->next[q->tail];
 403 +               q->next[q->tail] = q->next[d];
 404 +               q->allot[q->next[d]] += q->quantum;
 405 +               skb = q->qs[d].prev;
 406 +               len = skb->len;
 407 +               __skb_unlink(skb, &q->qs[d]);
 408 +               kfree_skb(skb);
 409 +               esfq_dec(q, d);
 410 +               sch->q.qlen--;
 411 +               q->ht[q->hash[d]] = q->depth;
 412 +               sch->qstats.drops++;
 413 +               sch->qstats.backlog -= len;
 414 +               return len;
 415 +       }
 416 +
 417 +       return 0;
 418 +}
 419 +
 420 +static void esfq_q_enqueue(struct sk_buff *skb, struct esfq_sched_data *q, unsigned int end)
 421 +{
 422 +       unsigned hash = esfq_hash(q, skb);
 423 +       unsigned depth = q->depth;
 424 +       esfq_index x;
 425 +
 426 +       x = q->ht[hash];
 427 +       if (x == depth) {
 428 +               q->ht[hash] = x = q->dep[depth].next;
 429 +               q->hash[x] = hash;
 430 +       }
 431 +
 432 +       if (end == ESFQ_TAIL)
 433 +               __skb_queue_tail(&q->qs[x], skb);
 434 +       else
 435 +               __skb_queue_head(&q->qs[x], skb);
 436 +
 437 +       esfq_inc(q, x);
 438 +       if (q->qs[x].qlen == 1) {               /* The flow is new */
 439 +               if (q->tail == depth) { /* It is the first flow */
 440 +                       q->tail = x;
 441 +                       q->next[x] = x;
 442 +                       q->allot[x] = q->quantum;
 443 +               } else {
 444 +                       q->next[x] = q->next[q->tail];
 445 +                       q->next[q->tail] = x;
 446 +                       q->tail = x;
 447 +               }
 448 +       }
 449 +}
 450 +
 451 +static int esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 452 +{
 453 +       struct esfq_sched_data *q = qdisc_priv(sch);
 454 +       esfq_q_enqueue(skb, q, ESFQ_TAIL);
 455 +       sch->qstats.backlog += skb->len;
 456 +       if (++sch->q.qlen < q->limit-1) {
 457 +               sch->bstats.bytes += skb->len;
 458 +               sch->bstats.packets++;
 459 +               return 0;
 460 +       }
 461 +
 462 +       sch->qstats.drops++;
 463 +       esfq_drop(sch);
 464 +       return NET_XMIT_CN;
 465 +}
 466 +
 467 +
 468 +static int esfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
 469 +{
 470 +       struct esfq_sched_data *q = qdisc_priv(sch);
 471 +       esfq_q_enqueue(skb, q, ESFQ_HEAD);
 472 +       sch->qstats.backlog += skb->len;
 473 +       if (++sch->q.qlen < q->limit - 1) {
 474 +               sch->qstats.requeues++;
 475 +               return 0;
 476 +       }
 477 +
 478 +       sch->qstats.drops++;
 479 +       esfq_drop(sch);
 480 +       return NET_XMIT_CN;
 481 +}
 482 +
 483 +static struct sk_buff *esfq_q_dequeue(struct esfq_sched_data *q)
 484 +{
 485 +       struct sk_buff *skb;
 486 +       unsigned depth = q->depth;
 487 +       esfq_index a, old_a;
 488 +
 489 +       /* No active slots */
 490 +       if (q->tail == depth)
 491 +               return NULL;
 492 +
 493 +       a = old_a = q->next[q->tail];
 494 +
 495 +       /* Grab packet */
 496 +       skb = __skb_dequeue(&q->qs[a]);
 497 +       esfq_dec(q, a);
 498 +
 499 +       /* Is the slot empty? */
 500 +       if (q->qs[a].qlen == 0) {
 501 +               q->ht[q->hash[a]] = depth;
 502 +               a = q->next[a];
 503 +               if (a == old_a) {
 504 +                       q->tail = depth;
 505 +                       return skb;
 506 +               }
 507 +               q->next[q->tail] = a;
 508 +               q->allot[a] += q->quantum;
 509 +       } else if ((q->allot[a] -= skb->len) <= 0) {
 510 +               q->tail = a;
 511 +               a = q->next[a];
 512 +               q->allot[a] += q->quantum;
 513 +       }
 514 +
 515 +       return skb;
 516 +}
 517 +
 518 +static struct sk_buff *esfq_dequeue(struct Qdisc* sch)
 519 +{
 520 +       struct esfq_sched_data *q = qdisc_priv(sch);
 521 +       struct sk_buff *skb;
 522 +
 523 +       skb = esfq_q_dequeue(q);
 524 +       if (skb == NULL)
 525 +               return NULL;
 526 +       sch->q.qlen--;
 527 +       sch->qstats.backlog -= skb->len;
 528 +       return skb;
 529 +}
 530 +
 531 +static void esfq_q_destroy(struct esfq_sched_data *q)
 532 +{
 533 +       del_timer(&q->perturb_timer);
 534 +       if(q->ht)
 535 +               kfree(q->ht);
 536 +       if(q->dep)
 537 +               kfree(q->dep);
 538 +       if(q->next)
 539 +               kfree(q->next);
 540 +       if(q->allot)
 541 +               kfree(q->allot);
 542 +       if(q->hash)
 543 +               kfree(q->hash);
 544 +       if(q->qs)
 545 +               kfree(q->qs);
 546 +}
 547 +
 548 +static void esfq_destroy(struct Qdisc *sch)
 549 +{
 550 +       struct esfq_sched_data *q = qdisc_priv(sch);
 551 +       esfq_q_destroy(q);
 552 +}
 553 +
 554 +
 555 +static void esfq_reset(struct Qdisc* sch)
 556 +{
 557 +       struct sk_buff *skb;
 558 +
 559 +       while ((skb = esfq_dequeue(sch)) != NULL)
 560 +               kfree_skb(skb);
 561 +}
 562 +
 563 +static void esfq_perturbation(unsigned long arg)
 564 +{
 565 +       struct Qdisc *sch = (struct Qdisc*)arg;
 566 +       struct esfq_sched_data *q = qdisc_priv(sch);
 567 +
 568 +       q->perturbation = net_random()&0x1F;
 569 +
 570 +       if (q->perturb_period) {
 571 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 572 +               add_timer(&q->perturb_timer);
 573 +       }
 574 +}
 575 +
 576 +static unsigned int esfq_check_hash(unsigned int kind)
 577 +{
 578 +       switch (kind) {
 579 +       case TCA_SFQ_HASH_CTORIGDST:
 580 +       case TCA_SFQ_HASH_CTORIGSRC:
 581 +       case TCA_SFQ_HASH_CTREPLDST:
 582 +       case TCA_SFQ_HASH_CTREPLSRC:
 583 +       case TCA_SFQ_HASH_CTNATCHG:
 584 +#ifndef CONFIG_NET_SCH_ESFQ_NFCT
 585 +       {
 586 +               if (net_ratelimit())
 587 +                       printk(KERN_WARNING "ESFQ: Conntrack hash types disabled in kernel config. Falling back to classic.\n");
 588 +               return TCA_SFQ_HASH_CLASSIC;
 589 +       }
 590 +#endif
 591 +       case TCA_SFQ_HASH_CLASSIC:
 592 +       case TCA_SFQ_HASH_DST:
 593 +       case TCA_SFQ_HASH_SRC:
 594 +       case TCA_SFQ_HASH_FWMARK:
 595 +               return kind;
 596 +       default:
 597 +       {
 598 +               if (net_ratelimit())
 599 +                       printk(KERN_WARNING "ESFQ: Unknown hash type. Falling back to classic.\n");
 600 +               return TCA_SFQ_HASH_CLASSIC;
 601 +       }
 602 +       }
 603 +}
 604 +
 605 +static int esfq_q_init(struct esfq_sched_data *q, struct rtattr *opt)
 606 +{
 607 +       struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 608 +       esfq_index p = ~0U/2;
 609 +       int i;
 610 +
 611 +       if (opt && opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
 612 +               return -EINVAL;
 613 +
 614 +       q->perturbation = 0;
 615 +       q->hash_kind = TCA_SFQ_HASH_CLASSIC;
 616 +       q->max_depth = 0;
 617 +       if (opt == NULL) {
 618 +               q->perturb_period = 0;
 619 +               q->hash_divisor = 1024;
 620 +               q->tail = q->limit = q->depth = 128;
 621 +
 622 +       } else {
 623 +               struct tc_esfq_qopt *ctl = RTA_DATA(opt);
 624 +               if (ctl->quantum)
 625 +                       q->quantum = ctl->quantum;
 626 +               q->perturb_period = ctl->perturb_period*HZ;
 627 +               q->hash_divisor = ctl->divisor ? : 1024;
 628 +               q->tail = q->limit = q->depth = ctl->flows ? : 128;
 629 +
 630 +               if ( q->depth > p - 1 )
 631 +                       return -EINVAL;
 632 +
 633 +               if (ctl->limit)
 634 +                       q->limit = min_t(u32, ctl->limit, q->depth);
 635 +
 636 +               if (ctl->hash_kind) {
 637 +                       q->hash_kind = esfq_check_hash(ctl->hash_kind);
 638 +               }
 639 +       }
 640 +
 641 +       q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL);
 642 +       if (!q->ht)
 643 +               goto err_case;
 644 +       q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL);
 645 +       if (!q->dep)
 646 +               goto err_case;
 647 +       q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL);
 648 +       if (!q->next)
 649 +               goto err_case;
 650 +       q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL);
 651 +       if (!q->allot)
 652 +               goto err_case;
 653 +       q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL);
 654 +       if (!q->hash)
 655 +               goto err_case;
 656 +       q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL);
 657 +       if (!q->qs)
 658 +               goto err_case;
 659 +
 660 +       for (i=0; i< q->hash_divisor; i++)
 661 +               q->ht[i] = q->depth;
 662 +       for (i=0; i<q->depth; i++) {
 663 +               skb_queue_head_init(&q->qs[i]);
 664 +               q->dep[i+q->depth].next = i+q->depth;
 665 +               q->dep[i+q->depth].prev = i+q->depth;
 666 +       }
 667 +
 668 +       for (i=0; i<q->depth; i++)
 669 +               esfq_link(q, i);
 670 +       return 0;
 671 +err_case:
 672 +       esfq_q_destroy(q);
 673 +       return -ENOBUFS;
 674 +}
 675 +
 676 +static int esfq_init(struct Qdisc *sch, struct rtattr *opt)
 677 +{
 678 +       struct esfq_sched_data *q = qdisc_priv(sch);
 679 +       int err;
 680 +
 681 +       q->quantum = psched_mtu(sch->dev); /* default */
 682 +       if ((err = esfq_q_init(q, opt)))
 683 +               return err;
 684 +
 685 +       init_timer(&q->perturb_timer);
 686 +       q->perturb_timer.data = (unsigned long)sch;
 687 +       q->perturb_timer.function = esfq_perturbation;
 688 +       if (q->perturb_period) {
 689 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 690 +               add_timer(&q->perturb_timer);
 691 +       }
 692 +
 693 +       return 0;
 694 +}
 695 +
 696 +static int esfq_change(struct Qdisc *sch, struct rtattr *opt)
 697 +{
 698 +       struct esfq_sched_data *q = qdisc_priv(sch);
 699 +       struct esfq_sched_data new;
 700 +       struct sk_buff *skb;
 701 +       int err;
 702 +
 703 +       /* set up new queue */
 704 +       memset(&new, 0, sizeof(struct esfq_sched_data));
 705 +       new.quantum = psched_mtu(sch->dev); /* default */
 706 +       if ((err = esfq_q_init(&new, opt)))
 707 +               return err;
 708 +
 709 +       /* copy all packets from the old queue to the new queue */
 710 +       sch_tree_lock(sch);
 711 +       while ((skb = esfq_q_dequeue(q)) != NULL)
 712 +               esfq_q_enqueue(skb, &new, ESFQ_TAIL);
 713 +
 714 +       /* clean up the old queue */
 715 +       esfq_q_destroy(q);
 716 +
 717 +       /* copy elements of the new queue into the old queue */
 718 +       q->perturb_period = new.perturb_period;
 719 +       q->quantum        = new.quantum;
 720 +       q->limit          = new.limit;
 721 +       q->depth          = new.depth;
 722 +       q->hash_divisor   = new.hash_divisor;
 723 +       q->hash_kind      = new.hash_kind;
 724 +       q->tail           = new.tail;
 725 +       q->max_depth      = new.max_depth;
 726 +       q->ht    = new.ht;
 727 +       q->dep   = new.dep;
 728 +       q->next  = new.next;
 729 +       q->allot = new.allot;
 730 +       q->hash  = new.hash;
 731 +       q->qs    = new.qs;
 732 +
 733 +       /* finish up */
 734 +       if (q->perturb_period) {
 735 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 736 +               add_timer(&q->perturb_timer);
 737 +       } else {
 738 +               q->perturbation = 0;
 739 +       }
 740 +       sch_tree_unlock(sch);
 741 +       return 0;
 742 +}
 743 +
 744 +static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb)
 745 +{
 746 +       struct esfq_sched_data *q = qdisc_priv(sch);
 747 +       unsigned char *b = skb->tail;
 748 +       struct tc_esfq_qopt opt;
 749 +
 750 +       opt.quantum = q->quantum;
 751 +       opt.perturb_period = q->perturb_period/HZ;
 752 +
 753 +       opt.limit = q->limit;
 754 +       opt.divisor = q->hash_divisor;
 755 +       opt.flows = q->depth;
 756 +       opt.hash_kind = q->hash_kind;
 757 +
 758 +       RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
 759 +
 760 +       return skb->len;
 761 +
 762 +rtattr_failure:
 763 +       skb_trim(skb, b - skb->data);
 764 +       return -1;
 765 +}
 766 +
 767 +static struct Qdisc_ops esfq_qdisc_ops =
 768 +{
 769 +       .next           =       NULL,
 770 +       .cl_ops         =       NULL,
 771 +       .id             =       "esfq",
 772 +       .priv_size      =       sizeof(struct esfq_sched_data),
 773 +       .enqueue        =       esfq_enqueue,
 774 +       .dequeue        =       esfq_dequeue,
 775 +       .requeue        =       esfq_requeue,
 776 +       .drop           =       esfq_drop,
 777 +       .init           =       esfq_init,
 778 +       .reset          =       esfq_reset,
 779 +       .destroy        =       esfq_destroy,
 780 +       .change         =       esfq_change,
 781 +       .dump           =       esfq_dump,
 782 +       .owner          =       THIS_MODULE,
 783 +};
 784 +
 785 +static int __init esfq_module_init(void)
 786 +{
 787 +       return register_qdisc(&esfq_qdisc_ops);
 788 +}
 789 +static void __exit esfq_module_exit(void)
 790 +{
 791 +       unregister_qdisc(&esfq_qdisc_ops);
 792 +}
 793 +module_init(esfq_module_init)
 794 +module_exit(esfq_module_exit)
 795 +MODULE_LICENSE("GPL");