target/linux/generic/patches-3.13/620-sched_esfq.patch

   1 --- a/include/uapi/linux/pkt_sched.h
   2 +++ b/include/uapi/linux/pkt_sched.h
   3 @@ -224,6 +224,33 @@ struct tc_sfq_xstats {
   4         __s32           allot;
   5  };
   6
   7 +/* ESFQ section */
   8 +
   9 +enum
  10 +{
  11 +        /* traditional */
  12 +       TCA_SFQ_HASH_CLASSIC,
  13 +       TCA_SFQ_HASH_DST,
  14 +       TCA_SFQ_HASH_SRC,
  15 +       TCA_SFQ_HASH_FWMARK,
  16 +       /* conntrack */
  17 +       TCA_SFQ_HASH_CTORIGDST,
  18 +       TCA_SFQ_HASH_CTORIGSRC,
  19 +       TCA_SFQ_HASH_CTREPLDST,
  20 +       TCA_SFQ_HASH_CTREPLSRC,
  21 +       TCA_SFQ_HASH_CTNATCHG,
  22 +};
  23 +
  24 +struct tc_esfq_qopt
  25 +{
  26 +       unsigned        quantum;        /* Bytes per round allocated to flow */
  27 +       int             perturb_period; /* Period of hash perturbation */
  28 +       __u32           limit;          /* Maximal packets in queue */
  29 +       unsigned        divisor;        /* Hash divisor  */
  30 +       unsigned        flows;          /* Maximal number of flows  */
  31 +       unsigned        hash_kind;      /* Hash function to use for flow identification */
  32 +};
  33 +
  34  /* RED section */
  35
  36  enum {
  37 --- a/net/sched/Kconfig
  38 +++ b/net/sched/Kconfig
  39 @@ -148,6 +148,37 @@ config NET_SCH_SFQ
  40           To compile this code as a module, choose M here: the
  41           module will be called sch_sfq.
  42
  43 +config NET_SCH_ESFQ
  44 +       tristate "Enhanced Stochastic Fairness Queueing (ESFQ)"
  45 +       ---help---
  46 +         Say Y here if you want to use the Enhanced Stochastic Fairness
  47 +         Queueing (ESFQ) packet scheduling algorithm for some of your network
  48 +         devices or as a leaf discipline for a classful qdisc such as HTB or
  49 +         CBQ (see the top of <file:net/sched/sch_esfq.c> for details and
  50 +         references to the SFQ algorithm).
  51 +
  52 +         This is an enchanced SFQ version which allows you to control some
  53 +         hardcoded values in the SFQ scheduler.
  54 +
  55 +         ESFQ also adds control of the hash function used to identify packet
  56 +         flows. The original SFQ discipline hashes by connection; ESFQ add
  57 +         several other hashing methods, such as by src IP or by dst IP, which
  58 +         can be more fair to users in some networking situations.
  59 +
  60 +         To compile this code as a module, choose M here: the
  61 +         module will be called sch_esfq.
  62 +
  63 +config NET_SCH_ESFQ_NFCT
  64 +       bool "Connection Tracking Hash Types"
  65 +       depends on NET_SCH_ESFQ && NF_CONNTRACK
  66 +       ---help---
  67 +         Say Y here to enable support for hashing based on netfilter connection
  68 +         tracking information. This is useful for a router that is also using
  69 +         NAT to connect privately-addressed hosts to the Internet. If you want
  70 +         to provide fair distribution of upstream bandwidth, ESFQ must use
  71 +         connection tracking information, since all outgoing packets will share
  72 +         the same source address.
  73 +
  74  config NET_SCH_TEQL
  75         tristate "True Link Equalizer (TEQL)"
  76         ---help---
  77 --- a/net/sched/Makefile
  78 +++ b/net/sched/Makefile
  79 @@ -26,6 +26,7 @@ obj-$(CONFIG_NET_SCH_INGRESS) += sch_ing
  80  obj-$(CONFIG_NET_SCH_DSMARK)   += sch_dsmark.o
  81  obj-$(CONFIG_NET_SCH_SFB)      += sch_sfb.o
  82  obj-$(CONFIG_NET_SCH_SFQ)      += sch_sfq.o
  83 +obj-$(CONFIG_NET_SCH_ESFQ)     += sch_esfq.o
  84  obj-$(CONFIG_NET_SCH_TBF)      += sch_tbf.o
  85  obj-$(CONFIG_NET_SCH_TEQL)     += sch_teql.o
  86  obj-$(CONFIG_NET_SCH_PRIO)     += sch_prio.o
  87 --- /dev/null
  88 +++ b/net/sched/sch_esfq.c
  89 @@ -0,0 +1,702 @@
  90 +/*
  91 + * net/sched/sch_esfq.c        Extended Stochastic Fairness Queueing discipline.
  92 + *
  93 + *             This program is free software; you can redistribute it and/or
  94 + *             modify it under the terms of the GNU General Public License
  95 + *             as published by the Free Software Foundation; either version
  96 + *             2 of the License, or (at your option) any later version.
  97 + *
  98 + * Authors:    Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  99 + *
 100 + * Changes:    Alexander Atanasov, <alex@ssi.bg>
 101 + *             Added dynamic depth,limit,divisor,hash_kind options.
 102 + *             Added dst and src hashes.
 103 + *
 104 + *             Alexander Clouter, <alex@digriz.org.uk>
 105 + *             Ported ESFQ to Linux 2.6.
 106 + *
 107 + *             Corey Hickey, <bugfood-c@fatooh.org>
 108 + *             Maintenance of the Linux 2.6 port.
 109 + *             Added fwmark hash (thanks to Robert Kurjata).
 110 + *             Added usage of jhash.
 111 + *             Added conntrack support.
 112 + *             Added ctnatchg hash (thanks to Ben Pfountz).
 113 + */
 114 +
 115 +#include <linux/module.h>
 116 +#include <asm/uaccess.h>
 117 +#include <linux/bitops.h>
 118 +#include <linux/types.h>
 119 +#include <linux/kernel.h>
 120 +#include <linux/jiffies.h>
 121 +#include <linux/string.h>
 122 +#include <linux/mm.h>
 123 +#include <linux/socket.h>
 124 +#include <linux/sockios.h>
 125 +#include <linux/in.h>
 126 +#include <linux/errno.h>
 127 +#include <linux/interrupt.h>
 128 +#include <linux/if_ether.h>
 129 +#include <linux/inet.h>
 130 +#include <linux/netdevice.h>
 131 +#include <linux/etherdevice.h>
 132 +#include <linux/notifier.h>
 133 +#include <linux/init.h>
 134 +#include <net/ip.h>
 135 +#include <net/netlink.h>
 136 +#include <linux/ipv6.h>
 137 +#include <net/route.h>
 138 +#include <linux/skbuff.h>
 139 +#include <net/sock.h>
 140 +#include <net/pkt_sched.h>
 141 +#include <linux/jhash.h>
 142 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 143 +#include <net/netfilter/nf_conntrack.h>
 144 +#endif
 145 +
 146 +/*     Stochastic Fairness Queuing algorithm.
 147 +       For more comments look at sch_sfq.c.
 148 +       The difference is that you can change limit, depth,
 149 +       hash table size and choose alternate hash types.
 150 +
 151 +       classic:        same as in sch_sfq.c
 152 +       dst:            destination IP address
 153 +       src:            source IP address
 154 +       fwmark:         netfilter mark value
 155 +       ctorigdst:      original destination IP address
 156 +       ctorigsrc:      original source IP address
 157 +       ctrepldst:      reply destination IP address
 158 +       ctreplsrc:      reply source IP
 159 +
 160 +*/
 161 +
 162 +#define ESFQ_HEAD 0
 163 +#define ESFQ_TAIL 1
 164 +
 165 +/* This type should contain at least SFQ_DEPTH*2 values */
 166 +typedef unsigned int esfq_index;
 167 +
 168 +struct esfq_head
 169 +{
 170 +       esfq_index      next;
 171 +       esfq_index      prev;
 172 +};
 173 +
 174 +struct esfq_sched_data
 175 +{
 176 +/* Parameters */
 177 +       int             perturb_period;
 178 +       unsigned        quantum;        /* Allotment per round: MUST BE >= MTU */
 179 +       int             limit;
 180 +       unsigned        depth;
 181 +       unsigned        hash_divisor;
 182 +       unsigned        hash_kind;
 183 +/* Variables */
 184 +       struct timer_list perturb_timer;
 185 +       int             perturbation;
 186 +       esfq_index      tail;           /* Index of current slot in round */
 187 +       esfq_index      max_depth;      /* Maximal depth */
 188 +
 189 +       esfq_index      *ht;                    /* Hash table */
 190 +       esfq_index      *next;                  /* Active slots link */
 191 +       short           *allot;                 /* Current allotment per slot */
 192 +       unsigned short  *hash;                  /* Hash value indexed by slots */
 193 +       struct sk_buff_head     *qs;            /* Slot queue */
 194 +       struct esfq_head        *dep;           /* Linked list of slots, indexed by depth */
 195 +};
 196 +
 197 +/* This contains the info we will hash. */
 198 +struct esfq_packet_info
 199 +{
 200 +       u32     proto;          /* protocol or port */
 201 +       u32     src;            /* source from packet header */
 202 +       u32     dst;            /* destination from packet header */
 203 +       u32     ctorigsrc;      /* original source from conntrack */
 204 +       u32     ctorigdst;      /* original destination from conntrack */
 205 +       u32     ctreplsrc;      /* reply source from conntrack */
 206 +       u32     ctrepldst;      /* reply destination from conntrack */
 207 +       u32     mark;           /* netfilter mark (fwmark) */
 208 +};
 209 +
 210 +static __inline__ unsigned esfq_jhash_1word(struct esfq_sched_data *q,u32 a)
 211 +{
 212 +       return jhash_1word(a, q->perturbation) & (q->hash_divisor-1);
 213 +}
 214 +
 215 +static __inline__ unsigned esfq_jhash_2words(struct esfq_sched_data *q, u32 a, u32 b)
 216 +{
 217 +       return jhash_2words(a, b, q->perturbation) & (q->hash_divisor-1);
 218 +}
 219 +
 220 +static __inline__ unsigned esfq_jhash_3words(struct esfq_sched_data *q, u32 a, u32 b, u32 c)
 221 +{
 222 +       return jhash_3words(a, b, c, q->perturbation) & (q->hash_divisor-1);
 223 +}
 224 +
 225 +static unsigned esfq_hash(struct esfq_sched_data *q, struct sk_buff *skb)
 226 +{
 227 +       struct esfq_packet_info info;
 228 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 229 +       enum ip_conntrack_info ctinfo;
 230 +       struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
 231 +#endif
 232 +
 233 +       switch (skb->protocol) {
 234 +       case __constant_htons(ETH_P_IP):
 235 +       {
 236 +               struct iphdr *iph = ip_hdr(skb);
 237 +               info.dst = iph->daddr;
 238 +               info.src = iph->saddr;
 239 +               if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
 240 +                   (iph->protocol == IPPROTO_TCP ||
 241 +                    iph->protocol == IPPROTO_UDP ||
 242 +                    iph->protocol == IPPROTO_SCTP ||
 243 +                    iph->protocol == IPPROTO_DCCP ||
 244 +                    iph->protocol == IPPROTO_ESP))
 245 +                       info.proto = *(((u32*)iph) + iph->ihl);
 246 +               else
 247 +                       info.proto = iph->protocol;
 248 +               break;
 249 +       }
 250 +       case __constant_htons(ETH_P_IPV6):
 251 +       {
 252 +               struct ipv6hdr *iph = ipv6_hdr(skb);
 253 +               /* Hash ipv6 addresses into a u32. This isn't ideal,
 254 +                * but the code is simple. */
 255 +               info.dst = jhash2(iph->daddr.s6_addr32, 4, q->perturbation);
 256 +               info.src = jhash2(iph->saddr.s6_addr32, 4, q->perturbation);
 257 +               if (iph->nexthdr == IPPROTO_TCP ||
 258 +                   iph->nexthdr == IPPROTO_UDP ||
 259 +                   iph->nexthdr == IPPROTO_SCTP ||
 260 +                   iph->nexthdr == IPPROTO_DCCP ||
 261 +                   iph->nexthdr == IPPROTO_ESP)
 262 +                       info.proto = *(u32*)&iph[1];
 263 +               else
 264 +                       info.proto = iph->nexthdr;
 265 +               break;
 266 +       }
 267 +       default:
 268 +               info.dst   = (u32)(unsigned long)skb_dst(skb);
 269 +               info.src   = (u32)(unsigned long)skb->sk;
 270 +               info.proto = skb->protocol;
 271 +       }
 272 +
 273 +       info.mark = skb->mark;
 274 +
 275 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 276 +       /* defaults if there is no conntrack info */
 277 +       info.ctorigsrc = info.src;
 278 +       info.ctorigdst = info.dst;
 279 +       info.ctreplsrc = info.dst;
 280 +       info.ctrepldst = info.src;
 281 +       /* collect conntrack info */
 282 +       if (ct && ct != &nf_conntrack_untracked) {
 283 +               if (skb->protocol == __constant_htons(ETH_P_IP)) {
 284 +                       info.ctorigsrc = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip;
 285 +                       info.ctorigdst = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.ip;
 286 +                       info.ctreplsrc = ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip;
 287 +                       info.ctrepldst = ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip;
 288 +               }
 289 +               else if (skb->protocol == __constant_htons(ETH_P_IPV6)) {
 290 +                       /* Again, hash ipv6 addresses into a single u32. */
 291 +                       info.ctorigsrc = jhash2(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.ip6, 4, q->perturbation);
 292 +                       info.ctorigdst = jhash2(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.ip6, 4, q->perturbation);
 293 +                       info.ctreplsrc = jhash2(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip6, 4, q->perturbation);
 294 +                       info.ctrepldst = jhash2(ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip6, 4, q->perturbation);
 295 +               }
 296 +
 297 +       }
 298 +#endif
 299 +
 300 +       switch(q->hash_kind) {
 301 +       case TCA_SFQ_HASH_CLASSIC:
 302 +               return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 303 +       case TCA_SFQ_HASH_DST:
 304 +               return esfq_jhash_1word(q, info.dst);
 305 +       case TCA_SFQ_HASH_SRC:
 306 +               return esfq_jhash_1word(q, info.src);
 307 +       case TCA_SFQ_HASH_FWMARK:
 308 +               return esfq_jhash_1word(q, info.mark);
 309 +#ifdef CONFIG_NET_SCH_ESFQ_NFCT
 310 +       case TCA_SFQ_HASH_CTORIGDST:
 311 +               return esfq_jhash_1word(q, info.ctorigdst);
 312 +       case TCA_SFQ_HASH_CTORIGSRC:
 313 +               return esfq_jhash_1word(q, info.ctorigsrc);
 314 +       case TCA_SFQ_HASH_CTREPLDST:
 315 +               return esfq_jhash_1word(q, info.ctrepldst);
 316 +       case TCA_SFQ_HASH_CTREPLSRC:
 317 +               return esfq_jhash_1word(q, info.ctreplsrc);
 318 +       case TCA_SFQ_HASH_CTNATCHG:
 319 +       {
 320 +               if (info.ctorigdst == info.ctreplsrc)
 321 +                       return esfq_jhash_1word(q, info.ctorigsrc);
 322 +               return esfq_jhash_1word(q, info.ctreplsrc);
 323 +       }
 324 +#endif
 325 +       default:
 326 +               if (net_ratelimit())
 327 +                       printk(KERN_WARNING "ESFQ: Unknown hash method. Falling back to classic.\n");
 328 +       }
 329 +       return esfq_jhash_3words(q, info.dst, info.src, info.proto);
 330 +}
 331 +
 332 +static inline void esfq_link(struct esfq_sched_data *q, esfq_index x)
 333 +{
 334 +       esfq_index p, n;
 335 +       int d = q->qs[x].qlen + q->depth;
 336 +
 337 +       p = d;
 338 +       n = q->dep[d].next;
 339 +       q->dep[x].next = n;
 340 +       q->dep[x].prev = p;
 341 +       q->dep[p].next = q->dep[n].prev = x;
 342 +}
 343 +
 344 +static inline void esfq_dec(struct esfq_sched_data *q, esfq_index x)
 345 +{
 346 +       esfq_index p, n;
 347 +
 348 +       n = q->dep[x].next;
 349 +       p = q->dep[x].prev;
 350 +       q->dep[p].next = n;
 351 +       q->dep[n].prev = p;
 352 +
 353 +       if (n == p && q->max_depth == q->qs[x].qlen + 1)
 354 +               q->max_depth--;
 355 +
 356 +       esfq_link(q, x);
 357 +}
 358 +
 359 +static inline void esfq_inc(struct esfq_sched_data *q, esfq_index x)
 360 +{
 361 +       esfq_index p, n;
 362 +       int d;
 363 +
 364 +       n = q->dep[x].next;
 365 +       p = q->dep[x].prev;
 366 +       q->dep[p].next = n;
 367 +       q->dep[n].prev = p;
 368 +       d = q->qs[x].qlen;
 369 +       if (q->max_depth < d)
 370 +               q->max_depth = d;
 371 +
 372 +       esfq_link(q, x);
 373 +}
 374 +
 375 +static unsigned int esfq_drop(struct Qdisc *sch)
 376 +{
 377 +       struct esfq_sched_data *q = qdisc_priv(sch);
 378 +       esfq_index d = q->max_depth;
 379 +       struct sk_buff *skb;
 380 +       unsigned int len;
 381 +
 382 +       /* Queue is full! Find the longest slot and
 383 +          drop a packet from it */
 384 +
 385 +       if (d > 1) {
 386 +               esfq_index x = q->dep[d+q->depth].next;
 387 +               skb = q->qs[x].prev;
 388 +               len = skb->len;
 389 +               __skb_unlink(skb, &q->qs[x]);
 390 +               kfree_skb(skb);
 391 +               esfq_dec(q, x);
 392 +               sch->q.qlen--;
 393 +               sch->qstats.drops++;
 394 +               sch->qstats.backlog -= len;
 395 +               return len;
 396 +       }
 397 +
 398 +       if (d == 1) {
 399 +               /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
 400 +               d = q->next[q->tail];
 401 +               q->next[q->tail] = q->next[d];
 402 +               q->allot[q->next[d]] += q->quantum;
 403 +               skb = q->qs[d].prev;
 404 +               len = skb->len;
 405 +               __skb_unlink(skb, &q->qs[d]);
 406 +               kfree_skb(skb);
 407 +               esfq_dec(q, d);
 408 +               sch->q.qlen--;
 409 +               q->ht[q->hash[d]] = q->depth;
 410 +               sch->qstats.drops++;
 411 +               sch->qstats.backlog -= len;
 412 +               return len;
 413 +       }
 414 +
 415 +       return 0;
 416 +}
 417 +
 418 +static void esfq_q_enqueue(struct sk_buff *skb, struct esfq_sched_data *q, unsigned int end)
 419 +{
 420 +       unsigned hash = esfq_hash(q, skb);
 421 +       unsigned depth = q->depth;
 422 +       esfq_index x;
 423 +
 424 +       x = q->ht[hash];
 425 +       if (x == depth) {
 426 +               q->ht[hash] = x = q->dep[depth].next;
 427 +               q->hash[x] = hash;
 428 +       }
 429 +
 430 +       if (end == ESFQ_TAIL)
 431 +               __skb_queue_tail(&q->qs[x], skb);
 432 +       else
 433 +               __skb_queue_head(&q->qs[x], skb);
 434 +
 435 +       esfq_inc(q, x);
 436 +       if (q->qs[x].qlen == 1) {               /* The flow is new */
 437 +               if (q->tail == depth) { /* It is the first flow */
 438 +                       q->tail = x;
 439 +                       q->next[x] = x;
 440 +                       q->allot[x] = q->quantum;
 441 +               } else {
 442 +                       q->next[x] = q->next[q->tail];
 443 +                       q->next[q->tail] = x;
 444 +                       q->tail = x;
 445 +               }
 446 +       }
 447 +}
 448 +
 449 +static int esfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 450 +{
 451 +       struct esfq_sched_data *q = qdisc_priv(sch);
 452 +       esfq_q_enqueue(skb, q, ESFQ_TAIL);
 453 +       sch->qstats.backlog += skb->len;
 454 +       if (++sch->q.qlen < q->limit-1) {
 455 +               sch->bstats.bytes += skb->len;
 456 +               sch->bstats.packets++;
 457 +               return 0;
 458 +       }
 459 +
 460 +       sch->qstats.drops++;
 461 +       esfq_drop(sch);
 462 +       return NET_XMIT_CN;
 463 +}
 464 +
 465 +static struct sk_buff *esfq_peek(struct Qdisc* sch)
 466 +{
 467 +       struct esfq_sched_data *q = qdisc_priv(sch);
 468 +       esfq_index a;
 469 +
 470 +       /* No active slots */
 471 +       if (q->tail == q->depth)
 472 +               return NULL;
 473 +
 474 +       a = q->next[q->tail];
 475 +       return skb_peek(&q->qs[a]);
 476 +}
 477 +
 478 +static struct sk_buff *esfq_q_dequeue(struct esfq_sched_data *q)
 479 +{
 480 +       struct sk_buff *skb;
 481 +       unsigned depth = q->depth;
 482 +       esfq_index a, old_a;
 483 +
 484 +       /* No active slots */
 485 +       if (q->tail == depth)
 486 +               return NULL;
 487 +
 488 +       a = old_a = q->next[q->tail];
 489 +
 490 +       /* Grab packet */
 491 +       skb = __skb_dequeue(&q->qs[a]);
 492 +       esfq_dec(q, a);
 493 +
 494 +       /* Is the slot empty? */
 495 +       if (q->qs[a].qlen == 0) {
 496 +               q->ht[q->hash[a]] = depth;
 497 +               a = q->next[a];
 498 +               if (a == old_a) {
 499 +                       q->tail = depth;
 500 +                       return skb;
 501 +               }
 502 +               q->next[q->tail] = a;
 503 +               q->allot[a] += q->quantum;
 504 +       } else if ((q->allot[a] -= skb->len) <= 0) {
 505 +               q->tail = a;
 506 +               a = q->next[a];
 507 +               q->allot[a] += q->quantum;
 508 +       }
 509 +
 510 +       return skb;
 511 +}
 512 +
 513 +static struct sk_buff *esfq_dequeue(struct Qdisc* sch)
 514 +{
 515 +       struct esfq_sched_data *q = qdisc_priv(sch);
 516 +       struct sk_buff *skb;
 517 +
 518 +       skb = esfq_q_dequeue(q);
 519 +       if (skb == NULL)
 520 +               return NULL;
 521 +       sch->q.qlen--;
 522 +       sch->qstats.backlog -= skb->len;
 523 +       return skb;
 524 +}
 525 +
 526 +static void esfq_q_destroy(struct esfq_sched_data *q)
 527 +{
 528 +       del_timer(&q->perturb_timer);
 529 +       if(q->ht)
 530 +               kfree(q->ht);
 531 +       if(q->dep)
 532 +               kfree(q->dep);
 533 +       if(q->next)
 534 +               kfree(q->next);
 535 +       if(q->allot)
 536 +               kfree(q->allot);
 537 +       if(q->hash)
 538 +               kfree(q->hash);
 539 +       if(q->qs)
 540 +               kfree(q->qs);
 541 +}
 542 +
 543 +static void esfq_destroy(struct Qdisc *sch)
 544 +{
 545 +       struct esfq_sched_data *q = qdisc_priv(sch);
 546 +       esfq_q_destroy(q);
 547 +}
 548 +
 549 +
 550 +static void esfq_reset(struct Qdisc* sch)
 551 +{
 552 +       struct sk_buff *skb;
 553 +
 554 +       while ((skb = esfq_dequeue(sch)) != NULL)
 555 +               kfree_skb(skb);
 556 +}
 557 +
 558 +static void esfq_perturbation(unsigned long arg)
 559 +{
 560 +       struct Qdisc *sch = (struct Qdisc*)arg;
 561 +       struct esfq_sched_data *q = qdisc_priv(sch);
 562 +
 563 +       q->perturbation = net_random()&0x1F;
 564 +
 565 +       if (q->perturb_period) {
 566 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 567 +               add_timer(&q->perturb_timer);
 568 +       }
 569 +}
 570 +
 571 +static unsigned int esfq_check_hash(unsigned int kind)
 572 +{
 573 +       switch (kind) {
 574 +       case TCA_SFQ_HASH_CTORIGDST:
 575 +       case TCA_SFQ_HASH_CTORIGSRC:
 576 +       case TCA_SFQ_HASH_CTREPLDST:
 577 +       case TCA_SFQ_HASH_CTREPLSRC:
 578 +       case TCA_SFQ_HASH_CTNATCHG:
 579 +#ifndef CONFIG_NET_SCH_ESFQ_NFCT
 580 +       {
 581 +               if (net_ratelimit())
 582 +                       printk(KERN_WARNING "ESFQ: Conntrack hash types disabled in kernel config. Falling back to classic.\n");
 583 +               return TCA_SFQ_HASH_CLASSIC;
 584 +       }
 585 +#endif
 586 +       case TCA_SFQ_HASH_CLASSIC:
 587 +       case TCA_SFQ_HASH_DST:
 588 +       case TCA_SFQ_HASH_SRC:
 589 +       case TCA_SFQ_HASH_FWMARK:
 590 +               return kind;
 591 +       default:
 592 +       {
 593 +               if (net_ratelimit())
 594 +                       printk(KERN_WARNING "ESFQ: Unknown hash type. Falling back to classic.\n");
 595 +               return TCA_SFQ_HASH_CLASSIC;
 596 +       }
 597 +       }
 598 +}
 599 +
 600 +static int esfq_q_init(struct esfq_sched_data *q, struct nlattr *opt)
 601 +{
 602 +       struct tc_esfq_qopt *ctl = nla_data(opt);
 603 +       esfq_index p = ~0U/2;
 604 +       int i;
 605 +
 606 +       if (opt && opt->nla_len < nla_attr_size(sizeof(*ctl)))
 607 +               return -EINVAL;
 608 +
 609 +       q->perturbation = 0;
 610 +       q->hash_kind = TCA_SFQ_HASH_CLASSIC;
 611 +       q->max_depth = 0;
 612 +       if (opt == NULL) {
 613 +               q->perturb_period = 0;
 614 +               q->hash_divisor = 1024;
 615 +               q->tail = q->limit = q->depth = 128;
 616 +
 617 +       } else {
 618 +               struct tc_esfq_qopt *ctl = nla_data(opt);
 619 +               if (ctl->quantum)
 620 +                       q->quantum = ctl->quantum;
 621 +               q->perturb_period = ctl->perturb_period*HZ;
 622 +               q->hash_divisor = ctl->divisor ? : 1024;
 623 +               q->tail = q->limit = q->depth = ctl->flows ? : 128;
 624 +
 625 +               if ( q->depth > p - 1 )
 626 +                       return -EINVAL;
 627 +
 628 +               if (ctl->limit)
 629 +                       q->limit = min_t(u32, ctl->limit, q->depth);
 630 +
 631 +               if (ctl->hash_kind) {
 632 +                       q->hash_kind = esfq_check_hash(ctl->hash_kind);
 633 +               }
 634 +       }
 635 +
 636 +       q->ht = kmalloc(q->hash_divisor*sizeof(esfq_index), GFP_KERNEL);
 637 +       if (!q->ht)
 638 +               goto err_case;
 639 +       q->dep = kmalloc((1+q->depth*2)*sizeof(struct esfq_head), GFP_KERNEL);
 640 +       if (!q->dep)
 641 +               goto err_case;
 642 +       q->next = kmalloc(q->depth*sizeof(esfq_index), GFP_KERNEL);
 643 +       if (!q->next)
 644 +               goto err_case;
 645 +       q->allot = kmalloc(q->depth*sizeof(short), GFP_KERNEL);
 646 +       if (!q->allot)
 647 +               goto err_case;
 648 +       q->hash = kmalloc(q->depth*sizeof(unsigned short), GFP_KERNEL);
 649 +       if (!q->hash)
 650 +               goto err_case;
 651 +       q->qs = kmalloc(q->depth*sizeof(struct sk_buff_head), GFP_KERNEL);
 652 +       if (!q->qs)
 653 +               goto err_case;
 654 +
 655 +       for (i=0; i< q->hash_divisor; i++)
 656 +               q->ht[i] = q->depth;
 657 +       for (i=0; i<q->depth; i++) {
 658 +               skb_queue_head_init(&q->qs[i]);
 659 +               q->dep[i+q->depth].next = i+q->depth;
 660 +               q->dep[i+q->depth].prev = i+q->depth;
 661 +       }
 662 +
 663 +       for (i=0; i<q->depth; i++)
 664 +               esfq_link(q, i);
 665 +       return 0;
 666 +err_case:
 667 +       esfq_q_destroy(q);
 668 +       return -ENOBUFS;
 669 +}
 670 +
 671 +static int esfq_init(struct Qdisc *sch, struct nlattr *opt)
 672 +{
 673 +       struct esfq_sched_data *q = qdisc_priv(sch);
 674 +       int err;
 675 +
 676 +       q->quantum = psched_mtu(qdisc_dev(sch)); /* default */
 677 +       if ((err = esfq_q_init(q, opt)))
 678 +               return err;
 679 +
 680 +       init_timer(&q->perturb_timer);
 681 +       q->perturb_timer.data = (unsigned long)sch;
 682 +       q->perturb_timer.function = esfq_perturbation;
 683 +       if (q->perturb_period) {
 684 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 685 +               add_timer(&q->perturb_timer);
 686 +       }
 687 +
 688 +       return 0;
 689 +}
 690 +
 691 +static int esfq_change(struct Qdisc *sch, struct nlattr *opt)
 692 +{
 693 +       struct esfq_sched_data *q = qdisc_priv(sch);
 694 +       struct esfq_sched_data new;
 695 +       struct sk_buff *skb;
 696 +       int err;
 697 +
 698 +       /* set up new queue */
 699 +       memset(&new, 0, sizeof(struct esfq_sched_data));
 700 +       new.quantum = psched_mtu(qdisc_dev(sch)); /* default */
 701 +       if ((err = esfq_q_init(&new, opt)))
 702 +               return err;
 703 +
 704 +       /* copy all packets from the old queue to the new queue */
 705 +       sch_tree_lock(sch);
 706 +       while ((skb = esfq_q_dequeue(q)) != NULL)
 707 +               esfq_q_enqueue(skb, &new, ESFQ_TAIL);
 708 +
 709 +       /* clean up the old queue */
 710 +       esfq_q_destroy(q);
 711 +
 712 +       /* copy elements of the new queue into the old queue */
 713 +       q->perturb_period = new.perturb_period;
 714 +       q->quantum        = new.quantum;
 715 +       q->limit          = new.limit;
 716 +       q->depth          = new.depth;
 717 +       q->hash_divisor   = new.hash_divisor;
 718 +       q->hash_kind      = new.hash_kind;
 719 +       q->tail           = new.tail;
 720 +       q->max_depth      = new.max_depth;
 721 +       q->ht    = new.ht;
 722 +       q->dep   = new.dep;
 723 +       q->next  = new.next;
 724 +       q->allot = new.allot;
 725 +       q->hash  = new.hash;
 726 +       q->qs    = new.qs;
 727 +
 728 +       /* finish up */
 729 +       if (q->perturb_period) {
 730 +               q->perturb_timer.expires = jiffies + q->perturb_period;
 731 +               add_timer(&q->perturb_timer);
 732 +       } else {
 733 +               q->perturbation = 0;
 734 +       }
 735 +       sch_tree_unlock(sch);
 736 +       return 0;
 737 +}
 738 +
 739 +static int esfq_dump(struct Qdisc *sch, struct sk_buff *skb)
 740 +{
 741 +       struct esfq_sched_data *q = qdisc_priv(sch);
 742 +       unsigned char *b = skb_tail_pointer(skb);
 743 +       struct tc_esfq_qopt opt;
 744 +
 745 +       opt.quantum = q->quantum;
 746 +       opt.perturb_period = q->perturb_period/HZ;
 747 +
 748 +       opt.limit = q->limit;
 749 +       opt.divisor = q->hash_divisor;
 750 +       opt.flows = q->depth;
 751 +       opt.hash_kind = q->hash_kind;
 752 +
 753 +       if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
 754 +               goto nla_put_failure;
 755 +
 756 +       return skb->len;
 757 +
 758 +nla_put_failure:
 759 +       nlmsg_trim(skb, b);
 760 +       return -1;
 761 +}
 762 +
 763 +static struct Qdisc_ops esfq_qdisc_ops =
 764 +{
 765 +       .next           =       NULL,
 766 +       .cl_ops         =       NULL,
 767 +       .id             =       "esfq",
 768 +       .priv_size      =       sizeof(struct esfq_sched_data),
 769 +       .enqueue        =       esfq_enqueue,
 770 +       .dequeue        =       esfq_dequeue,
 771 +       .peek           =       esfq_peek,
 772 +       .drop           =       esfq_drop,
 773 +       .init           =       esfq_init,
 774 +       .reset          =       esfq_reset,
 775 +       .destroy        =       esfq_destroy,
 776 +       .change         =       esfq_change,
 777 +       .dump           =       esfq_dump,
 778 +       .owner          =       THIS_MODULE,
 779 +};
 780 +
 781 +static int __init esfq_module_init(void)
 782 +{
 783 +       return register_qdisc(&esfq_qdisc_ops);
 784 +}
 785 +static void __exit esfq_module_exit(void)
 786 +{
 787 +       unregister_qdisc(&esfq_qdisc_ops);
 788 +}
 789 +module_init(esfq_module_init)
 790 +module_exit(esfq_module_exit)
 791 +MODULE_LICENSE("GPL");