root/net/sched/sch_sfb.c

DEFINITIONS

1. sfb_hash
2. prob_plus
3. prob_minus
4. increment_one_qlen
5. increment_qlen
6. decrement_one_qlen
7. decrement_qlen
8. decrement_prob
9. increment_prob
10. sfb_zero_all_buckets
11. sfb_compute_qlen
12. sfb_init_perturbation
13. sfb_swap_slot
14. sfb_rate_limit
15. sfb_classify
16. sfb_enqueue
17. sfb_dequeue
18. sfb_peek
19. sfb_reset
20. sfb_destroy
21. sfb_change
22. sfb_init
23. sfb_dump
24. sfb_dump_stats
25. sfb_dump_class
26. sfb_graft
27. sfb_leaf
28. sfb_find
29. sfb_unbind
30. sfb_change_class
31. sfb_delete
32. sfb_walk
33. sfb_tcf_block
34. sfb_bind
35. sfb_module_init
36. sfb_module_exit

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * net/sched/sch_sfb.c    Stochastic Fair Blue
   4  *
   5  * Copyright (c) 2008-2011 Juliusz Chroboczek <jch@pps.jussieu.fr>
   6  * Copyright (c) 2011 Eric Dumazet <eric.dumazet@gmail.com>
   7  *
   8  * W. Feng, D. Kandlur, D. Saha, K. Shin. Blue:
   9  * A New Class of Active Queue Management Algorithms.
  10  * U. Michigan CSE-TR-387-99, April 1999.
  11  *
  12  * http://www.thefengs.com/wuchang/blue/CSE-TR-387-99.pdf
  13  */
  14 
  15 #include <linux/module.h>
  16 #include <linux/types.h>
  17 #include <linux/kernel.h>
  18 #include <linux/errno.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/random.h>
  21 #include <linux/siphash.h>
  22 #include <net/ip.h>
  23 #include <net/pkt_sched.h>
  24 #include <net/pkt_cls.h>
  25 #include <net/inet_ecn.h>
  26 
  27 /*
  28  * SFB uses two B[l][n] : L x N arrays of bins (L levels, N bins per level)
  29  * This implementation uses L = 8 and N = 16
  30  * This permits us to split one 32bit hash (provided per packet by rxhash or
  31  * external classifier) into 8 subhashes of 4 bits.
  32  */
  33 #define SFB_BUCKET_SHIFT 4
  34 #define SFB_NUMBUCKETS  (1 << SFB_BUCKET_SHIFT) /* N bins per Level */
  35 #define SFB_BUCKET_MASK (SFB_NUMBUCKETS - 1)
  36 #define SFB_LEVELS      (32 / SFB_BUCKET_SHIFT) /* L */
  37 
  38 /* SFB algo uses a virtual queue, named "bin" */
  39 struct sfb_bucket {
  40         u16             qlen; /* length of virtual queue */
  41         u16             p_mark; /* marking probability */
  42 };
  43 
  44 /* We use a double buffering right before hash change
  45  * (Section 4.4 of SFB reference : moving hash functions)
  46  */
  47 struct sfb_bins {
  48         siphash_key_t     perturbation; /* siphash key */
  49         struct sfb_bucket bins[SFB_LEVELS][SFB_NUMBUCKETS];
  50 };
  51 
  52 struct sfb_sched_data {
  53         struct Qdisc    *qdisc;
  54         struct tcf_proto __rcu *filter_list;
  55         struct tcf_block *block;
  56         unsigned long   rehash_interval;
  57         unsigned long   warmup_time;    /* double buffering warmup time in jiffies */
  58         u32             max;
  59         u32             bin_size;       /* maximum queue length per bin */
  60         u32             increment;      /* d1 */
  61         u32             decrement;      /* d2 */
  62         u32             limit;          /* HARD maximal queue length */
  63         u32             penalty_rate;
  64         u32             penalty_burst;
  65         u32             tokens_avail;
  66         unsigned long   rehash_time;
  67         unsigned long   token_time;
  68 
  69         u8              slot;           /* current active bins (0 or 1) */
  70         bool            double_buffering;
  71         struct sfb_bins bins[2];
  72 
  73         struct {
  74                 u32     earlydrop;
  75                 u32     penaltydrop;
  76                 u32     bucketdrop;
  77                 u32     queuedrop;
  78                 u32     childdrop;      /* drops in child qdisc */
  79                 u32     marked;         /* ECN mark */
  80         } stats;
  81 };
  82 
  83 /*
  84  * Each queued skb might be hashed on one or two bins
  85  * We store in skb_cb the two hash values.
  86  * (A zero value means double buffering was not used)
  87  */
  88 struct sfb_skb_cb {
  89         u32 hashes[2];
  90 };
  91 
  92 static inline struct sfb_skb_cb *sfb_skb_cb(const struct sk_buff *skb)
  93 {
  94         qdisc_cb_private_validate(skb, sizeof(struct sfb_skb_cb));
  95         return (struct sfb_skb_cb *)qdisc_skb_cb(skb)->data;
  96 }
  97 
  98 /*
  99  * If using 'internal' SFB flow classifier, hash comes from skb rxhash
 100  * If using external classifier, hash comes from the classid.
 101  */
 102 static u32 sfb_hash(const struct sk_buff *skb, u32 slot)
 103 {
 104         return sfb_skb_cb(skb)->hashes[slot];
 105 }
 106 
 107 /* Probabilities are coded as Q0.16 fixed-point values,
 108  * with 0xFFFF representing 65535/65536 (almost 1.0)
 109  * Addition and subtraction are saturating in [0, 65535]
 110  */
 111 static u32 prob_plus(u32 p1, u32 p2)
 112 {
 113         u32 res = p1 + p2;
 114 
 115         return min_t(u32, res, SFB_MAX_PROB);
 116 }
 117 
 118 static u32 prob_minus(u32 p1, u32 p2)
 119 {
 120         return p1 > p2 ? p1 - p2 : 0;
 121 }
 122 
 123 static void increment_one_qlen(u32 sfbhash, u32 slot, struct sfb_sched_data *q)
 124 {
 125         int i;
 126         struct sfb_bucket *b = &q->bins[slot].bins[0][0];
 127 
 128         for (i = 0; i < SFB_LEVELS; i++) {
 129                 u32 hash = sfbhash & SFB_BUCKET_MASK;
 130 
 131                 sfbhash >>= SFB_BUCKET_SHIFT;
 132                 if (b[hash].qlen < 0xFFFF)
 133                         b[hash].qlen++;
 134                 b += SFB_NUMBUCKETS; /* next level */
 135         }
 136 }
 137 
 138 static void increment_qlen(const struct sk_buff *skb, struct sfb_sched_data *q)
 139 {
 140         u32 sfbhash;
 141 
 142         sfbhash = sfb_hash(skb, 0);
 143         if (sfbhash)
 144                 increment_one_qlen(sfbhash, 0, q);
 145 
 146         sfbhash = sfb_hash(skb, 1);
 147         if (sfbhash)
 148                 increment_one_qlen(sfbhash, 1, q);
 149 }
 150 
 151 static void decrement_one_qlen(u32 sfbhash, u32 slot,
 152                                struct sfb_sched_data *q)
 153 {
 154         int i;
 155         struct sfb_bucket *b = &q->bins[slot].bins[0][0];
 156 
 157         for (i = 0; i < SFB_LEVELS; i++) {
 158                 u32 hash = sfbhash & SFB_BUCKET_MASK;
 159 
 160                 sfbhash >>= SFB_BUCKET_SHIFT;
 161                 if (b[hash].qlen > 0)
 162                         b[hash].qlen--;
 163                 b += SFB_NUMBUCKETS; /* next level */
 164         }
 165 }
 166 
 167 static void decrement_qlen(const struct sk_buff *skb, struct sfb_sched_data *q)
 168 {
 169         u32 sfbhash;
 170 
 171         sfbhash = sfb_hash(skb, 0);
 172         if (sfbhash)
 173                 decrement_one_qlen(sfbhash, 0, q);
 174 
 175         sfbhash = sfb_hash(skb, 1);
 176         if (sfbhash)
 177                 decrement_one_qlen(sfbhash, 1, q);
 178 }
 179 
 180 static void decrement_prob(struct sfb_bucket *b, struct sfb_sched_data *q)
 181 {
 182         b->p_mark = prob_minus(b->p_mark, q->decrement);
 183 }
 184 
 185 static void increment_prob(struct sfb_bucket *b, struct sfb_sched_data *q)
 186 {
 187         b->p_mark = prob_plus(b->p_mark, q->increment);
 188 }
 189 
 190 static void sfb_zero_all_buckets(struct sfb_sched_data *q)
 191 {
 192         memset(&q->bins, 0, sizeof(q->bins));
 193 }
 194 
 195 /*
 196  * compute max qlen, max p_mark, and avg p_mark
 197  */
 198 static u32 sfb_compute_qlen(u32 *prob_r, u32 *avgpm_r, const struct sfb_sched_data *q)
 199 {
 200         int i;
 201         u32 qlen = 0, prob = 0, totalpm = 0;
 202         const struct sfb_bucket *b = &q->bins[q->slot].bins[0][0];
 203 
 204         for (i = 0; i < SFB_LEVELS * SFB_NUMBUCKETS; i++) {
 205                 if (qlen < b->qlen)
 206                         qlen = b->qlen;
 207                 totalpm += b->p_mark;
 208                 if (prob < b->p_mark)
 209                         prob = b->p_mark;
 210                 b++;
 211         }
 212         *prob_r = prob;
 213         *avgpm_r = totalpm / (SFB_LEVELS * SFB_NUMBUCKETS);
 214         return qlen;
 215 }
 216 
 217 
 218 static void sfb_init_perturbation(u32 slot, struct sfb_sched_data *q)
 219 {
 220         get_random_bytes(&q->bins[slot].perturbation,
 221                          sizeof(q->bins[slot].perturbation));
 222 }
 223 
 224 static void sfb_swap_slot(struct sfb_sched_data *q)
 225 {
 226         sfb_init_perturbation(q->slot, q);
 227         q->slot ^= 1;
 228         q->double_buffering = false;
 229 }
 230 
 231 /* Non elastic flows are allowed to use part of the bandwidth, expressed
 232  * in "penalty_rate" packets per second, with "penalty_burst" burst
 233  */
 234 static bool sfb_rate_limit(struct sk_buff *skb, struct sfb_sched_data *q)
 235 {
 236         if (q->penalty_rate == 0 || q->penalty_burst == 0)
 237                 return true;
 238 
 239         if (q->tokens_avail < 1) {
 240                 unsigned long age = min(10UL * HZ, jiffies - q->token_time);
 241 
 242                 q->tokens_avail = (age * q->penalty_rate) / HZ;
 243                 if (q->tokens_avail > q->penalty_burst)
 244                         q->tokens_avail = q->penalty_burst;
 245                 q->token_time = jiffies;
 246                 if (q->tokens_avail < 1)
 247                         return true;
 248         }
 249 
 250         q->tokens_avail--;
 251         return false;
 252 }
 253 
 254 static bool sfb_classify(struct sk_buff *skb, struct tcf_proto *fl,
 255                          int *qerr, u32 *salt)
 256 {
 257         struct tcf_result res;
 258         int result;
 259 
 260         result = tcf_classify(skb, fl, &res, false);
 261         if (result >= 0) {
 262 #ifdef CONFIG_NET_CLS_ACT
 263                 switch (result) {
 264                 case TC_ACT_STOLEN:
 265                 case TC_ACT_QUEUED:
 266                 case TC_ACT_TRAP:
 267                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 268                         /* fall through */
 269                 case TC_ACT_SHOT:
 270                         return false;
 271                 }
 272 #endif
 273                 *salt = TC_H_MIN(res.classid);
 274                 return true;
 275         }
 276         return false;
 277 }
 278 
 279 static int sfb_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 280                        struct sk_buff **to_free)
 281 {
 282 
 283         struct sfb_sched_data *q = qdisc_priv(sch);
 284         struct Qdisc *child = q->qdisc;
 285         struct tcf_proto *fl;
 286         int i;
 287         u32 p_min = ~0;
 288         u32 minqlen = ~0;
 289         u32 r, sfbhash;
 290         u32 slot = q->slot;
 291         int ret = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 292 
 293         if (unlikely(sch->q.qlen >= q->limit)) {
 294                 qdisc_qstats_overlimit(sch);
 295                 q->stats.queuedrop++;
 296                 goto drop;
 297         }
 298 
 299         if (q->rehash_interval > 0) {
 300                 unsigned long limit = q->rehash_time + q->rehash_interval;
 301 
 302                 if (unlikely(time_after(jiffies, limit))) {
 303                         sfb_swap_slot(q);
 304                         q->rehash_time = jiffies;
 305                 } else if (unlikely(!q->double_buffering && q->warmup_time > 0 &&
 306                                     time_after(jiffies, limit - q->warmup_time))) {
 307                         q->double_buffering = true;
 308                 }
 309         }
 310 
 311         fl = rcu_dereference_bh(q->filter_list);
 312         if (fl) {
 313                 u32 salt;
 314 
 315                 /* If using external classifiers, get result and record it. */
 316                 if (!sfb_classify(skb, fl, &ret, &salt))
 317                         goto other_drop;
 318                 sfbhash = siphash_1u32(salt, &q->bins[slot].perturbation);
 319         } else {
 320                 sfbhash = skb_get_hash_perturb(skb, &q->bins[slot].perturbation);
 321         }
 322 
 323 
 324         if (!sfbhash)
 325                 sfbhash = 1;
 326         sfb_skb_cb(skb)->hashes[slot] = sfbhash;
 327 
 328         for (i = 0; i < SFB_LEVELS; i++) {
 329                 u32 hash = sfbhash & SFB_BUCKET_MASK;
 330                 struct sfb_bucket *b = &q->bins[slot].bins[i][hash];
 331 
 332                 sfbhash >>= SFB_BUCKET_SHIFT;
 333                 if (b->qlen == 0)
 334                         decrement_prob(b, q);
 335                 else if (b->qlen >= q->bin_size)
 336                         increment_prob(b, q);
 337                 if (minqlen > b->qlen)
 338                         minqlen = b->qlen;
 339                 if (p_min > b->p_mark)
 340                         p_min = b->p_mark;
 341         }
 342 
 343         slot ^= 1;
 344         sfb_skb_cb(skb)->hashes[slot] = 0;
 345 
 346         if (unlikely(minqlen >= q->max)) {
 347                 qdisc_qstats_overlimit(sch);
 348                 q->stats.bucketdrop++;
 349                 goto drop;
 350         }
 351 
 352         if (unlikely(p_min >= SFB_MAX_PROB)) {
 353                 /* Inelastic flow */
 354                 if (q->double_buffering) {
 355                         sfbhash = skb_get_hash_perturb(skb,
 356                             &q->bins[slot].perturbation);
 357                         if (!sfbhash)
 358                                 sfbhash = 1;
 359                         sfb_skb_cb(skb)->hashes[slot] = sfbhash;
 360 
 361                         for (i = 0; i < SFB_LEVELS; i++) {
 362                                 u32 hash = sfbhash & SFB_BUCKET_MASK;
 363                                 struct sfb_bucket *b = &q->bins[slot].bins[i][hash];
 364 
 365                                 sfbhash >>= SFB_BUCKET_SHIFT;
 366                                 if (b->qlen == 0)
 367                                         decrement_prob(b, q);
 368                                 else if (b->qlen >= q->bin_size)
 369                                         increment_prob(b, q);
 370                         }
 371                 }
 372                 if (sfb_rate_limit(skb, q)) {
 373                         qdisc_qstats_overlimit(sch);
 374                         q->stats.penaltydrop++;
 375                         goto drop;
 376                 }
 377                 goto enqueue;
 378         }
 379 
 380         r = prandom_u32() & SFB_MAX_PROB;
 381 
 382         if (unlikely(r < p_min)) {
 383                 if (unlikely(p_min > SFB_MAX_PROB / 2)) {
 384                         /* If we're marking that many packets, then either
 385                          * this flow is unresponsive, or we're badly congested.
 386                          * In either case, we want to start dropping packets.
 387                          */
 388                         if (r < (p_min - SFB_MAX_PROB / 2) * 2) {
 389                                 q->stats.earlydrop++;
 390                                 goto drop;
 391                         }
 392                 }
 393                 if (INET_ECN_set_ce(skb)) {
 394                         q->stats.marked++;
 395                 } else {
 396                         q->stats.earlydrop++;
 397                         goto drop;
 398                 }
 399         }
 400 
 401 enqueue:
 402         ret = qdisc_enqueue(skb, child, to_free);
 403         if (likely(ret == NET_XMIT_SUCCESS)) {
 404                 qdisc_qstats_backlog_inc(sch, skb);
 405                 sch->q.qlen++;
 406                 increment_qlen(skb, q);
 407         } else if (net_xmit_drop_count(ret)) {
 408                 q->stats.childdrop++;
 409                 qdisc_qstats_drop(sch);
 410         }
 411         return ret;
 412 
 413 drop:
 414         qdisc_drop(skb, sch, to_free);
 415         return NET_XMIT_CN;
 416 other_drop:
 417         if (ret & __NET_XMIT_BYPASS)
 418                 qdisc_qstats_drop(sch);
 419         kfree_skb(skb);
 420         return ret;
 421 }
 422 
 423 static struct sk_buff *sfb_dequeue(struct Qdisc *sch)
 424 {
 425         struct sfb_sched_data *q = qdisc_priv(sch);
 426         struct Qdisc *child = q->qdisc;
 427         struct sk_buff *skb;
 428 
 429         skb = child->dequeue(q->qdisc);
 430 
 431         if (skb) {
 432                 qdisc_bstats_update(sch, skb);
 433                 qdisc_qstats_backlog_dec(sch, skb);
 434                 sch->q.qlen--;
 435                 decrement_qlen(skb, q);
 436         }
 437 
 438         return skb;
 439 }
 440 
 441 static struct sk_buff *sfb_peek(struct Qdisc *sch)
 442 {
 443         struct sfb_sched_data *q = qdisc_priv(sch);
 444         struct Qdisc *child = q->qdisc;
 445 
 446         return child->ops->peek(child);
 447 }
 448 
 449 /* No sfb_drop -- impossible since the child doesn't return the dropped skb. */
 450 
 451 static void sfb_reset(struct Qdisc *sch)
 452 {
 453         struct sfb_sched_data *q = qdisc_priv(sch);
 454 
 455         qdisc_reset(q->qdisc);
 456         sch->qstats.backlog = 0;
 457         sch->q.qlen = 0;
 458         q->slot = 0;
 459         q->double_buffering = false;
 460         sfb_zero_all_buckets(q);
 461         sfb_init_perturbation(0, q);
 462 }
 463 
 464 static void sfb_destroy(struct Qdisc *sch)
 465 {
 466         struct sfb_sched_data *q = qdisc_priv(sch);
 467 
 468         tcf_block_put(q->block);
 469         qdisc_put(q->qdisc);
 470 }
 471 
 472 static const struct nla_policy sfb_policy[TCA_SFB_MAX + 1] = {
 473         [TCA_SFB_PARMS] = { .len = sizeof(struct tc_sfb_qopt) },
 474 };
 475 
 476 static const struct tc_sfb_qopt sfb_default_ops = {
 477         .rehash_interval = 600 * MSEC_PER_SEC,
 478         .warmup_time = 60 * MSEC_PER_SEC,
 479         .limit = 0,
 480         .max = 25,
 481         .bin_size = 20,
 482         .increment = (SFB_MAX_PROB + 500) / 1000, /* 0.1 % */
 483         .decrement = (SFB_MAX_PROB + 3000) / 6000,
 484         .penalty_rate = 10,
 485         .penalty_burst = 20,
 486 };
 487 
 488 static int sfb_change(struct Qdisc *sch, struct nlattr *opt,
 489                       struct netlink_ext_ack *extack)
 490 {
 491         struct sfb_sched_data *q = qdisc_priv(sch);
 492         struct Qdisc *child, *old;
 493         struct nlattr *tb[TCA_SFB_MAX + 1];
 494         const struct tc_sfb_qopt *ctl = &sfb_default_ops;
 495         u32 limit;
 496         int err;
 497 
 498         if (opt) {
 499                 err = nla_parse_nested_deprecated(tb, TCA_SFB_MAX, opt,
 500                                                   sfb_policy, NULL);
 501                 if (err < 0)
 502                         return -EINVAL;
 503 
 504                 if (tb[TCA_SFB_PARMS] == NULL)
 505                         return -EINVAL;
 506 
 507                 ctl = nla_data(tb[TCA_SFB_PARMS]);
 508         }
 509 
 510         limit = ctl->limit;
 511         if (limit == 0)
 512                 limit = qdisc_dev(sch)->tx_queue_len;
 513 
 514         child = fifo_create_dflt(sch, &pfifo_qdisc_ops, limit, extack);
 515         if (IS_ERR(child))
 516                 return PTR_ERR(child);
 517 
 518         if (child != &noop_qdisc)
 519                 qdisc_hash_add(child, true);
 520         sch_tree_lock(sch);
 521 
 522         qdisc_purge_queue(q->qdisc);
 523         old = q->qdisc;
 524         q->qdisc = child;
 525 
 526         q->rehash_interval = msecs_to_jiffies(ctl->rehash_interval);
 527         q->warmup_time = msecs_to_jiffies(ctl->warmup_time);
 528         q->rehash_time = jiffies;
 529         q->limit = limit;
 530         q->increment = ctl->increment;
 531         q->decrement = ctl->decrement;
 532         q->max = ctl->max;
 533         q->bin_size = ctl->bin_size;
 534         q->penalty_rate = ctl->penalty_rate;
 535         q->penalty_burst = ctl->penalty_burst;
 536         q->tokens_avail = ctl->penalty_burst;
 537         q->token_time = jiffies;
 538 
 539         q->slot = 0;
 540         q->double_buffering = false;
 541         sfb_zero_all_buckets(q);
 542         sfb_init_perturbation(0, q);
 543         sfb_init_perturbation(1, q);
 544 
 545         sch_tree_unlock(sch);
 546         qdisc_put(old);
 547 
 548         return 0;
 549 }
 550 
 551 static int sfb_init(struct Qdisc *sch, struct nlattr *opt,
 552                     struct netlink_ext_ack *extack)
 553 {
 554         struct sfb_sched_data *q = qdisc_priv(sch);
 555         int err;
 556 
 557         err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
 558         if (err)
 559                 return err;
 560 
 561         q->qdisc = &noop_qdisc;
 562         return sfb_change(sch, opt, extack);
 563 }
 564 
 565 static int sfb_dump(struct Qdisc *sch, struct sk_buff *skb)
 566 {
 567         struct sfb_sched_data *q = qdisc_priv(sch);
 568         struct nlattr *opts;
 569         struct tc_sfb_qopt opt = {
 570                 .rehash_interval = jiffies_to_msecs(q->rehash_interval),
 571                 .warmup_time = jiffies_to_msecs(q->warmup_time),
 572                 .limit = q->limit,
 573                 .max = q->max,
 574                 .bin_size = q->bin_size,
 575                 .increment = q->increment,
 576                 .decrement = q->decrement,
 577                 .penalty_rate = q->penalty_rate,
 578                 .penalty_burst = q->penalty_burst,
 579         };
 580 
 581         sch->qstats.backlog = q->qdisc->qstats.backlog;
 582         opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
 583         if (opts == NULL)
 584                 goto nla_put_failure;
 585         if (nla_put(skb, TCA_SFB_PARMS, sizeof(opt), &opt))
 586                 goto nla_put_failure;
 587         return nla_nest_end(skb, opts);
 588 
 589 nla_put_failure:
 590         nla_nest_cancel(skb, opts);
 591         return -EMSGSIZE;
 592 }
 593 
 594 static int sfb_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 595 {
 596         struct sfb_sched_data *q = qdisc_priv(sch);
 597         struct tc_sfb_xstats st = {
 598                 .earlydrop = q->stats.earlydrop,
 599                 .penaltydrop = q->stats.penaltydrop,
 600                 .bucketdrop = q->stats.bucketdrop,
 601                 .queuedrop = q->stats.queuedrop,
 602                 .childdrop = q->stats.childdrop,
 603                 .marked = q->stats.marked,
 604         };
 605 
 606         st.maxqlen = sfb_compute_qlen(&st.maxprob, &st.avgprob, q);
 607 
 608         return gnet_stats_copy_app(d, &st, sizeof(st));
 609 }
 610 
 611 static int sfb_dump_class(struct Qdisc *sch, unsigned long cl,
 612                           struct sk_buff *skb, struct tcmsg *tcm)
 613 {
 614         return -ENOSYS;
 615 }
 616 
 617 static int sfb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
 618                      struct Qdisc **old, struct netlink_ext_ack *extack)
 619 {
 620         struct sfb_sched_data *q = qdisc_priv(sch);
 621 
 622         if (new == NULL)
 623                 new = &noop_qdisc;
 624 
 625         *old = qdisc_replace(sch, new, &q->qdisc);
 626         return 0;
 627 }
 628 
 629 static struct Qdisc *sfb_leaf(struct Qdisc *sch, unsigned long arg)
 630 {
 631         struct sfb_sched_data *q = qdisc_priv(sch);
 632 
 633         return q->qdisc;
 634 }
 635 
 636 static unsigned long sfb_find(struct Qdisc *sch, u32 classid)
 637 {
 638         return 1;
 639 }
 640 
 641 static void sfb_unbind(struct Qdisc *sch, unsigned long arg)
 642 {
 643 }
 644 
 645 static int sfb_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
 646                             struct nlattr **tca, unsigned long *arg,
 647                             struct netlink_ext_ack *extack)
 648 {
 649         return -ENOSYS;
 650 }
 651 
 652 static int sfb_delete(struct Qdisc *sch, unsigned long cl)
 653 {
 654         return -ENOSYS;
 655 }
 656 
 657 static void sfb_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 658 {
 659         if (!walker->stop) {
 660                 if (walker->count >= walker->skip)
 661                         if (walker->fn(sch, 1, walker) < 0) {
 662                                 walker->stop = 1;
 663                                 return;
 664                         }
 665                 walker->count++;
 666         }
 667 }
 668 
 669 static struct tcf_block *sfb_tcf_block(struct Qdisc *sch, unsigned long cl,
 670                                        struct netlink_ext_ack *extack)
 671 {
 672         struct sfb_sched_data *q = qdisc_priv(sch);
 673 
 674         if (cl)
 675                 return NULL;
 676         return q->block;
 677 }
 678 
 679 static unsigned long sfb_bind(struct Qdisc *sch, unsigned long parent,
 680                               u32 classid)
 681 {
 682         return 0;
 683 }
 684 
 685 
 686 static const struct Qdisc_class_ops sfb_class_ops = {
 687         .graft          =       sfb_graft,
 688         .leaf           =       sfb_leaf,
 689         .find           =       sfb_find,
 690         .change         =       sfb_change_class,
 691         .delete         =       sfb_delete,
 692         .walk           =       sfb_walk,
 693         .tcf_block      =       sfb_tcf_block,
 694         .bind_tcf       =       sfb_bind,
 695         .unbind_tcf     =       sfb_unbind,
 696         .dump           =       sfb_dump_class,
 697 };
 698 
 699 static struct Qdisc_ops sfb_qdisc_ops __read_mostly = {
 700         .id             =       "sfb",
 701         .priv_size      =       sizeof(struct sfb_sched_data),
 702         .cl_ops         =       &sfb_class_ops,
 703         .enqueue        =       sfb_enqueue,
 704         .dequeue        =       sfb_dequeue,
 705         .peek           =       sfb_peek,
 706         .init           =       sfb_init,
 707         .reset          =       sfb_reset,
 708         .destroy        =       sfb_destroy,
 709         .change         =       sfb_change,
 710         .dump           =       sfb_dump,
 711         .dump_stats     =       sfb_dump_stats,
 712         .owner          =       THIS_MODULE,
 713 };
 714 
 715 static int __init sfb_module_init(void)
 716 {
 717         return register_qdisc(&sfb_qdisc_ops);
 718 }
 719 
 720 static void __exit sfb_module_exit(void)
 721 {
 722         unregister_qdisc(&sfb_qdisc_ops);
 723 }
 724 
 725 module_init(sfb_module_init)
 726 module_exit(sfb_module_exit)
 727 
 728 MODULE_DESCRIPTION("Stochastic Fair Blue queue discipline");
 729 MODULE_AUTHOR("Juliusz Chroboczek");
 730 MODULE_AUTHOR("Eric Dumazet");
 731 MODULE_LICENSE("GPL");