net/sched/sch_tbf.c

   1 /*
   2  * net/sched/sch_tbf.c  Token Bucket Filter queue.
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  *              Dmitry Torokhov <dtor@mail.ru> - allow attaching inner qdiscs -
  11  *                                               original idea by Martin Devera
  12  *
  13  */
  14
  15 #include <linux/module.h>
  16 #include <linux/types.h>
  17 #include <linux/kernel.h>
  18 #include <linux/string.h>
  19 #include <linux/errno.h>
  20 #include <linux/skbuff.h>
  21 #include <net/netlink.h>
  22 #include <net/pkt_sched.h>
  23
  24
  25 /*      Simple Token Bucket Filter.
  26         =======================================
  27
  28         SOURCE.
  29         -------
  30
  31         None.
  32
  33         Description.
  34         ------------
  35
  36         A data flow obeys TBF with rate R and depth B, if for any
  37         time interval t_i...t_f the number of transmitted bits
  38         does not exceed B + R*(t_f-t_i).
  39
  40         Packetized version of this definition:
  41         The sequence of packets of sizes s_i served at moments t_i
  42         obeys TBF, if for any i<=k:
  43
  44         s_i+....+s_k <= B + R*(t_k - t_i)
  45
  46         Algorithm.
  47         ----------
  48
  49         Let N(t_i) be B/R initially and N(t) grow continuously with time as:
  50
  51         N(t+delta) = min{B/R, N(t) + delta}
  52
  53         If the first packet in queue has length S, it may be
  54         transmitted only at the time t_* when S/R <= N(t_*),
  55         and in this case N(t) jumps:
  56
  57         N(t_* + 0) = N(t_* - 0) - S/R.
  58
  59
  60
  61         Actually, QoS requires two TBF to be applied to a data stream.
  62         One of them controls steady state burst size, another
  63         one with rate P (peak rate) and depth M (equal to link MTU)
  64         limits bursts at a smaller time scale.
  65
  66         It is easy to see that P>R, and B>M. If P is infinity, this double
  67         TBF is equivalent to a single one.
  68
  69         When TBF works in reshaping mode, latency is estimated as:
  70
  71         lat = max ((L-B)/R, (L-M)/P)
  72
  73
  74         NOTES.
  75         ------
  76
  77         If TBF throttles, it starts a watchdog timer, which will wake it up
  78         when it is ready to transmit.
  79         Note that the minimal timer resolution is 1/HZ.
  80         If no new packets arrive during this period,
  81         or if the device is not awaken by EOI for some previous packet,
  82         TBF can stop its activity for 1/HZ.
  83
  84
  85         This means, that with depth B, the maximal rate is
  86
  87         R_crit = B*HZ
  88
  89         F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes.
  90
  91         Note that the peak rate TBF is much more tough: with MTU 1500
  92         P_crit = 150Kbytes/sec. So, if you need greater peak
  93         rates, use alpha with HZ=1000 :-)
  94
  95         With classful TBF, limit is just kept for backwards compatibility.
  96         It is passed to the default bfifo qdisc - if the inner qdisc is
  97         changed the limit is not effective anymore.
  98 */
  99
 100 struct tbf_sched_data
 101 {
 102 /* Parameters */
 103         u32             limit;          /* Maximal length of backlog: bytes */
 104         u32             buffer;         /* Token bucket depth/rate: MUST BE >= MTU/B */
 105         u32             mtu;
 106         u32             max_size;
 107         struct qdisc_rate_table *R_tab;
 108         struct qdisc_rate_table *P_tab;
 109
 110 /* Variables */
 111         long    tokens;                 /* Current number of B tokens */
 112         long    ptokens;                /* Current number of P tokens */
 113         psched_time_t   t_c;            /* Time check-point */
 114         struct Qdisc    *qdisc;         /* Inner qdisc, default - bfifo queue */
 115         struct qdisc_watchdog watchdog; /* Watchdog timer */
 116 };
 117
 118 #define L2T(q,L)   qdisc_l2t((q)->R_tab,L)
 119 #define L2T_P(q,L) qdisc_l2t((q)->P_tab,L)
 120
 121 static int tbf_enqueue(struct sk_buff *skb, struct Qdisc* sch)
 122 {
 123         struct tbf_sched_data *q = qdisc_priv(sch);
 124         int ret;
 125
 126         if (qdisc_pkt_len(skb) > q->max_size) {
 127                 sch->qstats.drops++;
 128 #ifdef CONFIG_NET_CLS_ACT
 129                 if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
 130 #endif
 131                         kfree_skb(skb);
 132
 133                 return NET_XMIT_DROP;
 134         }
 135
 136         ret = qdisc_enqueue(skb, q->qdisc);
 137         if (ret != 0) {
 138                 if (net_xmit_drop_count(ret))
 139                         sch->qstats.drops++;
 140                 return ret;
 141         }
 142
 143         sch->q.qlen++;
 144         sch->bstats.bytes += qdisc_pkt_len(skb);
 145         sch->bstats.packets++;
 146         return 0;
 147 }
 148
 149 static int tbf_requeue(struct sk_buff *skb, struct Qdisc* sch)
 150 {
 151         struct tbf_sched_data *q = qdisc_priv(sch);
 152         int ret;
 153
 154         if ((ret = q->qdisc->ops->requeue(skb, q->qdisc)) == 0) {
 155                 sch->q.qlen++;
 156                 sch->qstats.requeues++;
 157         }
 158
 159         return ret;
 160 }
 161
 162 static unsigned int tbf_drop(struct Qdisc* sch)
 163 {
 164         struct tbf_sched_data *q = qdisc_priv(sch);
 165         unsigned int len = 0;
 166
 167         if (q->qdisc->ops->drop && (len = q->qdisc->ops->drop(q->qdisc)) != 0) {
 168                 sch->q.qlen--;
 169                 sch->qstats.drops++;
 170         }
 171         return len;
 172 }
 173
 174 static struct sk_buff *tbf_dequeue(struct Qdisc* sch)
 175 {
 176         struct tbf_sched_data *q = qdisc_priv(sch);
 177         struct sk_buff *skb;
 178
 179         skb = q->qdisc->dequeue(q->qdisc);
 180
 181         if (skb) {
 182                 psched_time_t now;
 183                 long toks;
 184                 long ptoks = 0;
 185                 unsigned int len = qdisc_pkt_len(skb);
 186
 187                 now = psched_get_time();
 188                 toks = psched_tdiff_bounded(now, q->t_c, q->buffer);
 189
 190                 if (q->P_tab) {
 191                         ptoks = toks + q->ptokens;
 192                         if (ptoks > (long)q->mtu)
 193                                 ptoks = q->mtu;
 194                         ptoks -= L2T_P(q, len);
 195                 }
 196                 toks += q->tokens;
 197                 if (toks > (long)q->buffer)
 198                         toks = q->buffer;
 199                 toks -= L2T(q, len);
 200
 201                 if ((toks|ptoks) >= 0) {
 202                         q->t_c = now;
 203                         q->tokens = toks;
 204                         q->ptokens = ptoks;
 205                         sch->q.qlen--;
 206                         sch->flags &= ~TCQ_F_THROTTLED;
 207                         return skb;
 208                 }
 209
 210                 qdisc_watchdog_schedule(&q->watchdog,
 211                                         now + max_t(long, -toks, -ptoks));
 212
 213                 /* Maybe we have a shorter packet in the queue,
 214                    which can be sent now. It sounds cool,
 215                    but, however, this is wrong in principle.
 216                    We MUST NOT reorder packets under these circumstances.
 217
 218                    Really, if we split the flow into independent
 219                    subflows, it would be a very good solution.
 220                    This is the main idea of all FQ algorithms
 221                    (cf. CSZ, HPFQ, HFSC)
 222                  */
 223
 224                 if (q->qdisc->ops->requeue(skb, q->qdisc) != NET_XMIT_SUCCESS) {
 225                         /* When requeue fails skb is dropped */
 226                         qdisc_tree_decrease_qlen(q->qdisc, 1);
 227                         sch->qstats.drops++;
 228                 }
 229
 230                 sch->qstats.overlimits++;
 231         }
 232         return NULL;
 233 }
 234
 235 static void tbf_reset(struct Qdisc* sch)
 236 {
 237         struct tbf_sched_data *q = qdisc_priv(sch);
 238
 239         qdisc_reset(q->qdisc);
 240         sch->q.qlen = 0;
 241         q->t_c = psched_get_time();
 242         q->tokens = q->buffer;
 243         q->ptokens = q->mtu;
 244         qdisc_watchdog_cancel(&q->watchdog);
 245 }
 246
 247 static const struct nla_policy tbf_policy[TCA_TBF_MAX + 1] = {
 248         [TCA_TBF_PARMS] = { .len = sizeof(struct tc_tbf_qopt) },
 249         [TCA_TBF_RTAB]  = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
 250         [TCA_TBF_PTAB]  = { .type = NLA_BINARY, .len = TC_RTAB_SIZE },
 251 };
 252
 253 static int tbf_change(struct Qdisc* sch, struct nlattr *opt)
 254 {
 255         int err;
 256         struct tbf_sched_data *q = qdisc_priv(sch);
 257         struct nlattr *tb[TCA_TBF_PTAB + 1];
 258         struct tc_tbf_qopt *qopt;
 259         struct qdisc_rate_table *rtab = NULL;
 260         struct qdisc_rate_table *ptab = NULL;
 261         struct Qdisc *child = NULL;
 262         int max_size,n;
 263
 264         err = nla_parse_nested(tb, TCA_TBF_PTAB, opt, tbf_policy);
 265         if (err < 0)
 266                 return err;
 267
 268         err = -EINVAL;
 269         if (tb[TCA_TBF_PARMS] == NULL)
 270                 goto done;
 271
 272         qopt = nla_data(tb[TCA_TBF_PARMS]);
 273         rtab = qdisc_get_rtab(&qopt->rate, tb[TCA_TBF_RTAB]);
 274         if (rtab == NULL)
 275                 goto done;
 276
 277         if (qopt->peakrate.rate) {
 278                 if (qopt->peakrate.rate > qopt->rate.rate)
 279                         ptab = qdisc_get_rtab(&qopt->peakrate, tb[TCA_TBF_PTAB]);
 280                 if (ptab == NULL)
 281                         goto done;
 282         }
 283
 284         for (n = 0; n < 256; n++)
 285                 if (rtab->data[n] > qopt->buffer) break;
 286         max_size = (n << qopt->rate.cell_log)-1;
 287         if (ptab) {
 288                 int size;
 289
 290                 for (n = 0; n < 256; n++)
 291                         if (ptab->data[n] > qopt->mtu) break;
 292                 size = (n << qopt->peakrate.cell_log)-1;
 293                 if (size < max_size) max_size = size;
 294         }
 295         if (max_size < 0)
 296                 goto done;
 297
 298         if (qopt->limit > 0) {
 299                 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, qopt->limit);
 300                 if (IS_ERR(child)) {
 301                         err = PTR_ERR(child);
 302                         goto done;
 303                 }
 304         }
 305
 306         sch_tree_lock(sch);
 307         if (child) {
 308                 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
 309                 qdisc_destroy(xchg(&q->qdisc, child));
 310         }
 311         q->limit = qopt->limit;
 312         q->mtu = qopt->mtu;
 313         q->max_size = max_size;
 314         q->buffer = qopt->buffer;
 315         q->tokens = q->buffer;
 316         q->ptokens = q->mtu;
 317         rtab = xchg(&q->R_tab, rtab);
 318         ptab = xchg(&q->P_tab, ptab);
 319         sch_tree_unlock(sch);
 320         err = 0;
 321 done:
 322         if (rtab)
 323                 qdisc_put_rtab(rtab);
 324         if (ptab)
 325                 qdisc_put_rtab(ptab);
 326         return err;
 327 }
 328
 329 static int tbf_init(struct Qdisc* sch, struct nlattr *opt)
 330 {
 331         struct tbf_sched_data *q = qdisc_priv(sch);
 332
 333         if (opt == NULL)
 334                 return -EINVAL;
 335
 336         q->t_c = psched_get_time();
 337         qdisc_watchdog_init(&q->watchdog, sch);
 338         q->qdisc = &noop_qdisc;
 339
 340         return tbf_change(sch, opt);
 341 }
 342
 343 static void tbf_destroy(struct Qdisc *sch)
 344 {
 345         struct tbf_sched_data *q = qdisc_priv(sch);
 346
 347         qdisc_watchdog_cancel(&q->watchdog);
 348
 349         if (q->P_tab)
 350                 qdisc_put_rtab(q->P_tab);
 351         if (q->R_tab)
 352                 qdisc_put_rtab(q->R_tab);
 353
 354         qdisc_destroy(q->qdisc);
 355 }
 356
 357 static int tbf_dump(struct Qdisc *sch, struct sk_buff *skb)
 358 {
 359         struct tbf_sched_data *q = qdisc_priv(sch);
 360         struct nlattr *nest;
 361         struct tc_tbf_qopt opt;
 362
 363         nest = nla_nest_start(skb, TCA_OPTIONS);
 364         if (nest == NULL)
 365                 goto nla_put_failure;
 366
 367         opt.limit = q->limit;
 368         opt.rate = q->R_tab->rate;
 369         if (q->P_tab)
 370                 opt.peakrate = q->P_tab->rate;
 371         else
 372                 memset(&opt.peakrate, 0, sizeof(opt.peakrate));
 373         opt.mtu = q->mtu;
 374         opt.buffer = q->buffer;
 375         NLA_PUT(skb, TCA_TBF_PARMS, sizeof(opt), &opt);
 376
 377         nla_nest_end(skb, nest);
 378         return skb->len;
 379
 380 nla_put_failure:
 381         nla_nest_cancel(skb, nest);
 382         return -1;
 383 }
 384
 385 static int tbf_dump_class(struct Qdisc *sch, unsigned long cl,
 386                           struct sk_buff *skb, struct tcmsg *tcm)
 387 {
 388         struct tbf_sched_data *q = qdisc_priv(sch);
 389
 390         if (cl != 1)    /* only one class */
 391                 return -ENOENT;
 392
 393         tcm->tcm_handle |= TC_H_MIN(1);
 394         tcm->tcm_info = q->qdisc->handle;
 395
 396         return 0;
 397 }
 398
 399 static int tbf_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
 400                      struct Qdisc **old)
 401 {
 402         struct tbf_sched_data *q = qdisc_priv(sch);
 403
 404         if (new == NULL)
 405                 new = &noop_qdisc;
 406
 407         sch_tree_lock(sch);
 408         *old = xchg(&q->qdisc, new);
 409         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
 410         qdisc_reset(*old);
 411         sch_tree_unlock(sch);
 412
 413         return 0;
 414 }
 415
 416 static struct Qdisc *tbf_leaf(struct Qdisc *sch, unsigned long arg)
 417 {
 418         struct tbf_sched_data *q = qdisc_priv(sch);
 419         return q->qdisc;
 420 }
 421
 422 static unsigned long tbf_get(struct Qdisc *sch, u32 classid)
 423 {
 424         return 1;
 425 }
 426
 427 static void tbf_put(struct Qdisc *sch, unsigned long arg)
 428 {
 429 }
 430
 431 static int tbf_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
 432                             struct nlattr **tca, unsigned long *arg)
 433 {
 434         return -ENOSYS;
 435 }
 436
 437 static int tbf_delete(struct Qdisc *sch, unsigned long arg)
 438 {
 439         return -ENOSYS;
 440 }
 441
 442 static void tbf_walk(struct Qdisc *sch, struct qdisc_walker *walker)
 443 {
 444         if (!walker->stop) {
 445                 if (walker->count >= walker->skip)
 446                         if (walker->fn(sch, 1, walker) < 0) {
 447                                 walker->stop = 1;
 448                                 return;
 449                         }
 450                 walker->count++;
 451         }
 452 }
 453
 454 static struct tcf_proto **tbf_find_tcf(struct Qdisc *sch, unsigned long cl)
 455 {
 456         return NULL;
 457 }
 458
 459 static const struct Qdisc_class_ops tbf_class_ops =
 460 {
 461         .graft          =       tbf_graft,
 462         .leaf           =       tbf_leaf,
 463         .get            =       tbf_get,
 464         .put            =       tbf_put,
 465         .change         =       tbf_change_class,
 466         .delete         =       tbf_delete,
 467         .walk           =       tbf_walk,
 468         .tcf_chain      =       tbf_find_tcf,
 469         .dump           =       tbf_dump_class,
 470 };
 471
 472 static struct Qdisc_ops tbf_qdisc_ops __read_mostly = {
 473         .next           =       NULL,
 474         .cl_ops         =       &tbf_class_ops,
 475         .id             =       "tbf",
 476         .priv_size      =       sizeof(struct tbf_sched_data),
 477         .enqueue        =       tbf_enqueue,
 478         .dequeue        =       tbf_dequeue,
 479         .requeue        =       tbf_requeue,
 480         .drop           =       tbf_drop,
 481         .init           =       tbf_init,
 482         .reset          =       tbf_reset,
 483         .destroy        =       tbf_destroy,
 484         .change         =       tbf_change,
 485         .dump           =       tbf_dump,
 486         .owner          =       THIS_MODULE,
 487 };
 488
 489 static int __init tbf_module_init(void)
 490 {
 491         return register_qdisc(&tbf_qdisc_ops);
 492 }
 493
 494 static void __exit tbf_module_exit(void)
 495 {
 496         unregister_qdisc(&tbf_qdisc_ops);
 497 }
 498 module_init(tbf_module_init)
 499 module_exit(tbf_module_exit)
 500 MODULE_LICENSE("GPL");