net/sched/sch_prio.c

   1 /*
   2  * net/sched/sch_prio.c Simple 3-band priority "scheduler".
   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  * Fixes:       19990609: J Hadi Salim <hadi@nortelnetworks.com>:
  11  *              Init --  EINVAL when opt undefined
  12  */
  13
  14 #include <linux/module.h>
  15 #include <asm/uaccess.h>
  16 #include <asm/system.h>
  17 #include <linux/bitops.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/string.h>
  21 #include <linux/mm.h>
  22 #include <linux/socket.h>
  23 #include <linux/sockios.h>
  24 #include <linux/in.h>
  25 #include <linux/errno.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/if_ether.h>
  28 #include <linux/inet.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/notifier.h>
  32 #include <net/ip.h>
  33 #include <net/route.h>
  34 #include <linux/skbuff.h>
  35 #include <net/netlink.h>
  36 #include <net/sock.h>
  37 #include <net/pkt_sched.h>
  38
  39
  40 struct prio_sched_data
  41 {
  42         int bands;
  43         int curband; /* for round-robin */
  44         struct tcf_proto *filter_list;
  45         u8  prio2band[TC_PRIO_MAX+1];
  46         struct Qdisc *queues[TCQ_PRIO_BANDS];
  47         int mq;
  48 };
  49
  50
  51 static struct Qdisc *
  52 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
  53 {
  54         struct prio_sched_data *q = qdisc_priv(sch);
  55         u32 band = skb->priority;
  56         struct tcf_result res;
  57
  58         *qerr = NET_XMIT_BYPASS;
  59         if (TC_H_MAJ(skb->priority) != sch->handle) {
  60 #ifdef CONFIG_NET_CLS_ACT
  61                 switch (tc_classify(skb, q->filter_list, &res)) {
  62                 case TC_ACT_STOLEN:
  63                 case TC_ACT_QUEUED:
  64                         *qerr = NET_XMIT_SUCCESS;
  65                 case TC_ACT_SHOT:
  66                         return NULL;
  67                 }
  68
  69                 if (!q->filter_list ) {
  70 #else
  71                 if (!q->filter_list || tc_classify(skb, q->filter_list, &res)) {
  72 #endif
  73                         if (TC_H_MAJ(band))
  74                                 band = 0;
  75                         band = q->prio2band[band&TC_PRIO_MAX];
  76                         goto out;
  77                 }
  78                 band = res.classid;
  79         }
  80         band = TC_H_MIN(band) - 1;
  81         if (band >= q->bands)
  82                 band = q->prio2band[0];
  83 out:
  84         if (q->mq)
  85                 skb_set_queue_mapping(skb, band);
  86         return q->queues[band];
  87 }
  88
  89 static int
  90 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
  91 {
  92         struct Qdisc *qdisc;
  93         int ret;
  94
  95         qdisc = prio_classify(skb, sch, &ret);
  96 #ifdef CONFIG_NET_CLS_ACT
  97         if (qdisc == NULL) {
  98
  99                 if (ret == NET_XMIT_BYPASS)
 100                         sch->qstats.drops++;
 101                 kfree_skb(skb);
 102                 return ret;
 103         }
 104 #endif
 105
 106         if ((ret = qdisc->enqueue(skb, qdisc)) == NET_XMIT_SUCCESS) {
 107                 sch->bstats.bytes += skb->len;
 108                 sch->bstats.packets++;
 109                 sch->q.qlen++;
 110                 return NET_XMIT_SUCCESS;
 111         }
 112         sch->qstats.drops++;
 113         return ret;
 114 }
 115
 116
 117 static int
 118 prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
 119 {
 120         struct Qdisc *qdisc;
 121         int ret;
 122
 123         qdisc = prio_classify(skb, sch, &ret);
 124 #ifdef CONFIG_NET_CLS_ACT
 125         if (qdisc == NULL) {
 126                 if (ret == NET_XMIT_BYPASS)
 127                         sch->qstats.drops++;
 128                 kfree_skb(skb);
 129                 return ret;
 130         }
 131 #endif
 132
 133         if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
 134                 sch->q.qlen++;
 135                 sch->qstats.requeues++;
 136                 return 0;
 137         }
 138         sch->qstats.drops++;
 139         return NET_XMIT_DROP;
 140 }
 141
 142
 143 static struct sk_buff *
 144 prio_dequeue(struct Qdisc* sch)
 145 {
 146         struct sk_buff *skb;
 147         struct prio_sched_data *q = qdisc_priv(sch);
 148         int prio;
 149         struct Qdisc *qdisc;
 150
 151         for (prio = 0; prio < q->bands; prio++) {
 152                 /* Check if the target subqueue is available before
 153                  * pulling an skb.  This way we avoid excessive requeues
 154                  * for slower queues.
 155                  */
 156                 if (!netif_subqueue_stopped(sch->dev, (q->mq ? prio : 0))) {
 157                         qdisc = q->queues[prio];
 158                         skb = qdisc->dequeue(qdisc);
 159                         if (skb) {
 160                                 sch->q.qlen--;
 161                                 return skb;
 162                         }
 163                 }
 164         }
 165         return NULL;
 166
 167 }
 168
 169 static struct sk_buff *rr_dequeue(struct Qdisc* sch)
 170 {
 171         struct sk_buff *skb;
 172         struct prio_sched_data *q = qdisc_priv(sch);
 173         struct Qdisc *qdisc;
 174         int bandcount;
 175
 176         /* Only take one pass through the queues.  If nothing is available,
 177          * return nothing.
 178          */
 179         for (bandcount = 0; bandcount < q->bands; bandcount++) {
 180                 /* Check if the target subqueue is available before
 181                  * pulling an skb.  This way we avoid excessive requeues
 182                  * for slower queues.  If the queue is stopped, try the
 183                  * next queue.
 184                  */
 185                 if (!netif_subqueue_stopped(sch->dev,
 186                                             (q->mq ? q->curband : 0))) {
 187                         qdisc = q->queues[q->curband];
 188                         skb = qdisc->dequeue(qdisc);
 189                         if (skb) {
 190                                 sch->q.qlen--;
 191                                 q->curband++;
 192                                 if (q->curband >= q->bands)
 193                                         q->curband = 0;
 194                                 return skb;
 195                         }
 196                 }
 197                 q->curband++;
 198                 if (q->curband >= q->bands)
 199                         q->curband = 0;
 200         }
 201         return NULL;
 202 }
 203
 204 static unsigned int prio_drop(struct Qdisc* sch)
 205 {
 206         struct prio_sched_data *q = qdisc_priv(sch);
 207         int prio;
 208         unsigned int len;
 209         struct Qdisc *qdisc;
 210
 211         for (prio = q->bands-1; prio >= 0; prio--) {
 212                 qdisc = q->queues[prio];
 213                 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
 214                         sch->q.qlen--;
 215                         return len;
 216                 }
 217         }
 218         return 0;
 219 }
 220
 221
 222 static void
 223 prio_reset(struct Qdisc* sch)
 224 {
 225         int prio;
 226         struct prio_sched_data *q = qdisc_priv(sch);
 227
 228         for (prio=0; prio<q->bands; prio++)
 229                 qdisc_reset(q->queues[prio]);
 230         sch->q.qlen = 0;
 231 }
 232
 233 static void
 234 prio_destroy(struct Qdisc* sch)
 235 {
 236         int prio;
 237         struct prio_sched_data *q = qdisc_priv(sch);
 238
 239         tcf_destroy_chain(q->filter_list);
 240         for (prio=0; prio<q->bands; prio++)
 241                 qdisc_destroy(q->queues[prio]);
 242 }
 243
 244 static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
 245 {
 246         struct prio_sched_data *q = qdisc_priv(sch);
 247         struct tc_prio_qopt *qopt;
 248         struct rtattr *tb[TCA_PRIO_MAX];
 249         int i;
 250
 251         if (rtattr_parse_nested_compat(tb, TCA_PRIO_MAX, opt, qopt,
 252                                        sizeof(*qopt)))
 253                 return -EINVAL;
 254         q->bands = qopt->bands;
 255         /* If we're multiqueue, make sure the number of incoming bands
 256          * matches the number of queues on the device we're associating with.
 257          * If the number of bands requested is zero, then set q->bands to
 258          * dev->egress_subqueue_count.
 259          */
 260         q->mq = RTA_GET_FLAG(tb[TCA_PRIO_MQ - 1]);
 261         if (q->mq) {
 262                 if (sch->handle != TC_H_ROOT)
 263                         return -EINVAL;
 264                 if (netif_is_multiqueue(sch->dev)) {
 265                         if (q->bands == 0)
 266                                 q->bands = sch->dev->egress_subqueue_count;
 267                         else if (q->bands != sch->dev->egress_subqueue_count)
 268                                 return -EINVAL;
 269                 } else
 270                         return -EOPNOTSUPP;
 271         }
 272
 273         if (q->bands > TCQ_PRIO_BANDS || q->bands < 2)
 274                 return -EINVAL;
 275
 276         for (i=0; i<=TC_PRIO_MAX; i++) {
 277                 if (qopt->priomap[i] >= q->bands)
 278                         return -EINVAL;
 279         }
 280
 281         sch_tree_lock(sch);
 282         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
 283
 284         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
 285                 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
 286                 if (child != &noop_qdisc) {
 287                         qdisc_tree_decrease_qlen(child, child->q.qlen);
 288                         qdisc_destroy(child);
 289                 }
 290         }
 291         sch_tree_unlock(sch);
 292
 293         for (i=0; i<q->bands; i++) {
 294                 if (q->queues[i] == &noop_qdisc) {
 295                         struct Qdisc *child;
 296                         child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
 297                                                   TC_H_MAKE(sch->handle, i + 1));
 298                         if (child) {
 299                                 sch_tree_lock(sch);
 300                                 child = xchg(&q->queues[i], child);
 301
 302                                 if (child != &noop_qdisc) {
 303                                         qdisc_tree_decrease_qlen(child,
 304                                                                  child->q.qlen);
 305                                         qdisc_destroy(child);
 306                                 }
 307                                 sch_tree_unlock(sch);
 308                         }
 309                 }
 310         }
 311         return 0;
 312 }
 313
 314 static int prio_init(struct Qdisc *sch, struct rtattr *opt)
 315 {
 316         struct prio_sched_data *q = qdisc_priv(sch);
 317         int i;
 318
 319         for (i=0; i<TCQ_PRIO_BANDS; i++)
 320                 q->queues[i] = &noop_qdisc;
 321
 322         if (opt == NULL) {
 323                 return -EINVAL;
 324         } else {
 325                 int err;
 326
 327                 if ((err= prio_tune(sch, opt)) != 0)
 328                         return err;
 329         }
 330         return 0;
 331 }
 332
 333 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
 334 {
 335         struct prio_sched_data *q = qdisc_priv(sch);
 336         unsigned char *b = skb_tail_pointer(skb);
 337         struct rtattr *nest;
 338         struct tc_prio_qopt opt;
 339
 340         opt.bands = q->bands;
 341         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
 342
 343         nest = RTA_NEST_COMPAT(skb, TCA_OPTIONS, sizeof(opt), &opt);
 344         if (q->mq)
 345                 RTA_PUT_FLAG(skb, TCA_PRIO_MQ);
 346         RTA_NEST_COMPAT_END(skb, nest);
 347
 348         return skb->len;
 349
 350 rtattr_failure:
 351         nlmsg_trim(skb, b);
 352         return -1;
 353 }
 354
 355 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
 356                       struct Qdisc **old)
 357 {
 358         struct prio_sched_data *q = qdisc_priv(sch);
 359         unsigned long band = arg - 1;
 360
 361         if (band >= q->bands)
 362                 return -EINVAL;
 363
 364         if (new == NULL)
 365                 new = &noop_qdisc;
 366
 367         sch_tree_lock(sch);
 368         *old = q->queues[band];
 369         q->queues[band] = new;
 370         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
 371         qdisc_reset(*old);
 372         sch_tree_unlock(sch);
 373
 374         return 0;
 375 }
 376
 377 static struct Qdisc *
 378 prio_leaf(struct Qdisc *sch, unsigned long arg)
 379 {
 380         struct prio_sched_data *q = qdisc_priv(sch);
 381         unsigned long band = arg - 1;
 382
 383         if (band >= q->bands)
 384                 return NULL;
 385
 386         return q->queues[band];
 387 }
 388
 389 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
 390 {
 391         struct prio_sched_data *q = qdisc_priv(sch);
 392         unsigned long band = TC_H_MIN(classid);
 393
 394         if (band - 1 >= q->bands)
 395                 return 0;
 396         return band;
 397 }
 398
 399 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
 400 {
 401         return prio_get(sch, classid);
 402 }
 403
 404
 405 static void prio_put(struct Qdisc *q, unsigned long cl)
 406 {
 407         return;
 408 }
 409
 410 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct rtattr **tca, unsigned long *arg)
 411 {
 412         unsigned long cl = *arg;
 413         struct prio_sched_data *q = qdisc_priv(sch);
 414
 415         if (cl - 1 > q->bands)
 416                 return -ENOENT;
 417         return 0;
 418 }
 419
 420 static int prio_delete(struct Qdisc *sch, unsigned long cl)
 421 {
 422         struct prio_sched_data *q = qdisc_priv(sch);
 423         if (cl - 1 > q->bands)
 424                 return -ENOENT;
 425         return 0;
 426 }
 427
 428
 429 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
 430                            struct tcmsg *tcm)
 431 {
 432         struct prio_sched_data *q = qdisc_priv(sch);
 433
 434         if (cl - 1 > q->bands)
 435                 return -ENOENT;
 436         tcm->tcm_handle |= TC_H_MIN(cl);
 437         if (q->queues[cl-1])
 438                 tcm->tcm_info = q->queues[cl-1]->handle;
 439         return 0;
 440 }
 441
 442 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
 443                                  struct gnet_dump *d)
 444 {
 445         struct prio_sched_data *q = qdisc_priv(sch);
 446         struct Qdisc *cl_q;
 447
 448         cl_q = q->queues[cl - 1];
 449         if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
 450             gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
 451                 return -1;
 452
 453         return 0;
 454 }
 455
 456 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 457 {
 458         struct prio_sched_data *q = qdisc_priv(sch);
 459         int prio;
 460
 461         if (arg->stop)
 462                 return;
 463
 464         for (prio = 0; prio < q->bands; prio++) {
 465                 if (arg->count < arg->skip) {
 466                         arg->count++;
 467                         continue;
 468                 }
 469                 if (arg->fn(sch, prio+1, arg) < 0) {
 470                         arg->stop = 1;
 471                         break;
 472                 }
 473                 arg->count++;
 474         }
 475 }
 476
 477 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
 478 {
 479         struct prio_sched_data *q = qdisc_priv(sch);
 480
 481         if (cl)
 482                 return NULL;
 483         return &q->filter_list;
 484 }
 485
 486 static struct Qdisc_class_ops prio_class_ops = {
 487         .graft          =       prio_graft,
 488         .leaf           =       prio_leaf,
 489         .get            =       prio_get,
 490         .put            =       prio_put,
 491         .change         =       prio_change,
 492         .delete         =       prio_delete,
 493         .walk           =       prio_walk,
 494         .tcf_chain      =       prio_find_tcf,
 495         .bind_tcf       =       prio_bind,
 496         .unbind_tcf     =       prio_put,
 497         .dump           =       prio_dump_class,
 498         .dump_stats     =       prio_dump_class_stats,
 499 };
 500
 501 static struct Qdisc_ops prio_qdisc_ops = {
 502         .next           =       NULL,
 503         .cl_ops         =       &prio_class_ops,
 504         .id             =       "prio",
 505         .priv_size      =       sizeof(struct prio_sched_data),
 506         .enqueue        =       prio_enqueue,
 507         .dequeue        =       prio_dequeue,
 508         .requeue        =       prio_requeue,
 509         .drop           =       prio_drop,
 510         .init           =       prio_init,
 511         .reset          =       prio_reset,
 512         .destroy        =       prio_destroy,
 513         .change         =       prio_tune,
 514         .dump           =       prio_dump,
 515         .owner          =       THIS_MODULE,
 516 };
 517
 518 static struct Qdisc_ops rr_qdisc_ops = {
 519         .next           =       NULL,
 520         .cl_ops         =       &prio_class_ops,
 521         .id             =       "rr",
 522         .priv_size      =       sizeof(struct prio_sched_data),
 523         .enqueue        =       prio_enqueue,
 524         .dequeue        =       rr_dequeue,
 525         .requeue        =       prio_requeue,
 526         .drop           =       prio_drop,
 527         .init           =       prio_init,
 528         .reset          =       prio_reset,
 529         .destroy        =       prio_destroy,
 530         .change         =       prio_tune,
 531         .dump           =       prio_dump,
 532         .owner          =       THIS_MODULE,
 533 };
 534
 535 static int __init prio_module_init(void)
 536 {
 537         int err;
 538
 539         err = register_qdisc(&prio_qdisc_ops);
 540         if (err < 0)
 541                 return err;
 542         err = register_qdisc(&rr_qdisc_ops);
 543         if (err < 0)
 544                 unregister_qdisc(&prio_qdisc_ops);
 545         return err;
 546 }
 547
 548 static void __exit prio_module_exit(void)
 549 {
 550         unregister_qdisc(&prio_qdisc_ops);
 551         unregister_qdisc(&rr_qdisc_ops);
 552 }
 553
 554 module_init(prio_module_init)
 555 module_exit(prio_module_exit)
 556
 557 MODULE_LICENSE("GPL");
 558 MODULE_ALIAS("sch_rr");