net/sched/sch_drr.c

   1 /*
   2  * net/sched/sch_drr.c         Deficit Round Robin scheduler
   3  *
   4  * Copyright (c) 2008 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * version 2 as published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/slab.h>
  13 #include <linux/init.h>
  14 #include <linux/errno.h>
  15 #include <linux/netdevice.h>
  16 #include <linux/pkt_sched.h>
  17 #include <net/sch_generic.h>
  18 #include <net/pkt_sched.h>
  19 #include <net/pkt_cls.h>
  20
  21 struct drr_class {
  22         struct Qdisc_class_common       common;
  23         unsigned int                    refcnt;
  24         unsigned int                    filter_cnt;
  25
  26         struct gnet_stats_basic_packed          bstats;
  27         struct gnet_stats_queue         qstats;
  28         struct net_rate_estimator __rcu *rate_est;
  29         struct list_head                alist;
  30         struct Qdisc                    *qdisc;
  31
  32         u32                             quantum;
  33         u32                             deficit;
  34 };
  35
  36 struct drr_sched {
  37         struct list_head                active;
  38         struct tcf_proto __rcu          *filter_list;
  39         struct tcf_block                *block;
  40         struct Qdisc_class_hash         clhash;
  41 };
  42
  43 static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid)
  44 {
  45         struct drr_sched *q = qdisc_priv(sch);
  46         struct Qdisc_class_common *clc;
  47
  48         clc = qdisc_class_find(&q->clhash, classid);
  49         if (clc == NULL)
  50                 return NULL;
  51         return container_of(clc, struct drr_class, common);
  52 }
  53
  54 static void drr_purge_queue(struct drr_class *cl)
  55 {
  56         unsigned int len = cl->qdisc->q.qlen;
  57         unsigned int backlog = cl->qdisc->qstats.backlog;
  58
  59         qdisc_reset(cl->qdisc);
  60         qdisc_tree_reduce_backlog(cl->qdisc, len, backlog);
  61 }
  62
  63 static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = {
  64         [TCA_DRR_QUANTUM]       = { .type = NLA_U32 },
  65 };
  66
  67 static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
  68                             struct nlattr **tca, unsigned long *arg)
  69 {
  70         struct drr_sched *q = qdisc_priv(sch);
  71         struct drr_class *cl = (struct drr_class *)*arg;
  72         struct nlattr *opt = tca[TCA_OPTIONS];
  73         struct nlattr *tb[TCA_DRR_MAX + 1];
  74         u32 quantum;
  75         int err;
  76
  77         if (!opt)
  78                 return -EINVAL;
  79
  80         err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy, NULL);
  81         if (err < 0)
  82                 return err;
  83
  84         if (tb[TCA_DRR_QUANTUM]) {
  85                 quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]);
  86                 if (quantum == 0)
  87                         return -EINVAL;
  88         } else
  89                 quantum = psched_mtu(qdisc_dev(sch));
  90
  91         if (cl != NULL) {
  92                 if (tca[TCA_RATE]) {
  93                         err = gen_replace_estimator(&cl->bstats, NULL,
  94                                                     &cl->rate_est,
  95                                                     NULL,
  96                                                     qdisc_root_sleeping_running(sch),
  97                                                     tca[TCA_RATE]);
  98                         if (err)
  99                                 return err;
 100                 }
 101
 102                 sch_tree_lock(sch);
 103                 if (tb[TCA_DRR_QUANTUM])
 104                         cl->quantum = quantum;
 105                 sch_tree_unlock(sch);
 106
 107                 return 0;
 108         }
 109
 110         cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL);
 111         if (cl == NULL)
 112                 return -ENOBUFS;
 113
 114         cl->refcnt         = 1;
 115         cl->common.classid = classid;
 116         cl->quantum        = quantum;
 117         cl->qdisc          = qdisc_create_dflt(sch->dev_queue,
 118                                                &pfifo_qdisc_ops, classid);
 119         if (cl->qdisc == NULL)
 120                 cl->qdisc = &noop_qdisc;
 121         else
 122                 qdisc_hash_add(cl->qdisc, true);
 123
 124         if (tca[TCA_RATE]) {
 125                 err = gen_replace_estimator(&cl->bstats, NULL, &cl->rate_est,
 126                                             NULL,
 127                                             qdisc_root_sleeping_running(sch),
 128                                             tca[TCA_RATE]);
 129                 if (err) {
 130                         qdisc_destroy(cl->qdisc);
 131                         kfree(cl);
 132                         return err;
 133                 }
 134         }
 135
 136         sch_tree_lock(sch);
 137         qdisc_class_hash_insert(&q->clhash, &cl->common);
 138         sch_tree_unlock(sch);
 139
 140         qdisc_class_hash_grow(sch, &q->clhash);
 141
 142         *arg = (unsigned long)cl;
 143         return 0;
 144 }
 145
 146 static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl)
 147 {
 148         gen_kill_estimator(&cl->rate_est);
 149         qdisc_destroy(cl->qdisc);
 150         kfree(cl);
 151 }
 152
 153 static int drr_delete_class(struct Qdisc *sch, unsigned long arg)
 154 {
 155         struct drr_sched *q = qdisc_priv(sch);
 156         struct drr_class *cl = (struct drr_class *)arg;
 157
 158         if (cl->filter_cnt > 0)
 159                 return -EBUSY;
 160
 161         sch_tree_lock(sch);
 162
 163         drr_purge_queue(cl);
 164         qdisc_class_hash_remove(&q->clhash, &cl->common);
 165
 166         BUG_ON(--cl->refcnt == 0);
 167         /*
 168          * This shouldn't happen: we "hold" one cops->get() when called
 169          * from tc_ctl_tclass; the destroy method is done from cops->put().
 170          */
 171
 172         sch_tree_unlock(sch);
 173         return 0;
 174 }
 175
 176 static unsigned long drr_get_class(struct Qdisc *sch, u32 classid)
 177 {
 178         struct drr_class *cl = drr_find_class(sch, classid);
 179
 180         if (cl != NULL)
 181                 cl->refcnt++;
 182
 183         return (unsigned long)cl;
 184 }
 185
 186 static void drr_put_class(struct Qdisc *sch, unsigned long arg)
 187 {
 188         struct drr_class *cl = (struct drr_class *)arg;
 189
 190         if (--cl->refcnt == 0)
 191                 drr_destroy_class(sch, cl);
 192 }
 193
 194 static struct tcf_block *drr_tcf_block(struct Qdisc *sch, unsigned long cl)
 195 {
 196         struct drr_sched *q = qdisc_priv(sch);
 197
 198         if (cl)
 199                 return NULL;
 200
 201         return q->block;
 202 }
 203
 204 static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent,
 205                                   u32 classid)
 206 {
 207         struct drr_class *cl = drr_find_class(sch, classid);
 208
 209         if (cl != NULL)
 210                 cl->filter_cnt++;
 211
 212         return (unsigned long)cl;
 213 }
 214
 215 static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg)
 216 {
 217         struct drr_class *cl = (struct drr_class *)arg;
 218
 219         cl->filter_cnt--;
 220 }
 221
 222 static int drr_graft_class(struct Qdisc *sch, unsigned long arg,
 223                            struct Qdisc *new, struct Qdisc **old)
 224 {
 225         struct drr_class *cl = (struct drr_class *)arg;
 226
 227         if (new == NULL) {
 228                 new = qdisc_create_dflt(sch->dev_queue,
 229                                         &pfifo_qdisc_ops, cl->common.classid);
 230                 if (new == NULL)
 231                         new = &noop_qdisc;
 232         }
 233
 234         *old = qdisc_replace(sch, new, &cl->qdisc);
 235         return 0;
 236 }
 237
 238 static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg)
 239 {
 240         struct drr_class *cl = (struct drr_class *)arg;
 241
 242         return cl->qdisc;
 243 }
 244
 245 static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg)
 246 {
 247         struct drr_class *cl = (struct drr_class *)arg;
 248
 249         list_del(&cl->alist);
 250 }
 251
 252 static int drr_dump_class(struct Qdisc *sch, unsigned long arg,
 253                           struct sk_buff *skb, struct tcmsg *tcm)
 254 {
 255         struct drr_class *cl = (struct drr_class *)arg;
 256         struct nlattr *nest;
 257
 258         tcm->tcm_parent = TC_H_ROOT;
 259         tcm->tcm_handle = cl->common.classid;
 260         tcm->tcm_info   = cl->qdisc->handle;
 261
 262         nest = nla_nest_start(skb, TCA_OPTIONS);
 263         if (nest == NULL)
 264                 goto nla_put_failure;
 265         if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum))
 266                 goto nla_put_failure;
 267         return nla_nest_end(skb, nest);
 268
 269 nla_put_failure:
 270         nla_nest_cancel(skb, nest);
 271         return -EMSGSIZE;
 272 }
 273
 274 static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg,
 275                                 struct gnet_dump *d)
 276 {
 277         struct drr_class *cl = (struct drr_class *)arg;
 278         __u32 qlen = cl->qdisc->q.qlen;
 279         struct tc_drr_stats xstats;
 280
 281         memset(&xstats, 0, sizeof(xstats));
 282         if (qlen)
 283                 xstats.deficit = cl->deficit;
 284
 285         if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
 286                                   d, NULL, &cl->bstats) < 0 ||
 287             gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
 288             gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0)
 289                 return -1;
 290
 291         return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
 292 }
 293
 294 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 295 {
 296         struct drr_sched *q = qdisc_priv(sch);
 297         struct drr_class *cl;
 298         unsigned int i;
 299
 300         if (arg->stop)
 301                 return;
 302
 303         for (i = 0; i < q->clhash.hashsize; i++) {
 304                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 305                         if (arg->count < arg->skip) {
 306                                 arg->count++;
 307                                 continue;
 308                         }
 309                         if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
 310                                 arg->stop = 1;
 311                                 return;
 312                         }
 313                         arg->count++;
 314                 }
 315         }
 316 }
 317
 318 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch,
 319                                       int *qerr)
 320 {
 321         struct drr_sched *q = qdisc_priv(sch);
 322         struct drr_class *cl;
 323         struct tcf_result res;
 324         struct tcf_proto *fl;
 325         int result;
 326
 327         if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
 328                 cl = drr_find_class(sch, skb->priority);
 329                 if (cl != NULL)
 330                         return cl;
 331         }
 332
 333         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 334         fl = rcu_dereference_bh(q->filter_list);
 335         result = tcf_classify(skb, fl, &res, false);
 336         if (result >= 0) {
 337 #ifdef CONFIG_NET_CLS_ACT
 338                 switch (result) {
 339                 case TC_ACT_QUEUED:
 340                 case TC_ACT_STOLEN:
 341                 case TC_ACT_TRAP:
 342                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 343                 case TC_ACT_SHOT:
 344                         return NULL;
 345                 }
 346 #endif
 347                 cl = (struct drr_class *)res.class;
 348                 if (cl == NULL)
 349                         cl = drr_find_class(sch, res.classid);
 350                 return cl;
 351         }
 352         return NULL;
 353 }
 354
 355 static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 356                        struct sk_buff **to_free)
 357 {
 358         struct drr_sched *q = qdisc_priv(sch);
 359         struct drr_class *cl;
 360         int err = 0;
 361
 362         cl = drr_classify(skb, sch, &err);
 363         if (cl == NULL) {
 364                 if (err & __NET_XMIT_BYPASS)
 365                         qdisc_qstats_drop(sch);
 366                 __qdisc_drop(skb, to_free);
 367                 return err;
 368         }
 369
 370         err = qdisc_enqueue(skb, cl->qdisc, to_free);
 371         if (unlikely(err != NET_XMIT_SUCCESS)) {
 372                 if (net_xmit_drop_count(err)) {
 373                         cl->qstats.drops++;
 374                         qdisc_qstats_drop(sch);
 375                 }
 376                 return err;
 377         }
 378
 379         if (cl->qdisc->q.qlen == 1) {
 380                 list_add_tail(&cl->alist, &q->active);
 381                 cl->deficit = cl->quantum;
 382         }
 383
 384         qdisc_qstats_backlog_inc(sch, skb);
 385         sch->q.qlen++;
 386         return err;
 387 }
 388
 389 static struct sk_buff *drr_dequeue(struct Qdisc *sch)
 390 {
 391         struct drr_sched *q = qdisc_priv(sch);
 392         struct drr_class *cl;
 393         struct sk_buff *skb;
 394         unsigned int len;
 395
 396         if (list_empty(&q->active))
 397                 goto out;
 398         while (1) {
 399                 cl = list_first_entry(&q->active, struct drr_class, alist);
 400                 skb = cl->qdisc->ops->peek(cl->qdisc);
 401                 if (skb == NULL) {
 402                         qdisc_warn_nonwc(__func__, cl->qdisc);
 403                         goto out;
 404                 }
 405
 406                 len = qdisc_pkt_len(skb);
 407                 if (len <= cl->deficit) {
 408                         cl->deficit -= len;
 409                         skb = qdisc_dequeue_peeked(cl->qdisc);
 410                         if (unlikely(skb == NULL))
 411                                 goto out;
 412                         if (cl->qdisc->q.qlen == 0)
 413                                 list_del(&cl->alist);
 414
 415                         bstats_update(&cl->bstats, skb);
 416                         qdisc_bstats_update(sch, skb);
 417                         qdisc_qstats_backlog_dec(sch, skb);
 418                         sch->q.qlen--;
 419                         return skb;
 420                 }
 421
 422                 cl->deficit += cl->quantum;
 423                 list_move_tail(&cl->alist, &q->active);
 424         }
 425 out:
 426         return NULL;
 427 }
 428
 429 static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
 430 {
 431         struct drr_sched *q = qdisc_priv(sch);
 432         int err;
 433
 434         err = tcf_block_get(&q->block, &q->filter_list);
 435         if (err)
 436                 return err;
 437         err = qdisc_class_hash_init(&q->clhash);
 438         if (err < 0)
 439                 return err;
 440         INIT_LIST_HEAD(&q->active);
 441         return 0;
 442 }
 443
 444 static void drr_reset_qdisc(struct Qdisc *sch)
 445 {
 446         struct drr_sched *q = qdisc_priv(sch);
 447         struct drr_class *cl;
 448         unsigned int i;
 449
 450         for (i = 0; i < q->clhash.hashsize; i++) {
 451                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 452                         if (cl->qdisc->q.qlen)
 453                                 list_del(&cl->alist);
 454                         qdisc_reset(cl->qdisc);
 455                 }
 456         }
 457         sch->qstats.backlog = 0;
 458         sch->q.qlen = 0;
 459 }
 460
 461 static void drr_destroy_qdisc(struct Qdisc *sch)
 462 {
 463         struct drr_sched *q = qdisc_priv(sch);
 464         struct drr_class *cl;
 465         struct hlist_node *next;
 466         unsigned int i;
 467
 468         tcf_block_put(q->block);
 469
 470         for (i = 0; i < q->clhash.hashsize; i++) {
 471                 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
 472                                           common.hnode)
 473                         drr_destroy_class(sch, cl);
 474         }
 475         qdisc_class_hash_destroy(&q->clhash);
 476 }
 477
 478 static const struct Qdisc_class_ops drr_class_ops = {
 479         .change         = drr_change_class,
 480         .delete         = drr_delete_class,
 481         .get            = drr_get_class,
 482         .put            = drr_put_class,
 483         .tcf_block      = drr_tcf_block,
 484         .bind_tcf       = drr_bind_tcf,
 485         .unbind_tcf     = drr_unbind_tcf,
 486         .graft          = drr_graft_class,
 487         .leaf           = drr_class_leaf,
 488         .qlen_notify    = drr_qlen_notify,
 489         .dump           = drr_dump_class,
 490         .dump_stats     = drr_dump_class_stats,
 491         .walk           = drr_walk,
 492 };
 493
 494 static struct Qdisc_ops drr_qdisc_ops __read_mostly = {
 495         .cl_ops         = &drr_class_ops,
 496         .id             = "drr",
 497         .priv_size      = sizeof(struct drr_sched),
 498         .enqueue        = drr_enqueue,
 499         .dequeue        = drr_dequeue,
 500         .peek           = qdisc_peek_dequeued,
 501         .init           = drr_init_qdisc,
 502         .reset          = drr_reset_qdisc,
 503         .destroy        = drr_destroy_qdisc,
 504         .owner          = THIS_MODULE,
 505 };
 506
 507 static int __init drr_init(void)
 508 {
 509         return register_qdisc(&drr_qdisc_ops);
 510 }
 511
 512 static void __exit drr_exit(void)
 513 {
 514         unregister_qdisc(&drr_qdisc_ops);
 515 }
 516
 517 module_init(drr_init);
 518 module_exit(drr_exit);
 519 MODULE_LICENSE("GPL");