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 gnet_stats_rate_est64    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 Qdisc_class_hash         clhash;
  40 };
  41
  42 static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid)
  43 {
  44         struct drr_sched *q = qdisc_priv(sch);
  45         struct Qdisc_class_common *clc;
  46
  47         clc = qdisc_class_find(&q->clhash, classid);
  48         if (clc == NULL)
  49                 return NULL;
  50         return container_of(clc, struct drr_class, common);
  51 }
  52
  53 static void drr_purge_queue(struct drr_class *cl)
  54 {
  55         unsigned int len = cl->qdisc->q.qlen;
  56         unsigned int backlog = cl->qdisc->qstats.backlog;
  57
  58         qdisc_reset(cl->qdisc);
  59         qdisc_tree_reduce_backlog(cl->qdisc, len, backlog);
  60 }
  61
  62 static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = {
  63         [TCA_DRR_QUANTUM]       = { .type = NLA_U32 },
  64 };
  65
  66 static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
  67                             struct nlattr **tca, unsigned long *arg)
  68 {
  69         struct drr_sched *q = qdisc_priv(sch);
  70         struct drr_class *cl = (struct drr_class *)*arg;
  71         struct nlattr *opt = tca[TCA_OPTIONS];
  72         struct nlattr *tb[TCA_DRR_MAX + 1];
  73         u32 quantum;
  74         int err;
  75
  76         if (!opt)
  77                 return -EINVAL;
  78
  79         err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy);
  80         if (err < 0)
  81                 return err;
  82
  83         if (tb[TCA_DRR_QUANTUM]) {
  84                 quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]);
  85                 if (quantum == 0)
  86                         return -EINVAL;
  87         } else
  88                 quantum = psched_mtu(qdisc_dev(sch));
  89
  90         if (cl != NULL) {
  91                 if (tca[TCA_RATE]) {
  92                         err = gen_replace_estimator(&cl->bstats, NULL,
  93                                                     &cl->rate_est,
  94                                                     NULL,
  95                                                     qdisc_root_sleeping_running(sch),
  96                                                     tca[TCA_RATE]);
  97                         if (err)
  98                                 return err;
  99                 }
 100
 101                 sch_tree_lock(sch);
 102                 if (tb[TCA_DRR_QUANTUM])
 103                         cl->quantum = quantum;
 104                 sch_tree_unlock(sch);
 105
 106                 return 0;
 107         }
 108
 109         cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL);
 110         if (cl == NULL)
 111                 return -ENOBUFS;
 112
 113         cl->refcnt         = 1;
 114         cl->common.classid = classid;
 115         cl->quantum        = quantum;
 116         cl->qdisc          = qdisc_create_dflt(sch->dev_queue,
 117                                                &pfifo_qdisc_ops, classid);
 118         if (cl->qdisc == NULL)
 119                 cl->qdisc = &noop_qdisc;
 120
 121         if (tca[TCA_RATE]) {
 122                 err = gen_replace_estimator(&cl->bstats, NULL, &cl->rate_est,
 123                                             NULL,
 124                                             qdisc_root_sleeping_running(sch),
 125                                             tca[TCA_RATE]);
 126                 if (err) {
 127                         qdisc_destroy(cl->qdisc);
 128                         kfree(cl);
 129                         return err;
 130                 }
 131         }
 132
 133         sch_tree_lock(sch);
 134         qdisc_class_hash_insert(&q->clhash, &cl->common);
 135         sch_tree_unlock(sch);
 136
 137         qdisc_class_hash_grow(sch, &q->clhash);
 138
 139         *arg = (unsigned long)cl;
 140         return 0;
 141 }
 142
 143 static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl)
 144 {
 145         gen_kill_estimator(&cl->bstats, &cl->rate_est);
 146         qdisc_destroy(cl->qdisc);
 147         kfree(cl);
 148 }
 149
 150 static int drr_delete_class(struct Qdisc *sch, unsigned long arg)
 151 {
 152         struct drr_sched *q = qdisc_priv(sch);
 153         struct drr_class *cl = (struct drr_class *)arg;
 154
 155         if (cl->filter_cnt > 0)
 156                 return -EBUSY;
 157
 158         sch_tree_lock(sch);
 159
 160         drr_purge_queue(cl);
 161         qdisc_class_hash_remove(&q->clhash, &cl->common);
 162
 163         BUG_ON(--cl->refcnt == 0);
 164         /*
 165          * This shouldn't happen: we "hold" one cops->get() when called
 166          * from tc_ctl_tclass; the destroy method is done from cops->put().
 167          */
 168
 169         sch_tree_unlock(sch);
 170         return 0;
 171 }
 172
 173 static unsigned long drr_get_class(struct Qdisc *sch, u32 classid)
 174 {
 175         struct drr_class *cl = drr_find_class(sch, classid);
 176
 177         if (cl != NULL)
 178                 cl->refcnt++;
 179
 180         return (unsigned long)cl;
 181 }
 182
 183 static void drr_put_class(struct Qdisc *sch, unsigned long arg)
 184 {
 185         struct drr_class *cl = (struct drr_class *)arg;
 186
 187         if (--cl->refcnt == 0)
 188                 drr_destroy_class(sch, cl);
 189 }
 190
 191 static struct tcf_proto __rcu **drr_tcf_chain(struct Qdisc *sch,
 192                                               unsigned long cl)
 193 {
 194         struct drr_sched *q = qdisc_priv(sch);
 195
 196         if (cl)
 197                 return NULL;
 198
 199         return &q->filter_list;
 200 }
 201
 202 static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent,
 203                                   u32 classid)
 204 {
 205         struct drr_class *cl = drr_find_class(sch, classid);
 206
 207         if (cl != NULL)
 208                 cl->filter_cnt++;
 209
 210         return (unsigned long)cl;
 211 }
 212
 213 static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg)
 214 {
 215         struct drr_class *cl = (struct drr_class *)arg;
 216
 217         cl->filter_cnt--;
 218 }
 219
 220 static int drr_graft_class(struct Qdisc *sch, unsigned long arg,
 221                            struct Qdisc *new, struct Qdisc **old)
 222 {
 223         struct drr_class *cl = (struct drr_class *)arg;
 224
 225         if (new == NULL) {
 226                 new = qdisc_create_dflt(sch->dev_queue,
 227                                         &pfifo_qdisc_ops, cl->common.classid);
 228                 if (new == NULL)
 229                         new = &noop_qdisc;
 230         }
 231
 232         *old = qdisc_replace(sch, new, &cl->qdisc);
 233         return 0;
 234 }
 235
 236 static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg)
 237 {
 238         struct drr_class *cl = (struct drr_class *)arg;
 239
 240         return cl->qdisc;
 241 }
 242
 243 static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg)
 244 {
 245         struct drr_class *cl = (struct drr_class *)arg;
 246
 247         if (cl->qdisc->q.qlen == 0)
 248                 list_del(&cl->alist);
 249 }
 250
 251 static int drr_dump_class(struct Qdisc *sch, unsigned long arg,
 252                           struct sk_buff *skb, struct tcmsg *tcm)
 253 {
 254         struct drr_class *cl = (struct drr_class *)arg;
 255         struct nlattr *nest;
 256
 257         tcm->tcm_parent = TC_H_ROOT;
 258         tcm->tcm_handle = cl->common.classid;
 259         tcm->tcm_info   = cl->qdisc->handle;
 260
 261         nest = nla_nest_start(skb, TCA_OPTIONS);
 262         if (nest == NULL)
 263                 goto nla_put_failure;
 264         if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum))
 265                 goto nla_put_failure;
 266         return nla_nest_end(skb, nest);
 267
 268 nla_put_failure:
 269         nla_nest_cancel(skb, nest);
 270         return -EMSGSIZE;
 271 }
 272
 273 static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg,
 274                                 struct gnet_dump *d)
 275 {
 276         struct drr_class *cl = (struct drr_class *)arg;
 277         __u32 qlen = cl->qdisc->q.qlen;
 278         struct tc_drr_stats xstats;
 279
 280         memset(&xstats, 0, sizeof(xstats));
 281         if (qlen)
 282                 xstats.deficit = cl->deficit;
 283
 284         if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
 285                                   d, NULL, &cl->bstats) < 0 ||
 286             gnet_stats_copy_rate_est(d, &cl->bstats, &cl->rate_est) < 0 ||
 287             gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0)
 288                 return -1;
 289
 290         return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
 291 }
 292
 293 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 294 {
 295         struct drr_sched *q = qdisc_priv(sch);
 296         struct drr_class *cl;
 297         unsigned int i;
 298
 299         if (arg->stop)
 300                 return;
 301
 302         for (i = 0; i < q->clhash.hashsize; i++) {
 303                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 304                         if (arg->count < arg->skip) {
 305                                 arg->count++;
 306                                 continue;
 307                         }
 308                         if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
 309                                 arg->stop = 1;
 310                                 return;
 311                         }
 312                         arg->count++;
 313                 }
 314         }
 315 }
 316
 317 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch,
 318                                       int *qerr)
 319 {
 320         struct drr_sched *q = qdisc_priv(sch);
 321         struct drr_class *cl;
 322         struct tcf_result res;
 323         struct tcf_proto *fl;
 324         int result;
 325
 326         if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
 327                 cl = drr_find_class(sch, skb->priority);
 328                 if (cl != NULL)
 329                         return cl;
 330         }
 331
 332         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 333         fl = rcu_dereference_bh(q->filter_list);
 334         result = tc_classify(skb, fl, &res, false);
 335         if (result >= 0) {
 336 #ifdef CONFIG_NET_CLS_ACT
 337                 switch (result) {
 338                 case TC_ACT_QUEUED:
 339                 case TC_ACT_STOLEN:
 340                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 341                 case TC_ACT_SHOT:
 342                         return NULL;
 343                 }
 344 #endif
 345                 cl = (struct drr_class *)res.class;
 346                 if (cl == NULL)
 347                         cl = drr_find_class(sch, res.classid);
 348                 return cl;
 349         }
 350         return NULL;
 351 }
 352
 353 static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 354                        struct sk_buff **to_free)
 355 {
 356         struct drr_sched *q = qdisc_priv(sch);
 357         struct drr_class *cl;
 358         int err = 0;
 359
 360         cl = drr_classify(skb, sch, &err);
 361         if (cl == NULL) {
 362                 if (err & __NET_XMIT_BYPASS)
 363                         qdisc_qstats_drop(sch);
 364                 __qdisc_drop(skb, to_free);
 365                 return err;
 366         }
 367
 368         err = qdisc_enqueue(skb, cl->qdisc, to_free);
 369         if (unlikely(err != NET_XMIT_SUCCESS)) {
 370                 if (net_xmit_drop_count(err)) {
 371                         cl->qstats.drops++;
 372                         qdisc_qstats_drop(sch);
 373                 }
 374                 return err;
 375         }
 376
 377         if (cl->qdisc->q.qlen == 1) {
 378                 list_add_tail(&cl->alist, &q->active);
 379                 cl->deficit = cl->quantum;
 380         }
 381
 382         qdisc_qstats_backlog_inc(sch, skb);
 383         sch->q.qlen++;
 384         return err;
 385 }
 386
 387 static struct sk_buff *drr_dequeue(struct Qdisc *sch)
 388 {
 389         struct drr_sched *q = qdisc_priv(sch);
 390         struct drr_class *cl;
 391         struct sk_buff *skb;
 392         unsigned int len;
 393
 394         if (list_empty(&q->active))
 395                 goto out;
 396         while (1) {
 397                 cl = list_first_entry(&q->active, struct drr_class, alist);
 398                 skb = cl->qdisc->ops->peek(cl->qdisc);
 399                 if (skb == NULL) {
 400                         qdisc_warn_nonwc(__func__, cl->qdisc);
 401                         goto out;
 402                 }
 403
 404                 len = qdisc_pkt_len(skb);
 405                 if (len <= cl->deficit) {
 406                         cl->deficit -= len;
 407                         skb = qdisc_dequeue_peeked(cl->qdisc);
 408                         if (unlikely(skb == NULL))
 409                                 goto out;
 410                         if (cl->qdisc->q.qlen == 0)
 411                                 list_del(&cl->alist);
 412
 413                         bstats_update(&cl->bstats, skb);
 414                         qdisc_bstats_update(sch, skb);
 415                         qdisc_qstats_backlog_dec(sch, skb);
 416                         sch->q.qlen--;
 417                         return skb;
 418                 }
 419
 420                 cl->deficit += cl->quantum;
 421                 list_move_tail(&cl->alist, &q->active);
 422         }
 423 out:
 424         return NULL;
 425 }
 426
 427 static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
 428 {
 429         struct drr_sched *q = qdisc_priv(sch);
 430         int err;
 431
 432         err = qdisc_class_hash_init(&q->clhash);
 433         if (err < 0)
 434                 return err;
 435         INIT_LIST_HEAD(&q->active);
 436         return 0;
 437 }
 438
 439 static void drr_reset_qdisc(struct Qdisc *sch)
 440 {
 441         struct drr_sched *q = qdisc_priv(sch);
 442         struct drr_class *cl;
 443         unsigned int i;
 444
 445         for (i = 0; i < q->clhash.hashsize; i++) {
 446                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 447                         if (cl->qdisc->q.qlen)
 448                                 list_del(&cl->alist);
 449                         qdisc_reset(cl->qdisc);
 450                 }
 451         }
 452         sch->qstats.backlog = 0;
 453         sch->q.qlen = 0;
 454 }
 455
 456 static void drr_destroy_qdisc(struct Qdisc *sch)
 457 {
 458         struct drr_sched *q = qdisc_priv(sch);
 459         struct drr_class *cl;
 460         struct hlist_node *next;
 461         unsigned int i;
 462
 463         tcf_destroy_chain(&q->filter_list);
 464
 465         for (i = 0; i < q->clhash.hashsize; i++) {
 466                 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
 467                                           common.hnode)
 468                         drr_destroy_class(sch, cl);
 469         }
 470         qdisc_class_hash_destroy(&q->clhash);
 471 }
 472
 473 static const struct Qdisc_class_ops drr_class_ops = {
 474         .change         = drr_change_class,
 475         .delete         = drr_delete_class,
 476         .get            = drr_get_class,
 477         .put            = drr_put_class,
 478         .tcf_chain      = drr_tcf_chain,
 479         .bind_tcf       = drr_bind_tcf,
 480         .unbind_tcf     = drr_unbind_tcf,
 481         .graft          = drr_graft_class,
 482         .leaf           = drr_class_leaf,
 483         .qlen_notify    = drr_qlen_notify,
 484         .dump           = drr_dump_class,
 485         .dump_stats     = drr_dump_class_stats,
 486         .walk           = drr_walk,
 487 };
 488
 489 static struct Qdisc_ops drr_qdisc_ops __read_mostly = {
 490         .cl_ops         = &drr_class_ops,
 491         .id             = "drr",
 492         .priv_size      = sizeof(struct drr_sched),
 493         .enqueue        = drr_enqueue,
 494         .dequeue        = drr_dequeue,
 495         .peek           = qdisc_peek_dequeued,
 496         .init           = drr_init_qdisc,
 497         .reset          = drr_reset_qdisc,
 498         .destroy        = drr_destroy_qdisc,
 499         .owner          = THIS_MODULE,
 500 };
 501
 502 static int __init drr_init(void)
 503 {
 504         return register_qdisc(&drr_qdisc_ops);
 505 }
 506
 507 static void __exit drr_exit(void)
 508 {
 509         unregister_qdisc(&drr_qdisc_ops);
 510 }
 511
 512 module_init(drr_init);
 513 module_exit(drr_exit);
 514 MODULE_LICENSE("GPL");