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                                                     qdisc_root_sleeping_lock(sch),
  95                                                     tca[TCA_RATE]);
  96                         if (err)
  97                                 return err;
  98                 }
  99
 100                 sch_tree_lock(sch);
 101                 if (tb[TCA_DRR_QUANTUM])
 102                         cl->quantum = quantum;
 103                 sch_tree_unlock(sch);
 104
 105                 return 0;
 106         }
 107
 108         cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL);
 109         if (cl == NULL)
 110                 return -ENOBUFS;
 111
 112         cl->refcnt         = 1;
 113         cl->common.classid = classid;
 114         cl->quantum        = quantum;
 115         cl->qdisc          = qdisc_create_dflt(sch->dev_queue,
 116                                                &pfifo_qdisc_ops, classid);
 117         if (cl->qdisc == NULL)
 118                 cl->qdisc = &noop_qdisc;
 119
 120         if (tca[TCA_RATE]) {
 121                 err = gen_replace_estimator(&cl->bstats, NULL, &cl->rate_est,
 122                                             qdisc_root_sleeping_lock(sch),
 123                                             tca[TCA_RATE]);
 124                 if (err) {
 125                         qdisc_destroy(cl->qdisc);
 126                         kfree(cl);
 127                         return err;
 128                 }
 129         }
 130
 131         sch_tree_lock(sch);
 132         qdisc_class_hash_insert(&q->clhash, &cl->common);
 133         sch_tree_unlock(sch);
 134
 135         qdisc_class_hash_grow(sch, &q->clhash);
 136
 137         *arg = (unsigned long)cl;
 138         return 0;
 139 }
 140
 141 static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl)
 142 {
 143         gen_kill_estimator(&cl->bstats, &cl->rate_est);
 144         qdisc_destroy(cl->qdisc);
 145         kfree(cl);
 146 }
 147
 148 static int drr_delete_class(struct Qdisc *sch, unsigned long arg)
 149 {
 150         struct drr_sched *q = qdisc_priv(sch);
 151         struct drr_class *cl = (struct drr_class *)arg;
 152
 153         if (cl->filter_cnt > 0)
 154                 return -EBUSY;
 155
 156         sch_tree_lock(sch);
 157
 158         drr_purge_queue(cl);
 159         qdisc_class_hash_remove(&q->clhash, &cl->common);
 160
 161         BUG_ON(--cl->refcnt == 0);
 162         /*
 163          * This shouldn't happen: we "hold" one cops->get() when called
 164          * from tc_ctl_tclass; the destroy method is done from cops->put().
 165          */
 166
 167         sch_tree_unlock(sch);
 168         return 0;
 169 }
 170
 171 static unsigned long drr_get_class(struct Qdisc *sch, u32 classid)
 172 {
 173         struct drr_class *cl = drr_find_class(sch, classid);
 174
 175         if (cl != NULL)
 176                 cl->refcnt++;
 177
 178         return (unsigned long)cl;
 179 }
 180
 181 static void drr_put_class(struct Qdisc *sch, unsigned long arg)
 182 {
 183         struct drr_class *cl = (struct drr_class *)arg;
 184
 185         if (--cl->refcnt == 0)
 186                 drr_destroy_class(sch, cl);
 187 }
 188
 189 static struct tcf_proto __rcu **drr_tcf_chain(struct Qdisc *sch,
 190                                               unsigned long cl)
 191 {
 192         struct drr_sched *q = qdisc_priv(sch);
 193
 194         if (cl)
 195                 return NULL;
 196
 197         return &q->filter_list;
 198 }
 199
 200 static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent,
 201                                   u32 classid)
 202 {
 203         struct drr_class *cl = drr_find_class(sch, classid);
 204
 205         if (cl != NULL)
 206                 cl->filter_cnt++;
 207
 208         return (unsigned long)cl;
 209 }
 210
 211 static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg)
 212 {
 213         struct drr_class *cl = (struct drr_class *)arg;
 214
 215         cl->filter_cnt--;
 216 }
 217
 218 static int drr_graft_class(struct Qdisc *sch, unsigned long arg,
 219                            struct Qdisc *new, struct Qdisc **old)
 220 {
 221         struct drr_class *cl = (struct drr_class *)arg;
 222
 223         if (new == NULL) {
 224                 new = qdisc_create_dflt(sch->dev_queue,
 225                                         &pfifo_qdisc_ops, cl->common.classid);
 226                 if (new == NULL)
 227                         new = &noop_qdisc;
 228         }
 229
 230         *old = qdisc_replace(sch, new, &cl->qdisc);
 231         return 0;
 232 }
 233
 234 static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg)
 235 {
 236         struct drr_class *cl = (struct drr_class *)arg;
 237
 238         return cl->qdisc;
 239 }
 240
 241 static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg)
 242 {
 243         struct drr_class *cl = (struct drr_class *)arg;
 244
 245         if (cl->qdisc->q.qlen == 0)
 246                 list_del(&cl->alist);
 247 }
 248
 249 static int drr_dump_class(struct Qdisc *sch, unsigned long arg,
 250                           struct sk_buff *skb, struct tcmsg *tcm)
 251 {
 252         struct drr_class *cl = (struct drr_class *)arg;
 253         struct nlattr *nest;
 254
 255         tcm->tcm_parent = TC_H_ROOT;
 256         tcm->tcm_handle = cl->common.classid;
 257         tcm->tcm_info   = cl->qdisc->handle;
 258
 259         nest = nla_nest_start(skb, TCA_OPTIONS);
 260         if (nest == NULL)
 261                 goto nla_put_failure;
 262         if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum))
 263                 goto nla_put_failure;
 264         return nla_nest_end(skb, nest);
 265
 266 nla_put_failure:
 267         nla_nest_cancel(skb, nest);
 268         return -EMSGSIZE;
 269 }
 270
 271 static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg,
 272                                 struct gnet_dump *d)
 273 {
 274         struct drr_class *cl = (struct drr_class *)arg;
 275         __u32 qlen = cl->qdisc->q.qlen;
 276         struct tc_drr_stats xstats;
 277
 278         memset(&xstats, 0, sizeof(xstats));
 279         if (qlen)
 280                 xstats.deficit = cl->deficit;
 281
 282         if (gnet_stats_copy_basic(d, NULL, &cl->bstats) < 0 ||
 283             gnet_stats_copy_rate_est(d, &cl->bstats, &cl->rate_est) < 0 ||
 284             gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0)
 285                 return -1;
 286
 287         return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
 288 }
 289
 290 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 291 {
 292         struct drr_sched *q = qdisc_priv(sch);
 293         struct drr_class *cl;
 294         unsigned int i;
 295
 296         if (arg->stop)
 297                 return;
 298
 299         for (i = 0; i < q->clhash.hashsize; i++) {
 300                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 301                         if (arg->count < arg->skip) {
 302                                 arg->count++;
 303                                 continue;
 304                         }
 305                         if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
 306                                 arg->stop = 1;
 307                                 return;
 308                         }
 309                         arg->count++;
 310                 }
 311         }
 312 }
 313
 314 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch,
 315                                       int *qerr)
 316 {
 317         struct drr_sched *q = qdisc_priv(sch);
 318         struct drr_class *cl;
 319         struct tcf_result res;
 320         struct tcf_proto *fl;
 321         int result;
 322
 323         if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
 324                 cl = drr_find_class(sch, skb->priority);
 325                 if (cl != NULL)
 326                         return cl;
 327         }
 328
 329         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 330         fl = rcu_dereference_bh(q->filter_list);
 331         result = tc_classify(skb, fl, &res, false);
 332         if (result >= 0) {
 333 #ifdef CONFIG_NET_CLS_ACT
 334                 switch (result) {
 335                 case TC_ACT_QUEUED:
 336                 case TC_ACT_STOLEN:
 337                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 338                 case TC_ACT_SHOT:
 339                         return NULL;
 340                 }
 341 #endif
 342                 cl = (struct drr_class *)res.class;
 343                 if (cl == NULL)
 344                         cl = drr_find_class(sch, res.classid);
 345                 return cl;
 346         }
 347         return NULL;
 348 }
 349
 350 static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 351 {
 352         struct drr_sched *q = qdisc_priv(sch);
 353         struct drr_class *cl;
 354         int err = 0;
 355
 356         cl = drr_classify(skb, sch, &err);
 357         if (cl == NULL) {
 358                 if (err & __NET_XMIT_BYPASS)
 359                         qdisc_qstats_drop(sch);
 360                 kfree_skb(skb);
 361                 return err;
 362         }
 363
 364         err = qdisc_enqueue(skb, cl->qdisc);
 365         if (unlikely(err != NET_XMIT_SUCCESS)) {
 366                 if (net_xmit_drop_count(err)) {
 367                         cl->qstats.drops++;
 368                         qdisc_qstats_drop(sch);
 369                 }
 370                 return err;
 371         }
 372
 373         if (cl->qdisc->q.qlen == 1) {
 374                 list_add_tail(&cl->alist, &q->active);
 375                 cl->deficit = cl->quantum;
 376         }
 377
 378         qdisc_qstats_backlog_inc(sch, skb);
 379         sch->q.qlen++;
 380         return err;
 381 }
 382
 383 static struct sk_buff *drr_dequeue(struct Qdisc *sch)
 384 {
 385         struct drr_sched *q = qdisc_priv(sch);
 386         struct drr_class *cl;
 387         struct sk_buff *skb;
 388         unsigned int len;
 389
 390         if (list_empty(&q->active))
 391                 goto out;
 392         while (1) {
 393                 cl = list_first_entry(&q->active, struct drr_class, alist);
 394                 skb = cl->qdisc->ops->peek(cl->qdisc);
 395                 if (skb == NULL) {
 396                         qdisc_warn_nonwc(__func__, cl->qdisc);
 397                         goto out;
 398                 }
 399
 400                 len = qdisc_pkt_len(skb);
 401                 if (len <= cl->deficit) {
 402                         cl->deficit -= len;
 403                         skb = qdisc_dequeue_peeked(cl->qdisc);
 404                         if (unlikely(skb == NULL))
 405                                 goto out;
 406                         if (cl->qdisc->q.qlen == 0)
 407                                 list_del(&cl->alist);
 408
 409                         bstats_update(&cl->bstats, skb);
 410                         qdisc_bstats_update(sch, skb);
 411                         qdisc_qstats_backlog_dec(sch, skb);
 412                         sch->q.qlen--;
 413                         return skb;
 414                 }
 415
 416                 cl->deficit += cl->quantum;
 417                 list_move_tail(&cl->alist, &q->active);
 418         }
 419 out:
 420         return NULL;
 421 }
 422
 423 static unsigned int drr_drop(struct Qdisc *sch)
 424 {
 425         struct drr_sched *q = qdisc_priv(sch);
 426         struct drr_class *cl;
 427         unsigned int len;
 428
 429         list_for_each_entry(cl, &q->active, alist) {
 430                 if (cl->qdisc->ops->drop) {
 431                         len = cl->qdisc->ops->drop(cl->qdisc);
 432                         if (len > 0) {
 433                                 sch->qstats.backlog -= len;
 434                                 sch->q.qlen--;
 435                                 if (cl->qdisc->q.qlen == 0)
 436                                         list_del(&cl->alist);
 437                                 return len;
 438                         }
 439                 }
 440         }
 441         return 0;
 442 }
 443
 444 static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
 445 {
 446         struct drr_sched *q = qdisc_priv(sch);
 447         int err;
 448
 449         err = qdisc_class_hash_init(&q->clhash);
 450         if (err < 0)
 451                 return err;
 452         INIT_LIST_HEAD(&q->active);
 453         return 0;
 454 }
 455
 456 static void drr_reset_qdisc(struct Qdisc *sch)
 457 {
 458         struct drr_sched *q = qdisc_priv(sch);
 459         struct drr_class *cl;
 460         unsigned int i;
 461
 462         for (i = 0; i < q->clhash.hashsize; i++) {
 463                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 464                         if (cl->qdisc->q.qlen)
 465                                 list_del(&cl->alist);
 466                         qdisc_reset(cl->qdisc);
 467                 }
 468         }
 469         sch->qstats.backlog = 0;
 470         sch->q.qlen = 0;
 471 }
 472
 473 static void drr_destroy_qdisc(struct Qdisc *sch)
 474 {
 475         struct drr_sched *q = qdisc_priv(sch);
 476         struct drr_class *cl;
 477         struct hlist_node *next;
 478         unsigned int i;
 479
 480         tcf_destroy_chain(&q->filter_list);
 481
 482         for (i = 0; i < q->clhash.hashsize; i++) {
 483                 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
 484                                           common.hnode)
 485                         drr_destroy_class(sch, cl);
 486         }
 487         qdisc_class_hash_destroy(&q->clhash);
 488 }
 489
 490 static const struct Qdisc_class_ops drr_class_ops = {
 491         .change         = drr_change_class,
 492         .delete         = drr_delete_class,
 493         .get            = drr_get_class,
 494         .put            = drr_put_class,
 495         .tcf_chain      = drr_tcf_chain,
 496         .bind_tcf       = drr_bind_tcf,
 497         .unbind_tcf     = drr_unbind_tcf,
 498         .graft          = drr_graft_class,
 499         .leaf           = drr_class_leaf,
 500         .qlen_notify    = drr_qlen_notify,
 501         .dump           = drr_dump_class,
 502         .dump_stats     = drr_dump_class_stats,
 503         .walk           = drr_walk,
 504 };
 505
 506 static struct Qdisc_ops drr_qdisc_ops __read_mostly = {
 507         .cl_ops         = &drr_class_ops,
 508         .id             = "drr",
 509         .priv_size      = sizeof(struct drr_sched),
 510         .enqueue        = drr_enqueue,
 511         .dequeue        = drr_dequeue,
 512         .peek           = qdisc_peek_dequeued,
 513         .drop           = drr_drop,
 514         .init           = drr_init_qdisc,
 515         .reset          = drr_reset_qdisc,
 516         .destroy        = drr_destroy_qdisc,
 517         .owner          = THIS_MODULE,
 518 };
 519
 520 static int __init drr_init(void)
 521 {
 522         return register_qdisc(&drr_qdisc_ops);
 523 }
 524
 525 static void __exit drr_exit(void)
 526 {
 527         unregister_qdisc(&drr_qdisc_ops);
 528 }
 529
 530 module_init(drr_init);
 531 module_exit(drr_exit);
 532 MODULE_LICENSE("GPL");