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         if (cl->qdisc->q.qlen == 0)
 250                 list_del(&cl->alist);
 251 }
 252
 253 static int drr_dump_class(struct Qdisc *sch, unsigned long arg,
 254                           struct sk_buff *skb, struct tcmsg *tcm)
 255 {
 256         struct drr_class *cl = (struct drr_class *)arg;
 257         struct nlattr *nest;
 258
 259         tcm->tcm_parent = TC_H_ROOT;
 260         tcm->tcm_handle = cl->common.classid;
 261         tcm->tcm_info   = cl->qdisc->handle;
 262
 263         nest = nla_nest_start(skb, TCA_OPTIONS);
 264         if (nest == NULL)
 265                 goto nla_put_failure;
 266         if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum))
 267                 goto nla_put_failure;
 268         return nla_nest_end(skb, nest);
 269
 270 nla_put_failure:
 271         nla_nest_cancel(skb, nest);
 272         return -EMSGSIZE;
 273 }
 274
 275 static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg,
 276                                 struct gnet_dump *d)
 277 {
 278         struct drr_class *cl = (struct drr_class *)arg;
 279         __u32 qlen = cl->qdisc->q.qlen;
 280         struct tc_drr_stats xstats;
 281
 282         memset(&xstats, 0, sizeof(xstats));
 283         if (qlen)
 284                 xstats.deficit = cl->deficit;
 285
 286         if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
 287                                   d, NULL, &cl->bstats) < 0 ||
 288             gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
 289             gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0)
 290                 return -1;
 291
 292         return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
 293 }
 294
 295 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 296 {
 297         struct drr_sched *q = qdisc_priv(sch);
 298         struct drr_class *cl;
 299         unsigned int i;
 300
 301         if (arg->stop)
 302                 return;
 303
 304         for (i = 0; i < q->clhash.hashsize; i++) {
 305                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 306                         if (arg->count < arg->skip) {
 307                                 arg->count++;
 308                                 continue;
 309                         }
 310                         if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
 311                                 arg->stop = 1;
 312                                 return;
 313                         }
 314                         arg->count++;
 315                 }
 316         }
 317 }
 318
 319 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch,
 320                                       int *qerr)
 321 {
 322         struct drr_sched *q = qdisc_priv(sch);
 323         struct drr_class *cl;
 324         struct tcf_result res;
 325         struct tcf_proto *fl;
 326         int result;
 327
 328         if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
 329                 cl = drr_find_class(sch, skb->priority);
 330                 if (cl != NULL)
 331                         return cl;
 332         }
 333
 334         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 335         fl = rcu_dereference_bh(q->filter_list);
 336         result = tcf_classify(skb, fl, &res, false);
 337         if (result >= 0) {
 338 #ifdef CONFIG_NET_CLS_ACT
 339                 switch (result) {
 340                 case TC_ACT_QUEUED:
 341                 case TC_ACT_STOLEN:
 342                 case TC_ACT_TRAP:
 343                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 344                 case TC_ACT_SHOT:
 345                         return NULL;
 346                 }
 347 #endif
 348                 cl = (struct drr_class *)res.class;
 349                 if (cl == NULL)
 350                         cl = drr_find_class(sch, res.classid);
 351                 return cl;
 352         }
 353         return NULL;
 354 }
 355
 356 static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 357                        struct sk_buff **to_free)
 358 {
 359         struct drr_sched *q = qdisc_priv(sch);
 360         struct drr_class *cl;
 361         int err = 0;
 362
 363         cl = drr_classify(skb, sch, &err);
 364         if (cl == NULL) {
 365                 if (err & __NET_XMIT_BYPASS)
 366                         qdisc_qstats_drop(sch);
 367                 __qdisc_drop(skb, to_free);
 368                 return err;
 369         }
 370
 371         err = qdisc_enqueue(skb, cl->qdisc, to_free);
 372         if (unlikely(err != NET_XMIT_SUCCESS)) {
 373                 if (net_xmit_drop_count(err)) {
 374                         cl->qstats.drops++;
 375                         qdisc_qstats_drop(sch);
 376                 }
 377                 return err;
 378         }
 379
 380         if (cl->qdisc->q.qlen == 1) {
 381                 list_add_tail(&cl->alist, &q->active);
 382                 cl->deficit = cl->quantum;
 383         }
 384
 385         qdisc_qstats_backlog_inc(sch, skb);
 386         sch->q.qlen++;
 387         return err;
 388 }
 389
 390 static struct sk_buff *drr_dequeue(struct Qdisc *sch)
 391 {
 392         struct drr_sched *q = qdisc_priv(sch);
 393         struct drr_class *cl;
 394         struct sk_buff *skb;
 395         unsigned int len;
 396
 397         if (list_empty(&q->active))
 398                 goto out;
 399         while (1) {
 400                 cl = list_first_entry(&q->active, struct drr_class, alist);
 401                 skb = cl->qdisc->ops->peek(cl->qdisc);
 402                 if (skb == NULL) {
 403                         qdisc_warn_nonwc(__func__, cl->qdisc);
 404                         goto out;
 405                 }
 406
 407                 len = qdisc_pkt_len(skb);
 408                 if (len <= cl->deficit) {
 409                         cl->deficit -= len;
 410                         skb = qdisc_dequeue_peeked(cl->qdisc);
 411                         if (unlikely(skb == NULL))
 412                                 goto out;
 413                         if (cl->qdisc->q.qlen == 0)
 414                                 list_del(&cl->alist);
 415
 416                         bstats_update(&cl->bstats, skb);
 417                         qdisc_bstats_update(sch, skb);
 418                         qdisc_qstats_backlog_dec(sch, skb);
 419                         sch->q.qlen--;
 420                         return skb;
 421                 }
 422
 423                 cl->deficit += cl->quantum;
 424                 list_move_tail(&cl->alist, &q->active);
 425         }
 426 out:
 427         return NULL;
 428 }
 429
 430 static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
 431 {
 432         struct drr_sched *q = qdisc_priv(sch);
 433         int err;
 434
 435         err = tcf_block_get(&q->block, &q->filter_list);
 436         if (err)
 437                 return err;
 438         err = qdisc_class_hash_init(&q->clhash);
 439         if (err < 0)
 440                 return err;
 441         INIT_LIST_HEAD(&q->active);
 442         return 0;
 443 }
 444
 445 static void drr_reset_qdisc(struct Qdisc *sch)
 446 {
 447         struct drr_sched *q = qdisc_priv(sch);
 448         struct drr_class *cl;
 449         unsigned int i;
 450
 451         for (i = 0; i < q->clhash.hashsize; i++) {
 452                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 453                         if (cl->qdisc->q.qlen)
 454                                 list_del(&cl->alist);
 455                         qdisc_reset(cl->qdisc);
 456                 }
 457         }
 458         sch->qstats.backlog = 0;
 459         sch->q.qlen = 0;
 460 }
 461
 462 static void drr_destroy_qdisc(struct Qdisc *sch)
 463 {
 464         struct drr_sched *q = qdisc_priv(sch);
 465         struct drr_class *cl;
 466         struct hlist_node *next;
 467         unsigned int i;
 468
 469         tcf_block_put(q->block);
 470
 471         for (i = 0; i < q->clhash.hashsize; i++) {
 472                 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
 473                                           common.hnode)
 474                         drr_destroy_class(sch, cl);
 475         }
 476         qdisc_class_hash_destroy(&q->clhash);
 477 }
 478
 479 static const struct Qdisc_class_ops drr_class_ops = {
 480         .change         = drr_change_class,
 481         .delete         = drr_delete_class,
 482         .get            = drr_get_class,
 483         .put            = drr_put_class,
 484         .tcf_block      = drr_tcf_block,
 485         .bind_tcf       = drr_bind_tcf,
 486         .unbind_tcf     = drr_unbind_tcf,
 487         .graft          = drr_graft_class,
 488         .leaf           = drr_class_leaf,
 489         .qlen_notify    = drr_qlen_notify,
 490         .dump           = drr_dump_class,
 491         .dump_stats     = drr_dump_class_stats,
 492         .walk           = drr_walk,
 493 };
 494
 495 static struct Qdisc_ops drr_qdisc_ops __read_mostly = {
 496         .cl_ops         = &drr_class_ops,
 497         .id             = "drr",
 498         .priv_size      = sizeof(struct drr_sched),
 499         .enqueue        = drr_enqueue,
 500         .dequeue        = drr_dequeue,
 501         .peek           = qdisc_peek_dequeued,
 502         .init           = drr_init_qdisc,
 503         .reset          = drr_reset_qdisc,
 504         .destroy        = drr_destroy_qdisc,
 505         .owner          = THIS_MODULE,
 506 };
 507
 508 static int __init drr_init(void)
 509 {
 510         return register_qdisc(&drr_qdisc_ops);
 511 }
 512
 513 static void __exit drr_exit(void)
 514 {
 515         unregister_qdisc(&drr_qdisc_ops);
 516 }
 517
 518 module_init(drr_init);
 519 module_exit(drr_exit);
 520 MODULE_LICENSE("GPL");