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
  57         qdisc_reset(cl->qdisc);
  58         qdisc_tree_decrease_qlen(cl->qdisc, len);
  59 }
  60
  61 static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = {
  62         [TCA_DRR_QUANTUM]       = { .type = NLA_U32 },
  63 };
  64
  65 static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
  66                             struct nlattr **tca, unsigned long *arg)
  67 {
  68         struct drr_sched *q = qdisc_priv(sch);
  69         struct drr_class *cl = (struct drr_class *)*arg;
  70         struct nlattr *opt = tca[TCA_OPTIONS];
  71         struct nlattr *tb[TCA_DRR_MAX + 1];
  72         u32 quantum;
  73         int err;
  74
  75         if (!opt)
  76                 return -EINVAL;
  77
  78         err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy);
  79         if (err < 0)
  80                 return err;
  81
  82         if (tb[TCA_DRR_QUANTUM]) {
  83                 quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]);
  84                 if (quantum == 0)
  85                         return -EINVAL;
  86         } else
  87                 quantum = psched_mtu(qdisc_dev(sch));
  88
  89         if (cl != NULL) {
  90                 if (tca[TCA_RATE]) {
  91                         err = gen_replace_estimator(&cl->bstats, NULL,
  92                                                     &cl->rate_est,
  93                                                     qdisc_root_sleeping_lock(sch),
  94                                                     tca[TCA_RATE]);
  95                         if (err)
  96                                 return err;
  97                 }
  98
  99                 sch_tree_lock(sch);
 100                 if (tb[TCA_DRR_QUANTUM])
 101                         cl->quantum = quantum;
 102                 sch_tree_unlock(sch);
 103
 104                 return 0;
 105         }
 106
 107         cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL);
 108         if (cl == NULL)
 109                 return -ENOBUFS;
 110
 111         cl->refcnt         = 1;
 112         cl->common.classid = classid;
 113         cl->quantum        = quantum;
 114         cl->qdisc          = qdisc_create_dflt(sch->dev_queue,
 115                                                &pfifo_qdisc_ops, classid);
 116         if (cl->qdisc == NULL)
 117                 cl->qdisc = &noop_qdisc;
 118
 119         if (tca[TCA_RATE]) {
 120                 err = gen_replace_estimator(&cl->bstats, NULL, &cl->rate_est,
 121                                             qdisc_root_sleeping_lock(sch),
 122                                             tca[TCA_RATE]);
 123                 if (err) {
 124                         qdisc_destroy(cl->qdisc);
 125                         kfree(cl);
 126                         return err;
 127                 }
 128         }
 129
 130         sch_tree_lock(sch);
 131         qdisc_class_hash_insert(&q->clhash, &cl->common);
 132         sch_tree_unlock(sch);
 133
 134         qdisc_class_hash_grow(sch, &q->clhash);
 135
 136         *arg = (unsigned long)cl;
 137         return 0;
 138 }
 139
 140 static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl)
 141 {
 142         gen_kill_estimator(&cl->bstats, &cl->rate_est);
 143         qdisc_destroy(cl->qdisc);
 144         kfree(cl);
 145 }
 146
 147 static int drr_delete_class(struct Qdisc *sch, unsigned long arg)
 148 {
 149         struct drr_sched *q = qdisc_priv(sch);
 150         struct drr_class *cl = (struct drr_class *)arg;
 151
 152         if (cl->filter_cnt > 0)
 153                 return -EBUSY;
 154
 155         sch_tree_lock(sch);
 156
 157         drr_purge_queue(cl);
 158         qdisc_class_hash_remove(&q->clhash, &cl->common);
 159
 160         BUG_ON(--cl->refcnt == 0);
 161         /*
 162          * This shouldn't happen: we "hold" one cops->get() when called
 163          * from tc_ctl_tclass; the destroy method is done from cops->put().
 164          */
 165
 166         sch_tree_unlock(sch);
 167         return 0;
 168 }
 169
 170 static unsigned long drr_get_class(struct Qdisc *sch, u32 classid)
 171 {
 172         struct drr_class *cl = drr_find_class(sch, classid);
 173
 174         if (cl != NULL)
 175                 cl->refcnt++;
 176
 177         return (unsigned long)cl;
 178 }
 179
 180 static void drr_put_class(struct Qdisc *sch, unsigned long arg)
 181 {
 182         struct drr_class *cl = (struct drr_class *)arg;
 183
 184         if (--cl->refcnt == 0)
 185                 drr_destroy_class(sch, cl);
 186 }
 187
 188 static struct tcf_proto __rcu **drr_tcf_chain(struct Qdisc *sch,
 189                                               unsigned long cl)
 190 {
 191         struct drr_sched *q = qdisc_priv(sch);
 192
 193         if (cl)
 194                 return NULL;
 195
 196         return &q->filter_list;
 197 }
 198
 199 static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent,
 200                                   u32 classid)
 201 {
 202         struct drr_class *cl = drr_find_class(sch, classid);
 203
 204         if (cl != NULL)
 205                 cl->filter_cnt++;
 206
 207         return (unsigned long)cl;
 208 }
 209
 210 static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg)
 211 {
 212         struct drr_class *cl = (struct drr_class *)arg;
 213
 214         cl->filter_cnt--;
 215 }
 216
 217 static int drr_graft_class(struct Qdisc *sch, unsigned long arg,
 218                            struct Qdisc *new, struct Qdisc **old)
 219 {
 220         struct drr_class *cl = (struct drr_class *)arg;
 221
 222         if (new == NULL) {
 223                 new = qdisc_create_dflt(sch->dev_queue,
 224                                         &pfifo_qdisc_ops, cl->common.classid);
 225                 if (new == NULL)
 226                         new = &noop_qdisc;
 227         }
 228
 229         *old = qdisc_replace(sch, new, &cl->qdisc);
 230         return 0;
 231 }
 232
 233 static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg)
 234 {
 235         struct drr_class *cl = (struct drr_class *)arg;
 236
 237         return cl->qdisc;
 238 }
 239
 240 static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg)
 241 {
 242         struct drr_class *cl = (struct drr_class *)arg;
 243
 244         if (cl->qdisc->q.qlen == 0)
 245                 list_del(&cl->alist);
 246 }
 247
 248 static int drr_dump_class(struct Qdisc *sch, unsigned long arg,
 249                           struct sk_buff *skb, struct tcmsg *tcm)
 250 {
 251         struct drr_class *cl = (struct drr_class *)arg;
 252         struct nlattr *nest;
 253
 254         tcm->tcm_parent = TC_H_ROOT;
 255         tcm->tcm_handle = cl->common.classid;
 256         tcm->tcm_info   = cl->qdisc->handle;
 257
 258         nest = nla_nest_start(skb, TCA_OPTIONS);
 259         if (nest == NULL)
 260                 goto nla_put_failure;
 261         if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum))
 262                 goto nla_put_failure;
 263         return nla_nest_end(skb, nest);
 264
 265 nla_put_failure:
 266         nla_nest_cancel(skb, nest);
 267         return -EMSGSIZE;
 268 }
 269
 270 static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg,
 271                                 struct gnet_dump *d)
 272 {
 273         struct drr_class *cl = (struct drr_class *)arg;
 274         __u32 qlen = cl->qdisc->q.qlen;
 275         struct tc_drr_stats xstats;
 276
 277         memset(&xstats, 0, sizeof(xstats));
 278         if (qlen)
 279                 xstats.deficit = cl->deficit;
 280
 281         if (gnet_stats_copy_basic(d, NULL, &cl->bstats) < 0 ||
 282             gnet_stats_copy_rate_est(d, &cl->bstats, &cl->rate_est) < 0 ||
 283             gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0)
 284                 return -1;
 285
 286         return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
 287 }
 288
 289 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 290 {
 291         struct drr_sched *q = qdisc_priv(sch);
 292         struct drr_class *cl;
 293         unsigned int i;
 294
 295         if (arg->stop)
 296                 return;
 297
 298         for (i = 0; i < q->clhash.hashsize; i++) {
 299                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 300                         if (arg->count < arg->skip) {
 301                                 arg->count++;
 302                                 continue;
 303                         }
 304                         if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
 305                                 arg->stop = 1;
 306                                 return;
 307                         }
 308                         arg->count++;
 309                 }
 310         }
 311 }
 312
 313 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch,
 314                                       int *qerr)
 315 {
 316         struct drr_sched *q = qdisc_priv(sch);
 317         struct drr_class *cl;
 318         struct tcf_result res;
 319         struct tcf_proto *fl;
 320         int result;
 321
 322         if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
 323                 cl = drr_find_class(sch, skb->priority);
 324                 if (cl != NULL)
 325                         return cl;
 326         }
 327
 328         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 329         fl = rcu_dereference_bh(q->filter_list);
 330         result = tc_classify(skb, fl, &res, false);
 331         if (result >= 0) {
 332 #ifdef CONFIG_NET_CLS_ACT
 333                 switch (result) {
 334                 case TC_ACT_QUEUED:
 335                 case TC_ACT_STOLEN:
 336                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 337                 case TC_ACT_SHOT:
 338                         return NULL;
 339                 }
 340 #endif
 341                 cl = (struct drr_class *)res.class;
 342                 if (cl == NULL)
 343                         cl = drr_find_class(sch, res.classid);
 344                 return cl;
 345         }
 346         return NULL;
 347 }
 348
 349 static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 350 {
 351         struct drr_sched *q = qdisc_priv(sch);
 352         struct drr_class *cl;
 353         int err = 0;
 354
 355         cl = drr_classify(skb, sch, &err);
 356         if (cl == NULL) {
 357                 if (err & __NET_XMIT_BYPASS)
 358                         qdisc_qstats_drop(sch);
 359                 kfree_skb(skb);
 360                 return err;
 361         }
 362
 363         err = qdisc_enqueue(skb, cl->qdisc);
 364         if (unlikely(err != NET_XMIT_SUCCESS)) {
 365                 if (net_xmit_drop_count(err)) {
 366                         cl->qstats.drops++;
 367                         qdisc_qstats_drop(sch);
 368                 }
 369                 return err;
 370         }
 371
 372         if (cl->qdisc->q.qlen == 1) {
 373                 list_add_tail(&cl->alist, &q->active);
 374                 cl->deficit = cl->quantum;
 375         }
 376
 377         sch->q.qlen++;
 378         return err;
 379 }
 380
 381 static struct sk_buff *drr_dequeue(struct Qdisc *sch)
 382 {
 383         struct drr_sched *q = qdisc_priv(sch);
 384         struct drr_class *cl;
 385         struct sk_buff *skb;
 386         unsigned int len;
 387
 388         if (list_empty(&q->active))
 389                 goto out;
 390         while (1) {
 391                 cl = list_first_entry(&q->active, struct drr_class, alist);
 392                 skb = cl->qdisc->ops->peek(cl->qdisc);
 393                 if (skb == NULL) {
 394                         qdisc_warn_nonwc(__func__, cl->qdisc);
 395                         goto out;
 396                 }
 397
 398                 len = qdisc_pkt_len(skb);
 399                 if (len <= cl->deficit) {
 400                         cl->deficit -= len;
 401                         skb = qdisc_dequeue_peeked(cl->qdisc);
 402                         if (unlikely(skb == NULL))
 403                                 goto out;
 404                         if (cl->qdisc->q.qlen == 0)
 405                                 list_del(&cl->alist);
 406
 407                         bstats_update(&cl->bstats, skb);
 408                         qdisc_bstats_update(sch, skb);
 409                         sch->q.qlen--;
 410                         return skb;
 411                 }
 412
 413                 cl->deficit += cl->quantum;
 414                 list_move_tail(&cl->alist, &q->active);
 415         }
 416 out:
 417         return NULL;
 418 }
 419
 420 static unsigned int drr_drop(struct Qdisc *sch)
 421 {
 422         struct drr_sched *q = qdisc_priv(sch);
 423         struct drr_class *cl;
 424         unsigned int len;
 425
 426         list_for_each_entry(cl, &q->active, alist) {
 427                 if (cl->qdisc->ops->drop) {
 428                         len = cl->qdisc->ops->drop(cl->qdisc);
 429                         if (len > 0) {
 430                                 sch->q.qlen--;
 431                                 if (cl->qdisc->q.qlen == 0)
 432                                         list_del(&cl->alist);
 433                                 return len;
 434                         }
 435                 }
 436         }
 437         return 0;
 438 }
 439
 440 static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
 441 {
 442         struct drr_sched *q = qdisc_priv(sch);
 443         int err;
 444
 445         err = qdisc_class_hash_init(&q->clhash);
 446         if (err < 0)
 447                 return err;
 448         INIT_LIST_HEAD(&q->active);
 449         return 0;
 450 }
 451
 452 static void drr_reset_qdisc(struct Qdisc *sch)
 453 {
 454         struct drr_sched *q = qdisc_priv(sch);
 455         struct drr_class *cl;
 456         unsigned int i;
 457
 458         for (i = 0; i < q->clhash.hashsize; i++) {
 459                 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
 460                         if (cl->qdisc->q.qlen)
 461                                 list_del(&cl->alist);
 462                         qdisc_reset(cl->qdisc);
 463                 }
 464         }
 465         sch->q.qlen = 0;
 466 }
 467
 468 static void drr_destroy_qdisc(struct Qdisc *sch)
 469 {
 470         struct drr_sched *q = qdisc_priv(sch);
 471         struct drr_class *cl;
 472         struct hlist_node *next;
 473         unsigned int i;
 474
 475         tcf_destroy_chain(&q->filter_list);
 476
 477         for (i = 0; i < q->clhash.hashsize; i++) {
 478                 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
 479                                           common.hnode)
 480                         drr_destroy_class(sch, cl);
 481         }
 482         qdisc_class_hash_destroy(&q->clhash);
 483 }
 484
 485 static const struct Qdisc_class_ops drr_class_ops = {
 486         .change         = drr_change_class,
 487         .delete         = drr_delete_class,
 488         .get            = drr_get_class,
 489         .put            = drr_put_class,
 490         .tcf_chain      = drr_tcf_chain,
 491         .bind_tcf       = drr_bind_tcf,
 492         .unbind_tcf     = drr_unbind_tcf,
 493         .graft          = drr_graft_class,
 494         .leaf           = drr_class_leaf,
 495         .qlen_notify    = drr_qlen_notify,
 496         .dump           = drr_dump_class,
 497         .dump_stats     = drr_dump_class_stats,
 498         .walk           = drr_walk,
 499 };
 500
 501 static struct Qdisc_ops drr_qdisc_ops __read_mostly = {
 502         .cl_ops         = &drr_class_ops,
 503         .id             = "drr",
 504         .priv_size      = sizeof(struct drr_sched),
 505         .enqueue        = drr_enqueue,
 506         .dequeue        = drr_dequeue,
 507         .peek           = qdisc_peek_dequeued,
 508         .drop           = drr_drop,
 509         .init           = drr_init_qdisc,
 510         .reset          = drr_reset_qdisc,
 511         .destroy        = drr_destroy_qdisc,
 512         .owner          = THIS_MODULE,
 513 };
 514
 515 static int __init drr_init(void)
 516 {
 517         return register_qdisc(&drr_qdisc_ops);
 518 }
 519
 520 static void __exit drr_exit(void)
 521 {
 522         unregister_qdisc(&drr_qdisc_ops);
 523 }
 524
 525 module_init(drr_init);
 526 module_exit(drr_exit);
 527 MODULE_LICENSE("GPL");