net/sched/sch_prio.c

   1 /*
   2  * net/sched/sch_prio.c Simple 3-band priority "scheduler".
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              as published by the Free Software Foundation; either version
   7  *              2 of the License, or (at your option) any later version.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  * Fixes:       19990609: J Hadi Salim <hadi@nortelnetworks.com>:
  11  *              Init --  EINVAL when opt undefined
  12  */
  13
  14 #include <linux/module.h>
  15 #include <linux/slab.h>
  16 #include <linux/types.h>
  17 #include <linux/kernel.h>
  18 #include <linux/string.h>
  19 #include <linux/errno.h>
  20 #include <linux/skbuff.h>
  21 #include <net/netlink.h>
  22 #include <net/pkt_sched.h>
  23 #include <net/pkt_cls.h>
  24
  25 struct prio_sched_data {
  26         int bands;
  27         struct tcf_proto __rcu *filter_list;
  28         struct tcf_block *block;
  29         u8  prio2band[TC_PRIO_MAX+1];
  30         struct Qdisc *queues[TCQ_PRIO_BANDS];
  31 };
  32
  33
  34 static struct Qdisc *
  35 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
  36 {
  37         struct prio_sched_data *q = qdisc_priv(sch);
  38         u32 band = skb->priority;
  39         struct tcf_result res;
  40         struct tcf_proto *fl;
  41         int err;
  42
  43         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
  44         if (TC_H_MAJ(skb->priority) != sch->handle) {
  45                 fl = rcu_dereference_bh(q->filter_list);
  46                 err = tcf_classify(skb, fl, &res, false);
  47 #ifdef CONFIG_NET_CLS_ACT
  48                 switch (err) {
  49                 case TC_ACT_STOLEN:
  50                 case TC_ACT_QUEUED:
  51                 case TC_ACT_TRAP:
  52                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
  53                         /* fall through */
  54                 case TC_ACT_SHOT:
  55                         return NULL;
  56                 }
  57 #endif
  58                 if (!fl || err < 0) {
  59                         if (TC_H_MAJ(band))
  60                                 band = 0;
  61                         return q->queues[q->prio2band[band & TC_PRIO_MAX]];
  62                 }
  63                 band = res.classid;
  64         }
  65         band = TC_H_MIN(band) - 1;
  66         if (band >= q->bands)
  67                 return q->queues[q->prio2band[0]];
  68
  69         return q->queues[band];
  70 }
  71
  72 static int
  73 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free)
  74 {
  75         struct Qdisc *qdisc;
  76         int ret;
  77
  78         qdisc = prio_classify(skb, sch, &ret);
  79 #ifdef CONFIG_NET_CLS_ACT
  80         if (qdisc == NULL) {
  81
  82                 if (ret & __NET_XMIT_BYPASS)
  83                         qdisc_qstats_drop(sch);
  84                 __qdisc_drop(skb, to_free);
  85                 return ret;
  86         }
  87 #endif
  88
  89         ret = qdisc_enqueue(skb, qdisc, to_free);
  90         if (ret == NET_XMIT_SUCCESS) {
  91                 qdisc_qstats_backlog_inc(sch, skb);
  92                 sch->q.qlen++;
  93                 return NET_XMIT_SUCCESS;
  94         }
  95         if (net_xmit_drop_count(ret))
  96                 qdisc_qstats_drop(sch);
  97         return ret;
  98 }
  99
 100 static struct sk_buff *prio_peek(struct Qdisc *sch)
 101 {
 102         struct prio_sched_data *q = qdisc_priv(sch);
 103         int prio;
 104
 105         for (prio = 0; prio < q->bands; prio++) {
 106                 struct Qdisc *qdisc = q->queues[prio];
 107                 struct sk_buff *skb = qdisc->ops->peek(qdisc);
 108                 if (skb)
 109                         return skb;
 110         }
 111         return NULL;
 112 }
 113
 114 static struct sk_buff *prio_dequeue(struct Qdisc *sch)
 115 {
 116         struct prio_sched_data *q = qdisc_priv(sch);
 117         int prio;
 118
 119         for (prio = 0; prio < q->bands; prio++) {
 120                 struct Qdisc *qdisc = q->queues[prio];
 121                 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
 122                 if (skb) {
 123                         qdisc_bstats_update(sch, skb);
 124                         qdisc_qstats_backlog_dec(sch, skb);
 125                         sch->q.qlen--;
 126                         return skb;
 127                 }
 128         }
 129         return NULL;
 130
 131 }
 132
 133 static void
 134 prio_reset(struct Qdisc *sch)
 135 {
 136         int prio;
 137         struct prio_sched_data *q = qdisc_priv(sch);
 138
 139         for (prio = 0; prio < q->bands; prio++)
 140                 qdisc_reset(q->queues[prio]);
 141         sch->qstats.backlog = 0;
 142         sch->q.qlen = 0;
 143 }
 144
 145 static int prio_offload(struct Qdisc *sch, bool enable)
 146 {
 147         struct prio_sched_data *q = qdisc_priv(sch);
 148         struct net_device *dev = qdisc_dev(sch);
 149         struct tc_prio_qopt_offload opt = {
 150                 .handle = sch->handle,
 151                 .parent = sch->parent,
 152         };
 153
 154         if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
 155                 return -EOPNOTSUPP;
 156
 157         if (enable) {
 158                 opt.command = TC_PRIO_REPLACE;
 159                 opt.replace_params.bands = q->bands;
 160                 memcpy(&opt.replace_params.priomap, q->prio2band,
 161                        TC_PRIO_MAX + 1);
 162                 opt.replace_params.qstats = &sch->qstats;
 163         } else {
 164                 opt.command = TC_PRIO_DESTROY;
 165         }
 166
 167         return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_PRIO, &opt);
 168 }
 169
 170 static void
 171 prio_destroy(struct Qdisc *sch)
 172 {
 173         int prio;
 174         struct prio_sched_data *q = qdisc_priv(sch);
 175
 176         tcf_block_put(q->block);
 177         prio_offload(sch, false);
 178         for (prio = 0; prio < q->bands; prio++)
 179                 qdisc_destroy(q->queues[prio]);
 180 }
 181
 182 static int prio_tune(struct Qdisc *sch, struct nlattr *opt,
 183                      struct netlink_ext_ack *extack)
 184 {
 185         struct prio_sched_data *q = qdisc_priv(sch);
 186         struct Qdisc *queues[TCQ_PRIO_BANDS];
 187         int oldbands = q->bands, i;
 188         struct tc_prio_qopt *qopt;
 189
 190         if (nla_len(opt) < sizeof(*qopt))
 191                 return -EINVAL;
 192         qopt = nla_data(opt);
 193
 194         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
 195                 return -EINVAL;
 196
 197         for (i = 0; i <= TC_PRIO_MAX; i++) {
 198                 if (qopt->priomap[i] >= qopt->bands)
 199                         return -EINVAL;
 200         }
 201
 202         /* Before commit, make sure we can allocate all new qdiscs */
 203         for (i = oldbands; i < qopt->bands; i++) {
 204                 queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
 205                                               TC_H_MAKE(sch->handle, i + 1),
 206                                               extack);
 207                 if (!queues[i]) {
 208                         while (i > oldbands)
 209                                 qdisc_destroy(queues[--i]);
 210                         return -ENOMEM;
 211                 }
 212         }
 213
 214         sch_tree_lock(sch);
 215         q->bands = qopt->bands;
 216         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
 217
 218         for (i = q->bands; i < oldbands; i++) {
 219                 struct Qdisc *child = q->queues[i];
 220
 221                 qdisc_tree_reduce_backlog(child, child->q.qlen,
 222                                           child->qstats.backlog);
 223                 qdisc_destroy(child);
 224         }
 225
 226         for (i = oldbands; i < q->bands; i++) {
 227                 q->queues[i] = queues[i];
 228                 if (q->queues[i] != &noop_qdisc)
 229                         qdisc_hash_add(q->queues[i], true);
 230         }
 231
 232         sch_tree_unlock(sch);
 233         prio_offload(sch, true);
 234         return 0;
 235 }
 236
 237 static int prio_init(struct Qdisc *sch, struct nlattr *opt,
 238                      struct netlink_ext_ack *extack)
 239 {
 240         struct prio_sched_data *q = qdisc_priv(sch);
 241         int err;
 242
 243         if (!opt)
 244                 return -EINVAL;
 245
 246         err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
 247         if (err)
 248                 return err;
 249
 250         return prio_tune(sch, opt, extack);
 251 }
 252
 253 static int prio_dump_offload(struct Qdisc *sch)
 254 {
 255         struct net_device *dev = qdisc_dev(sch);
 256         struct tc_prio_qopt_offload hw_stats = {
 257                 .command = TC_PRIO_STATS,
 258                 .handle = sch->handle,
 259                 .parent = sch->parent,
 260                 {
 261                         .stats = {
 262                                 .bstats = &sch->bstats,
 263                                 .qstats = &sch->qstats,
 264                         },
 265                 },
 266         };
 267         int err;
 268
 269         sch->flags &= ~TCQ_F_OFFLOADED;
 270         if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
 271                 return 0;
 272
 273         err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_PRIO,
 274                                             &hw_stats);
 275         if (err == -EOPNOTSUPP)
 276                 return 0;
 277
 278         if (!err)
 279                 sch->flags |= TCQ_F_OFFLOADED;
 280
 281         return err;
 282 }
 283
 284 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
 285 {
 286         struct prio_sched_data *q = qdisc_priv(sch);
 287         unsigned char *b = skb_tail_pointer(skb);
 288         struct tc_prio_qopt opt;
 289         int err;
 290
 291         opt.bands = q->bands;
 292         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
 293
 294         err = prio_dump_offload(sch);
 295         if (err)
 296                 goto nla_put_failure;
 297
 298         if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
 299                 goto nla_put_failure;
 300
 301         return skb->len;
 302
 303 nla_put_failure:
 304         nlmsg_trim(skb, b);
 305         return -1;
 306 }
 307
 308 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
 309                       struct Qdisc **old, struct netlink_ext_ack *extack)
 310 {
 311         struct prio_sched_data *q = qdisc_priv(sch);
 312         unsigned long band = arg - 1;
 313
 314         if (new == NULL)
 315                 new = &noop_qdisc;
 316
 317         *old = qdisc_replace(sch, new, &q->queues[band]);
 318         return 0;
 319 }
 320
 321 static struct Qdisc *
 322 prio_leaf(struct Qdisc *sch, unsigned long arg)
 323 {
 324         struct prio_sched_data *q = qdisc_priv(sch);
 325         unsigned long band = arg - 1;
 326
 327         return q->queues[band];
 328 }
 329
 330 static unsigned long prio_find(struct Qdisc *sch, u32 classid)
 331 {
 332         struct prio_sched_data *q = qdisc_priv(sch);
 333         unsigned long band = TC_H_MIN(classid);
 334
 335         if (band - 1 >= q->bands)
 336                 return 0;
 337         return band;
 338 }
 339
 340 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
 341 {
 342         return prio_find(sch, classid);
 343 }
 344
 345
 346 static void prio_unbind(struct Qdisc *q, unsigned long cl)
 347 {
 348 }
 349
 350 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
 351                            struct tcmsg *tcm)
 352 {
 353         struct prio_sched_data *q = qdisc_priv(sch);
 354
 355         tcm->tcm_handle |= TC_H_MIN(cl);
 356         tcm->tcm_info = q->queues[cl-1]->handle;
 357         return 0;
 358 }
 359
 360 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
 361                                  struct gnet_dump *d)
 362 {
 363         struct prio_sched_data *q = qdisc_priv(sch);
 364         struct Qdisc *cl_q;
 365
 366         cl_q = q->queues[cl - 1];
 367         if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
 368                                   d, NULL, &cl_q->bstats) < 0 ||
 369             gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
 370                 return -1;
 371
 372         return 0;
 373 }
 374
 375 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 376 {
 377         struct prio_sched_data *q = qdisc_priv(sch);
 378         int prio;
 379
 380         if (arg->stop)
 381                 return;
 382
 383         for (prio = 0; prio < q->bands; prio++) {
 384                 if (arg->count < arg->skip) {
 385                         arg->count++;
 386                         continue;
 387                 }
 388                 if (arg->fn(sch, prio + 1, arg) < 0) {
 389                         arg->stop = 1;
 390                         break;
 391                 }
 392                 arg->count++;
 393         }
 394 }
 395
 396 static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl,
 397                                         struct netlink_ext_ack *extack)
 398 {
 399         struct prio_sched_data *q = qdisc_priv(sch);
 400
 401         if (cl)
 402                 return NULL;
 403         return q->block;
 404 }
 405
 406 static const struct Qdisc_class_ops prio_class_ops = {
 407         .graft          =       prio_graft,
 408         .leaf           =       prio_leaf,
 409         .find           =       prio_find,
 410         .walk           =       prio_walk,
 411         .tcf_block      =       prio_tcf_block,
 412         .bind_tcf       =       prio_bind,
 413         .unbind_tcf     =       prio_unbind,
 414         .dump           =       prio_dump_class,
 415         .dump_stats     =       prio_dump_class_stats,
 416 };
 417
 418 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
 419         .next           =       NULL,
 420         .cl_ops         =       &prio_class_ops,
 421         .id             =       "prio",
 422         .priv_size      =       sizeof(struct prio_sched_data),
 423         .enqueue        =       prio_enqueue,
 424         .dequeue        =       prio_dequeue,
 425         .peek           =       prio_peek,
 426         .init           =       prio_init,
 427         .reset          =       prio_reset,
 428         .destroy        =       prio_destroy,
 429         .change         =       prio_tune,
 430         .dump           =       prio_dump,
 431         .owner          =       THIS_MODULE,
 432 };
 433
 434 static int __init prio_module_init(void)
 435 {
 436         return register_qdisc(&prio_qdisc_ops);
 437 }
 438
 439 static void __exit prio_module_exit(void)
 440 {
 441         unregister_qdisc(&prio_qdisc_ops);
 442 }
 443
 444 module_init(prio_module_init)
 445 module_exit(prio_module_exit)
 446
 447 MODULE_LICENSE("GPL");