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
  24
  25 struct prio_sched_data {
  26         int bands;
  27         struct tcf_proto __rcu *filter_list;
  28         u8  prio2band[TC_PRIO_MAX+1];
  29         struct Qdisc *queues[TCQ_PRIO_BANDS];
  30 };
  31
  32
  33 static struct Qdisc *
  34 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
  35 {
  36         struct prio_sched_data *q = qdisc_priv(sch);
  37         u32 band = skb->priority;
  38         struct tcf_result res;
  39         struct tcf_proto *fl;
  40         int err;
  41
  42         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
  43         if (TC_H_MAJ(skb->priority) != sch->handle) {
  44                 fl = rcu_dereference_bh(q->filter_list);
  45                 err = tc_classify(skb, fl, &res, false);
  46 #ifdef CONFIG_NET_CLS_ACT
  47                 switch (err) {
  48                 case TC_ACT_STOLEN:
  49                 case TC_ACT_QUEUED:
  50                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
  51                 case TC_ACT_SHOT:
  52                         return NULL;
  53                 }
  54 #endif
  55                 if (!fl || err < 0) {
  56                         if (TC_H_MAJ(band))
  57                                 band = 0;
  58                         return q->queues[q->prio2band[band & TC_PRIO_MAX]];
  59                 }
  60                 band = res.classid;
  61         }
  62         band = TC_H_MIN(band) - 1;
  63         if (band >= q->bands)
  64                 return q->queues[q->prio2band[0]];
  65
  66         return q->queues[band];
  67 }
  68
  69 static int
  70 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
  71 {
  72         struct Qdisc *qdisc;
  73         int ret;
  74
  75         qdisc = prio_classify(skb, sch, &ret);
  76 #ifdef CONFIG_NET_CLS_ACT
  77         if (qdisc == NULL) {
  78
  79                 if (ret & __NET_XMIT_BYPASS)
  80                         qdisc_qstats_drop(sch);
  81                 kfree_skb(skb);
  82                 return ret;
  83         }
  84 #endif
  85
  86         ret = qdisc_enqueue(skb, qdisc);
  87         if (ret == NET_XMIT_SUCCESS) {
  88                 sch->q.qlen++;
  89                 return NET_XMIT_SUCCESS;
  90         }
  91         if (net_xmit_drop_count(ret))
  92                 qdisc_qstats_drop(sch);
  93         return ret;
  94 }
  95
  96 static struct sk_buff *prio_peek(struct Qdisc *sch)
  97 {
  98         struct prio_sched_data *q = qdisc_priv(sch);
  99         int prio;
 100
 101         for (prio = 0; prio < q->bands; prio++) {
 102                 struct Qdisc *qdisc = q->queues[prio];
 103                 struct sk_buff *skb = qdisc->ops->peek(qdisc);
 104                 if (skb)
 105                         return skb;
 106         }
 107         return NULL;
 108 }
 109
 110 static struct sk_buff *prio_dequeue(struct Qdisc *sch)
 111 {
 112         struct prio_sched_data *q = qdisc_priv(sch);
 113         int prio;
 114
 115         for (prio = 0; prio < q->bands; prio++) {
 116                 struct Qdisc *qdisc = q->queues[prio];
 117                 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
 118                 if (skb) {
 119                         qdisc_bstats_update(sch, skb);
 120                         sch->q.qlen--;
 121                         return skb;
 122                 }
 123         }
 124         return NULL;
 125
 126 }
 127
 128 static void
 129 prio_reset(struct Qdisc *sch)
 130 {
 131         int prio;
 132         struct prio_sched_data *q = qdisc_priv(sch);
 133
 134         for (prio = 0; prio < q->bands; prio++)
 135                 qdisc_reset(q->queues[prio]);
 136         sch->q.qlen = 0;
 137 }
 138
 139 static void
 140 prio_destroy(struct Qdisc *sch)
 141 {
 142         int prio;
 143         struct prio_sched_data *q = qdisc_priv(sch);
 144
 145         tcf_destroy_chain(&q->filter_list);
 146         for (prio = 0; prio < q->bands; prio++)
 147                 qdisc_destroy(q->queues[prio]);
 148 }
 149
 150 static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
 151 {
 152         struct prio_sched_data *q = qdisc_priv(sch);
 153         struct tc_prio_qopt *qopt;
 154         int i;
 155
 156         if (nla_len(opt) < sizeof(*qopt))
 157                 return -EINVAL;
 158         qopt = nla_data(opt);
 159
 160         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
 161                 return -EINVAL;
 162
 163         for (i = 0; i <= TC_PRIO_MAX; i++) {
 164                 if (qopt->priomap[i] >= qopt->bands)
 165                         return -EINVAL;
 166         }
 167
 168         sch_tree_lock(sch);
 169         q->bands = qopt->bands;
 170         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
 171
 172         for (i = q->bands; i < TCQ_PRIO_BANDS; i++) {
 173                 struct Qdisc *child = q->queues[i];
 174                 q->queues[i] = &noop_qdisc;
 175                 if (child != &noop_qdisc) {
 176                         qdisc_tree_reduce_backlog(child, child->q.qlen, child->qstats.backlog);
 177                         qdisc_destroy(child);
 178                 }
 179         }
 180         sch_tree_unlock(sch);
 181
 182         for (i = 0; i < q->bands; i++) {
 183                 if (q->queues[i] == &noop_qdisc) {
 184                         struct Qdisc *child, *old;
 185
 186                         child = qdisc_create_dflt(sch->dev_queue,
 187                                                   &pfifo_qdisc_ops,
 188                                                   TC_H_MAKE(sch->handle, i + 1));
 189                         if (child) {
 190                                 sch_tree_lock(sch);
 191                                 old = q->queues[i];
 192                                 q->queues[i] = child;
 193
 194                                 if (old != &noop_qdisc) {
 195                                         qdisc_tree_reduce_backlog(old,
 196                                                                   old->q.qlen,
 197                                                                   old->qstats.backlog);
 198                                         qdisc_destroy(old);
 199                                 }
 200                                 sch_tree_unlock(sch);
 201                         }
 202                 }
 203         }
 204         return 0;
 205 }
 206
 207 static int prio_init(struct Qdisc *sch, struct nlattr *opt)
 208 {
 209         struct prio_sched_data *q = qdisc_priv(sch);
 210         int i;
 211
 212         for (i = 0; i < TCQ_PRIO_BANDS; i++)
 213                 q->queues[i] = &noop_qdisc;
 214
 215         if (opt == NULL) {
 216                 return -EINVAL;
 217         } else {
 218                 int err;
 219
 220                 if ((err = prio_tune(sch, opt)) != 0)
 221                         return err;
 222         }
 223         return 0;
 224 }
 225
 226 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
 227 {
 228         struct prio_sched_data *q = qdisc_priv(sch);
 229         unsigned char *b = skb_tail_pointer(skb);
 230         struct tc_prio_qopt opt;
 231
 232         opt.bands = q->bands;
 233         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
 234
 235         if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
 236                 goto nla_put_failure;
 237
 238         return skb->len;
 239
 240 nla_put_failure:
 241         nlmsg_trim(skb, b);
 242         return -1;
 243 }
 244
 245 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
 246                       struct Qdisc **old)
 247 {
 248         struct prio_sched_data *q = qdisc_priv(sch);
 249         unsigned long band = arg - 1;
 250
 251         if (new == NULL)
 252                 new = &noop_qdisc;
 253
 254         *old = qdisc_replace(sch, new, &q->queues[band]);
 255         return 0;
 256 }
 257
 258 static struct Qdisc *
 259 prio_leaf(struct Qdisc *sch, unsigned long arg)
 260 {
 261         struct prio_sched_data *q = qdisc_priv(sch);
 262         unsigned long band = arg - 1;
 263
 264         return q->queues[band];
 265 }
 266
 267 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
 268 {
 269         struct prio_sched_data *q = qdisc_priv(sch);
 270         unsigned long band = TC_H_MIN(classid);
 271
 272         if (band - 1 >= q->bands)
 273                 return 0;
 274         return band;
 275 }
 276
 277 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
 278 {
 279         return prio_get(sch, classid);
 280 }
 281
 282
 283 static void prio_put(struct Qdisc *q, unsigned long cl)
 284 {
 285 }
 286
 287 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
 288                            struct tcmsg *tcm)
 289 {
 290         struct prio_sched_data *q = qdisc_priv(sch);
 291
 292         tcm->tcm_handle |= TC_H_MIN(cl);
 293         tcm->tcm_info = q->queues[cl-1]->handle;
 294         return 0;
 295 }
 296
 297 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
 298                                  struct gnet_dump *d)
 299 {
 300         struct prio_sched_data *q = qdisc_priv(sch);
 301         struct Qdisc *cl_q;
 302
 303         cl_q = q->queues[cl - 1];
 304         if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
 305                                   d, NULL, &cl_q->bstats) < 0 ||
 306             gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
 307                 return -1;
 308
 309         return 0;
 310 }
 311
 312 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 313 {
 314         struct prio_sched_data *q = qdisc_priv(sch);
 315         int prio;
 316
 317         if (arg->stop)
 318                 return;
 319
 320         for (prio = 0; prio < q->bands; prio++) {
 321                 if (arg->count < arg->skip) {
 322                         arg->count++;
 323                         continue;
 324                 }
 325                 if (arg->fn(sch, prio + 1, arg) < 0) {
 326                         arg->stop = 1;
 327                         break;
 328                 }
 329                 arg->count++;
 330         }
 331 }
 332
 333 static struct tcf_proto __rcu **prio_find_tcf(struct Qdisc *sch,
 334                                               unsigned long cl)
 335 {
 336         struct prio_sched_data *q = qdisc_priv(sch);
 337
 338         if (cl)
 339                 return NULL;
 340         return &q->filter_list;
 341 }
 342
 343 static const struct Qdisc_class_ops prio_class_ops = {
 344         .graft          =       prio_graft,
 345         .leaf           =       prio_leaf,
 346         .get            =       prio_get,
 347         .put            =       prio_put,
 348         .walk           =       prio_walk,
 349         .tcf_chain      =       prio_find_tcf,
 350         .bind_tcf       =       prio_bind,
 351         .unbind_tcf     =       prio_put,
 352         .dump           =       prio_dump_class,
 353         .dump_stats     =       prio_dump_class_stats,
 354 };
 355
 356 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
 357         .next           =       NULL,
 358         .cl_ops         =       &prio_class_ops,
 359         .id             =       "prio",
 360         .priv_size      =       sizeof(struct prio_sched_data),
 361         .enqueue        =       prio_enqueue,
 362         .dequeue        =       prio_dequeue,
 363         .peek           =       prio_peek,
 364         .init           =       prio_init,
 365         .reset          =       prio_reset,
 366         .destroy        =       prio_destroy,
 367         .change         =       prio_tune,
 368         .dump           =       prio_dump,
 369         .owner          =       THIS_MODULE,
 370 };
 371
 372 static int __init prio_module_init(void)
 373 {
 374         return register_qdisc(&prio_qdisc_ops);
 375 }
 376
 377 static void __exit prio_module_exit(void)
 378 {
 379         unregister_qdisc(&prio_qdisc_ops);
 380 }
 381
 382 module_init(prio_module_init)
 383 module_exit(prio_module_exit)
 384
 385 MODULE_LICENSE("GPL");