net/sched/sch_multiq.c

   1 /*
   2  * Copyright (c) 2008, Intel Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Author: Alexander Duyck <alexander.h.duyck@intel.com>
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/slab.h>
  21 #include <linux/types.h>
  22 #include <linux/kernel.h>
  23 #include <linux/string.h>
  24 #include <linux/errno.h>
  25 #include <linux/skbuff.h>
  26 #include <net/netlink.h>
  27 #include <net/pkt_sched.h>
  28
  29
  30 struct multiq_sched_data {
  31         u16 bands;
  32         u16 max_bands;
  33         u16 curband;
  34         struct tcf_proto __rcu *filter_list;
  35         struct Qdisc **queues;
  36 };
  37
  38
  39 static struct Qdisc *
  40 multiq_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
  41 {
  42         struct multiq_sched_data *q = qdisc_priv(sch);
  43         u32 band;
  44         struct tcf_result res;
  45         struct tcf_proto *fl = rcu_dereference_bh(q->filter_list);
  46         int err;
  47
  48         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
  49         err = tc_classify(skb, fl, &res, false);
  50 #ifdef CONFIG_NET_CLS_ACT
  51         switch (err) {
  52         case TC_ACT_STOLEN:
  53         case TC_ACT_QUEUED:
  54                 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
  55         case TC_ACT_SHOT:
  56                 return NULL;
  57         }
  58 #endif
  59         band = skb_get_queue_mapping(skb);
  60
  61         if (band >= q->bands)
  62                 return q->queues[0];
  63
  64         return q->queues[band];
  65 }
  66
  67 static int
  68 multiq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
  69 {
  70         struct Qdisc *qdisc;
  71         int ret;
  72
  73         qdisc = multiq_classify(skb, sch, &ret);
  74 #ifdef CONFIG_NET_CLS_ACT
  75         if (qdisc == NULL) {
  76
  77                 if (ret & __NET_XMIT_BYPASS)
  78                         qdisc_qstats_drop(sch);
  79                 kfree_skb(skb);
  80                 return ret;
  81         }
  82 #endif
  83
  84         ret = qdisc_enqueue(skb, qdisc);
  85         if (ret == NET_XMIT_SUCCESS) {
  86                 sch->q.qlen++;
  87                 return NET_XMIT_SUCCESS;
  88         }
  89         if (net_xmit_drop_count(ret))
  90                 qdisc_qstats_drop(sch);
  91         return ret;
  92 }
  93
  94 static struct sk_buff *multiq_dequeue(struct Qdisc *sch)
  95 {
  96         struct multiq_sched_data *q = qdisc_priv(sch);
  97         struct Qdisc *qdisc;
  98         struct sk_buff *skb;
  99         int band;
 100
 101         for (band = 0; band < q->bands; band++) {
 102                 /* cycle through bands to ensure fairness */
 103                 q->curband++;
 104                 if (q->curband >= q->bands)
 105                         q->curband = 0;
 106
 107                 /* Check that target subqueue is available before
 108                  * pulling an skb to avoid head-of-line blocking.
 109                  */
 110                 if (!netif_xmit_stopped(
 111                     netdev_get_tx_queue(qdisc_dev(sch), q->curband))) {
 112                         qdisc = q->queues[q->curband];
 113                         skb = qdisc->dequeue(qdisc);
 114                         if (skb) {
 115                                 qdisc_bstats_update(sch, skb);
 116                                 sch->q.qlen--;
 117                                 return skb;
 118                         }
 119                 }
 120         }
 121         return NULL;
 122
 123 }
 124
 125 static struct sk_buff *multiq_peek(struct Qdisc *sch)
 126 {
 127         struct multiq_sched_data *q = qdisc_priv(sch);
 128         unsigned int curband = q->curband;
 129         struct Qdisc *qdisc;
 130         struct sk_buff *skb;
 131         int band;
 132
 133         for (band = 0; band < q->bands; band++) {
 134                 /* cycle through bands to ensure fairness */
 135                 curband++;
 136                 if (curband >= q->bands)
 137                         curband = 0;
 138
 139                 /* Check that target subqueue is available before
 140                  * pulling an skb to avoid head-of-line blocking.
 141                  */
 142                 if (!netif_xmit_stopped(
 143                     netdev_get_tx_queue(qdisc_dev(sch), curband))) {
 144                         qdisc = q->queues[curband];
 145                         skb = qdisc->ops->peek(qdisc);
 146                         if (skb)
 147                                 return skb;
 148                 }
 149         }
 150         return NULL;
 151
 152 }
 153
 154 static unsigned int multiq_drop(struct Qdisc *sch)
 155 {
 156         struct multiq_sched_data *q = qdisc_priv(sch);
 157         int band;
 158         unsigned int len;
 159         struct Qdisc *qdisc;
 160
 161         for (band = q->bands - 1; band >= 0; band--) {
 162                 qdisc = q->queues[band];
 163                 if (qdisc->ops->drop) {
 164                         len = qdisc->ops->drop(qdisc);
 165                         if (len != 0) {
 166                                 sch->q.qlen--;
 167                                 return len;
 168                         }
 169                 }
 170         }
 171         return 0;
 172 }
 173
 174
 175 static void
 176 multiq_reset(struct Qdisc *sch)
 177 {
 178         u16 band;
 179         struct multiq_sched_data *q = qdisc_priv(sch);
 180
 181         for (band = 0; band < q->bands; band++)
 182                 qdisc_reset(q->queues[band]);
 183         sch->q.qlen = 0;
 184         q->curband = 0;
 185 }
 186
 187 static void
 188 multiq_destroy(struct Qdisc *sch)
 189 {
 190         int band;
 191         struct multiq_sched_data *q = qdisc_priv(sch);
 192
 193         tcf_destroy_chain(&q->filter_list);
 194         for (band = 0; band < q->bands; band++)
 195                 qdisc_destroy(q->queues[band]);
 196
 197         kfree(q->queues);
 198 }
 199
 200 static int multiq_tune(struct Qdisc *sch, struct nlattr *opt)
 201 {
 202         struct multiq_sched_data *q = qdisc_priv(sch);
 203         struct tc_multiq_qopt *qopt;
 204         int i;
 205
 206         if (!netif_is_multiqueue(qdisc_dev(sch)))
 207                 return -EOPNOTSUPP;
 208         if (nla_len(opt) < sizeof(*qopt))
 209                 return -EINVAL;
 210
 211         qopt = nla_data(opt);
 212
 213         qopt->bands = qdisc_dev(sch)->real_num_tx_queues;
 214
 215         sch_tree_lock(sch);
 216         q->bands = qopt->bands;
 217         for (i = q->bands; i < q->max_bands; i++) {
 218                 if (q->queues[i] != &noop_qdisc) {
 219                         struct Qdisc *child = q->queues[i];
 220                         q->queues[i] = &noop_qdisc;
 221                         qdisc_tree_reduce_backlog(child, child->q.qlen,
 222                                                   child->qstats.backlog);
 223                         qdisc_destroy(child);
 224                 }
 225         }
 226
 227         sch_tree_unlock(sch);
 228
 229         for (i = 0; i < q->bands; i++) {
 230                 if (q->queues[i] == &noop_qdisc) {
 231                         struct Qdisc *child, *old;
 232                         child = qdisc_create_dflt(sch->dev_queue,
 233                                                   &pfifo_qdisc_ops,
 234                                                   TC_H_MAKE(sch->handle,
 235                                                             i + 1));
 236                         if (child) {
 237                                 sch_tree_lock(sch);
 238                                 old = q->queues[i];
 239                                 q->queues[i] = child;
 240
 241                                 if (old != &noop_qdisc) {
 242                                         qdisc_tree_reduce_backlog(old,
 243                                                                   old->q.qlen,
 244                                                                   old->qstats.backlog);
 245                                         qdisc_destroy(old);
 246                                 }
 247                                 sch_tree_unlock(sch);
 248                         }
 249                 }
 250         }
 251         return 0;
 252 }
 253
 254 static int multiq_init(struct Qdisc *sch, struct nlattr *opt)
 255 {
 256         struct multiq_sched_data *q = qdisc_priv(sch);
 257         int i, err;
 258
 259         q->queues = NULL;
 260
 261         if (opt == NULL)
 262                 return -EINVAL;
 263
 264         q->max_bands = qdisc_dev(sch)->num_tx_queues;
 265
 266         q->queues = kcalloc(q->max_bands, sizeof(struct Qdisc *), GFP_KERNEL);
 267         if (!q->queues)
 268                 return -ENOBUFS;
 269         for (i = 0; i < q->max_bands; i++)
 270                 q->queues[i] = &noop_qdisc;
 271
 272         err = multiq_tune(sch, opt);
 273
 274         if (err)
 275                 kfree(q->queues);
 276
 277         return err;
 278 }
 279
 280 static int multiq_dump(struct Qdisc *sch, struct sk_buff *skb)
 281 {
 282         struct multiq_sched_data *q = qdisc_priv(sch);
 283         unsigned char *b = skb_tail_pointer(skb);
 284         struct tc_multiq_qopt opt;
 285
 286         opt.bands = q->bands;
 287         opt.max_bands = q->max_bands;
 288
 289         if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
 290                 goto nla_put_failure;
 291
 292         return skb->len;
 293
 294 nla_put_failure:
 295         nlmsg_trim(skb, b);
 296         return -1;
 297 }
 298
 299 static int multiq_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
 300                       struct Qdisc **old)
 301 {
 302         struct multiq_sched_data *q = qdisc_priv(sch);
 303         unsigned long band = arg - 1;
 304
 305         if (new == NULL)
 306                 new = &noop_qdisc;
 307
 308         *old = qdisc_replace(sch, new, &q->queues[band]);
 309         return 0;
 310 }
 311
 312 static struct Qdisc *
 313 multiq_leaf(struct Qdisc *sch, unsigned long arg)
 314 {
 315         struct multiq_sched_data *q = qdisc_priv(sch);
 316         unsigned long band = arg - 1;
 317
 318         return q->queues[band];
 319 }
 320
 321 static unsigned long multiq_get(struct Qdisc *sch, u32 classid)
 322 {
 323         struct multiq_sched_data *q = qdisc_priv(sch);
 324         unsigned long band = TC_H_MIN(classid);
 325
 326         if (band - 1 >= q->bands)
 327                 return 0;
 328         return band;
 329 }
 330
 331 static unsigned long multiq_bind(struct Qdisc *sch, unsigned long parent,
 332                                  u32 classid)
 333 {
 334         return multiq_get(sch, classid);
 335 }
 336
 337
 338 static void multiq_put(struct Qdisc *q, unsigned long cl)
 339 {
 340 }
 341
 342 static int multiq_dump_class(struct Qdisc *sch, unsigned long cl,
 343                              struct sk_buff *skb, struct tcmsg *tcm)
 344 {
 345         struct multiq_sched_data *q = qdisc_priv(sch);
 346
 347         tcm->tcm_handle |= TC_H_MIN(cl);
 348         tcm->tcm_info = q->queues[cl - 1]->handle;
 349         return 0;
 350 }
 351
 352 static int multiq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
 353                                  struct gnet_dump *d)
 354 {
 355         struct multiq_sched_data *q = qdisc_priv(sch);
 356         struct Qdisc *cl_q;
 357
 358         cl_q = q->queues[cl - 1];
 359         if (gnet_stats_copy_basic(d, NULL, &cl_q->bstats) < 0 ||
 360             gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
 361                 return -1;
 362
 363         return 0;
 364 }
 365
 366 static void multiq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 367 {
 368         struct multiq_sched_data *q = qdisc_priv(sch);
 369         int band;
 370
 371         if (arg->stop)
 372                 return;
 373
 374         for (band = 0; band < q->bands; band++) {
 375                 if (arg->count < arg->skip) {
 376                         arg->count++;
 377                         continue;
 378                 }
 379                 if (arg->fn(sch, band + 1, arg) < 0) {
 380                         arg->stop = 1;
 381                         break;
 382                 }
 383                 arg->count++;
 384         }
 385 }
 386
 387 static struct tcf_proto __rcu **multiq_find_tcf(struct Qdisc *sch,
 388                                                 unsigned long cl)
 389 {
 390         struct multiq_sched_data *q = qdisc_priv(sch);
 391
 392         if (cl)
 393                 return NULL;
 394         return &q->filter_list;
 395 }
 396
 397 static const struct Qdisc_class_ops multiq_class_ops = {
 398         .graft          =       multiq_graft,
 399         .leaf           =       multiq_leaf,
 400         .get            =       multiq_get,
 401         .put            =       multiq_put,
 402         .walk           =       multiq_walk,
 403         .tcf_chain      =       multiq_find_tcf,
 404         .bind_tcf       =       multiq_bind,
 405         .unbind_tcf     =       multiq_put,
 406         .dump           =       multiq_dump_class,
 407         .dump_stats     =       multiq_dump_class_stats,
 408 };
 409
 410 static struct Qdisc_ops multiq_qdisc_ops __read_mostly = {
 411         .next           =       NULL,
 412         .cl_ops         =       &multiq_class_ops,
 413         .id             =       "multiq",
 414         .priv_size      =       sizeof(struct multiq_sched_data),
 415         .enqueue        =       multiq_enqueue,
 416         .dequeue        =       multiq_dequeue,
 417         .peek           =       multiq_peek,
 418         .drop           =       multiq_drop,
 419         .init           =       multiq_init,
 420         .reset          =       multiq_reset,
 421         .destroy        =       multiq_destroy,
 422         .change         =       multiq_tune,
 423         .dump           =       multiq_dump,
 424         .owner          =       THIS_MODULE,
 425 };
 426
 427 static int __init multiq_module_init(void)
 428 {
 429         return register_qdisc(&multiq_qdisc_ops);
 430 }
 431
 432 static void __exit multiq_module_exit(void)
 433 {
 434         unregister_qdisc(&multiq_qdisc_ops);
 435 }
 436
 437 module_init(multiq_module_init)
 438 module_exit(multiq_module_exit)
 439
 440 MODULE_LICENSE("GPL");