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/types.h> 16 #include <linux/kernel.h> 17 #include <linux/string.h> 18 #include <linux/errno.h> 19 #include <linux/skbuff.h> 20 #include <net/netlink.h> 21 #include <net/pkt_sched.h> 22 23 24 struct prio_sched_data 25 { 26 int bands; 27 int curband; /* for round-robin */ 28 struct tcf_proto *filter_list; 29 u8 prio2band[TC_PRIO_MAX+1]; 30 struct Qdisc *queues[TCQ_PRIO_BANDS]; 31 int mq; 32 }; 33 34 35 static struct Qdisc * 36 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr) 37 { 38 struct prio_sched_data *q = qdisc_priv(sch); 39 u32 band = skb->priority; 40 struct tcf_result res; 41 int err; 42 43 *qerr = NET_XMIT_BYPASS; 44 if (TC_H_MAJ(skb->priority) != sch->handle) { 45 err = tc_classify(skb, q->filter_list, &res); 46 #ifdef CONFIG_NET_CLS_ACT 47 switch (err) { 48 case TC_ACT_STOLEN: 49 case TC_ACT_QUEUED: 50 *qerr = NET_XMIT_SUCCESS; 51 case TC_ACT_SHOT: 52 return NULL; 53 } 54 #endif 55 if (!q->filter_list || err < 0) { 56 if (TC_H_MAJ(band)) 57 band = 0; 58 band = q->prio2band[band&TC_PRIO_MAX]; 59 goto out; 60 } 61 band = res.classid; 62 } 63 band = TC_H_MIN(band) - 1; 64 if (band >= q->bands) 65 band = q->prio2band[0]; 66 out: 67 if (q->mq) 68 skb_set_queue_mapping(skb, band); 69 return q->queues[band]; 70 } 71 72 static int 73 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch) 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 sch->qstats.drops++; 84 kfree_skb(skb); 85 return ret; 86 } 87 #endif 88 89 if ((ret = qdisc->enqueue(skb, qdisc)) == NET_XMIT_SUCCESS) { 90 sch->bstats.bytes += skb->len; 91 sch->bstats.packets++; 92 sch->q.qlen++; 93 return NET_XMIT_SUCCESS; 94 } 95 sch->qstats.drops++; 96 return ret; 97 } 98 99 100 static int 101 prio_requeue(struct sk_buff *skb, struct Qdisc* sch) 102 { 103 struct Qdisc *qdisc; 104 int ret; 105 106 qdisc = prio_classify(skb, sch, &ret); 107 #ifdef CONFIG_NET_CLS_ACT 108 if (qdisc == NULL) { 109 if (ret == NET_XMIT_BYPASS) 110 sch->qstats.drops++; 111 kfree_skb(skb); 112 return ret; 113 } 114 #endif 115 116 if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) { 117 sch->q.qlen++; 118 sch->qstats.requeues++; 119 return 0; 120 } 121 sch->qstats.drops++; 122 return NET_XMIT_DROP; 123 } 124 125 126 static struct sk_buff * 127 prio_dequeue(struct Qdisc* sch) 128 { 129 struct sk_buff *skb; 130 struct prio_sched_data *q = qdisc_priv(sch); 131 int prio; 132 struct Qdisc *qdisc; 133 134 for (prio = 0; prio < q->bands; prio++) { 135 /* Check if the target subqueue is available before 136 * pulling an skb. This way we avoid excessive requeues 137 * for slower queues. 138 */ 139 if (!__netif_subqueue_stopped(sch->dev, (q->mq ? prio : 0))) { 140 qdisc = q->queues[prio]; 141 skb = qdisc->dequeue(qdisc); 142 if (skb) { 143 sch->q.qlen--; 144 return skb; 145 } 146 } 147 } 148 return NULL; 149 150 } 151 152 static struct sk_buff *rr_dequeue(struct Qdisc* sch) 153 { 154 struct sk_buff *skb; 155 struct prio_sched_data *q = qdisc_priv(sch); 156 struct Qdisc *qdisc; 157 int bandcount; 158 159 /* Only take one pass through the queues. If nothing is available, 160 * return nothing. 161 */ 162 for (bandcount = 0; bandcount < q->bands; bandcount++) { 163 /* Check if the target subqueue is available before 164 * pulling an skb. This way we avoid excessive requeues 165 * for slower queues. If the queue is stopped, try the 166 * next queue. 167 */ 168 if (!__netif_subqueue_stopped(sch->dev, 169 (q->mq ? q->curband : 0))) { 170 qdisc = q->queues[q->curband]; 171 skb = qdisc->dequeue(qdisc); 172 if (skb) { 173 sch->q.qlen--; 174 q->curband++; 175 if (q->curband >= q->bands) 176 q->curband = 0; 177 return skb; 178 } 179 } 180 q->curband++; 181 if (q->curband >= q->bands) 182 q->curband = 0; 183 } 184 return NULL; 185 } 186 187 static unsigned int prio_drop(struct Qdisc* sch) 188 { 189 struct prio_sched_data *q = qdisc_priv(sch); 190 int prio; 191 unsigned int len; 192 struct Qdisc *qdisc; 193 194 for (prio = q->bands-1; prio >= 0; prio--) { 195 qdisc = q->queues[prio]; 196 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) { 197 sch->q.qlen--; 198 return len; 199 } 200 } 201 return 0; 202 } 203 204 205 static void 206 prio_reset(struct Qdisc* sch) 207 { 208 int prio; 209 struct prio_sched_data *q = qdisc_priv(sch); 210 211 for (prio=0; prio<q->bands; prio++) 212 qdisc_reset(q->queues[prio]); 213 sch->q.qlen = 0; 214 } 215 216 static void 217 prio_destroy(struct Qdisc* sch) 218 { 219 int prio; 220 struct prio_sched_data *q = qdisc_priv(sch); 221 222 tcf_destroy_chain(q->filter_list); 223 for (prio=0; prio<q->bands; prio++) 224 qdisc_destroy(q->queues[prio]); 225 } 226 227 static int prio_tune(struct Qdisc *sch, struct nlattr *opt) 228 { 229 struct prio_sched_data *q = qdisc_priv(sch); 230 struct tc_prio_qopt *qopt; 231 struct nlattr *tb[TCA_PRIO_MAX + 1]; 232 int i; 233 234 if (nla_parse_nested_compat(tb, TCA_PRIO_MAX, opt, NULL, qopt, 235 sizeof(*qopt))) 236 return -EINVAL; 237 q->bands = qopt->bands; 238 /* If we're multiqueue, make sure the number of incoming bands 239 * matches the number of queues on the device we're associating with. 240 * If the number of bands requested is zero, then set q->bands to 241 * dev->egress_subqueue_count. Also, the root qdisc must be the 242 * only one that is enabled for multiqueue, since it's the only one 243 * that interacts with the underlying device. 244 */ 245 q->mq = nla_get_flag(tb[TCA_PRIO_MQ]); 246 if (q->mq) { 247 if (sch->parent != TC_H_ROOT) 248 return -EINVAL; 249 if (netif_is_multiqueue(sch->dev)) { 250 if (q->bands == 0) 251 q->bands = sch->dev->egress_subqueue_count; 252 else if (q->bands != sch->dev->egress_subqueue_count) 253 return -EINVAL; 254 } else 255 return -EOPNOTSUPP; 256 } 257 258 if (q->bands > TCQ_PRIO_BANDS || q->bands < 2) 259 return -EINVAL; 260 261 for (i=0; i<=TC_PRIO_MAX; i++) { 262 if (qopt->priomap[i] >= q->bands) 263 return -EINVAL; 264 } 265 266 sch_tree_lock(sch); 267 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1); 268 269 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) { 270 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc); 271 if (child != &noop_qdisc) { 272 qdisc_tree_decrease_qlen(child, child->q.qlen); 273 qdisc_destroy(child); 274 } 275 } 276 sch_tree_unlock(sch); 277 278 for (i=0; i<q->bands; i++) { 279 if (q->queues[i] == &noop_qdisc) { 280 struct Qdisc *child; 281 child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, 282 TC_H_MAKE(sch->handle, i + 1)); 283 if (child) { 284 sch_tree_lock(sch); 285 child = xchg(&q->queues[i], child); 286 287 if (child != &noop_qdisc) { 288 qdisc_tree_decrease_qlen(child, 289 child->q.qlen); 290 qdisc_destroy(child); 291 } 292 sch_tree_unlock(sch); 293 } 294 } 295 } 296 return 0; 297 } 298 299 static int prio_init(struct Qdisc *sch, struct nlattr *opt) 300 { 301 struct prio_sched_data *q = qdisc_priv(sch); 302 int i; 303 304 for (i=0; i<TCQ_PRIO_BANDS; i++) 305 q->queues[i] = &noop_qdisc; 306 307 if (opt == NULL) { 308 return -EINVAL; 309 } else { 310 int err; 311 312 if ((err= prio_tune(sch, opt)) != 0) 313 return err; 314 } 315 return 0; 316 } 317 318 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb) 319 { 320 struct prio_sched_data *q = qdisc_priv(sch); 321 unsigned char *b = skb_tail_pointer(skb); 322 struct nlattr *nest; 323 struct tc_prio_qopt opt; 324 325 opt.bands = q->bands; 326 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1); 327 328 nest = nla_nest_compat_start(skb, TCA_OPTIONS, sizeof(opt), &opt); 329 if (nest == NULL) 330 goto nla_put_failure; 331 if (q->mq) { 332 if (nla_put_flag(skb, TCA_PRIO_MQ) < 0) 333 goto nla_put_failure; 334 } 335 nla_nest_compat_end(skb, nest); 336 337 return skb->len; 338 339 nla_put_failure: 340 nlmsg_trim(skb, b); 341 return -1; 342 } 343 344 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 345 struct Qdisc **old) 346 { 347 struct prio_sched_data *q = qdisc_priv(sch); 348 unsigned long band = arg - 1; 349 350 if (band >= q->bands) 351 return -EINVAL; 352 353 if (new == NULL) 354 new = &noop_qdisc; 355 356 sch_tree_lock(sch); 357 *old = q->queues[band]; 358 q->queues[band] = new; 359 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); 360 qdisc_reset(*old); 361 sch_tree_unlock(sch); 362 363 return 0; 364 } 365 366 static struct Qdisc * 367 prio_leaf(struct Qdisc *sch, unsigned long arg) 368 { 369 struct prio_sched_data *q = qdisc_priv(sch); 370 unsigned long band = arg - 1; 371 372 if (band >= q->bands) 373 return NULL; 374 375 return q->queues[band]; 376 } 377 378 static unsigned long prio_get(struct Qdisc *sch, u32 classid) 379 { 380 struct prio_sched_data *q = qdisc_priv(sch); 381 unsigned long band = TC_H_MIN(classid); 382 383 if (band - 1 >= q->bands) 384 return 0; 385 return band; 386 } 387 388 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid) 389 { 390 return prio_get(sch, classid); 391 } 392 393 394 static void prio_put(struct Qdisc *q, unsigned long cl) 395 { 396 return; 397 } 398 399 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct nlattr **tca, unsigned long *arg) 400 { 401 unsigned long cl = *arg; 402 struct prio_sched_data *q = qdisc_priv(sch); 403 404 if (cl - 1 > q->bands) 405 return -ENOENT; 406 return 0; 407 } 408 409 static int prio_delete(struct Qdisc *sch, unsigned long cl) 410 { 411 struct prio_sched_data *q = qdisc_priv(sch); 412 if (cl - 1 > q->bands) 413 return -ENOENT; 414 return 0; 415 } 416 417 418 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb, 419 struct tcmsg *tcm) 420 { 421 struct prio_sched_data *q = qdisc_priv(sch); 422 423 if (cl - 1 > q->bands) 424 return -ENOENT; 425 tcm->tcm_handle |= TC_H_MIN(cl); 426 if (q->queues[cl-1]) 427 tcm->tcm_info = q->queues[cl-1]->handle; 428 return 0; 429 } 430 431 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl, 432 struct gnet_dump *d) 433 { 434 struct prio_sched_data *q = qdisc_priv(sch); 435 struct Qdisc *cl_q; 436 437 cl_q = q->queues[cl - 1]; 438 if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 || 439 gnet_stats_copy_queue(d, &cl_q->qstats) < 0) 440 return -1; 441 442 return 0; 443 } 444 445 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg) 446 { 447 struct prio_sched_data *q = qdisc_priv(sch); 448 int prio; 449 450 if (arg->stop) 451 return; 452 453 for (prio = 0; prio < q->bands; prio++) { 454 if (arg->count < arg->skip) { 455 arg->count++; 456 continue; 457 } 458 if (arg->fn(sch, prio+1, arg) < 0) { 459 arg->stop = 1; 460 break; 461 } 462 arg->count++; 463 } 464 } 465 466 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl) 467 { 468 struct prio_sched_data *q = qdisc_priv(sch); 469 470 if (cl) 471 return NULL; 472 return &q->filter_list; 473 } 474 475 static const struct Qdisc_class_ops prio_class_ops = { 476 .graft = prio_graft, 477 .leaf = prio_leaf, 478 .get = prio_get, 479 .put = prio_put, 480 .change = prio_change, 481 .delete = prio_delete, 482 .walk = prio_walk, 483 .tcf_chain = prio_find_tcf, 484 .bind_tcf = prio_bind, 485 .unbind_tcf = prio_put, 486 .dump = prio_dump_class, 487 .dump_stats = prio_dump_class_stats, 488 }; 489 490 static struct Qdisc_ops prio_qdisc_ops __read_mostly = { 491 .next = NULL, 492 .cl_ops = &prio_class_ops, 493 .id = "prio", 494 .priv_size = sizeof(struct prio_sched_data), 495 .enqueue = prio_enqueue, 496 .dequeue = prio_dequeue, 497 .requeue = prio_requeue, 498 .drop = prio_drop, 499 .init = prio_init, 500 .reset = prio_reset, 501 .destroy = prio_destroy, 502 .change = prio_tune, 503 .dump = prio_dump, 504 .owner = THIS_MODULE, 505 }; 506 507 static struct Qdisc_ops rr_qdisc_ops __read_mostly = { 508 .next = NULL, 509 .cl_ops = &prio_class_ops, 510 .id = "rr", 511 .priv_size = sizeof(struct prio_sched_data), 512 .enqueue = prio_enqueue, 513 .dequeue = rr_dequeue, 514 .requeue = prio_requeue, 515 .drop = prio_drop, 516 .init = prio_init, 517 .reset = prio_reset, 518 .destroy = prio_destroy, 519 .change = prio_tune, 520 .dump = prio_dump, 521 .owner = THIS_MODULE, 522 }; 523 524 static int __init prio_module_init(void) 525 { 526 int err; 527 528 err = register_qdisc(&prio_qdisc_ops); 529 if (err < 0) 530 return err; 531 err = register_qdisc(&rr_qdisc_ops); 532 if (err < 0) 533 unregister_qdisc(&prio_qdisc_ops); 534 return err; 535 } 536 537 static void __exit prio_module_exit(void) 538 { 539 unregister_qdisc(&prio_qdisc_ops); 540 unregister_qdisc(&rr_qdisc_ops); 541 } 542 543 module_init(prio_module_init) 544 module_exit(prio_module_exit) 545 546 MODULE_LICENSE("GPL"); 547 MODULE_ALIAS("sch_rr"); 548