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/init.h> 13 #include <linux/errno.h> 14 #include <linux/netdevice.h> 15 #include <linux/pkt_sched.h> 16 #include <net/sch_generic.h> 17 #include <net/pkt_sched.h> 18 #include <net/pkt_cls.h> 19 20 struct drr_class { 21 struct Qdisc_class_common common; 22 unsigned int refcnt; 23 unsigned int filter_cnt; 24 25 struct gnet_stats_basic_packed bstats; 26 struct gnet_stats_queue qstats; 27 struct gnet_stats_rate_est rate_est; 28 struct list_head alist; 29 struct Qdisc *qdisc; 30 31 u32 quantum; 32 u32 deficit; 33 }; 34 35 struct drr_sched { 36 struct list_head active; 37 struct tcf_proto *filter_list; 38 struct Qdisc_class_hash clhash; 39 }; 40 41 static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid) 42 { 43 struct drr_sched *q = qdisc_priv(sch); 44 struct Qdisc_class_common *clc; 45 46 clc = qdisc_class_find(&q->clhash, classid); 47 if (clc == NULL) 48 return NULL; 49 return container_of(clc, struct drr_class, common); 50 } 51 52 static void drr_purge_queue(struct drr_class *cl) 53 { 54 unsigned int len = cl->qdisc->q.qlen; 55 56 qdisc_reset(cl->qdisc); 57 qdisc_tree_decrease_qlen(cl->qdisc, len); 58 } 59 60 static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = { 61 [TCA_DRR_QUANTUM] = { .type = NLA_U32 }, 62 }; 63 64 static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid, 65 struct nlattr **tca, unsigned long *arg) 66 { 67 struct drr_sched *q = qdisc_priv(sch); 68 struct drr_class *cl = (struct drr_class *)*arg; 69 struct nlattr *opt = tca[TCA_OPTIONS]; 70 struct nlattr *tb[TCA_DRR_MAX + 1]; 71 u32 quantum; 72 int err; 73 74 if (!opt) 75 return -EINVAL; 76 77 err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy); 78 if (err < 0) 79 return err; 80 81 if (tb[TCA_DRR_QUANTUM]) { 82 quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]); 83 if (quantum == 0) 84 return -EINVAL; 85 } else 86 quantum = psched_mtu(qdisc_dev(sch)); 87 88 if (cl != NULL) { 89 if (tca[TCA_RATE]) { 90 err = gen_replace_estimator(&cl->bstats, &cl->rate_est, 91 qdisc_root_sleeping_lock(sch), 92 tca[TCA_RATE]); 93 if (err) 94 return err; 95 } 96 97 sch_tree_lock(sch); 98 if (tb[TCA_DRR_QUANTUM]) 99 cl->quantum = quantum; 100 sch_tree_unlock(sch); 101 102 return 0; 103 } 104 105 cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL); 106 if (cl == NULL) 107 return -ENOBUFS; 108 109 cl->refcnt = 1; 110 cl->common.classid = classid; 111 cl->quantum = quantum; 112 cl->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, 113 &pfifo_qdisc_ops, classid); 114 if (cl->qdisc == NULL) 115 cl->qdisc = &noop_qdisc; 116 117 if (tca[TCA_RATE]) { 118 err = gen_replace_estimator(&cl->bstats, &cl->rate_est, 119 qdisc_root_sleeping_lock(sch), 120 tca[TCA_RATE]); 121 if (err) { 122 qdisc_destroy(cl->qdisc); 123 kfree(cl); 124 return err; 125 } 126 } 127 128 sch_tree_lock(sch); 129 qdisc_class_hash_insert(&q->clhash, &cl->common); 130 sch_tree_unlock(sch); 131 132 qdisc_class_hash_grow(sch, &q->clhash); 133 134 *arg = (unsigned long)cl; 135 return 0; 136 } 137 138 static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl) 139 { 140 gen_kill_estimator(&cl->bstats, &cl->rate_est); 141 qdisc_destroy(cl->qdisc); 142 kfree(cl); 143 } 144 145 static int drr_delete_class(struct Qdisc *sch, unsigned long arg) 146 { 147 struct drr_sched *q = qdisc_priv(sch); 148 struct drr_class *cl = (struct drr_class *)arg; 149 150 if (cl->filter_cnt > 0) 151 return -EBUSY; 152 153 sch_tree_lock(sch); 154 155 drr_purge_queue(cl); 156 qdisc_class_hash_remove(&q->clhash, &cl->common); 157 158 BUG_ON(--cl->refcnt == 0); 159 /* 160 * This shouldn't happen: we "hold" one cops->get() when called 161 * from tc_ctl_tclass; the destroy method is done from cops->put(). 162 */ 163 164 sch_tree_unlock(sch); 165 return 0; 166 } 167 168 static unsigned long drr_get_class(struct Qdisc *sch, u32 classid) 169 { 170 struct drr_class *cl = drr_find_class(sch, classid); 171 172 if (cl != NULL) 173 cl->refcnt++; 174 175 return (unsigned long)cl; 176 } 177 178 static void drr_put_class(struct Qdisc *sch, unsigned long arg) 179 { 180 struct drr_class *cl = (struct drr_class *)arg; 181 182 if (--cl->refcnt == 0) 183 drr_destroy_class(sch, cl); 184 } 185 186 static struct tcf_proto **drr_tcf_chain(struct Qdisc *sch, unsigned long cl) 187 { 188 struct drr_sched *q = qdisc_priv(sch); 189 190 if (cl) 191 return NULL; 192 193 return &q->filter_list; 194 } 195 196 static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent, 197 u32 classid) 198 { 199 struct drr_class *cl = drr_find_class(sch, classid); 200 201 if (cl != NULL) 202 cl->filter_cnt++; 203 204 return (unsigned long)cl; 205 } 206 207 static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg) 208 { 209 struct drr_class *cl = (struct drr_class *)arg; 210 211 cl->filter_cnt--; 212 } 213 214 static int drr_graft_class(struct Qdisc *sch, unsigned long arg, 215 struct Qdisc *new, struct Qdisc **old) 216 { 217 struct drr_class *cl = (struct drr_class *)arg; 218 219 if (new == NULL) { 220 new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, 221 &pfifo_qdisc_ops, cl->common.classid); 222 if (new == NULL) 223 new = &noop_qdisc; 224 } 225 226 sch_tree_lock(sch); 227 drr_purge_queue(cl); 228 *old = cl->qdisc; 229 cl->qdisc = new; 230 sch_tree_unlock(sch); 231 return 0; 232 } 233 234 static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg) 235 { 236 struct drr_class *cl = (struct drr_class *)arg; 237 238 return cl->qdisc; 239 } 240 241 static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg) 242 { 243 struct drr_class *cl = (struct drr_class *)arg; 244 245 if (cl->qdisc->q.qlen == 0) 246 list_del(&cl->alist); 247 } 248 249 static int drr_dump_class(struct Qdisc *sch, unsigned long arg, 250 struct sk_buff *skb, struct tcmsg *tcm) 251 { 252 struct drr_class *cl = (struct drr_class *)arg; 253 struct nlattr *nest; 254 255 tcm->tcm_parent = TC_H_ROOT; 256 tcm->tcm_handle = cl->common.classid; 257 tcm->tcm_info = cl->qdisc->handle; 258 259 nest = nla_nest_start(skb, TCA_OPTIONS); 260 if (nest == NULL) 261 goto nla_put_failure; 262 NLA_PUT_U32(skb, TCA_DRR_QUANTUM, cl->quantum); 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 struct tc_drr_stats xstats; 275 276 memset(&xstats, 0, sizeof(xstats)); 277 if (cl->qdisc->q.qlen) { 278 xstats.deficit = cl->deficit; 279 cl->qdisc->qstats.qlen = cl->qdisc->q.qlen; 280 } 281 282 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || 283 gnet_stats_copy_rate_est(d, &cl->bstats, &cl->rate_est) < 0 || 284 gnet_stats_copy_queue(d, &cl->qdisc->qstats) < 0) 285 return -1; 286 287 return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); 288 } 289 290 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg) 291 { 292 struct drr_sched *q = qdisc_priv(sch); 293 struct drr_class *cl; 294 struct hlist_node *n; 295 unsigned int i; 296 297 if (arg->stop) 298 return; 299 300 for (i = 0; i < q->clhash.hashsize; i++) { 301 hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { 302 if (arg->count < arg->skip) { 303 arg->count++; 304 continue; 305 } 306 if (arg->fn(sch, (unsigned long)cl, arg) < 0) { 307 arg->stop = 1; 308 return; 309 } 310 arg->count++; 311 } 312 } 313 } 314 315 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch, 316 int *qerr) 317 { 318 struct drr_sched *q = qdisc_priv(sch); 319 struct drr_class *cl; 320 struct tcf_result res; 321 int result; 322 323 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) { 324 cl = drr_find_class(sch, skb->priority); 325 if (cl != NULL) 326 return cl; 327 } 328 329 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 330 result = tc_classify(skb, q->filter_list, &res); 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 unsigned int len; 354 int err; 355 356 cl = drr_classify(skb, sch, &err); 357 if (cl == NULL) { 358 if (err & __NET_XMIT_BYPASS) 359 sch->qstats.drops++; 360 kfree_skb(skb); 361 return err; 362 } 363 364 len = qdisc_pkt_len(skb); 365 err = qdisc_enqueue(skb, cl->qdisc); 366 if (unlikely(err != NET_XMIT_SUCCESS)) { 367 if (net_xmit_drop_count(err)) { 368 cl->qstats.drops++; 369 sch->qstats.drops++; 370 } 371 return err; 372 } 373 374 if (cl->qdisc->q.qlen == 1) { 375 list_add_tail(&cl->alist, &q->active); 376 cl->deficit = cl->quantum; 377 } 378 379 cl->bstats.packets++; 380 cl->bstats.bytes += len; 381 sch->bstats.packets++; 382 sch->bstats.bytes += len; 383 384 sch->q.qlen++; 385 return err; 386 } 387 388 static struct sk_buff *drr_dequeue(struct Qdisc *sch) 389 { 390 struct drr_sched *q = qdisc_priv(sch); 391 struct drr_class *cl; 392 struct sk_buff *skb; 393 unsigned int len; 394 395 if (list_empty(&q->active)) 396 goto out; 397 while (1) { 398 cl = list_first_entry(&q->active, struct drr_class, alist); 399 skb = cl->qdisc->ops->peek(cl->qdisc); 400 if (skb == NULL) 401 goto out; 402 403 len = qdisc_pkt_len(skb); 404 if (len <= cl->deficit) { 405 cl->deficit -= len; 406 skb = qdisc_dequeue_peeked(cl->qdisc); 407 if (cl->qdisc->q.qlen == 0) 408 list_del(&cl->alist); 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 struct hlist_node *n; 457 unsigned int i; 458 459 for (i = 0; i < q->clhash.hashsize; i++) { 460 hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { 461 if (cl->qdisc->q.qlen) 462 list_del(&cl->alist); 463 qdisc_reset(cl->qdisc); 464 } 465 } 466 sch->q.qlen = 0; 467 } 468 469 static void drr_destroy_qdisc(struct Qdisc *sch) 470 { 471 struct drr_sched *q = qdisc_priv(sch); 472 struct drr_class *cl; 473 struct hlist_node *n, *next; 474 unsigned int i; 475 476 tcf_destroy_chain(&q->filter_list); 477 478 for (i = 0; i < q->clhash.hashsize; i++) { 479 hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i], 480 common.hnode) 481 drr_destroy_class(sch, cl); 482 } 483 qdisc_class_hash_destroy(&q->clhash); 484 } 485 486 static const struct Qdisc_class_ops drr_class_ops = { 487 .change = drr_change_class, 488 .delete = drr_delete_class, 489 .get = drr_get_class, 490 .put = drr_put_class, 491 .tcf_chain = drr_tcf_chain, 492 .bind_tcf = drr_bind_tcf, 493 .unbind_tcf = drr_unbind_tcf, 494 .graft = drr_graft_class, 495 .leaf = drr_class_leaf, 496 .qlen_notify = drr_qlen_notify, 497 .dump = drr_dump_class, 498 .dump_stats = drr_dump_class_stats, 499 .walk = drr_walk, 500 }; 501 502 static struct Qdisc_ops drr_qdisc_ops __read_mostly = { 503 .cl_ops = &drr_class_ops, 504 .id = "drr", 505 .priv_size = sizeof(struct drr_sched), 506 .enqueue = drr_enqueue, 507 .dequeue = drr_dequeue, 508 .peek = qdisc_peek_dequeued, 509 .drop = drr_drop, 510 .init = drr_init_qdisc, 511 .reset = drr_reset_qdisc, 512 .destroy = drr_destroy_qdisc, 513 .owner = THIS_MODULE, 514 }; 515 516 static int __init drr_init(void) 517 { 518 return register_qdisc(&drr_qdisc_ops); 519 } 520 521 static void __exit drr_exit(void) 522 { 523 unregister_qdisc(&drr_qdisc_ops); 524 } 525 526 module_init(drr_init); 527 module_exit(drr_exit); 528 MODULE_LICENSE("GPL"); 529