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/slab.h> 13 #include <linux/init.h> 14 #include <linux/errno.h> 15 #include <linux/netdevice.h> 16 #include <linux/pkt_sched.h> 17 #include <net/sch_generic.h> 18 #include <net/pkt_sched.h> 19 #include <net/pkt_cls.h> 20 21 struct drr_class { 22 struct Qdisc_class_common common; 23 unsigned int filter_cnt; 24 25 struct gnet_stats_basic_packed bstats; 26 struct gnet_stats_queue qstats; 27 struct net_rate_estimator __rcu *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 __rcu *filter_list; 38 struct tcf_block *block; 39 struct Qdisc_class_hash clhash; 40 }; 41 42 static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid) 43 { 44 struct drr_sched *q = qdisc_priv(sch); 45 struct Qdisc_class_common *clc; 46 47 clc = qdisc_class_find(&q->clhash, classid); 48 if (clc == NULL) 49 return NULL; 50 return container_of(clc, struct drr_class, common); 51 } 52 53 static void drr_purge_queue(struct drr_class *cl) 54 { 55 unsigned int len = cl->qdisc->q.qlen; 56 unsigned int backlog = cl->qdisc->qstats.backlog; 57 58 qdisc_reset(cl->qdisc); 59 qdisc_tree_reduce_backlog(cl->qdisc, len, backlog); 60 } 61 62 static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = { 63 [TCA_DRR_QUANTUM] = { .type = NLA_U32 }, 64 }; 65 66 static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid, 67 struct nlattr **tca, unsigned long *arg, 68 struct netlink_ext_ack *extack) 69 { 70 struct drr_sched *q = qdisc_priv(sch); 71 struct drr_class *cl = (struct drr_class *)*arg; 72 struct nlattr *opt = tca[TCA_OPTIONS]; 73 struct nlattr *tb[TCA_DRR_MAX + 1]; 74 u32 quantum; 75 int err; 76 77 if (!opt) { 78 NL_SET_ERR_MSG(extack, "DRR options are required for this operation"); 79 return -EINVAL; 80 } 81 82 err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy, extack); 83 if (err < 0) 84 return err; 85 86 if (tb[TCA_DRR_QUANTUM]) { 87 quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]); 88 if (quantum == 0) { 89 NL_SET_ERR_MSG(extack, "Specified DRR quantum cannot be zero"); 90 return -EINVAL; 91 } 92 } else 93 quantum = psched_mtu(qdisc_dev(sch)); 94 95 if (cl != NULL) { 96 if (tca[TCA_RATE]) { 97 err = gen_replace_estimator(&cl->bstats, NULL, 98 &cl->rate_est, 99 NULL, 100 qdisc_root_sleeping_running(sch), 101 tca[TCA_RATE]); 102 if (err) { 103 NL_SET_ERR_MSG(extack, "Failed to replace estimator"); 104 return err; 105 } 106 } 107 108 sch_tree_lock(sch); 109 if (tb[TCA_DRR_QUANTUM]) 110 cl->quantum = quantum; 111 sch_tree_unlock(sch); 112 113 return 0; 114 } 115 116 cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL); 117 if (cl == NULL) 118 return -ENOBUFS; 119 120 cl->common.classid = classid; 121 cl->quantum = quantum; 122 cl->qdisc = qdisc_create_dflt(sch->dev_queue, 123 &pfifo_qdisc_ops, classid, 124 NULL); 125 if (cl->qdisc == NULL) 126 cl->qdisc = &noop_qdisc; 127 else 128 qdisc_hash_add(cl->qdisc, true); 129 130 if (tca[TCA_RATE]) { 131 err = gen_replace_estimator(&cl->bstats, NULL, &cl->rate_est, 132 NULL, 133 qdisc_root_sleeping_running(sch), 134 tca[TCA_RATE]); 135 if (err) { 136 NL_SET_ERR_MSG(extack, "Failed to replace estimator"); 137 qdisc_destroy(cl->qdisc); 138 kfree(cl); 139 return err; 140 } 141 } 142 143 sch_tree_lock(sch); 144 qdisc_class_hash_insert(&q->clhash, &cl->common); 145 sch_tree_unlock(sch); 146 147 qdisc_class_hash_grow(sch, &q->clhash); 148 149 *arg = (unsigned long)cl; 150 return 0; 151 } 152 153 static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl) 154 { 155 gen_kill_estimator(&cl->rate_est); 156 qdisc_destroy(cl->qdisc); 157 kfree(cl); 158 } 159 160 static int drr_delete_class(struct Qdisc *sch, unsigned long arg) 161 { 162 struct drr_sched *q = qdisc_priv(sch); 163 struct drr_class *cl = (struct drr_class *)arg; 164 165 if (cl->filter_cnt > 0) 166 return -EBUSY; 167 168 sch_tree_lock(sch); 169 170 drr_purge_queue(cl); 171 qdisc_class_hash_remove(&q->clhash, &cl->common); 172 173 sch_tree_unlock(sch); 174 175 drr_destroy_class(sch, cl); 176 return 0; 177 } 178 179 static unsigned long drr_search_class(struct Qdisc *sch, u32 classid) 180 { 181 return (unsigned long)drr_find_class(sch, classid); 182 } 183 184 static struct tcf_block *drr_tcf_block(struct Qdisc *sch, unsigned long cl, 185 struct netlink_ext_ack *extack) 186 { 187 struct drr_sched *q = qdisc_priv(sch); 188 189 if (cl) { 190 NL_SET_ERR_MSG(extack, "DRR classid must be zero"); 191 return NULL; 192 } 193 194 return q->block; 195 } 196 197 static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent, 198 u32 classid) 199 { 200 struct drr_class *cl = drr_find_class(sch, classid); 201 202 if (cl != NULL) 203 cl->filter_cnt++; 204 205 return (unsigned long)cl; 206 } 207 208 static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg) 209 { 210 struct drr_class *cl = (struct drr_class *)arg; 211 212 cl->filter_cnt--; 213 } 214 215 static int drr_graft_class(struct Qdisc *sch, unsigned long arg, 216 struct Qdisc *new, struct Qdisc **old, 217 struct netlink_ext_ack *extack) 218 { 219 struct drr_class *cl = (struct drr_class *)arg; 220 221 if (new == NULL) { 222 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 223 cl->common.classid, NULL); 224 if (new == NULL) 225 new = &noop_qdisc; 226 } 227 228 *old = qdisc_replace(sch, new, &cl->qdisc); 229 return 0; 230 } 231 232 static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg) 233 { 234 struct drr_class *cl = (struct drr_class *)arg; 235 236 return cl->qdisc; 237 } 238 239 static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg) 240 { 241 struct drr_class *cl = (struct drr_class *)arg; 242 243 list_del(&cl->alist); 244 } 245 246 static int drr_dump_class(struct Qdisc *sch, unsigned long arg, 247 struct sk_buff *skb, struct tcmsg *tcm) 248 { 249 struct drr_class *cl = (struct drr_class *)arg; 250 struct nlattr *nest; 251 252 tcm->tcm_parent = TC_H_ROOT; 253 tcm->tcm_handle = cl->common.classid; 254 tcm->tcm_info = cl->qdisc->handle; 255 256 nest = nla_nest_start(skb, TCA_OPTIONS); 257 if (nest == NULL) 258 goto nla_put_failure; 259 if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum)) 260 goto nla_put_failure; 261 return nla_nest_end(skb, nest); 262 263 nla_put_failure: 264 nla_nest_cancel(skb, nest); 265 return -EMSGSIZE; 266 } 267 268 static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg, 269 struct gnet_dump *d) 270 { 271 struct drr_class *cl = (struct drr_class *)arg; 272 __u32 qlen = cl->qdisc->q.qlen; 273 struct tc_drr_stats xstats; 274 275 memset(&xstats, 0, sizeof(xstats)); 276 if (qlen) 277 xstats.deficit = cl->deficit; 278 279 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), 280 d, NULL, &cl->bstats) < 0 || 281 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || 282 gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0) 283 return -1; 284 285 return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); 286 } 287 288 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg) 289 { 290 struct drr_sched *q = qdisc_priv(sch); 291 struct drr_class *cl; 292 unsigned int i; 293 294 if (arg->stop) 295 return; 296 297 for (i = 0; i < q->clhash.hashsize; i++) { 298 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 299 if (arg->count < arg->skip) { 300 arg->count++; 301 continue; 302 } 303 if (arg->fn(sch, (unsigned long)cl, arg) < 0) { 304 arg->stop = 1; 305 return; 306 } 307 arg->count++; 308 } 309 } 310 } 311 312 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch, 313 int *qerr) 314 { 315 struct drr_sched *q = qdisc_priv(sch); 316 struct drr_class *cl; 317 struct tcf_result res; 318 struct tcf_proto *fl; 319 int result; 320 321 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) { 322 cl = drr_find_class(sch, skb->priority); 323 if (cl != NULL) 324 return cl; 325 } 326 327 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 328 fl = rcu_dereference_bh(q->filter_list); 329 result = tcf_classify(skb, fl, &res, false); 330 if (result >= 0) { 331 #ifdef CONFIG_NET_CLS_ACT 332 switch (result) { 333 case TC_ACT_QUEUED: 334 case TC_ACT_STOLEN: 335 case TC_ACT_TRAP: 336 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 337 /* fall through */ 338 case TC_ACT_SHOT: 339 return NULL; 340 } 341 #endif 342 cl = (struct drr_class *)res.class; 343 if (cl == NULL) 344 cl = drr_find_class(sch, res.classid); 345 return cl; 346 } 347 return NULL; 348 } 349 350 static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch, 351 struct sk_buff **to_free) 352 { 353 struct drr_sched *q = qdisc_priv(sch); 354 struct drr_class *cl; 355 int err = 0; 356 357 cl = drr_classify(skb, sch, &err); 358 if (cl == NULL) { 359 if (err & __NET_XMIT_BYPASS) 360 qdisc_qstats_drop(sch); 361 __qdisc_drop(skb, to_free); 362 return err; 363 } 364 365 err = qdisc_enqueue(skb, cl->qdisc, to_free); 366 if (unlikely(err != NET_XMIT_SUCCESS)) { 367 if (net_xmit_drop_count(err)) { 368 cl->qstats.drops++; 369 qdisc_qstats_drop(sch); 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 qdisc_qstats_backlog_inc(sch, skb); 380 sch->q.qlen++; 381 return err; 382 } 383 384 static struct sk_buff *drr_dequeue(struct Qdisc *sch) 385 { 386 struct drr_sched *q = qdisc_priv(sch); 387 struct drr_class *cl; 388 struct sk_buff *skb; 389 unsigned int len; 390 391 if (list_empty(&q->active)) 392 goto out; 393 while (1) { 394 cl = list_first_entry(&q->active, struct drr_class, alist); 395 skb = cl->qdisc->ops->peek(cl->qdisc); 396 if (skb == NULL) { 397 qdisc_warn_nonwc(__func__, cl->qdisc); 398 goto out; 399 } 400 401 len = qdisc_pkt_len(skb); 402 if (len <= cl->deficit) { 403 cl->deficit -= len; 404 skb = qdisc_dequeue_peeked(cl->qdisc); 405 if (unlikely(skb == NULL)) 406 goto out; 407 if (cl->qdisc->q.qlen == 0) 408 list_del(&cl->alist); 409 410 bstats_update(&cl->bstats, skb); 411 qdisc_bstats_update(sch, skb); 412 qdisc_qstats_backlog_dec(sch, skb); 413 sch->q.qlen--; 414 return skb; 415 } 416 417 cl->deficit += cl->quantum; 418 list_move_tail(&cl->alist, &q->active); 419 } 420 out: 421 return NULL; 422 } 423 424 static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt, 425 struct netlink_ext_ack *extack) 426 { 427 struct drr_sched *q = qdisc_priv(sch); 428 int err; 429 430 err = tcf_block_get(&q->block, &q->filter_list, sch, extack); 431 if (err) 432 return err; 433 err = qdisc_class_hash_init(&q->clhash); 434 if (err < 0) 435 return err; 436 INIT_LIST_HEAD(&q->active); 437 return 0; 438 } 439 440 static void drr_reset_qdisc(struct Qdisc *sch) 441 { 442 struct drr_sched *q = qdisc_priv(sch); 443 struct drr_class *cl; 444 unsigned int i; 445 446 for (i = 0; i < q->clhash.hashsize; i++) { 447 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 448 if (cl->qdisc->q.qlen) 449 list_del(&cl->alist); 450 qdisc_reset(cl->qdisc); 451 } 452 } 453 sch->qstats.backlog = 0; 454 sch->q.qlen = 0; 455 } 456 457 static void drr_destroy_qdisc(struct Qdisc *sch) 458 { 459 struct drr_sched *q = qdisc_priv(sch); 460 struct drr_class *cl; 461 struct hlist_node *next; 462 unsigned int i; 463 464 tcf_block_put(q->block); 465 466 for (i = 0; i < q->clhash.hashsize; i++) { 467 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], 468 common.hnode) 469 drr_destroy_class(sch, cl); 470 } 471 qdisc_class_hash_destroy(&q->clhash); 472 } 473 474 static const struct Qdisc_class_ops drr_class_ops = { 475 .change = drr_change_class, 476 .delete = drr_delete_class, 477 .find = drr_search_class, 478 .tcf_block = drr_tcf_block, 479 .bind_tcf = drr_bind_tcf, 480 .unbind_tcf = drr_unbind_tcf, 481 .graft = drr_graft_class, 482 .leaf = drr_class_leaf, 483 .qlen_notify = drr_qlen_notify, 484 .dump = drr_dump_class, 485 .dump_stats = drr_dump_class_stats, 486 .walk = drr_walk, 487 }; 488 489 static struct Qdisc_ops drr_qdisc_ops __read_mostly = { 490 .cl_ops = &drr_class_ops, 491 .id = "drr", 492 .priv_size = sizeof(struct drr_sched), 493 .enqueue = drr_enqueue, 494 .dequeue = drr_dequeue, 495 .peek = qdisc_peek_dequeued, 496 .init = drr_init_qdisc, 497 .reset = drr_reset_qdisc, 498 .destroy = drr_destroy_qdisc, 499 .owner = THIS_MODULE, 500 }; 501 502 static int __init drr_init(void) 503 { 504 return register_qdisc(&drr_qdisc_ops); 505 } 506 507 static void __exit drr_exit(void) 508 { 509 unregister_qdisc(&drr_qdisc_ops); 510 } 511 512 module_init(drr_init); 513 module_exit(drr_exit); 514 MODULE_LICENSE("GPL"); 515