Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 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 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 *tb[TCA_DRR_MAX + 1]; |
| 70 | u32 quantum; |
| 71 | int err; |
| 72 | |
| 73 | err = nla_parse_nested(tb, TCA_DRR_MAX, tca[TCA_OPTIONS], drr_policy); |
| 74 | if (err < 0) |
| 75 | return err; |
| 76 | |
| 77 | if (tb[TCA_DRR_QUANTUM]) { |
| 78 | quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]); |
| 79 | if (quantum == 0) |
| 80 | return -EINVAL; |
| 81 | } else |
| 82 | quantum = psched_mtu(qdisc_dev(sch)); |
| 83 | |
| 84 | if (cl != NULL) { |
Stephen Hemminger | 71bcb09 | 2008-11-25 21:13:31 -0800 | [diff] [blame^] | 85 | if (tca[TCA_RATE]) { |
| 86 | err = gen_replace_estimator(&cl->bstats, &cl->rate_est, |
| 87 | qdisc_root_sleeping_lock(sch), |
| 88 | tca[TCA_RATE]); |
| 89 | if (err) |
| 90 | return err; |
| 91 | } |
| 92 | |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 93 | sch_tree_lock(sch); |
| 94 | if (tb[TCA_DRR_QUANTUM]) |
| 95 | cl->quantum = quantum; |
| 96 | sch_tree_unlock(sch); |
| 97 | |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 98 | return 0; |
| 99 | } |
| 100 | |
| 101 | cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL); |
| 102 | if (cl == NULL) |
| 103 | return -ENOBUFS; |
| 104 | |
| 105 | cl->refcnt = 1; |
| 106 | cl->common.classid = classid; |
| 107 | cl->quantum = quantum; |
| 108 | cl->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, |
| 109 | &pfifo_qdisc_ops, classid); |
| 110 | if (cl->qdisc == NULL) |
| 111 | cl->qdisc = &noop_qdisc; |
| 112 | |
Stephen Hemminger | 71bcb09 | 2008-11-25 21:13:31 -0800 | [diff] [blame^] | 113 | if (tca[TCA_RATE]) { |
| 114 | err = gen_replace_estimator(&cl->bstats, &cl->rate_est, |
| 115 | qdisc_root_sleeping_lock(sch), |
| 116 | tca[TCA_RATE]); |
| 117 | if (err) { |
| 118 | qdisc_destroy(cl->qdisc); |
| 119 | kfree(cl); |
| 120 | return err; |
| 121 | } |
| 122 | } |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 123 | |
| 124 | sch_tree_lock(sch); |
| 125 | qdisc_class_hash_insert(&q->clhash, &cl->common); |
| 126 | sch_tree_unlock(sch); |
| 127 | |
| 128 | qdisc_class_hash_grow(sch, &q->clhash); |
| 129 | |
| 130 | *arg = (unsigned long)cl; |
| 131 | return 0; |
| 132 | } |
| 133 | |
| 134 | static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl) |
| 135 | { |
| 136 | gen_kill_estimator(&cl->bstats, &cl->rate_est); |
| 137 | qdisc_destroy(cl->qdisc); |
| 138 | kfree(cl); |
| 139 | } |
| 140 | |
| 141 | static int drr_delete_class(struct Qdisc *sch, unsigned long arg) |
| 142 | { |
| 143 | struct drr_sched *q = qdisc_priv(sch); |
| 144 | struct drr_class *cl = (struct drr_class *)arg; |
| 145 | |
| 146 | if (cl->filter_cnt > 0) |
| 147 | return -EBUSY; |
| 148 | |
| 149 | sch_tree_lock(sch); |
| 150 | |
| 151 | drr_purge_queue(cl); |
| 152 | qdisc_class_hash_remove(&q->clhash, &cl->common); |
| 153 | |
| 154 | if (--cl->refcnt == 0) |
| 155 | drr_destroy_class(sch, cl); |
| 156 | |
| 157 | sch_tree_unlock(sch); |
| 158 | return 0; |
| 159 | } |
| 160 | |
| 161 | static unsigned long drr_get_class(struct Qdisc *sch, u32 classid) |
| 162 | { |
| 163 | struct drr_class *cl = drr_find_class(sch, classid); |
| 164 | |
| 165 | if (cl != NULL) |
| 166 | cl->refcnt++; |
| 167 | |
| 168 | return (unsigned long)cl; |
| 169 | } |
| 170 | |
| 171 | static void drr_put_class(struct Qdisc *sch, unsigned long arg) |
| 172 | { |
| 173 | struct drr_class *cl = (struct drr_class *)arg; |
| 174 | |
| 175 | if (--cl->refcnt == 0) |
| 176 | drr_destroy_class(sch, cl); |
| 177 | } |
| 178 | |
| 179 | static struct tcf_proto **drr_tcf_chain(struct Qdisc *sch, unsigned long cl) |
| 180 | { |
| 181 | struct drr_sched *q = qdisc_priv(sch); |
| 182 | |
| 183 | if (cl) |
| 184 | return NULL; |
| 185 | |
| 186 | return &q->filter_list; |
| 187 | } |
| 188 | |
| 189 | static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent, |
| 190 | u32 classid) |
| 191 | { |
| 192 | struct drr_class *cl = drr_find_class(sch, classid); |
| 193 | |
| 194 | if (cl != NULL) |
| 195 | cl->filter_cnt++; |
| 196 | |
| 197 | return (unsigned long)cl; |
| 198 | } |
| 199 | |
| 200 | static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg) |
| 201 | { |
| 202 | struct drr_class *cl = (struct drr_class *)arg; |
| 203 | |
| 204 | cl->filter_cnt--; |
| 205 | } |
| 206 | |
| 207 | static int drr_graft_class(struct Qdisc *sch, unsigned long arg, |
| 208 | struct Qdisc *new, struct Qdisc **old) |
| 209 | { |
| 210 | struct drr_class *cl = (struct drr_class *)arg; |
| 211 | |
| 212 | if (new == NULL) { |
| 213 | new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, |
| 214 | &pfifo_qdisc_ops, cl->common.classid); |
| 215 | if (new == NULL) |
| 216 | new = &noop_qdisc; |
| 217 | } |
| 218 | |
| 219 | sch_tree_lock(sch); |
| 220 | drr_purge_queue(cl); |
Patrick McHardy | b94c8af | 2008-11-20 04:11:36 -0800 | [diff] [blame] | 221 | *old = cl->qdisc; |
| 222 | cl->qdisc = new; |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 223 | sch_tree_unlock(sch); |
| 224 | return 0; |
| 225 | } |
| 226 | |
| 227 | static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg) |
| 228 | { |
| 229 | struct drr_class *cl = (struct drr_class *)arg; |
| 230 | |
| 231 | return cl->qdisc; |
| 232 | } |
| 233 | |
| 234 | static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg) |
| 235 | { |
| 236 | struct drr_class *cl = (struct drr_class *)arg; |
| 237 | |
| 238 | if (cl->qdisc->q.qlen == 0) |
| 239 | list_del(&cl->alist); |
| 240 | } |
| 241 | |
| 242 | static int drr_dump_class(struct Qdisc *sch, unsigned long arg, |
| 243 | struct sk_buff *skb, struct tcmsg *tcm) |
| 244 | { |
| 245 | struct drr_class *cl = (struct drr_class *)arg; |
| 246 | struct nlattr *nest; |
| 247 | |
| 248 | tcm->tcm_parent = TC_H_ROOT; |
| 249 | tcm->tcm_handle = cl->common.classid; |
| 250 | tcm->tcm_info = cl->qdisc->handle; |
| 251 | |
| 252 | nest = nla_nest_start(skb, TCA_OPTIONS); |
| 253 | if (nest == NULL) |
| 254 | goto nla_put_failure; |
| 255 | NLA_PUT_U32(skb, TCA_DRR_QUANTUM, cl->quantum); |
| 256 | return nla_nest_end(skb, nest); |
| 257 | |
| 258 | nla_put_failure: |
| 259 | nla_nest_cancel(skb, nest); |
| 260 | return -EMSGSIZE; |
| 261 | } |
| 262 | |
| 263 | static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg, |
| 264 | struct gnet_dump *d) |
| 265 | { |
| 266 | struct drr_class *cl = (struct drr_class *)arg; |
| 267 | struct tc_drr_stats xstats; |
| 268 | |
| 269 | memset(&xstats, 0, sizeof(xstats)); |
| 270 | if (cl->qdisc->q.qlen) |
| 271 | xstats.deficit = cl->deficit; |
| 272 | |
| 273 | if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || |
| 274 | gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || |
| 275 | gnet_stats_copy_queue(d, &cl->qdisc->qstats) < 0) |
| 276 | return -1; |
| 277 | |
| 278 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); |
| 279 | } |
| 280 | |
| 281 | static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg) |
| 282 | { |
| 283 | struct drr_sched *q = qdisc_priv(sch); |
| 284 | struct drr_class *cl; |
| 285 | struct hlist_node *n; |
| 286 | unsigned int i; |
| 287 | |
| 288 | if (arg->stop) |
| 289 | return; |
| 290 | |
| 291 | for (i = 0; i < q->clhash.hashsize; i++) { |
| 292 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { |
| 293 | if (arg->count < arg->skip) { |
| 294 | arg->count++; |
| 295 | continue; |
| 296 | } |
| 297 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { |
| 298 | arg->stop = 1; |
| 299 | return; |
| 300 | } |
| 301 | arg->count++; |
| 302 | } |
| 303 | } |
| 304 | } |
| 305 | |
| 306 | static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch, |
| 307 | int *qerr) |
| 308 | { |
| 309 | struct drr_sched *q = qdisc_priv(sch); |
| 310 | struct drr_class *cl; |
| 311 | struct tcf_result res; |
| 312 | int result; |
| 313 | |
| 314 | if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) { |
| 315 | cl = drr_find_class(sch, skb->priority); |
| 316 | if (cl != NULL) |
| 317 | return cl; |
| 318 | } |
| 319 | |
| 320 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
| 321 | result = tc_classify(skb, q->filter_list, &res); |
| 322 | if (result >= 0) { |
| 323 | #ifdef CONFIG_NET_CLS_ACT |
| 324 | switch (result) { |
| 325 | case TC_ACT_QUEUED: |
| 326 | case TC_ACT_STOLEN: |
| 327 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
| 328 | case TC_ACT_SHOT: |
| 329 | return NULL; |
| 330 | } |
| 331 | #endif |
| 332 | cl = (struct drr_class *)res.class; |
| 333 | if (cl == NULL) |
| 334 | cl = drr_find_class(sch, res.classid); |
| 335 | return cl; |
| 336 | } |
| 337 | return NULL; |
| 338 | } |
| 339 | |
| 340 | static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch) |
| 341 | { |
| 342 | struct drr_sched *q = qdisc_priv(sch); |
| 343 | struct drr_class *cl; |
| 344 | unsigned int len; |
| 345 | int err; |
| 346 | |
| 347 | cl = drr_classify(skb, sch, &err); |
| 348 | if (cl == NULL) { |
| 349 | if (err & __NET_XMIT_BYPASS) |
| 350 | sch->qstats.drops++; |
| 351 | kfree_skb(skb); |
| 352 | return err; |
| 353 | } |
| 354 | |
| 355 | len = qdisc_pkt_len(skb); |
| 356 | err = qdisc_enqueue(skb, cl->qdisc); |
| 357 | if (unlikely(err != NET_XMIT_SUCCESS)) { |
| 358 | if (net_xmit_drop_count(err)) { |
| 359 | cl->qstats.drops++; |
| 360 | sch->qstats.drops++; |
| 361 | } |
| 362 | return err; |
| 363 | } |
| 364 | |
| 365 | if (cl->qdisc->q.qlen == 1) { |
| 366 | list_add_tail(&cl->alist, &q->active); |
| 367 | cl->deficit = cl->quantum; |
| 368 | } |
| 369 | |
| 370 | cl->bstats.packets++; |
| 371 | cl->bstats.bytes += len; |
| 372 | sch->bstats.packets++; |
| 373 | sch->bstats.bytes += len; |
| 374 | |
| 375 | sch->q.qlen++; |
| 376 | return err; |
| 377 | } |
| 378 | |
| 379 | static struct sk_buff *drr_dequeue(struct Qdisc *sch) |
| 380 | { |
| 381 | struct drr_sched *q = qdisc_priv(sch); |
| 382 | struct drr_class *cl; |
| 383 | struct sk_buff *skb; |
| 384 | unsigned int len; |
| 385 | |
Patrick McHardy | 3f0947c | 2008-11-24 15:46:08 -0800 | [diff] [blame] | 386 | if (list_empty(&q->active)) |
| 387 | goto out; |
| 388 | while (1) { |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 389 | cl = list_first_entry(&q->active, struct drr_class, alist); |
| 390 | skb = cl->qdisc->ops->peek(cl->qdisc); |
| 391 | if (skb == NULL) |
Patrick McHardy | 3f0947c | 2008-11-24 15:46:08 -0800 | [diff] [blame] | 392 | goto out; |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 393 | |
| 394 | len = qdisc_pkt_len(skb); |
| 395 | if (len <= cl->deficit) { |
| 396 | cl->deficit -= len; |
| 397 | skb = qdisc_dequeue_peeked(cl->qdisc); |
| 398 | if (cl->qdisc->q.qlen == 0) |
| 399 | list_del(&cl->alist); |
| 400 | sch->q.qlen--; |
| 401 | return skb; |
| 402 | } |
| 403 | |
| 404 | cl->deficit += cl->quantum; |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 405 | list_move_tail(&cl->alist, &q->active); |
| 406 | } |
Patrick McHardy | 3f0947c | 2008-11-24 15:46:08 -0800 | [diff] [blame] | 407 | out: |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 408 | return NULL; |
| 409 | } |
| 410 | |
| 411 | static unsigned int drr_drop(struct Qdisc *sch) |
| 412 | { |
| 413 | struct drr_sched *q = qdisc_priv(sch); |
| 414 | struct drr_class *cl; |
| 415 | unsigned int len; |
| 416 | |
| 417 | list_for_each_entry(cl, &q->active, alist) { |
| 418 | if (cl->qdisc->ops->drop) { |
| 419 | len = cl->qdisc->ops->drop(cl->qdisc); |
| 420 | if (len > 0) { |
Jarek Poplawski | 98aa9c8 | 2008-11-21 04:37:27 -0800 | [diff] [blame] | 421 | sch->q.qlen--; |
Patrick McHardy | 13d2a1d | 2008-11-20 04:10:00 -0800 | [diff] [blame] | 422 | if (cl->qdisc->q.qlen == 0) |
| 423 | list_del(&cl->alist); |
| 424 | return len; |
| 425 | } |
| 426 | } |
| 427 | } |
| 428 | return 0; |
| 429 | } |
| 430 | |
| 431 | static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt) |
| 432 | { |
| 433 | struct drr_sched *q = qdisc_priv(sch); |
| 434 | int err; |
| 435 | |
| 436 | err = qdisc_class_hash_init(&q->clhash); |
| 437 | if (err < 0) |
| 438 | return err; |
| 439 | INIT_LIST_HEAD(&q->active); |
| 440 | return 0; |
| 441 | } |
| 442 | |
| 443 | static void drr_reset_qdisc(struct Qdisc *sch) |
| 444 | { |
| 445 | struct drr_sched *q = qdisc_priv(sch); |
| 446 | struct drr_class *cl; |
| 447 | struct hlist_node *n; |
| 448 | unsigned int i; |
| 449 | |
| 450 | for (i = 0; i < q->clhash.hashsize; i++) { |
| 451 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { |
| 452 | if (cl->qdisc->q.qlen) |
| 453 | list_del(&cl->alist); |
| 454 | qdisc_reset(cl->qdisc); |
| 455 | } |
| 456 | } |
| 457 | sch->q.qlen = 0; |
| 458 | } |
| 459 | |
| 460 | static void drr_destroy_qdisc(struct Qdisc *sch) |
| 461 | { |
| 462 | struct drr_sched *q = qdisc_priv(sch); |
| 463 | struct drr_class *cl; |
| 464 | struct hlist_node *n, *next; |
| 465 | unsigned int i; |
| 466 | |
| 467 | tcf_destroy_chain(&q->filter_list); |
| 468 | |
| 469 | for (i = 0; i < q->clhash.hashsize; i++) { |
| 470 | hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i], |
| 471 | common.hnode) |
| 472 | drr_destroy_class(sch, cl); |
| 473 | } |
| 474 | qdisc_class_hash_destroy(&q->clhash); |
| 475 | } |
| 476 | |
| 477 | static const struct Qdisc_class_ops drr_class_ops = { |
| 478 | .change = drr_change_class, |
| 479 | .delete = drr_delete_class, |
| 480 | .get = drr_get_class, |
| 481 | .put = drr_put_class, |
| 482 | .tcf_chain = drr_tcf_chain, |
| 483 | .bind_tcf = drr_bind_tcf, |
| 484 | .unbind_tcf = drr_unbind_tcf, |
| 485 | .graft = drr_graft_class, |
| 486 | .leaf = drr_class_leaf, |
| 487 | .qlen_notify = drr_qlen_notify, |
| 488 | .dump = drr_dump_class, |
| 489 | .dump_stats = drr_dump_class_stats, |
| 490 | .walk = drr_walk, |
| 491 | }; |
| 492 | |
| 493 | static struct Qdisc_ops drr_qdisc_ops __read_mostly = { |
| 494 | .cl_ops = &drr_class_ops, |
| 495 | .id = "drr", |
| 496 | .priv_size = sizeof(struct drr_sched), |
| 497 | .enqueue = drr_enqueue, |
| 498 | .dequeue = drr_dequeue, |
| 499 | .peek = qdisc_peek_dequeued, |
| 500 | .drop = drr_drop, |
| 501 | .init = drr_init_qdisc, |
| 502 | .reset = drr_reset_qdisc, |
| 503 | .destroy = drr_destroy_qdisc, |
| 504 | .owner = THIS_MODULE, |
| 505 | }; |
| 506 | |
| 507 | static int __init drr_init(void) |
| 508 | { |
| 509 | return register_qdisc(&drr_qdisc_ops); |
| 510 | } |
| 511 | |
| 512 | static void __exit drr_exit(void) |
| 513 | { |
| 514 | unregister_qdisc(&drr_qdisc_ops); |
| 515 | } |
| 516 | |
| 517 | module_init(drr_init); |
| 518 | module_exit(drr_exit); |
| 519 | MODULE_LICENSE("GPL"); |