1 /* 2 * net/sched/sch_red.c Random Early Detection queue. 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 * 11 * Changes: 12 * J Hadi Salim 980914: computation fixes 13 * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly. 14 * J Hadi Salim 980816: ECN support 15 */ 16 17 #include <linux/module.h> 18 #include <linux/types.h> 19 #include <linux/kernel.h> 20 #include <linux/skbuff.h> 21 #include <net/pkt_sched.h> 22 #include <net/pkt_cls.h> 23 #include <net/inet_ecn.h> 24 #include <net/red.h> 25 26 27 /* Parameters, settable by user: 28 ----------------------------- 29 30 limit - bytes (must be > qth_max + burst) 31 32 Hard limit on queue length, should be chosen >qth_max 33 to allow packet bursts. This parameter does not 34 affect the algorithms behaviour and can be chosen 35 arbitrarily high (well, less than ram size) 36 Really, this limit will never be reached 37 if RED works correctly. 38 */ 39 40 struct red_sched_data { 41 u32 limit; /* HARD maximal queue length */ 42 unsigned char flags; 43 struct timer_list adapt_timer; 44 struct Qdisc *sch; 45 struct red_parms parms; 46 struct red_vars vars; 47 struct red_stats stats; 48 struct Qdisc *qdisc; 49 }; 50 51 static inline int red_use_ecn(struct red_sched_data *q) 52 { 53 return q->flags & TC_RED_ECN; 54 } 55 56 static inline int red_use_harddrop(struct red_sched_data *q) 57 { 58 return q->flags & TC_RED_HARDDROP; 59 } 60 61 static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch, 62 struct sk_buff **to_free) 63 { 64 struct red_sched_data *q = qdisc_priv(sch); 65 struct Qdisc *child = q->qdisc; 66 int ret; 67 68 q->vars.qavg = red_calc_qavg(&q->parms, 69 &q->vars, 70 child->qstats.backlog); 71 72 if (red_is_idling(&q->vars)) 73 red_end_of_idle_period(&q->vars); 74 75 switch (red_action(&q->parms, &q->vars, q->vars.qavg)) { 76 case RED_DONT_MARK: 77 break; 78 79 case RED_PROB_MARK: 80 qdisc_qstats_overlimit(sch); 81 if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) { 82 q->stats.prob_drop++; 83 goto congestion_drop; 84 } 85 86 q->stats.prob_mark++; 87 break; 88 89 case RED_HARD_MARK: 90 qdisc_qstats_overlimit(sch); 91 if (red_use_harddrop(q) || !red_use_ecn(q) || 92 !INET_ECN_set_ce(skb)) { 93 q->stats.forced_drop++; 94 goto congestion_drop; 95 } 96 97 q->stats.forced_mark++; 98 break; 99 } 100 101 ret = qdisc_enqueue(skb, child, to_free); 102 if (likely(ret == NET_XMIT_SUCCESS)) { 103 qdisc_qstats_backlog_inc(sch, skb); 104 sch->q.qlen++; 105 } else if (net_xmit_drop_count(ret)) { 106 q->stats.pdrop++; 107 qdisc_qstats_drop(sch); 108 } 109 return ret; 110 111 congestion_drop: 112 qdisc_drop(skb, sch, to_free); 113 return NET_XMIT_CN; 114 } 115 116 static struct sk_buff *red_dequeue(struct Qdisc *sch) 117 { 118 struct sk_buff *skb; 119 struct red_sched_data *q = qdisc_priv(sch); 120 struct Qdisc *child = q->qdisc; 121 122 skb = child->dequeue(child); 123 if (skb) { 124 qdisc_bstats_update(sch, skb); 125 qdisc_qstats_backlog_dec(sch, skb); 126 sch->q.qlen--; 127 } else { 128 if (!red_is_idling(&q->vars)) 129 red_start_of_idle_period(&q->vars); 130 } 131 return skb; 132 } 133 134 static struct sk_buff *red_peek(struct Qdisc *sch) 135 { 136 struct red_sched_data *q = qdisc_priv(sch); 137 struct Qdisc *child = q->qdisc; 138 139 return child->ops->peek(child); 140 } 141 142 static void red_reset(struct Qdisc *sch) 143 { 144 struct red_sched_data *q = qdisc_priv(sch); 145 146 qdisc_reset(q->qdisc); 147 sch->qstats.backlog = 0; 148 sch->q.qlen = 0; 149 red_restart(&q->vars); 150 } 151 152 static int red_offload(struct Qdisc *sch, bool enable) 153 { 154 struct red_sched_data *q = qdisc_priv(sch); 155 struct net_device *dev = qdisc_dev(sch); 156 struct tc_red_qopt_offload opt = { 157 .handle = sch->handle, 158 .parent = sch->parent, 159 }; 160 161 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 162 return -EOPNOTSUPP; 163 164 if (enable) { 165 opt.command = TC_RED_REPLACE; 166 opt.set.min = q->parms.qth_min >> q->parms.Wlog; 167 opt.set.max = q->parms.qth_max >> q->parms.Wlog; 168 opt.set.probability = q->parms.max_P; 169 opt.set.is_ecn = red_use_ecn(q); 170 } else { 171 opt.command = TC_RED_DESTROY; 172 } 173 174 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt); 175 } 176 177 static void red_destroy(struct Qdisc *sch) 178 { 179 struct red_sched_data *q = qdisc_priv(sch); 180 181 del_timer_sync(&q->adapt_timer); 182 red_offload(sch, false); 183 qdisc_destroy(q->qdisc); 184 } 185 186 static const struct nla_policy red_policy[TCA_RED_MAX + 1] = { 187 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) }, 188 [TCA_RED_STAB] = { .len = RED_STAB_SIZE }, 189 [TCA_RED_MAX_P] = { .type = NLA_U32 }, 190 }; 191 192 static int red_change(struct Qdisc *sch, struct nlattr *opt) 193 { 194 struct red_sched_data *q = qdisc_priv(sch); 195 struct nlattr *tb[TCA_RED_MAX + 1]; 196 struct tc_red_qopt *ctl; 197 struct Qdisc *child = NULL; 198 int err; 199 u32 max_P; 200 201 if (opt == NULL) 202 return -EINVAL; 203 204 err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL); 205 if (err < 0) 206 return err; 207 208 if (tb[TCA_RED_PARMS] == NULL || 209 tb[TCA_RED_STAB] == NULL) 210 return -EINVAL; 211 212 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0; 213 214 ctl = nla_data(tb[TCA_RED_PARMS]); 215 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog)) 216 return -EINVAL; 217 218 if (ctl->limit > 0) { 219 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit); 220 if (IS_ERR(child)) 221 return PTR_ERR(child); 222 } 223 224 if (child != &noop_qdisc) 225 qdisc_hash_add(child, true); 226 sch_tree_lock(sch); 227 q->flags = ctl->flags; 228 q->limit = ctl->limit; 229 if (child) { 230 qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen, 231 q->qdisc->qstats.backlog); 232 qdisc_destroy(q->qdisc); 233 q->qdisc = child; 234 } 235 236 red_set_parms(&q->parms, 237 ctl->qth_min, ctl->qth_max, ctl->Wlog, 238 ctl->Plog, ctl->Scell_log, 239 nla_data(tb[TCA_RED_STAB]), 240 max_P); 241 red_set_vars(&q->vars); 242 243 del_timer(&q->adapt_timer); 244 if (ctl->flags & TC_RED_ADAPTATIVE) 245 mod_timer(&q->adapt_timer, jiffies + HZ/2); 246 247 if (!q->qdisc->q.qlen) 248 red_start_of_idle_period(&q->vars); 249 250 sch_tree_unlock(sch); 251 red_offload(sch, true); 252 return 0; 253 } 254 255 static inline void red_adaptative_timer(struct timer_list *t) 256 { 257 struct red_sched_data *q = from_timer(q, t, adapt_timer); 258 struct Qdisc *sch = q->sch; 259 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 260 261 spin_lock(root_lock); 262 red_adaptative_algo(&q->parms, &q->vars); 263 mod_timer(&q->adapt_timer, jiffies + HZ/2); 264 spin_unlock(root_lock); 265 } 266 267 static int red_init(struct Qdisc *sch, struct nlattr *opt) 268 { 269 struct red_sched_data *q = qdisc_priv(sch); 270 271 q->qdisc = &noop_qdisc; 272 q->sch = sch; 273 timer_setup(&q->adapt_timer, red_adaptative_timer, 0); 274 return red_change(sch, opt); 275 } 276 277 static int red_dump_offload(struct Qdisc *sch, struct tc_red_qopt *opt) 278 { 279 struct net_device *dev = qdisc_dev(sch); 280 struct tc_red_qopt_offload hw_stats = { 281 .command = TC_RED_STATS, 282 .handle = sch->handle, 283 .parent = sch->parent, 284 { 285 .stats.bstats = &sch->bstats, 286 .stats.qstats = &sch->qstats, 287 }, 288 }; 289 int err; 290 291 opt->flags &= ~TC_RED_OFFLOADED; 292 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 293 return 0; 294 295 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, 296 &hw_stats); 297 if (err == -EOPNOTSUPP) 298 return 0; 299 300 if (!err) 301 opt->flags |= TC_RED_OFFLOADED; 302 303 return err; 304 } 305 306 static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 307 { 308 struct red_sched_data *q = qdisc_priv(sch); 309 struct nlattr *opts = NULL; 310 struct tc_red_qopt opt = { 311 .limit = q->limit, 312 .flags = q->flags, 313 .qth_min = q->parms.qth_min >> q->parms.Wlog, 314 .qth_max = q->parms.qth_max >> q->parms.Wlog, 315 .Wlog = q->parms.Wlog, 316 .Plog = q->parms.Plog, 317 .Scell_log = q->parms.Scell_log, 318 }; 319 int err; 320 321 sch->qstats.backlog = q->qdisc->qstats.backlog; 322 err = red_dump_offload(sch, &opt); 323 if (err) 324 goto nla_put_failure; 325 326 opts = nla_nest_start(skb, TCA_OPTIONS); 327 if (opts == NULL) 328 goto nla_put_failure; 329 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || 330 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P)) 331 goto nla_put_failure; 332 return nla_nest_end(skb, opts); 333 334 nla_put_failure: 335 nla_nest_cancel(skb, opts); 336 return -EMSGSIZE; 337 } 338 339 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 340 { 341 struct red_sched_data *q = qdisc_priv(sch); 342 struct net_device *dev = qdisc_dev(sch); 343 struct tc_red_xstats st = { 344 .early = q->stats.prob_drop + q->stats.forced_drop, 345 .pdrop = q->stats.pdrop, 346 .other = q->stats.other, 347 .marked = q->stats.prob_mark + q->stats.forced_mark, 348 }; 349 350 if (tc_can_offload(dev) && dev->netdev_ops->ndo_setup_tc) { 351 struct red_stats hw_stats = {0}; 352 struct tc_red_qopt_offload hw_stats_request = { 353 .command = TC_RED_XSTATS, 354 .handle = sch->handle, 355 .parent = sch->parent, 356 { 357 .xstats = &hw_stats, 358 }, 359 }; 360 if (!dev->netdev_ops->ndo_setup_tc(dev, 361 TC_SETUP_QDISC_RED, 362 &hw_stats_request)) { 363 st.early += hw_stats.prob_drop + hw_stats.forced_drop; 364 st.pdrop += hw_stats.pdrop; 365 st.other += hw_stats.other; 366 st.marked += hw_stats.prob_mark + hw_stats.forced_mark; 367 } 368 } 369 370 return gnet_stats_copy_app(d, &st, sizeof(st)); 371 } 372 373 static int red_dump_class(struct Qdisc *sch, unsigned long cl, 374 struct sk_buff *skb, struct tcmsg *tcm) 375 { 376 struct red_sched_data *q = qdisc_priv(sch); 377 378 tcm->tcm_handle |= TC_H_MIN(1); 379 tcm->tcm_info = q->qdisc->handle; 380 return 0; 381 } 382 383 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 384 struct Qdisc **old) 385 { 386 struct red_sched_data *q = qdisc_priv(sch); 387 388 if (new == NULL) 389 new = &noop_qdisc; 390 391 *old = qdisc_replace(sch, new, &q->qdisc); 392 return 0; 393 } 394 395 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) 396 { 397 struct red_sched_data *q = qdisc_priv(sch); 398 return q->qdisc; 399 } 400 401 static unsigned long red_find(struct Qdisc *sch, u32 classid) 402 { 403 return 1; 404 } 405 406 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) 407 { 408 if (!walker->stop) { 409 if (walker->count >= walker->skip) 410 if (walker->fn(sch, 1, walker) < 0) { 411 walker->stop = 1; 412 return; 413 } 414 walker->count++; 415 } 416 } 417 418 static const struct Qdisc_class_ops red_class_ops = { 419 .graft = red_graft, 420 .leaf = red_leaf, 421 .find = red_find, 422 .walk = red_walk, 423 .dump = red_dump_class, 424 }; 425 426 static struct Qdisc_ops red_qdisc_ops __read_mostly = { 427 .id = "red", 428 .priv_size = sizeof(struct red_sched_data), 429 .cl_ops = &red_class_ops, 430 .enqueue = red_enqueue, 431 .dequeue = red_dequeue, 432 .peek = red_peek, 433 .init = red_init, 434 .reset = red_reset, 435 .destroy = red_destroy, 436 .change = red_change, 437 .dump = red_dump, 438 .dump_stats = red_dump_stats, 439 .owner = THIS_MODULE, 440 }; 441 442 static int __init red_module_init(void) 443 { 444 return register_qdisc(&red_qdisc_ops); 445 } 446 447 static void __exit red_module_exit(void) 448 { 449 unregister_qdisc(&red_qdisc_ops); 450 } 451 452 module_init(red_module_init) 453 module_exit(red_module_exit) 454 455 MODULE_LICENSE("GPL"); 456