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 int err; 161 162 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 163 return -EOPNOTSUPP; 164 165 if (enable) { 166 opt.command = TC_RED_REPLACE; 167 opt.set.min = q->parms.qth_min >> q->parms.Wlog; 168 opt.set.max = q->parms.qth_max >> q->parms.Wlog; 169 opt.set.probability = q->parms.max_P; 170 opt.set.is_ecn = red_use_ecn(q); 171 } else { 172 opt.command = TC_RED_DESTROY; 173 } 174 175 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt); 176 177 if (!err && enable) 178 sch->flags |= TCQ_F_OFFLOADED; 179 else 180 sch->flags &= ~TCQ_F_OFFLOADED; 181 182 return err; 183 } 184 185 static void red_destroy(struct Qdisc *sch) 186 { 187 struct red_sched_data *q = qdisc_priv(sch); 188 189 del_timer_sync(&q->adapt_timer); 190 red_offload(sch, false); 191 qdisc_destroy(q->qdisc); 192 } 193 194 static const struct nla_policy red_policy[TCA_RED_MAX + 1] = { 195 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) }, 196 [TCA_RED_STAB] = { .len = RED_STAB_SIZE }, 197 [TCA_RED_MAX_P] = { .type = NLA_U32 }, 198 }; 199 200 static int red_change(struct Qdisc *sch, struct nlattr *opt, 201 struct netlink_ext_ack *extack) 202 { 203 struct red_sched_data *q = qdisc_priv(sch); 204 struct nlattr *tb[TCA_RED_MAX + 1]; 205 struct tc_red_qopt *ctl; 206 struct Qdisc *child = NULL; 207 int err; 208 u32 max_P; 209 210 if (opt == NULL) 211 return -EINVAL; 212 213 err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL); 214 if (err < 0) 215 return err; 216 217 if (tb[TCA_RED_PARMS] == NULL || 218 tb[TCA_RED_STAB] == NULL) 219 return -EINVAL; 220 221 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0; 222 223 ctl = nla_data(tb[TCA_RED_PARMS]); 224 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog)) 225 return -EINVAL; 226 227 if (ctl->limit > 0) { 228 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit, 229 extack); 230 if (IS_ERR(child)) 231 return PTR_ERR(child); 232 } 233 234 if (child != &noop_qdisc) 235 qdisc_hash_add(child, true); 236 sch_tree_lock(sch); 237 q->flags = ctl->flags; 238 q->limit = ctl->limit; 239 if (child) { 240 qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen, 241 q->qdisc->qstats.backlog); 242 qdisc_destroy(q->qdisc); 243 q->qdisc = child; 244 } 245 246 red_set_parms(&q->parms, 247 ctl->qth_min, ctl->qth_max, ctl->Wlog, 248 ctl->Plog, ctl->Scell_log, 249 nla_data(tb[TCA_RED_STAB]), 250 max_P); 251 red_set_vars(&q->vars); 252 253 del_timer(&q->adapt_timer); 254 if (ctl->flags & TC_RED_ADAPTATIVE) 255 mod_timer(&q->adapt_timer, jiffies + HZ/2); 256 257 if (!q->qdisc->q.qlen) 258 red_start_of_idle_period(&q->vars); 259 260 sch_tree_unlock(sch); 261 red_offload(sch, true); 262 return 0; 263 } 264 265 static inline void red_adaptative_timer(struct timer_list *t) 266 { 267 struct red_sched_data *q = from_timer(q, t, adapt_timer); 268 struct Qdisc *sch = q->sch; 269 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 270 271 spin_lock(root_lock); 272 red_adaptative_algo(&q->parms, &q->vars); 273 mod_timer(&q->adapt_timer, jiffies + HZ/2); 274 spin_unlock(root_lock); 275 } 276 277 static int red_init(struct Qdisc *sch, struct nlattr *opt, 278 struct netlink_ext_ack *extack) 279 { 280 struct red_sched_data *q = qdisc_priv(sch); 281 282 q->qdisc = &noop_qdisc; 283 q->sch = sch; 284 timer_setup(&q->adapt_timer, red_adaptative_timer, 0); 285 return red_change(sch, opt, extack); 286 } 287 288 static int red_dump_offload_stats(struct Qdisc *sch, struct tc_red_qopt *opt) 289 { 290 struct net_device *dev = qdisc_dev(sch); 291 struct tc_red_qopt_offload hw_stats = { 292 .command = TC_RED_STATS, 293 .handle = sch->handle, 294 .parent = sch->parent, 295 { 296 .stats.bstats = &sch->bstats, 297 .stats.qstats = &sch->qstats, 298 }, 299 }; 300 301 if (!(sch->flags & TCQ_F_OFFLOADED)) 302 return 0; 303 304 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, 305 &hw_stats); 306 } 307 308 static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 309 { 310 struct red_sched_data *q = qdisc_priv(sch); 311 struct nlattr *opts = NULL; 312 struct tc_red_qopt opt = { 313 .limit = q->limit, 314 .flags = q->flags, 315 .qth_min = q->parms.qth_min >> q->parms.Wlog, 316 .qth_max = q->parms.qth_max >> q->parms.Wlog, 317 .Wlog = q->parms.Wlog, 318 .Plog = q->parms.Plog, 319 .Scell_log = q->parms.Scell_log, 320 }; 321 int err; 322 323 sch->qstats.backlog = q->qdisc->qstats.backlog; 324 err = red_dump_offload_stats(sch, &opt); 325 if (err) 326 goto nla_put_failure; 327 328 opts = nla_nest_start(skb, TCA_OPTIONS); 329 if (opts == NULL) 330 goto nla_put_failure; 331 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || 332 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P)) 333 goto nla_put_failure; 334 return nla_nest_end(skb, opts); 335 336 nla_put_failure: 337 nla_nest_cancel(skb, opts); 338 return -EMSGSIZE; 339 } 340 341 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 342 { 343 struct red_sched_data *q = qdisc_priv(sch); 344 struct net_device *dev = qdisc_dev(sch); 345 struct tc_red_xstats st = { 346 .early = q->stats.prob_drop + q->stats.forced_drop, 347 .pdrop = q->stats.pdrop, 348 .other = q->stats.other, 349 .marked = q->stats.prob_mark + q->stats.forced_mark, 350 }; 351 352 if (sch->flags & TCQ_F_OFFLOADED) { 353 struct red_stats hw_stats = {0}; 354 struct tc_red_qopt_offload hw_stats_request = { 355 .command = TC_RED_XSTATS, 356 .handle = sch->handle, 357 .parent = sch->parent, 358 { 359 .xstats = &hw_stats, 360 }, 361 }; 362 if (!dev->netdev_ops->ndo_setup_tc(dev, 363 TC_SETUP_QDISC_RED, 364 &hw_stats_request)) { 365 st.early += hw_stats.prob_drop + hw_stats.forced_drop; 366 st.pdrop += hw_stats.pdrop; 367 st.other += hw_stats.other; 368 st.marked += hw_stats.prob_mark + hw_stats.forced_mark; 369 } 370 } 371 372 return gnet_stats_copy_app(d, &st, sizeof(st)); 373 } 374 375 static int red_dump_class(struct Qdisc *sch, unsigned long cl, 376 struct sk_buff *skb, struct tcmsg *tcm) 377 { 378 struct red_sched_data *q = qdisc_priv(sch); 379 380 tcm->tcm_handle |= TC_H_MIN(1); 381 tcm->tcm_info = q->qdisc->handle; 382 return 0; 383 } 384 385 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 386 struct Qdisc **old, struct netlink_ext_ack *extack) 387 { 388 struct red_sched_data *q = qdisc_priv(sch); 389 390 if (new == NULL) 391 new = &noop_qdisc; 392 393 *old = qdisc_replace(sch, new, &q->qdisc); 394 return 0; 395 } 396 397 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) 398 { 399 struct red_sched_data *q = qdisc_priv(sch); 400 return q->qdisc; 401 } 402 403 static unsigned long red_find(struct Qdisc *sch, u32 classid) 404 { 405 return 1; 406 } 407 408 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) 409 { 410 if (!walker->stop) { 411 if (walker->count >= walker->skip) 412 if (walker->fn(sch, 1, walker) < 0) { 413 walker->stop = 1; 414 return; 415 } 416 walker->count++; 417 } 418 } 419 420 static const struct Qdisc_class_ops red_class_ops = { 421 .graft = red_graft, 422 .leaf = red_leaf, 423 .find = red_find, 424 .walk = red_walk, 425 .dump = red_dump_class, 426 }; 427 428 static struct Qdisc_ops red_qdisc_ops __read_mostly = { 429 .id = "red", 430 .priv_size = sizeof(struct red_sched_data), 431 .cl_ops = &red_class_ops, 432 .enqueue = red_enqueue, 433 .dequeue = red_dequeue, 434 .peek = red_peek, 435 .init = red_init, 436 .reset = red_reset, 437 .destroy = red_destroy, 438 .change = red_change, 439 .dump = red_dump, 440 .dump_stats = red_dump_stats, 441 .owner = THIS_MODULE, 442 }; 443 444 static int __init red_module_init(void) 445 { 446 return register_qdisc(&red_qdisc_ops); 447 } 448 449 static void __exit red_module_exit(void) 450 { 451 unregister_qdisc(&red_qdisc_ops); 452 } 453 454 module_init(red_module_init) 455 module_exit(red_module_exit) 456 457 MODULE_LICENSE("GPL"); 458