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 struct netlink_ext_ack *extack) 194 { 195 struct red_sched_data *q = qdisc_priv(sch); 196 struct nlattr *tb[TCA_RED_MAX + 1]; 197 struct tc_red_qopt *ctl; 198 struct Qdisc *child = NULL; 199 int err; 200 u32 max_P; 201 202 if (opt == NULL) 203 return -EINVAL; 204 205 err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL); 206 if (err < 0) 207 return err; 208 209 if (tb[TCA_RED_PARMS] == NULL || 210 tb[TCA_RED_STAB] == NULL) 211 return -EINVAL; 212 213 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0; 214 215 ctl = nla_data(tb[TCA_RED_PARMS]); 216 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog)) 217 return -EINVAL; 218 219 if (ctl->limit > 0) { 220 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit, 221 extack); 222 if (IS_ERR(child)) 223 return PTR_ERR(child); 224 } 225 226 if (child != &noop_qdisc) 227 qdisc_hash_add(child, true); 228 sch_tree_lock(sch); 229 q->flags = ctl->flags; 230 q->limit = ctl->limit; 231 if (child) { 232 qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen, 233 q->qdisc->qstats.backlog); 234 qdisc_destroy(q->qdisc); 235 q->qdisc = child; 236 } 237 238 red_set_parms(&q->parms, 239 ctl->qth_min, ctl->qth_max, ctl->Wlog, 240 ctl->Plog, ctl->Scell_log, 241 nla_data(tb[TCA_RED_STAB]), 242 max_P); 243 red_set_vars(&q->vars); 244 245 del_timer(&q->adapt_timer); 246 if (ctl->flags & TC_RED_ADAPTATIVE) 247 mod_timer(&q->adapt_timer, jiffies + HZ/2); 248 249 if (!q->qdisc->q.qlen) 250 red_start_of_idle_period(&q->vars); 251 252 sch_tree_unlock(sch); 253 red_offload(sch, true); 254 return 0; 255 } 256 257 static inline void red_adaptative_timer(struct timer_list *t) 258 { 259 struct red_sched_data *q = from_timer(q, t, adapt_timer); 260 struct Qdisc *sch = q->sch; 261 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 262 263 spin_lock(root_lock); 264 red_adaptative_algo(&q->parms, &q->vars); 265 mod_timer(&q->adapt_timer, jiffies + HZ/2); 266 spin_unlock(root_lock); 267 } 268 269 static int red_init(struct Qdisc *sch, struct nlattr *opt, 270 struct netlink_ext_ack *extack) 271 { 272 struct red_sched_data *q = qdisc_priv(sch); 273 274 q->qdisc = &noop_qdisc; 275 q->sch = sch; 276 timer_setup(&q->adapt_timer, red_adaptative_timer, 0); 277 return red_change(sch, opt, extack); 278 } 279 280 static int red_dump_offload_stats(struct Qdisc *sch, struct tc_red_qopt *opt) 281 { 282 struct net_device *dev = qdisc_dev(sch); 283 struct tc_red_qopt_offload hw_stats = { 284 .command = TC_RED_STATS, 285 .handle = sch->handle, 286 .parent = sch->parent, 287 { 288 .stats.bstats = &sch->bstats, 289 .stats.qstats = &sch->qstats, 290 }, 291 }; 292 int err; 293 294 sch->flags &= ~TCQ_F_OFFLOADED; 295 296 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 297 return 0; 298 299 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, 300 &hw_stats); 301 if (err == -EOPNOTSUPP) 302 return 0; 303 304 if (!err) 305 sch->flags |= TCQ_F_OFFLOADED; 306 307 return err; 308 } 309 310 static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 311 { 312 struct red_sched_data *q = qdisc_priv(sch); 313 struct nlattr *opts = NULL; 314 struct tc_red_qopt opt = { 315 .limit = q->limit, 316 .flags = q->flags, 317 .qth_min = q->parms.qth_min >> q->parms.Wlog, 318 .qth_max = q->parms.qth_max >> q->parms.Wlog, 319 .Wlog = q->parms.Wlog, 320 .Plog = q->parms.Plog, 321 .Scell_log = q->parms.Scell_log, 322 }; 323 int err; 324 325 sch->qstats.backlog = q->qdisc->qstats.backlog; 326 err = red_dump_offload_stats(sch, &opt); 327 if (err) 328 goto nla_put_failure; 329 330 opts = nla_nest_start(skb, TCA_OPTIONS); 331 if (opts == NULL) 332 goto nla_put_failure; 333 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || 334 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P)) 335 goto nla_put_failure; 336 return nla_nest_end(skb, opts); 337 338 nla_put_failure: 339 nla_nest_cancel(skb, opts); 340 return -EMSGSIZE; 341 } 342 343 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 344 { 345 struct red_sched_data *q = qdisc_priv(sch); 346 struct net_device *dev = qdisc_dev(sch); 347 struct tc_red_xstats st = { 348 .early = q->stats.prob_drop + q->stats.forced_drop, 349 .pdrop = q->stats.pdrop, 350 .other = q->stats.other, 351 .marked = q->stats.prob_mark + q->stats.forced_mark, 352 }; 353 354 if (sch->flags & TCQ_F_OFFLOADED) { 355 struct red_stats hw_stats = {0}; 356 struct tc_red_qopt_offload hw_stats_request = { 357 .command = TC_RED_XSTATS, 358 .handle = sch->handle, 359 .parent = sch->parent, 360 { 361 .xstats = &hw_stats, 362 }, 363 }; 364 if (!dev->netdev_ops->ndo_setup_tc(dev, 365 TC_SETUP_QDISC_RED, 366 &hw_stats_request)) { 367 st.early += hw_stats.prob_drop + hw_stats.forced_drop; 368 st.pdrop += hw_stats.pdrop; 369 st.other += hw_stats.other; 370 st.marked += hw_stats.prob_mark + hw_stats.forced_mark; 371 } 372 } 373 374 return gnet_stats_copy_app(d, &st, sizeof(st)); 375 } 376 377 static int red_dump_class(struct Qdisc *sch, unsigned long cl, 378 struct sk_buff *skb, struct tcmsg *tcm) 379 { 380 struct red_sched_data *q = qdisc_priv(sch); 381 382 tcm->tcm_handle |= TC_H_MIN(1); 383 tcm->tcm_info = q->qdisc->handle; 384 return 0; 385 } 386 387 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 388 struct Qdisc **old, struct netlink_ext_ack *extack) 389 { 390 struct red_sched_data *q = qdisc_priv(sch); 391 392 if (new == NULL) 393 new = &noop_qdisc; 394 395 *old = qdisc_replace(sch, new, &q->qdisc); 396 return 0; 397 } 398 399 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) 400 { 401 struct red_sched_data *q = qdisc_priv(sch); 402 return q->qdisc; 403 } 404 405 static unsigned long red_find(struct Qdisc *sch, u32 classid) 406 { 407 return 1; 408 } 409 410 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) 411 { 412 if (!walker->stop) { 413 if (walker->count >= walker->skip) 414 if (walker->fn(sch, 1, walker) < 0) { 415 walker->stop = 1; 416 return; 417 } 418 walker->count++; 419 } 420 } 421 422 static const struct Qdisc_class_ops red_class_ops = { 423 .graft = red_graft, 424 .leaf = red_leaf, 425 .find = red_find, 426 .walk = red_walk, 427 .dump = red_dump_class, 428 }; 429 430 static struct Qdisc_ops red_qdisc_ops __read_mostly = { 431 .id = "red", 432 .priv_size = sizeof(struct red_sched_data), 433 .cl_ops = &red_class_ops, 434 .enqueue = red_enqueue, 435 .dequeue = red_dequeue, 436 .peek = red_peek, 437 .init = red_init, 438 .reset = red_reset, 439 .destroy = red_destroy, 440 .change = red_change, 441 .dump = red_dump, 442 .dump_stats = red_dump_stats, 443 .owner = THIS_MODULE, 444 }; 445 446 static int __init red_module_init(void) 447 { 448 return register_qdisc(&red_qdisc_ops); 449 } 450 451 static void __exit red_module_exit(void) 452 { 453 unregister_qdisc(&red_qdisc_ops); 454 } 455 456 module_init(red_module_init) 457 module_exit(red_module_exit) 458 459 MODULE_LICENSE("GPL"); 460