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.limit = q->limit; 170 opt.set.is_ecn = red_use_ecn(q); 171 opt.set.is_harddrop = red_use_harddrop(q); 172 opt.set.qstats = &sch->qstats; 173 } else { 174 opt.command = TC_RED_DESTROY; 175 } 176 177 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt); 178 } 179 180 static void red_destroy(struct Qdisc *sch) 181 { 182 struct red_sched_data *q = qdisc_priv(sch); 183 184 del_timer_sync(&q->adapt_timer); 185 red_offload(sch, false); 186 qdisc_put(q->qdisc); 187 } 188 189 static const struct nla_policy red_policy[TCA_RED_MAX + 1] = { 190 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) }, 191 [TCA_RED_STAB] = { .len = RED_STAB_SIZE }, 192 [TCA_RED_MAX_P] = { .type = NLA_U32 }, 193 }; 194 195 static int red_change(struct Qdisc *sch, struct nlattr *opt, 196 struct netlink_ext_ack *extack) 197 { 198 struct Qdisc *old_child = NULL, *child = NULL; 199 struct red_sched_data *q = qdisc_priv(sch); 200 struct nlattr *tb[TCA_RED_MAX + 1]; 201 struct tc_red_qopt *ctl; 202 int err; 203 u32 max_P; 204 205 if (opt == NULL) 206 return -EINVAL; 207 208 err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL); 209 if (err < 0) 210 return err; 211 212 if (tb[TCA_RED_PARMS] == NULL || 213 tb[TCA_RED_STAB] == NULL) 214 return -EINVAL; 215 216 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0; 217 218 ctl = nla_data(tb[TCA_RED_PARMS]); 219 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog)) 220 return -EINVAL; 221 222 if (ctl->limit > 0) { 223 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit, 224 extack); 225 if (IS_ERR(child)) 226 return PTR_ERR(child); 227 228 /* child is fifo, no need to check for noop_qdisc */ 229 qdisc_hash_add(child, true); 230 } 231 232 sch_tree_lock(sch); 233 q->flags = ctl->flags; 234 q->limit = ctl->limit; 235 if (child) { 236 qdisc_tree_flush_backlog(q->qdisc); 237 old_child = q->qdisc; 238 q->qdisc = child; 239 } 240 241 red_set_parms(&q->parms, 242 ctl->qth_min, ctl->qth_max, ctl->Wlog, 243 ctl->Plog, ctl->Scell_log, 244 nla_data(tb[TCA_RED_STAB]), 245 max_P); 246 red_set_vars(&q->vars); 247 248 del_timer(&q->adapt_timer); 249 if (ctl->flags & TC_RED_ADAPTATIVE) 250 mod_timer(&q->adapt_timer, jiffies + HZ/2); 251 252 if (!q->qdisc->q.qlen) 253 red_start_of_idle_period(&q->vars); 254 255 sch_tree_unlock(sch); 256 257 red_offload(sch, true); 258 259 if (old_child) 260 qdisc_put(old_child); 261 return 0; 262 } 263 264 static inline void red_adaptative_timer(struct timer_list *t) 265 { 266 struct red_sched_data *q = from_timer(q, t, adapt_timer); 267 struct Qdisc *sch = q->sch; 268 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 269 270 spin_lock(root_lock); 271 red_adaptative_algo(&q->parms, &q->vars); 272 mod_timer(&q->adapt_timer, jiffies + HZ/2); 273 spin_unlock(root_lock); 274 } 275 276 static int red_init(struct Qdisc *sch, struct nlattr *opt, 277 struct netlink_ext_ack *extack) 278 { 279 struct red_sched_data *q = qdisc_priv(sch); 280 281 q->qdisc = &noop_qdisc; 282 q->sch = sch; 283 timer_setup(&q->adapt_timer, red_adaptative_timer, 0); 284 return red_change(sch, opt, extack); 285 } 286 287 static int red_dump_offload_stats(struct Qdisc *sch) 288 { 289 struct tc_red_qopt_offload hw_stats = { 290 .command = TC_RED_STATS, 291 .handle = sch->handle, 292 .parent = sch->parent, 293 { 294 .stats.bstats = &sch->bstats, 295 .stats.qstats = &sch->qstats, 296 }, 297 }; 298 299 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_RED, &hw_stats); 300 } 301 302 static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 303 { 304 struct red_sched_data *q = qdisc_priv(sch); 305 struct nlattr *opts = NULL; 306 struct tc_red_qopt opt = { 307 .limit = q->limit, 308 .flags = q->flags, 309 .qth_min = q->parms.qth_min >> q->parms.Wlog, 310 .qth_max = q->parms.qth_max >> q->parms.Wlog, 311 .Wlog = q->parms.Wlog, 312 .Plog = q->parms.Plog, 313 .Scell_log = q->parms.Scell_log, 314 }; 315 int err; 316 317 err = red_dump_offload_stats(sch); 318 if (err) 319 goto nla_put_failure; 320 321 opts = nla_nest_start_noflag(skb, TCA_OPTIONS); 322 if (opts == NULL) 323 goto nla_put_failure; 324 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || 325 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P)) 326 goto nla_put_failure; 327 return nla_nest_end(skb, opts); 328 329 nla_put_failure: 330 nla_nest_cancel(skb, opts); 331 return -EMSGSIZE; 332 } 333 334 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 335 { 336 struct red_sched_data *q = qdisc_priv(sch); 337 struct net_device *dev = qdisc_dev(sch); 338 struct tc_red_xstats st = {0}; 339 340 if (sch->flags & TCQ_F_OFFLOADED) { 341 struct tc_red_qopt_offload hw_stats_request = { 342 .command = TC_RED_XSTATS, 343 .handle = sch->handle, 344 .parent = sch->parent, 345 { 346 .xstats = &q->stats, 347 }, 348 }; 349 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, 350 &hw_stats_request); 351 } 352 st.early = q->stats.prob_drop + q->stats.forced_drop; 353 st.pdrop = q->stats.pdrop; 354 st.other = q->stats.other; 355 st.marked = q->stats.prob_mark + q->stats.forced_mark; 356 357 return gnet_stats_copy_app(d, &st, sizeof(st)); 358 } 359 360 static int red_dump_class(struct Qdisc *sch, unsigned long cl, 361 struct sk_buff *skb, struct tcmsg *tcm) 362 { 363 struct red_sched_data *q = qdisc_priv(sch); 364 365 tcm->tcm_handle |= TC_H_MIN(1); 366 tcm->tcm_info = q->qdisc->handle; 367 return 0; 368 } 369 370 static void red_graft_offload(struct Qdisc *sch, 371 struct Qdisc *new, struct Qdisc *old, 372 struct netlink_ext_ack *extack) 373 { 374 struct tc_red_qopt_offload graft_offload = { 375 .handle = sch->handle, 376 .parent = sch->parent, 377 .child_handle = new->handle, 378 .command = TC_RED_GRAFT, 379 }; 380 381 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, old, 382 TC_SETUP_QDISC_RED, &graft_offload, extack); 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); 394 395 red_graft_offload(sch, new, *old, extack); 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