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