1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net/sched/sch_red.c Random Early Detection queue. 4 * 5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 6 * 7 * Changes: 8 * J Hadi Salim 980914: computation fixes 9 * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly. 10 * J Hadi Salim 980816: ECN support 11 */ 12 13 #include <linux/module.h> 14 #include <linux/types.h> 15 #include <linux/kernel.h> 16 #include <linux/skbuff.h> 17 #include <net/pkt_sched.h> 18 #include <net/pkt_cls.h> 19 #include <net/inet_ecn.h> 20 #include <net/red.h> 21 22 23 /* Parameters, settable by user: 24 ----------------------------- 25 26 limit - bytes (must be > qth_max + burst) 27 28 Hard limit on queue length, should be chosen >qth_max 29 to allow packet bursts. This parameter does not 30 affect the algorithms behaviour and can be chosen 31 arbitrarily high (well, less than ram size) 32 Really, this limit will never be reached 33 if RED works correctly. 34 */ 35 36 struct red_sched_data { 37 u32 limit; /* HARD maximal queue length */ 38 unsigned char flags; 39 struct timer_list adapt_timer; 40 struct Qdisc *sch; 41 struct red_parms parms; 42 struct red_vars vars; 43 struct red_stats stats; 44 struct Qdisc *qdisc; 45 }; 46 47 static inline int red_use_ecn(struct red_sched_data *q) 48 { 49 return q->flags & TC_RED_ECN; 50 } 51 52 static inline int red_use_harddrop(struct red_sched_data *q) 53 { 54 return q->flags & TC_RED_HARDDROP; 55 } 56 57 static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch, 58 struct sk_buff **to_free) 59 { 60 struct red_sched_data *q = qdisc_priv(sch); 61 struct Qdisc *child = q->qdisc; 62 int ret; 63 64 q->vars.qavg = red_calc_qavg(&q->parms, 65 &q->vars, 66 child->qstats.backlog); 67 68 if (red_is_idling(&q->vars)) 69 red_end_of_idle_period(&q->vars); 70 71 switch (red_action(&q->parms, &q->vars, q->vars.qavg)) { 72 case RED_DONT_MARK: 73 break; 74 75 case RED_PROB_MARK: 76 qdisc_qstats_overlimit(sch); 77 if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) { 78 q->stats.prob_drop++; 79 goto congestion_drop; 80 } 81 82 q->stats.prob_mark++; 83 break; 84 85 case RED_HARD_MARK: 86 qdisc_qstats_overlimit(sch); 87 if (red_use_harddrop(q) || !red_use_ecn(q) || 88 !INET_ECN_set_ce(skb)) { 89 q->stats.forced_drop++; 90 goto congestion_drop; 91 } 92 93 q->stats.forced_mark++; 94 break; 95 } 96 97 ret = qdisc_enqueue(skb, child, to_free); 98 if (likely(ret == NET_XMIT_SUCCESS)) { 99 qdisc_qstats_backlog_inc(sch, skb); 100 sch->q.qlen++; 101 } else if (net_xmit_drop_count(ret)) { 102 q->stats.pdrop++; 103 qdisc_qstats_drop(sch); 104 } 105 return ret; 106 107 congestion_drop: 108 qdisc_drop(skb, sch, to_free); 109 return NET_XMIT_CN; 110 } 111 112 static struct sk_buff *red_dequeue(struct Qdisc *sch) 113 { 114 struct sk_buff *skb; 115 struct red_sched_data *q = qdisc_priv(sch); 116 struct Qdisc *child = q->qdisc; 117 118 skb = child->dequeue(child); 119 if (skb) { 120 qdisc_bstats_update(sch, skb); 121 qdisc_qstats_backlog_dec(sch, skb); 122 sch->q.qlen--; 123 } else { 124 if (!red_is_idling(&q->vars)) 125 red_start_of_idle_period(&q->vars); 126 } 127 return skb; 128 } 129 130 static struct sk_buff *red_peek(struct Qdisc *sch) 131 { 132 struct red_sched_data *q = qdisc_priv(sch); 133 struct Qdisc *child = q->qdisc; 134 135 return child->ops->peek(child); 136 } 137 138 static void red_reset(struct Qdisc *sch) 139 { 140 struct red_sched_data *q = qdisc_priv(sch); 141 142 qdisc_reset(q->qdisc); 143 sch->qstats.backlog = 0; 144 sch->q.qlen = 0; 145 red_restart(&q->vars); 146 } 147 148 static int red_offload(struct Qdisc *sch, bool enable) 149 { 150 struct red_sched_data *q = qdisc_priv(sch); 151 struct net_device *dev = qdisc_dev(sch); 152 struct tc_red_qopt_offload opt = { 153 .handle = sch->handle, 154 .parent = sch->parent, 155 }; 156 157 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 158 return -EOPNOTSUPP; 159 160 if (enable) { 161 opt.command = TC_RED_REPLACE; 162 opt.set.min = q->parms.qth_min >> q->parms.Wlog; 163 opt.set.max = q->parms.qth_max >> q->parms.Wlog; 164 opt.set.probability = q->parms.max_P; 165 opt.set.limit = q->limit; 166 opt.set.is_ecn = red_use_ecn(q); 167 opt.set.is_harddrop = red_use_harddrop(q); 168 opt.set.qstats = &sch->qstats; 169 } else { 170 opt.command = TC_RED_DESTROY; 171 } 172 173 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt); 174 } 175 176 static void red_destroy(struct Qdisc *sch) 177 { 178 struct red_sched_data *q = qdisc_priv(sch); 179 180 del_timer_sync(&q->adapt_timer); 181 red_offload(sch, false); 182 qdisc_put(q->qdisc); 183 } 184 185 static const struct nla_policy red_policy[TCA_RED_MAX + 1] = { 186 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) }, 187 [TCA_RED_STAB] = { .len = RED_STAB_SIZE }, 188 [TCA_RED_MAX_P] = { .type = NLA_U32 }, 189 }; 190 191 static int red_change(struct Qdisc *sch, struct nlattr *opt, 192 struct netlink_ext_ack *extack) 193 { 194 struct Qdisc *old_child = NULL, *child = NULL; 195 struct red_sched_data *q = qdisc_priv(sch); 196 struct nlattr *tb[TCA_RED_MAX + 1]; 197 struct tc_red_qopt *ctl; 198 int err; 199 u32 max_P; 200 201 if (opt == NULL) 202 return -EINVAL; 203 204 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy, 205 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 /* child is fifo, no need to check for noop_qdisc */ 226 qdisc_hash_add(child, true); 227 } 228 229 sch_tree_lock(sch); 230 q->flags = ctl->flags; 231 q->limit = ctl->limit; 232 if (child) { 233 qdisc_tree_flush_backlog(q->qdisc); 234 old_child = 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); 253 254 red_offload(sch, true); 255 256 if (old_child) 257 qdisc_put(old_child); 258 return 0; 259 } 260 261 static inline void red_adaptative_timer(struct timer_list *t) 262 { 263 struct red_sched_data *q = from_timer(q, t, adapt_timer); 264 struct Qdisc *sch = q->sch; 265 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 266 267 spin_lock(root_lock); 268 red_adaptative_algo(&q->parms, &q->vars); 269 mod_timer(&q->adapt_timer, jiffies + HZ/2); 270 spin_unlock(root_lock); 271 } 272 273 static int red_init(struct Qdisc *sch, struct nlattr *opt, 274 struct netlink_ext_ack *extack) 275 { 276 struct red_sched_data *q = qdisc_priv(sch); 277 278 q->qdisc = &noop_qdisc; 279 q->sch = sch; 280 timer_setup(&q->adapt_timer, red_adaptative_timer, 0); 281 return red_change(sch, opt, extack); 282 } 283 284 static int red_dump_offload_stats(struct Qdisc *sch) 285 { 286 struct tc_red_qopt_offload hw_stats = { 287 .command = TC_RED_STATS, 288 .handle = sch->handle, 289 .parent = sch->parent, 290 { 291 .stats.bstats = &sch->bstats, 292 .stats.qstats = &sch->qstats, 293 }, 294 }; 295 296 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_RED, &hw_stats); 297 } 298 299 static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 300 { 301 struct red_sched_data *q = qdisc_priv(sch); 302 struct nlattr *opts = NULL; 303 struct tc_red_qopt opt = { 304 .limit = q->limit, 305 .flags = q->flags, 306 .qth_min = q->parms.qth_min >> q->parms.Wlog, 307 .qth_max = q->parms.qth_max >> q->parms.Wlog, 308 .Wlog = q->parms.Wlog, 309 .Plog = q->parms.Plog, 310 .Scell_log = q->parms.Scell_log, 311 }; 312 int err; 313 314 err = red_dump_offload_stats(sch); 315 if (err) 316 goto nla_put_failure; 317 318 opts = nla_nest_start_noflag(skb, TCA_OPTIONS); 319 if (opts == NULL) 320 goto nla_put_failure; 321 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || 322 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P)) 323 goto nla_put_failure; 324 return nla_nest_end(skb, opts); 325 326 nla_put_failure: 327 nla_nest_cancel(skb, opts); 328 return -EMSGSIZE; 329 } 330 331 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 332 { 333 struct red_sched_data *q = qdisc_priv(sch); 334 struct net_device *dev = qdisc_dev(sch); 335 struct tc_red_xstats st = {0}; 336 337 if (sch->flags & TCQ_F_OFFLOADED) { 338 struct tc_red_qopt_offload hw_stats_request = { 339 .command = TC_RED_XSTATS, 340 .handle = sch->handle, 341 .parent = sch->parent, 342 { 343 .xstats = &q->stats, 344 }, 345 }; 346 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, 347 &hw_stats_request); 348 } 349 st.early = q->stats.prob_drop + q->stats.forced_drop; 350 st.pdrop = q->stats.pdrop; 351 st.other = q->stats.other; 352 st.marked = q->stats.prob_mark + q->stats.forced_mark; 353 354 return gnet_stats_copy_app(d, &st, sizeof(st)); 355 } 356 357 static int red_dump_class(struct Qdisc *sch, unsigned long cl, 358 struct sk_buff *skb, struct tcmsg *tcm) 359 { 360 struct red_sched_data *q = qdisc_priv(sch); 361 362 tcm->tcm_handle |= TC_H_MIN(1); 363 tcm->tcm_info = q->qdisc->handle; 364 return 0; 365 } 366 367 static void red_graft_offload(struct Qdisc *sch, 368 struct Qdisc *new, struct Qdisc *old, 369 struct netlink_ext_ack *extack) 370 { 371 struct tc_red_qopt_offload graft_offload = { 372 .handle = sch->handle, 373 .parent = sch->parent, 374 .child_handle = new->handle, 375 .command = TC_RED_GRAFT, 376 }; 377 378 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, old, 379 TC_SETUP_QDISC_RED, &graft_offload, extack); 380 } 381 382 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 383 struct Qdisc **old, struct netlink_ext_ack *extack) 384 { 385 struct red_sched_data *q = qdisc_priv(sch); 386 387 if (new == NULL) 388 new = &noop_qdisc; 389 390 *old = qdisc_replace(sch, new, &q->qdisc); 391 392 red_graft_offload(sch, new, *old, extack); 393 return 0; 394 } 395 396 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) 397 { 398 struct red_sched_data *q = qdisc_priv(sch); 399 return q->qdisc; 400 } 401 402 static unsigned long red_find(struct Qdisc *sch, u32 classid) 403 { 404 return 1; 405 } 406 407 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) 408 { 409 if (!walker->stop) { 410 if (walker->count >= walker->skip) 411 if (walker->fn(sch, 1, walker) < 0) { 412 walker->stop = 1; 413 return; 414 } 415 walker->count++; 416 } 417 } 418 419 static const struct Qdisc_class_ops red_class_ops = { 420 .graft = red_graft, 421 .leaf = red_leaf, 422 .find = red_find, 423 .walk = red_walk, 424 .dump = red_dump_class, 425 }; 426 427 static struct Qdisc_ops red_qdisc_ops __read_mostly = { 428 .id = "red", 429 .priv_size = sizeof(struct red_sched_data), 430 .cl_ops = &red_class_ops, 431 .enqueue = red_enqueue, 432 .dequeue = red_dequeue, 433 .peek = red_peek, 434 .init = red_init, 435 .reset = red_reset, 436 .destroy = red_destroy, 437 .change = red_change, 438 .dump = red_dump, 439 .dump_stats = red_dump_stats, 440 .owner = THIS_MODULE, 441 }; 442 443 static int __init red_module_init(void) 444 { 445 return register_qdisc(&red_qdisc_ops); 446 } 447 448 static void __exit red_module_exit(void) 449 { 450 unregister_qdisc(&red_qdisc_ops); 451 } 452 453 module_init(red_module_init) 454 module_exit(red_module_exit) 455 456 MODULE_LICENSE("GPL"); 457