sch_red.c (13735B)
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 36struct red_sched_data { 37 u32 limit; /* HARD maximal queue length */ 38 39 unsigned char flags; 40 /* Non-flags in tc_red_qopt.flags. */ 41 unsigned char userbits; 42 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 struct tcf_qevent qe_early_drop; 50 struct tcf_qevent qe_mark; 51}; 52 53#define TC_RED_SUPPORTED_FLAGS (TC_RED_HISTORIC_FLAGS | TC_RED_NODROP) 54 55static inline int red_use_ecn(struct red_sched_data *q) 56{ 57 return q->flags & TC_RED_ECN; 58} 59 60static inline int red_use_harddrop(struct red_sched_data *q) 61{ 62 return q->flags & TC_RED_HARDDROP; 63} 64 65static int red_use_nodrop(struct red_sched_data *q) 66{ 67 return q->flags & TC_RED_NODROP; 68} 69 70static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch, 71 struct sk_buff **to_free) 72{ 73 struct red_sched_data *q = qdisc_priv(sch); 74 struct Qdisc *child = q->qdisc; 75 int ret; 76 77 q->vars.qavg = red_calc_qavg(&q->parms, 78 &q->vars, 79 child->qstats.backlog); 80 81 if (red_is_idling(&q->vars)) 82 red_end_of_idle_period(&q->vars); 83 84 switch (red_action(&q->parms, &q->vars, q->vars.qavg)) { 85 case RED_DONT_MARK: 86 break; 87 88 case RED_PROB_MARK: 89 qdisc_qstats_overlimit(sch); 90 if (!red_use_ecn(q)) { 91 q->stats.prob_drop++; 92 goto congestion_drop; 93 } 94 95 if (INET_ECN_set_ce(skb)) { 96 q->stats.prob_mark++; 97 skb = tcf_qevent_handle(&q->qe_mark, sch, skb, to_free, &ret); 98 if (!skb) 99 return NET_XMIT_CN | ret; 100 } else if (!red_use_nodrop(q)) { 101 q->stats.prob_drop++; 102 goto congestion_drop; 103 } 104 105 /* Non-ECT packet in ECN nodrop mode: queue it. */ 106 break; 107 108 case RED_HARD_MARK: 109 qdisc_qstats_overlimit(sch); 110 if (red_use_harddrop(q) || !red_use_ecn(q)) { 111 q->stats.forced_drop++; 112 goto congestion_drop; 113 } 114 115 if (INET_ECN_set_ce(skb)) { 116 q->stats.forced_mark++; 117 skb = tcf_qevent_handle(&q->qe_mark, sch, skb, to_free, &ret); 118 if (!skb) 119 return NET_XMIT_CN | ret; 120 } else if (!red_use_nodrop(q)) { 121 q->stats.forced_drop++; 122 goto congestion_drop; 123 } 124 125 /* Non-ECT packet in ECN nodrop mode: queue it. */ 126 break; 127 } 128 129 ret = qdisc_enqueue(skb, child, to_free); 130 if (likely(ret == NET_XMIT_SUCCESS)) { 131 qdisc_qstats_backlog_inc(sch, skb); 132 sch->q.qlen++; 133 } else if (net_xmit_drop_count(ret)) { 134 q->stats.pdrop++; 135 qdisc_qstats_drop(sch); 136 } 137 return ret; 138 139congestion_drop: 140 skb = tcf_qevent_handle(&q->qe_early_drop, sch, skb, to_free, &ret); 141 if (!skb) 142 return NET_XMIT_CN | ret; 143 144 qdisc_drop(skb, sch, to_free); 145 return NET_XMIT_CN; 146} 147 148static struct sk_buff *red_dequeue(struct Qdisc *sch) 149{ 150 struct sk_buff *skb; 151 struct red_sched_data *q = qdisc_priv(sch); 152 struct Qdisc *child = q->qdisc; 153 154 skb = child->dequeue(child); 155 if (skb) { 156 qdisc_bstats_update(sch, skb); 157 qdisc_qstats_backlog_dec(sch, skb); 158 sch->q.qlen--; 159 } else { 160 if (!red_is_idling(&q->vars)) 161 red_start_of_idle_period(&q->vars); 162 } 163 return skb; 164} 165 166static struct sk_buff *red_peek(struct Qdisc *sch) 167{ 168 struct red_sched_data *q = qdisc_priv(sch); 169 struct Qdisc *child = q->qdisc; 170 171 return child->ops->peek(child); 172} 173 174static void red_reset(struct Qdisc *sch) 175{ 176 struct red_sched_data *q = qdisc_priv(sch); 177 178 qdisc_reset(q->qdisc); 179 sch->qstats.backlog = 0; 180 sch->q.qlen = 0; 181 red_restart(&q->vars); 182} 183 184static int red_offload(struct Qdisc *sch, bool enable) 185{ 186 struct red_sched_data *q = qdisc_priv(sch); 187 struct net_device *dev = qdisc_dev(sch); 188 struct tc_red_qopt_offload opt = { 189 .handle = sch->handle, 190 .parent = sch->parent, 191 }; 192 193 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 194 return -EOPNOTSUPP; 195 196 if (enable) { 197 opt.command = TC_RED_REPLACE; 198 opt.set.min = q->parms.qth_min >> q->parms.Wlog; 199 opt.set.max = q->parms.qth_max >> q->parms.Wlog; 200 opt.set.probability = q->parms.max_P; 201 opt.set.limit = q->limit; 202 opt.set.is_ecn = red_use_ecn(q); 203 opt.set.is_harddrop = red_use_harddrop(q); 204 opt.set.is_nodrop = red_use_nodrop(q); 205 opt.set.qstats = &sch->qstats; 206 } else { 207 opt.command = TC_RED_DESTROY; 208 } 209 210 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt); 211} 212 213static void red_destroy(struct Qdisc *sch) 214{ 215 struct red_sched_data *q = qdisc_priv(sch); 216 217 tcf_qevent_destroy(&q->qe_mark, sch); 218 tcf_qevent_destroy(&q->qe_early_drop, sch); 219 del_timer_sync(&q->adapt_timer); 220 red_offload(sch, false); 221 qdisc_put(q->qdisc); 222} 223 224static const struct nla_policy red_policy[TCA_RED_MAX + 1] = { 225 [TCA_RED_UNSPEC] = { .strict_start_type = TCA_RED_FLAGS }, 226 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) }, 227 [TCA_RED_STAB] = { .len = RED_STAB_SIZE }, 228 [TCA_RED_MAX_P] = { .type = NLA_U32 }, 229 [TCA_RED_FLAGS] = NLA_POLICY_BITFIELD32(TC_RED_SUPPORTED_FLAGS), 230 [TCA_RED_EARLY_DROP_BLOCK] = { .type = NLA_U32 }, 231 [TCA_RED_MARK_BLOCK] = { .type = NLA_U32 }, 232}; 233 234static int __red_change(struct Qdisc *sch, struct nlattr **tb, 235 struct netlink_ext_ack *extack) 236{ 237 struct Qdisc *old_child = NULL, *child = NULL; 238 struct red_sched_data *q = qdisc_priv(sch); 239 struct nla_bitfield32 flags_bf; 240 struct tc_red_qopt *ctl; 241 unsigned char userbits; 242 unsigned char flags; 243 int err; 244 u32 max_P; 245 u8 *stab; 246 247 if (tb[TCA_RED_PARMS] == NULL || 248 tb[TCA_RED_STAB] == NULL) 249 return -EINVAL; 250 251 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0; 252 253 ctl = nla_data(tb[TCA_RED_PARMS]); 254 stab = nla_data(tb[TCA_RED_STAB]); 255 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog, 256 ctl->Scell_log, stab)) 257 return -EINVAL; 258 259 err = red_get_flags(ctl->flags, TC_RED_HISTORIC_FLAGS, 260 tb[TCA_RED_FLAGS], TC_RED_SUPPORTED_FLAGS, 261 &flags_bf, &userbits, extack); 262 if (err) 263 return err; 264 265 if (ctl->limit > 0) { 266 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit, 267 extack); 268 if (IS_ERR(child)) 269 return PTR_ERR(child); 270 271 /* child is fifo, no need to check for noop_qdisc */ 272 qdisc_hash_add(child, true); 273 } 274 275 sch_tree_lock(sch); 276 277 flags = (q->flags & ~flags_bf.selector) | flags_bf.value; 278 err = red_validate_flags(flags, extack); 279 if (err) 280 goto unlock_out; 281 282 q->flags = flags; 283 q->userbits = userbits; 284 q->limit = ctl->limit; 285 if (child) { 286 qdisc_tree_flush_backlog(q->qdisc); 287 old_child = q->qdisc; 288 q->qdisc = child; 289 } 290 291 red_set_parms(&q->parms, 292 ctl->qth_min, ctl->qth_max, ctl->Wlog, 293 ctl->Plog, ctl->Scell_log, 294 stab, 295 max_P); 296 red_set_vars(&q->vars); 297 298 del_timer(&q->adapt_timer); 299 if (ctl->flags & TC_RED_ADAPTATIVE) 300 mod_timer(&q->adapt_timer, jiffies + HZ/2); 301 302 if (!q->qdisc->q.qlen) 303 red_start_of_idle_period(&q->vars); 304 305 sch_tree_unlock(sch); 306 307 red_offload(sch, true); 308 309 if (old_child) 310 qdisc_put(old_child); 311 return 0; 312 313unlock_out: 314 sch_tree_unlock(sch); 315 if (child) 316 qdisc_put(child); 317 return err; 318} 319 320static inline void red_adaptative_timer(struct timer_list *t) 321{ 322 struct red_sched_data *q = from_timer(q, t, adapt_timer); 323 struct Qdisc *sch = q->sch; 324 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 325 326 spin_lock(root_lock); 327 red_adaptative_algo(&q->parms, &q->vars); 328 mod_timer(&q->adapt_timer, jiffies + HZ/2); 329 spin_unlock(root_lock); 330} 331 332static int red_init(struct Qdisc *sch, struct nlattr *opt, 333 struct netlink_ext_ack *extack) 334{ 335 struct red_sched_data *q = qdisc_priv(sch); 336 struct nlattr *tb[TCA_RED_MAX + 1]; 337 int err; 338 339 q->qdisc = &noop_qdisc; 340 q->sch = sch; 341 timer_setup(&q->adapt_timer, red_adaptative_timer, 0); 342 343 if (!opt) 344 return -EINVAL; 345 346 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy, 347 extack); 348 if (err < 0) 349 return err; 350 351 err = __red_change(sch, tb, extack); 352 if (err) 353 return err; 354 355 err = tcf_qevent_init(&q->qe_early_drop, sch, 356 FLOW_BLOCK_BINDER_TYPE_RED_EARLY_DROP, 357 tb[TCA_RED_EARLY_DROP_BLOCK], extack); 358 if (err) 359 return err; 360 361 return tcf_qevent_init(&q->qe_mark, sch, 362 FLOW_BLOCK_BINDER_TYPE_RED_MARK, 363 tb[TCA_RED_MARK_BLOCK], extack); 364} 365 366static int red_change(struct Qdisc *sch, struct nlattr *opt, 367 struct netlink_ext_ack *extack) 368{ 369 struct red_sched_data *q = qdisc_priv(sch); 370 struct nlattr *tb[TCA_RED_MAX + 1]; 371 int err; 372 373 if (!opt) 374 return -EINVAL; 375 376 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy, 377 extack); 378 if (err < 0) 379 return err; 380 381 err = tcf_qevent_validate_change(&q->qe_early_drop, 382 tb[TCA_RED_EARLY_DROP_BLOCK], extack); 383 if (err) 384 return err; 385 386 err = tcf_qevent_validate_change(&q->qe_mark, 387 tb[TCA_RED_MARK_BLOCK], extack); 388 if (err) 389 return err; 390 391 return __red_change(sch, tb, extack); 392} 393 394static int red_dump_offload_stats(struct Qdisc *sch) 395{ 396 struct tc_red_qopt_offload hw_stats = { 397 .command = TC_RED_STATS, 398 .handle = sch->handle, 399 .parent = sch->parent, 400 { 401 .stats.bstats = &sch->bstats, 402 .stats.qstats = &sch->qstats, 403 }, 404 }; 405 406 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_RED, &hw_stats); 407} 408 409static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 410{ 411 struct red_sched_data *q = qdisc_priv(sch); 412 struct nlattr *opts = NULL; 413 struct tc_red_qopt opt = { 414 .limit = q->limit, 415 .flags = (q->flags & TC_RED_HISTORIC_FLAGS) | 416 q->userbits, 417 .qth_min = q->parms.qth_min >> q->parms.Wlog, 418 .qth_max = q->parms.qth_max >> q->parms.Wlog, 419 .Wlog = q->parms.Wlog, 420 .Plog = q->parms.Plog, 421 .Scell_log = q->parms.Scell_log, 422 }; 423 int err; 424 425 err = red_dump_offload_stats(sch); 426 if (err) 427 goto nla_put_failure; 428 429 opts = nla_nest_start_noflag(skb, TCA_OPTIONS); 430 if (opts == NULL) 431 goto nla_put_failure; 432 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || 433 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P) || 434 nla_put_bitfield32(skb, TCA_RED_FLAGS, 435 q->flags, TC_RED_SUPPORTED_FLAGS) || 436 tcf_qevent_dump(skb, TCA_RED_MARK_BLOCK, &q->qe_mark) || 437 tcf_qevent_dump(skb, TCA_RED_EARLY_DROP_BLOCK, &q->qe_early_drop)) 438 goto nla_put_failure; 439 return nla_nest_end(skb, opts); 440 441nla_put_failure: 442 nla_nest_cancel(skb, opts); 443 return -EMSGSIZE; 444} 445 446static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 447{ 448 struct red_sched_data *q = qdisc_priv(sch); 449 struct net_device *dev = qdisc_dev(sch); 450 struct tc_red_xstats st = {0}; 451 452 if (sch->flags & TCQ_F_OFFLOADED) { 453 struct tc_red_qopt_offload hw_stats_request = { 454 .command = TC_RED_XSTATS, 455 .handle = sch->handle, 456 .parent = sch->parent, 457 { 458 .xstats = &q->stats, 459 }, 460 }; 461 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, 462 &hw_stats_request); 463 } 464 st.early = q->stats.prob_drop + q->stats.forced_drop; 465 st.pdrop = q->stats.pdrop; 466 st.other = q->stats.other; 467 st.marked = q->stats.prob_mark + q->stats.forced_mark; 468 469 return gnet_stats_copy_app(d, &st, sizeof(st)); 470} 471 472static int red_dump_class(struct Qdisc *sch, unsigned long cl, 473 struct sk_buff *skb, struct tcmsg *tcm) 474{ 475 struct red_sched_data *q = qdisc_priv(sch); 476 477 tcm->tcm_handle |= TC_H_MIN(1); 478 tcm->tcm_info = q->qdisc->handle; 479 return 0; 480} 481 482static void red_graft_offload(struct Qdisc *sch, 483 struct Qdisc *new, struct Qdisc *old, 484 struct netlink_ext_ack *extack) 485{ 486 struct tc_red_qopt_offload graft_offload = { 487 .handle = sch->handle, 488 .parent = sch->parent, 489 .child_handle = new->handle, 490 .command = TC_RED_GRAFT, 491 }; 492 493 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, old, 494 TC_SETUP_QDISC_RED, &graft_offload, extack); 495} 496 497static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 498 struct Qdisc **old, struct netlink_ext_ack *extack) 499{ 500 struct red_sched_data *q = qdisc_priv(sch); 501 502 if (new == NULL) 503 new = &noop_qdisc; 504 505 *old = qdisc_replace(sch, new, &q->qdisc); 506 507 red_graft_offload(sch, new, *old, extack); 508 return 0; 509} 510 511static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) 512{ 513 struct red_sched_data *q = qdisc_priv(sch); 514 return q->qdisc; 515} 516 517static unsigned long red_find(struct Qdisc *sch, u32 classid) 518{ 519 return 1; 520} 521 522static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) 523{ 524 if (!walker->stop) { 525 if (walker->count >= walker->skip) 526 if (walker->fn(sch, 1, walker) < 0) { 527 walker->stop = 1; 528 return; 529 } 530 walker->count++; 531 } 532} 533 534static const struct Qdisc_class_ops red_class_ops = { 535 .graft = red_graft, 536 .leaf = red_leaf, 537 .find = red_find, 538 .walk = red_walk, 539 .dump = red_dump_class, 540}; 541 542static struct Qdisc_ops red_qdisc_ops __read_mostly = { 543 .id = "red", 544 .priv_size = sizeof(struct red_sched_data), 545 .cl_ops = &red_class_ops, 546 .enqueue = red_enqueue, 547 .dequeue = red_dequeue, 548 .peek = red_peek, 549 .init = red_init, 550 .reset = red_reset, 551 .destroy = red_destroy, 552 .change = red_change, 553 .dump = red_dump, 554 .dump_stats = red_dump_stats, 555 .owner = THIS_MODULE, 556}; 557 558static int __init red_module_init(void) 559{ 560 return register_qdisc(&red_qdisc_ops); 561} 562 563static void __exit red_module_exit(void) 564{ 565 unregister_qdisc(&red_qdisc_ops); 566} 567 568module_init(red_module_init) 569module_exit(red_module_exit) 570 571MODULE_LICENSE("GPL");