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 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 55 static inline int red_use_ecn(struct red_sched_data *q) 56 { 57 return q->flags & TC_RED_ECN; 58 } 59 60 static inline int red_use_harddrop(struct red_sched_data *q) 61 { 62 return q->flags & TC_RED_HARDDROP; 63 } 64 65 static int red_use_nodrop(struct red_sched_data *q) 66 { 67 return q->flags & TC_RED_NODROP; 68 } 69 70 static 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; 138 139 congestion_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 } 147 148 static 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 } 165 166 static 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 } 173 174 static 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 } 183 184 static 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 } 212 213 static 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 } 223 224 static 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 }; 233 234 static 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; 245 246 if (tb[TCA_RED_PARMS] == NULL || 247 tb[TCA_RED_STAB] == NULL) 248 return -EINVAL; 249 250 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0; 251 252 ctl = nla_data(tb[TCA_RED_PARMS]); 253 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog)) 254 return -EINVAL; 255 256 err = red_get_flags(ctl->flags, TC_RED_HISTORIC_FLAGS, 257 tb[TCA_RED_FLAGS], TC_RED_SUPPORTED_FLAGS, 258 &flags_bf, &userbits, extack); 259 if (err) 260 return err; 261 262 if (ctl->limit > 0) { 263 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit, 264 extack); 265 if (IS_ERR(child)) 266 return PTR_ERR(child); 267 268 /* child is fifo, no need to check for noop_qdisc */ 269 qdisc_hash_add(child, true); 270 } 271 272 sch_tree_lock(sch); 273 274 flags = (q->flags & ~flags_bf.selector) | flags_bf.value; 275 err = red_validate_flags(flags, extack); 276 if (err) 277 goto unlock_out; 278 279 q->flags = flags; 280 q->userbits = userbits; 281 q->limit = ctl->limit; 282 if (child) { 283 qdisc_tree_flush_backlog(q->qdisc); 284 old_child = q->qdisc; 285 q->qdisc = child; 286 } 287 288 red_set_parms(&q->parms, 289 ctl->qth_min, ctl->qth_max, ctl->Wlog, 290 ctl->Plog, ctl->Scell_log, 291 nla_data(tb[TCA_RED_STAB]), 292 max_P); 293 red_set_vars(&q->vars); 294 295 del_timer(&q->adapt_timer); 296 if (ctl->flags & TC_RED_ADAPTATIVE) 297 mod_timer(&q->adapt_timer, jiffies + HZ/2); 298 299 if (!q->qdisc->q.qlen) 300 red_start_of_idle_period(&q->vars); 301 302 sch_tree_unlock(sch); 303 304 red_offload(sch, true); 305 306 if (old_child) 307 qdisc_put(old_child); 308 return 0; 309 310 unlock_out: 311 sch_tree_unlock(sch); 312 if (child) 313 qdisc_put(child); 314 return err; 315 } 316 317 static inline void red_adaptative_timer(struct timer_list *t) 318 { 319 struct red_sched_data *q = from_timer(q, t, adapt_timer); 320 struct Qdisc *sch = q->sch; 321 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 322 323 spin_lock(root_lock); 324 red_adaptative_algo(&q->parms, &q->vars); 325 mod_timer(&q->adapt_timer, jiffies + HZ/2); 326 spin_unlock(root_lock); 327 } 328 329 static int red_init(struct Qdisc *sch, struct nlattr *opt, 330 struct netlink_ext_ack *extack) 331 { 332 struct red_sched_data *q = qdisc_priv(sch); 333 struct nlattr *tb[TCA_RED_MAX + 1]; 334 int err; 335 336 q->qdisc = &noop_qdisc; 337 q->sch = sch; 338 timer_setup(&q->adapt_timer, red_adaptative_timer, 0); 339 340 if (!opt) 341 return -EINVAL; 342 343 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy, 344 extack); 345 if (err < 0) 346 return err; 347 348 err = __red_change(sch, tb, extack); 349 if (err) 350 return err; 351 352 err = tcf_qevent_init(&q->qe_early_drop, sch, 353 FLOW_BLOCK_BINDER_TYPE_RED_EARLY_DROP, 354 tb[TCA_RED_EARLY_DROP_BLOCK], extack); 355 if (err) 356 return err; 357 358 return tcf_qevent_init(&q->qe_mark, sch, 359 FLOW_BLOCK_BINDER_TYPE_RED_MARK, 360 tb[TCA_RED_MARK_BLOCK], extack); 361 } 362 363 static int red_change(struct Qdisc *sch, struct nlattr *opt, 364 struct netlink_ext_ack *extack) 365 { 366 struct red_sched_data *q = qdisc_priv(sch); 367 struct nlattr *tb[TCA_RED_MAX + 1]; 368 int err; 369 370 if (!opt) 371 return -EINVAL; 372 373 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy, 374 extack); 375 if (err < 0) 376 return err; 377 378 err = tcf_qevent_validate_change(&q->qe_early_drop, 379 tb[TCA_RED_EARLY_DROP_BLOCK], extack); 380 if (err) 381 return err; 382 383 err = tcf_qevent_validate_change(&q->qe_mark, 384 tb[TCA_RED_MARK_BLOCK], extack); 385 if (err) 386 return err; 387 388 return __red_change(sch, tb, extack); 389 } 390 391 static int red_dump_offload_stats(struct Qdisc *sch) 392 { 393 struct tc_red_qopt_offload hw_stats = { 394 .command = TC_RED_STATS, 395 .handle = sch->handle, 396 .parent = sch->parent, 397 { 398 .stats.bstats = &sch->bstats, 399 .stats.qstats = &sch->qstats, 400 }, 401 }; 402 403 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_RED, &hw_stats); 404 } 405 406 static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 407 { 408 struct red_sched_data *q = qdisc_priv(sch); 409 struct nlattr *opts = NULL; 410 struct tc_red_qopt opt = { 411 .limit = q->limit, 412 .flags = (q->flags & TC_RED_HISTORIC_FLAGS) | 413 q->userbits, 414 .qth_min = q->parms.qth_min >> q->parms.Wlog, 415 .qth_max = q->parms.qth_max >> q->parms.Wlog, 416 .Wlog = q->parms.Wlog, 417 .Plog = q->parms.Plog, 418 .Scell_log = q->parms.Scell_log, 419 }; 420 int err; 421 422 err = red_dump_offload_stats(sch); 423 if (err) 424 goto nla_put_failure; 425 426 opts = nla_nest_start_noflag(skb, TCA_OPTIONS); 427 if (opts == NULL) 428 goto nla_put_failure; 429 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || 430 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P) || 431 nla_put_bitfield32(skb, TCA_RED_FLAGS, 432 q->flags, TC_RED_SUPPORTED_FLAGS) || 433 tcf_qevent_dump(skb, TCA_RED_MARK_BLOCK, &q->qe_mark) || 434 tcf_qevent_dump(skb, TCA_RED_EARLY_DROP_BLOCK, &q->qe_early_drop)) 435 goto nla_put_failure; 436 return nla_nest_end(skb, opts); 437 438 nla_put_failure: 439 nla_nest_cancel(skb, opts); 440 return -EMSGSIZE; 441 } 442 443 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 444 { 445 struct red_sched_data *q = qdisc_priv(sch); 446 struct net_device *dev = qdisc_dev(sch); 447 struct tc_red_xstats st = {0}; 448 449 if (sch->flags & TCQ_F_OFFLOADED) { 450 struct tc_red_qopt_offload hw_stats_request = { 451 .command = TC_RED_XSTATS, 452 .handle = sch->handle, 453 .parent = sch->parent, 454 { 455 .xstats = &q->stats, 456 }, 457 }; 458 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, 459 &hw_stats_request); 460 } 461 st.early = q->stats.prob_drop + q->stats.forced_drop; 462 st.pdrop = q->stats.pdrop; 463 st.other = q->stats.other; 464 st.marked = q->stats.prob_mark + q->stats.forced_mark; 465 466 return gnet_stats_copy_app(d, &st, sizeof(st)); 467 } 468 469 static int red_dump_class(struct Qdisc *sch, unsigned long cl, 470 struct sk_buff *skb, struct tcmsg *tcm) 471 { 472 struct red_sched_data *q = qdisc_priv(sch); 473 474 tcm->tcm_handle |= TC_H_MIN(1); 475 tcm->tcm_info = q->qdisc->handle; 476 return 0; 477 } 478 479 static void red_graft_offload(struct Qdisc *sch, 480 struct Qdisc *new, struct Qdisc *old, 481 struct netlink_ext_ack *extack) 482 { 483 struct tc_red_qopt_offload graft_offload = { 484 .handle = sch->handle, 485 .parent = sch->parent, 486 .child_handle = new->handle, 487 .command = TC_RED_GRAFT, 488 }; 489 490 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, old, 491 TC_SETUP_QDISC_RED, &graft_offload, extack); 492 } 493 494 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 495 struct Qdisc **old, struct netlink_ext_ack *extack) 496 { 497 struct red_sched_data *q = qdisc_priv(sch); 498 499 if (new == NULL) 500 new = &noop_qdisc; 501 502 *old = qdisc_replace(sch, new, &q->qdisc); 503 504 red_graft_offload(sch, new, *old, extack); 505 return 0; 506 } 507 508 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) 509 { 510 struct red_sched_data *q = qdisc_priv(sch); 511 return q->qdisc; 512 } 513 514 static unsigned long red_find(struct Qdisc *sch, u32 classid) 515 { 516 return 1; 517 } 518 519 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) 520 { 521 if (!walker->stop) { 522 if (walker->count >= walker->skip) 523 if (walker->fn(sch, 1, walker) < 0) { 524 walker->stop = 1; 525 return; 526 } 527 walker->count++; 528 } 529 } 530 531 static const struct Qdisc_class_ops red_class_ops = { 532 .graft = red_graft, 533 .leaf = red_leaf, 534 .find = red_find, 535 .walk = red_walk, 536 .dump = red_dump_class, 537 }; 538 539 static struct Qdisc_ops red_qdisc_ops __read_mostly = { 540 .id = "red", 541 .priv_size = sizeof(struct red_sched_data), 542 .cl_ops = &red_class_ops, 543 .enqueue = red_enqueue, 544 .dequeue = red_dequeue, 545 .peek = red_peek, 546 .init = red_init, 547 .reset = red_reset, 548 .destroy = red_destroy, 549 .change = red_change, 550 .dump = red_dump, 551 .dump_stats = red_dump_stats, 552 .owner = THIS_MODULE, 553 }; 554 555 static int __init red_module_init(void) 556 { 557 return register_qdisc(&red_qdisc_ops); 558 } 559 560 static void __exit red_module_exit(void) 561 { 562 unregister_qdisc(&red_qdisc_ops); 563 } 564 565 module_init(red_module_init) 566 module_exit(red_module_exit) 567 568 MODULE_LICENSE("GPL"); 569