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 red_restart(&q->vars); 180 } 181 182 static int red_offload(struct Qdisc *sch, bool enable) 183 { 184 struct red_sched_data *q = qdisc_priv(sch); 185 struct net_device *dev = qdisc_dev(sch); 186 struct tc_red_qopt_offload opt = { 187 .handle = sch->handle, 188 .parent = sch->parent, 189 }; 190 191 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc) 192 return -EOPNOTSUPP; 193 194 if (enable) { 195 opt.command = TC_RED_REPLACE; 196 opt.set.min = q->parms.qth_min >> q->parms.Wlog; 197 opt.set.max = q->parms.qth_max >> q->parms.Wlog; 198 opt.set.probability = q->parms.max_P; 199 opt.set.limit = q->limit; 200 opt.set.is_ecn = red_use_ecn(q); 201 opt.set.is_harddrop = red_use_harddrop(q); 202 opt.set.is_nodrop = red_use_nodrop(q); 203 opt.set.qstats = &sch->qstats; 204 } else { 205 opt.command = TC_RED_DESTROY; 206 } 207 208 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, &opt); 209 } 210 211 static void red_destroy(struct Qdisc *sch) 212 { 213 struct red_sched_data *q = qdisc_priv(sch); 214 215 tcf_qevent_destroy(&q->qe_mark, sch); 216 tcf_qevent_destroy(&q->qe_early_drop, sch); 217 del_timer_sync(&q->adapt_timer); 218 red_offload(sch, false); 219 qdisc_put(q->qdisc); 220 } 221 222 static const struct nla_policy red_policy[TCA_RED_MAX + 1] = { 223 [TCA_RED_UNSPEC] = { .strict_start_type = TCA_RED_FLAGS }, 224 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) }, 225 [TCA_RED_STAB] = { .len = RED_STAB_SIZE }, 226 [TCA_RED_MAX_P] = { .type = NLA_U32 }, 227 [TCA_RED_FLAGS] = NLA_POLICY_BITFIELD32(TC_RED_SUPPORTED_FLAGS), 228 [TCA_RED_EARLY_DROP_BLOCK] = { .type = NLA_U32 }, 229 [TCA_RED_MARK_BLOCK] = { .type = NLA_U32 }, 230 }; 231 232 static int __red_change(struct Qdisc *sch, struct nlattr **tb, 233 struct netlink_ext_ack *extack) 234 { 235 struct Qdisc *old_child = NULL, *child = NULL; 236 struct red_sched_data *q = qdisc_priv(sch); 237 struct nla_bitfield32 flags_bf; 238 struct tc_red_qopt *ctl; 239 unsigned char userbits; 240 unsigned char flags; 241 int err; 242 u32 max_P; 243 u8 *stab; 244 245 if (tb[TCA_RED_PARMS] == NULL || 246 tb[TCA_RED_STAB] == NULL) 247 return -EINVAL; 248 249 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0; 250 251 ctl = nla_data(tb[TCA_RED_PARMS]); 252 stab = nla_data(tb[TCA_RED_STAB]); 253 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog, 254 ctl->Scell_log, stab)) 255 return -EINVAL; 256 257 err = red_get_flags(ctl->flags, TC_RED_HISTORIC_FLAGS, 258 tb[TCA_RED_FLAGS], TC_RED_SUPPORTED_FLAGS, 259 &flags_bf, &userbits, extack); 260 if (err) 261 return err; 262 263 if (ctl->limit > 0) { 264 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit, 265 extack); 266 if (IS_ERR(child)) 267 return PTR_ERR(child); 268 269 /* child is fifo, no need to check for noop_qdisc */ 270 qdisc_hash_add(child, true); 271 } 272 273 sch_tree_lock(sch); 274 275 flags = (q->flags & ~flags_bf.selector) | flags_bf.value; 276 err = red_validate_flags(flags, extack); 277 if (err) 278 goto unlock_out; 279 280 q->flags = flags; 281 q->userbits = userbits; 282 q->limit = ctl->limit; 283 if (child) { 284 qdisc_tree_flush_backlog(q->qdisc); 285 old_child = q->qdisc; 286 q->qdisc = child; 287 } 288 289 red_set_parms(&q->parms, 290 ctl->qth_min, ctl->qth_max, ctl->Wlog, 291 ctl->Plog, ctl->Scell_log, 292 stab, 293 max_P); 294 red_set_vars(&q->vars); 295 296 del_timer(&q->adapt_timer); 297 if (ctl->flags & TC_RED_ADAPTATIVE) 298 mod_timer(&q->adapt_timer, jiffies + HZ/2); 299 300 if (!q->qdisc->q.qlen) 301 red_start_of_idle_period(&q->vars); 302 303 sch_tree_unlock(sch); 304 305 red_offload(sch, true); 306 307 if (old_child) 308 qdisc_put(old_child); 309 return 0; 310 311 unlock_out: 312 sch_tree_unlock(sch); 313 if (child) 314 qdisc_put(child); 315 return err; 316 } 317 318 static inline void red_adaptative_timer(struct timer_list *t) 319 { 320 struct red_sched_data *q = from_timer(q, t, adapt_timer); 321 struct Qdisc *sch = q->sch; 322 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 323 324 spin_lock(root_lock); 325 red_adaptative_algo(&q->parms, &q->vars); 326 mod_timer(&q->adapt_timer, jiffies + HZ/2); 327 spin_unlock(root_lock); 328 } 329 330 static int red_init(struct Qdisc *sch, struct nlattr *opt, 331 struct netlink_ext_ack *extack) 332 { 333 struct red_sched_data *q = qdisc_priv(sch); 334 struct nlattr *tb[TCA_RED_MAX + 1]; 335 int err; 336 337 q->qdisc = &noop_qdisc; 338 q->sch = sch; 339 timer_setup(&q->adapt_timer, red_adaptative_timer, 0); 340 341 if (!opt) 342 return -EINVAL; 343 344 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy, 345 extack); 346 if (err < 0) 347 return err; 348 349 err = __red_change(sch, tb, extack); 350 if (err) 351 return err; 352 353 err = tcf_qevent_init(&q->qe_early_drop, sch, 354 FLOW_BLOCK_BINDER_TYPE_RED_EARLY_DROP, 355 tb[TCA_RED_EARLY_DROP_BLOCK], extack); 356 if (err) 357 return err; 358 359 return tcf_qevent_init(&q->qe_mark, sch, 360 FLOW_BLOCK_BINDER_TYPE_RED_MARK, 361 tb[TCA_RED_MARK_BLOCK], extack); 362 } 363 364 static int red_change(struct Qdisc *sch, struct nlattr *opt, 365 struct netlink_ext_ack *extack) 366 { 367 struct red_sched_data *q = qdisc_priv(sch); 368 struct nlattr *tb[TCA_RED_MAX + 1]; 369 int err; 370 371 if (!opt) 372 return -EINVAL; 373 374 err = nla_parse_nested_deprecated(tb, TCA_RED_MAX, opt, red_policy, 375 extack); 376 if (err < 0) 377 return err; 378 379 err = tcf_qevent_validate_change(&q->qe_early_drop, 380 tb[TCA_RED_EARLY_DROP_BLOCK], extack); 381 if (err) 382 return err; 383 384 err = tcf_qevent_validate_change(&q->qe_mark, 385 tb[TCA_RED_MARK_BLOCK], extack); 386 if (err) 387 return err; 388 389 return __red_change(sch, tb, extack); 390 } 391 392 static int red_dump_offload_stats(struct Qdisc *sch) 393 { 394 struct tc_red_qopt_offload hw_stats = { 395 .command = TC_RED_STATS, 396 .handle = sch->handle, 397 .parent = sch->parent, 398 { 399 .stats.bstats = &sch->bstats, 400 .stats.qstats = &sch->qstats, 401 }, 402 }; 403 404 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_RED, &hw_stats); 405 } 406 407 static int red_dump(struct Qdisc *sch, struct sk_buff *skb) 408 { 409 struct red_sched_data *q = qdisc_priv(sch); 410 struct nlattr *opts = NULL; 411 struct tc_red_qopt opt = { 412 .limit = q->limit, 413 .flags = (q->flags & TC_RED_HISTORIC_FLAGS) | 414 q->userbits, 415 .qth_min = q->parms.qth_min >> q->parms.Wlog, 416 .qth_max = q->parms.qth_max >> q->parms.Wlog, 417 .Wlog = q->parms.Wlog, 418 .Plog = q->parms.Plog, 419 .Scell_log = q->parms.Scell_log, 420 }; 421 int err; 422 423 err = red_dump_offload_stats(sch); 424 if (err) 425 goto nla_put_failure; 426 427 opts = nla_nest_start_noflag(skb, TCA_OPTIONS); 428 if (opts == NULL) 429 goto nla_put_failure; 430 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || 431 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P) || 432 nla_put_bitfield32(skb, TCA_RED_FLAGS, 433 q->flags, TC_RED_SUPPORTED_FLAGS) || 434 tcf_qevent_dump(skb, TCA_RED_MARK_BLOCK, &q->qe_mark) || 435 tcf_qevent_dump(skb, TCA_RED_EARLY_DROP_BLOCK, &q->qe_early_drop)) 436 goto nla_put_failure; 437 return nla_nest_end(skb, opts); 438 439 nla_put_failure: 440 nla_nest_cancel(skb, opts); 441 return -EMSGSIZE; 442 } 443 444 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 445 { 446 struct red_sched_data *q = qdisc_priv(sch); 447 struct net_device *dev = qdisc_dev(sch); 448 struct tc_red_xstats st = {0}; 449 450 if (sch->flags & TCQ_F_OFFLOADED) { 451 struct tc_red_qopt_offload hw_stats_request = { 452 .command = TC_RED_XSTATS, 453 .handle = sch->handle, 454 .parent = sch->parent, 455 { 456 .xstats = &q->stats, 457 }, 458 }; 459 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_RED, 460 &hw_stats_request); 461 } 462 st.early = q->stats.prob_drop + q->stats.forced_drop; 463 st.pdrop = q->stats.pdrop; 464 st.other = q->stats.other; 465 st.marked = q->stats.prob_mark + q->stats.forced_mark; 466 467 return gnet_stats_copy_app(d, &st, sizeof(st)); 468 } 469 470 static int red_dump_class(struct Qdisc *sch, unsigned long cl, 471 struct sk_buff *skb, struct tcmsg *tcm) 472 { 473 struct red_sched_data *q = qdisc_priv(sch); 474 475 tcm->tcm_handle |= TC_H_MIN(1); 476 tcm->tcm_info = q->qdisc->handle; 477 return 0; 478 } 479 480 static void red_graft_offload(struct Qdisc *sch, 481 struct Qdisc *new, struct Qdisc *old, 482 struct netlink_ext_ack *extack) 483 { 484 struct tc_red_qopt_offload graft_offload = { 485 .handle = sch->handle, 486 .parent = sch->parent, 487 .child_handle = new->handle, 488 .command = TC_RED_GRAFT, 489 }; 490 491 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, old, 492 TC_SETUP_QDISC_RED, &graft_offload, extack); 493 } 494 495 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 496 struct Qdisc **old, struct netlink_ext_ack *extack) 497 { 498 struct red_sched_data *q = qdisc_priv(sch); 499 500 if (new == NULL) 501 new = &noop_qdisc; 502 503 *old = qdisc_replace(sch, new, &q->qdisc); 504 505 red_graft_offload(sch, new, *old, extack); 506 return 0; 507 } 508 509 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) 510 { 511 struct red_sched_data *q = qdisc_priv(sch); 512 return q->qdisc; 513 } 514 515 static unsigned long red_find(struct Qdisc *sch, u32 classid) 516 { 517 return 1; 518 } 519 520 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) 521 { 522 if (!walker->stop) { 523 if (walker->count >= walker->skip) 524 if (walker->fn(sch, 1, walker) < 0) { 525 walker->stop = 1; 526 return; 527 } 528 walker->count++; 529 } 530 } 531 532 static const struct Qdisc_class_ops red_class_ops = { 533 .graft = red_graft, 534 .leaf = red_leaf, 535 .find = red_find, 536 .walk = red_walk, 537 .dump = red_dump_class, 538 }; 539 540 static struct Qdisc_ops red_qdisc_ops __read_mostly = { 541 .id = "red", 542 .priv_size = sizeof(struct red_sched_data), 543 .cl_ops = &red_class_ops, 544 .enqueue = red_enqueue, 545 .dequeue = red_dequeue, 546 .peek = red_peek, 547 .init = red_init, 548 .reset = red_reset, 549 .destroy = red_destroy, 550 .change = red_change, 551 .dump = red_dump, 552 .dump_stats = red_dump_stats, 553 .owner = THIS_MODULE, 554 }; 555 556 static int __init red_module_init(void) 557 { 558 return register_qdisc(&red_qdisc_ops); 559 } 560 561 static void __exit red_module_exit(void) 562 { 563 unregister_qdisc(&red_qdisc_ops); 564 } 565 566 module_init(red_module_init) 567 module_exit(red_module_exit) 568 569 MODULE_LICENSE("GPL"); 570