1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net/sched/sch_tbf.c Token Bucket Filter queue. 4 * 5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 6 * Dmitry Torokhov <dtor@mail.ru> - allow attaching inner qdiscs - 7 * original idea by Martin Devera 8 */ 9 10 #include <linux/module.h> 11 #include <linux/types.h> 12 #include <linux/kernel.h> 13 #include <linux/string.h> 14 #include <linux/errno.h> 15 #include <linux/skbuff.h> 16 #include <net/netlink.h> 17 #include <net/sch_generic.h> 18 #include <net/pkt_sched.h> 19 20 21 /* Simple Token Bucket Filter. 22 ======================================= 23 24 SOURCE. 25 ------- 26 27 None. 28 29 Description. 30 ------------ 31 32 A data flow obeys TBF with rate R and depth B, if for any 33 time interval t_i...t_f the number of transmitted bits 34 does not exceed B + R*(t_f-t_i). 35 36 Packetized version of this definition: 37 The sequence of packets of sizes s_i served at moments t_i 38 obeys TBF, if for any i<=k: 39 40 s_i+....+s_k <= B + R*(t_k - t_i) 41 42 Algorithm. 43 ---------- 44 45 Let N(t_i) be B/R initially and N(t) grow continuously with time as: 46 47 N(t+delta) = min{B/R, N(t) + delta} 48 49 If the first packet in queue has length S, it may be 50 transmitted only at the time t_* when S/R <= N(t_*), 51 and in this case N(t) jumps: 52 53 N(t_* + 0) = N(t_* - 0) - S/R. 54 55 56 57 Actually, QoS requires two TBF to be applied to a data stream. 58 One of them controls steady state burst size, another 59 one with rate P (peak rate) and depth M (equal to link MTU) 60 limits bursts at a smaller time scale. 61 62 It is easy to see that P>R, and B>M. If P is infinity, this double 63 TBF is equivalent to a single one. 64 65 When TBF works in reshaping mode, latency is estimated as: 66 67 lat = max ((L-B)/R, (L-M)/P) 68 69 70 NOTES. 71 ------ 72 73 If TBF throttles, it starts a watchdog timer, which will wake it up 74 when it is ready to transmit. 75 Note that the minimal timer resolution is 1/HZ. 76 If no new packets arrive during this period, 77 or if the device is not awaken by EOI for some previous packet, 78 TBF can stop its activity for 1/HZ. 79 80 81 This means, that with depth B, the maximal rate is 82 83 R_crit = B*HZ 84 85 F.e. for 10Mbit ethernet and HZ=100 the minimal allowed B is ~10Kbytes. 86 87 Note that the peak rate TBF is much more tough: with MTU 1500 88 P_crit = 150Kbytes/sec. So, if you need greater peak 89 rates, use alpha with HZ=1000 :-) 90 91 With classful TBF, limit is just kept for backwards compatibility. 92 It is passed to the default bfifo qdisc - if the inner qdisc is 93 changed the limit is not effective anymore. 94 */ 95 96 struct tbf_sched_data { 97 /* Parameters */ 98 u32 limit; /* Maximal length of backlog: bytes */ 99 u32 max_size; 100 s64 buffer; /* Token bucket depth/rate: MUST BE >= MTU/B */ 101 s64 mtu; 102 struct psched_ratecfg rate; 103 struct psched_ratecfg peak; 104 105 /* Variables */ 106 s64 tokens; /* Current number of B tokens */ 107 s64 ptokens; /* Current number of P tokens */ 108 s64 t_c; /* Time check-point */ 109 struct Qdisc *qdisc; /* Inner qdisc, default - bfifo queue */ 110 struct qdisc_watchdog watchdog; /* Watchdog timer */ 111 }; 112 113 114 /* Time to Length, convert time in ns to length in bytes 115 * to determinate how many bytes can be sent in given time. 116 */ 117 static u64 psched_ns_t2l(const struct psched_ratecfg *r, 118 u64 time_in_ns) 119 { 120 /* The formula is : 121 * len = (time_in_ns * r->rate_bytes_ps) / NSEC_PER_SEC 122 */ 123 u64 len = time_in_ns * r->rate_bytes_ps; 124 125 do_div(len, NSEC_PER_SEC); 126 127 if (unlikely(r->linklayer == TC_LINKLAYER_ATM)) { 128 do_div(len, 53); 129 len = len * 48; 130 } 131 132 if (len > r->overhead) 133 len -= r->overhead; 134 else 135 len = 0; 136 137 return len; 138 } 139 140 /* GSO packet is too big, segment it so that tbf can transmit 141 * each segment in time 142 */ 143 static int tbf_segment(struct sk_buff *skb, struct Qdisc *sch, 144 struct sk_buff **to_free) 145 { 146 struct tbf_sched_data *q = qdisc_priv(sch); 147 struct sk_buff *segs, *nskb; 148 netdev_features_t features = netif_skb_features(skb); 149 unsigned int len = 0, prev_len = qdisc_pkt_len(skb); 150 int ret, nb; 151 152 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); 153 154 if (IS_ERR_OR_NULL(segs)) 155 return qdisc_drop(skb, sch, to_free); 156 157 nb = 0; 158 while (segs) { 159 nskb = segs->next; 160 skb_mark_not_on_list(segs); 161 qdisc_skb_cb(segs)->pkt_len = segs->len; 162 len += segs->len; 163 ret = qdisc_enqueue(segs, q->qdisc, to_free); 164 if (ret != NET_XMIT_SUCCESS) { 165 if (net_xmit_drop_count(ret)) 166 qdisc_qstats_drop(sch); 167 } else { 168 nb++; 169 } 170 segs = nskb; 171 } 172 sch->q.qlen += nb; 173 if (nb > 1) 174 qdisc_tree_reduce_backlog(sch, 1 - nb, prev_len - len); 175 consume_skb(skb); 176 return nb > 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP; 177 } 178 179 static int tbf_enqueue(struct sk_buff *skb, struct Qdisc *sch, 180 struct sk_buff **to_free) 181 { 182 struct tbf_sched_data *q = qdisc_priv(sch); 183 unsigned int len = qdisc_pkt_len(skb); 184 int ret; 185 186 if (qdisc_pkt_len(skb) > q->max_size) { 187 if (skb_is_gso(skb) && 188 skb_gso_validate_mac_len(skb, q->max_size)) 189 return tbf_segment(skb, sch, to_free); 190 return qdisc_drop(skb, sch, to_free); 191 } 192 ret = qdisc_enqueue(skb, q->qdisc, to_free); 193 if (ret != NET_XMIT_SUCCESS) { 194 if (net_xmit_drop_count(ret)) 195 qdisc_qstats_drop(sch); 196 return ret; 197 } 198 199 sch->qstats.backlog += len; 200 sch->q.qlen++; 201 return NET_XMIT_SUCCESS; 202 } 203 204 static bool tbf_peak_present(const struct tbf_sched_data *q) 205 { 206 return q->peak.rate_bytes_ps; 207 } 208 209 static struct sk_buff *tbf_dequeue(struct Qdisc *sch) 210 { 211 struct tbf_sched_data *q = qdisc_priv(sch); 212 struct sk_buff *skb; 213 214 skb = q->qdisc->ops->peek(q->qdisc); 215 216 if (skb) { 217 s64 now; 218 s64 toks; 219 s64 ptoks = 0; 220 unsigned int len = qdisc_pkt_len(skb); 221 222 now = ktime_get_ns(); 223 toks = min_t(s64, now - q->t_c, q->buffer); 224 225 if (tbf_peak_present(q)) { 226 ptoks = toks + q->ptokens; 227 if (ptoks > q->mtu) 228 ptoks = q->mtu; 229 ptoks -= (s64) psched_l2t_ns(&q->peak, len); 230 } 231 toks += q->tokens; 232 if (toks > q->buffer) 233 toks = q->buffer; 234 toks -= (s64) psched_l2t_ns(&q->rate, len); 235 236 if ((toks|ptoks) >= 0) { 237 skb = qdisc_dequeue_peeked(q->qdisc); 238 if (unlikely(!skb)) 239 return NULL; 240 241 q->t_c = now; 242 q->tokens = toks; 243 q->ptokens = ptoks; 244 qdisc_qstats_backlog_dec(sch, skb); 245 sch->q.qlen--; 246 qdisc_bstats_update(sch, skb); 247 return skb; 248 } 249 250 qdisc_watchdog_schedule_ns(&q->watchdog, 251 now + max_t(long, -toks, -ptoks)); 252 253 /* Maybe we have a shorter packet in the queue, 254 which can be sent now. It sounds cool, 255 but, however, this is wrong in principle. 256 We MUST NOT reorder packets under these circumstances. 257 258 Really, if we split the flow into independent 259 subflows, it would be a very good solution. 260 This is the main idea of all FQ algorithms 261 (cf. CSZ, HPFQ, HFSC) 262 */ 263 264 qdisc_qstats_overlimit(sch); 265 } 266 return NULL; 267 } 268 269 static void tbf_reset(struct Qdisc *sch) 270 { 271 struct tbf_sched_data *q = qdisc_priv(sch); 272 273 qdisc_reset(q->qdisc); 274 sch->qstats.backlog = 0; 275 sch->q.qlen = 0; 276 q->t_c = ktime_get_ns(); 277 q->tokens = q->buffer; 278 q->ptokens = q->mtu; 279 qdisc_watchdog_cancel(&q->watchdog); 280 } 281 282 static const struct nla_policy tbf_policy[TCA_TBF_MAX + 1] = { 283 [TCA_TBF_PARMS] = { .len = sizeof(struct tc_tbf_qopt) }, 284 [TCA_TBF_RTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, 285 [TCA_TBF_PTAB] = { .type = NLA_BINARY, .len = TC_RTAB_SIZE }, 286 [TCA_TBF_RATE64] = { .type = NLA_U64 }, 287 [TCA_TBF_PRATE64] = { .type = NLA_U64 }, 288 [TCA_TBF_BURST] = { .type = NLA_U32 }, 289 [TCA_TBF_PBURST] = { .type = NLA_U32 }, 290 }; 291 292 static int tbf_change(struct Qdisc *sch, struct nlattr *opt, 293 struct netlink_ext_ack *extack) 294 { 295 int err; 296 struct tbf_sched_data *q = qdisc_priv(sch); 297 struct nlattr *tb[TCA_TBF_MAX + 1]; 298 struct tc_tbf_qopt *qopt; 299 struct Qdisc *child = NULL; 300 struct psched_ratecfg rate; 301 struct psched_ratecfg peak; 302 u64 max_size; 303 s64 buffer, mtu; 304 u64 rate64 = 0, prate64 = 0; 305 306 err = nla_parse_nested_deprecated(tb, TCA_TBF_MAX, opt, tbf_policy, 307 NULL); 308 if (err < 0) 309 return err; 310 311 err = -EINVAL; 312 if (tb[TCA_TBF_PARMS] == NULL) 313 goto done; 314 315 qopt = nla_data(tb[TCA_TBF_PARMS]); 316 if (qopt->rate.linklayer == TC_LINKLAYER_UNAWARE) 317 qdisc_put_rtab(qdisc_get_rtab(&qopt->rate, 318 tb[TCA_TBF_RTAB], 319 NULL)); 320 321 if (qopt->peakrate.linklayer == TC_LINKLAYER_UNAWARE) 322 qdisc_put_rtab(qdisc_get_rtab(&qopt->peakrate, 323 tb[TCA_TBF_PTAB], 324 NULL)); 325 326 buffer = min_t(u64, PSCHED_TICKS2NS(qopt->buffer), ~0U); 327 mtu = min_t(u64, PSCHED_TICKS2NS(qopt->mtu), ~0U); 328 329 if (tb[TCA_TBF_RATE64]) 330 rate64 = nla_get_u64(tb[TCA_TBF_RATE64]); 331 psched_ratecfg_precompute(&rate, &qopt->rate, rate64); 332 333 if (tb[TCA_TBF_BURST]) { 334 max_size = nla_get_u32(tb[TCA_TBF_BURST]); 335 buffer = psched_l2t_ns(&rate, max_size); 336 } else { 337 max_size = min_t(u64, psched_ns_t2l(&rate, buffer), ~0U); 338 } 339 340 if (qopt->peakrate.rate) { 341 if (tb[TCA_TBF_PRATE64]) 342 prate64 = nla_get_u64(tb[TCA_TBF_PRATE64]); 343 psched_ratecfg_precompute(&peak, &qopt->peakrate, prate64); 344 if (peak.rate_bytes_ps <= rate.rate_bytes_ps) { 345 pr_warn_ratelimited("sch_tbf: peakrate %llu is lower than or equals to rate %llu !\n", 346 peak.rate_bytes_ps, rate.rate_bytes_ps); 347 err = -EINVAL; 348 goto done; 349 } 350 351 if (tb[TCA_TBF_PBURST]) { 352 u32 pburst = nla_get_u32(tb[TCA_TBF_PBURST]); 353 max_size = min_t(u32, max_size, pburst); 354 mtu = psched_l2t_ns(&peak, pburst); 355 } else { 356 max_size = min_t(u64, max_size, psched_ns_t2l(&peak, mtu)); 357 } 358 } else { 359 memset(&peak, 0, sizeof(peak)); 360 } 361 362 if (max_size < psched_mtu(qdisc_dev(sch))) 363 pr_warn_ratelimited("sch_tbf: burst %llu is lower than device %s mtu (%u) !\n", 364 max_size, qdisc_dev(sch)->name, 365 psched_mtu(qdisc_dev(sch))); 366 367 if (!max_size) { 368 err = -EINVAL; 369 goto done; 370 } 371 372 if (q->qdisc != &noop_qdisc) { 373 err = fifo_set_limit(q->qdisc, qopt->limit); 374 if (err) 375 goto done; 376 } else if (qopt->limit > 0) { 377 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, qopt->limit, 378 extack); 379 if (IS_ERR(child)) { 380 err = PTR_ERR(child); 381 goto done; 382 } 383 384 /* child is fifo, no need to check for noop_qdisc */ 385 qdisc_hash_add(child, true); 386 } 387 388 sch_tree_lock(sch); 389 if (child) { 390 qdisc_tree_flush_backlog(q->qdisc); 391 qdisc_put(q->qdisc); 392 q->qdisc = child; 393 } 394 q->limit = qopt->limit; 395 if (tb[TCA_TBF_PBURST]) 396 q->mtu = mtu; 397 else 398 q->mtu = PSCHED_TICKS2NS(qopt->mtu); 399 q->max_size = max_size; 400 if (tb[TCA_TBF_BURST]) 401 q->buffer = buffer; 402 else 403 q->buffer = PSCHED_TICKS2NS(qopt->buffer); 404 q->tokens = q->buffer; 405 q->ptokens = q->mtu; 406 407 memcpy(&q->rate, &rate, sizeof(struct psched_ratecfg)); 408 memcpy(&q->peak, &peak, sizeof(struct psched_ratecfg)); 409 410 sch_tree_unlock(sch); 411 err = 0; 412 done: 413 return err; 414 } 415 416 static int tbf_init(struct Qdisc *sch, struct nlattr *opt, 417 struct netlink_ext_ack *extack) 418 { 419 struct tbf_sched_data *q = qdisc_priv(sch); 420 421 qdisc_watchdog_init(&q->watchdog, sch); 422 q->qdisc = &noop_qdisc; 423 424 if (!opt) 425 return -EINVAL; 426 427 q->t_c = ktime_get_ns(); 428 429 return tbf_change(sch, opt, extack); 430 } 431 432 static void tbf_destroy(struct Qdisc *sch) 433 { 434 struct tbf_sched_data *q = qdisc_priv(sch); 435 436 qdisc_watchdog_cancel(&q->watchdog); 437 qdisc_put(q->qdisc); 438 } 439 440 static int tbf_dump(struct Qdisc *sch, struct sk_buff *skb) 441 { 442 struct tbf_sched_data *q = qdisc_priv(sch); 443 struct nlattr *nest; 444 struct tc_tbf_qopt opt; 445 446 sch->qstats.backlog = q->qdisc->qstats.backlog; 447 nest = nla_nest_start_noflag(skb, TCA_OPTIONS); 448 if (nest == NULL) 449 goto nla_put_failure; 450 451 opt.limit = q->limit; 452 psched_ratecfg_getrate(&opt.rate, &q->rate); 453 if (tbf_peak_present(q)) 454 psched_ratecfg_getrate(&opt.peakrate, &q->peak); 455 else 456 memset(&opt.peakrate, 0, sizeof(opt.peakrate)); 457 opt.mtu = PSCHED_NS2TICKS(q->mtu); 458 opt.buffer = PSCHED_NS2TICKS(q->buffer); 459 if (nla_put(skb, TCA_TBF_PARMS, sizeof(opt), &opt)) 460 goto nla_put_failure; 461 if (q->rate.rate_bytes_ps >= (1ULL << 32) && 462 nla_put_u64_64bit(skb, TCA_TBF_RATE64, q->rate.rate_bytes_ps, 463 TCA_TBF_PAD)) 464 goto nla_put_failure; 465 if (tbf_peak_present(q) && 466 q->peak.rate_bytes_ps >= (1ULL << 32) && 467 nla_put_u64_64bit(skb, TCA_TBF_PRATE64, q->peak.rate_bytes_ps, 468 TCA_TBF_PAD)) 469 goto nla_put_failure; 470 471 return nla_nest_end(skb, nest); 472 473 nla_put_failure: 474 nla_nest_cancel(skb, nest); 475 return -1; 476 } 477 478 static int tbf_dump_class(struct Qdisc *sch, unsigned long cl, 479 struct sk_buff *skb, struct tcmsg *tcm) 480 { 481 struct tbf_sched_data *q = qdisc_priv(sch); 482 483 tcm->tcm_handle |= TC_H_MIN(1); 484 tcm->tcm_info = q->qdisc->handle; 485 486 return 0; 487 } 488 489 static int tbf_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, 490 struct Qdisc **old, struct netlink_ext_ack *extack) 491 { 492 struct tbf_sched_data *q = qdisc_priv(sch); 493 494 if (new == NULL) 495 new = &noop_qdisc; 496 497 *old = qdisc_replace(sch, new, &q->qdisc); 498 return 0; 499 } 500 501 static struct Qdisc *tbf_leaf(struct Qdisc *sch, unsigned long arg) 502 { 503 struct tbf_sched_data *q = qdisc_priv(sch); 504 return q->qdisc; 505 } 506 507 static unsigned long tbf_find(struct Qdisc *sch, u32 classid) 508 { 509 return 1; 510 } 511 512 static void tbf_walk(struct Qdisc *sch, struct qdisc_walker *walker) 513 { 514 if (!walker->stop) { 515 if (walker->count >= walker->skip) 516 if (walker->fn(sch, 1, walker) < 0) { 517 walker->stop = 1; 518 return; 519 } 520 walker->count++; 521 } 522 } 523 524 static const struct Qdisc_class_ops tbf_class_ops = { 525 .graft = tbf_graft, 526 .leaf = tbf_leaf, 527 .find = tbf_find, 528 .walk = tbf_walk, 529 .dump = tbf_dump_class, 530 }; 531 532 static struct Qdisc_ops tbf_qdisc_ops __read_mostly = { 533 .next = NULL, 534 .cl_ops = &tbf_class_ops, 535 .id = "tbf", 536 .priv_size = sizeof(struct tbf_sched_data), 537 .enqueue = tbf_enqueue, 538 .dequeue = tbf_dequeue, 539 .peek = qdisc_peek_dequeued, 540 .init = tbf_init, 541 .reset = tbf_reset, 542 .destroy = tbf_destroy, 543 .change = tbf_change, 544 .dump = tbf_dump, 545 .owner = THIS_MODULE, 546 }; 547 548 static int __init tbf_module_init(void) 549 { 550 return register_qdisc(&tbf_qdisc_ops); 551 } 552 553 static void __exit tbf_module_exit(void) 554 { 555 unregister_qdisc(&tbf_qdisc_ops); 556 } 557 module_init(tbf_module_init) 558 module_exit(tbf_module_exit) 559 MODULE_LICENSE("GPL"); 560