1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Fair Queue CoDel discipline 4 * 5 * Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com> 6 */ 7 8 #include <linux/module.h> 9 #include <linux/types.h> 10 #include <linux/kernel.h> 11 #include <linux/jiffies.h> 12 #include <linux/string.h> 13 #include <linux/in.h> 14 #include <linux/errno.h> 15 #include <linux/init.h> 16 #include <linux/skbuff.h> 17 #include <linux/slab.h> 18 #include <linux/vmalloc.h> 19 #include <net/netlink.h> 20 #include <net/pkt_sched.h> 21 #include <net/pkt_cls.h> 22 #include <net/codel.h> 23 #include <net/codel_impl.h> 24 #include <net/codel_qdisc.h> 25 26 /* Fair Queue CoDel. 27 * 28 * Principles : 29 * Packets are classified (internal classifier or external) on flows. 30 * This is a Stochastic model (as we use a hash, several flows 31 * might be hashed on same slot) 32 * Each flow has a CoDel managed queue. 33 * Flows are linked onto two (Round Robin) lists, 34 * so that new flows have priority on old ones. 35 * 36 * For a given flow, packets are not reordered (CoDel uses a FIFO) 37 * head drops only. 38 * ECN capability is on by default. 39 * Low memory footprint (64 bytes per flow) 40 */ 41 42 struct fq_codel_flow { 43 struct sk_buff *head; 44 struct sk_buff *tail; 45 struct list_head flowchain; 46 int deficit; 47 struct codel_vars cvars; 48 }; /* please try to keep this structure <= 64 bytes */ 49 50 struct fq_codel_sched_data { 51 struct tcf_proto __rcu *filter_list; /* optional external classifier */ 52 struct tcf_block *block; 53 struct fq_codel_flow *flows; /* Flows table [flows_cnt] */ 54 u32 *backlogs; /* backlog table [flows_cnt] */ 55 u32 flows_cnt; /* number of flows */ 56 u32 quantum; /* psched_mtu(qdisc_dev(sch)); */ 57 u32 drop_batch_size; 58 u32 memory_limit; 59 struct codel_params cparams; 60 struct codel_stats cstats; 61 u32 memory_usage; 62 u32 drop_overmemory; 63 u32 drop_overlimit; 64 u32 new_flow_count; 65 66 struct list_head new_flows; /* list of new flows */ 67 struct list_head old_flows; /* list of old flows */ 68 }; 69 70 static unsigned int fq_codel_hash(const struct fq_codel_sched_data *q, 71 struct sk_buff *skb) 72 { 73 return reciprocal_scale(skb_get_hash(skb), q->flows_cnt); 74 } 75 76 static unsigned int fq_codel_classify(struct sk_buff *skb, struct Qdisc *sch, 77 int *qerr) 78 { 79 struct fq_codel_sched_data *q = qdisc_priv(sch); 80 struct tcf_proto *filter; 81 struct tcf_result res; 82 int result; 83 84 if (TC_H_MAJ(skb->priority) == sch->handle && 85 TC_H_MIN(skb->priority) > 0 && 86 TC_H_MIN(skb->priority) <= q->flows_cnt) 87 return TC_H_MIN(skb->priority); 88 89 filter = rcu_dereference_bh(q->filter_list); 90 if (!filter) 91 return fq_codel_hash(q, skb) + 1; 92 93 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 94 result = tcf_classify(skb, NULL, filter, &res, false); 95 if (result >= 0) { 96 #ifdef CONFIG_NET_CLS_ACT 97 switch (result) { 98 case TC_ACT_STOLEN: 99 case TC_ACT_QUEUED: 100 case TC_ACT_TRAP: 101 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 102 fallthrough; 103 case TC_ACT_SHOT: 104 return 0; 105 } 106 #endif 107 if (TC_H_MIN(res.classid) <= q->flows_cnt) 108 return TC_H_MIN(res.classid); 109 } 110 return 0; 111 } 112 113 /* helper functions : might be changed when/if skb use a standard list_head */ 114 115 /* remove one skb from head of slot queue */ 116 static inline struct sk_buff *dequeue_head(struct fq_codel_flow *flow) 117 { 118 struct sk_buff *skb = flow->head; 119 120 WRITE_ONCE(flow->head, skb->next); 121 skb_mark_not_on_list(skb); 122 return skb; 123 } 124 125 /* add skb to flow queue (tail add) */ 126 static inline void flow_queue_add(struct fq_codel_flow *flow, 127 struct sk_buff *skb) 128 { 129 if (flow->head == NULL) 130 WRITE_ONCE(flow->head, skb); 131 else 132 flow->tail->next = skb; 133 flow->tail = skb; 134 skb->next = NULL; 135 } 136 137 static unsigned int fq_codel_drop(struct Qdisc *sch, unsigned int max_packets, 138 struct sk_buff **to_free) 139 { 140 struct fq_codel_sched_data *q = qdisc_priv(sch); 141 struct sk_buff *skb; 142 unsigned int maxbacklog = 0, idx = 0, i, len; 143 struct fq_codel_flow *flow; 144 unsigned int threshold; 145 unsigned int mem = 0; 146 147 /* Queue is full! Find the fat flow and drop packet(s) from it. 148 * This might sound expensive, but with 1024 flows, we scan 149 * 4KB of memory, and we dont need to handle a complex tree 150 * in fast path (packet queue/enqueue) with many cache misses. 151 * In stress mode, we'll try to drop 64 packets from the flow, 152 * amortizing this linear lookup to one cache line per drop. 153 */ 154 for (i = 0; i < q->flows_cnt; i++) { 155 if (q->backlogs[i] > maxbacklog) { 156 maxbacklog = q->backlogs[i]; 157 idx = i; 158 } 159 } 160 161 /* Our goal is to drop half of this fat flow backlog */ 162 threshold = maxbacklog >> 1; 163 164 flow = &q->flows[idx]; 165 len = 0; 166 i = 0; 167 do { 168 skb = dequeue_head(flow); 169 len += qdisc_pkt_len(skb); 170 mem += get_codel_cb(skb)->mem_usage; 171 tcf_set_qdisc_drop_reason(skb, QDISC_DROP_OVERLIMIT); 172 __qdisc_drop(skb, to_free); 173 } while (++i < max_packets && len < threshold); 174 175 /* Tell codel to increase its signal strength also */ 176 WRITE_ONCE(flow->cvars.count, flow->cvars.count + i); 177 WRITE_ONCE(q->backlogs[idx], q->backlogs[idx] - len); 178 q->memory_usage -= mem; 179 __qdisc_qstats_drop(sch, i); 180 qstats_backlog_sub(sch, len); 181 WRITE_ONCE(sch->q.qlen, sch->q.qlen - i); 182 return idx; 183 } 184 185 static int fq_codel_enqueue(struct sk_buff *skb, struct Qdisc *sch, 186 struct sk_buff **to_free) 187 { 188 struct fq_codel_sched_data *q = qdisc_priv(sch); 189 unsigned int idx, prev_backlog, prev_qlen; 190 struct fq_codel_flow *flow; 191 int ret; 192 unsigned int pkt_len; 193 bool memory_limited; 194 195 idx = fq_codel_classify(skb, sch, &ret); 196 if (idx == 0) { 197 if (ret & __NET_XMIT_BYPASS) 198 qdisc_qstats_drop(sch); 199 __qdisc_drop(skb, to_free); 200 return ret; 201 } 202 idx--; 203 204 codel_set_enqueue_time(skb); 205 flow = &q->flows[idx]; 206 flow_queue_add(flow, skb); 207 WRITE_ONCE(q->backlogs[idx], q->backlogs[idx] + qdisc_pkt_len(skb)); 208 qdisc_qstats_backlog_inc(sch, skb); 209 210 if (list_empty(&flow->flowchain)) { 211 list_add_tail(&flow->flowchain, &q->new_flows); 212 q->new_flow_count++; 213 WRITE_ONCE(flow->deficit, q->quantum); 214 } 215 get_codel_cb(skb)->mem_usage = is_skb_wmem(skb) ? 0 : skb->truesize; 216 q->memory_usage += get_codel_cb(skb)->mem_usage; 217 memory_limited = q->memory_usage > q->memory_limit; 218 qdisc_qlen_inc(sch); 219 if (sch->q.qlen <= sch->limit && !memory_limited) 220 return NET_XMIT_SUCCESS; 221 222 prev_backlog = sch->qstats.backlog; 223 prev_qlen = sch->q.qlen; 224 225 /* save this packet length as it might be dropped by fq_codel_drop() */ 226 pkt_len = qdisc_pkt_len(skb); 227 /* fq_codel_drop() is quite expensive, as it performs a linear search 228 * in q->backlogs[] to find a fat flow. 229 * So instead of dropping a single packet, drop half of its backlog 230 * with a 64 packets limit to not add a too big cpu spike here. 231 */ 232 ret = fq_codel_drop(sch, q->drop_batch_size, to_free); 233 234 prev_qlen -= sch->q.qlen; 235 prev_backlog -= sch->qstats.backlog; 236 q->drop_overlimit += prev_qlen; 237 if (memory_limited) 238 q->drop_overmemory += prev_qlen; 239 240 /* As we dropped packet(s), better let upper stack know this. 241 * If we dropped a packet for this flow, return NET_XMIT_CN, 242 * but in this case, our parents wont increase their backlogs. 243 */ 244 if (ret == idx) { 245 qdisc_tree_reduce_backlog(sch, prev_qlen - 1, 246 prev_backlog - pkt_len); 247 return NET_XMIT_CN; 248 } 249 qdisc_tree_reduce_backlog(sch, prev_qlen, prev_backlog); 250 return NET_XMIT_SUCCESS; 251 } 252 253 /* This is the specific function called from codel_dequeue() 254 * to dequeue a packet from queue. Note: backlog is handled in 255 * codel, we dont need to reduce it here. 256 */ 257 static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx) 258 { 259 struct Qdisc *sch = ctx; 260 struct fq_codel_sched_data *q = qdisc_priv(sch); 261 struct fq_codel_flow *flow; 262 struct sk_buff *skb = NULL; 263 264 flow = container_of(vars, struct fq_codel_flow, cvars); 265 if (flow->head) { 266 skb = dequeue_head(flow); 267 WRITE_ONCE(q->backlogs[flow - q->flows], 268 q->backlogs[flow - q->flows] - qdisc_pkt_len(skb)); 269 q->memory_usage -= get_codel_cb(skb)->mem_usage; 270 qdisc_qlen_dec(sch); 271 qdisc_qstats_backlog_dec(sch, skb); 272 } 273 return skb; 274 } 275 276 static void drop_func(struct sk_buff *skb, void *ctx) 277 { 278 struct Qdisc *sch = ctx; 279 280 qdisc_dequeue_drop(sch, skb, QDISC_DROP_CONGESTED); 281 qdisc_qstats_drop(sch); 282 } 283 284 static struct sk_buff *fq_codel_dequeue(struct Qdisc *sch) 285 { 286 struct fq_codel_sched_data *q = qdisc_priv(sch); 287 struct sk_buff *skb; 288 struct fq_codel_flow *flow; 289 struct list_head *head; 290 291 begin: 292 head = &q->new_flows; 293 if (list_empty(head)) { 294 head = &q->old_flows; 295 if (list_empty(head)) 296 return NULL; 297 } 298 flow = list_first_entry(head, struct fq_codel_flow, flowchain); 299 300 if (flow->deficit <= 0) { 301 WRITE_ONCE(flow->deficit, flow->deficit + q->quantum); 302 list_move_tail(&flow->flowchain, &q->old_flows); 303 goto begin; 304 } 305 306 skb = codel_dequeue(sch, &sch->qstats.backlog, &q->cparams, 307 &flow->cvars, &q->cstats, qdisc_pkt_len, 308 codel_get_enqueue_time, drop_func, dequeue_func); 309 310 if (!skb) { 311 /* force a pass through old_flows to prevent starvation */ 312 if ((head == &q->new_flows) && !list_empty(&q->old_flows)) 313 list_move_tail(&flow->flowchain, &q->old_flows); 314 else 315 list_del_init(&flow->flowchain); 316 goto begin; 317 } 318 qdisc_bstats_update(sch, skb); 319 WRITE_ONCE(flow->deficit, flow->deficit - qdisc_pkt_len(skb)); 320 321 if (q->cstats.drop_count) { 322 qdisc_tree_reduce_backlog(sch, q->cstats.drop_count, 323 q->cstats.drop_len); 324 q->cstats.drop_count = 0; 325 q->cstats.drop_len = 0; 326 } 327 return skb; 328 } 329 330 static void fq_codel_flow_purge(struct fq_codel_flow *flow) 331 { 332 rtnl_kfree_skbs(flow->head, flow->tail); 333 WRITE_ONCE(flow->head, NULL); 334 } 335 336 static void fq_codel_reset(struct Qdisc *sch) 337 { 338 struct fq_codel_sched_data *q = qdisc_priv(sch); 339 int i; 340 341 INIT_LIST_HEAD(&q->new_flows); 342 INIT_LIST_HEAD(&q->old_flows); 343 for (i = 0; i < q->flows_cnt; i++) { 344 struct fq_codel_flow *flow = q->flows + i; 345 346 fq_codel_flow_purge(flow); 347 INIT_LIST_HEAD(&flow->flowchain); 348 codel_vars_init(&flow->cvars); 349 } 350 memset(q->backlogs, 0, q->flows_cnt * sizeof(u32)); 351 q->memory_usage = 0; 352 } 353 354 static const struct nla_policy fq_codel_policy[TCA_FQ_CODEL_MAX + 1] = { 355 [TCA_FQ_CODEL_TARGET] = { .type = NLA_U32 }, 356 [TCA_FQ_CODEL_LIMIT] = { .type = NLA_U32 }, 357 [TCA_FQ_CODEL_INTERVAL] = { .type = NLA_U32 }, 358 [TCA_FQ_CODEL_ECN] = { .type = NLA_U32 }, 359 [TCA_FQ_CODEL_FLOWS] = { .type = NLA_U32 }, 360 [TCA_FQ_CODEL_QUANTUM] = { .type = NLA_U32 }, 361 [TCA_FQ_CODEL_CE_THRESHOLD] = { .type = NLA_U32 }, 362 [TCA_FQ_CODEL_DROP_BATCH_SIZE] = { .type = NLA_U32 }, 363 [TCA_FQ_CODEL_MEMORY_LIMIT] = { .type = NLA_U32 }, 364 [TCA_FQ_CODEL_CE_THRESHOLD_SELECTOR] = { .type = NLA_U8 }, 365 [TCA_FQ_CODEL_CE_THRESHOLD_MASK] = { .type = NLA_U8 }, 366 }; 367 368 static int fq_codel_change(struct Qdisc *sch, struct nlattr *opt, 369 struct netlink_ext_ack *extack) 370 { 371 unsigned int dropped_pkts = 0, dropped_bytes = 0; 372 struct fq_codel_sched_data *q = qdisc_priv(sch); 373 struct nlattr *tb[TCA_FQ_CODEL_MAX + 1]; 374 u32 quantum = 0; 375 int err; 376 377 err = nla_parse_nested_deprecated(tb, TCA_FQ_CODEL_MAX, opt, 378 fq_codel_policy, NULL); 379 if (err < 0) 380 return err; 381 if (tb[TCA_FQ_CODEL_FLOWS]) { 382 if (q->flows) 383 return -EINVAL; 384 q->flows_cnt = nla_get_u32(tb[TCA_FQ_CODEL_FLOWS]); 385 if (!q->flows_cnt || 386 q->flows_cnt > 65536) 387 return -EINVAL; 388 } 389 if (tb[TCA_FQ_CODEL_QUANTUM]) { 390 quantum = max(256U, nla_get_u32(tb[TCA_FQ_CODEL_QUANTUM])); 391 if (quantum > FQ_CODEL_QUANTUM_MAX) { 392 NL_SET_ERR_MSG(extack, "Invalid quantum"); 393 return -EINVAL; 394 } 395 } 396 sch_tree_lock(sch); 397 398 if (tb[TCA_FQ_CODEL_TARGET]) { 399 u64 target = nla_get_u32(tb[TCA_FQ_CODEL_TARGET]); 400 401 WRITE_ONCE(q->cparams.target, 402 (target * NSEC_PER_USEC) >> CODEL_SHIFT); 403 } 404 405 if (tb[TCA_FQ_CODEL_CE_THRESHOLD]) { 406 u64 val = nla_get_u32(tb[TCA_FQ_CODEL_CE_THRESHOLD]); 407 408 WRITE_ONCE(q->cparams.ce_threshold, 409 (val * NSEC_PER_USEC) >> CODEL_SHIFT); 410 } 411 412 if (tb[TCA_FQ_CODEL_CE_THRESHOLD_SELECTOR]) 413 WRITE_ONCE(q->cparams.ce_threshold_selector, 414 nla_get_u8(tb[TCA_FQ_CODEL_CE_THRESHOLD_SELECTOR])); 415 if (tb[TCA_FQ_CODEL_CE_THRESHOLD_MASK]) 416 WRITE_ONCE(q->cparams.ce_threshold_mask, 417 nla_get_u8(tb[TCA_FQ_CODEL_CE_THRESHOLD_MASK])); 418 419 if (tb[TCA_FQ_CODEL_INTERVAL]) { 420 u64 interval = nla_get_u32(tb[TCA_FQ_CODEL_INTERVAL]); 421 422 WRITE_ONCE(q->cparams.interval, 423 (interval * NSEC_PER_USEC) >> CODEL_SHIFT); 424 } 425 426 if (tb[TCA_FQ_CODEL_LIMIT]) 427 WRITE_ONCE(sch->limit, 428 nla_get_u32(tb[TCA_FQ_CODEL_LIMIT])); 429 430 if (tb[TCA_FQ_CODEL_ECN]) 431 WRITE_ONCE(q->cparams.ecn, 432 !!nla_get_u32(tb[TCA_FQ_CODEL_ECN])); 433 434 if (quantum) 435 WRITE_ONCE(q->quantum, quantum); 436 437 if (tb[TCA_FQ_CODEL_DROP_BATCH_SIZE]) 438 WRITE_ONCE(q->drop_batch_size, 439 max(1U, nla_get_u32(tb[TCA_FQ_CODEL_DROP_BATCH_SIZE]))); 440 441 if (tb[TCA_FQ_CODEL_MEMORY_LIMIT]) 442 WRITE_ONCE(q->memory_limit, 443 min(1U << 31, nla_get_u32(tb[TCA_FQ_CODEL_MEMORY_LIMIT]))); 444 445 while (sch->q.qlen > sch->limit || 446 q->memory_usage > q->memory_limit) { 447 struct sk_buff *skb = qdisc_dequeue_internal(sch, false); 448 449 if (!skb) 450 break; 451 452 dropped_pkts++; 453 dropped_bytes += qdisc_pkt_len(skb); 454 rtnl_kfree_skbs(skb, skb); 455 } 456 qdisc_tree_reduce_backlog(sch, dropped_pkts, dropped_bytes); 457 458 sch_tree_unlock(sch); 459 return 0; 460 } 461 462 static void fq_codel_destroy(struct Qdisc *sch) 463 { 464 struct fq_codel_sched_data *q = qdisc_priv(sch); 465 466 tcf_block_put(q->block); 467 kvfree(q->backlogs); 468 kvfree(q->flows); 469 } 470 471 static int fq_codel_init(struct Qdisc *sch, struct nlattr *opt, 472 struct netlink_ext_ack *extack) 473 { 474 struct fq_codel_sched_data *q = qdisc_priv(sch); 475 int i; 476 int err; 477 478 sch->limit = 10*1024; 479 q->flows_cnt = 1024; 480 q->memory_limit = 32 << 20; /* 32 MBytes */ 481 q->drop_batch_size = 64; 482 q->quantum = psched_mtu(qdisc_dev(sch)); 483 INIT_LIST_HEAD(&q->new_flows); 484 INIT_LIST_HEAD(&q->old_flows); 485 codel_params_init(&q->cparams); 486 codel_stats_init(&q->cstats); 487 q->cparams.ecn = true; 488 q->cparams.mtu = psched_mtu(qdisc_dev(sch)); 489 490 if (opt) { 491 err = fq_codel_change(sch, opt, extack); 492 if (err) 493 goto init_failure; 494 } 495 496 err = tcf_block_get(&q->block, &q->filter_list, sch, extack); 497 if (err) 498 goto init_failure; 499 500 if (!q->flows) { 501 q->flows = kvzalloc_objs(struct fq_codel_flow, q->flows_cnt); 502 if (!q->flows) { 503 err = -ENOMEM; 504 goto init_failure; 505 } 506 q->backlogs = kvcalloc(q->flows_cnt, sizeof(u32), GFP_KERNEL); 507 if (!q->backlogs) { 508 err = -ENOMEM; 509 goto alloc_failure; 510 } 511 for (i = 0; i < q->flows_cnt; i++) { 512 struct fq_codel_flow *flow = q->flows + i; 513 514 INIT_LIST_HEAD(&flow->flowchain); 515 codel_vars_init(&flow->cvars); 516 } 517 } 518 if (sch->limit >= 1) 519 sch->flags |= TCQ_F_CAN_BYPASS; 520 else 521 sch->flags &= ~TCQ_F_CAN_BYPASS; 522 523 sch->flags |= TCQ_F_DEQUEUE_DROPS; 524 525 return 0; 526 527 alloc_failure: 528 kvfree(q->flows); 529 q->flows = NULL; 530 init_failure: 531 q->flows_cnt = 0; 532 return err; 533 } 534 535 static int fq_codel_dump(struct Qdisc *sch, struct sk_buff *skb) 536 { 537 struct fq_codel_sched_data *q = qdisc_priv(sch); 538 codel_time_t ce_threshold; 539 struct nlattr *opts; 540 541 opts = nla_nest_start_noflag(skb, TCA_OPTIONS); 542 if (opts == NULL) 543 goto nla_put_failure; 544 545 if (nla_put_u32(skb, TCA_FQ_CODEL_TARGET, 546 codel_time_to_us(READ_ONCE(q->cparams.target))) || 547 nla_put_u32(skb, TCA_FQ_CODEL_LIMIT, 548 READ_ONCE(sch->limit)) || 549 nla_put_u32(skb, TCA_FQ_CODEL_INTERVAL, 550 codel_time_to_us(READ_ONCE(q->cparams.interval))) || 551 nla_put_u32(skb, TCA_FQ_CODEL_ECN, 552 READ_ONCE(q->cparams.ecn)) || 553 nla_put_u32(skb, TCA_FQ_CODEL_QUANTUM, 554 READ_ONCE(q->quantum)) || 555 nla_put_u32(skb, TCA_FQ_CODEL_DROP_BATCH_SIZE, 556 READ_ONCE(q->drop_batch_size)) || 557 nla_put_u32(skb, TCA_FQ_CODEL_MEMORY_LIMIT, 558 READ_ONCE(q->memory_limit)) || 559 nla_put_u32(skb, TCA_FQ_CODEL_FLOWS, 560 READ_ONCE(q->flows_cnt))) 561 goto nla_put_failure; 562 563 ce_threshold = READ_ONCE(q->cparams.ce_threshold); 564 if (ce_threshold != CODEL_DISABLED_THRESHOLD) { 565 if (nla_put_u32(skb, TCA_FQ_CODEL_CE_THRESHOLD, 566 codel_time_to_us(ce_threshold))) 567 goto nla_put_failure; 568 if (nla_put_u8(skb, TCA_FQ_CODEL_CE_THRESHOLD_SELECTOR, 569 READ_ONCE(q->cparams.ce_threshold_selector))) 570 goto nla_put_failure; 571 if (nla_put_u8(skb, TCA_FQ_CODEL_CE_THRESHOLD_MASK, 572 READ_ONCE(q->cparams.ce_threshold_mask))) 573 goto nla_put_failure; 574 } 575 576 return nla_nest_end(skb, opts); 577 578 nla_put_failure: 579 return -1; 580 } 581 582 static int fq_codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 583 { 584 struct fq_codel_sched_data *q = qdisc_priv(sch); 585 struct tc_fq_codel_xstats st = { 586 .type = TCA_FQ_CODEL_XSTATS_QDISC, 587 }; 588 struct list_head *pos; 589 590 sch_tree_lock(sch); 591 592 st.qdisc_stats.maxpacket = q->cstats.maxpacket; 593 st.qdisc_stats.drop_overlimit = q->drop_overlimit; 594 st.qdisc_stats.ecn_mark = q->cstats.ecn_mark; 595 st.qdisc_stats.new_flow_count = q->new_flow_count; 596 st.qdisc_stats.ce_mark = q->cstats.ce_mark; 597 st.qdisc_stats.memory_usage = q->memory_usage; 598 st.qdisc_stats.drop_overmemory = q->drop_overmemory; 599 600 list_for_each(pos, &q->new_flows) 601 st.qdisc_stats.new_flows_len++; 602 603 list_for_each(pos, &q->old_flows) 604 st.qdisc_stats.old_flows_len++; 605 sch_tree_unlock(sch); 606 607 return gnet_stats_copy_app(d, &st, sizeof(st)); 608 } 609 610 static struct Qdisc *fq_codel_leaf(struct Qdisc *sch, unsigned long arg) 611 { 612 return NULL; 613 } 614 615 static unsigned long fq_codel_find(struct Qdisc *sch, u32 classid) 616 { 617 return 0; 618 } 619 620 static unsigned long fq_codel_bind(struct Qdisc *sch, unsigned long parent, 621 u32 classid) 622 { 623 return 0; 624 } 625 626 static void fq_codel_unbind(struct Qdisc *q, unsigned long cl) 627 { 628 } 629 630 static struct tcf_block *fq_codel_tcf_block(struct Qdisc *sch, unsigned long cl, 631 struct netlink_ext_ack *extack) 632 { 633 struct fq_codel_sched_data *q = qdisc_priv(sch); 634 635 if (cl) 636 return NULL; 637 return q->block; 638 } 639 640 static int fq_codel_dump_class(struct Qdisc *sch, unsigned long cl, 641 struct sk_buff *skb, struct tcmsg *tcm) 642 { 643 tcm->tcm_handle |= TC_H_MIN(cl); 644 return 0; 645 } 646 647 static int fq_codel_dump_class_stats(struct Qdisc *sch, unsigned long cl, 648 struct gnet_dump *d) 649 { 650 struct fq_codel_sched_data *q = qdisc_priv(sch); 651 u32 idx = cl - 1; 652 struct gnet_stats_queue qs = { 0 }; 653 struct tc_fq_codel_xstats xstats; 654 655 if (idx < q->flows_cnt) { 656 const struct fq_codel_flow *flow = &q->flows[idx]; 657 const struct sk_buff *skb; 658 659 memset(&xstats, 0, sizeof(xstats)); 660 xstats.type = TCA_FQ_CODEL_XSTATS_CLASS; 661 xstats.class_stats.deficit = READ_ONCE(flow->deficit); 662 xstats.class_stats.ldelay = 663 codel_time_to_us(READ_ONCE(flow->cvars.ldelay)); 664 xstats.class_stats.count = READ_ONCE(flow->cvars.count); 665 xstats.class_stats.lastcount = READ_ONCE(flow->cvars.lastcount); 666 xstats.class_stats.dropping = READ_ONCE(flow->cvars.dropping); 667 if (xstats.class_stats.dropping) { 668 codel_tdiff_t delta = READ_ONCE(flow->cvars.drop_next) - 669 codel_get_time(); 670 671 xstats.class_stats.drop_next = (delta >= 0) ? 672 codel_time_to_us(delta) : 673 -codel_time_to_us(-delta); 674 } 675 if (READ_ONCE(flow->head)) { 676 sch_tree_lock(sch); 677 skb = flow->head; 678 while (skb) { 679 qs.qlen++; 680 skb = skb->next; 681 } 682 sch_tree_unlock(sch); 683 } 684 qs.backlog = READ_ONCE(q->backlogs[idx]); 685 qs.drops = 0; 686 } 687 if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0) 688 return -1; 689 if (idx < q->flows_cnt) 690 return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); 691 return 0; 692 } 693 694 static void fq_codel_walk(struct Qdisc *sch, struct qdisc_walker *arg) 695 { 696 struct fq_codel_sched_data *q = qdisc_priv(sch); 697 unsigned int i; 698 699 if (arg->stop) 700 return; 701 702 for (i = 0; i < q->flows_cnt; i++) { 703 if (list_empty(&q->flows[i].flowchain)) { 704 arg->count++; 705 continue; 706 } 707 if (!tc_qdisc_stats_dump(sch, i + 1, arg)) 708 break; 709 } 710 } 711 712 static const struct Qdisc_class_ops fq_codel_class_ops = { 713 .leaf = fq_codel_leaf, 714 .find = fq_codel_find, 715 .tcf_block = fq_codel_tcf_block, 716 .bind_tcf = fq_codel_bind, 717 .unbind_tcf = fq_codel_unbind, 718 .dump = fq_codel_dump_class, 719 .dump_stats = fq_codel_dump_class_stats, 720 .walk = fq_codel_walk, 721 }; 722 723 static struct Qdisc_ops fq_codel_qdisc_ops __read_mostly = { 724 .cl_ops = &fq_codel_class_ops, 725 .id = "fq_codel", 726 .priv_size = sizeof(struct fq_codel_sched_data), 727 .enqueue = fq_codel_enqueue, 728 .dequeue = fq_codel_dequeue, 729 .peek = qdisc_peek_dequeued, 730 .init = fq_codel_init, 731 .reset = fq_codel_reset, 732 .destroy = fq_codel_destroy, 733 .change = fq_codel_change, 734 .dump = fq_codel_dump, 735 .dump_stats = fq_codel_dump_stats, 736 .owner = THIS_MODULE, 737 }; 738 MODULE_ALIAS_NET_SCH("fq_codel"); 739 740 static int __init fq_codel_module_init(void) 741 { 742 return register_qdisc(&fq_codel_qdisc_ops); 743 } 744 745 static void __exit fq_codel_module_exit(void) 746 { 747 unregister_qdisc(&fq_codel_qdisc_ops); 748 } 749 750 module_init(fq_codel_module_init) 751 module_exit(fq_codel_module_exit) 752 MODULE_AUTHOR("Eric Dumazet"); 753 MODULE_LICENSE("GPL"); 754 MODULE_DESCRIPTION("Fair Queue CoDel discipline"); 755