net/sched/sch_fq_pie.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Flow Queue PIE discipline
   3  *
   4  * Copyright (C) 2019 Mohit P. Tahiliani <tahiliani@nitk.edu.in>
   5  * Copyright (C) 2019 Sachin D. Patil <sdp.sachin@gmail.com>
   6  * Copyright (C) 2019 V. Saicharan <vsaicharan1998@gmail.com>
   7  * Copyright (C) 2019 Mohit Bhasi <mohitbhasi1998@gmail.com>
   8  * Copyright (C) 2019 Leslie Monis <lesliemonis@gmail.com>
   9  * Copyright (C) 2019 Gautam Ramakrishnan <gautamramk@gmail.com>
  10  */
  11
  12 #include <linux/jhash.h>
  13 #include <linux/sizes.h>
  14 #include <linux/vmalloc.h>
  15 #include <net/pkt_cls.h>
  16 #include <net/pie.h>
  17
  18 /* Flow Queue PIE
  19  *
  20  * Principles:
  21  *   - Packets are classified on flows.
  22  *   - This is a Stochastic model (as we use a hash, several flows might
  23  *                                 be hashed to the same slot)
  24  *   - Each flow has a PIE managed queue.
  25  *   - Flows are linked onto two (Round Robin) lists,
  26  *     so that new flows have priority on old ones.
  27  *   - For a given flow, packets are not reordered.
  28  *   - Drops during enqueue only.
  29  *   - ECN capability is off by default.
  30  *   - ECN threshold (if ECN is enabled) is at 10% by default.
  31  *   - Uses timestamps to calculate queue delay by default.
  32  */
  33
  34 /**
  35  * struct fq_pie_flow - contains data for each flow
  36  * @vars:       pie vars associated with the flow
  37  * @deficit:    number of remaining byte credits
  38  * @backlog:    size of data in the flow
  39  * @qlen:       number of packets in the flow
  40  * @flowchain:  flowchain for the flow
  41  * @head:       first packet in the flow
  42  * @tail:       last packet in the flow
  43  */
  44 struct fq_pie_flow {
  45         struct pie_vars vars;
  46         s32 deficit;
  47         u32 backlog;
  48         u32 qlen;
  49         struct list_head flowchain;
  50         struct sk_buff *head;
  51         struct sk_buff *tail;
  52 };
  53
  54 struct fq_pie_sched_data {
  55         struct tcf_proto __rcu *filter_list; /* optional external classifier */
  56         struct tcf_block *block;
  57         struct fq_pie_flow *flows;
  58         struct Qdisc *sch;
  59         struct list_head old_flows;
  60         struct list_head new_flows;
  61         struct pie_params p_params;
  62         u32 ecn_prob;
  63         u32 flows_cnt;
  64         u32 quantum;
  65         u32 memory_limit;
  66         u32 new_flow_count;
  67         u32 memory_usage;
  68         u32 overmemory;
  69         struct pie_stats stats;
  70         struct timer_list adapt_timer;
  71 };
  72
  73 static unsigned int fq_pie_hash(const struct fq_pie_sched_data *q,
  74                                 struct sk_buff *skb)
  75 {
  76         return reciprocal_scale(skb_get_hash(skb), q->flows_cnt);
  77 }
  78
  79 static unsigned int fq_pie_classify(struct sk_buff *skb, struct Qdisc *sch,
  80                                     int *qerr)
  81 {
  82         struct fq_pie_sched_data *q = qdisc_priv(sch);
  83         struct tcf_proto *filter;
  84         struct tcf_result res;
  85         int result;
  86
  87         if (TC_H_MAJ(skb->priority) == sch->handle &&
  88             TC_H_MIN(skb->priority) > 0 &&
  89             TC_H_MIN(skb->priority) <= q->flows_cnt)
  90                 return TC_H_MIN(skb->priority);
  91
  92         filter = rcu_dereference_bh(q->filter_list);
  93         if (!filter)
  94                 return fq_pie_hash(q, skb) + 1;
  95
  96         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
  97         result = tcf_classify(skb, filter, &res, false);
  98         if (result >= 0) {
  99 #ifdef CONFIG_NET_CLS_ACT
 100                 switch (result) {
 101                 case TC_ACT_STOLEN:
 102                 case TC_ACT_QUEUED:
 103                 case TC_ACT_TRAP:
 104                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 105                         fallthrough;
 106                 case TC_ACT_SHOT:
 107                         return 0;
 108                 }
 109 #endif
 110                 if (TC_H_MIN(res.classid) <= q->flows_cnt)
 111                         return TC_H_MIN(res.classid);
 112         }
 113         return 0;
 114 }
 115
 116 /* add skb to flow queue (tail add) */
 117 static inline void flow_queue_add(struct fq_pie_flow *flow,
 118                                   struct sk_buff *skb)
 119 {
 120         if (!flow->head)
 121                 flow->head = skb;
 122         else
 123                 flow->tail->next = skb;
 124         flow->tail = skb;
 125         skb->next = NULL;
 126 }
 127
 128 static int fq_pie_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 129                                 struct sk_buff **to_free)
 130 {
 131         struct fq_pie_sched_data *q = qdisc_priv(sch);
 132         struct fq_pie_flow *sel_flow;
 133         int ret;
 134         u8 memory_limited = false;
 135         u8 enqueue = false;
 136         u32 pkt_len;
 137         u32 idx;
 138
 139         /* Classifies packet into corresponding flow */
 140         idx = fq_pie_classify(skb, sch, &ret);
 141         sel_flow = &q->flows[idx];
 142
 143         /* Checks whether adding a new packet would exceed memory limit */
 144         get_pie_cb(skb)->mem_usage = skb->truesize;
 145         memory_limited = q->memory_usage > q->memory_limit + skb->truesize;
 146
 147         /* Checks if the qdisc is full */
 148         if (unlikely(qdisc_qlen(sch) >= sch->limit)) {
 149                 q->stats.overlimit++;
 150                 goto out;
 151         } else if (unlikely(memory_limited)) {
 152                 q->overmemory++;
 153         }
 154
 155         if (!pie_drop_early(sch, &q->p_params, &sel_flow->vars,
 156                             sel_flow->backlog, skb->len)) {
 157                 enqueue = true;
 158         } else if (q->p_params.ecn &&
 159                    sel_flow->vars.prob <= (MAX_PROB / 100) * q->ecn_prob &&
 160                    INET_ECN_set_ce(skb)) {
 161                 /* If packet is ecn capable, mark it if drop probability
 162                  * is lower than the parameter ecn_prob, else drop it.
 163                  */
 164                 q->stats.ecn_mark++;
 165                 enqueue = true;
 166         }
 167         if (enqueue) {
 168                 /* Set enqueue time only when dq_rate_estimator is disabled. */
 169                 if (!q->p_params.dq_rate_estimator)
 170                         pie_set_enqueue_time(skb);
 171
 172                 pkt_len = qdisc_pkt_len(skb);
 173                 q->stats.packets_in++;
 174                 q->memory_usage += skb->truesize;
 175                 sch->qstats.backlog += pkt_len;
 176                 sch->q.qlen++;
 177                 flow_queue_add(sel_flow, skb);
 178                 if (list_empty(&sel_flow->flowchain)) {
 179                         list_add_tail(&sel_flow->flowchain, &q->new_flows);
 180                         q->new_flow_count++;
 181                         sel_flow->deficit = q->quantum;
 182                         sel_flow->qlen = 0;
 183                         sel_flow->backlog = 0;
 184                 }
 185                 sel_flow->qlen++;
 186                 sel_flow->backlog += pkt_len;
 187                 return NET_XMIT_SUCCESS;
 188         }
 189 out:
 190         q->stats.dropped++;
 191         sel_flow->vars.accu_prob = 0;
 192         __qdisc_drop(skb, to_free);
 193         qdisc_qstats_drop(sch);
 194         return NET_XMIT_CN;
 195 }
 196
 197 static const struct nla_policy fq_pie_policy[TCA_FQ_PIE_MAX + 1] = {
 198         [TCA_FQ_PIE_LIMIT]              = {.type = NLA_U32},
 199         [TCA_FQ_PIE_FLOWS]              = {.type = NLA_U32},
 200         [TCA_FQ_PIE_TARGET]             = {.type = NLA_U32},
 201         [TCA_FQ_PIE_TUPDATE]            = {.type = NLA_U32},
 202         [TCA_FQ_PIE_ALPHA]              = {.type = NLA_U32},
 203         [TCA_FQ_PIE_BETA]               = {.type = NLA_U32},
 204         [TCA_FQ_PIE_QUANTUM]            = {.type = NLA_U32},
 205         [TCA_FQ_PIE_MEMORY_LIMIT]       = {.type = NLA_U32},
 206         [TCA_FQ_PIE_ECN_PROB]           = {.type = NLA_U32},
 207         [TCA_FQ_PIE_ECN]                = {.type = NLA_U32},
 208         [TCA_FQ_PIE_BYTEMODE]           = {.type = NLA_U32},
 209         [TCA_FQ_PIE_DQ_RATE_ESTIMATOR]  = {.type = NLA_U32},
 210 };
 211
 212 static inline struct sk_buff *dequeue_head(struct fq_pie_flow *flow)
 213 {
 214         struct sk_buff *skb = flow->head;
 215
 216         flow->head = skb->next;
 217         skb->next = NULL;
 218         return skb;
 219 }
 220
 221 static struct sk_buff *fq_pie_qdisc_dequeue(struct Qdisc *sch)
 222 {
 223         struct fq_pie_sched_data *q = qdisc_priv(sch);
 224         struct sk_buff *skb = NULL;
 225         struct fq_pie_flow *flow;
 226         struct list_head *head;
 227         u32 pkt_len;
 228
 229 begin:
 230         head = &q->new_flows;
 231         if (list_empty(head)) {
 232                 head = &q->old_flows;
 233                 if (list_empty(head))
 234                         return NULL;
 235         }
 236
 237         flow = list_first_entry(head, struct fq_pie_flow, flowchain);
 238         /* Flow has exhausted all its credits */
 239         if (flow->deficit <= 0) {
 240                 flow->deficit += q->quantum;
 241                 list_move_tail(&flow->flowchain, &q->old_flows);
 242                 goto begin;
 243         }
 244
 245         if (flow->head) {
 246                 skb = dequeue_head(flow);
 247                 pkt_len = qdisc_pkt_len(skb);
 248                 sch->qstats.backlog -= pkt_len;
 249                 sch->q.qlen--;
 250                 qdisc_bstats_update(sch, skb);
 251         }
 252
 253         if (!skb) {
 254                 /* force a pass through old_flows to prevent starvation */
 255                 if (head == &q->new_flows && !list_empty(&q->old_flows))
 256                         list_move_tail(&flow->flowchain, &q->old_flows);
 257                 else
 258                         list_del_init(&flow->flowchain);
 259                 goto begin;
 260         }
 261
 262         flow->qlen--;
 263         flow->deficit -= pkt_len;
 264         flow->backlog -= pkt_len;
 265         q->memory_usage -= get_pie_cb(skb)->mem_usage;
 266         pie_process_dequeue(skb, &q->p_params, &flow->vars, flow->backlog);
 267         return skb;
 268 }
 269
 270 static int fq_pie_change(struct Qdisc *sch, struct nlattr *opt,
 271                          struct netlink_ext_ack *extack)
 272 {
 273         struct fq_pie_sched_data *q = qdisc_priv(sch);
 274         struct nlattr *tb[TCA_FQ_PIE_MAX + 1];
 275         unsigned int len_dropped = 0;
 276         unsigned int num_dropped = 0;
 277         int err;
 278
 279         if (!opt)
 280                 return -EINVAL;
 281
 282         err = nla_parse_nested(tb, TCA_FQ_PIE_MAX, opt, fq_pie_policy, extack);
 283         if (err < 0)
 284                 return err;
 285
 286         sch_tree_lock(sch);
 287         if (tb[TCA_FQ_PIE_LIMIT]) {
 288                 u32 limit = nla_get_u32(tb[TCA_FQ_PIE_LIMIT]);
 289
 290                 q->p_params.limit = limit;
 291                 sch->limit = limit;
 292         }
 293         if (tb[TCA_FQ_PIE_FLOWS]) {
 294                 if (q->flows) {
 295                         NL_SET_ERR_MSG_MOD(extack,
 296                                            "Number of flows cannot be changed");
 297                         goto flow_error;
 298                 }
 299                 q->flows_cnt = nla_get_u32(tb[TCA_FQ_PIE_FLOWS]);
 300                 if (!q->flows_cnt || q->flows_cnt >= 65536) {
 301                         NL_SET_ERR_MSG_MOD(extack,
 302                                            "Number of flows must range in [1..65535]");
 303                         goto flow_error;
 304                 }
 305         }
 306
 307         /* convert from microseconds to pschedtime */
 308         if (tb[TCA_FQ_PIE_TARGET]) {
 309                 /* target is in us */
 310                 u32 target = nla_get_u32(tb[TCA_FQ_PIE_TARGET]);
 311
 312                 /* convert to pschedtime */
 313                 q->p_params.target =
 314                         PSCHED_NS2TICKS((u64)target * NSEC_PER_USEC);
 315         }
 316
 317         /* tupdate is in jiffies */
 318         if (tb[TCA_FQ_PIE_TUPDATE])
 319                 q->p_params.tupdate =
 320                         usecs_to_jiffies(nla_get_u32(tb[TCA_FQ_PIE_TUPDATE]));
 321
 322         if (tb[TCA_FQ_PIE_ALPHA])
 323                 q->p_params.alpha = nla_get_u32(tb[TCA_FQ_PIE_ALPHA]);
 324
 325         if (tb[TCA_FQ_PIE_BETA])
 326                 q->p_params.beta = nla_get_u32(tb[TCA_FQ_PIE_BETA]);
 327
 328         if (tb[TCA_FQ_PIE_QUANTUM])
 329                 q->quantum = nla_get_u32(tb[TCA_FQ_PIE_QUANTUM]);
 330
 331         if (tb[TCA_FQ_PIE_MEMORY_LIMIT])
 332                 q->memory_limit = nla_get_u32(tb[TCA_FQ_PIE_MEMORY_LIMIT]);
 333
 334         if (tb[TCA_FQ_PIE_ECN_PROB])
 335                 q->ecn_prob = nla_get_u32(tb[TCA_FQ_PIE_ECN_PROB]);
 336
 337         if (tb[TCA_FQ_PIE_ECN])
 338                 q->p_params.ecn = nla_get_u32(tb[TCA_FQ_PIE_ECN]);
 339
 340         if (tb[TCA_FQ_PIE_BYTEMODE])
 341                 q->p_params.bytemode = nla_get_u32(tb[TCA_FQ_PIE_BYTEMODE]);
 342
 343         if (tb[TCA_FQ_PIE_DQ_RATE_ESTIMATOR])
 344                 q->p_params.dq_rate_estimator =
 345                         nla_get_u32(tb[TCA_FQ_PIE_DQ_RATE_ESTIMATOR]);
 346
 347         /* Drop excess packets if new limit is lower */
 348         while (sch->q.qlen > sch->limit) {
 349                 struct sk_buff *skb = fq_pie_qdisc_dequeue(sch);
 350
 351                 len_dropped += qdisc_pkt_len(skb);
 352                 num_dropped += 1;
 353                 rtnl_kfree_skbs(skb, skb);
 354         }
 355         qdisc_tree_reduce_backlog(sch, num_dropped, len_dropped);
 356
 357         sch_tree_unlock(sch);
 358         return 0;
 359
 360 flow_error:
 361         sch_tree_unlock(sch);
 362         return -EINVAL;
 363 }
 364
 365 static void fq_pie_timer(struct timer_list *t)
 366 {
 367         struct fq_pie_sched_data *q = from_timer(q, t, adapt_timer);
 368         struct Qdisc *sch = q->sch;
 369         spinlock_t *root_lock; /* to lock qdisc for probability calculations */
 370         u16 idx;
 371
 372         root_lock = qdisc_lock(qdisc_root_sleeping(sch));
 373         spin_lock(root_lock);
 374
 375         for (idx = 0; idx < q->flows_cnt; idx++)
 376                 pie_calculate_probability(&q->p_params, &q->flows[idx].vars,
 377                                           q->flows[idx].backlog);
 378
 379         /* reset the timer to fire after 'tupdate' jiffies. */
 380         if (q->p_params.tupdate)
 381                 mod_timer(&q->adapt_timer, jiffies + q->p_params.tupdate);
 382
 383         spin_unlock(root_lock);
 384 }
 385
 386 static int fq_pie_init(struct Qdisc *sch, struct nlattr *opt,
 387                        struct netlink_ext_ack *extack)
 388 {
 389         struct fq_pie_sched_data *q = qdisc_priv(sch);
 390         int err;
 391         u16 idx;
 392
 393         pie_params_init(&q->p_params);
 394         sch->limit = 10 * 1024;
 395         q->p_params.limit = sch->limit;
 396         q->quantum = psched_mtu(qdisc_dev(sch));
 397         q->sch = sch;
 398         q->ecn_prob = 10;
 399         q->flows_cnt = 1024;
 400         q->memory_limit = SZ_32M;
 401
 402         INIT_LIST_HEAD(&q->new_flows);
 403         INIT_LIST_HEAD(&q->old_flows);
 404         timer_setup(&q->adapt_timer, fq_pie_timer, 0);
 405
 406         if (opt) {
 407                 err = fq_pie_change(sch, opt, extack);
 408
 409                 if (err)
 410                         return err;
 411         }
 412
 413         err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
 414         if (err)
 415                 goto init_failure;
 416
 417         q->flows = kvcalloc(q->flows_cnt, sizeof(struct fq_pie_flow),
 418                             GFP_KERNEL);
 419         if (!q->flows) {
 420                 err = -ENOMEM;
 421                 goto init_failure;
 422         }
 423         for (idx = 0; idx < q->flows_cnt; idx++) {
 424                 struct fq_pie_flow *flow = q->flows + idx;
 425
 426                 INIT_LIST_HEAD(&flow->flowchain);
 427                 pie_vars_init(&flow->vars);
 428         }
 429
 430         mod_timer(&q->adapt_timer, jiffies + HZ / 2);
 431
 432         return 0;
 433
 434 init_failure:
 435         q->flows_cnt = 0;
 436
 437         return err;
 438 }
 439
 440 static int fq_pie_dump(struct Qdisc *sch, struct sk_buff *skb)
 441 {
 442         struct fq_pie_sched_data *q = qdisc_priv(sch);
 443         struct nlattr *opts;
 444
 445         opts = nla_nest_start(skb, TCA_OPTIONS);
 446         if (!opts)
 447                 return -EMSGSIZE;
 448
 449         /* convert target from pschedtime to us */
 450         if (nla_put_u32(skb, TCA_FQ_PIE_LIMIT, sch->limit) ||
 451             nla_put_u32(skb, TCA_FQ_PIE_FLOWS, q->flows_cnt) ||
 452             nla_put_u32(skb, TCA_FQ_PIE_TARGET,
 453                         ((u32)PSCHED_TICKS2NS(q->p_params.target)) /
 454                         NSEC_PER_USEC) ||
 455             nla_put_u32(skb, TCA_FQ_PIE_TUPDATE,
 456                         jiffies_to_usecs(q->p_params.tupdate)) ||
 457             nla_put_u32(skb, TCA_FQ_PIE_ALPHA, q->p_params.alpha) ||
 458             nla_put_u32(skb, TCA_FQ_PIE_BETA, q->p_params.beta) ||
 459             nla_put_u32(skb, TCA_FQ_PIE_QUANTUM, q->quantum) ||
 460             nla_put_u32(skb, TCA_FQ_PIE_MEMORY_LIMIT, q->memory_limit) ||
 461             nla_put_u32(skb, TCA_FQ_PIE_ECN_PROB, q->ecn_prob) ||
 462             nla_put_u32(skb, TCA_FQ_PIE_ECN, q->p_params.ecn) ||
 463             nla_put_u32(skb, TCA_FQ_PIE_BYTEMODE, q->p_params.bytemode) ||
 464             nla_put_u32(skb, TCA_FQ_PIE_DQ_RATE_ESTIMATOR,
 465                         q->p_params.dq_rate_estimator))
 466                 goto nla_put_failure;
 467
 468         return nla_nest_end(skb, opts);
 469
 470 nla_put_failure:
 471         nla_nest_cancel(skb, opts);
 472         return -EMSGSIZE;
 473 }
 474
 475 static int fq_pie_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 476 {
 477         struct fq_pie_sched_data *q = qdisc_priv(sch);
 478         struct tc_fq_pie_xstats st = {
 479                 .packets_in     = q->stats.packets_in,
 480                 .overlimit      = q->stats.overlimit,
 481                 .overmemory     = q->overmemory,
 482                 .dropped        = q->stats.dropped,
 483                 .ecn_mark       = q->stats.ecn_mark,
 484                 .new_flow_count = q->new_flow_count,
 485                 .memory_usage   = q->memory_usage,
 486         };
 487         struct list_head *pos;
 488
 489         sch_tree_lock(sch);
 490         list_for_each(pos, &q->new_flows)
 491                 st.new_flows_len++;
 492
 493         list_for_each(pos, &q->old_flows)
 494                 st.old_flows_len++;
 495         sch_tree_unlock(sch);
 496
 497         return gnet_stats_copy_app(d, &st, sizeof(st));
 498 }
 499
 500 static void fq_pie_reset(struct Qdisc *sch)
 501 {
 502         struct fq_pie_sched_data *q = qdisc_priv(sch);
 503         u16 idx;
 504
 505         INIT_LIST_HEAD(&q->new_flows);
 506         INIT_LIST_HEAD(&q->old_flows);
 507         for (idx = 0; idx < q->flows_cnt; idx++) {
 508                 struct fq_pie_flow *flow = q->flows + idx;
 509
 510                 /* Removes all packets from flow */
 511                 rtnl_kfree_skbs(flow->head, flow->tail);
 512                 flow->head = NULL;
 513
 514                 INIT_LIST_HEAD(&flow->flowchain);
 515                 pie_vars_init(&flow->vars);
 516         }
 517
 518         sch->q.qlen = 0;
 519         sch->qstats.backlog = 0;
 520 }
 521
 522 static void fq_pie_destroy(struct Qdisc *sch)
 523 {
 524         struct fq_pie_sched_data *q = qdisc_priv(sch);
 525
 526         tcf_block_put(q->block);
 527         del_timer_sync(&q->adapt_timer);
 528         kvfree(q->flows);
 529 }
 530
 531 static struct Qdisc_ops fq_pie_qdisc_ops __read_mostly = {
 532         .id             = "fq_pie",
 533         .priv_size      = sizeof(struct fq_pie_sched_data),
 534         .enqueue        = fq_pie_qdisc_enqueue,
 535         .dequeue        = fq_pie_qdisc_dequeue,
 536         .peek           = qdisc_peek_dequeued,
 537         .init           = fq_pie_init,
 538         .destroy        = fq_pie_destroy,
 539         .reset          = fq_pie_reset,
 540         .change         = fq_pie_change,
 541         .dump           = fq_pie_dump,
 542         .dump_stats     = fq_pie_dump_stats,
 543         .owner          = THIS_MODULE,
 544 };
 545
 546 static int __init fq_pie_module_init(void)
 547 {
 548         return register_qdisc(&fq_pie_qdisc_ops);
 549 }
 550
 551 static void __exit fq_pie_module_exit(void)
 552 {
 553         unregister_qdisc(&fq_pie_qdisc_ops);
 554 }
 555
 556 module_init(fq_pie_module_init);
 557 module_exit(fq_pie_module_exit);
 558
 559 MODULE_DESCRIPTION("Flow Queue Proportional Integral controller Enhanced (FQ-PIE)");
 560 MODULE_AUTHOR("Mohit P. Tahiliani");
 561 MODULE_LICENSE("GPL");