linux/net/sched/sch_qfq.c
<<
>>
Prefs
   1/*
   2 * net/sched/sch_qfq.c         Quick Fair Queueing Scheduler.
   3 *
   4 * Copyright (c) 2009 Fabio Checconi, Luigi Rizzo, and Paolo Valente.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License
   8 * version 2 as published by the Free Software Foundation.
   9 */
  10
  11#include <linux/module.h>
  12#include <linux/init.h>
  13#include <linux/bitops.h>
  14#include <linux/errno.h>
  15#include <linux/netdevice.h>
  16#include <linux/pkt_sched.h>
  17#include <net/sch_generic.h>
  18#include <net/pkt_sched.h>
  19#include <net/pkt_cls.h>
  20
  21
  22/*  Quick Fair Queueing
  23    ===================
  24
  25    Sources:
  26
  27    Fabio Checconi, Luigi Rizzo, and Paolo Valente: "QFQ: Efficient
  28    Packet Scheduling with Tight Bandwidth Distribution Guarantees."
  29
  30    See also:
  31    http://retis.sssup.it/~fabio/linux/qfq/
  32 */
  33
  34/*
  35
  36  Virtual time computations.
  37
  38  S, F and V are all computed in fixed point arithmetic with
  39  FRAC_BITS decimal bits.
  40
  41  QFQ_MAX_INDEX is the maximum index allowed for a group. We need
  42        one bit per index.
  43  QFQ_MAX_WSHIFT is the maximum power of two supported as a weight.
  44
  45  The layout of the bits is as below:
  46
  47                   [ MTU_SHIFT ][      FRAC_BITS    ]
  48                   [ MAX_INDEX    ][ MIN_SLOT_SHIFT ]
  49                                 ^.__grp->index = 0
  50                                 *.__grp->slot_shift
  51
  52  where MIN_SLOT_SHIFT is derived by difference from the others.
  53
  54  The max group index corresponds to Lmax/w_min, where
  55  Lmax=1<<MTU_SHIFT, w_min = 1 .
  56  From this, and knowing how many groups (MAX_INDEX) we want,
  57  we can derive the shift corresponding to each group.
  58
  59  Because we often need to compute
  60        F = S + len/w_i  and V = V + len/wsum
  61  instead of storing w_i store the value
  62        inv_w = (1<<FRAC_BITS)/w_i
  63  so we can do F = S + len * inv_w * wsum.
  64  We use W_TOT in the formulas so we can easily move between
  65  static and adaptive weight sum.
  66
  67  The per-scheduler-instance data contain all the data structures
  68  for the scheduler: bitmaps and bucket lists.
  69
  70 */
  71
  72/*
  73 * Maximum number of consecutive slots occupied by backlogged classes
  74 * inside a group.
  75 */
  76#define QFQ_MAX_SLOTS   32
  77
  78/*
  79 * Shifts used for class<->group mapping.  We allow class weights that are
  80 * in the range [1, 2^MAX_WSHIFT], and we try to map each class i to the
  81 * group with the smallest index that can support the L_i / r_i configured
  82 * for the class.
  83 *
  84 * grp->index is the index of the group; and grp->slot_shift
  85 * is the shift for the corresponding (scaled) sigma_i.
  86 */
  87#define QFQ_MAX_INDEX           24
  88#define QFQ_MAX_WSHIFT          12
  89
  90#define QFQ_MAX_WEIGHT          (1<<QFQ_MAX_WSHIFT)
  91#define QFQ_MAX_WSUM            (16*QFQ_MAX_WEIGHT)
  92
  93#define FRAC_BITS               30      /* fixed point arithmetic */
  94#define ONE_FP                  (1UL << FRAC_BITS)
  95#define IWSUM                   (ONE_FP/QFQ_MAX_WSUM)
  96
  97#define QFQ_MTU_SHIFT           16      /* to support TSO/GSO */
  98#define QFQ_MIN_SLOT_SHIFT      (FRAC_BITS + QFQ_MTU_SHIFT - QFQ_MAX_INDEX)
  99#define QFQ_MIN_LMAX            256     /* min possible lmax for a class */
 100
 101/*
 102 * Possible group states.  These values are used as indexes for the bitmaps
 103 * array of struct qfq_queue.
 104 */
 105enum qfq_state { ER, IR, EB, IB, QFQ_MAX_STATE };
 106
 107struct qfq_group;
 108
 109struct qfq_class {
 110        struct Qdisc_class_common common;
 111
 112        unsigned int refcnt;
 113        unsigned int filter_cnt;
 114
 115        struct gnet_stats_basic_packed bstats;
 116        struct gnet_stats_queue qstats;
 117        struct gnet_stats_rate_est rate_est;
 118        struct Qdisc *qdisc;
 119
 120        struct hlist_node next; /* Link for the slot list. */
 121        u64 S, F;               /* flow timestamps (exact) */
 122
 123        /* group we belong to. In principle we would need the index,
 124         * which is log_2(lmax/weight), but we never reference it
 125         * directly, only the group.
 126         */
 127        struct qfq_group *grp;
 128
 129        /* these are copied from the flowset. */
 130        u32     inv_w;          /* ONE_FP/weight */
 131        u32     lmax;           /* Max packet size for this flow. */
 132};
 133
 134struct qfq_group {
 135        u64 S, F;                       /* group timestamps (approx). */
 136        unsigned int slot_shift;        /* Slot shift. */
 137        unsigned int index;             /* Group index. */
 138        unsigned int front;             /* Index of the front slot. */
 139        unsigned long full_slots;       /* non-empty slots */
 140
 141        /* Array of RR lists of active classes. */
 142        struct hlist_head slots[QFQ_MAX_SLOTS];
 143};
 144
 145struct qfq_sched {
 146        struct tcf_proto *filter_list;
 147        struct Qdisc_class_hash clhash;
 148
 149        u64             V;              /* Precise virtual time. */
 150        u32             wsum;           /* weight sum */
 151
 152        unsigned long bitmaps[QFQ_MAX_STATE];       /* Group bitmaps. */
 153        struct qfq_group groups[QFQ_MAX_INDEX + 1]; /* The groups. */
 154};
 155
 156static struct qfq_class *qfq_find_class(struct Qdisc *sch, u32 classid)
 157{
 158        struct qfq_sched *q = qdisc_priv(sch);
 159        struct Qdisc_class_common *clc;
 160
 161        clc = qdisc_class_find(&q->clhash, classid);
 162        if (clc == NULL)
 163                return NULL;
 164        return container_of(clc, struct qfq_class, common);
 165}
 166
 167static void qfq_purge_queue(struct qfq_class *cl)
 168{
 169        unsigned int len = cl->qdisc->q.qlen;
 170
 171        qdisc_reset(cl->qdisc);
 172        qdisc_tree_decrease_qlen(cl->qdisc, len);
 173}
 174
 175static const struct nla_policy qfq_policy[TCA_QFQ_MAX + 1] = {
 176        [TCA_QFQ_WEIGHT] = { .type = NLA_U32 },
 177        [TCA_QFQ_LMAX] = { .type = NLA_U32 },
 178};
 179
 180/*
 181 * Calculate a flow index, given its weight and maximum packet length.
 182 * index = log_2(maxlen/weight) but we need to apply the scaling.
 183 * This is used only once at flow creation.
 184 */
 185static int qfq_calc_index(u32 inv_w, unsigned int maxlen)
 186{
 187        u64 slot_size = (u64)maxlen * inv_w;
 188        unsigned long size_map;
 189        int index = 0;
 190
 191        size_map = slot_size >> QFQ_MIN_SLOT_SHIFT;
 192        if (!size_map)
 193                goto out;
 194
 195        index = __fls(size_map) + 1;    /* basically a log_2 */
 196        index -= !(slot_size - (1ULL << (index + QFQ_MIN_SLOT_SHIFT - 1)));
 197
 198        if (index < 0)
 199                index = 0;
 200out:
 201        pr_debug("qfq calc_index: W = %lu, L = %u, I = %d\n",
 202                 (unsigned long) ONE_FP/inv_w, maxlen, index);
 203
 204        return index;
 205}
 206
 207/* Length of the next packet (0 if the queue is empty). */
 208static unsigned int qdisc_peek_len(struct Qdisc *sch)
 209{
 210        struct sk_buff *skb;
 211
 212        skb = sch->ops->peek(sch);
 213        return skb ? qdisc_pkt_len(skb) : 0;
 214}
 215
 216static void qfq_deactivate_class(struct qfq_sched *, struct qfq_class *);
 217static void qfq_activate_class(struct qfq_sched *q, struct qfq_class *cl,
 218                               unsigned int len);
 219
 220static void qfq_update_class_params(struct qfq_sched *q, struct qfq_class *cl,
 221                                    u32 lmax, u32 inv_w, int delta_w)
 222{
 223        int i;
 224
 225        /* update qfq-specific data */
 226        cl->lmax = lmax;
 227        cl->inv_w = inv_w;
 228        i = qfq_calc_index(cl->inv_w, cl->lmax);
 229
 230        cl->grp = &q->groups[i];
 231
 232        q->wsum += delta_w;
 233}
 234
 235static void qfq_update_reactivate_class(struct qfq_sched *q,
 236                                        struct qfq_class *cl,
 237                                        u32 inv_w, u32 lmax, int delta_w)
 238{
 239        bool need_reactivation = false;
 240        int i = qfq_calc_index(inv_w, lmax);
 241
 242        if (&q->groups[i] != cl->grp && cl->qdisc->q.qlen > 0) {
 243                /*
 244                 * shift cl->F back, to not charge the
 245                 * class for the not-yet-served head
 246                 * packet
 247                 */
 248                cl->F = cl->S;
 249                /* remove class from its slot in the old group */
 250                qfq_deactivate_class(q, cl);
 251                need_reactivation = true;
 252        }
 253
 254        qfq_update_class_params(q, cl, lmax, inv_w, delta_w);
 255
 256        if (need_reactivation) /* activate in new group */
 257                qfq_activate_class(q, cl, qdisc_peek_len(cl->qdisc));
 258}
 259
 260
 261static int qfq_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
 262                            struct nlattr **tca, unsigned long *arg)
 263{
 264        struct qfq_sched *q = qdisc_priv(sch);
 265        struct qfq_class *cl = (struct qfq_class *)*arg;
 266        struct nlattr *tb[TCA_QFQ_MAX + 1];
 267        u32 weight, lmax, inv_w;
 268        int err;
 269        int delta_w;
 270
 271        if (tca[TCA_OPTIONS] == NULL) {
 272                pr_notice("qfq: no options\n");
 273                return -EINVAL;
 274        }
 275
 276        err = nla_parse_nested(tb, TCA_QFQ_MAX, tca[TCA_OPTIONS], qfq_policy);
 277        if (err < 0)
 278                return err;
 279
 280        if (tb[TCA_QFQ_WEIGHT]) {
 281                weight = nla_get_u32(tb[TCA_QFQ_WEIGHT]);
 282                if (!weight || weight > (1UL << QFQ_MAX_WSHIFT)) {
 283                        pr_notice("qfq: invalid weight %u\n", weight);
 284                        return -EINVAL;
 285                }
 286        } else
 287                weight = 1;
 288
 289        inv_w = ONE_FP / weight;
 290        weight = ONE_FP / inv_w;
 291        delta_w = weight - (cl ? ONE_FP / cl->inv_w : 0);
 292        if (q->wsum + delta_w > QFQ_MAX_WSUM) {
 293                pr_notice("qfq: total weight out of range (%u + %u)\n",
 294                          delta_w, q->wsum);
 295                return -EINVAL;
 296        }
 297
 298        if (tb[TCA_QFQ_LMAX]) {
 299                lmax = nla_get_u32(tb[TCA_QFQ_LMAX]);
 300                if (lmax < QFQ_MIN_LMAX || lmax > (1UL << QFQ_MTU_SHIFT)) {
 301                        pr_notice("qfq: invalid max length %u\n", lmax);
 302                        return -EINVAL;
 303                }
 304        } else
 305                lmax = psched_mtu(qdisc_dev(sch));
 306
 307        if (cl != NULL) {
 308                if (tca[TCA_RATE]) {
 309                        err = gen_replace_estimator(&cl->bstats, &cl->rate_est,
 310                                                    qdisc_root_sleeping_lock(sch),
 311                                                    tca[TCA_RATE]);
 312                        if (err)
 313                                return err;
 314                }
 315
 316                if (lmax == cl->lmax && inv_w == cl->inv_w)
 317                        return 0; /* nothing to update */
 318
 319                sch_tree_lock(sch);
 320                qfq_update_reactivate_class(q, cl, inv_w, lmax, delta_w);
 321                sch_tree_unlock(sch);
 322
 323                return 0;
 324        }
 325
 326        cl = kzalloc(sizeof(struct qfq_class), GFP_KERNEL);
 327        if (cl == NULL)
 328                return -ENOBUFS;
 329
 330        cl->refcnt = 1;
 331        cl->common.classid = classid;
 332
 333        qfq_update_class_params(q, cl, lmax, inv_w, delta_w);
 334
 335        cl->qdisc = qdisc_create_dflt(sch->dev_queue,
 336                                      &pfifo_qdisc_ops, classid);
 337        if (cl->qdisc == NULL)
 338                cl->qdisc = &noop_qdisc;
 339
 340        if (tca[TCA_RATE]) {
 341                err = gen_new_estimator(&cl->bstats, &cl->rate_est,
 342                                        qdisc_root_sleeping_lock(sch),
 343                                        tca[TCA_RATE]);
 344                if (err) {
 345                        qdisc_destroy(cl->qdisc);
 346                        kfree(cl);
 347                        return err;
 348                }
 349        }
 350
 351        sch_tree_lock(sch);
 352        qdisc_class_hash_insert(&q->clhash, &cl->common);
 353        sch_tree_unlock(sch);
 354
 355        qdisc_class_hash_grow(sch, &q->clhash);
 356
 357        *arg = (unsigned long)cl;
 358        return 0;
 359}
 360
 361static void qfq_destroy_class(struct Qdisc *sch, struct qfq_class *cl)
 362{
 363        struct qfq_sched *q = qdisc_priv(sch);
 364
 365        if (cl->inv_w) {
 366                q->wsum -= ONE_FP / cl->inv_w;
 367                cl->inv_w = 0;
 368        }
 369
 370        gen_kill_estimator(&cl->bstats, &cl->rate_est);
 371        qdisc_destroy(cl->qdisc);
 372        kfree(cl);
 373}
 374
 375static int qfq_delete_class(struct Qdisc *sch, unsigned long arg)
 376{
 377        struct qfq_sched *q = qdisc_priv(sch);
 378        struct qfq_class *cl = (struct qfq_class *)arg;
 379
 380        if (cl->filter_cnt > 0)
 381                return -EBUSY;
 382
 383        sch_tree_lock(sch);
 384
 385        qfq_purge_queue(cl);
 386        qdisc_class_hash_remove(&q->clhash, &cl->common);
 387
 388        BUG_ON(--cl->refcnt == 0);
 389        /*
 390         * This shouldn't happen: we "hold" one cops->get() when called
 391         * from tc_ctl_tclass; the destroy method is done from cops->put().
 392         */
 393
 394        sch_tree_unlock(sch);
 395        return 0;
 396}
 397
 398static unsigned long qfq_get_class(struct Qdisc *sch, u32 classid)
 399{
 400        struct qfq_class *cl = qfq_find_class(sch, classid);
 401
 402        if (cl != NULL)
 403                cl->refcnt++;
 404
 405        return (unsigned long)cl;
 406}
 407
 408static void qfq_put_class(struct Qdisc *sch, unsigned long arg)
 409{
 410        struct qfq_class *cl = (struct qfq_class *)arg;
 411
 412        if (--cl->refcnt == 0)
 413                qfq_destroy_class(sch, cl);
 414}
 415
 416static struct tcf_proto **qfq_tcf_chain(struct Qdisc *sch, unsigned long cl)
 417{
 418        struct qfq_sched *q = qdisc_priv(sch);
 419
 420        if (cl)
 421                return NULL;
 422
 423        return &q->filter_list;
 424}
 425
 426static unsigned long qfq_bind_tcf(struct Qdisc *sch, unsigned long parent,
 427                                  u32 classid)
 428{
 429        struct qfq_class *cl = qfq_find_class(sch, classid);
 430
 431        if (cl != NULL)
 432                cl->filter_cnt++;
 433
 434        return (unsigned long)cl;
 435}
 436
 437static void qfq_unbind_tcf(struct Qdisc *sch, unsigned long arg)
 438{
 439        struct qfq_class *cl = (struct qfq_class *)arg;
 440
 441        cl->filter_cnt--;
 442}
 443
 444static int qfq_graft_class(struct Qdisc *sch, unsigned long arg,
 445                           struct Qdisc *new, struct Qdisc **old)
 446{
 447        struct qfq_class *cl = (struct qfq_class *)arg;
 448
 449        if (new == NULL) {
 450                new = qdisc_create_dflt(sch->dev_queue,
 451                                        &pfifo_qdisc_ops, cl->common.classid);
 452                if (new == NULL)
 453                        new = &noop_qdisc;
 454        }
 455
 456        sch_tree_lock(sch);
 457        qfq_purge_queue(cl);
 458        *old = cl->qdisc;
 459        cl->qdisc = new;
 460        sch_tree_unlock(sch);
 461        return 0;
 462}
 463
 464static struct Qdisc *qfq_class_leaf(struct Qdisc *sch, unsigned long arg)
 465{
 466        struct qfq_class *cl = (struct qfq_class *)arg;
 467
 468        return cl->qdisc;
 469}
 470
 471static int qfq_dump_class(struct Qdisc *sch, unsigned long arg,
 472                          struct sk_buff *skb, struct tcmsg *tcm)
 473{
 474        struct qfq_class *cl = (struct qfq_class *)arg;
 475        struct nlattr *nest;
 476
 477        tcm->tcm_parent = TC_H_ROOT;
 478        tcm->tcm_handle = cl->common.classid;
 479        tcm->tcm_info   = cl->qdisc->handle;
 480
 481        nest = nla_nest_start(skb, TCA_OPTIONS);
 482        if (nest == NULL)
 483                goto nla_put_failure;
 484        if (nla_put_u32(skb, TCA_QFQ_WEIGHT, ONE_FP/cl->inv_w) ||
 485            nla_put_u32(skb, TCA_QFQ_LMAX, cl->lmax))
 486                goto nla_put_failure;
 487        return nla_nest_end(skb, nest);
 488
 489nla_put_failure:
 490        nla_nest_cancel(skb, nest);
 491        return -EMSGSIZE;
 492}
 493
 494static int qfq_dump_class_stats(struct Qdisc *sch, unsigned long arg,
 495                                struct gnet_dump *d)
 496{
 497        struct qfq_class *cl = (struct qfq_class *)arg;
 498        struct tc_qfq_stats xstats;
 499
 500        memset(&xstats, 0, sizeof(xstats));
 501        cl->qdisc->qstats.qlen = cl->qdisc->q.qlen;
 502
 503        xstats.weight = ONE_FP/cl->inv_w;
 504        xstats.lmax = cl->lmax;
 505
 506        if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
 507            gnet_stats_copy_rate_est(d, &cl->bstats, &cl->rate_est) < 0 ||
 508            gnet_stats_copy_queue(d, &cl->qdisc->qstats) < 0)
 509                return -1;
 510
 511        return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
 512}
 513
 514static void qfq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 515{
 516        struct qfq_sched *q = qdisc_priv(sch);
 517        struct qfq_class *cl;
 518        struct hlist_node *n;
 519        unsigned int i;
 520
 521        if (arg->stop)
 522                return;
 523
 524        for (i = 0; i < q->clhash.hashsize; i++) {
 525                hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) {
 526                        if (arg->count < arg->skip) {
 527                                arg->count++;
 528                                continue;
 529                        }
 530                        if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
 531                                arg->stop = 1;
 532                                return;
 533                        }
 534                        arg->count++;
 535                }
 536        }
 537}
 538
 539static struct qfq_class *qfq_classify(struct sk_buff *skb, struct Qdisc *sch,
 540                                      int *qerr)
 541{
 542        struct qfq_sched *q = qdisc_priv(sch);
 543        struct qfq_class *cl;
 544        struct tcf_result res;
 545        int result;
 546
 547        if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
 548                pr_debug("qfq_classify: found %d\n", skb->priority);
 549                cl = qfq_find_class(sch, skb->priority);
 550                if (cl != NULL)
 551                        return cl;
 552        }
 553
 554        *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
 555        result = tc_classify(skb, q->filter_list, &res);
 556        if (result >= 0) {
 557#ifdef CONFIG_NET_CLS_ACT
 558                switch (result) {
 559                case TC_ACT_QUEUED:
 560                case TC_ACT_STOLEN:
 561                        *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
 562                case TC_ACT_SHOT:
 563                        return NULL;
 564                }
 565#endif
 566                cl = (struct qfq_class *)res.class;
 567                if (cl == NULL)
 568                        cl = qfq_find_class(sch, res.classid);
 569                return cl;
 570        }
 571
 572        return NULL;
 573}
 574
 575/* Generic comparison function, handling wraparound. */
 576static inline int qfq_gt(u64 a, u64 b)
 577{
 578        return (s64)(a - b) > 0;
 579}
 580
 581/* Round a precise timestamp to its slotted value. */
 582static inline u64 qfq_round_down(u64 ts, unsigned int shift)
 583{
 584        return ts & ~((1ULL << shift) - 1);
 585}
 586
 587/* return the pointer to the group with lowest index in the bitmap */
 588static inline struct qfq_group *qfq_ffs(struct qfq_sched *q,
 589                                        unsigned long bitmap)
 590{
 591        int index = __ffs(bitmap);
 592        return &q->groups[index];
 593}
 594/* Calculate a mask to mimic what would be ffs_from(). */
 595static inline unsigned long mask_from(unsigned long bitmap, int from)
 596{
 597        return bitmap & ~((1UL << from) - 1);
 598}
 599
 600/*
 601 * The state computation relies on ER=0, IR=1, EB=2, IB=3
 602 * First compute eligibility comparing grp->S, q->V,
 603 * then check if someone is blocking us and possibly add EB
 604 */
 605static int qfq_calc_state(struct qfq_sched *q, const struct qfq_group *grp)
 606{
 607        /* if S > V we are not eligible */
 608        unsigned int state = qfq_gt(grp->S, q->V);
 609        unsigned long mask = mask_from(q->bitmaps[ER], grp->index);
 610        struct qfq_group *next;
 611
 612        if (mask) {
 613                next = qfq_ffs(q, mask);
 614                if (qfq_gt(grp->F, next->F))
 615                        state |= EB;
 616        }
 617
 618        return state;
 619}
 620
 621
 622/*
 623 * In principle
 624 *      q->bitmaps[dst] |= q->bitmaps[src] & mask;
 625 *      q->bitmaps[src] &= ~mask;
 626 * but we should make sure that src != dst
 627 */
 628static inline void qfq_move_groups(struct qfq_sched *q, unsigned long mask,
 629                                   int src, int dst)
 630{
 631        q->bitmaps[dst] |= q->bitmaps[src] & mask;
 632        q->bitmaps[src] &= ~mask;
 633}
 634
 635static void qfq_unblock_groups(struct qfq_sched *q, int index, u64 old_F)
 636{
 637        unsigned long mask = mask_from(q->bitmaps[ER], index + 1);
 638        struct qfq_group *next;
 639
 640        if (mask) {
 641                next = qfq_ffs(q, mask);
 642                if (!qfq_gt(next->F, old_F))
 643                        return;
 644        }
 645
 646        mask = (1UL << index) - 1;
 647        qfq_move_groups(q, mask, EB, ER);
 648        qfq_move_groups(q, mask, IB, IR);
 649}
 650
 651/*
 652 * perhaps
 653 *
 654        old_V ^= q->V;
 655        old_V >>= QFQ_MIN_SLOT_SHIFT;
 656        if (old_V) {
 657                ...
 658        }
 659 *
 660 */
 661static void qfq_make_eligible(struct qfq_sched *q, u64 old_V)
 662{
 663        unsigned long vslot = q->V >> QFQ_MIN_SLOT_SHIFT;
 664        unsigned long old_vslot = old_V >> QFQ_MIN_SLOT_SHIFT;
 665
 666        if (vslot != old_vslot) {
 667                unsigned long mask = (1UL << fls(vslot ^ old_vslot)) - 1;
 668                qfq_move_groups(q, mask, IR, ER);
 669                qfq_move_groups(q, mask, IB, EB);
 670        }
 671}
 672
 673
 674/*
 675 * If the weight and lmax (max_pkt_size) of the classes do not change,
 676 * then QFQ guarantees that the slot index is never higher than
 677 * 2 + ((1<<QFQ_MTU_SHIFT)/QFQ_MIN_LMAX) * (QFQ_MAX_WEIGHT/QFQ_MAX_WSUM).
 678 *
 679 * With the current values of the above constants, the index is
 680 * then guaranteed to never be higher than 2 + 256 * (1 / 16) = 18.
 681 *
 682 * When the weight of a class is increased or the lmax of the class is
 683 * decreased, a new class with smaller slot size may happen to be
 684 * activated. The activation of this class should be properly delayed
 685 * to when the service of the class has finished in the ideal system
 686 * tracked by QFQ. If the activation of the class is not delayed to
 687 * this reference time instant, then this class may be unjustly served
 688 * before other classes waiting for service. This may cause
 689 * (unfrequently) the above bound to the slot index to be violated for
 690 * some of these unlucky classes.
 691 *
 692 * Instead of delaying the activation of the new class, which is quite
 693 * complex, the following inaccurate but simple solution is used: if
 694 * the slot index is higher than QFQ_MAX_SLOTS-2, then the timestamps
 695 * of the class are shifted backward so as to let the slot index
 696 * become equal to QFQ_MAX_SLOTS-2. This threshold is used because, if
 697 * the slot index is above it, then the data structure implementing
 698 * the bucket list either gets immediately corrupted or may get
 699 * corrupted on a possible next packet arrival that causes the start
 700 * time of the group to be shifted backward.
 701 */
 702static void qfq_slot_insert(struct qfq_group *grp, struct qfq_class *cl,
 703                            u64 roundedS)
 704{
 705        u64 slot = (roundedS - grp->S) >> grp->slot_shift;
 706        unsigned int i; /* slot index in the bucket list */
 707
 708        if (unlikely(slot > QFQ_MAX_SLOTS - 2)) {
 709                u64 deltaS = roundedS - grp->S -
 710                        ((u64)(QFQ_MAX_SLOTS - 2)<<grp->slot_shift);
 711                cl->S -= deltaS;
 712                cl->F -= deltaS;
 713                slot = QFQ_MAX_SLOTS - 2;
 714        }
 715
 716        i = (grp->front + slot) % QFQ_MAX_SLOTS;
 717
 718        hlist_add_head(&cl->next, &grp->slots[i]);
 719        __set_bit(slot, &grp->full_slots);
 720}
 721
 722/* Maybe introduce hlist_first_entry?? */
 723static struct qfq_class *qfq_slot_head(struct qfq_group *grp)
 724{
 725        return hlist_entry(grp->slots[grp->front].first,
 726                           struct qfq_class, next);
 727}
 728
 729/*
 730 * remove the entry from the slot
 731 */
 732static void qfq_front_slot_remove(struct qfq_group *grp)
 733{
 734        struct qfq_class *cl = qfq_slot_head(grp);
 735
 736        BUG_ON(!cl);
 737        hlist_del(&cl->next);
 738        if (hlist_empty(&grp->slots[grp->front]))
 739                __clear_bit(0, &grp->full_slots);
 740}
 741
 742/*
 743 * Returns the first full queue in a group. As a side effect,
 744 * adjust the bucket list so the first non-empty bucket is at
 745 * position 0 in full_slots.
 746 */
 747static struct qfq_class *qfq_slot_scan(struct qfq_group *grp)
 748{
 749        unsigned int i;
 750
 751        pr_debug("qfq slot_scan: grp %u full %#lx\n",
 752                 grp->index, grp->full_slots);
 753
 754        if (grp->full_slots == 0)
 755                return NULL;
 756
 757        i = __ffs(grp->full_slots);  /* zero based */
 758        if (i > 0) {
 759                grp->front = (grp->front + i) % QFQ_MAX_SLOTS;
 760                grp->full_slots >>= i;
 761        }
 762
 763        return qfq_slot_head(grp);
 764}
 765
 766/*
 767 * adjust the bucket list. When the start time of a group decreases,
 768 * we move the index down (modulo QFQ_MAX_SLOTS) so we don't need to
 769 * move the objects. The mask of occupied slots must be shifted
 770 * because we use ffs() to find the first non-empty slot.
 771 * This covers decreases in the group's start time, but what about
 772 * increases of the start time ?
 773 * Here too we should make sure that i is less than 32
 774 */
 775static void qfq_slot_rotate(struct qfq_group *grp, u64 roundedS)
 776{
 777        unsigned int i = (grp->S - roundedS) >> grp->slot_shift;
 778
 779        grp->full_slots <<= i;
 780        grp->front = (grp->front - i) % QFQ_MAX_SLOTS;
 781}
 782
 783static void qfq_update_eligible(struct qfq_sched *q, u64 old_V)
 784{
 785        struct qfq_group *grp;
 786        unsigned long ineligible;
 787
 788        ineligible = q->bitmaps[IR] | q->bitmaps[IB];
 789        if (ineligible) {
 790                if (!q->bitmaps[ER]) {
 791                        grp = qfq_ffs(q, ineligible);
 792                        if (qfq_gt(grp->S, q->V))
 793                                q->V = grp->S;
 794                }
 795                qfq_make_eligible(q, old_V);
 796        }
 797}
 798
 799/*
 800 * Updates the class, returns true if also the group needs to be updated.
 801 */
 802static bool qfq_update_class(struct qfq_group *grp, struct qfq_class *cl)
 803{
 804        unsigned int len = qdisc_peek_len(cl->qdisc);
 805
 806        cl->S = cl->F;
 807        if (!len)
 808                qfq_front_slot_remove(grp);     /* queue is empty */
 809        else {
 810                u64 roundedS;
 811
 812                cl->F = cl->S + (u64)len * cl->inv_w;
 813                roundedS = qfq_round_down(cl->S, grp->slot_shift);
 814                if (roundedS == grp->S)
 815                        return false;
 816
 817                qfq_front_slot_remove(grp);
 818                qfq_slot_insert(grp, cl, roundedS);
 819        }
 820
 821        return true;
 822}
 823
 824static struct sk_buff *qfq_dequeue(struct Qdisc *sch)
 825{
 826        struct qfq_sched *q = qdisc_priv(sch);
 827        struct qfq_group *grp;
 828        struct qfq_class *cl;
 829        struct sk_buff *skb;
 830        unsigned int len;
 831        u64 old_V;
 832
 833        if (!q->bitmaps[ER])
 834                return NULL;
 835
 836        grp = qfq_ffs(q, q->bitmaps[ER]);
 837
 838        cl = qfq_slot_head(grp);
 839        skb = qdisc_dequeue_peeked(cl->qdisc);
 840        if (!skb) {
 841                WARN_ONCE(1, "qfq_dequeue: non-workconserving leaf\n");
 842                return NULL;
 843        }
 844
 845        sch->q.qlen--;
 846        qdisc_bstats_update(sch, skb);
 847
 848        old_V = q->V;
 849        len = qdisc_pkt_len(skb);
 850        q->V += (u64)len * IWSUM;
 851        pr_debug("qfq dequeue: len %u F %lld now %lld\n",
 852                 len, (unsigned long long) cl->F, (unsigned long long) q->V);
 853
 854        if (qfq_update_class(grp, cl)) {
 855                u64 old_F = grp->F;
 856
 857                cl = qfq_slot_scan(grp);
 858                if (!cl)
 859                        __clear_bit(grp->index, &q->bitmaps[ER]);
 860                else {
 861                        u64 roundedS = qfq_round_down(cl->S, grp->slot_shift);
 862                        unsigned int s;
 863
 864                        if (grp->S == roundedS)
 865                                goto skip_unblock;
 866                        grp->S = roundedS;
 867                        grp->F = roundedS + (2ULL << grp->slot_shift);
 868                        __clear_bit(grp->index, &q->bitmaps[ER]);
 869                        s = qfq_calc_state(q, grp);
 870                        __set_bit(grp->index, &q->bitmaps[s]);
 871                }
 872
 873                qfq_unblock_groups(q, grp->index, old_F);
 874        }
 875
 876skip_unblock:
 877        qfq_update_eligible(q, old_V);
 878
 879        return skb;
 880}
 881
 882/*
 883 * Assign a reasonable start time for a new flow k in group i.
 884 * Admissible values for \hat(F) are multiples of \sigma_i
 885 * no greater than V+\sigma_i . Larger values mean that
 886 * we had a wraparound so we consider the timestamp to be stale.
 887 *
 888 * If F is not stale and F >= V then we set S = F.
 889 * Otherwise we should assign S = V, but this may violate
 890 * the ordering in ER. So, if we have groups in ER, set S to
 891 * the F_j of the first group j which would be blocking us.
 892 * We are guaranteed not to move S backward because
 893 * otherwise our group i would still be blocked.
 894 */
 895static void qfq_update_start(struct qfq_sched *q, struct qfq_class *cl)
 896{
 897        unsigned long mask;
 898        u64 limit, roundedF;
 899        int slot_shift = cl->grp->slot_shift;
 900
 901        roundedF = qfq_round_down(cl->F, slot_shift);
 902        limit = qfq_round_down(q->V, slot_shift) + (1ULL << slot_shift);
 903
 904        if (!qfq_gt(cl->F, q->V) || qfq_gt(roundedF, limit)) {
 905                /* timestamp was stale */
 906                mask = mask_from(q->bitmaps[ER], cl->grp->index);
 907                if (mask) {
 908                        struct qfq_group *next = qfq_ffs(q, mask);
 909                        if (qfq_gt(roundedF, next->F)) {
 910                                if (qfq_gt(limit, next->F))
 911                                        cl->S = next->F;
 912                                else /* preserve timestamp correctness */
 913                                        cl->S = limit;
 914                                return;
 915                        }
 916                }
 917                cl->S = q->V;
 918        } else  /* timestamp is not stale */
 919                cl->S = cl->F;
 920}
 921
 922static int qfq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 923{
 924        struct qfq_sched *q = qdisc_priv(sch);
 925        struct qfq_class *cl;
 926        int err = 0;
 927
 928        cl = qfq_classify(skb, sch, &err);
 929        if (cl == NULL) {
 930                if (err & __NET_XMIT_BYPASS)
 931                        sch->qstats.drops++;
 932                kfree_skb(skb);
 933                return err;
 934        }
 935        pr_debug("qfq_enqueue: cl = %x\n", cl->common.classid);
 936
 937        if (unlikely(cl->lmax < qdisc_pkt_len(skb))) {
 938                pr_debug("qfq: increasing maxpkt from %u to %u for class %u",
 939                          cl->lmax, qdisc_pkt_len(skb), cl->common.classid);
 940                qfq_update_reactivate_class(q, cl, cl->inv_w,
 941                                            qdisc_pkt_len(skb), 0);
 942        }
 943
 944        err = qdisc_enqueue(skb, cl->qdisc);
 945        if (unlikely(err != NET_XMIT_SUCCESS)) {
 946                pr_debug("qfq_enqueue: enqueue failed %d\n", err);
 947                if (net_xmit_drop_count(err)) {
 948                        cl->qstats.drops++;
 949                        sch->qstats.drops++;
 950                }
 951                return err;
 952        }
 953
 954        bstats_update(&cl->bstats, skb);
 955        ++sch->q.qlen;
 956
 957        /* If the new skb is not the head of queue, then done here. */
 958        if (cl->qdisc->q.qlen != 1)
 959                return err;
 960
 961        /* If reach this point, queue q was idle */
 962        qfq_activate_class(q, cl, qdisc_pkt_len(skb));
 963
 964        return err;
 965}
 966
 967/*
 968 * Handle class switch from idle to backlogged.
 969 */
 970static void qfq_activate_class(struct qfq_sched *q, struct qfq_class *cl,
 971                               unsigned int pkt_len)
 972{
 973        struct qfq_group *grp = cl->grp;
 974        u64 roundedS;
 975        int s;
 976
 977        qfq_update_start(q, cl);
 978
 979        /* compute new finish time and rounded start. */
 980        cl->F = cl->S + (u64)pkt_len * cl->inv_w;
 981        roundedS = qfq_round_down(cl->S, grp->slot_shift);
 982
 983        /*
 984         * insert cl in the correct bucket.
 985         * If cl->S >= grp->S we don't need to adjust the
 986         * bucket list and simply go to the insertion phase.
 987         * Otherwise grp->S is decreasing, we must make room
 988         * in the bucket list, and also recompute the group state.
 989         * Finally, if there were no flows in this group and nobody
 990         * was in ER make sure to adjust V.
 991         */
 992        if (grp->full_slots) {
 993                if (!qfq_gt(grp->S, cl->S))
 994                        goto skip_update;
 995
 996                /* create a slot for this cl->S */
 997                qfq_slot_rotate(grp, roundedS);
 998                /* group was surely ineligible, remove */
 999                __clear_bit(grp->index, &q->bitmaps[IR]);
1000                __clear_bit(grp->index, &q->bitmaps[IB]);
1001        } else if (!q->bitmaps[ER] && qfq_gt(roundedS, q->V))
1002                q->V = roundedS;
1003
1004        grp->S = roundedS;
1005        grp->F = roundedS + (2ULL << grp->slot_shift);
1006        s = qfq_calc_state(q, grp);
1007        __set_bit(grp->index, &q->bitmaps[s]);
1008
1009        pr_debug("qfq enqueue: new state %d %#lx S %lld F %lld V %lld\n",
1010                 s, q->bitmaps[s],
1011                 (unsigned long long) cl->S,
1012                 (unsigned long long) cl->F,
1013                 (unsigned long long) q->V);
1014
1015skip_update:
1016        qfq_slot_insert(grp, cl, roundedS);
1017}
1018
1019
1020static void qfq_slot_remove(struct qfq_sched *q, struct qfq_group *grp,
1021                            struct qfq_class *cl)
1022{
1023        unsigned int i, offset;
1024        u64 roundedS;
1025
1026        roundedS = qfq_round_down(cl->S, grp->slot_shift);
1027        offset = (roundedS - grp->S) >> grp->slot_shift;
1028        i = (grp->front + offset) % QFQ_MAX_SLOTS;
1029
1030        hlist_del(&cl->next);
1031        if (hlist_empty(&grp->slots[i]))
1032                __clear_bit(offset, &grp->full_slots);
1033}
1034
1035/*
1036 * called to forcibly destroy a queue.
1037 * If the queue is not in the front bucket, or if it has
1038 * other queues in the front bucket, we can simply remove
1039 * the queue with no other side effects.
1040 * Otherwise we must propagate the event up.
1041 */
1042static void qfq_deactivate_class(struct qfq_sched *q, struct qfq_class *cl)
1043{
1044        struct qfq_group *grp = cl->grp;
1045        unsigned long mask;
1046        u64 roundedS;
1047        int s;
1048
1049        cl->F = cl->S;
1050        qfq_slot_remove(q, grp, cl);
1051
1052        if (!grp->full_slots) {
1053                __clear_bit(grp->index, &q->bitmaps[IR]);
1054                __clear_bit(grp->index, &q->bitmaps[EB]);
1055                __clear_bit(grp->index, &q->bitmaps[IB]);
1056
1057                if (test_bit(grp->index, &q->bitmaps[ER]) &&
1058                    !(q->bitmaps[ER] & ~((1UL << grp->index) - 1))) {
1059                        mask = q->bitmaps[ER] & ((1UL << grp->index) - 1);
1060                        if (mask)
1061                                mask = ~((1UL << __fls(mask)) - 1);
1062                        else
1063                                mask = ~0UL;
1064                        qfq_move_groups(q, mask, EB, ER);
1065                        qfq_move_groups(q, mask, IB, IR);
1066                }
1067                __clear_bit(grp->index, &q->bitmaps[ER]);
1068        } else if (hlist_empty(&grp->slots[grp->front])) {
1069                cl = qfq_slot_scan(grp);
1070                roundedS = qfq_round_down(cl->S, grp->slot_shift);
1071                if (grp->S != roundedS) {
1072                        __clear_bit(grp->index, &q->bitmaps[ER]);
1073                        __clear_bit(grp->index, &q->bitmaps[IR]);
1074                        __clear_bit(grp->index, &q->bitmaps[EB]);
1075                        __clear_bit(grp->index, &q->bitmaps[IB]);
1076                        grp->S = roundedS;
1077                        grp->F = roundedS + (2ULL << grp->slot_shift);
1078                        s = qfq_calc_state(q, grp);
1079                        __set_bit(grp->index, &q->bitmaps[s]);
1080                }
1081        }
1082
1083        qfq_update_eligible(q, q->V);
1084}
1085
1086static void qfq_qlen_notify(struct Qdisc *sch, unsigned long arg)
1087{
1088        struct qfq_sched *q = qdisc_priv(sch);
1089        struct qfq_class *cl = (struct qfq_class *)arg;
1090
1091        if (cl->qdisc->q.qlen == 0)
1092                qfq_deactivate_class(q, cl);
1093}
1094
1095static unsigned int qfq_drop(struct Qdisc *sch)
1096{
1097        struct qfq_sched *q = qdisc_priv(sch);
1098        struct qfq_group *grp;
1099        unsigned int i, j, len;
1100
1101        for (i = 0; i <= QFQ_MAX_INDEX; i++) {
1102                grp = &q->groups[i];
1103                for (j = 0; j < QFQ_MAX_SLOTS; j++) {
1104                        struct qfq_class *cl;
1105                        struct hlist_node *n;
1106
1107                        hlist_for_each_entry(cl, n, &grp->slots[j], next) {
1108
1109                                if (!cl->qdisc->ops->drop)
1110                                        continue;
1111
1112                                len = cl->qdisc->ops->drop(cl->qdisc);
1113                                if (len > 0) {
1114                                        sch->q.qlen--;
1115                                        if (!cl->qdisc->q.qlen)
1116                                                qfq_deactivate_class(q, cl);
1117
1118                                        return len;
1119                                }
1120                        }
1121                }
1122        }
1123
1124        return 0;
1125}
1126
1127static int qfq_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
1128{
1129        struct qfq_sched *q = qdisc_priv(sch);
1130        struct qfq_group *grp;
1131        int i, j, err;
1132
1133        err = qdisc_class_hash_init(&q->clhash);
1134        if (err < 0)
1135                return err;
1136
1137        for (i = 0; i <= QFQ_MAX_INDEX; i++) {
1138                grp = &q->groups[i];
1139                grp->index = i;
1140                grp->slot_shift = QFQ_MTU_SHIFT + FRAC_BITS
1141                                   - (QFQ_MAX_INDEX - i);
1142                for (j = 0; j < QFQ_MAX_SLOTS; j++)
1143                        INIT_HLIST_HEAD(&grp->slots[j]);
1144        }
1145
1146        return 0;
1147}
1148
1149static void qfq_reset_qdisc(struct Qdisc *sch)
1150{
1151        struct qfq_sched *q = qdisc_priv(sch);
1152        struct qfq_group *grp;
1153        struct qfq_class *cl;
1154        struct hlist_node *n, *tmp;
1155        unsigned int i, j;
1156
1157        for (i = 0; i <= QFQ_MAX_INDEX; i++) {
1158                grp = &q->groups[i];
1159                for (j = 0; j < QFQ_MAX_SLOTS; j++) {
1160                        hlist_for_each_entry_safe(cl, n, tmp,
1161                                                  &grp->slots[j], next) {
1162                                qfq_deactivate_class(q, cl);
1163                        }
1164                }
1165        }
1166
1167        for (i = 0; i < q->clhash.hashsize; i++) {
1168                hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode)
1169                        qdisc_reset(cl->qdisc);
1170        }
1171        sch->q.qlen = 0;
1172}
1173
1174static void qfq_destroy_qdisc(struct Qdisc *sch)
1175{
1176        struct qfq_sched *q = qdisc_priv(sch);
1177        struct qfq_class *cl;
1178        struct hlist_node *n, *next;
1179        unsigned int i;
1180
1181        tcf_destroy_chain(&q->filter_list);
1182
1183        for (i = 0; i < q->clhash.hashsize; i++) {
1184                hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i],
1185                                          common.hnode) {
1186                        qfq_destroy_class(sch, cl);
1187                }
1188        }
1189        qdisc_class_hash_destroy(&q->clhash);
1190}
1191
1192static const struct Qdisc_class_ops qfq_class_ops = {
1193        .change         = qfq_change_class,
1194        .delete         = qfq_delete_class,
1195        .get            = qfq_get_class,
1196        .put            = qfq_put_class,
1197        .tcf_chain      = qfq_tcf_chain,
1198        .bind_tcf       = qfq_bind_tcf,
1199        .unbind_tcf     = qfq_unbind_tcf,
1200        .graft          = qfq_graft_class,
1201        .leaf           = qfq_class_leaf,
1202        .qlen_notify    = qfq_qlen_notify,
1203        .dump           = qfq_dump_class,
1204        .dump_stats     = qfq_dump_class_stats,
1205        .walk           = qfq_walk,
1206};
1207
1208static struct Qdisc_ops qfq_qdisc_ops __read_mostly = {
1209        .cl_ops         = &qfq_class_ops,
1210        .id             = "qfq",
1211        .priv_size      = sizeof(struct qfq_sched),
1212        .enqueue        = qfq_enqueue,
1213        .dequeue        = qfq_dequeue,
1214        .peek           = qdisc_peek_dequeued,
1215        .drop           = qfq_drop,
1216        .init           = qfq_init_qdisc,
1217        .reset          = qfq_reset_qdisc,
1218        .destroy        = qfq_destroy_qdisc,
1219        .owner          = THIS_MODULE,
1220};
1221
1222static int __init qfq_init(void)
1223{
1224        return register_qdisc(&qfq_qdisc_ops);
1225}
1226
1227static void __exit qfq_exit(void)
1228{
1229        unregister_qdisc(&qfq_qdisc_ops);
1230}
1231
1232module_init(qfq_init);
1233module_exit(qfq_exit);
1234MODULE_LICENSE("GPL");
1235
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.