net/sched/sch_gred.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/sch_gred.c Generic Random Early Detection queue.
   4  *
   5  * Authors:    J Hadi Salim (hadi@cyberus.ca) 1998-2002
   6  *
   7  *             991129: -  Bug fix with grio mode
   8  *                     - a better sing. AvgQ mode with Grio(WRED)
   9  *                     - A finer grained VQ dequeue based on suggestion
  10  *                       from Ren Liu
  11  *                     - More error checks
  12  *
  13  *  For all the glorious comments look at include/net/red.h
  14  */
  15
  16 #include <linux/slab.h>
  17 #include <linux/module.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/skbuff.h>
  21 #include <net/pkt_cls.h>
  22 #include <net/pkt_sched.h>
  23 #include <net/red.h>
  24
  25 #define GRED_DEF_PRIO (MAX_DPs / 2)
  26 #define GRED_VQ_MASK (MAX_DPs - 1)
  27
  28 #define GRED_VQ_RED_FLAGS       (TC_RED_ECN | TC_RED_HARDDROP)
  29
  30 struct gred_sched_data;
  31 struct gred_sched;
  32
  33 struct gred_sched_data {
  34         u32             limit;          /* HARD maximal queue length    */
  35         u32             DP;             /* the drop parameters */
  36         u32             red_flags;      /* virtualQ version of red_flags */
  37         u64             bytesin;        /* bytes seen on virtualQ so far*/
  38         u32             packetsin;      /* packets seen on virtualQ so far*/
  39         u32             backlog;        /* bytes on the virtualQ */
  40         u8              prio;           /* the prio of this vq */
  41
  42         struct red_parms parms;
  43         struct red_vars  vars;
  44         struct red_stats stats;
  45 };
  46
  47 enum {
  48         GRED_WRED_MODE = 1,
  49         GRED_RIO_MODE,
  50 };
  51
  52 struct gred_sched {
  53         struct gred_sched_data *tab[MAX_DPs];
  54         unsigned long   flags;
  55         u32             red_flags;
  56         u32             DPs;
  57         u32             def;
  58         struct red_vars wred_set;
  59         struct tc_gred_qopt_offload *opt;
  60 };
  61
  62 static inline int gred_wred_mode(struct gred_sched *table)
  63 {
  64         return test_bit(GRED_WRED_MODE, &table->flags);
  65 }
  66
  67 static inline void gred_enable_wred_mode(struct gred_sched *table)
  68 {
  69         __set_bit(GRED_WRED_MODE, &table->flags);
  70 }
  71
  72 static inline void gred_disable_wred_mode(struct gred_sched *table)
  73 {
  74         __clear_bit(GRED_WRED_MODE, &table->flags);
  75 }
  76
  77 static inline int gred_rio_mode(struct gred_sched *table)
  78 {
  79         return test_bit(GRED_RIO_MODE, &table->flags);
  80 }
  81
  82 static inline void gred_enable_rio_mode(struct gred_sched *table)
  83 {
  84         __set_bit(GRED_RIO_MODE, &table->flags);
  85 }
  86
  87 static inline void gred_disable_rio_mode(struct gred_sched *table)
  88 {
  89         __clear_bit(GRED_RIO_MODE, &table->flags);
  90 }
  91
  92 static inline int gred_wred_mode_check(struct Qdisc *sch)
  93 {
  94         struct gred_sched *table = qdisc_priv(sch);
  95         int i;
  96
  97         /* Really ugly O(n^2) but shouldn't be necessary too frequent. */
  98         for (i = 0; i < table->DPs; i++) {
  99                 struct gred_sched_data *q = table->tab[i];
 100                 int n;
 101
 102                 if (q == NULL)
 103                         continue;
 104
 105                 for (n = i + 1; n < table->DPs; n++)
 106                         if (table->tab[n] && table->tab[n]->prio == q->prio)
 107                                 return 1;
 108         }
 109
 110         return 0;
 111 }
 112
 113 static inline unsigned int gred_backlog(struct gred_sched *table,
 114                                         struct gred_sched_data *q,
 115                                         struct Qdisc *sch)
 116 {
 117         if (gred_wred_mode(table))
 118                 return sch->qstats.backlog;
 119         else
 120                 return q->backlog;
 121 }
 122
 123 static inline u16 tc_index_to_dp(struct sk_buff *skb)
 124 {
 125         return skb->tc_index & GRED_VQ_MASK;
 126 }
 127
 128 static inline void gred_load_wred_set(const struct gred_sched *table,
 129                                       struct gred_sched_data *q)
 130 {
 131         q->vars.qavg = table->wred_set.qavg;
 132         q->vars.qidlestart = table->wred_set.qidlestart;
 133 }
 134
 135 static inline void gred_store_wred_set(struct gred_sched *table,
 136                                        struct gred_sched_data *q)
 137 {
 138         table->wred_set.qavg = q->vars.qavg;
 139         table->wred_set.qidlestart = q->vars.qidlestart;
 140 }
 141
 142 static int gred_use_ecn(struct gred_sched_data *q)
 143 {
 144         return q->red_flags & TC_RED_ECN;
 145 }
 146
 147 static int gred_use_harddrop(struct gred_sched_data *q)
 148 {
 149         return q->red_flags & TC_RED_HARDDROP;
 150 }
 151
 152 static bool gred_per_vq_red_flags_used(struct gred_sched *table)
 153 {
 154         unsigned int i;
 155
 156         /* Local per-vq flags couldn't have been set unless global are 0 */
 157         if (table->red_flags)
 158                 return false;
 159         for (i = 0; i < MAX_DPs; i++)
 160                 if (table->tab[i] && table->tab[i]->red_flags)
 161                         return true;
 162         return false;
 163 }
 164
 165 static int gred_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 166                         struct sk_buff **to_free)
 167 {
 168         struct gred_sched_data *q = NULL;
 169         struct gred_sched *t = qdisc_priv(sch);
 170         unsigned long qavg = 0;
 171         u16 dp = tc_index_to_dp(skb);
 172
 173         if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
 174                 dp = t->def;
 175
 176                 q = t->tab[dp];
 177                 if (!q) {
 178                         /* Pass through packets not assigned to a DP
 179                          * if no default DP has been configured. This
 180                          * allows for DP flows to be left untouched.
 181                          */
 182                         if (likely(sch->qstats.backlog + qdisc_pkt_len(skb) <=
 183                                         sch->limit))
 184                                 return qdisc_enqueue_tail(skb, sch);
 185                         else
 186                                 goto drop;
 187                 }
 188
 189                 /* fix tc_index? --could be controversial but needed for
 190                    requeueing */
 191                 skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
 192         }
 193
 194         /* sum up all the qaves of prios < ours to get the new qave */
 195         if (!gred_wred_mode(t) && gred_rio_mode(t)) {
 196                 int i;
 197
 198                 for (i = 0; i < t->DPs; i++) {
 199                         if (t->tab[i] && t->tab[i]->prio < q->prio &&
 200                             !red_is_idling(&t->tab[i]->vars))
 201                                 qavg += t->tab[i]->vars.qavg;
 202                 }
 203
 204         }
 205
 206         q->packetsin++;
 207         q->bytesin += qdisc_pkt_len(skb);
 208
 209         if (gred_wred_mode(t))
 210                 gred_load_wred_set(t, q);
 211
 212         q->vars.qavg = red_calc_qavg(&q->parms,
 213                                      &q->vars,
 214                                      gred_backlog(t, q, sch));
 215
 216         if (red_is_idling(&q->vars))
 217                 red_end_of_idle_period(&q->vars);
 218
 219         if (gred_wred_mode(t))
 220                 gred_store_wred_set(t, q);
 221
 222         switch (red_action(&q->parms, &q->vars, q->vars.qavg + qavg)) {
 223         case RED_DONT_MARK:
 224                 break;
 225
 226         case RED_PROB_MARK:
 227                 qdisc_qstats_overlimit(sch);
 228                 if (!gred_use_ecn(q) || !INET_ECN_set_ce(skb)) {
 229                         q->stats.prob_drop++;
 230                         goto congestion_drop;
 231                 }
 232
 233                 q->stats.prob_mark++;
 234                 break;
 235
 236         case RED_HARD_MARK:
 237                 qdisc_qstats_overlimit(sch);
 238                 if (gred_use_harddrop(q) || !gred_use_ecn(q) ||
 239                     !INET_ECN_set_ce(skb)) {
 240                         q->stats.forced_drop++;
 241                         goto congestion_drop;
 242                 }
 243                 q->stats.forced_mark++;
 244                 break;
 245         }
 246
 247         if (gred_backlog(t, q, sch) + qdisc_pkt_len(skb) <= q->limit) {
 248                 q->backlog += qdisc_pkt_len(skb);
 249                 return qdisc_enqueue_tail(skb, sch);
 250         }
 251
 252         q->stats.pdrop++;
 253 drop:
 254         return qdisc_drop(skb, sch, to_free);
 255
 256 congestion_drop:
 257         qdisc_drop(skb, sch, to_free);
 258         return NET_XMIT_CN;
 259 }
 260
 261 static struct sk_buff *gred_dequeue(struct Qdisc *sch)
 262 {
 263         struct sk_buff *skb;
 264         struct gred_sched *t = qdisc_priv(sch);
 265
 266         skb = qdisc_dequeue_head(sch);
 267
 268         if (skb) {
 269                 struct gred_sched_data *q;
 270                 u16 dp = tc_index_to_dp(skb);
 271
 272                 if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
 273                         net_warn_ratelimited("GRED: Unable to relocate VQ 0x%x after dequeue, screwing up backlog\n",
 274                                              tc_index_to_dp(skb));
 275                 } else {
 276                         q->backlog -= qdisc_pkt_len(skb);
 277
 278                         if (gred_wred_mode(t)) {
 279                                 if (!sch->qstats.backlog)
 280                                         red_start_of_idle_period(&t->wred_set);
 281                         } else {
 282                                 if (!q->backlog)
 283                                         red_start_of_idle_period(&q->vars);
 284                         }
 285                 }
 286
 287                 return skb;
 288         }
 289
 290         return NULL;
 291 }
 292
 293 static void gred_reset(struct Qdisc *sch)
 294 {
 295         int i;
 296         struct gred_sched *t = qdisc_priv(sch);
 297
 298         qdisc_reset_queue(sch);
 299
 300         for (i = 0; i < t->DPs; i++) {
 301                 struct gred_sched_data *q = t->tab[i];
 302
 303                 if (!q)
 304                         continue;
 305
 306                 red_restart(&q->vars);
 307                 q->backlog = 0;
 308         }
 309 }
 310
 311 static void gred_offload(struct Qdisc *sch, enum tc_gred_command command)
 312 {
 313         struct gred_sched *table = qdisc_priv(sch);
 314         struct net_device *dev = qdisc_dev(sch);
 315         struct tc_gred_qopt_offload *opt = table->opt;
 316
 317         if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
 318                 return;
 319
 320         memset(opt, 0, sizeof(*opt));
 321         opt->command = command;
 322         opt->handle = sch->handle;
 323         opt->parent = sch->parent;
 324
 325         if (command == TC_GRED_REPLACE) {
 326                 unsigned int i;
 327
 328                 opt->set.grio_on = gred_rio_mode(table);
 329                 opt->set.wred_on = gred_wred_mode(table);
 330                 opt->set.dp_cnt = table->DPs;
 331                 opt->set.dp_def = table->def;
 332
 333                 for (i = 0; i < table->DPs; i++) {
 334                         struct gred_sched_data *q = table->tab[i];
 335
 336                         if (!q)
 337                                 continue;
 338                         opt->set.tab[i].present = true;
 339                         opt->set.tab[i].limit = q->limit;
 340                         opt->set.tab[i].prio = q->prio;
 341                         opt->set.tab[i].min = q->parms.qth_min >> q->parms.Wlog;
 342                         opt->set.tab[i].max = q->parms.qth_max >> q->parms.Wlog;
 343                         opt->set.tab[i].is_ecn = gred_use_ecn(q);
 344                         opt->set.tab[i].is_harddrop = gred_use_harddrop(q);
 345                         opt->set.tab[i].probability = q->parms.max_P;
 346                         opt->set.tab[i].backlog = &q->backlog;
 347                 }
 348                 opt->set.qstats = &sch->qstats;
 349         }
 350
 351         dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_GRED, opt);
 352 }
 353
 354 static int gred_offload_dump_stats(struct Qdisc *sch)
 355 {
 356         struct gred_sched *table = qdisc_priv(sch);
 357         struct tc_gred_qopt_offload *hw_stats;
 358         u64 bytes = 0, packets = 0;
 359         unsigned int i;
 360         int ret;
 361
 362         hw_stats = kzalloc(sizeof(*hw_stats), GFP_KERNEL);
 363         if (!hw_stats)
 364                 return -ENOMEM;
 365
 366         hw_stats->command = TC_GRED_STATS;
 367         hw_stats->handle = sch->handle;
 368         hw_stats->parent = sch->parent;
 369
 370         for (i = 0; i < MAX_DPs; i++) {
 371                 gnet_stats_basic_sync_init(&hw_stats->stats.bstats[i]);
 372                 if (table->tab[i])
 373                         hw_stats->stats.xstats[i] = &table->tab[i]->stats;
 374         }
 375
 376         ret = qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_GRED, hw_stats);
 377         /* Even if driver returns failure adjust the stats - in case offload
 378          * ended but driver still wants to adjust the values.
 379          */
 380         for (i = 0; i < MAX_DPs; i++) {
 381                 if (!table->tab[i])
 382                         continue;
 383                 table->tab[i]->packetsin += u64_stats_read(&hw_stats->stats.bstats[i].packets);
 384                 table->tab[i]->bytesin += u64_stats_read(&hw_stats->stats.bstats[i].bytes);
 385                 table->tab[i]->backlog += hw_stats->stats.qstats[i].backlog;
 386
 387                 bytes += u64_stats_read(&hw_stats->stats.bstats[i].bytes);
 388                 packets += u64_stats_read(&hw_stats->stats.bstats[i].packets);
 389                 sch->qstats.qlen += hw_stats->stats.qstats[i].qlen;
 390                 sch->qstats.backlog += hw_stats->stats.qstats[i].backlog;
 391                 sch->qstats.drops += hw_stats->stats.qstats[i].drops;
 392                 sch->qstats.requeues += hw_stats->stats.qstats[i].requeues;
 393                 sch->qstats.overlimits += hw_stats->stats.qstats[i].overlimits;
 394         }
 395         _bstats_update(&sch->bstats, bytes, packets);
 396
 397         kfree(hw_stats);
 398         return ret;
 399 }
 400
 401 static inline void gred_destroy_vq(struct gred_sched_data *q)
 402 {
 403         kfree(q);
 404 }
 405
 406 static int gred_change_table_def(struct Qdisc *sch, struct nlattr *dps,
 407                                  struct netlink_ext_ack *extack)
 408 {
 409         struct gred_sched *table = qdisc_priv(sch);
 410         struct tc_gred_sopt *sopt;
 411         bool red_flags_changed;
 412         int i;
 413
 414         if (!dps)
 415                 return -EINVAL;
 416
 417         sopt = nla_data(dps);
 418
 419         if (sopt->DPs > MAX_DPs) {
 420                 NL_SET_ERR_MSG_MOD(extack, "number of virtual queues too high");
 421                 return -EINVAL;
 422         }
 423         if (sopt->DPs == 0) {
 424                 NL_SET_ERR_MSG_MOD(extack,
 425                                    "number of virtual queues can't be 0");
 426                 return -EINVAL;
 427         }
 428         if (sopt->def_DP >= sopt->DPs) {
 429                 NL_SET_ERR_MSG_MOD(extack, "default virtual queue above virtual queue count");
 430                 return -EINVAL;
 431         }
 432         if (sopt->flags && gred_per_vq_red_flags_used(table)) {
 433                 NL_SET_ERR_MSG_MOD(extack, "can't set per-Qdisc RED flags when per-virtual queue flags are used");
 434                 return -EINVAL;
 435         }
 436
 437         sch_tree_lock(sch);
 438         table->DPs = sopt->DPs;
 439         table->def = sopt->def_DP;
 440         red_flags_changed = table->red_flags != sopt->flags;
 441         table->red_flags = sopt->flags;
 442
 443         /*
 444          * Every entry point to GRED is synchronized with the above code
 445          * and the DP is checked against DPs, i.e. shadowed VQs can no
 446          * longer be found so we can unlock right here.
 447          */
 448         sch_tree_unlock(sch);
 449
 450         if (sopt->grio) {
 451                 gred_enable_rio_mode(table);
 452                 gred_disable_wred_mode(table);
 453                 if (gred_wred_mode_check(sch))
 454                         gred_enable_wred_mode(table);
 455         } else {
 456                 gred_disable_rio_mode(table);
 457                 gred_disable_wred_mode(table);
 458         }
 459
 460         if (red_flags_changed)
 461                 for (i = 0; i < table->DPs; i++)
 462                         if (table->tab[i])
 463                                 table->tab[i]->red_flags =
 464                                         table->red_flags & GRED_VQ_RED_FLAGS;
 465
 466         for (i = table->DPs; i < MAX_DPs; i++) {
 467                 if (table->tab[i]) {
 468                         pr_warn("GRED: Warning: Destroying shadowed VQ 0x%x\n",
 469                                 i);
 470                         gred_destroy_vq(table->tab[i]);
 471                         table->tab[i] = NULL;
 472                 }
 473         }
 474
 475         gred_offload(sch, TC_GRED_REPLACE);
 476         return 0;
 477 }
 478
 479 static inline int gred_change_vq(struct Qdisc *sch, int dp,
 480                                  struct tc_gred_qopt *ctl, int prio,
 481                                  u8 *stab, u32 max_P,
 482                                  struct gred_sched_data **prealloc,
 483                                  struct netlink_ext_ack *extack)
 484 {
 485         struct gred_sched *table = qdisc_priv(sch);
 486         struct gred_sched_data *q = table->tab[dp];
 487
 488         if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Scell_log, stab)) {
 489                 NL_SET_ERR_MSG_MOD(extack, "invalid RED parameters");
 490                 return -EINVAL;
 491         }
 492
 493         if (!q) {
 494                 table->tab[dp] = q = *prealloc;
 495                 *prealloc = NULL;
 496                 if (!q)
 497                         return -ENOMEM;
 498                 q->red_flags = table->red_flags & GRED_VQ_RED_FLAGS;
 499         }
 500
 501         q->DP = dp;
 502         q->prio = prio;
 503         if (ctl->limit > sch->limit)
 504                 q->limit = sch->limit;
 505         else
 506                 q->limit = ctl->limit;
 507
 508         if (q->backlog == 0)
 509                 red_end_of_idle_period(&q->vars);
 510
 511         red_set_parms(&q->parms,
 512                       ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
 513                       ctl->Scell_log, stab, max_P);
 514         red_set_vars(&q->vars);
 515         return 0;
 516 }
 517
 518 static const struct nla_policy gred_vq_policy[TCA_GRED_VQ_MAX + 1] = {
 519         [TCA_GRED_VQ_DP]        = { .type = NLA_U32 },
 520         [TCA_GRED_VQ_FLAGS]     = { .type = NLA_U32 },
 521 };
 522
 523 static const struct nla_policy gred_vqe_policy[TCA_GRED_VQ_ENTRY_MAX + 1] = {
 524         [TCA_GRED_VQ_ENTRY]     = { .type = NLA_NESTED },
 525 };
 526
 527 static const struct nla_policy gred_policy[TCA_GRED_MAX + 1] = {
 528         [TCA_GRED_PARMS]        = { .len = sizeof(struct tc_gred_qopt) },
 529         [TCA_GRED_STAB]         = { .len = 256 },
 530         [TCA_GRED_DPS]          = { .len = sizeof(struct tc_gred_sopt) },
 531         [TCA_GRED_MAX_P]        = { .type = NLA_U32 },
 532         [TCA_GRED_LIMIT]        = { .type = NLA_U32 },
 533         [TCA_GRED_VQ_LIST]      = { .type = NLA_NESTED },
 534 };
 535
 536 static void gred_vq_apply(struct gred_sched *table, const struct nlattr *entry)
 537 {
 538         struct nlattr *tb[TCA_GRED_VQ_MAX + 1];
 539         u32 dp;
 540
 541         nla_parse_nested_deprecated(tb, TCA_GRED_VQ_MAX, entry,
 542                                     gred_vq_policy, NULL);
 543
 544         dp = nla_get_u32(tb[TCA_GRED_VQ_DP]);
 545
 546         if (tb[TCA_GRED_VQ_FLAGS])
 547                 table->tab[dp]->red_flags = nla_get_u32(tb[TCA_GRED_VQ_FLAGS]);
 548 }
 549
 550 static void gred_vqs_apply(struct gred_sched *table, struct nlattr *vqs)
 551 {
 552         const struct nlattr *attr;
 553         int rem;
 554
 555         nla_for_each_nested(attr, vqs, rem) {
 556                 switch (nla_type(attr)) {
 557                 case TCA_GRED_VQ_ENTRY:
 558                         gred_vq_apply(table, attr);
 559                         break;
 560                 }
 561         }
 562 }
 563
 564 static int gred_vq_validate(struct gred_sched *table, u32 cdp,
 565                             const struct nlattr *entry,
 566                             struct netlink_ext_ack *extack)
 567 {
 568         struct nlattr *tb[TCA_GRED_VQ_MAX + 1];
 569         int err;
 570         u32 dp;
 571
 572         err = nla_parse_nested_deprecated(tb, TCA_GRED_VQ_MAX, entry,
 573                                           gred_vq_policy, extack);
 574         if (err < 0)
 575                 return err;
 576
 577         if (!tb[TCA_GRED_VQ_DP]) {
 578                 NL_SET_ERR_MSG_MOD(extack, "Virtual queue with no index specified");
 579                 return -EINVAL;
 580         }
 581         dp = nla_get_u32(tb[TCA_GRED_VQ_DP]);
 582         if (dp >= table->DPs) {
 583                 NL_SET_ERR_MSG_MOD(extack, "Virtual queue with index out of bounds");
 584                 return -EINVAL;
 585         }
 586         if (dp != cdp && !table->tab[dp]) {
 587                 NL_SET_ERR_MSG_MOD(extack, "Virtual queue not yet instantiated");
 588                 return -EINVAL;
 589         }
 590
 591         if (tb[TCA_GRED_VQ_FLAGS]) {
 592                 u32 red_flags = nla_get_u32(tb[TCA_GRED_VQ_FLAGS]);
 593
 594                 if (table->red_flags && table->red_flags != red_flags) {
 595                         NL_SET_ERR_MSG_MOD(extack, "can't change per-virtual queue RED flags when per-Qdisc flags are used");
 596                         return -EINVAL;
 597                 }
 598                 if (red_flags & ~GRED_VQ_RED_FLAGS) {
 599                         NL_SET_ERR_MSG_MOD(extack,
 600                                            "invalid RED flags specified");
 601                         return -EINVAL;
 602                 }
 603         }
 604
 605         return 0;
 606 }
 607
 608 static int gred_vqs_validate(struct gred_sched *table, u32 cdp,
 609                              struct nlattr *vqs, struct netlink_ext_ack *extack)
 610 {
 611         const struct nlattr *attr;
 612         int rem, err;
 613
 614         err = nla_validate_nested_deprecated(vqs, TCA_GRED_VQ_ENTRY_MAX,
 615                                              gred_vqe_policy, extack);
 616         if (err < 0)
 617                 return err;
 618
 619         nla_for_each_nested(attr, vqs, rem) {
 620                 switch (nla_type(attr)) {
 621                 case TCA_GRED_VQ_ENTRY:
 622                         err = gred_vq_validate(table, cdp, attr, extack);
 623                         if (err)
 624                                 return err;
 625                         break;
 626                 default:
 627                         NL_SET_ERR_MSG_MOD(extack, "GRED_VQ_LIST can contain only entry attributes");
 628                         return -EINVAL;
 629                 }
 630         }
 631
 632         if (rem > 0) {
 633                 NL_SET_ERR_MSG_MOD(extack, "Trailing data after parsing virtual queue list");
 634                 return -EINVAL;
 635         }
 636
 637         return 0;
 638 }
 639
 640 static int gred_change(struct Qdisc *sch, struct nlattr *opt,
 641                        struct netlink_ext_ack *extack)
 642 {
 643         struct gred_sched *table = qdisc_priv(sch);
 644         struct tc_gred_qopt *ctl;
 645         struct nlattr *tb[TCA_GRED_MAX + 1];
 646         int err, prio = GRED_DEF_PRIO;
 647         u8 *stab;
 648         u32 max_P;
 649         struct gred_sched_data *prealloc;
 650
 651         err = nla_parse_nested_deprecated(tb, TCA_GRED_MAX, opt, gred_policy,
 652                                           extack);
 653         if (err < 0)
 654                 return err;
 655
 656         if (tb[TCA_GRED_PARMS] == NULL && tb[TCA_GRED_STAB] == NULL) {
 657                 if (tb[TCA_GRED_LIMIT] != NULL)
 658                         sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
 659                 return gred_change_table_def(sch, tb[TCA_GRED_DPS], extack);
 660         }
 661
 662         if (tb[TCA_GRED_PARMS] == NULL ||
 663             tb[TCA_GRED_STAB] == NULL ||
 664             tb[TCA_GRED_LIMIT] != NULL) {
 665                 NL_SET_ERR_MSG_MOD(extack, "can't configure Qdisc and virtual queue at the same time");
 666                 return -EINVAL;
 667         }
 668
 669         max_P = tb[TCA_GRED_MAX_P] ? nla_get_u32(tb[TCA_GRED_MAX_P]) : 0;
 670
 671         ctl = nla_data(tb[TCA_GRED_PARMS]);
 672         stab = nla_data(tb[TCA_GRED_STAB]);
 673
 674         if (ctl->DP >= table->DPs) {
 675                 NL_SET_ERR_MSG_MOD(extack, "virtual queue index above virtual queue count");
 676                 return -EINVAL;
 677         }
 678
 679         if (tb[TCA_GRED_VQ_LIST]) {
 680                 err = gred_vqs_validate(table, ctl->DP, tb[TCA_GRED_VQ_LIST],
 681                                         extack);
 682                 if (err)
 683                         return err;
 684         }
 685
 686         if (gred_rio_mode(table)) {
 687                 if (ctl->prio == 0) {
 688                         int def_prio = GRED_DEF_PRIO;
 689
 690                         if (table->tab[table->def])
 691                                 def_prio = table->tab[table->def]->prio;
 692
 693                         printk(KERN_DEBUG "GRED: DP %u does not have a prio "
 694                                "setting default to %d\n", ctl->DP, def_prio);
 695
 696                         prio = def_prio;
 697                 } else
 698                         prio = ctl->prio;
 699         }
 700
 701         prealloc = kzalloc(sizeof(*prealloc), GFP_KERNEL);
 702         sch_tree_lock(sch);
 703
 704         err = gred_change_vq(sch, ctl->DP, ctl, prio, stab, max_P, &prealloc,
 705                              extack);
 706         if (err < 0)
 707                 goto err_unlock_free;
 708
 709         if (tb[TCA_GRED_VQ_LIST])
 710                 gred_vqs_apply(table, tb[TCA_GRED_VQ_LIST]);
 711
 712         if (gred_rio_mode(table)) {
 713                 gred_disable_wred_mode(table);
 714                 if (gred_wred_mode_check(sch))
 715                         gred_enable_wred_mode(table);
 716         }
 717
 718         sch_tree_unlock(sch);
 719         kfree(prealloc);
 720
 721         gred_offload(sch, TC_GRED_REPLACE);
 722         return 0;
 723
 724 err_unlock_free:
 725         sch_tree_unlock(sch);
 726         kfree(prealloc);
 727         return err;
 728 }
 729
 730 static int gred_init(struct Qdisc *sch, struct nlattr *opt,
 731                      struct netlink_ext_ack *extack)
 732 {
 733         struct gred_sched *table = qdisc_priv(sch);
 734         struct nlattr *tb[TCA_GRED_MAX + 1];
 735         int err;
 736
 737         if (!opt)
 738                 return -EINVAL;
 739
 740         err = nla_parse_nested_deprecated(tb, TCA_GRED_MAX, opt, gred_policy,
 741                                           extack);
 742         if (err < 0)
 743                 return err;
 744
 745         if (tb[TCA_GRED_PARMS] || tb[TCA_GRED_STAB]) {
 746                 NL_SET_ERR_MSG_MOD(extack,
 747                                    "virtual queue configuration can't be specified at initialization time");
 748                 return -EINVAL;
 749         }
 750
 751         if (tb[TCA_GRED_LIMIT])
 752                 sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
 753         else
 754                 sch->limit = qdisc_dev(sch)->tx_queue_len
 755                              * psched_mtu(qdisc_dev(sch));
 756
 757         if (qdisc_dev(sch)->netdev_ops->ndo_setup_tc) {
 758                 table->opt = kzalloc(sizeof(*table->opt), GFP_KERNEL);
 759                 if (!table->opt)
 760                         return -ENOMEM;
 761         }
 762
 763         return gred_change_table_def(sch, tb[TCA_GRED_DPS], extack);
 764 }
 765
 766 static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
 767 {
 768         struct gred_sched *table = qdisc_priv(sch);
 769         struct nlattr *parms, *vqs, *opts = NULL;
 770         int i;
 771         u32 max_p[MAX_DPs];
 772         struct tc_gred_sopt sopt = {
 773                 .DPs    = table->DPs,
 774                 .def_DP = table->def,
 775                 .grio   = gred_rio_mode(table),
 776                 .flags  = table->red_flags,
 777         };
 778
 779         if (gred_offload_dump_stats(sch))
 780                 goto nla_put_failure;
 781
 782         opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
 783         if (opts == NULL)
 784                 goto nla_put_failure;
 785         if (nla_put(skb, TCA_GRED_DPS, sizeof(sopt), &sopt))
 786                 goto nla_put_failure;
 787
 788         for (i = 0; i < MAX_DPs; i++) {
 789                 struct gred_sched_data *q = table->tab[i];
 790
 791                 max_p[i] = q ? q->parms.max_P : 0;
 792         }
 793         if (nla_put(skb, TCA_GRED_MAX_P, sizeof(max_p), max_p))
 794                 goto nla_put_failure;
 795
 796         if (nla_put_u32(skb, TCA_GRED_LIMIT, sch->limit))
 797                 goto nla_put_failure;
 798
 799         /* Old style all-in-one dump of VQs */
 800         parms = nla_nest_start_noflag(skb, TCA_GRED_PARMS);
 801         if (parms == NULL)
 802                 goto nla_put_failure;
 803
 804         for (i = 0; i < MAX_DPs; i++) {
 805                 struct gred_sched_data *q = table->tab[i];
 806                 struct tc_gred_qopt opt;
 807                 unsigned long qavg;
 808
 809                 memset(&opt, 0, sizeof(opt));
 810
 811                 if (!q) {
 812                         /* hack -- fix at some point with proper message
 813                            This is how we indicate to tc that there is no VQ
 814                            at this DP */
 815
 816                         opt.DP = MAX_DPs + i;
 817                         goto append_opt;
 818                 }
 819
 820                 opt.limit       = q->limit;
 821                 opt.DP          = q->DP;
 822                 opt.backlog     = gred_backlog(table, q, sch);
 823                 opt.prio        = q->prio;
 824                 opt.qth_min     = q->parms.qth_min >> q->parms.Wlog;
 825                 opt.qth_max     = q->parms.qth_max >> q->parms.Wlog;
 826                 opt.Wlog        = q->parms.Wlog;
 827                 opt.Plog        = q->parms.Plog;
 828                 opt.Scell_log   = q->parms.Scell_log;
 829                 opt.early       = q->stats.prob_drop;
 830                 opt.forced      = q->stats.forced_drop;
 831                 opt.pdrop       = q->stats.pdrop;
 832                 opt.packets     = q->packetsin;
 833                 opt.bytesin     = q->bytesin;
 834
 835                 if (gred_wred_mode(table))
 836                         gred_load_wred_set(table, q);
 837
 838                 qavg = red_calc_qavg(&q->parms, &q->vars,
 839                                      q->vars.qavg >> q->parms.Wlog);
 840                 opt.qave = qavg >> q->parms.Wlog;
 841
 842 append_opt:
 843                 if (nla_append(skb, sizeof(opt), &opt) < 0)
 844                         goto nla_put_failure;
 845         }
 846
 847         nla_nest_end(skb, parms);
 848
 849         /* Dump the VQs again, in more structured way */
 850         vqs = nla_nest_start_noflag(skb, TCA_GRED_VQ_LIST);
 851         if (!vqs)
 852                 goto nla_put_failure;
 853
 854         for (i = 0; i < MAX_DPs; i++) {
 855                 struct gred_sched_data *q = table->tab[i];
 856                 struct nlattr *vq;
 857
 858                 if (!q)
 859                         continue;
 860
 861                 vq = nla_nest_start_noflag(skb, TCA_GRED_VQ_ENTRY);
 862                 if (!vq)
 863                         goto nla_put_failure;
 864
 865                 if (nla_put_u32(skb, TCA_GRED_VQ_DP, q->DP))
 866                         goto nla_put_failure;
 867
 868                 if (nla_put_u32(skb, TCA_GRED_VQ_FLAGS, q->red_flags))
 869                         goto nla_put_failure;
 870
 871                 /* Stats */
 872                 if (nla_put_u64_64bit(skb, TCA_GRED_VQ_STAT_BYTES, q->bytesin,
 873                                       TCA_GRED_VQ_PAD))
 874                         goto nla_put_failure;
 875                 if (nla_put_u32(skb, TCA_GRED_VQ_STAT_PACKETS, q->packetsin))
 876                         goto nla_put_failure;
 877                 if (nla_put_u32(skb, TCA_GRED_VQ_STAT_BACKLOG,
 878                                 gred_backlog(table, q, sch)))
 879                         goto nla_put_failure;
 880                 if (nla_put_u32(skb, TCA_GRED_VQ_STAT_PROB_DROP,
 881                                 q->stats.prob_drop))
 882                         goto nla_put_failure;
 883                 if (nla_put_u32(skb, TCA_GRED_VQ_STAT_PROB_MARK,
 884                                 q->stats.prob_mark))
 885                         goto nla_put_failure;
 886                 if (nla_put_u32(skb, TCA_GRED_VQ_STAT_FORCED_DROP,
 887                                 q->stats.forced_drop))
 888                         goto nla_put_failure;
 889                 if (nla_put_u32(skb, TCA_GRED_VQ_STAT_FORCED_MARK,
 890                                 q->stats.forced_mark))
 891                         goto nla_put_failure;
 892                 if (nla_put_u32(skb, TCA_GRED_VQ_STAT_PDROP, q->stats.pdrop))
 893                         goto nla_put_failure;
 894
 895                 nla_nest_end(skb, vq);
 896         }
 897         nla_nest_end(skb, vqs);
 898
 899         return nla_nest_end(skb, opts);
 900
 901 nla_put_failure:
 902         nla_nest_cancel(skb, opts);
 903         return -EMSGSIZE;
 904 }
 905
 906 static void gred_destroy(struct Qdisc *sch)
 907 {
 908         struct gred_sched *table = qdisc_priv(sch);
 909         int i;
 910
 911         for (i = 0; i < table->DPs; i++)
 912                 gred_destroy_vq(table->tab[i]);
 913
 914         gred_offload(sch, TC_GRED_DESTROY);
 915         kfree(table->opt);
 916 }
 917
 918 static struct Qdisc_ops gred_qdisc_ops __read_mostly = {
 919         .id             =       "gred",
 920         .priv_size      =       sizeof(struct gred_sched),
 921         .enqueue        =       gred_enqueue,
 922         .dequeue        =       gred_dequeue,
 923         .peek           =       qdisc_peek_head,
 924         .init           =       gred_init,
 925         .reset          =       gred_reset,
 926         .destroy        =       gred_destroy,
 927         .change         =       gred_change,
 928         .dump           =       gred_dump,
 929         .owner          =       THIS_MODULE,
 930 };
 931
 932 static int __init gred_module_init(void)
 933 {
 934         return register_qdisc(&gred_qdisc_ops);
 935 }
 936
 937 static void __exit gred_module_exit(void)
 938 {
 939         unregister_qdisc(&gred_qdisc_ops);
 940 }
 941
 942 module_init(gred_module_init)
 943 module_exit(gred_module_exit)
 944
 945 MODULE_LICENSE("GPL");