kernel/net/sched/sch_fq.c

   1 /*
   2  * net/sched/sch_fq.c Fair Queue Packet Scheduler (per flow pacing)
   3  *
   4  *  Copyright (C) 2013-2015 Eric Dumazet <edumazet@google.com>
   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  *      as published by the Free Software Foundation; either version
   9  *      2 of the License, or (at your option) any later version.
  10  *
  11  *  Meant to be mostly used for locally generated traffic :
  12  *  Fast classification depends on skb->sk being set before reaching us.
  13  *  If not, (router workload), we use rxhash as fallback, with 32 bits wide hash.
  14  *  All packets belonging to a socket are considered as a 'flow'.
  15  *
  16  *  Flows are dynamically allocated and stored in a hash table of RB trees
  17  *  They are also part of one Round Robin 'queues' (new or old flows)
  18  *
  19  *  Burst avoidance (aka pacing) capability :
  20  *
  21  *  Transport (eg TCP) can set in sk->sk_pacing_rate a rate, enqueue a
  22  *  bunch of packets, and this packet scheduler adds delay between
  23  *  packets to respect rate limitation.
  24  *
  25  *  enqueue() :
  26  *   - lookup one RB tree (out of 1024 or more) to find the flow.
  27  *     If non existent flow, create it, add it to the tree.
  28  *     Add skb to the per flow list of skb (fifo).
  29  *   - Use a special fifo for high prio packets
  30  *
  31  *  dequeue() : serves flows in Round Robin
  32  *  Note : When a flow becomes empty, we do not immediately remove it from
  33  *  rb trees, for performance reasons (its expected to send additional packets,
  34  *  or SLAB cache will reuse socket for another flow)
  35  */
  36
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/kernel.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/string.h>
  42 #include <linux/in.h>
  43 #include <linux/errno.h>
  44 #include <linux/init.h>
  45 #include <linux/skbuff.h>
  46 #include <linux/slab.h>
  47 #include <linux/rbtree.h>
  48 #include <linux/hash.h>
  49 #include <linux/prefetch.h>
  50 #include <linux/vmalloc.h>
  51 #include <net/netlink.h>
  52 #include <net/pkt_sched.h>
  53 #include <net/sock.h>
  54 #include <net/tcp_states.h>
  55 #include <net/tcp.h>
  56
  57 /*
  58  * Per flow structure, dynamically allocated
  59  */
  60 struct fq_flow {
  61         struct sk_buff  *head;          /* list of skbs for this flow : first skb */
  62         union {
  63                 struct sk_buff *tail;   /* last skb in the list */
  64                 unsigned long  age;     /* jiffies when flow was emptied, for gc */
  65         };
  66         struct rb_node  fq_node;        /* anchor in fq_root[] trees */
  67         struct sock     *sk;
  68         int             qlen;           /* number of packets in flow queue */
  69         int             credit;
  70         u32             socket_hash;    /* sk_hash */
  71         struct fq_flow *next;           /* next pointer in RR lists, or &detached */
  72
  73         struct rb_node  rate_node;      /* anchor in q->delayed tree */
  74         u64             time_next_packet;
  75 };
  76
  77 struct fq_flow_head {
  78         struct fq_flow *first;
  79         struct fq_flow *last;
  80 };
  81
  82 struct fq_sched_data {
  83         struct fq_flow_head new_flows;
  84
  85         struct fq_flow_head old_flows;
  86
  87         struct rb_root  delayed;        /* for rate limited flows */
  88         u64             time_next_delayed_flow;
  89
  90         struct fq_flow  internal;       /* for non classified or high prio packets */
  91         u32             quantum;
  92         u32             initial_quantum;
  93         u32             flow_refill_delay;
  94         u32             flow_max_rate;  /* optional max rate per flow */
  95         u32             flow_plimit;    /* max packets per flow */
  96         u32             orphan_mask;    /* mask for orphaned skb */
  97         struct rb_root  *fq_root;
  98         u8              rate_enable;
  99         u8              fq_trees_log;
 100
 101         u32             flows;
 102         u32             inactive_flows;
 103         u32             throttled_flows;
 104
 105         u64             stat_gc_flows;
 106         u64             stat_internal_packets;
 107         u64             stat_tcp_retrans;
 108         u64             stat_throttled;
 109         u64             stat_flows_plimit;
 110         u64             stat_pkts_too_long;
 111         u64             stat_allocation_errors;
 112         struct qdisc_watchdog watchdog;
 113 };
 114
 115 /* special value to mark a detached flow (not on old/new list) */
 116 static struct fq_flow detached, throttled;
 117
 118 static void fq_flow_set_detached(struct fq_flow *f)
 119 {
 120         f->next = &detached;
 121         f->age = jiffies;
 122 }
 123
 124 static bool fq_flow_is_detached(const struct fq_flow *f)
 125 {
 126         return f->next == &detached;
 127 }
 128
 129 static void fq_flow_set_throttled(struct fq_sched_data *q, struct fq_flow *f)
 130 {
 131         struct rb_node **p = &q->delayed.rb_node, *parent = NULL;
 132
 133         while (*p) {
 134                 struct fq_flow *aux;
 135
 136                 parent = *p;
 137                 aux = container_of(parent, struct fq_flow, rate_node);
 138                 if (f->time_next_packet >= aux->time_next_packet)
 139                         p = &parent->rb_right;
 140                 else
 141                         p = &parent->rb_left;
 142         }
 143         rb_link_node(&f->rate_node, parent, p);
 144         rb_insert_color(&f->rate_node, &q->delayed);
 145         q->throttled_flows++;
 146         q->stat_throttled++;
 147
 148         f->next = &throttled;
 149         if (q->time_next_delayed_flow > f->time_next_packet)
 150                 q->time_next_delayed_flow = f->time_next_packet;
 151 }
 152
 153
 154 static struct kmem_cache *fq_flow_cachep __read_mostly;
 155
 156 static void fq_flow_add_tail(struct fq_flow_head *head, struct fq_flow *flow)
 157 {
 158         if (head->first)
 159                 head->last->next = flow;
 160         else
 161                 head->first = flow;
 162         head->last = flow;
 163         flow->next = NULL;
 164 }
 165
 166 /* limit number of collected flows per round */
 167 #define FQ_GC_MAX 8
 168 #define FQ_GC_AGE (3*HZ)
 169
 170 static bool fq_gc_candidate(const struct fq_flow *f)
 171 {
 172         return fq_flow_is_detached(f) &&
 173                time_after(jiffies, f->age + FQ_GC_AGE);
 174 }
 175
 176 static void fq_gc(struct fq_sched_data *q,
 177                   struct rb_root *root,
 178                   struct sock *sk)
 179 {
 180         struct fq_flow *f, *tofree[FQ_GC_MAX];
 181         struct rb_node **p, *parent;
 182         int fcnt = 0;
 183
 184         p = &root->rb_node;
 185         parent = NULL;
 186         while (*p) {
 187                 parent = *p;
 188
 189                 f = container_of(parent, struct fq_flow, fq_node);
 190                 if (f->sk == sk)
 191                         break;
 192
 193                 if (fq_gc_candidate(f)) {
 194                         tofree[fcnt++] = f;
 195                         if (fcnt == FQ_GC_MAX)
 196                                 break;
 197                 }
 198
 199                 if (f->sk > sk)
 200                         p = &parent->rb_right;
 201                 else
 202                         p = &parent->rb_left;
 203         }
 204
 205         q->flows -= fcnt;
 206         q->inactive_flows -= fcnt;
 207         q->stat_gc_flows += fcnt;
 208         while (fcnt) {
 209                 struct fq_flow *f = tofree[--fcnt];
 210
 211                 rb_erase(&f->fq_node, root);
 212                 kmem_cache_free(fq_flow_cachep, f);
 213         }
 214 }
 215
 216 static struct fq_flow *fq_classify(struct sk_buff *skb, struct fq_sched_data *q)
 217 {
 218         struct rb_node **p, *parent;
 219         struct sock *sk = skb->sk;
 220         struct rb_root *root;
 221         struct fq_flow *f;
 222
 223         /* warning: no starvation prevention... */
 224         if (unlikely((skb->priority & TC_PRIO_MAX) == TC_PRIO_CONTROL))
 225                 return &q->internal;
 226
 227         /* SYNACK messages are attached to a TCP_NEW_SYN_RECV request socket
 228          * or a listener (SYNCOOKIE mode)
 229          * 1) request sockets are not full blown,
 230          *    they do not contain sk_pacing_rate
 231          * 2) They are not part of a 'flow' yet
 232          * 3) We do not want to rate limit them (eg SYNFLOOD attack),
 233          *    especially if the listener set SO_MAX_PACING_RATE
 234          * 4) We pretend they are orphaned
 235          */
 236         if (!sk || sk_listener(sk)) {
 237                 unsigned long hash = skb_get_hash(skb) & q->orphan_mask;
 238
 239                 /* By forcing low order bit to 1, we make sure to not
 240                  * collide with a local flow (socket pointers are word aligned)
 241                  */
 242                 sk = (struct sock *)((hash << 1) | 1UL);
 243                 skb_orphan(skb);
 244         }
 245
 246         root = &q->fq_root[hash_32((u32)(long)sk, q->fq_trees_log)];
 247
 248         if (q->flows >= (2U << q->fq_trees_log) &&
 249             q->inactive_flows > q->flows/2)
 250                 fq_gc(q, root, sk);
 251
 252         p = &root->rb_node;
 253         parent = NULL;
 254         while (*p) {
 255                 parent = *p;
 256
 257                 f = container_of(parent, struct fq_flow, fq_node);
 258                 if (f->sk == sk) {
 259                         /* socket might have been reallocated, so check
 260                          * if its sk_hash is the same.
 261                          * It not, we need to refill credit with
 262                          * initial quantum
 263                          */
 264                         if (unlikely(skb->sk &&
 265                                      f->socket_hash != sk->sk_hash)) {
 266                                 f->credit = q->initial_quantum;
 267                                 f->socket_hash = sk->sk_hash;
 268                                 f->time_next_packet = 0ULL;
 269                         }
 270                         return f;
 271                 }
 272                 if (f->sk > sk)
 273                         p = &parent->rb_right;
 274                 else
 275                         p = &parent->rb_left;
 276         }
 277
 278         f = kmem_cache_zalloc(fq_flow_cachep, GFP_ATOMIC | __GFP_NOWARN);
 279         if (unlikely(!f)) {
 280                 q->stat_allocation_errors++;
 281                 return &q->internal;
 282         }
 283         fq_flow_set_detached(f);
 284         f->sk = sk;
 285         if (skb->sk)
 286                 f->socket_hash = sk->sk_hash;
 287         f->credit = q->initial_quantum;
 288
 289         rb_link_node(&f->fq_node, parent, p);
 290         rb_insert_color(&f->fq_node, root);
 291
 292         q->flows++;
 293         q->inactive_flows++;
 294         return f;
 295 }
 296
 297
 298 /* remove one skb from head of flow queue */
 299 static struct sk_buff *fq_dequeue_head(struct Qdisc *sch, struct fq_flow *flow)
 300 {
 301         struct sk_buff *skb = flow->head;
 302
 303         if (skb) {
 304                 flow->head = skb->next;
 305                 skb->next = NULL;
 306                 flow->qlen--;
 307                 qdisc_qstats_backlog_dec(sch, skb);
 308                 sch->q.qlen--;
 309         }
 310         return skb;
 311 }
 312
 313 /* We might add in the future detection of retransmits
 314  * For the time being, just return false
 315  */
 316 static bool skb_is_retransmit(struct sk_buff *skb)
 317 {
 318         return false;
 319 }
 320
 321 /* add skb to flow queue
 322  * flow queue is a linked list, kind of FIFO, except for TCP retransmits
 323  * We special case tcp retransmits to be transmitted before other packets.
 324  * We rely on fact that TCP retransmits are unlikely, so we do not waste
 325  * a separate queue or a pointer.
 326  * head->  [retrans pkt 1]
 327  *         [retrans pkt 2]
 328  *         [ normal pkt 1]
 329  *         [ normal pkt 2]
 330  *         [ normal pkt 3]
 331  * tail->  [ normal pkt 4]
 332  */
 333 static void flow_queue_add(struct fq_flow *flow, struct sk_buff *skb)
 334 {
 335         struct sk_buff *prev, *head = flow->head;
 336
 337         skb->next = NULL;
 338         if (!head) {
 339                 flow->head = skb;
 340                 flow->tail = skb;
 341                 return;
 342         }
 343         if (likely(!skb_is_retransmit(skb))) {
 344                 flow->tail->next = skb;
 345                 flow->tail = skb;
 346                 return;
 347         }
 348
 349         /* This skb is a tcp retransmit,
 350          * find the last retrans packet in the queue
 351          */
 352         prev = NULL;
 353         while (skb_is_retransmit(head)) {
 354                 prev = head;
 355                 head = head->next;
 356                 if (!head)
 357                         break;
 358         }
 359         if (!prev) { /* no rtx packet in queue, become the new head */
 360                 skb->next = flow->head;
 361                 flow->head = skb;
 362         } else {
 363                 if (prev == flow->tail)
 364                         flow->tail = skb;
 365                 else
 366                         skb->next = prev->next;
 367                 prev->next = skb;
 368         }
 369 }
 370
 371 static int fq_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 372 {
 373         struct fq_sched_data *q = qdisc_priv(sch);
 374         struct fq_flow *f;
 375
 376         if (unlikely(sch->q.qlen >= sch->limit))
 377                 return qdisc_drop(skb, sch);
 378
 379         f = fq_classify(skb, q);
 380         if (unlikely(f->qlen >= q->flow_plimit && f != &q->internal)) {
 381                 q->stat_flows_plimit++;
 382                 return qdisc_drop(skb, sch);
 383         }
 384
 385         f->qlen++;
 386         if (skb_is_retransmit(skb))
 387                 q->stat_tcp_retrans++;
 388         qdisc_qstats_backlog_inc(sch, skb);
 389         if (fq_flow_is_detached(f)) {
 390                 fq_flow_add_tail(&q->new_flows, f);
 391                 if (time_after(jiffies, f->age + q->flow_refill_delay))
 392                         f->credit = max_t(u32, f->credit, q->quantum);
 393                 q->inactive_flows--;
 394         }
 395
 396         /* Note: this overwrites f->age */
 397         flow_queue_add(f, skb);
 398
 399         if (unlikely(f == &q->internal)) {
 400                 q->stat_internal_packets++;
 401         }
 402         sch->q.qlen++;
 403
 404         return NET_XMIT_SUCCESS;
 405 }
 406
 407 static void fq_check_throttled(struct fq_sched_data *q, u64 now)
 408 {
 409         struct rb_node *p;
 410
 411         if (q->time_next_delayed_flow > now)
 412                 return;
 413
 414         q->time_next_delayed_flow = ~0ULL;
 415         while ((p = rb_first(&q->delayed)) != NULL) {
 416                 struct fq_flow *f = container_of(p, struct fq_flow, rate_node);
 417
 418                 if (f->time_next_packet > now) {
 419                         q->time_next_delayed_flow = f->time_next_packet;
 420                         break;
 421                 }
 422                 rb_erase(p, &q->delayed);
 423                 q->throttled_flows--;
 424                 fq_flow_add_tail(&q->old_flows, f);
 425         }
 426 }
 427
 428 static struct sk_buff *fq_dequeue(struct Qdisc *sch)
 429 {
 430         struct fq_sched_data *q = qdisc_priv(sch);
 431         u64 now = ktime_get_ns();
 432         struct fq_flow_head *head;
 433         struct sk_buff *skb;
 434         struct fq_flow *f;
 435         u32 rate;
 436
 437         skb = fq_dequeue_head(sch, &q->internal);
 438         if (skb)
 439                 goto out;
 440         fq_check_throttled(q, now);
 441 begin:
 442         head = &q->new_flows;
 443         if (!head->first) {
 444                 head = &q->old_flows;
 445                 if (!head->first) {
 446                         if (q->time_next_delayed_flow != ~0ULL)
 447                                 qdisc_watchdog_schedule_ns(&q->watchdog,
 448                                                            q->time_next_delayed_flow,
 449                                                            false);
 450                         return NULL;
 451                 }
 452         }
 453         f = head->first;
 454
 455         if (f->credit <= 0) {
 456                 f->credit += q->quantum;
 457                 head->first = f->next;
 458                 fq_flow_add_tail(&q->old_flows, f);
 459                 goto begin;
 460         }
 461
 462         skb = f->head;
 463         if (unlikely(skb && now < f->time_next_packet &&
 464                      !skb_is_tcp_pure_ack(skb))) {
 465                 head->first = f->next;
 466                 fq_flow_set_throttled(q, f);
 467                 goto begin;
 468         }
 469
 470         skb = fq_dequeue_head(sch, f);
 471         if (!skb) {
 472                 head->first = f->next;
 473                 /* force a pass through old_flows to prevent starvation */
 474                 if ((head == &q->new_flows) && q->old_flows.first) {
 475                         fq_flow_add_tail(&q->old_flows, f);
 476                 } else {
 477                         fq_flow_set_detached(f);
 478                         q->inactive_flows++;
 479                 }
 480                 goto begin;
 481         }
 482         prefetch(&skb->end);
 483         f->credit -= qdisc_pkt_len(skb);
 484
 485         if (f->credit > 0 || !q->rate_enable)
 486                 goto out;
 487
 488         /* Do not pace locally generated ack packets */
 489         if (skb_is_tcp_pure_ack(skb))
 490                 goto out;
 491
 492         rate = q->flow_max_rate;
 493         if (skb->sk)
 494                 rate = min(skb->sk->sk_pacing_rate, rate);
 495
 496         if (rate != ~0U) {
 497                 u32 plen = max(qdisc_pkt_len(skb), q->quantum);
 498                 u64 len = (u64)plen * NSEC_PER_SEC;
 499
 500                 if (likely(rate))
 501                         do_div(len, rate);
 502                 /* Since socket rate can change later,
 503                  * clamp the delay to 1 second.
 504                  * Really, providers of too big packets should be fixed !
 505                  */
 506                 if (unlikely(len > NSEC_PER_SEC)) {
 507                         len = NSEC_PER_SEC;
 508                         q->stat_pkts_too_long++;
 509                 }
 510
 511                 f->time_next_packet = now + len;
 512         }
 513 out:
 514         qdisc_bstats_update(sch, skb);
 515         return skb;
 516 }
 517
 518 static void fq_reset(struct Qdisc *sch)
 519 {
 520         struct fq_sched_data *q = qdisc_priv(sch);
 521         struct rb_root *root;
 522         struct sk_buff *skb;
 523         struct rb_node *p;
 524         struct fq_flow *f;
 525         unsigned int idx;
 526
 527         while ((skb = fq_dequeue_head(sch, &q->internal)) != NULL)
 528                 kfree_skb(skb);
 529
 530         if (!q->fq_root)
 531                 return;
 532
 533         for (idx = 0; idx < (1U << q->fq_trees_log); idx++) {
 534                 root = &q->fq_root[idx];
 535                 while ((p = rb_first(root)) != NULL) {
 536                         f = container_of(p, struct fq_flow, fq_node);
 537                         rb_erase(p, root);
 538
 539                         while ((skb = fq_dequeue_head(sch, f)) != NULL)
 540                                 kfree_skb(skb);
 541
 542                         kmem_cache_free(fq_flow_cachep, f);
 543                 }
 544         }
 545         q->new_flows.first      = NULL;
 546         q->old_flows.first      = NULL;
 547         q->delayed              = RB_ROOT;
 548         q->flows                = 0;
 549         q->inactive_flows       = 0;
 550         q->throttled_flows      = 0;
 551 }
 552
 553 static void fq_rehash(struct fq_sched_data *q,
 554                       struct rb_root *old_array, u32 old_log,
 555                       struct rb_root *new_array, u32 new_log)
 556 {
 557         struct rb_node *op, **np, *parent;
 558         struct rb_root *oroot, *nroot;
 559         struct fq_flow *of, *nf;
 560         int fcnt = 0;
 561         u32 idx;
 562
 563         for (idx = 0; idx < (1U << old_log); idx++) {
 564                 oroot = &old_array[idx];
 565                 while ((op = rb_first(oroot)) != NULL) {
 566                         rb_erase(op, oroot);
 567                         of = container_of(op, struct fq_flow, fq_node);
 568                         if (fq_gc_candidate(of)) {
 569                                 fcnt++;
 570                                 kmem_cache_free(fq_flow_cachep, of);
 571                                 continue;
 572                         }
 573                         nroot = &new_array[hash_32((u32)(long)of->sk, new_log)];
 574
 575                         np = &nroot->rb_node;
 576                         parent = NULL;
 577                         while (*np) {
 578                                 parent = *np;
 579
 580                                 nf = container_of(parent, struct fq_flow, fq_node);
 581                                 BUG_ON(nf->sk == of->sk);
 582
 583                                 if (nf->sk > of->sk)
 584                                         np = &parent->rb_right;
 585                                 else
 586                                         np = &parent->rb_left;
 587                         }
 588
 589                         rb_link_node(&of->fq_node, parent, np);
 590                         rb_insert_color(&of->fq_node, nroot);
 591                 }
 592         }
 593         q->flows -= fcnt;
 594         q->inactive_flows -= fcnt;
 595         q->stat_gc_flows += fcnt;
 596 }
 597
 598 static void *fq_alloc_node(size_t sz, int node)
 599 {
 600         void *ptr;
 601
 602         ptr = kmalloc_node(sz, GFP_KERNEL | __GFP_REPEAT | __GFP_NOWARN, node);
 603         if (!ptr)
 604                 ptr = vmalloc_node(sz, node);
 605         return ptr;
 606 }
 607
 608 static void fq_free(void *addr)
 609 {
 610         kvfree(addr);
 611 }
 612
 613 static int fq_resize(struct Qdisc *sch, u32 log)
 614 {
 615         struct fq_sched_data *q = qdisc_priv(sch);
 616         struct rb_root *array;
 617         void *old_fq_root;
 618         u32 idx;
 619
 620         if (q->fq_root && log == q->fq_trees_log)
 621                 return 0;
 622
 623         /* If XPS was setup, we can allocate memory on right NUMA node */
 624         array = fq_alloc_node(sizeof(struct rb_root) << log,
 625                               netdev_queue_numa_node_read(sch->dev_queue));
 626         if (!array)
 627                 return -ENOMEM;
 628
 629         for (idx = 0; idx < (1U << log); idx++)
 630                 array[idx] = RB_ROOT;
 631
 632         sch_tree_lock(sch);
 633
 634         old_fq_root = q->fq_root;
 635         if (old_fq_root)
 636                 fq_rehash(q, old_fq_root, q->fq_trees_log, array, log);
 637
 638         q->fq_root = array;
 639         q->fq_trees_log = log;
 640
 641         sch_tree_unlock(sch);
 642
 643         fq_free(old_fq_root);
 644
 645         return 0;
 646 }
 647
 648 static const struct nla_policy fq_policy[TCA_FQ_MAX + 1] = {
 649         [TCA_FQ_PLIMIT]                 = { .type = NLA_U32 },
 650         [TCA_FQ_FLOW_PLIMIT]            = { .type = NLA_U32 },
 651         [TCA_FQ_QUANTUM]                = { .type = NLA_U32 },
 652         [TCA_FQ_INITIAL_QUANTUM]        = { .type = NLA_U32 },
 653         [TCA_FQ_RATE_ENABLE]            = { .type = NLA_U32 },
 654         [TCA_FQ_FLOW_DEFAULT_RATE]      = { .type = NLA_U32 },
 655         [TCA_FQ_FLOW_MAX_RATE]          = { .type = NLA_U32 },
 656         [TCA_FQ_BUCKETS_LOG]            = { .type = NLA_U32 },
 657         [TCA_FQ_FLOW_REFILL_DELAY]      = { .type = NLA_U32 },
 658 };
 659
 660 static int fq_change(struct Qdisc *sch, struct nlattr *opt)
 661 {
 662         struct fq_sched_data *q = qdisc_priv(sch);
 663         struct nlattr *tb[TCA_FQ_MAX + 1];
 664         int err, drop_count = 0;
 665         u32 fq_log;
 666
 667         if (!opt)
 668                 return -EINVAL;
 669
 670         err = nla_parse_nested(tb, TCA_FQ_MAX, opt, fq_policy);
 671         if (err < 0)
 672                 return err;
 673
 674         sch_tree_lock(sch);
 675
 676         fq_log = q->fq_trees_log;
 677
 678         if (tb[TCA_FQ_BUCKETS_LOG]) {
 679                 u32 nval = nla_get_u32(tb[TCA_FQ_BUCKETS_LOG]);
 680
 681                 if (nval >= 1 && nval <= ilog2(256*1024))
 682                         fq_log = nval;
 683                 else
 684                         err = -EINVAL;
 685         }
 686         if (tb[TCA_FQ_PLIMIT])
 687                 sch->limit = nla_get_u32(tb[TCA_FQ_PLIMIT]);
 688
 689         if (tb[TCA_FQ_FLOW_PLIMIT])
 690                 q->flow_plimit = nla_get_u32(tb[TCA_FQ_FLOW_PLIMIT]);
 691
 692         if (tb[TCA_FQ_QUANTUM]) {
 693                 u32 quantum = nla_get_u32(tb[TCA_FQ_QUANTUM]);
 694
 695                 if (quantum > 0)
 696                         q->quantum = quantum;
 697                 else
 698                         err = -EINVAL;
 699         }
 700
 701         if (tb[TCA_FQ_INITIAL_QUANTUM])
 702                 q->initial_quantum = nla_get_u32(tb[TCA_FQ_INITIAL_QUANTUM]);
 703
 704         if (tb[TCA_FQ_FLOW_DEFAULT_RATE])
 705                 pr_warn_ratelimited("sch_fq: defrate %u ignored.\n",
 706                                     nla_get_u32(tb[TCA_FQ_FLOW_DEFAULT_RATE]));
 707
 708         if (tb[TCA_FQ_FLOW_MAX_RATE])
 709                 q->flow_max_rate = nla_get_u32(tb[TCA_FQ_FLOW_MAX_RATE]);
 710
 711         if (tb[TCA_FQ_RATE_ENABLE]) {
 712                 u32 enable = nla_get_u32(tb[TCA_FQ_RATE_ENABLE]);
 713
 714                 if (enable <= 1)
 715                         q->rate_enable = enable;
 716                 else
 717                         err = -EINVAL;
 718         }
 719
 720         if (tb[TCA_FQ_FLOW_REFILL_DELAY]) {
 721                 u32 usecs_delay = nla_get_u32(tb[TCA_FQ_FLOW_REFILL_DELAY]) ;
 722
 723                 q->flow_refill_delay = usecs_to_jiffies(usecs_delay);
 724         }
 725
 726         if (tb[TCA_FQ_ORPHAN_MASK])
 727                 q->orphan_mask = nla_get_u32(tb[TCA_FQ_ORPHAN_MASK]);
 728
 729         if (!err) {
 730                 sch_tree_unlock(sch);
 731                 err = fq_resize(sch, fq_log);
 732                 sch_tree_lock(sch);
 733         }
 734         while (sch->q.qlen > sch->limit) {
 735                 struct sk_buff *skb = fq_dequeue(sch);
 736
 737                 if (!skb)
 738                         break;
 739                 kfree_skb(skb);
 740                 drop_count++;
 741         }
 742         qdisc_tree_decrease_qlen(sch, drop_count);
 743
 744         sch_tree_unlock(sch);
 745         return err;
 746 }
 747
 748 static void fq_destroy(struct Qdisc *sch)
 749 {
 750         struct fq_sched_data *q = qdisc_priv(sch);
 751
 752         fq_reset(sch);
 753         fq_free(q->fq_root);
 754         qdisc_watchdog_cancel(&q->watchdog);
 755 }
 756
 757 static int fq_init(struct Qdisc *sch, struct nlattr *opt)
 758 {
 759         struct fq_sched_data *q = qdisc_priv(sch);
 760         int err;
 761
 762         sch->limit              = 10000;
 763         q->flow_plimit          = 100;
 764         q->quantum              = 2 * psched_mtu(qdisc_dev(sch));
 765         q->initial_quantum      = 10 * psched_mtu(qdisc_dev(sch));
 766         q->flow_refill_delay    = msecs_to_jiffies(40);
 767         q->flow_max_rate        = ~0U;
 768         q->rate_enable          = 1;
 769         q->new_flows.first      = NULL;
 770         q->old_flows.first      = NULL;
 771         q->delayed              = RB_ROOT;
 772         q->fq_root              = NULL;
 773         q->fq_trees_log         = ilog2(1024);
 774         q->orphan_mask          = 1024 - 1;
 775         qdisc_watchdog_init(&q->watchdog, sch);
 776
 777         if (opt)
 778                 err = fq_change(sch, opt);
 779         else
 780                 err = fq_resize(sch, q->fq_trees_log);
 781
 782         return err;
 783 }
 784
 785 static int fq_dump(struct Qdisc *sch, struct sk_buff *skb)
 786 {
 787         struct fq_sched_data *q = qdisc_priv(sch);
 788         struct nlattr *opts;
 789
 790         opts = nla_nest_start(skb, TCA_OPTIONS);
 791         if (opts == NULL)
 792                 goto nla_put_failure;
 793
 794         /* TCA_FQ_FLOW_DEFAULT_RATE is not used anymore */
 795
 796         if (nla_put_u32(skb, TCA_FQ_PLIMIT, sch->limit) ||
 797             nla_put_u32(skb, TCA_FQ_FLOW_PLIMIT, q->flow_plimit) ||
 798             nla_put_u32(skb, TCA_FQ_QUANTUM, q->quantum) ||
 799             nla_put_u32(skb, TCA_FQ_INITIAL_QUANTUM, q->initial_quantum) ||
 800             nla_put_u32(skb, TCA_FQ_RATE_ENABLE, q->rate_enable) ||
 801             nla_put_u32(skb, TCA_FQ_FLOW_MAX_RATE, q->flow_max_rate) ||
 802             nla_put_u32(skb, TCA_FQ_FLOW_REFILL_DELAY,
 803                         jiffies_to_usecs(q->flow_refill_delay)) ||
 804             nla_put_u32(skb, TCA_FQ_ORPHAN_MASK, q->orphan_mask) ||
 805             nla_put_u32(skb, TCA_FQ_BUCKETS_LOG, q->fq_trees_log))
 806                 goto nla_put_failure;
 807
 808         return nla_nest_end(skb, opts);
 809
 810 nla_put_failure:
 811         return -1;
 812 }
 813
 814 static int fq_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
 815 {
 816         struct fq_sched_data *q = qdisc_priv(sch);
 817         u64 now = ktime_get_ns();
 818         struct tc_fq_qd_stats st = {
 819                 .gc_flows               = q->stat_gc_flows,
 820                 .highprio_packets       = q->stat_internal_packets,
 821                 .tcp_retrans            = q->stat_tcp_retrans,
 822                 .throttled              = q->stat_throttled,
 823                 .flows_plimit           = q->stat_flows_plimit,
 824                 .pkts_too_long          = q->stat_pkts_too_long,
 825                 .allocation_errors      = q->stat_allocation_errors,
 826                 .flows                  = q->flows,
 827                 .inactive_flows         = q->inactive_flows,
 828                 .throttled_flows        = q->throttled_flows,
 829                 .time_next_delayed_flow = q->time_next_delayed_flow - now,
 830         };
 831
 832         return gnet_stats_copy_app(d, &st, sizeof(st));
 833 }
 834
 835 static struct Qdisc_ops fq_qdisc_ops __read_mostly = {
 836         .id             =       "fq",
 837         .priv_size      =       sizeof(struct fq_sched_data),
 838
 839         .enqueue        =       fq_enqueue,
 840         .dequeue        =       fq_dequeue,
 841         .peek           =       qdisc_peek_dequeued,
 842         .init           =       fq_init,
 843         .reset          =       fq_reset,
 844         .destroy        =       fq_destroy,
 845         .change         =       fq_change,
 846         .dump           =       fq_dump,
 847         .dump_stats     =       fq_dump_stats,
 848         .owner          =       THIS_MODULE,
 849 };
 850
 851 static int __init fq_module_init(void)
 852 {
 853         int ret;
 854
 855         fq_flow_cachep = kmem_cache_create("fq_flow_cache",
 856                                            sizeof(struct fq_flow),
 857                                            0, 0, NULL);
 858         if (!fq_flow_cachep)
 859                 return -ENOMEM;
 860
 861         ret = register_qdisc(&fq_qdisc_ops);
 862         if (ret)
 863                 kmem_cache_destroy(fq_flow_cachep);
 864         return ret;
 865 }
 866
 867 static void __exit fq_module_exit(void)
 868 {
 869         unregister_qdisc(&fq_qdisc_ops);
 870         kmem_cache_destroy(fq_flow_cachep);
 871 }
 872
 873 module_init(fq_module_init)
 874 module_exit(fq_module_exit)
 875 MODULE_AUTHOR("Eric Dumazet");
 876 MODULE_LICENSE("GPL");