1 /* 2 * net/sched/sch_sfq.c Stochastic Fairness Queueing discipline. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 */ 11 12 #include <linux/module.h> 13 #include <linux/types.h> 14 #include <linux/kernel.h> 15 #include <linux/jiffies.h> 16 #include <linux/string.h> 17 #include <linux/in.h> 18 #include <linux/errno.h> 19 #include <linux/init.h> 20 #include <linux/skbuff.h> 21 #include <linux/jhash.h> 22 #include <linux/slab.h> 23 #include <linux/vmalloc.h> 24 #include <net/netlink.h> 25 #include <net/pkt_sched.h> 26 #include <net/pkt_cls.h> 27 #include <net/red.h> 28 29 30 /* Stochastic Fairness Queuing algorithm. 31 ======================================= 32 33 Source: 34 Paul E. McKenney "Stochastic Fairness Queuing", 35 IEEE INFOCOMM'90 Proceedings, San Francisco, 1990. 36 37 Paul E. McKenney "Stochastic Fairness Queuing", 38 "Interworking: Research and Experience", v.2, 1991, p.113-131. 39 40 41 See also: 42 M. Shreedhar and George Varghese "Efficient Fair 43 Queuing using Deficit Round Robin", Proc. SIGCOMM 95. 44 45 46 This is not the thing that is usually called (W)FQ nowadays. 47 It does not use any timestamp mechanism, but instead 48 processes queues in round-robin order. 49 50 ADVANTAGE: 51 52 - It is very cheap. Both CPU and memory requirements are minimal. 53 54 DRAWBACKS: 55 56 - "Stochastic" -> It is not 100% fair. 57 When hash collisions occur, several flows are considered as one. 58 59 - "Round-robin" -> It introduces larger delays than virtual clock 60 based schemes, and should not be used for isolating interactive 61 traffic from non-interactive. It means, that this scheduler 62 should be used as leaf of CBQ or P3, which put interactive traffic 63 to higher priority band. 64 65 We still need true WFQ for top level CSZ, but using WFQ 66 for the best effort traffic is absolutely pointless: 67 SFQ is superior for this purpose. 68 69 IMPLEMENTATION: 70 This implementation limits : 71 - maximal queue length per flow to 127 packets. 72 - max mtu to 2^18-1; 73 - max 65408 flows, 74 - number of hash buckets to 65536. 75 76 It is easy to increase these values, but not in flight. */ 77 78 #define SFQ_MAX_DEPTH 127 /* max number of packets per flow */ 79 #define SFQ_DEFAULT_FLOWS 128 80 #define SFQ_MAX_FLOWS (0x10000 - SFQ_MAX_DEPTH - 1) /* max number of flows */ 81 #define SFQ_EMPTY_SLOT 0xffff 82 #define SFQ_DEFAULT_HASH_DIVISOR 1024 83 84 /* We use 16 bits to store allot, and want to handle packets up to 64K 85 * Scale allot by 8 (1<<3) so that no overflow occurs. 86 */ 87 #define SFQ_ALLOT_SHIFT 3 88 #define SFQ_ALLOT_SIZE(X) DIV_ROUND_UP(X, 1 << SFQ_ALLOT_SHIFT) 89 90 /* This type should contain at least SFQ_MAX_DEPTH + 1 + SFQ_MAX_FLOWS values */ 91 typedef u16 sfq_index; 92 93 /* 94 * We dont use pointers to save space. 95 * Small indexes [0 ... SFQ_MAX_FLOWS - 1] are 'pointers' to slots[] array 96 * while following values [SFQ_MAX_FLOWS ... SFQ_MAX_FLOWS + SFQ_MAX_DEPTH] 97 * are 'pointers' to dep[] array 98 */ 99 struct sfq_head { 100 sfq_index next; 101 sfq_index prev; 102 }; 103 104 struct sfq_slot { 105 struct sk_buff *skblist_next; 106 struct sk_buff *skblist_prev; 107 sfq_index qlen; /* number of skbs in skblist */ 108 sfq_index next; /* next slot in sfq RR chain */ 109 struct sfq_head dep; /* anchor in dep[] chains */ 110 unsigned short hash; /* hash value (index in ht[]) */ 111 short allot; /* credit for this slot */ 112 113 unsigned int backlog; 114 struct red_vars vars; 115 }; 116 117 struct sfq_sched_data { 118 /* frequently used fields */ 119 int limit; /* limit of total number of packets in this qdisc */ 120 unsigned int divisor; /* number of slots in hash table */ 121 u8 headdrop; 122 u8 maxdepth; /* limit of packets per flow */ 123 124 u32 perturbation; 125 u8 cur_depth; /* depth of longest slot */ 126 u8 flags; 127 unsigned short scaled_quantum; /* SFQ_ALLOT_SIZE(quantum) */ 128 struct tcf_proto __rcu *filter_list; 129 struct tcf_block *block; 130 sfq_index *ht; /* Hash table ('divisor' slots) */ 131 struct sfq_slot *slots; /* Flows table ('maxflows' entries) */ 132 133 struct red_parms *red_parms; 134 struct tc_sfqred_stats stats; 135 struct sfq_slot *tail; /* current slot in round */ 136 137 struct sfq_head dep[SFQ_MAX_DEPTH + 1]; 138 /* Linked lists of slots, indexed by depth 139 * dep[0] : list of unused flows 140 * dep[1] : list of flows with 1 packet 141 * dep[X] : list of flows with X packets 142 */ 143 144 unsigned int maxflows; /* number of flows in flows array */ 145 int perturb_period; 146 unsigned int quantum; /* Allotment per round: MUST BE >= MTU */ 147 struct timer_list perturb_timer; 148 }; 149 150 /* 151 * sfq_head are either in a sfq_slot or in dep[] array 152 */ 153 static inline struct sfq_head *sfq_dep_head(struct sfq_sched_data *q, sfq_index val) 154 { 155 if (val < SFQ_MAX_FLOWS) 156 return &q->slots[val].dep; 157 return &q->dep[val - SFQ_MAX_FLOWS]; 158 } 159 160 static unsigned int sfq_hash(const struct sfq_sched_data *q, 161 const struct sk_buff *skb) 162 { 163 return skb_get_hash_perturb(skb, q->perturbation) & (q->divisor - 1); 164 } 165 166 static unsigned int sfq_classify(struct sk_buff *skb, struct Qdisc *sch, 167 int *qerr) 168 { 169 struct sfq_sched_data *q = qdisc_priv(sch); 170 struct tcf_result res; 171 struct tcf_proto *fl; 172 int result; 173 174 if (TC_H_MAJ(skb->priority) == sch->handle && 175 TC_H_MIN(skb->priority) > 0 && 176 TC_H_MIN(skb->priority) <= q->divisor) 177 return TC_H_MIN(skb->priority); 178 179 fl = rcu_dereference_bh(q->filter_list); 180 if (!fl) 181 return sfq_hash(q, skb) + 1; 182 183 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 184 result = tcf_classify(skb, fl, &res, false); 185 if (result >= 0) { 186 #ifdef CONFIG_NET_CLS_ACT 187 switch (result) { 188 case TC_ACT_STOLEN: 189 case TC_ACT_QUEUED: 190 case TC_ACT_TRAP: 191 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 192 case TC_ACT_SHOT: 193 return 0; 194 } 195 #endif 196 if (TC_H_MIN(res.classid) <= q->divisor) 197 return TC_H_MIN(res.classid); 198 } 199 return 0; 200 } 201 202 /* 203 * x : slot number [0 .. SFQ_MAX_FLOWS - 1] 204 */ 205 static inline void sfq_link(struct sfq_sched_data *q, sfq_index x) 206 { 207 sfq_index p, n; 208 struct sfq_slot *slot = &q->slots[x]; 209 int qlen = slot->qlen; 210 211 p = qlen + SFQ_MAX_FLOWS; 212 n = q->dep[qlen].next; 213 214 slot->dep.next = n; 215 slot->dep.prev = p; 216 217 q->dep[qlen].next = x; /* sfq_dep_head(q, p)->next = x */ 218 sfq_dep_head(q, n)->prev = x; 219 } 220 221 #define sfq_unlink(q, x, n, p) \ 222 do { \ 223 n = q->slots[x].dep.next; \ 224 p = q->slots[x].dep.prev; \ 225 sfq_dep_head(q, p)->next = n; \ 226 sfq_dep_head(q, n)->prev = p; \ 227 } while (0) 228 229 230 static inline void sfq_dec(struct sfq_sched_data *q, sfq_index x) 231 { 232 sfq_index p, n; 233 int d; 234 235 sfq_unlink(q, x, n, p); 236 237 d = q->slots[x].qlen--; 238 if (n == p && q->cur_depth == d) 239 q->cur_depth--; 240 sfq_link(q, x); 241 } 242 243 static inline void sfq_inc(struct sfq_sched_data *q, sfq_index x) 244 { 245 sfq_index p, n; 246 int d; 247 248 sfq_unlink(q, x, n, p); 249 250 d = ++q->slots[x].qlen; 251 if (q->cur_depth < d) 252 q->cur_depth = d; 253 sfq_link(q, x); 254 } 255 256 /* helper functions : might be changed when/if skb use a standard list_head */ 257 258 /* remove one skb from tail of slot queue */ 259 static inline struct sk_buff *slot_dequeue_tail(struct sfq_slot *slot) 260 { 261 struct sk_buff *skb = slot->skblist_prev; 262 263 slot->skblist_prev = skb->prev; 264 skb->prev->next = (struct sk_buff *)slot; 265 skb->next = skb->prev = NULL; 266 return skb; 267 } 268 269 /* remove one skb from head of slot queue */ 270 static inline struct sk_buff *slot_dequeue_head(struct sfq_slot *slot) 271 { 272 struct sk_buff *skb = slot->skblist_next; 273 274 slot->skblist_next = skb->next; 275 skb->next->prev = (struct sk_buff *)slot; 276 skb->next = skb->prev = NULL; 277 return skb; 278 } 279 280 static inline void slot_queue_init(struct sfq_slot *slot) 281 { 282 memset(slot, 0, sizeof(*slot)); 283 slot->skblist_prev = slot->skblist_next = (struct sk_buff *)slot; 284 } 285 286 /* add skb to slot queue (tail add) */ 287 static inline void slot_queue_add(struct sfq_slot *slot, struct sk_buff *skb) 288 { 289 skb->prev = slot->skblist_prev; 290 skb->next = (struct sk_buff *)slot; 291 slot->skblist_prev->next = skb; 292 slot->skblist_prev = skb; 293 } 294 295 static unsigned int sfq_drop(struct Qdisc *sch) 296 { 297 struct sfq_sched_data *q = qdisc_priv(sch); 298 sfq_index x, d = q->cur_depth; 299 struct sk_buff *skb; 300 unsigned int len; 301 struct sfq_slot *slot; 302 303 /* Queue is full! Find the longest slot and drop tail packet from it */ 304 if (d > 1) { 305 x = q->dep[d].next; 306 slot = &q->slots[x]; 307 drop: 308 skb = q->headdrop ? slot_dequeue_head(slot) : slot_dequeue_tail(slot); 309 len = qdisc_pkt_len(skb); 310 slot->backlog -= len; 311 sfq_dec(q, x); 312 sch->q.qlen--; 313 qdisc_qstats_drop(sch); 314 qdisc_qstats_backlog_dec(sch, skb); 315 kfree_skb(skb); 316 return len; 317 } 318 319 if (d == 1) { 320 /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */ 321 x = q->tail->next; 322 slot = &q->slots[x]; 323 q->tail->next = slot->next; 324 q->ht[slot->hash] = SFQ_EMPTY_SLOT; 325 goto drop; 326 } 327 328 return 0; 329 } 330 331 /* Is ECN parameter configured */ 332 static int sfq_prob_mark(const struct sfq_sched_data *q) 333 { 334 return q->flags & TC_RED_ECN; 335 } 336 337 /* Should packets over max threshold just be marked */ 338 static int sfq_hard_mark(const struct sfq_sched_data *q) 339 { 340 return (q->flags & (TC_RED_ECN | TC_RED_HARDDROP)) == TC_RED_ECN; 341 } 342 343 static int sfq_headdrop(const struct sfq_sched_data *q) 344 { 345 return q->headdrop; 346 } 347 348 static int 349 sfq_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free) 350 { 351 struct sfq_sched_data *q = qdisc_priv(sch); 352 unsigned int hash, dropped; 353 sfq_index x, qlen; 354 struct sfq_slot *slot; 355 int uninitialized_var(ret); 356 struct sk_buff *head; 357 int delta; 358 359 hash = sfq_classify(skb, sch, &ret); 360 if (hash == 0) { 361 if (ret & __NET_XMIT_BYPASS) 362 qdisc_qstats_drop(sch); 363 kfree_skb(skb); 364 return ret; 365 } 366 hash--; 367 368 x = q->ht[hash]; 369 slot = &q->slots[x]; 370 if (x == SFQ_EMPTY_SLOT) { 371 x = q->dep[0].next; /* get a free slot */ 372 if (x >= SFQ_MAX_FLOWS) 373 return qdisc_drop(skb, sch, to_free); 374 q->ht[hash] = x; 375 slot = &q->slots[x]; 376 slot->hash = hash; 377 slot->backlog = 0; /* should already be 0 anyway... */ 378 red_set_vars(&slot->vars); 379 goto enqueue; 380 } 381 if (q->red_parms) { 382 slot->vars.qavg = red_calc_qavg_no_idle_time(q->red_parms, 383 &slot->vars, 384 slot->backlog); 385 switch (red_action(q->red_parms, 386 &slot->vars, 387 slot->vars.qavg)) { 388 case RED_DONT_MARK: 389 break; 390 391 case RED_PROB_MARK: 392 qdisc_qstats_overlimit(sch); 393 if (sfq_prob_mark(q)) { 394 /* We know we have at least one packet in queue */ 395 if (sfq_headdrop(q) && 396 INET_ECN_set_ce(slot->skblist_next)) { 397 q->stats.prob_mark_head++; 398 break; 399 } 400 if (INET_ECN_set_ce(skb)) { 401 q->stats.prob_mark++; 402 break; 403 } 404 } 405 q->stats.prob_drop++; 406 goto congestion_drop; 407 408 case RED_HARD_MARK: 409 qdisc_qstats_overlimit(sch); 410 if (sfq_hard_mark(q)) { 411 /* We know we have at least one packet in queue */ 412 if (sfq_headdrop(q) && 413 INET_ECN_set_ce(slot->skblist_next)) { 414 q->stats.forced_mark_head++; 415 break; 416 } 417 if (INET_ECN_set_ce(skb)) { 418 q->stats.forced_mark++; 419 break; 420 } 421 } 422 q->stats.forced_drop++; 423 goto congestion_drop; 424 } 425 } 426 427 if (slot->qlen >= q->maxdepth) { 428 congestion_drop: 429 if (!sfq_headdrop(q)) 430 return qdisc_drop(skb, sch, to_free); 431 432 /* We know we have at least one packet in queue */ 433 head = slot_dequeue_head(slot); 434 delta = qdisc_pkt_len(head) - qdisc_pkt_len(skb); 435 sch->qstats.backlog -= delta; 436 slot->backlog -= delta; 437 qdisc_drop(head, sch, to_free); 438 439 slot_queue_add(slot, skb); 440 qdisc_tree_reduce_backlog(sch, 0, delta); 441 return NET_XMIT_CN; 442 } 443 444 enqueue: 445 qdisc_qstats_backlog_inc(sch, skb); 446 slot->backlog += qdisc_pkt_len(skb); 447 slot_queue_add(slot, skb); 448 sfq_inc(q, x); 449 if (slot->qlen == 1) { /* The flow is new */ 450 if (q->tail == NULL) { /* It is the first flow */ 451 slot->next = x; 452 } else { 453 slot->next = q->tail->next; 454 q->tail->next = x; 455 } 456 /* We put this flow at the end of our flow list. 457 * This might sound unfair for a new flow to wait after old ones, 458 * but we could endup servicing new flows only, and freeze old ones. 459 */ 460 q->tail = slot; 461 /* We could use a bigger initial quantum for new flows */ 462 slot->allot = q->scaled_quantum; 463 } 464 if (++sch->q.qlen <= q->limit) 465 return NET_XMIT_SUCCESS; 466 467 qlen = slot->qlen; 468 dropped = sfq_drop(sch); 469 /* Return Congestion Notification only if we dropped a packet 470 * from this flow. 471 */ 472 if (qlen != slot->qlen) { 473 qdisc_tree_reduce_backlog(sch, 0, dropped - qdisc_pkt_len(skb)); 474 return NET_XMIT_CN; 475 } 476 477 /* As we dropped a packet, better let upper stack know this */ 478 qdisc_tree_reduce_backlog(sch, 1, dropped); 479 return NET_XMIT_SUCCESS; 480 } 481 482 static struct sk_buff * 483 sfq_dequeue(struct Qdisc *sch) 484 { 485 struct sfq_sched_data *q = qdisc_priv(sch); 486 struct sk_buff *skb; 487 sfq_index a, next_a; 488 struct sfq_slot *slot; 489 490 /* No active slots */ 491 if (q->tail == NULL) 492 return NULL; 493 494 next_slot: 495 a = q->tail->next; 496 slot = &q->slots[a]; 497 if (slot->allot <= 0) { 498 q->tail = slot; 499 slot->allot += q->scaled_quantum; 500 goto next_slot; 501 } 502 skb = slot_dequeue_head(slot); 503 sfq_dec(q, a); 504 qdisc_bstats_update(sch, skb); 505 sch->q.qlen--; 506 qdisc_qstats_backlog_dec(sch, skb); 507 slot->backlog -= qdisc_pkt_len(skb); 508 /* Is the slot empty? */ 509 if (slot->qlen == 0) { 510 q->ht[slot->hash] = SFQ_EMPTY_SLOT; 511 next_a = slot->next; 512 if (a == next_a) { 513 q->tail = NULL; /* no more active slots */ 514 return skb; 515 } 516 q->tail->next = next_a; 517 } else { 518 slot->allot -= SFQ_ALLOT_SIZE(qdisc_pkt_len(skb)); 519 } 520 return skb; 521 } 522 523 static void 524 sfq_reset(struct Qdisc *sch) 525 { 526 struct sk_buff *skb; 527 528 while ((skb = sfq_dequeue(sch)) != NULL) 529 rtnl_kfree_skbs(skb, skb); 530 } 531 532 /* 533 * When q->perturbation is changed, we rehash all queued skbs 534 * to avoid OOO (Out Of Order) effects. 535 * We dont use sfq_dequeue()/sfq_enqueue() because we dont want to change 536 * counters. 537 */ 538 static void sfq_rehash(struct Qdisc *sch) 539 { 540 struct sfq_sched_data *q = qdisc_priv(sch); 541 struct sk_buff *skb; 542 int i; 543 struct sfq_slot *slot; 544 struct sk_buff_head list; 545 int dropped = 0; 546 unsigned int drop_len = 0; 547 548 __skb_queue_head_init(&list); 549 550 for (i = 0; i < q->maxflows; i++) { 551 slot = &q->slots[i]; 552 if (!slot->qlen) 553 continue; 554 while (slot->qlen) { 555 skb = slot_dequeue_head(slot); 556 sfq_dec(q, i); 557 __skb_queue_tail(&list, skb); 558 } 559 slot->backlog = 0; 560 red_set_vars(&slot->vars); 561 q->ht[slot->hash] = SFQ_EMPTY_SLOT; 562 } 563 q->tail = NULL; 564 565 while ((skb = __skb_dequeue(&list)) != NULL) { 566 unsigned int hash = sfq_hash(q, skb); 567 sfq_index x = q->ht[hash]; 568 569 slot = &q->slots[x]; 570 if (x == SFQ_EMPTY_SLOT) { 571 x = q->dep[0].next; /* get a free slot */ 572 if (x >= SFQ_MAX_FLOWS) { 573 drop: 574 qdisc_qstats_backlog_dec(sch, skb); 575 drop_len += qdisc_pkt_len(skb); 576 kfree_skb(skb); 577 dropped++; 578 continue; 579 } 580 q->ht[hash] = x; 581 slot = &q->slots[x]; 582 slot->hash = hash; 583 } 584 if (slot->qlen >= q->maxdepth) 585 goto drop; 586 slot_queue_add(slot, skb); 587 if (q->red_parms) 588 slot->vars.qavg = red_calc_qavg(q->red_parms, 589 &slot->vars, 590 slot->backlog); 591 slot->backlog += qdisc_pkt_len(skb); 592 sfq_inc(q, x); 593 if (slot->qlen == 1) { /* The flow is new */ 594 if (q->tail == NULL) { /* It is the first flow */ 595 slot->next = x; 596 } else { 597 slot->next = q->tail->next; 598 q->tail->next = x; 599 } 600 q->tail = slot; 601 slot->allot = q->scaled_quantum; 602 } 603 } 604 sch->q.qlen -= dropped; 605 qdisc_tree_reduce_backlog(sch, dropped, drop_len); 606 } 607 608 static void sfq_perturbation(unsigned long arg) 609 { 610 struct Qdisc *sch = (struct Qdisc *)arg; 611 struct sfq_sched_data *q = qdisc_priv(sch); 612 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); 613 614 spin_lock(root_lock); 615 q->perturbation = prandom_u32(); 616 if (!q->filter_list && q->tail) 617 sfq_rehash(sch); 618 spin_unlock(root_lock); 619 620 if (q->perturb_period) 621 mod_timer(&q->perturb_timer, jiffies + q->perturb_period); 622 } 623 624 static int sfq_change(struct Qdisc *sch, struct nlattr *opt) 625 { 626 struct sfq_sched_data *q = qdisc_priv(sch); 627 struct tc_sfq_qopt *ctl = nla_data(opt); 628 struct tc_sfq_qopt_v1 *ctl_v1 = NULL; 629 unsigned int qlen, dropped = 0; 630 struct red_parms *p = NULL; 631 632 if (opt->nla_len < nla_attr_size(sizeof(*ctl))) 633 return -EINVAL; 634 if (opt->nla_len >= nla_attr_size(sizeof(*ctl_v1))) 635 ctl_v1 = nla_data(opt); 636 if (ctl->divisor && 637 (!is_power_of_2(ctl->divisor) || ctl->divisor > 65536)) 638 return -EINVAL; 639 if (ctl_v1 && ctl_v1->qth_min) { 640 p = kmalloc(sizeof(*p), GFP_KERNEL); 641 if (!p) 642 return -ENOMEM; 643 } 644 sch_tree_lock(sch); 645 if (ctl->quantum) { 646 q->quantum = ctl->quantum; 647 q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum); 648 } 649 q->perturb_period = ctl->perturb_period * HZ; 650 if (ctl->flows) 651 q->maxflows = min_t(u32, ctl->flows, SFQ_MAX_FLOWS); 652 if (ctl->divisor) { 653 q->divisor = ctl->divisor; 654 q->maxflows = min_t(u32, q->maxflows, q->divisor); 655 } 656 if (ctl_v1) { 657 if (ctl_v1->depth) 658 q->maxdepth = min_t(u32, ctl_v1->depth, SFQ_MAX_DEPTH); 659 if (p) { 660 swap(q->red_parms, p); 661 red_set_parms(q->red_parms, 662 ctl_v1->qth_min, ctl_v1->qth_max, 663 ctl_v1->Wlog, 664 ctl_v1->Plog, ctl_v1->Scell_log, 665 NULL, 666 ctl_v1->max_P); 667 } 668 q->flags = ctl_v1->flags; 669 q->headdrop = ctl_v1->headdrop; 670 } 671 if (ctl->limit) { 672 q->limit = min_t(u32, ctl->limit, q->maxdepth * q->maxflows); 673 q->maxflows = min_t(u32, q->maxflows, q->limit); 674 } 675 676 qlen = sch->q.qlen; 677 while (sch->q.qlen > q->limit) 678 dropped += sfq_drop(sch); 679 qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped); 680 681 del_timer(&q->perturb_timer); 682 if (q->perturb_period) { 683 mod_timer(&q->perturb_timer, jiffies + q->perturb_period); 684 q->perturbation = prandom_u32(); 685 } 686 sch_tree_unlock(sch); 687 kfree(p); 688 return 0; 689 } 690 691 static void *sfq_alloc(size_t sz) 692 { 693 return kvmalloc(sz, GFP_KERNEL); 694 } 695 696 static void sfq_free(void *addr) 697 { 698 kvfree(addr); 699 } 700 701 static void sfq_destroy(struct Qdisc *sch) 702 { 703 struct sfq_sched_data *q = qdisc_priv(sch); 704 705 tcf_block_put(q->block); 706 q->perturb_period = 0; 707 del_timer_sync(&q->perturb_timer); 708 sfq_free(q->ht); 709 sfq_free(q->slots); 710 kfree(q->red_parms); 711 } 712 713 static int sfq_init(struct Qdisc *sch, struct nlattr *opt) 714 { 715 struct sfq_sched_data *q = qdisc_priv(sch); 716 int i; 717 int err; 718 719 err = tcf_block_get(&q->block, &q->filter_list); 720 if (err) 721 return err; 722 723 setup_deferrable_timer(&q->perturb_timer, sfq_perturbation, 724 (unsigned long)sch); 725 726 for (i = 0; i < SFQ_MAX_DEPTH + 1; i++) { 727 q->dep[i].next = i + SFQ_MAX_FLOWS; 728 q->dep[i].prev = i + SFQ_MAX_FLOWS; 729 } 730 731 q->limit = SFQ_MAX_DEPTH; 732 q->maxdepth = SFQ_MAX_DEPTH; 733 q->cur_depth = 0; 734 q->tail = NULL; 735 q->divisor = SFQ_DEFAULT_HASH_DIVISOR; 736 q->maxflows = SFQ_DEFAULT_FLOWS; 737 q->quantum = psched_mtu(qdisc_dev(sch)); 738 q->scaled_quantum = SFQ_ALLOT_SIZE(q->quantum); 739 q->perturb_period = 0; 740 q->perturbation = prandom_u32(); 741 742 if (opt) { 743 int err = sfq_change(sch, opt); 744 if (err) 745 return err; 746 } 747 748 q->ht = sfq_alloc(sizeof(q->ht[0]) * q->divisor); 749 q->slots = sfq_alloc(sizeof(q->slots[0]) * q->maxflows); 750 if (!q->ht || !q->slots) { 751 /* Note: sfq_destroy() will be called by our caller */ 752 return -ENOMEM; 753 } 754 755 for (i = 0; i < q->divisor; i++) 756 q->ht[i] = SFQ_EMPTY_SLOT; 757 758 for (i = 0; i < q->maxflows; i++) { 759 slot_queue_init(&q->slots[i]); 760 sfq_link(q, i); 761 } 762 if (q->limit >= 1) 763 sch->flags |= TCQ_F_CAN_BYPASS; 764 else 765 sch->flags &= ~TCQ_F_CAN_BYPASS; 766 return 0; 767 } 768 769 static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb) 770 { 771 struct sfq_sched_data *q = qdisc_priv(sch); 772 unsigned char *b = skb_tail_pointer(skb); 773 struct tc_sfq_qopt_v1 opt; 774 struct red_parms *p = q->red_parms; 775 776 memset(&opt, 0, sizeof(opt)); 777 opt.v0.quantum = q->quantum; 778 opt.v0.perturb_period = q->perturb_period / HZ; 779 opt.v0.limit = q->limit; 780 opt.v0.divisor = q->divisor; 781 opt.v0.flows = q->maxflows; 782 opt.depth = q->maxdepth; 783 opt.headdrop = q->headdrop; 784 785 if (p) { 786 opt.qth_min = p->qth_min >> p->Wlog; 787 opt.qth_max = p->qth_max >> p->Wlog; 788 opt.Wlog = p->Wlog; 789 opt.Plog = p->Plog; 790 opt.Scell_log = p->Scell_log; 791 opt.max_P = p->max_P; 792 } 793 memcpy(&opt.stats, &q->stats, sizeof(opt.stats)); 794 opt.flags = q->flags; 795 796 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) 797 goto nla_put_failure; 798 799 return skb->len; 800 801 nla_put_failure: 802 nlmsg_trim(skb, b); 803 return -1; 804 } 805 806 static struct Qdisc *sfq_leaf(struct Qdisc *sch, unsigned long arg) 807 { 808 return NULL; 809 } 810 811 static unsigned long sfq_get(struct Qdisc *sch, u32 classid) 812 { 813 return 0; 814 } 815 816 static unsigned long sfq_bind(struct Qdisc *sch, unsigned long parent, 817 u32 classid) 818 { 819 /* we cannot bypass queue discipline anymore */ 820 sch->flags &= ~TCQ_F_CAN_BYPASS; 821 return 0; 822 } 823 824 static void sfq_put(struct Qdisc *q, unsigned long cl) 825 { 826 } 827 828 static struct tcf_block *sfq_tcf_block(struct Qdisc *sch, unsigned long cl) 829 { 830 struct sfq_sched_data *q = qdisc_priv(sch); 831 832 if (cl) 833 return NULL; 834 return q->block; 835 } 836 837 static int sfq_dump_class(struct Qdisc *sch, unsigned long cl, 838 struct sk_buff *skb, struct tcmsg *tcm) 839 { 840 tcm->tcm_handle |= TC_H_MIN(cl); 841 return 0; 842 } 843 844 static int sfq_dump_class_stats(struct Qdisc *sch, unsigned long cl, 845 struct gnet_dump *d) 846 { 847 struct sfq_sched_data *q = qdisc_priv(sch); 848 sfq_index idx = q->ht[cl - 1]; 849 struct gnet_stats_queue qs = { 0 }; 850 struct tc_sfq_xstats xstats = { 0 }; 851 852 if (idx != SFQ_EMPTY_SLOT) { 853 const struct sfq_slot *slot = &q->slots[idx]; 854 855 xstats.allot = slot->allot << SFQ_ALLOT_SHIFT; 856 qs.qlen = slot->qlen; 857 qs.backlog = slot->backlog; 858 } 859 if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0) 860 return -1; 861 return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); 862 } 863 864 static void sfq_walk(struct Qdisc *sch, struct qdisc_walker *arg) 865 { 866 struct sfq_sched_data *q = qdisc_priv(sch); 867 unsigned int i; 868 869 if (arg->stop) 870 return; 871 872 for (i = 0; i < q->divisor; i++) { 873 if (q->ht[i] == SFQ_EMPTY_SLOT || 874 arg->count < arg->skip) { 875 arg->count++; 876 continue; 877 } 878 if (arg->fn(sch, i + 1, arg) < 0) { 879 arg->stop = 1; 880 break; 881 } 882 arg->count++; 883 } 884 } 885 886 static const struct Qdisc_class_ops sfq_class_ops = { 887 .leaf = sfq_leaf, 888 .get = sfq_get, 889 .put = sfq_put, 890 .tcf_block = sfq_tcf_block, 891 .bind_tcf = sfq_bind, 892 .unbind_tcf = sfq_put, 893 .dump = sfq_dump_class, 894 .dump_stats = sfq_dump_class_stats, 895 .walk = sfq_walk, 896 }; 897 898 static struct Qdisc_ops sfq_qdisc_ops __read_mostly = { 899 .cl_ops = &sfq_class_ops, 900 .id = "sfq", 901 .priv_size = sizeof(struct sfq_sched_data), 902 .enqueue = sfq_enqueue, 903 .dequeue = sfq_dequeue, 904 .peek = qdisc_peek_dequeued, 905 .init = sfq_init, 906 .reset = sfq_reset, 907 .destroy = sfq_destroy, 908 .change = NULL, 909 .dump = sfq_dump, 910 .owner = THIS_MODULE, 911 }; 912 913 static int __init sfq_module_init(void) 914 { 915 return register_qdisc(&sfq_qdisc_ops); 916 } 917 static void __exit sfq_module_exit(void) 918 { 919 unregister_qdisc(&sfq_qdisc_ops); 920 } 921 module_init(sfq_module_init) 922 module_exit(sfq_module_exit) 923 MODULE_LICENSE("GPL"); 924