2 * net/sched/sch_sfq.c Stochastic Fairness Queueing discipline.
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.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
12 #include <linux/config.h>
13 #include <linux/module.h>
14 #include <asm/uaccess.h>
15 #include <asm/system.h>
16 #include <linux/bitops.h>
17 #include <linux/types.h>
18 #include <linux/kernel.h>
19 #include <linux/jiffies.h>
20 #include <linux/string.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
25 #include <linux/errno.h>
26 #include <linux/interrupt.h>
27 #include <linux/if_ether.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
31 #include <linux/notifier.h>
32 #include <linux/init.h>
34 #include <linux/ipv6.h>
35 #include <net/route.h>
36 #include <linux/skbuff.h>
38 #include <net/pkt_sched.h>
41 /* Stochastic Fairness Queuing algorithm.
42 =======================================
45 Paul E. McKenney "Stochastic Fairness Queuing",
46 IEEE INFOCOMM'90 Proceedings, San Francisco, 1990.
48 Paul E. McKenney "Stochastic Fairness Queuing",
49 "Interworking: Research and Experience", v.2, 1991, p.113-131.
53 M. Shreedhar and George Varghese "Efficient Fair
54 Queuing using Deficit Round Robin", Proc. SIGCOMM 95.
57 This is not the thing that is usually called (W)FQ nowadays.
58 It does not use any timestamp mechanism, but instead
59 processes queues in round-robin order.
63 - It is very cheap. Both CPU and memory requirements are minimal.
67 - "Stochastic" -> It is not 100% fair.
68 When hash collisions occur, several flows are considered as one.
70 - "Round-robin" -> It introduces larger delays than virtual clock
71 based schemes, and should not be used for isolating interactive
72 traffic from non-interactive. It means, that this scheduler
73 should be used as leaf of CBQ or P3, which put interactive traffic
74 to higher priority band.
76 We still need true WFQ for top level CSZ, but using WFQ
77 for the best effort traffic is absolutely pointless:
78 SFQ is superior for this purpose.
81 This implementation limits maximal queue length to 128;
82 maximal mtu to 2^15-1; number of hash buckets to 1024.
83 The only goal of this restrictions was that all data
84 fit into one 4K page :-). Struct sfq_sched_data is
85 organized in anti-cache manner: all the data for a bucket
86 are scattered over different locations. This is not good,
87 but it allowed me to put it into 4K.
89 It is easy to increase these values, but not in flight. */
92 #define SFQ_HASH_DIVISOR 1024
94 /* This type should contain at least SFQ_DEPTH*2 values */
95 typedef unsigned char sfq_index;
103 struct sfq_sched_data
107 unsigned quantum; /* Allotment per round: MUST BE >= MTU */
111 struct timer_list perturb_timer;
113 sfq_index tail; /* Index of current slot in round */
114 sfq_index max_depth; /* Maximal depth */
116 sfq_index ht[SFQ_HASH_DIVISOR]; /* Hash table */
117 sfq_index next[SFQ_DEPTH]; /* Active slots link */
118 short allot[SFQ_DEPTH]; /* Current allotment per slot */
119 unsigned short hash[SFQ_DEPTH]; /* Hash value indexed by slots */
120 struct sk_buff_head qs[SFQ_DEPTH]; /* Slot queue */
121 struct sfq_head dep[SFQ_DEPTH*2]; /* Linked list of slots, indexed by depth */
124 static __inline__ unsigned sfq_fold_hash(struct sfq_sched_data *q, u32 h, u32 h1)
126 int pert = q->perturbation;
128 /* Have we any rotation primitives? If not, WHY? */
129 h ^= (h1<<pert) ^ (h1>>(0x1F - pert));
134 static unsigned sfq_hash(struct sfq_sched_data *q, struct sk_buff *skb)
138 switch (skb->protocol) {
139 case __constant_htons(ETH_P_IP):
141 struct iphdr *iph = skb->nh.iph;
143 h2 = iph->saddr^iph->protocol;
144 if (!(iph->frag_off&htons(IP_MF|IP_OFFSET)) &&
145 (iph->protocol == IPPROTO_TCP ||
146 iph->protocol == IPPROTO_UDP ||
147 iph->protocol == IPPROTO_SCTP ||
148 iph->protocol == IPPROTO_DCCP ||
149 iph->protocol == IPPROTO_ESP))
150 h2 ^= *(((u32*)iph) + iph->ihl);
153 case __constant_htons(ETH_P_IPV6):
155 struct ipv6hdr *iph = skb->nh.ipv6h;
156 h = iph->daddr.s6_addr32[3];
157 h2 = iph->saddr.s6_addr32[3]^iph->nexthdr;
158 if (iph->nexthdr == IPPROTO_TCP ||
159 iph->nexthdr == IPPROTO_UDP ||
160 iph->nexthdr == IPPROTO_SCTP ||
161 iph->nexthdr == IPPROTO_DCCP ||
162 iph->nexthdr == IPPROTO_ESP)
163 h2 ^= *(u32*)&iph[1];
167 h = (u32)(unsigned long)skb->dst^skb->protocol;
168 h2 = (u32)(unsigned long)skb->sk;
170 return sfq_fold_hash(q, h, h2);
173 static inline void sfq_link(struct sfq_sched_data *q, sfq_index x)
176 int d = q->qs[x].qlen + SFQ_DEPTH;
182 q->dep[p].next = q->dep[n].prev = x;
185 static inline void sfq_dec(struct sfq_sched_data *q, sfq_index x)
194 if (n == p && q->max_depth == q->qs[x].qlen + 1)
200 static inline void sfq_inc(struct sfq_sched_data *q, sfq_index x)
210 if (q->max_depth < d)
216 static unsigned int sfq_drop(struct Qdisc *sch)
218 struct sfq_sched_data *q = qdisc_priv(sch);
219 sfq_index d = q->max_depth;
223 /* Queue is full! Find the longest slot and
224 drop a packet from it */
227 sfq_index x = q->dep[d+SFQ_DEPTH].next;
230 __skb_unlink(skb, &q->qs[x]);
235 sch->qstats.backlog -= len;
240 /* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
241 d = q->next[q->tail];
242 q->next[q->tail] = q->next[d];
243 q->allot[q->next[d]] += q->quantum;
246 __skb_unlink(skb, &q->qs[d]);
250 q->ht[q->hash[d]] = SFQ_DEPTH;
252 sch->qstats.backlog -= len;
260 sfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
262 struct sfq_sched_data *q = qdisc_priv(sch);
263 unsigned hash = sfq_hash(q, skb);
267 if (x == SFQ_DEPTH) {
268 q->ht[hash] = x = q->dep[SFQ_DEPTH].next;
271 sch->qstats.backlog += skb->len;
272 __skb_queue_tail(&q->qs[x], skb);
274 if (q->qs[x].qlen == 1) { /* The flow is new */
275 if (q->tail == SFQ_DEPTH) { /* It is the first flow */
278 q->allot[x] = q->quantum;
280 q->next[x] = q->next[q->tail];
281 q->next[q->tail] = x;
285 if (++sch->q.qlen < q->limit-1) {
286 sch->bstats.bytes += skb->len;
287 sch->bstats.packets++;
296 sfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
298 struct sfq_sched_data *q = qdisc_priv(sch);
299 unsigned hash = sfq_hash(q, skb);
303 if (x == SFQ_DEPTH) {
304 q->ht[hash] = x = q->dep[SFQ_DEPTH].next;
307 sch->qstats.backlog += skb->len;
308 __skb_queue_head(&q->qs[x], skb);
310 if (q->qs[x].qlen == 1) { /* The flow is new */
311 if (q->tail == SFQ_DEPTH) { /* It is the first flow */
314 q->allot[x] = q->quantum;
316 q->next[x] = q->next[q->tail];
317 q->next[q->tail] = x;
321 if (++sch->q.qlen < q->limit - 1) {
322 sch->qstats.requeues++;
334 static struct sk_buff *
335 sfq_dequeue(struct Qdisc* sch)
337 struct sfq_sched_data *q = qdisc_priv(sch);
341 /* No active slots */
342 if (q->tail == SFQ_DEPTH)
345 a = old_a = q->next[q->tail];
348 skb = __skb_dequeue(&q->qs[a]);
351 sch->qstats.backlog -= skb->len;
353 /* Is the slot empty? */
354 if (q->qs[a].qlen == 0) {
355 q->ht[q->hash[a]] = SFQ_DEPTH;
361 q->next[q->tail] = a;
362 q->allot[a] += q->quantum;
363 } else if ((q->allot[a] -= skb->len) <= 0) {
366 q->allot[a] += q->quantum;
372 sfq_reset(struct Qdisc* sch)
376 while ((skb = sfq_dequeue(sch)) != NULL)
380 static void sfq_perturbation(unsigned long arg)
382 struct Qdisc *sch = (struct Qdisc*)arg;
383 struct sfq_sched_data *q = qdisc_priv(sch);
385 q->perturbation = net_random()&0x1F;
387 if (q->perturb_period) {
388 q->perturb_timer.expires = jiffies + q->perturb_period;
389 add_timer(&q->perturb_timer);
393 static int sfq_change(struct Qdisc *sch, struct rtattr *opt)
395 struct sfq_sched_data *q = qdisc_priv(sch);
396 struct tc_sfq_qopt *ctl = RTA_DATA(opt);
398 if (opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
402 q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
403 q->perturb_period = ctl->perturb_period*HZ;
405 q->limit = min_t(u32, ctl->limit, SFQ_DEPTH);
407 while (sch->q.qlen >= q->limit-1)
410 del_timer(&q->perturb_timer);
411 if (q->perturb_period) {
412 q->perturb_timer.expires = jiffies + q->perturb_period;
413 add_timer(&q->perturb_timer);
415 sch_tree_unlock(sch);
419 static int sfq_init(struct Qdisc *sch, struct rtattr *opt)
421 struct sfq_sched_data *q = qdisc_priv(sch);
424 init_timer(&q->perturb_timer);
425 q->perturb_timer.data = (unsigned long)sch;
426 q->perturb_timer.function = sfq_perturbation;
428 for (i=0; i<SFQ_HASH_DIVISOR; i++)
429 q->ht[i] = SFQ_DEPTH;
430 for (i=0; i<SFQ_DEPTH; i++) {
431 skb_queue_head_init(&q->qs[i]);
432 q->dep[i+SFQ_DEPTH].next = i+SFQ_DEPTH;
433 q->dep[i+SFQ_DEPTH].prev = i+SFQ_DEPTH;
435 q->limit = SFQ_DEPTH;
439 q->quantum = psched_mtu(sch->dev);
440 q->perturb_period = 0;
442 int err = sfq_change(sch, opt);
446 for (i=0; i<SFQ_DEPTH; i++)
451 static void sfq_destroy(struct Qdisc *sch)
453 struct sfq_sched_data *q = qdisc_priv(sch);
454 del_timer(&q->perturb_timer);
457 static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb)
459 struct sfq_sched_data *q = qdisc_priv(sch);
460 unsigned char *b = skb->tail;
461 struct tc_sfq_qopt opt;
463 opt.quantum = q->quantum;
464 opt.perturb_period = q->perturb_period/HZ;
466 opt.limit = q->limit;
467 opt.divisor = SFQ_HASH_DIVISOR;
468 opt.flows = q->limit;
470 RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
475 skb_trim(skb, b - skb->data);
479 static struct Qdisc_ops sfq_qdisc_ops = {
483 .priv_size = sizeof(struct sfq_sched_data),
484 .enqueue = sfq_enqueue,
485 .dequeue = sfq_dequeue,
486 .requeue = sfq_requeue,
490 .destroy = sfq_destroy,
493 .owner = THIS_MODULE,
496 static int __init sfq_module_init(void)
498 return register_qdisc(&sfq_qdisc_ops);
500 static void __exit sfq_module_exit(void)
502 unregister_qdisc(&sfq_qdisc_ops);
504 module_init(sfq_module_init)
505 module_exit(sfq_module_exit)
506 MODULE_LICENSE("GPL");