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/config.h>
13 #include <linux/module.h>
14 #include <asm/uaccess.h>
15 #include <asm/system.h>
16 #include <asm/bitops.h>
17 #include <linux/types.h>
18 #include <linux/kernel.h>
19 #include <linux/sched.h>
20 #include <linux/string.h>
21 #include <linux/mm.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/in.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>
33 #include <net/ip.h>
34 #include <linux/ipv6.h>
35 #include <net/route.h>
36 #include <linux/skbuff.h>
37 #include <net/sock.h>
38 #include <net/pkt_sched.h>
39 
40 
41 /*	Stochastic Fairness Queuing algorithm.
42 	=======================================
43 
44 	Source:
45 	Paul E. McKenney "Stochastic Fairness Queuing",
46 	IEEE INFOCOMM'90 Proceedings, San Francisco, 1990.
47 
48 	Paul E. McKenney "Stochastic Fairness Queuing",
49 	"Interworking: Research and Experience", v.2, 1991, p.113-131.
50 
51 
52 	See also:
53 	M. Shreedhar and George Varghese "Efficient Fair
54 	Queuing using Deficit Round Robin", Proc. SIGCOMM 95.
55 
56 
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.
60 
61 	ADVANTAGE:
62 
63 	- It is very cheap. Both CPU and memory requirements are minimal.
64 
65 	DRAWBACKS:
66 
67 	- "Stochastic" -> It is not 100% fair.
68 	When hash collisions occur, several flows are considered as one.
69 
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.
75 
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.
79 
80 	IMPLEMENTATION:
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.
88 
89 	It is easy to increase these values, but not in flight.  */
90 
91 #define SFQ_DEPTH		128
92 #define SFQ_HASH_DIVISOR	1024
93 
94 /* This type should contain at least SFQ_DEPTH*2 values */
95 typedef unsigned char sfq_index;
96 
97 struct sfq_head
98 {
99 	sfq_index	next;
100 	sfq_index	prev;
101 };
102 
103 struct sfq_sched_data
104 {
105 /* Parameters */
106 	int		perturb_period;
107 	unsigned	quantum;	/* Allotment per round: MUST BE >= MTU */
108 	int		limit;
109 
110 /* Variables */
111 	struct timer_list perturb_timer;
112 	int		perturbation;
113 	sfq_index	tail;		/* Index of current slot in round */
114 	sfq_index	max_depth;	/* Maximal depth */
115 
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 */
122 };
123 
sfq_fold_hash(struct sfq_sched_data * q,u32 h,u32 h1)124 static __inline__ unsigned sfq_fold_hash(struct sfq_sched_data *q, u32 h, u32 h1)
125 {
126 	int pert = q->perturbation;
127 
128 	/* Have we any rotation primitives? If not, WHY? */
129 	h ^= (h1<<pert) ^ (h1>>(0x1F - pert));
130 	h ^= h>>10;
131 	return h & 0x3FF;
132 }
133 
sfq_hash(struct sfq_sched_data * q,struct sk_buff * skb)134 static unsigned sfq_hash(struct sfq_sched_data *q, struct sk_buff *skb)
135 {
136 	u32 h, h2;
137 
138 	switch (skb->protocol) {
139 	case __constant_htons(ETH_P_IP):
140 	{
141 		struct iphdr *iph = skb->nh.iph;
142 		h = iph->daddr;
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_ESP))
148 			h2 ^= *(((u32*)iph) + iph->ihl);
149 		break;
150 	}
151 	case __constant_htons(ETH_P_IPV6):
152 	{
153 		struct ipv6hdr *iph = skb->nh.ipv6h;
154 		h = iph->daddr.s6_addr32[3];
155 		h2 = iph->saddr.s6_addr32[3]^iph->nexthdr;
156 		if (iph->nexthdr == IPPROTO_TCP ||
157 		    iph->nexthdr == IPPROTO_UDP ||
158 		    iph->nexthdr == IPPROTO_ESP)
159 			h2 ^= *(u32*)&iph[1];
160 		break;
161 	}
162 	default:
163 		h = (u32)(unsigned long)skb->dst^skb->protocol;
164 		h2 = (u32)(unsigned long)skb->sk;
165 	}
166 	return sfq_fold_hash(q, h, h2);
167 }
168 
sfq_link(struct sfq_sched_data * q,sfq_index x)169 extern __inline__ void sfq_link(struct sfq_sched_data *q, sfq_index x)
170 {
171 	sfq_index p, n;
172 	int d = q->qs[x].qlen + SFQ_DEPTH;
173 
174 	p = d;
175 	n = q->dep[d].next;
176 	q->dep[x].next = n;
177 	q->dep[x].prev = p;
178 	q->dep[p].next = q->dep[n].prev = x;
179 }
180 
sfq_dec(struct sfq_sched_data * q,sfq_index x)181 extern __inline__ void sfq_dec(struct sfq_sched_data *q, sfq_index x)
182 {
183 	sfq_index p, n;
184 
185 	n = q->dep[x].next;
186 	p = q->dep[x].prev;
187 	q->dep[p].next = n;
188 	q->dep[n].prev = p;
189 
190 	if (n == p && q->max_depth == q->qs[x].qlen + 1)
191 		q->max_depth--;
192 
193 	sfq_link(q, x);
194 }
195 
sfq_inc(struct sfq_sched_data * q,sfq_index x)196 extern __inline__ void sfq_inc(struct sfq_sched_data *q, sfq_index x)
197 {
198 	sfq_index p, n;
199 	int d;
200 
201 	n = q->dep[x].next;
202 	p = q->dep[x].prev;
203 	q->dep[p].next = n;
204 	q->dep[n].prev = p;
205 	d = q->qs[x].qlen;
206 	if (q->max_depth < d)
207 		q->max_depth = d;
208 
209 	sfq_link(q, x);
210 }
211 
sfq_drop(struct Qdisc * sch)212 static unsigned int sfq_drop(struct Qdisc *sch)
213 {
214 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
215 	sfq_index d = q->max_depth;
216 	struct sk_buff *skb;
217 	unsigned int len;
218 
219 	/* Queue is full! Find the longest slot and
220 	   drop a packet from it */
221 
222 	if (d > 1) {
223 		sfq_index x = q->dep[d+SFQ_DEPTH].next;
224 		skb = q->qs[x].prev;
225 		len = skb->len;
226 		__skb_unlink(skb, &q->qs[x]);
227 		kfree_skb(skb);
228 		sfq_dec(q, x);
229 		sch->q.qlen--;
230 		sch->stats.drops++;
231 		return len;
232 	}
233 
234 	if (d == 1) {
235 		/* It is difficult to believe, but ALL THE SLOTS HAVE LENGTH 1. */
236 		d = q->next[q->tail];
237 		q->next[q->tail] = q->next[d];
238 		q->allot[q->next[d]] += q->quantum;
239 		skb = q->qs[d].prev;
240 		len = skb->len;
241 		__skb_unlink(skb, &q->qs[d]);
242 		kfree_skb(skb);
243 		sfq_dec(q, d);
244 		sch->q.qlen--;
245 		q->ht[q->hash[d]] = SFQ_DEPTH;
246 		sch->stats.drops++;
247 		return len;
248 	}
249 
250 	return 0;
251 }
252 
253 static int
sfq_enqueue(struct sk_buff * skb,struct Qdisc * sch)254 sfq_enqueue(struct sk_buff *skb, struct Qdisc* sch)
255 {
256 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
257 	unsigned hash = sfq_hash(q, skb);
258 	sfq_index x;
259 
260 	x = q->ht[hash];
261 	if (x == SFQ_DEPTH) {
262 		q->ht[hash] = x = q->dep[SFQ_DEPTH].next;
263 		q->hash[x] = hash;
264 	}
265 	__skb_queue_tail(&q->qs[x], skb);
266 	sfq_inc(q, x);
267 	if (q->qs[x].qlen == 1) {		/* The flow is new */
268 		if (q->tail == SFQ_DEPTH) {	/* It is the first flow */
269 			q->tail = x;
270 			q->next[x] = x;
271 			q->allot[x] = q->quantum;
272 		} else {
273 			q->next[x] = q->next[q->tail];
274 			q->next[q->tail] = x;
275 			q->tail = x;
276 		}
277 	}
278 	if (++sch->q.qlen < q->limit-1) {
279 		sch->stats.bytes += skb->len;
280 		sch->stats.packets++;
281 		return 0;
282 	}
283 
284 	sfq_drop(sch);
285 	return NET_XMIT_CN;
286 }
287 
288 static int
sfq_requeue(struct sk_buff * skb,struct Qdisc * sch)289 sfq_requeue(struct sk_buff *skb, struct Qdisc* sch)
290 {
291 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
292 	unsigned hash = sfq_hash(q, skb);
293 	sfq_index x;
294 
295 	x = q->ht[hash];
296 	if (x == SFQ_DEPTH) {
297 		q->ht[hash] = x = q->dep[SFQ_DEPTH].next;
298 		q->hash[x] = hash;
299 	}
300 	__skb_queue_head(&q->qs[x], skb);
301 	sfq_inc(q, x);
302 	if (q->qs[x].qlen == 1) {		/* The flow is new */
303 		if (q->tail == SFQ_DEPTH) {	/* It is the first flow */
304 			q->tail = x;
305 			q->next[x] = x;
306 			q->allot[x] = q->quantum;
307 		} else {
308 			q->next[x] = q->next[q->tail];
309 			q->next[q->tail] = x;
310 			q->tail = x;
311 		}
312 	}
313 	if (++sch->q.qlen < q->limit - 1)
314 		return 0;
315 
316 	sch->stats.drops++;
317 	sfq_drop(sch);
318 	return NET_XMIT_CN;
319 }
320 
321 
322 
323 
324 static struct sk_buff *
sfq_dequeue(struct Qdisc * sch)325 sfq_dequeue(struct Qdisc* sch)
326 {
327 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
328 	struct sk_buff *skb;
329 	sfq_index a, old_a;
330 
331 	/* No active slots */
332 	if (q->tail == SFQ_DEPTH)
333 		return NULL;
334 
335 	a = old_a = q->next[q->tail];
336 
337 	/* Grab packet */
338 	skb = __skb_dequeue(&q->qs[a]);
339 	sfq_dec(q, a);
340 	sch->q.qlen--;
341 
342 	/* Is the slot empty? */
343 	if (q->qs[a].qlen == 0) {
344 		q->ht[q->hash[a]] = SFQ_DEPTH;
345 		a = q->next[a];
346 		if (a == old_a) {
347 			q->tail = SFQ_DEPTH;
348 			return skb;
349 		}
350 		q->next[q->tail] = a;
351 		q->allot[a] += q->quantum;
352 	} else if ((q->allot[a] -= skb->len) <= 0) {
353 		q->tail = a;
354 		a = q->next[a];
355 		q->allot[a] += q->quantum;
356 	}
357 	return skb;
358 }
359 
360 static void
sfq_reset(struct Qdisc * sch)361 sfq_reset(struct Qdisc* sch)
362 {
363 	struct sk_buff *skb;
364 
365 	while ((skb = sfq_dequeue(sch)) != NULL)
366 		kfree_skb(skb);
367 }
368 
sfq_perturbation(unsigned long arg)369 static void sfq_perturbation(unsigned long arg)
370 {
371 	struct Qdisc *sch = (struct Qdisc*)arg;
372 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
373 
374 	q->perturbation = net_random()&0x1F;
375 	q->perturb_timer.expires = jiffies + q->perturb_period;
376 
377 	if (q->perturb_period) {
378 		q->perturb_timer.expires = jiffies + q->perturb_period;
379 		add_timer(&q->perturb_timer);
380 	}
381 }
382 
sfq_change(struct Qdisc * sch,struct rtattr * opt)383 static int sfq_change(struct Qdisc *sch, struct rtattr *opt)
384 {
385 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
386 	struct tc_sfq_qopt *ctl = RTA_DATA(opt);
387 
388 	if (opt->rta_len < RTA_LENGTH(sizeof(*ctl)))
389 		return -EINVAL;
390 
391 	sch_tree_lock(sch);
392 	q->quantum = ctl->quantum ? : psched_mtu(sch->dev);
393 	q->perturb_period = ctl->perturb_period*HZ;
394 	if (ctl->limit)
395 		q->limit = min_t(u32, ctl->limit, SFQ_DEPTH);
396 
397 	while (sch->q.qlen >= q->limit-1)
398 		sfq_drop(sch);
399 
400 	del_timer(&q->perturb_timer);
401 	if (q->perturb_period) {
402 		q->perturb_timer.expires = jiffies + q->perturb_period;
403 		add_timer(&q->perturb_timer);
404 	}
405 	sch_tree_unlock(sch);
406 	return 0;
407 }
408 
sfq_init(struct Qdisc * sch,struct rtattr * opt)409 static int sfq_init(struct Qdisc *sch, struct rtattr *opt)
410 {
411 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
412 	int i;
413 
414 	q->perturb_timer.data = (unsigned long)sch;
415 	q->perturb_timer.function = sfq_perturbation;
416 	init_timer(&q->perturb_timer);
417 
418 	for (i=0; i<SFQ_HASH_DIVISOR; i++)
419 		q->ht[i] = SFQ_DEPTH;
420 	for (i=0; i<SFQ_DEPTH; i++) {
421 		skb_queue_head_init(&q->qs[i]);
422 		q->dep[i+SFQ_DEPTH].next = i+SFQ_DEPTH;
423 		q->dep[i+SFQ_DEPTH].prev = i+SFQ_DEPTH;
424 	}
425 	q->limit = SFQ_DEPTH;
426 	q->max_depth = 0;
427 	q->tail = SFQ_DEPTH;
428 	if (opt == NULL) {
429 		q->quantum = psched_mtu(sch->dev);
430 		q->perturb_period = 0;
431 	} else {
432 		int err = sfq_change(sch, opt);
433 		if (err)
434 			return err;
435 	}
436 	for (i=0; i<SFQ_DEPTH; i++)
437 		sfq_link(q, i);
438 	MOD_INC_USE_COUNT;
439 	return 0;
440 }
441 
sfq_destroy(struct Qdisc * sch)442 static void sfq_destroy(struct Qdisc *sch)
443 {
444 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
445 	del_timer(&q->perturb_timer);
446 	MOD_DEC_USE_COUNT;
447 }
448 
sfq_dump(struct Qdisc * sch,struct sk_buff * skb)449 static int sfq_dump(struct Qdisc *sch, struct sk_buff *skb)
450 {
451 	struct sfq_sched_data *q = (struct sfq_sched_data *)sch->data;
452 	unsigned char	 *b = skb->tail;
453 	struct tc_sfq_qopt opt;
454 
455 	opt.quantum = q->quantum;
456 	opt.perturb_period = q->perturb_period/HZ;
457 
458 	opt.limit = q->limit;
459 	opt.divisor = SFQ_HASH_DIVISOR;
460 	opt.flows = q->limit;
461 
462 	RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
463 
464 	return skb->len;
465 
466 rtattr_failure:
467 	skb_trim(skb, b - skb->data);
468 	return -1;
469 }
470 
471 struct Qdisc_ops sfq_qdisc_ops =
472 {
473 	NULL,
474 	NULL,
475 	"sfq",
476 	sizeof(struct sfq_sched_data),
477 
478 	sfq_enqueue,
479 	sfq_dequeue,
480 	sfq_requeue,
481 	sfq_drop,
482 
483 	sfq_init,
484 	sfq_reset,
485 	sfq_destroy,
486 	NULL, /* sfq_change */
487 
488 	sfq_dump,
489 };
490 
491 #ifdef MODULE
init_module(void)492 int init_module(void)
493 {
494 	return register_qdisc(&sfq_qdisc_ops);
495 }
496 
cleanup_module(void)497 void cleanup_module(void)
498 {
499 	unregister_qdisc(&sfq_qdisc_ops);
500 }
501 #endif
502 MODULE_LICENSE("GPL");
503