xref: /linux/net/sched/sch_sfb.c (revision 79790b6818e96c58fe2bffee1b418c16e64e7b80)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * net/sched/sch_sfb.c	  Stochastic Fair Blue
4  *
5  * Copyright (c) 2008-2011 Juliusz Chroboczek <jch@pps.jussieu.fr>
6  * Copyright (c) 2011 Eric Dumazet <eric.dumazet@gmail.com>
7  *
8  * W. Feng, D. Kandlur, D. Saha, K. Shin. Blue:
9  * A New Class of Active Queue Management Algorithms.
10  * U. Michigan CSE-TR-387-99, April 1999.
11  *
12  * http://www.thefengs.com/wuchang/blue/CSE-TR-387-99.pdf
13  */
14 
15 #include <linux/module.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <linux/random.h>
21 #include <linux/siphash.h>
22 #include <net/ip.h>
23 #include <net/pkt_sched.h>
24 #include <net/pkt_cls.h>
25 #include <net/inet_ecn.h>
26 
27 /*
28  * SFB uses two B[l][n] : L x N arrays of bins (L levels, N bins per level)
29  * This implementation uses L = 8 and N = 16
30  * This permits us to split one 32bit hash (provided per packet by rxhash or
31  * external classifier) into 8 subhashes of 4 bits.
32  */
33 #define SFB_BUCKET_SHIFT 4
34 #define SFB_NUMBUCKETS	(1 << SFB_BUCKET_SHIFT) /* N bins per Level */
35 #define SFB_BUCKET_MASK (SFB_NUMBUCKETS - 1)
36 #define SFB_LEVELS	(32 / SFB_BUCKET_SHIFT) /* L */
37 
38 /* SFB algo uses a virtual queue, named "bin" */
39 struct sfb_bucket {
40 	u16		qlen; /* length of virtual queue */
41 	u16		p_mark; /* marking probability */
42 };
43 
44 /* We use a double buffering right before hash change
45  * (Section 4.4 of SFB reference : moving hash functions)
46  */
47 struct sfb_bins {
48 	siphash_key_t	  perturbation; /* siphash key */
49 	struct sfb_bucket bins[SFB_LEVELS][SFB_NUMBUCKETS];
50 };
51 
52 struct sfb_sched_data {
53 	struct Qdisc	*qdisc;
54 	struct tcf_proto __rcu *filter_list;
55 	struct tcf_block *block;
56 	unsigned long	rehash_interval;
57 	unsigned long	warmup_time;	/* double buffering warmup time in jiffies */
58 	u32		max;
59 	u32		bin_size;	/* maximum queue length per bin */
60 	u32		increment;	/* d1 */
61 	u32		decrement;	/* d2 */
62 	u32		limit;		/* HARD maximal queue length */
63 	u32		penalty_rate;
64 	u32		penalty_burst;
65 	u32		tokens_avail;
66 	unsigned long	rehash_time;
67 	unsigned long	token_time;
68 
69 	u8		slot;		/* current active bins (0 or 1) */
70 	bool		double_buffering;
71 	struct sfb_bins bins[2];
72 
73 	struct {
74 		u32	earlydrop;
75 		u32	penaltydrop;
76 		u32	bucketdrop;
77 		u32	queuedrop;
78 		u32	childdrop;	/* drops in child qdisc */
79 		u32	marked;		/* ECN mark */
80 	} stats;
81 };
82 
83 /*
84  * Each queued skb might be hashed on one or two bins
85  * We store in skb_cb the two hash values.
86  * (A zero value means double buffering was not used)
87  */
88 struct sfb_skb_cb {
89 	u32 hashes[2];
90 };
91 
sfb_skb_cb(const struct sk_buff * skb)92 static inline struct sfb_skb_cb *sfb_skb_cb(const struct sk_buff *skb)
93 {
94 	qdisc_cb_private_validate(skb, sizeof(struct sfb_skb_cb));
95 	return (struct sfb_skb_cb *)qdisc_skb_cb(skb)->data;
96 }
97 
98 /*
99  * If using 'internal' SFB flow classifier, hash comes from skb rxhash
100  * If using external classifier, hash comes from the classid.
101  */
sfb_hash(const struct sk_buff * skb,u32 slot)102 static u32 sfb_hash(const struct sk_buff *skb, u32 slot)
103 {
104 	return sfb_skb_cb(skb)->hashes[slot];
105 }
106 
107 /* Probabilities are coded as Q0.16 fixed-point values,
108  * with 0xFFFF representing 65535/65536 (almost 1.0)
109  * Addition and subtraction are saturating in [0, 65535]
110  */
prob_plus(u32 p1,u32 p2)111 static u32 prob_plus(u32 p1, u32 p2)
112 {
113 	u32 res = p1 + p2;
114 
115 	return min_t(u32, res, SFB_MAX_PROB);
116 }
117 
prob_minus(u32 p1,u32 p2)118 static u32 prob_minus(u32 p1, u32 p2)
119 {
120 	return p1 > p2 ? p1 - p2 : 0;
121 }
122 
increment_one_qlen(u32 sfbhash,u32 slot,struct sfb_sched_data * q)123 static void increment_one_qlen(u32 sfbhash, u32 slot, struct sfb_sched_data *q)
124 {
125 	int i;
126 	struct sfb_bucket *b = &q->bins[slot].bins[0][0];
127 
128 	for (i = 0; i < SFB_LEVELS; i++) {
129 		u32 hash = sfbhash & SFB_BUCKET_MASK;
130 
131 		sfbhash >>= SFB_BUCKET_SHIFT;
132 		if (b[hash].qlen < 0xFFFF)
133 			b[hash].qlen++;
134 		b += SFB_NUMBUCKETS; /* next level */
135 	}
136 }
137 
increment_qlen(const struct sfb_skb_cb * cb,struct sfb_sched_data * q)138 static void increment_qlen(const struct sfb_skb_cb *cb, struct sfb_sched_data *q)
139 {
140 	u32 sfbhash;
141 
142 	sfbhash = cb->hashes[0];
143 	if (sfbhash)
144 		increment_one_qlen(sfbhash, 0, q);
145 
146 	sfbhash = cb->hashes[1];
147 	if (sfbhash)
148 		increment_one_qlen(sfbhash, 1, q);
149 }
150 
decrement_one_qlen(u32 sfbhash,u32 slot,struct sfb_sched_data * q)151 static void decrement_one_qlen(u32 sfbhash, u32 slot,
152 			       struct sfb_sched_data *q)
153 {
154 	int i;
155 	struct sfb_bucket *b = &q->bins[slot].bins[0][0];
156 
157 	for (i = 0; i < SFB_LEVELS; i++) {
158 		u32 hash = sfbhash & SFB_BUCKET_MASK;
159 
160 		sfbhash >>= SFB_BUCKET_SHIFT;
161 		if (b[hash].qlen > 0)
162 			b[hash].qlen--;
163 		b += SFB_NUMBUCKETS; /* next level */
164 	}
165 }
166 
decrement_qlen(const struct sk_buff * skb,struct sfb_sched_data * q)167 static void decrement_qlen(const struct sk_buff *skb, struct sfb_sched_data *q)
168 {
169 	u32 sfbhash;
170 
171 	sfbhash = sfb_hash(skb, 0);
172 	if (sfbhash)
173 		decrement_one_qlen(sfbhash, 0, q);
174 
175 	sfbhash = sfb_hash(skb, 1);
176 	if (sfbhash)
177 		decrement_one_qlen(sfbhash, 1, q);
178 }
179 
decrement_prob(struct sfb_bucket * b,struct sfb_sched_data * q)180 static void decrement_prob(struct sfb_bucket *b, struct sfb_sched_data *q)
181 {
182 	b->p_mark = prob_minus(b->p_mark, q->decrement);
183 }
184 
increment_prob(struct sfb_bucket * b,struct sfb_sched_data * q)185 static void increment_prob(struct sfb_bucket *b, struct sfb_sched_data *q)
186 {
187 	b->p_mark = prob_plus(b->p_mark, q->increment);
188 }
189 
sfb_zero_all_buckets(struct sfb_sched_data * q)190 static void sfb_zero_all_buckets(struct sfb_sched_data *q)
191 {
192 	memset(&q->bins, 0, sizeof(q->bins));
193 }
194 
195 /*
196  * compute max qlen, max p_mark, and avg p_mark
197  */
sfb_compute_qlen(u32 * prob_r,u32 * avgpm_r,const struct sfb_sched_data * q)198 static u32 sfb_compute_qlen(u32 *prob_r, u32 *avgpm_r, const struct sfb_sched_data *q)
199 {
200 	int i;
201 	u32 qlen = 0, prob = 0, totalpm = 0;
202 	const struct sfb_bucket *b = &q->bins[q->slot].bins[0][0];
203 
204 	for (i = 0; i < SFB_LEVELS * SFB_NUMBUCKETS; i++) {
205 		if (qlen < b->qlen)
206 			qlen = b->qlen;
207 		totalpm += b->p_mark;
208 		if (prob < b->p_mark)
209 			prob = b->p_mark;
210 		b++;
211 	}
212 	*prob_r = prob;
213 	*avgpm_r = totalpm / (SFB_LEVELS * SFB_NUMBUCKETS);
214 	return qlen;
215 }
216 
217 
sfb_init_perturbation(u32 slot,struct sfb_sched_data * q)218 static void sfb_init_perturbation(u32 slot, struct sfb_sched_data *q)
219 {
220 	get_random_bytes(&q->bins[slot].perturbation,
221 			 sizeof(q->bins[slot].perturbation));
222 }
223 
sfb_swap_slot(struct sfb_sched_data * q)224 static void sfb_swap_slot(struct sfb_sched_data *q)
225 {
226 	sfb_init_perturbation(q->slot, q);
227 	q->slot ^= 1;
228 	q->double_buffering = false;
229 }
230 
231 /* Non elastic flows are allowed to use part of the bandwidth, expressed
232  * in "penalty_rate" packets per second, with "penalty_burst" burst
233  */
sfb_rate_limit(struct sk_buff * skb,struct sfb_sched_data * q)234 static bool sfb_rate_limit(struct sk_buff *skb, struct sfb_sched_data *q)
235 {
236 	if (q->penalty_rate == 0 || q->penalty_burst == 0)
237 		return true;
238 
239 	if (q->tokens_avail < 1) {
240 		unsigned long age = min(10UL * HZ, jiffies - q->token_time);
241 
242 		q->tokens_avail = (age * q->penalty_rate) / HZ;
243 		if (q->tokens_avail > q->penalty_burst)
244 			q->tokens_avail = q->penalty_burst;
245 		q->token_time = jiffies;
246 		if (q->tokens_avail < 1)
247 			return true;
248 	}
249 
250 	q->tokens_avail--;
251 	return false;
252 }
253 
sfb_classify(struct sk_buff * skb,struct tcf_proto * fl,int * qerr,u32 * salt)254 static bool sfb_classify(struct sk_buff *skb, struct tcf_proto *fl,
255 			 int *qerr, u32 *salt)
256 {
257 	struct tcf_result res;
258 	int result;
259 
260 	result = tcf_classify(skb, NULL, fl, &res, false);
261 	if (result >= 0) {
262 #ifdef CONFIG_NET_CLS_ACT
263 		switch (result) {
264 		case TC_ACT_STOLEN:
265 		case TC_ACT_QUEUED:
266 		case TC_ACT_TRAP:
267 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
268 			fallthrough;
269 		case TC_ACT_SHOT:
270 			return false;
271 		}
272 #endif
273 		*salt = TC_H_MIN(res.classid);
274 		return true;
275 	}
276 	return false;
277 }
278 
sfb_enqueue(struct sk_buff * skb,struct Qdisc * sch,struct sk_buff ** to_free)279 static int sfb_enqueue(struct sk_buff *skb, struct Qdisc *sch,
280 		       struct sk_buff **to_free)
281 {
282 
283 	struct sfb_sched_data *q = qdisc_priv(sch);
284 	unsigned int len = qdisc_pkt_len(skb);
285 	struct Qdisc *child = q->qdisc;
286 	struct tcf_proto *fl;
287 	struct sfb_skb_cb cb;
288 	int i;
289 	u32 p_min = ~0;
290 	u32 minqlen = ~0;
291 	u32 r, sfbhash;
292 	u32 slot = q->slot;
293 	int ret = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
294 
295 	if (unlikely(sch->q.qlen >= q->limit)) {
296 		qdisc_qstats_overlimit(sch);
297 		q->stats.queuedrop++;
298 		goto drop;
299 	}
300 
301 	if (q->rehash_interval > 0) {
302 		unsigned long limit = q->rehash_time + q->rehash_interval;
303 
304 		if (unlikely(time_after(jiffies, limit))) {
305 			sfb_swap_slot(q);
306 			q->rehash_time = jiffies;
307 		} else if (unlikely(!q->double_buffering && q->warmup_time > 0 &&
308 				    time_after(jiffies, limit - q->warmup_time))) {
309 			q->double_buffering = true;
310 		}
311 	}
312 
313 	fl = rcu_dereference_bh(q->filter_list);
314 	if (fl) {
315 		u32 salt;
316 
317 		/* If using external classifiers, get result and record it. */
318 		if (!sfb_classify(skb, fl, &ret, &salt))
319 			goto other_drop;
320 		sfbhash = siphash_1u32(salt, &q->bins[slot].perturbation);
321 	} else {
322 		sfbhash = skb_get_hash_perturb(skb, &q->bins[slot].perturbation);
323 	}
324 
325 
326 	if (!sfbhash)
327 		sfbhash = 1;
328 	sfb_skb_cb(skb)->hashes[slot] = sfbhash;
329 
330 	for (i = 0; i < SFB_LEVELS; i++) {
331 		u32 hash = sfbhash & SFB_BUCKET_MASK;
332 		struct sfb_bucket *b = &q->bins[slot].bins[i][hash];
333 
334 		sfbhash >>= SFB_BUCKET_SHIFT;
335 		if (b->qlen == 0)
336 			decrement_prob(b, q);
337 		else if (b->qlen >= q->bin_size)
338 			increment_prob(b, q);
339 		if (minqlen > b->qlen)
340 			minqlen = b->qlen;
341 		if (p_min > b->p_mark)
342 			p_min = b->p_mark;
343 	}
344 
345 	slot ^= 1;
346 	sfb_skb_cb(skb)->hashes[slot] = 0;
347 
348 	if (unlikely(minqlen >= q->max)) {
349 		qdisc_qstats_overlimit(sch);
350 		q->stats.bucketdrop++;
351 		goto drop;
352 	}
353 
354 	if (unlikely(p_min >= SFB_MAX_PROB)) {
355 		/* Inelastic flow */
356 		if (q->double_buffering) {
357 			sfbhash = skb_get_hash_perturb(skb,
358 			    &q->bins[slot].perturbation);
359 			if (!sfbhash)
360 				sfbhash = 1;
361 			sfb_skb_cb(skb)->hashes[slot] = sfbhash;
362 
363 			for (i = 0; i < SFB_LEVELS; i++) {
364 				u32 hash = sfbhash & SFB_BUCKET_MASK;
365 				struct sfb_bucket *b = &q->bins[slot].bins[i][hash];
366 
367 				sfbhash >>= SFB_BUCKET_SHIFT;
368 				if (b->qlen == 0)
369 					decrement_prob(b, q);
370 				else if (b->qlen >= q->bin_size)
371 					increment_prob(b, q);
372 			}
373 		}
374 		if (sfb_rate_limit(skb, q)) {
375 			qdisc_qstats_overlimit(sch);
376 			q->stats.penaltydrop++;
377 			goto drop;
378 		}
379 		goto enqueue;
380 	}
381 
382 	r = get_random_u16() & SFB_MAX_PROB;
383 
384 	if (unlikely(r < p_min)) {
385 		if (unlikely(p_min > SFB_MAX_PROB / 2)) {
386 			/* If we're marking that many packets, then either
387 			 * this flow is unresponsive, or we're badly congested.
388 			 * In either case, we want to start dropping packets.
389 			 */
390 			if (r < (p_min - SFB_MAX_PROB / 2) * 2) {
391 				q->stats.earlydrop++;
392 				goto drop;
393 			}
394 		}
395 		if (INET_ECN_set_ce(skb)) {
396 			q->stats.marked++;
397 		} else {
398 			q->stats.earlydrop++;
399 			goto drop;
400 		}
401 	}
402 
403 enqueue:
404 	memcpy(&cb, sfb_skb_cb(skb), sizeof(cb));
405 	ret = qdisc_enqueue(skb, child, to_free);
406 	if (likely(ret == NET_XMIT_SUCCESS)) {
407 		sch->qstats.backlog += len;
408 		sch->q.qlen++;
409 		increment_qlen(&cb, q);
410 	} else if (net_xmit_drop_count(ret)) {
411 		q->stats.childdrop++;
412 		qdisc_qstats_drop(sch);
413 	}
414 	return ret;
415 
416 drop:
417 	qdisc_drop(skb, sch, to_free);
418 	return NET_XMIT_CN;
419 other_drop:
420 	if (ret & __NET_XMIT_BYPASS)
421 		qdisc_qstats_drop(sch);
422 	kfree_skb(skb);
423 	return ret;
424 }
425 
sfb_dequeue(struct Qdisc * sch)426 static struct sk_buff *sfb_dequeue(struct Qdisc *sch)
427 {
428 	struct sfb_sched_data *q = qdisc_priv(sch);
429 	struct Qdisc *child = q->qdisc;
430 	struct sk_buff *skb;
431 
432 	skb = child->dequeue(q->qdisc);
433 
434 	if (skb) {
435 		qdisc_bstats_update(sch, skb);
436 		qdisc_qstats_backlog_dec(sch, skb);
437 		sch->q.qlen--;
438 		decrement_qlen(skb, q);
439 	}
440 
441 	return skb;
442 }
443 
sfb_peek(struct Qdisc * sch)444 static struct sk_buff *sfb_peek(struct Qdisc *sch)
445 {
446 	struct sfb_sched_data *q = qdisc_priv(sch);
447 	struct Qdisc *child = q->qdisc;
448 
449 	return child->ops->peek(child);
450 }
451 
452 /* No sfb_drop -- impossible since the child doesn't return the dropped skb. */
453 
sfb_reset(struct Qdisc * sch)454 static void sfb_reset(struct Qdisc *sch)
455 {
456 	struct sfb_sched_data *q = qdisc_priv(sch);
457 
458 	if (likely(q->qdisc))
459 		qdisc_reset(q->qdisc);
460 	q->slot = 0;
461 	q->double_buffering = false;
462 	sfb_zero_all_buckets(q);
463 	sfb_init_perturbation(0, q);
464 }
465 
sfb_destroy(struct Qdisc * sch)466 static void sfb_destroy(struct Qdisc *sch)
467 {
468 	struct sfb_sched_data *q = qdisc_priv(sch);
469 
470 	tcf_block_put(q->block);
471 	qdisc_put(q->qdisc);
472 }
473 
474 static const struct nla_policy sfb_policy[TCA_SFB_MAX + 1] = {
475 	[TCA_SFB_PARMS]	= { .len = sizeof(struct tc_sfb_qopt) },
476 };
477 
478 static const struct tc_sfb_qopt sfb_default_ops = {
479 	.rehash_interval = 600 * MSEC_PER_SEC,
480 	.warmup_time = 60 * MSEC_PER_SEC,
481 	.limit = 0,
482 	.max = 25,
483 	.bin_size = 20,
484 	.increment = (SFB_MAX_PROB + 500) / 1000, /* 0.1 % */
485 	.decrement = (SFB_MAX_PROB + 3000) / 6000,
486 	.penalty_rate = 10,
487 	.penalty_burst = 20,
488 };
489 
sfb_change(struct Qdisc * sch,struct nlattr * opt,struct netlink_ext_ack * extack)490 static int sfb_change(struct Qdisc *sch, struct nlattr *opt,
491 		      struct netlink_ext_ack *extack)
492 {
493 	struct sfb_sched_data *q = qdisc_priv(sch);
494 	struct Qdisc *child, *old;
495 	struct nlattr *tb[TCA_SFB_MAX + 1];
496 	const struct tc_sfb_qopt *ctl = &sfb_default_ops;
497 	u32 limit;
498 	int err;
499 
500 	if (opt) {
501 		err = nla_parse_nested_deprecated(tb, TCA_SFB_MAX, opt,
502 						  sfb_policy, NULL);
503 		if (err < 0)
504 			return -EINVAL;
505 
506 		if (tb[TCA_SFB_PARMS] == NULL)
507 			return -EINVAL;
508 
509 		ctl = nla_data(tb[TCA_SFB_PARMS]);
510 	}
511 
512 	limit = ctl->limit;
513 	if (limit == 0)
514 		limit = qdisc_dev(sch)->tx_queue_len;
515 
516 	child = fifo_create_dflt(sch, &pfifo_qdisc_ops, limit, extack);
517 	if (IS_ERR(child))
518 		return PTR_ERR(child);
519 
520 	if (child != &noop_qdisc)
521 		qdisc_hash_add(child, true);
522 	sch_tree_lock(sch);
523 
524 	qdisc_purge_queue(q->qdisc);
525 	old = q->qdisc;
526 	q->qdisc = child;
527 
528 	q->rehash_interval = msecs_to_jiffies(ctl->rehash_interval);
529 	q->warmup_time = msecs_to_jiffies(ctl->warmup_time);
530 	q->rehash_time = jiffies;
531 	q->limit = limit;
532 	q->increment = ctl->increment;
533 	q->decrement = ctl->decrement;
534 	q->max = ctl->max;
535 	q->bin_size = ctl->bin_size;
536 	q->penalty_rate = ctl->penalty_rate;
537 	q->penalty_burst = ctl->penalty_burst;
538 	q->tokens_avail = ctl->penalty_burst;
539 	q->token_time = jiffies;
540 
541 	q->slot = 0;
542 	q->double_buffering = false;
543 	sfb_zero_all_buckets(q);
544 	sfb_init_perturbation(0, q);
545 	sfb_init_perturbation(1, q);
546 
547 	sch_tree_unlock(sch);
548 	qdisc_put(old);
549 
550 	return 0;
551 }
552 
sfb_init(struct Qdisc * sch,struct nlattr * opt,struct netlink_ext_ack * extack)553 static int sfb_init(struct Qdisc *sch, struct nlattr *opt,
554 		    struct netlink_ext_ack *extack)
555 {
556 	struct sfb_sched_data *q = qdisc_priv(sch);
557 	int err;
558 
559 	err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
560 	if (err)
561 		return err;
562 
563 	q->qdisc = &noop_qdisc;
564 	return sfb_change(sch, opt, extack);
565 }
566 
sfb_dump(struct Qdisc * sch,struct sk_buff * skb)567 static int sfb_dump(struct Qdisc *sch, struct sk_buff *skb)
568 {
569 	struct sfb_sched_data *q = qdisc_priv(sch);
570 	struct nlattr *opts;
571 	struct tc_sfb_qopt opt = {
572 		.rehash_interval = jiffies_to_msecs(q->rehash_interval),
573 		.warmup_time = jiffies_to_msecs(q->warmup_time),
574 		.limit = q->limit,
575 		.max = q->max,
576 		.bin_size = q->bin_size,
577 		.increment = q->increment,
578 		.decrement = q->decrement,
579 		.penalty_rate = q->penalty_rate,
580 		.penalty_burst = q->penalty_burst,
581 	};
582 
583 	sch->qstats.backlog = q->qdisc->qstats.backlog;
584 	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
585 	if (opts == NULL)
586 		goto nla_put_failure;
587 	if (nla_put(skb, TCA_SFB_PARMS, sizeof(opt), &opt))
588 		goto nla_put_failure;
589 	return nla_nest_end(skb, opts);
590 
591 nla_put_failure:
592 	nla_nest_cancel(skb, opts);
593 	return -EMSGSIZE;
594 }
595 
sfb_dump_stats(struct Qdisc * sch,struct gnet_dump * d)596 static int sfb_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
597 {
598 	struct sfb_sched_data *q = qdisc_priv(sch);
599 	struct tc_sfb_xstats st = {
600 		.earlydrop = q->stats.earlydrop,
601 		.penaltydrop = q->stats.penaltydrop,
602 		.bucketdrop = q->stats.bucketdrop,
603 		.queuedrop = q->stats.queuedrop,
604 		.childdrop = q->stats.childdrop,
605 		.marked = q->stats.marked,
606 	};
607 
608 	st.maxqlen = sfb_compute_qlen(&st.maxprob, &st.avgprob, q);
609 
610 	return gnet_stats_copy_app(d, &st, sizeof(st));
611 }
612 
sfb_dump_class(struct Qdisc * sch,unsigned long cl,struct sk_buff * skb,struct tcmsg * tcm)613 static int sfb_dump_class(struct Qdisc *sch, unsigned long cl,
614 			  struct sk_buff *skb, struct tcmsg *tcm)
615 {
616 	return -ENOSYS;
617 }
618 
sfb_graft(struct Qdisc * sch,unsigned long arg,struct Qdisc * new,struct Qdisc ** old,struct netlink_ext_ack * extack)619 static int sfb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
620 		     struct Qdisc **old, struct netlink_ext_ack *extack)
621 {
622 	struct sfb_sched_data *q = qdisc_priv(sch);
623 
624 	if (new == NULL)
625 		new = &noop_qdisc;
626 
627 	*old = qdisc_replace(sch, new, &q->qdisc);
628 	return 0;
629 }
630 
sfb_leaf(struct Qdisc * sch,unsigned long arg)631 static struct Qdisc *sfb_leaf(struct Qdisc *sch, unsigned long arg)
632 {
633 	struct sfb_sched_data *q = qdisc_priv(sch);
634 
635 	return q->qdisc;
636 }
637 
sfb_find(struct Qdisc * sch,u32 classid)638 static unsigned long sfb_find(struct Qdisc *sch, u32 classid)
639 {
640 	return 1;
641 }
642 
sfb_unbind(struct Qdisc * sch,unsigned long arg)643 static void sfb_unbind(struct Qdisc *sch, unsigned long arg)
644 {
645 }
646 
sfb_change_class(struct Qdisc * sch,u32 classid,u32 parentid,struct nlattr ** tca,unsigned long * arg,struct netlink_ext_ack * extack)647 static int sfb_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
648 			    struct nlattr **tca, unsigned long *arg,
649 			    struct netlink_ext_ack *extack)
650 {
651 	return -ENOSYS;
652 }
653 
sfb_delete(struct Qdisc * sch,unsigned long cl,struct netlink_ext_ack * extack)654 static int sfb_delete(struct Qdisc *sch, unsigned long cl,
655 		      struct netlink_ext_ack *extack)
656 {
657 	return -ENOSYS;
658 }
659 
sfb_walk(struct Qdisc * sch,struct qdisc_walker * walker)660 static void sfb_walk(struct Qdisc *sch, struct qdisc_walker *walker)
661 {
662 	if (!walker->stop) {
663 		tc_qdisc_stats_dump(sch, 1, walker);
664 	}
665 }
666 
sfb_tcf_block(struct Qdisc * sch,unsigned long cl,struct netlink_ext_ack * extack)667 static struct tcf_block *sfb_tcf_block(struct Qdisc *sch, unsigned long cl,
668 				       struct netlink_ext_ack *extack)
669 {
670 	struct sfb_sched_data *q = qdisc_priv(sch);
671 
672 	if (cl)
673 		return NULL;
674 	return q->block;
675 }
676 
sfb_bind(struct Qdisc * sch,unsigned long parent,u32 classid)677 static unsigned long sfb_bind(struct Qdisc *sch, unsigned long parent,
678 			      u32 classid)
679 {
680 	return 0;
681 }
682 
683 
684 static const struct Qdisc_class_ops sfb_class_ops = {
685 	.graft		=	sfb_graft,
686 	.leaf		=	sfb_leaf,
687 	.find		=	sfb_find,
688 	.change		=	sfb_change_class,
689 	.delete		=	sfb_delete,
690 	.walk		=	sfb_walk,
691 	.tcf_block	=	sfb_tcf_block,
692 	.bind_tcf	=	sfb_bind,
693 	.unbind_tcf	=	sfb_unbind,
694 	.dump		=	sfb_dump_class,
695 };
696 
697 static struct Qdisc_ops sfb_qdisc_ops __read_mostly = {
698 	.id		=	"sfb",
699 	.priv_size	=	sizeof(struct sfb_sched_data),
700 	.cl_ops		=	&sfb_class_ops,
701 	.enqueue	=	sfb_enqueue,
702 	.dequeue	=	sfb_dequeue,
703 	.peek		=	sfb_peek,
704 	.init		=	sfb_init,
705 	.reset		=	sfb_reset,
706 	.destroy	=	sfb_destroy,
707 	.change		=	sfb_change,
708 	.dump		=	sfb_dump,
709 	.dump_stats	=	sfb_dump_stats,
710 	.owner		=	THIS_MODULE,
711 };
712 MODULE_ALIAS_NET_SCH("sfb");
713 
sfb_module_init(void)714 static int __init sfb_module_init(void)
715 {
716 	return register_qdisc(&sfb_qdisc_ops);
717 }
718 
sfb_module_exit(void)719 static void __exit sfb_module_exit(void)
720 {
721 	unregister_qdisc(&sfb_qdisc_ops);
722 }
723 
724 module_init(sfb_module_init)
725 module_exit(sfb_module_exit)
726 
727 MODULE_DESCRIPTION("Stochastic Fair Blue queue discipline");
728 MODULE_AUTHOR("Juliusz Chroboczek");
729 MODULE_AUTHOR("Eric Dumazet");
730 MODULE_LICENSE("GPL");
731