]>
Commit | Line | Data |
---|---|---|
e13e02a3 ED |
1 | /* |
2 | * net/sched/sch_sfb.c Stochastic Fair Blue | |
3 | * | |
4 | * Copyright (c) 2008-2011 Juliusz Chroboczek <jch@pps.jussieu.fr> | |
5 | * Copyright (c) 2011 Eric Dumazet <eric.dumazet@gmail.com> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * version 2 as published by the Free Software Foundation. | |
10 | * | |
11 | * W. Feng, D. Kandlur, D. Saha, K. Shin. Blue: | |
12 | * A New Class of Active Queue Management Algorithms. | |
13 | * U. Michigan CSE-TR-387-99, April 1999. | |
14 | * | |
15 | * http://www.thefengs.com/wuchang/blue/CSE-TR-387-99.pdf | |
16 | * | |
17 | */ | |
18 | ||
19 | #include <linux/module.h> | |
20 | #include <linux/types.h> | |
21 | #include <linux/kernel.h> | |
22 | #include <linux/errno.h> | |
23 | #include <linux/skbuff.h> | |
24 | #include <linux/random.h> | |
25 | #include <linux/jhash.h> | |
26 | #include <net/ip.h> | |
27 | #include <net/pkt_sched.h> | |
28 | #include <net/inet_ecn.h> | |
a00bd469 | 29 | #include <net/flow_keys.h> |
e13e02a3 ED |
30 | |
31 | /* | |
32 | * SFB uses two B[l][n] : L x N arrays of bins (L levels, N bins per level) | |
33 | * This implementation uses L = 8 and N = 16 | |
34 | * This permits us to split one 32bit hash (provided per packet by rxhash or | |
35 | * external classifier) into 8 subhashes of 4 bits. | |
36 | */ | |
37 | #define SFB_BUCKET_SHIFT 4 | |
38 | #define SFB_NUMBUCKETS (1 << SFB_BUCKET_SHIFT) /* N bins per Level */ | |
39 | #define SFB_BUCKET_MASK (SFB_NUMBUCKETS - 1) | |
40 | #define SFB_LEVELS (32 / SFB_BUCKET_SHIFT) /* L */ | |
41 | ||
42 | /* SFB algo uses a virtual queue, named "bin" */ | |
43 | struct sfb_bucket { | |
44 | u16 qlen; /* length of virtual queue */ | |
45 | u16 p_mark; /* marking probability */ | |
46 | }; | |
47 | ||
48 | /* We use a double buffering right before hash change | |
49 | * (Section 4.4 of SFB reference : moving hash functions) | |
50 | */ | |
51 | struct sfb_bins { | |
52 | u32 perturbation; /* jhash perturbation */ | |
53 | struct sfb_bucket bins[SFB_LEVELS][SFB_NUMBUCKETS]; | |
54 | }; | |
55 | ||
56 | struct sfb_sched_data { | |
57 | struct Qdisc *qdisc; | |
58 | struct tcf_proto *filter_list; | |
59 | unsigned long rehash_interval; | |
60 | unsigned long warmup_time; /* double buffering warmup time in jiffies */ | |
61 | u32 max; | |
62 | u32 bin_size; /* maximum queue length per bin */ | |
63 | u32 increment; /* d1 */ | |
64 | u32 decrement; /* d2 */ | |
65 | u32 limit; /* HARD maximal queue length */ | |
66 | u32 penalty_rate; | |
67 | u32 penalty_burst; | |
68 | u32 tokens_avail; | |
69 | unsigned long rehash_time; | |
70 | unsigned long token_time; | |
71 | ||
72 | u8 slot; /* current active bins (0 or 1) */ | |
73 | bool double_buffering; | |
74 | struct sfb_bins bins[2]; | |
75 | ||
76 | struct { | |
77 | u32 earlydrop; | |
78 | u32 penaltydrop; | |
79 | u32 bucketdrop; | |
80 | u32 queuedrop; | |
81 | u32 childdrop; /* drops in child qdisc */ | |
82 | u32 marked; /* ECN mark */ | |
83 | } stats; | |
84 | }; | |
85 | ||
86 | /* | |
87 | * Each queued skb might be hashed on one or two bins | |
88 | * We store in skb_cb the two hash values. | |
89 | * (A zero value means double buffering was not used) | |
90 | */ | |
91 | struct sfb_skb_cb { | |
92 | u32 hashes[2]; | |
93 | }; | |
94 | ||
95 | static inline struct sfb_skb_cb *sfb_skb_cb(const struct sk_buff *skb) | |
96 | { | |
97 | BUILD_BUG_ON(sizeof(skb->cb) < | |
98 | sizeof(struct qdisc_skb_cb) + sizeof(struct sfb_skb_cb)); | |
99 | return (struct sfb_skb_cb *)qdisc_skb_cb(skb)->data; | |
100 | } | |
101 | ||
102 | /* | |
103 | * If using 'internal' SFB flow classifier, hash comes from skb rxhash | |
104 | * If using external classifier, hash comes from the classid. | |
105 | */ | |
106 | static u32 sfb_hash(const struct sk_buff *skb, u32 slot) | |
107 | { | |
108 | return sfb_skb_cb(skb)->hashes[slot]; | |
109 | } | |
110 | ||
111 | /* Probabilities are coded as Q0.16 fixed-point values, | |
112 | * with 0xFFFF representing 65535/65536 (almost 1.0) | |
113 | * Addition and subtraction are saturating in [0, 65535] | |
114 | */ | |
115 | static u32 prob_plus(u32 p1, u32 p2) | |
116 | { | |
117 | u32 res = p1 + p2; | |
118 | ||
119 | return min_t(u32, res, SFB_MAX_PROB); | |
120 | } | |
121 | ||
122 | static u32 prob_minus(u32 p1, u32 p2) | |
123 | { | |
124 | return p1 > p2 ? p1 - p2 : 0; | |
125 | } | |
126 | ||
127 | static void increment_one_qlen(u32 sfbhash, u32 slot, struct sfb_sched_data *q) | |
128 | { | |
129 | int i; | |
130 | struct sfb_bucket *b = &q->bins[slot].bins[0][0]; | |
131 | ||
132 | for (i = 0; i < SFB_LEVELS; i++) { | |
133 | u32 hash = sfbhash & SFB_BUCKET_MASK; | |
134 | ||
135 | sfbhash >>= SFB_BUCKET_SHIFT; | |
136 | if (b[hash].qlen < 0xFFFF) | |
137 | b[hash].qlen++; | |
138 | b += SFB_NUMBUCKETS; /* next level */ | |
139 | } | |
140 | } | |
141 | ||
142 | static void increment_qlen(const struct sk_buff *skb, struct sfb_sched_data *q) | |
143 | { | |
144 | u32 sfbhash; | |
145 | ||
146 | sfbhash = sfb_hash(skb, 0); | |
147 | if (sfbhash) | |
148 | increment_one_qlen(sfbhash, 0, q); | |
149 | ||
150 | sfbhash = sfb_hash(skb, 1); | |
151 | if (sfbhash) | |
152 | increment_one_qlen(sfbhash, 1, q); | |
153 | } | |
154 | ||
155 | static void decrement_one_qlen(u32 sfbhash, u32 slot, | |
156 | struct sfb_sched_data *q) | |
157 | { | |
158 | int i; | |
159 | struct sfb_bucket *b = &q->bins[slot].bins[0][0]; | |
160 | ||
161 | for (i = 0; i < SFB_LEVELS; i++) { | |
162 | u32 hash = sfbhash & SFB_BUCKET_MASK; | |
163 | ||
164 | sfbhash >>= SFB_BUCKET_SHIFT; | |
165 | if (b[hash].qlen > 0) | |
166 | b[hash].qlen--; | |
167 | b += SFB_NUMBUCKETS; /* next level */ | |
168 | } | |
169 | } | |
170 | ||
171 | static void decrement_qlen(const struct sk_buff *skb, struct sfb_sched_data *q) | |
172 | { | |
173 | u32 sfbhash; | |
174 | ||
175 | sfbhash = sfb_hash(skb, 0); | |
176 | if (sfbhash) | |
177 | decrement_one_qlen(sfbhash, 0, q); | |
178 | ||
179 | sfbhash = sfb_hash(skb, 1); | |
180 | if (sfbhash) | |
181 | decrement_one_qlen(sfbhash, 1, q); | |
182 | } | |
183 | ||
184 | static void decrement_prob(struct sfb_bucket *b, struct sfb_sched_data *q) | |
185 | { | |
186 | b->p_mark = prob_minus(b->p_mark, q->decrement); | |
187 | } | |
188 | ||
189 | static void increment_prob(struct sfb_bucket *b, struct sfb_sched_data *q) | |
190 | { | |
191 | b->p_mark = prob_plus(b->p_mark, q->increment); | |
192 | } | |
193 | ||
194 | static void sfb_zero_all_buckets(struct sfb_sched_data *q) | |
195 | { | |
196 | memset(&q->bins, 0, sizeof(q->bins)); | |
197 | } | |
198 | ||
199 | /* | |
200 | * compute max qlen, max p_mark, and avg p_mark | |
201 | */ | |
202 | static u32 sfb_compute_qlen(u32 *prob_r, u32 *avgpm_r, const struct sfb_sched_data *q) | |
203 | { | |
204 | int i; | |
205 | u32 qlen = 0, prob = 0, totalpm = 0; | |
206 | const struct sfb_bucket *b = &q->bins[q->slot].bins[0][0]; | |
207 | ||
208 | for (i = 0; i < SFB_LEVELS * SFB_NUMBUCKETS; i++) { | |
209 | if (qlen < b->qlen) | |
210 | qlen = b->qlen; | |
211 | totalpm += b->p_mark; | |
212 | if (prob < b->p_mark) | |
213 | prob = b->p_mark; | |
214 | b++; | |
215 | } | |
216 | *prob_r = prob; | |
217 | *avgpm_r = totalpm / (SFB_LEVELS * SFB_NUMBUCKETS); | |
218 | return qlen; | |
219 | } | |
220 | ||
221 | ||
222 | static void sfb_init_perturbation(u32 slot, struct sfb_sched_data *q) | |
223 | { | |
224 | q->bins[slot].perturbation = net_random(); | |
225 | } | |
226 | ||
227 | static void sfb_swap_slot(struct sfb_sched_data *q) | |
228 | { | |
229 | sfb_init_perturbation(q->slot, q); | |
230 | q->slot ^= 1; | |
231 | q->double_buffering = false; | |
232 | } | |
233 | ||
234 | /* Non elastic flows are allowed to use part of the bandwidth, expressed | |
235 | * in "penalty_rate" packets per second, with "penalty_burst" burst | |
236 | */ | |
237 | static bool sfb_rate_limit(struct sk_buff *skb, struct sfb_sched_data *q) | |
238 | { | |
239 | if (q->penalty_rate == 0 || q->penalty_burst == 0) | |
240 | return true; | |
241 | ||
242 | if (q->tokens_avail < 1) { | |
243 | unsigned long age = min(10UL * HZ, jiffies - q->token_time); | |
244 | ||
245 | q->tokens_avail = (age * q->penalty_rate) / HZ; | |
246 | if (q->tokens_avail > q->penalty_burst) | |
247 | q->tokens_avail = q->penalty_burst; | |
248 | q->token_time = jiffies; | |
249 | if (q->tokens_avail < 1) | |
250 | return true; | |
251 | } | |
252 | ||
253 | q->tokens_avail--; | |
254 | return false; | |
255 | } | |
256 | ||
257 | static bool sfb_classify(struct sk_buff *skb, struct sfb_sched_data *q, | |
258 | int *qerr, u32 *salt) | |
259 | { | |
260 | struct tcf_result res; | |
261 | int result; | |
262 | ||
263 | result = tc_classify(skb, q->filter_list, &res); | |
264 | if (result >= 0) { | |
265 | #ifdef CONFIG_NET_CLS_ACT | |
266 | switch (result) { | |
267 | case TC_ACT_STOLEN: | |
268 | case TC_ACT_QUEUED: | |
269 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; | |
270 | case TC_ACT_SHOT: | |
271 | return false; | |
272 | } | |
273 | #endif | |
274 | *salt = TC_H_MIN(res.classid); | |
275 | return true; | |
276 | } | |
277 | return false; | |
278 | } | |
279 | ||
280 | static int sfb_enqueue(struct sk_buff *skb, struct Qdisc *sch) | |
281 | { | |
282 | ||
283 | struct sfb_sched_data *q = qdisc_priv(sch); | |
284 | struct Qdisc *child = q->qdisc; | |
285 | int i; | |
286 | u32 p_min = ~0; | |
287 | u32 minqlen = ~0; | |
288 | u32 r, slot, salt, sfbhash; | |
289 | int ret = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
a00bd469 | 290 | struct flow_keys keys; |
e13e02a3 | 291 | |
363437f4 ED |
292 | if (unlikely(sch->q.qlen >= q->limit)) { |
293 | sch->qstats.overlimits++; | |
294 | q->stats.queuedrop++; | |
295 | goto drop; | |
296 | } | |
297 | ||
e13e02a3 ED |
298 | if (q->rehash_interval > 0) { |
299 | unsigned long limit = q->rehash_time + q->rehash_interval; | |
300 | ||
301 | if (unlikely(time_after(jiffies, limit))) { | |
302 | sfb_swap_slot(q); | |
303 | q->rehash_time = jiffies; | |
304 | } else if (unlikely(!q->double_buffering && q->warmup_time > 0 && | |
305 | time_after(jiffies, limit - q->warmup_time))) { | |
306 | q->double_buffering = true; | |
307 | } | |
308 | } | |
309 | ||
310 | if (q->filter_list) { | |
311 | /* If using external classifiers, get result and record it. */ | |
312 | if (!sfb_classify(skb, q, &ret, &salt)) | |
313 | goto other_drop; | |
a00bd469 ED |
314 | keys.src = salt; |
315 | keys.dst = 0; | |
316 | keys.ports = 0; | |
e13e02a3 | 317 | } else { |
a00bd469 | 318 | skb_flow_dissect(skb, &keys); |
e13e02a3 ED |
319 | } |
320 | ||
321 | slot = q->slot; | |
322 | ||
a00bd469 ED |
323 | sfbhash = jhash_3words((__force u32)keys.dst, |
324 | (__force u32)keys.src, | |
325 | (__force u32)keys.ports, | |
326 | q->bins[slot].perturbation); | |
e13e02a3 ED |
327 | if (!sfbhash) |
328 | sfbhash = 1; | |
329 | sfb_skb_cb(skb)->hashes[slot] = sfbhash; | |
330 | ||
331 | for (i = 0; i < SFB_LEVELS; i++) { | |
332 | u32 hash = sfbhash & SFB_BUCKET_MASK; | |
333 | struct sfb_bucket *b = &q->bins[slot].bins[i][hash]; | |
334 | ||
335 | sfbhash >>= SFB_BUCKET_SHIFT; | |
336 | if (b->qlen == 0) | |
337 | decrement_prob(b, q); | |
338 | else if (b->qlen >= q->bin_size) | |
339 | increment_prob(b, q); | |
340 | if (minqlen > b->qlen) | |
341 | minqlen = b->qlen; | |
342 | if (p_min > b->p_mark) | |
343 | p_min = b->p_mark; | |
344 | } | |
345 | ||
346 | slot ^= 1; | |
347 | sfb_skb_cb(skb)->hashes[slot] = 0; | |
348 | ||
363437f4 | 349 | if (unlikely(minqlen >= q->max)) { |
e13e02a3 | 350 | sch->qstats.overlimits++; |
363437f4 | 351 | q->stats.bucketdrop++; |
e13e02a3 ED |
352 | goto drop; |
353 | } | |
354 | ||
355 | if (unlikely(p_min >= SFB_MAX_PROB)) { | |
356 | /* Inelastic flow */ | |
357 | if (q->double_buffering) { | |
a00bd469 ED |
358 | sfbhash = jhash_3words((__force u32)keys.dst, |
359 | (__force u32)keys.src, | |
360 | (__force u32)keys.ports, | |
361 | q->bins[slot].perturbation); | |
e13e02a3 ED |
362 | if (!sfbhash) |
363 | sfbhash = 1; | |
364 | sfb_skb_cb(skb)->hashes[slot] = sfbhash; | |
365 | ||
366 | for (i = 0; i < SFB_LEVELS; i++) { | |
367 | u32 hash = sfbhash & SFB_BUCKET_MASK; | |
368 | struct sfb_bucket *b = &q->bins[slot].bins[i][hash]; | |
369 | ||
370 | sfbhash >>= SFB_BUCKET_SHIFT; | |
371 | if (b->qlen == 0) | |
372 | decrement_prob(b, q); | |
373 | else if (b->qlen >= q->bin_size) | |
374 | increment_prob(b, q); | |
375 | } | |
376 | } | |
377 | if (sfb_rate_limit(skb, q)) { | |
378 | sch->qstats.overlimits++; | |
379 | q->stats.penaltydrop++; | |
380 | goto drop; | |
381 | } | |
382 | goto enqueue; | |
383 | } | |
384 | ||
385 | r = net_random() & SFB_MAX_PROB; | |
386 | ||
387 | if (unlikely(r < p_min)) { | |
388 | if (unlikely(p_min > SFB_MAX_PROB / 2)) { | |
389 | /* If we're marking that many packets, then either | |
390 | * this flow is unresponsive, or we're badly congested. | |
391 | * In either case, we want to start dropping packets. | |
392 | */ | |
393 | if (r < (p_min - SFB_MAX_PROB / 2) * 2) { | |
394 | q->stats.earlydrop++; | |
395 | goto drop; | |
396 | } | |
397 | } | |
398 | if (INET_ECN_set_ce(skb)) { | |
399 | q->stats.marked++; | |
400 | } else { | |
401 | q->stats.earlydrop++; | |
402 | goto drop; | |
403 | } | |
404 | } | |
405 | ||
406 | enqueue: | |
407 | ret = qdisc_enqueue(skb, child); | |
408 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
409 | sch->q.qlen++; | |
410 | increment_qlen(skb, q); | |
411 | } else if (net_xmit_drop_count(ret)) { | |
412 | q->stats.childdrop++; | |
413 | sch->qstats.drops++; | |
414 | } | |
415 | return ret; | |
416 | ||
417 | drop: | |
418 | qdisc_drop(skb, sch); | |
419 | return NET_XMIT_CN; | |
420 | other_drop: | |
421 | if (ret & __NET_XMIT_BYPASS) | |
422 | sch->qstats.drops++; | |
423 | kfree_skb(skb); | |
424 | return ret; | |
425 | } | |
426 | ||
427 | static struct sk_buff *sfb_dequeue(struct Qdisc *sch) | |
428 | { | |
429 | struct sfb_sched_data *q = qdisc_priv(sch); | |
430 | struct Qdisc *child = q->qdisc; | |
431 | struct sk_buff *skb; | |
432 | ||
433 | skb = child->dequeue(q->qdisc); | |
434 | ||
435 | if (skb) { | |
436 | qdisc_bstats_update(sch, skb); | |
437 | sch->q.qlen--; | |
438 | decrement_qlen(skb, q); | |
439 | } | |
440 | ||
441 | return skb; | |
442 | } | |
443 | ||
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 | ||
454 | static void sfb_reset(struct Qdisc *sch) | |
455 | { | |
456 | struct sfb_sched_data *q = qdisc_priv(sch); | |
457 | ||
458 | qdisc_reset(q->qdisc); | |
459 | sch->q.qlen = 0; | |
460 | q->slot = 0; | |
461 | q->double_buffering = false; | |
462 | sfb_zero_all_buckets(q); | |
463 | sfb_init_perturbation(0, q); | |
464 | } | |
465 | ||
466 | static void sfb_destroy(struct Qdisc *sch) | |
467 | { | |
468 | struct sfb_sched_data *q = qdisc_priv(sch); | |
469 | ||
470 | tcf_destroy_chain(&q->filter_list); | |
471 | qdisc_destroy(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 | ||
490 | static int sfb_change(struct Qdisc *sch, struct nlattr *opt) | |
491 | { | |
492 | struct sfb_sched_data *q = qdisc_priv(sch); | |
493 | struct Qdisc *child; | |
494 | struct nlattr *tb[TCA_SFB_MAX + 1]; | |
495 | const struct tc_sfb_qopt *ctl = &sfb_default_ops; | |
496 | u32 limit; | |
497 | int err; | |
498 | ||
499 | if (opt) { | |
500 | err = nla_parse_nested(tb, TCA_SFB_MAX, opt, sfb_policy); | |
501 | if (err < 0) | |
502 | return -EINVAL; | |
503 | ||
504 | if (tb[TCA_SFB_PARMS] == NULL) | |
505 | return -EINVAL; | |
506 | ||
507 | ctl = nla_data(tb[TCA_SFB_PARMS]); | |
508 | } | |
509 | ||
510 | limit = ctl->limit; | |
511 | if (limit == 0) | |
512 | limit = max_t(u32, qdisc_dev(sch)->tx_queue_len, 1); | |
513 | ||
514 | child = fifo_create_dflt(sch, &pfifo_qdisc_ops, limit); | |
515 | if (IS_ERR(child)) | |
516 | return PTR_ERR(child); | |
517 | ||
518 | sch_tree_lock(sch); | |
519 | ||
520 | qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen); | |
521 | qdisc_destroy(q->qdisc); | |
522 | q->qdisc = child; | |
523 | ||
524 | q->rehash_interval = msecs_to_jiffies(ctl->rehash_interval); | |
525 | q->warmup_time = msecs_to_jiffies(ctl->warmup_time); | |
526 | q->rehash_time = jiffies; | |
527 | q->limit = limit; | |
528 | q->increment = ctl->increment; | |
529 | q->decrement = ctl->decrement; | |
530 | q->max = ctl->max; | |
531 | q->bin_size = ctl->bin_size; | |
532 | q->penalty_rate = ctl->penalty_rate; | |
533 | q->penalty_burst = ctl->penalty_burst; | |
534 | q->tokens_avail = ctl->penalty_burst; | |
535 | q->token_time = jiffies; | |
536 | ||
537 | q->slot = 0; | |
538 | q->double_buffering = false; | |
539 | sfb_zero_all_buckets(q); | |
540 | sfb_init_perturbation(0, q); | |
541 | sfb_init_perturbation(1, q); | |
542 | ||
543 | sch_tree_unlock(sch); | |
544 | ||
545 | return 0; | |
546 | } | |
547 | ||
548 | static int sfb_init(struct Qdisc *sch, struct nlattr *opt) | |
549 | { | |
550 | struct sfb_sched_data *q = qdisc_priv(sch); | |
551 | ||
552 | q->qdisc = &noop_qdisc; | |
553 | return sfb_change(sch, opt); | |
554 | } | |
555 | ||
556 | static int sfb_dump(struct Qdisc *sch, struct sk_buff *skb) | |
557 | { | |
558 | struct sfb_sched_data *q = qdisc_priv(sch); | |
559 | struct nlattr *opts; | |
560 | struct tc_sfb_qopt opt = { | |
561 | .rehash_interval = jiffies_to_msecs(q->rehash_interval), | |
562 | .warmup_time = jiffies_to_msecs(q->warmup_time), | |
563 | .limit = q->limit, | |
564 | .max = q->max, | |
565 | .bin_size = q->bin_size, | |
566 | .increment = q->increment, | |
567 | .decrement = q->decrement, | |
568 | .penalty_rate = q->penalty_rate, | |
569 | .penalty_burst = q->penalty_burst, | |
570 | }; | |
571 | ||
572 | sch->qstats.backlog = q->qdisc->qstats.backlog; | |
573 | opts = nla_nest_start(skb, TCA_OPTIONS); | |
574 | NLA_PUT(skb, TCA_SFB_PARMS, sizeof(opt), &opt); | |
575 | return nla_nest_end(skb, opts); | |
576 | ||
577 | nla_put_failure: | |
578 | nla_nest_cancel(skb, opts); | |
579 | return -EMSGSIZE; | |
580 | } | |
581 | ||
582 | static int sfb_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
583 | { | |
584 | struct sfb_sched_data *q = qdisc_priv(sch); | |
585 | struct tc_sfb_xstats st = { | |
586 | .earlydrop = q->stats.earlydrop, | |
587 | .penaltydrop = q->stats.penaltydrop, | |
588 | .bucketdrop = q->stats.bucketdrop, | |
589 | .queuedrop = q->stats.queuedrop, | |
590 | .childdrop = q->stats.childdrop, | |
591 | .marked = q->stats.marked, | |
592 | }; | |
593 | ||
594 | st.maxqlen = sfb_compute_qlen(&st.maxprob, &st.avgprob, q); | |
595 | ||
596 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
597 | } | |
598 | ||
599 | static int sfb_dump_class(struct Qdisc *sch, unsigned long cl, | |
600 | struct sk_buff *skb, struct tcmsg *tcm) | |
601 | { | |
602 | return -ENOSYS; | |
603 | } | |
604 | ||
605 | static int sfb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
606 | struct Qdisc **old) | |
607 | { | |
608 | struct sfb_sched_data *q = qdisc_priv(sch); | |
609 | ||
610 | if (new == NULL) | |
611 | new = &noop_qdisc; | |
612 | ||
613 | sch_tree_lock(sch); | |
614 | *old = q->qdisc; | |
615 | q->qdisc = new; | |
616 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); | |
617 | qdisc_reset(*old); | |
618 | sch_tree_unlock(sch); | |
619 | return 0; | |
620 | } | |
621 | ||
622 | static struct Qdisc *sfb_leaf(struct Qdisc *sch, unsigned long arg) | |
623 | { | |
624 | struct sfb_sched_data *q = qdisc_priv(sch); | |
625 | ||
626 | return q->qdisc; | |
627 | } | |
628 | ||
629 | static unsigned long sfb_get(struct Qdisc *sch, u32 classid) | |
630 | { | |
631 | return 1; | |
632 | } | |
633 | ||
634 | static void sfb_put(struct Qdisc *sch, unsigned long arg) | |
635 | { | |
636 | } | |
637 | ||
638 | static int sfb_change_class(struct Qdisc *sch, u32 classid, u32 parentid, | |
639 | struct nlattr **tca, unsigned long *arg) | |
640 | { | |
641 | return -ENOSYS; | |
642 | } | |
643 | ||
644 | static int sfb_delete(struct Qdisc *sch, unsigned long cl) | |
645 | { | |
646 | return -ENOSYS; | |
647 | } | |
648 | ||
649 | static void sfb_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
650 | { | |
651 | if (!walker->stop) { | |
652 | if (walker->count >= walker->skip) | |
653 | if (walker->fn(sch, 1, walker) < 0) { | |
654 | walker->stop = 1; | |
655 | return; | |
656 | } | |
657 | walker->count++; | |
658 | } | |
659 | } | |
660 | ||
661 | static struct tcf_proto **sfb_find_tcf(struct Qdisc *sch, unsigned long cl) | |
662 | { | |
663 | struct sfb_sched_data *q = qdisc_priv(sch); | |
664 | ||
665 | if (cl) | |
666 | return NULL; | |
667 | return &q->filter_list; | |
668 | } | |
669 | ||
670 | static unsigned long sfb_bind(struct Qdisc *sch, unsigned long parent, | |
671 | u32 classid) | |
672 | { | |
673 | return 0; | |
674 | } | |
675 | ||
676 | ||
677 | static const struct Qdisc_class_ops sfb_class_ops = { | |
678 | .graft = sfb_graft, | |
679 | .leaf = sfb_leaf, | |
680 | .get = sfb_get, | |
681 | .put = sfb_put, | |
682 | .change = sfb_change_class, | |
683 | .delete = sfb_delete, | |
684 | .walk = sfb_walk, | |
685 | .tcf_chain = sfb_find_tcf, | |
686 | .bind_tcf = sfb_bind, | |
687 | .unbind_tcf = sfb_put, | |
688 | .dump = sfb_dump_class, | |
689 | }; | |
690 | ||
691 | static struct Qdisc_ops sfb_qdisc_ops __read_mostly = { | |
692 | .id = "sfb", | |
693 | .priv_size = sizeof(struct sfb_sched_data), | |
694 | .cl_ops = &sfb_class_ops, | |
695 | .enqueue = sfb_enqueue, | |
696 | .dequeue = sfb_dequeue, | |
697 | .peek = sfb_peek, | |
698 | .init = sfb_init, | |
699 | .reset = sfb_reset, | |
700 | .destroy = sfb_destroy, | |
701 | .change = sfb_change, | |
702 | .dump = sfb_dump, | |
703 | .dump_stats = sfb_dump_stats, | |
704 | .owner = THIS_MODULE, | |
705 | }; | |
706 | ||
707 | static int __init sfb_module_init(void) | |
708 | { | |
709 | return register_qdisc(&sfb_qdisc_ops); | |
710 | } | |
711 | ||
712 | static void __exit sfb_module_exit(void) | |
713 | { | |
714 | unregister_qdisc(&sfb_qdisc_ops); | |
715 | } | |
716 | ||
717 | module_init(sfb_module_init) | |
718 | module_exit(sfb_module_exit) | |
719 | ||
720 | MODULE_DESCRIPTION("Stochastic Fair Blue queue discipline"); | |
721 | MODULE_AUTHOR("Juliusz Chroboczek"); | |
722 | MODULE_AUTHOR("Eric Dumazet"); | |
723 | MODULE_LICENSE("GPL"); |