Commit | Line | Data |
---|---|---|
13d2a1d2 PM |
1 | /* |
2 | * net/sched/sch_drr.c Deficit Round Robin scheduler | |
3 | * | |
4 | * Copyright (c) 2008 Patrick McHardy <kaber@trash.net> | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * version 2 as published by the Free Software Foundation. | |
9 | */ | |
10 | ||
11 | #include <linux/module.h> | |
12 | #include <linux/init.h> | |
13 | #include <linux/errno.h> | |
14 | #include <linux/netdevice.h> | |
15 | #include <linux/pkt_sched.h> | |
16 | #include <net/sch_generic.h> | |
17 | #include <net/pkt_sched.h> | |
18 | #include <net/pkt_cls.h> | |
19 | ||
20 | struct drr_class { | |
21 | struct Qdisc_class_common common; | |
22 | unsigned int refcnt; | |
23 | unsigned int filter_cnt; | |
24 | ||
25 | struct gnet_stats_basic bstats; | |
26 | struct gnet_stats_queue qstats; | |
27 | struct gnet_stats_rate_est rate_est; | |
28 | struct list_head alist; | |
29 | struct Qdisc *qdisc; | |
30 | ||
31 | u32 quantum; | |
32 | u32 deficit; | |
33 | }; | |
34 | ||
35 | struct drr_sched { | |
36 | struct list_head active; | |
37 | struct tcf_proto *filter_list; | |
38 | struct Qdisc_class_hash clhash; | |
39 | }; | |
40 | ||
41 | static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid) | |
42 | { | |
43 | struct drr_sched *q = qdisc_priv(sch); | |
44 | struct Qdisc_class_common *clc; | |
45 | ||
46 | clc = qdisc_class_find(&q->clhash, classid); | |
47 | if (clc == NULL) | |
48 | return NULL; | |
49 | return container_of(clc, struct drr_class, common); | |
50 | } | |
51 | ||
52 | static void drr_purge_queue(struct drr_class *cl) | |
53 | { | |
54 | unsigned int len = cl->qdisc->q.qlen; | |
55 | ||
56 | qdisc_reset(cl->qdisc); | |
57 | qdisc_tree_decrease_qlen(cl->qdisc, len); | |
58 | } | |
59 | ||
60 | static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = { | |
61 | [TCA_DRR_QUANTUM] = { .type = NLA_U32 }, | |
62 | }; | |
63 | ||
64 | static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid, | |
65 | struct nlattr **tca, unsigned long *arg) | |
66 | { | |
67 | struct drr_sched *q = qdisc_priv(sch); | |
68 | struct drr_class *cl = (struct drr_class *)*arg; | |
1844f747 | 69 | struct nlattr *opt = tca[TCA_OPTIONS]; |
13d2a1d2 PM |
70 | struct nlattr *tb[TCA_DRR_MAX + 1]; |
71 | u32 quantum; | |
72 | int err; | |
73 | ||
1844f747 JP |
74 | if (!opt) |
75 | return -EINVAL; | |
76 | ||
77 | err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy); | |
13d2a1d2 PM |
78 | if (err < 0) |
79 | return err; | |
80 | ||
81 | if (tb[TCA_DRR_QUANTUM]) { | |
82 | quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]); | |
83 | if (quantum == 0) | |
84 | return -EINVAL; | |
85 | } else | |
86 | quantum = psched_mtu(qdisc_dev(sch)); | |
87 | ||
88 | if (cl != NULL) { | |
71bcb09a SH |
89 | if (tca[TCA_RATE]) { |
90 | err = gen_replace_estimator(&cl->bstats, &cl->rate_est, | |
91 | qdisc_root_sleeping_lock(sch), | |
92 | tca[TCA_RATE]); | |
93 | if (err) | |
94 | return err; | |
95 | } | |
96 | ||
13d2a1d2 PM |
97 | sch_tree_lock(sch); |
98 | if (tb[TCA_DRR_QUANTUM]) | |
99 | cl->quantum = quantum; | |
100 | sch_tree_unlock(sch); | |
101 | ||
13d2a1d2 PM |
102 | return 0; |
103 | } | |
104 | ||
105 | cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL); | |
106 | if (cl == NULL) | |
107 | return -ENOBUFS; | |
108 | ||
109 | cl->refcnt = 1; | |
110 | cl->common.classid = classid; | |
111 | cl->quantum = quantum; | |
112 | cl->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, | |
113 | &pfifo_qdisc_ops, classid); | |
114 | if (cl->qdisc == NULL) | |
115 | cl->qdisc = &noop_qdisc; | |
116 | ||
71bcb09a SH |
117 | if (tca[TCA_RATE]) { |
118 | err = gen_replace_estimator(&cl->bstats, &cl->rate_est, | |
119 | qdisc_root_sleeping_lock(sch), | |
120 | tca[TCA_RATE]); | |
121 | if (err) { | |
122 | qdisc_destroy(cl->qdisc); | |
123 | kfree(cl); | |
124 | return err; | |
125 | } | |
126 | } | |
13d2a1d2 PM |
127 | |
128 | sch_tree_lock(sch); | |
129 | qdisc_class_hash_insert(&q->clhash, &cl->common); | |
130 | sch_tree_unlock(sch); | |
131 | ||
132 | qdisc_class_hash_grow(sch, &q->clhash); | |
133 | ||
134 | *arg = (unsigned long)cl; | |
135 | return 0; | |
136 | } | |
137 | ||
138 | static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl) | |
139 | { | |
140 | gen_kill_estimator(&cl->bstats, &cl->rate_est); | |
141 | qdisc_destroy(cl->qdisc); | |
142 | kfree(cl); | |
143 | } | |
144 | ||
145 | static int drr_delete_class(struct Qdisc *sch, unsigned long arg) | |
146 | { | |
147 | struct drr_sched *q = qdisc_priv(sch); | |
148 | struct drr_class *cl = (struct drr_class *)arg; | |
149 | ||
150 | if (cl->filter_cnt > 0) | |
151 | return -EBUSY; | |
152 | ||
153 | sch_tree_lock(sch); | |
154 | ||
155 | drr_purge_queue(cl); | |
156 | qdisc_class_hash_remove(&q->clhash, &cl->common); | |
157 | ||
158 | if (--cl->refcnt == 0) | |
159 | drr_destroy_class(sch, cl); | |
160 | ||
161 | sch_tree_unlock(sch); | |
162 | return 0; | |
163 | } | |
164 | ||
165 | static unsigned long drr_get_class(struct Qdisc *sch, u32 classid) | |
166 | { | |
167 | struct drr_class *cl = drr_find_class(sch, classid); | |
168 | ||
169 | if (cl != NULL) | |
170 | cl->refcnt++; | |
171 | ||
172 | return (unsigned long)cl; | |
173 | } | |
174 | ||
175 | static void drr_put_class(struct Qdisc *sch, unsigned long arg) | |
176 | { | |
177 | struct drr_class *cl = (struct drr_class *)arg; | |
178 | ||
179 | if (--cl->refcnt == 0) | |
180 | drr_destroy_class(sch, cl); | |
181 | } | |
182 | ||
183 | static struct tcf_proto **drr_tcf_chain(struct Qdisc *sch, unsigned long cl) | |
184 | { | |
185 | struct drr_sched *q = qdisc_priv(sch); | |
186 | ||
187 | if (cl) | |
188 | return NULL; | |
189 | ||
190 | return &q->filter_list; | |
191 | } | |
192 | ||
193 | static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent, | |
194 | u32 classid) | |
195 | { | |
196 | struct drr_class *cl = drr_find_class(sch, classid); | |
197 | ||
198 | if (cl != NULL) | |
199 | cl->filter_cnt++; | |
200 | ||
201 | return (unsigned long)cl; | |
202 | } | |
203 | ||
204 | static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg) | |
205 | { | |
206 | struct drr_class *cl = (struct drr_class *)arg; | |
207 | ||
208 | cl->filter_cnt--; | |
209 | } | |
210 | ||
211 | static int drr_graft_class(struct Qdisc *sch, unsigned long arg, | |
212 | struct Qdisc *new, struct Qdisc **old) | |
213 | { | |
214 | struct drr_class *cl = (struct drr_class *)arg; | |
215 | ||
216 | if (new == NULL) { | |
217 | new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, | |
218 | &pfifo_qdisc_ops, cl->common.classid); | |
219 | if (new == NULL) | |
220 | new = &noop_qdisc; | |
221 | } | |
222 | ||
223 | sch_tree_lock(sch); | |
224 | drr_purge_queue(cl); | |
b94c8afc PM |
225 | *old = cl->qdisc; |
226 | cl->qdisc = new; | |
13d2a1d2 PM |
227 | sch_tree_unlock(sch); |
228 | return 0; | |
229 | } | |
230 | ||
231 | static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg) | |
232 | { | |
233 | struct drr_class *cl = (struct drr_class *)arg; | |
234 | ||
235 | return cl->qdisc; | |
236 | } | |
237 | ||
238 | static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg) | |
239 | { | |
240 | struct drr_class *cl = (struct drr_class *)arg; | |
241 | ||
242 | if (cl->qdisc->q.qlen == 0) | |
243 | list_del(&cl->alist); | |
244 | } | |
245 | ||
246 | static int drr_dump_class(struct Qdisc *sch, unsigned long arg, | |
247 | struct sk_buff *skb, struct tcmsg *tcm) | |
248 | { | |
249 | struct drr_class *cl = (struct drr_class *)arg; | |
250 | struct nlattr *nest; | |
251 | ||
252 | tcm->tcm_parent = TC_H_ROOT; | |
253 | tcm->tcm_handle = cl->common.classid; | |
254 | tcm->tcm_info = cl->qdisc->handle; | |
255 | ||
256 | nest = nla_nest_start(skb, TCA_OPTIONS); | |
257 | if (nest == NULL) | |
258 | goto nla_put_failure; | |
259 | NLA_PUT_U32(skb, TCA_DRR_QUANTUM, cl->quantum); | |
260 | return nla_nest_end(skb, nest); | |
261 | ||
262 | nla_put_failure: | |
263 | nla_nest_cancel(skb, nest); | |
264 | return -EMSGSIZE; | |
265 | } | |
266 | ||
267 | static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg, | |
268 | struct gnet_dump *d) | |
269 | { | |
270 | struct drr_class *cl = (struct drr_class *)arg; | |
271 | struct tc_drr_stats xstats; | |
272 | ||
273 | memset(&xstats, 0, sizeof(xstats)); | |
274 | if (cl->qdisc->q.qlen) | |
275 | xstats.deficit = cl->deficit; | |
276 | ||
277 | if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || | |
278 | gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || | |
279 | gnet_stats_copy_queue(d, &cl->qdisc->qstats) < 0) | |
280 | return -1; | |
281 | ||
282 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
283 | } | |
284 | ||
285 | static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
286 | { | |
287 | struct drr_sched *q = qdisc_priv(sch); | |
288 | struct drr_class *cl; | |
289 | struct hlist_node *n; | |
290 | unsigned int i; | |
291 | ||
292 | if (arg->stop) | |
293 | return; | |
294 | ||
295 | for (i = 0; i < q->clhash.hashsize; i++) { | |
296 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { | |
297 | if (arg->count < arg->skip) { | |
298 | arg->count++; | |
299 | continue; | |
300 | } | |
301 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
302 | arg->stop = 1; | |
303 | return; | |
304 | } | |
305 | arg->count++; | |
306 | } | |
307 | } | |
308 | } | |
309 | ||
310 | static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch, | |
311 | int *qerr) | |
312 | { | |
313 | struct drr_sched *q = qdisc_priv(sch); | |
314 | struct drr_class *cl; | |
315 | struct tcf_result res; | |
316 | int result; | |
317 | ||
318 | if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) { | |
319 | cl = drr_find_class(sch, skb->priority); | |
320 | if (cl != NULL) | |
321 | return cl; | |
322 | } | |
323 | ||
324 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
325 | result = tc_classify(skb, q->filter_list, &res); | |
326 | if (result >= 0) { | |
327 | #ifdef CONFIG_NET_CLS_ACT | |
328 | switch (result) { | |
329 | case TC_ACT_QUEUED: | |
330 | case TC_ACT_STOLEN: | |
331 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; | |
332 | case TC_ACT_SHOT: | |
333 | return NULL; | |
334 | } | |
335 | #endif | |
336 | cl = (struct drr_class *)res.class; | |
337 | if (cl == NULL) | |
338 | cl = drr_find_class(sch, res.classid); | |
339 | return cl; | |
340 | } | |
341 | return NULL; | |
342 | } | |
343 | ||
344 | static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch) | |
345 | { | |
346 | struct drr_sched *q = qdisc_priv(sch); | |
347 | struct drr_class *cl; | |
348 | unsigned int len; | |
349 | int err; | |
350 | ||
351 | cl = drr_classify(skb, sch, &err); | |
352 | if (cl == NULL) { | |
353 | if (err & __NET_XMIT_BYPASS) | |
354 | sch->qstats.drops++; | |
355 | kfree_skb(skb); | |
356 | return err; | |
357 | } | |
358 | ||
359 | len = qdisc_pkt_len(skb); | |
360 | err = qdisc_enqueue(skb, cl->qdisc); | |
361 | if (unlikely(err != NET_XMIT_SUCCESS)) { | |
362 | if (net_xmit_drop_count(err)) { | |
363 | cl->qstats.drops++; | |
364 | sch->qstats.drops++; | |
365 | } | |
366 | return err; | |
367 | } | |
368 | ||
369 | if (cl->qdisc->q.qlen == 1) { | |
370 | list_add_tail(&cl->alist, &q->active); | |
371 | cl->deficit = cl->quantum; | |
372 | } | |
373 | ||
374 | cl->bstats.packets++; | |
375 | cl->bstats.bytes += len; | |
376 | sch->bstats.packets++; | |
377 | sch->bstats.bytes += len; | |
378 | ||
379 | sch->q.qlen++; | |
380 | return err; | |
381 | } | |
382 | ||
383 | static struct sk_buff *drr_dequeue(struct Qdisc *sch) | |
384 | { | |
385 | struct drr_sched *q = qdisc_priv(sch); | |
386 | struct drr_class *cl; | |
387 | struct sk_buff *skb; | |
388 | unsigned int len; | |
389 | ||
3f0947c3 PM |
390 | if (list_empty(&q->active)) |
391 | goto out; | |
392 | while (1) { | |
13d2a1d2 PM |
393 | cl = list_first_entry(&q->active, struct drr_class, alist); |
394 | skb = cl->qdisc->ops->peek(cl->qdisc); | |
395 | if (skb == NULL) | |
3f0947c3 | 396 | goto out; |
13d2a1d2 PM |
397 | |
398 | len = qdisc_pkt_len(skb); | |
399 | if (len <= cl->deficit) { | |
400 | cl->deficit -= len; | |
401 | skb = qdisc_dequeue_peeked(cl->qdisc); | |
402 | if (cl->qdisc->q.qlen == 0) | |
403 | list_del(&cl->alist); | |
404 | sch->q.qlen--; | |
405 | return skb; | |
406 | } | |
407 | ||
408 | cl->deficit += cl->quantum; | |
13d2a1d2 PM |
409 | list_move_tail(&cl->alist, &q->active); |
410 | } | |
3f0947c3 | 411 | out: |
13d2a1d2 PM |
412 | return NULL; |
413 | } | |
414 | ||
415 | static unsigned int drr_drop(struct Qdisc *sch) | |
416 | { | |
417 | struct drr_sched *q = qdisc_priv(sch); | |
418 | struct drr_class *cl; | |
419 | unsigned int len; | |
420 | ||
421 | list_for_each_entry(cl, &q->active, alist) { | |
422 | if (cl->qdisc->ops->drop) { | |
423 | len = cl->qdisc->ops->drop(cl->qdisc); | |
424 | if (len > 0) { | |
98aa9c80 | 425 | sch->q.qlen--; |
13d2a1d2 PM |
426 | if (cl->qdisc->q.qlen == 0) |
427 | list_del(&cl->alist); | |
428 | return len; | |
429 | } | |
430 | } | |
431 | } | |
432 | return 0; | |
433 | } | |
434 | ||
435 | static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt) | |
436 | { | |
437 | struct drr_sched *q = qdisc_priv(sch); | |
438 | int err; | |
439 | ||
440 | err = qdisc_class_hash_init(&q->clhash); | |
441 | if (err < 0) | |
442 | return err; | |
443 | INIT_LIST_HEAD(&q->active); | |
444 | return 0; | |
445 | } | |
446 | ||
447 | static void drr_reset_qdisc(struct Qdisc *sch) | |
448 | { | |
449 | struct drr_sched *q = qdisc_priv(sch); | |
450 | struct drr_class *cl; | |
451 | struct hlist_node *n; | |
452 | unsigned int i; | |
453 | ||
454 | for (i = 0; i < q->clhash.hashsize; i++) { | |
455 | hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) { | |
456 | if (cl->qdisc->q.qlen) | |
457 | list_del(&cl->alist); | |
458 | qdisc_reset(cl->qdisc); | |
459 | } | |
460 | } | |
461 | sch->q.qlen = 0; | |
462 | } | |
463 | ||
464 | static void drr_destroy_qdisc(struct Qdisc *sch) | |
465 | { | |
466 | struct drr_sched *q = qdisc_priv(sch); | |
467 | struct drr_class *cl; | |
468 | struct hlist_node *n, *next; | |
469 | unsigned int i; | |
470 | ||
471 | tcf_destroy_chain(&q->filter_list); | |
472 | ||
473 | for (i = 0; i < q->clhash.hashsize; i++) { | |
474 | hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i], | |
475 | common.hnode) | |
476 | drr_destroy_class(sch, cl); | |
477 | } | |
478 | qdisc_class_hash_destroy(&q->clhash); | |
479 | } | |
480 | ||
481 | static const struct Qdisc_class_ops drr_class_ops = { | |
482 | .change = drr_change_class, | |
483 | .delete = drr_delete_class, | |
484 | .get = drr_get_class, | |
485 | .put = drr_put_class, | |
486 | .tcf_chain = drr_tcf_chain, | |
487 | .bind_tcf = drr_bind_tcf, | |
488 | .unbind_tcf = drr_unbind_tcf, | |
489 | .graft = drr_graft_class, | |
490 | .leaf = drr_class_leaf, | |
491 | .qlen_notify = drr_qlen_notify, | |
492 | .dump = drr_dump_class, | |
493 | .dump_stats = drr_dump_class_stats, | |
494 | .walk = drr_walk, | |
495 | }; | |
496 | ||
497 | static struct Qdisc_ops drr_qdisc_ops __read_mostly = { | |
498 | .cl_ops = &drr_class_ops, | |
499 | .id = "drr", | |
500 | .priv_size = sizeof(struct drr_sched), | |
501 | .enqueue = drr_enqueue, | |
502 | .dequeue = drr_dequeue, | |
503 | .peek = qdisc_peek_dequeued, | |
504 | .drop = drr_drop, | |
505 | .init = drr_init_qdisc, | |
506 | .reset = drr_reset_qdisc, | |
507 | .destroy = drr_destroy_qdisc, | |
508 | .owner = THIS_MODULE, | |
509 | }; | |
510 | ||
511 | static int __init drr_init(void) | |
512 | { | |
513 | return register_qdisc(&drr_qdisc_ops); | |
514 | } | |
515 | ||
516 | static void __exit drr_exit(void) | |
517 | { | |
518 | unregister_qdisc(&drr_qdisc_ops); | |
519 | } | |
520 | ||
521 | module_init(drr_init); | |
522 | module_exit(drr_exit); | |
523 | MODULE_LICENSE("GPL"); |