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> | |
5a0e3ad6 | 12 | #include <linux/slab.h> |
13d2a1d2 PM |
13 | #include <linux/init.h> |
14 | #include <linux/errno.h> | |
15 | #include <linux/netdevice.h> | |
16 | #include <linux/pkt_sched.h> | |
17 | #include <net/sch_generic.h> | |
18 | #include <net/pkt_sched.h> | |
19 | #include <net/pkt_cls.h> | |
20 | ||
21 | struct drr_class { | |
22 | struct Qdisc_class_common common; | |
13d2a1d2 PM |
23 | unsigned int filter_cnt; |
24 | ||
c1a8f1f1 | 25 | struct gnet_stats_basic_packed bstats; |
13d2a1d2 | 26 | struct gnet_stats_queue qstats; |
1c0d32fd | 27 | struct net_rate_estimator __rcu *rate_est; |
13d2a1d2 PM |
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; | |
25d8c0d5 | 37 | struct tcf_proto __rcu *filter_list; |
6529eaba | 38 | struct tcf_block *block; |
13d2a1d2 PM |
39 | struct Qdisc_class_hash clhash; |
40 | }; | |
41 | ||
42 | static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid) | |
43 | { | |
44 | struct drr_sched *q = qdisc_priv(sch); | |
45 | struct Qdisc_class_common *clc; | |
46 | ||
47 | clc = qdisc_class_find(&q->clhash, classid); | |
48 | if (clc == NULL) | |
49 | return NULL; | |
50 | return container_of(clc, struct drr_class, common); | |
51 | } | |
52 | ||
53 | static void drr_purge_queue(struct drr_class *cl) | |
54 | { | |
55 | unsigned int len = cl->qdisc->q.qlen; | |
2ccccf5f | 56 | unsigned int backlog = cl->qdisc->qstats.backlog; |
13d2a1d2 PM |
57 | |
58 | qdisc_reset(cl->qdisc); | |
2ccccf5f | 59 | qdisc_tree_reduce_backlog(cl->qdisc, len, backlog); |
13d2a1d2 PM |
60 | } |
61 | ||
62 | static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = { | |
63 | [TCA_DRR_QUANTUM] = { .type = NLA_U32 }, | |
64 | }; | |
65 | ||
66 | static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid, | |
793d81d6 AA |
67 | struct nlattr **tca, unsigned long *arg, |
68 | struct netlink_ext_ack *extack) | |
13d2a1d2 PM |
69 | { |
70 | struct drr_sched *q = qdisc_priv(sch); | |
71 | struct drr_class *cl = (struct drr_class *)*arg; | |
1844f747 | 72 | struct nlattr *opt = tca[TCA_OPTIONS]; |
13d2a1d2 PM |
73 | struct nlattr *tb[TCA_DRR_MAX + 1]; |
74 | u32 quantum; | |
75 | int err; | |
76 | ||
a7c31693 AA |
77 | if (!opt) { |
78 | NL_SET_ERR_MSG(extack, "DRR options are required for this operation"); | |
1844f747 | 79 | return -EINVAL; |
a7c31693 | 80 | } |
1844f747 | 81 | |
a7c31693 | 82 | err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy, extack); |
13d2a1d2 PM |
83 | if (err < 0) |
84 | return err; | |
85 | ||
86 | if (tb[TCA_DRR_QUANTUM]) { | |
87 | quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]); | |
a7c31693 AA |
88 | if (quantum == 0) { |
89 | NL_SET_ERR_MSG(extack, "Specified DRR quantum cannot be zero"); | |
13d2a1d2 | 90 | return -EINVAL; |
a7c31693 | 91 | } |
13d2a1d2 PM |
92 | } else |
93 | quantum = psched_mtu(qdisc_dev(sch)); | |
94 | ||
95 | if (cl != NULL) { | |
71bcb09a | 96 | if (tca[TCA_RATE]) { |
22e0f8b9 JF |
97 | err = gen_replace_estimator(&cl->bstats, NULL, |
98 | &cl->rate_est, | |
edb09eb1 ED |
99 | NULL, |
100 | qdisc_root_sleeping_running(sch), | |
71bcb09a | 101 | tca[TCA_RATE]); |
a7c31693 AA |
102 | if (err) { |
103 | NL_SET_ERR_MSG(extack, "Failed to replace estimator"); | |
71bcb09a | 104 | return err; |
a7c31693 | 105 | } |
71bcb09a SH |
106 | } |
107 | ||
13d2a1d2 PM |
108 | sch_tree_lock(sch); |
109 | if (tb[TCA_DRR_QUANTUM]) | |
110 | cl->quantum = quantum; | |
111 | sch_tree_unlock(sch); | |
112 | ||
13d2a1d2 PM |
113 | return 0; |
114 | } | |
115 | ||
116 | cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL); | |
117 | if (cl == NULL) | |
118 | return -ENOBUFS; | |
119 | ||
13d2a1d2 PM |
120 | cl->common.classid = classid; |
121 | cl->quantum = quantum; | |
3511c913 | 122 | cl->qdisc = qdisc_create_dflt(sch->dev_queue, |
a38a9882 AA |
123 | &pfifo_qdisc_ops, classid, |
124 | NULL); | |
13d2a1d2 PM |
125 | if (cl->qdisc == NULL) |
126 | cl->qdisc = &noop_qdisc; | |
49b49971 JK |
127 | else |
128 | qdisc_hash_add(cl->qdisc, true); | |
13d2a1d2 | 129 | |
71bcb09a | 130 | if (tca[TCA_RATE]) { |
22e0f8b9 | 131 | err = gen_replace_estimator(&cl->bstats, NULL, &cl->rate_est, |
edb09eb1 ED |
132 | NULL, |
133 | qdisc_root_sleeping_running(sch), | |
71bcb09a SH |
134 | tca[TCA_RATE]); |
135 | if (err) { | |
a7c31693 | 136 | NL_SET_ERR_MSG(extack, "Failed to replace estimator"); |
86bd446b | 137 | qdisc_put(cl->qdisc); |
71bcb09a SH |
138 | kfree(cl); |
139 | return err; | |
140 | } | |
141 | } | |
13d2a1d2 PM |
142 | |
143 | sch_tree_lock(sch); | |
144 | qdisc_class_hash_insert(&q->clhash, &cl->common); | |
145 | sch_tree_unlock(sch); | |
146 | ||
147 | qdisc_class_hash_grow(sch, &q->clhash); | |
148 | ||
149 | *arg = (unsigned long)cl; | |
150 | return 0; | |
151 | } | |
152 | ||
153 | static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl) | |
154 | { | |
1c0d32fd | 155 | gen_kill_estimator(&cl->rate_est); |
86bd446b | 156 | qdisc_put(cl->qdisc); |
13d2a1d2 PM |
157 | kfree(cl); |
158 | } | |
159 | ||
160 | static int drr_delete_class(struct Qdisc *sch, unsigned long arg) | |
161 | { | |
162 | struct drr_sched *q = qdisc_priv(sch); | |
163 | struct drr_class *cl = (struct drr_class *)arg; | |
164 | ||
165 | if (cl->filter_cnt > 0) | |
166 | return -EBUSY; | |
167 | ||
168 | sch_tree_lock(sch); | |
169 | ||
170 | drr_purge_queue(cl); | |
171 | qdisc_class_hash_remove(&q->clhash, &cl->common); | |
172 | ||
13d2a1d2 | 173 | sch_tree_unlock(sch); |
13d2a1d2 | 174 | |
143976ce WC |
175 | drr_destroy_class(sch, cl); |
176 | return 0; | |
13d2a1d2 PM |
177 | } |
178 | ||
143976ce | 179 | static unsigned long drr_search_class(struct Qdisc *sch, u32 classid) |
13d2a1d2 | 180 | { |
143976ce | 181 | return (unsigned long)drr_find_class(sch, classid); |
13d2a1d2 PM |
182 | } |
183 | ||
cbaacc4e AA |
184 | static struct tcf_block *drr_tcf_block(struct Qdisc *sch, unsigned long cl, |
185 | struct netlink_ext_ack *extack) | |
13d2a1d2 PM |
186 | { |
187 | struct drr_sched *q = qdisc_priv(sch); | |
188 | ||
a7c31693 AA |
189 | if (cl) { |
190 | NL_SET_ERR_MSG(extack, "DRR classid must be zero"); | |
13d2a1d2 | 191 | return NULL; |
a7c31693 | 192 | } |
13d2a1d2 | 193 | |
6529eaba | 194 | return q->block; |
13d2a1d2 PM |
195 | } |
196 | ||
197 | static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent, | |
198 | u32 classid) | |
199 | { | |
200 | struct drr_class *cl = drr_find_class(sch, classid); | |
201 | ||
202 | if (cl != NULL) | |
203 | cl->filter_cnt++; | |
204 | ||
205 | return (unsigned long)cl; | |
206 | } | |
207 | ||
208 | static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg) | |
209 | { | |
210 | struct drr_class *cl = (struct drr_class *)arg; | |
211 | ||
212 | cl->filter_cnt--; | |
213 | } | |
214 | ||
215 | static int drr_graft_class(struct Qdisc *sch, unsigned long arg, | |
653d6fd6 AA |
216 | struct Qdisc *new, struct Qdisc **old, |
217 | struct netlink_ext_ack *extack) | |
13d2a1d2 PM |
218 | { |
219 | struct drr_class *cl = (struct drr_class *)arg; | |
220 | ||
221 | if (new == NULL) { | |
a38a9882 AA |
222 | new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, |
223 | cl->common.classid, NULL); | |
13d2a1d2 PM |
224 | if (new == NULL) |
225 | new = &noop_qdisc; | |
226 | } | |
227 | ||
86a7996c | 228 | *old = qdisc_replace(sch, new, &cl->qdisc); |
13d2a1d2 PM |
229 | return 0; |
230 | } | |
231 | ||
232 | static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg) | |
233 | { | |
234 | struct drr_class *cl = (struct drr_class *)arg; | |
235 | ||
236 | return cl->qdisc; | |
237 | } | |
238 | ||
239 | static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg) | |
240 | { | |
241 | struct drr_class *cl = (struct drr_class *)arg; | |
242 | ||
95946658 | 243 | list_del(&cl->alist); |
13d2a1d2 PM |
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; | |
1b34ec43 DM |
259 | if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum)) |
260 | goto nla_put_failure; | |
13d2a1d2 PM |
261 | return nla_nest_end(skb, nest); |
262 | ||
263 | nla_put_failure: | |
264 | nla_nest_cancel(skb, nest); | |
265 | return -EMSGSIZE; | |
266 | } | |
267 | ||
268 | static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg, | |
269 | struct gnet_dump *d) | |
270 | { | |
271 | struct drr_class *cl = (struct drr_class *)arg; | |
64015853 | 272 | __u32 qlen = cl->qdisc->q.qlen; |
13d2a1d2 PM |
273 | struct tc_drr_stats xstats; |
274 | ||
275 | memset(&xstats, 0, sizeof(xstats)); | |
64015853 | 276 | if (qlen) |
13d2a1d2 PM |
277 | xstats.deficit = cl->deficit; |
278 | ||
edb09eb1 ED |
279 | if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), |
280 | d, NULL, &cl->bstats) < 0 || | |
1c0d32fd | 281 | gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || |
b0ab6f92 | 282 | gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0) |
13d2a1d2 PM |
283 | return -1; |
284 | ||
285 | return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); | |
286 | } | |
287 | ||
288 | static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg) | |
289 | { | |
290 | struct drr_sched *q = qdisc_priv(sch); | |
291 | struct drr_class *cl; | |
13d2a1d2 PM |
292 | unsigned int i; |
293 | ||
294 | if (arg->stop) | |
295 | return; | |
296 | ||
297 | for (i = 0; i < q->clhash.hashsize; i++) { | |
b67bfe0d | 298 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { |
13d2a1d2 PM |
299 | if (arg->count < arg->skip) { |
300 | arg->count++; | |
301 | continue; | |
302 | } | |
303 | if (arg->fn(sch, (unsigned long)cl, arg) < 0) { | |
304 | arg->stop = 1; | |
305 | return; | |
306 | } | |
307 | arg->count++; | |
308 | } | |
309 | } | |
310 | } | |
311 | ||
312 | static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch, | |
313 | int *qerr) | |
314 | { | |
315 | struct drr_sched *q = qdisc_priv(sch); | |
316 | struct drr_class *cl; | |
317 | struct tcf_result res; | |
25d8c0d5 | 318 | struct tcf_proto *fl; |
13d2a1d2 PM |
319 | int result; |
320 | ||
321 | if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) { | |
322 | cl = drr_find_class(sch, skb->priority); | |
323 | if (cl != NULL) | |
324 | return cl; | |
325 | } | |
326 | ||
327 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; | |
25d8c0d5 | 328 | fl = rcu_dereference_bh(q->filter_list); |
87d83093 | 329 | result = tcf_classify(skb, fl, &res, false); |
13d2a1d2 PM |
330 | if (result >= 0) { |
331 | #ifdef CONFIG_NET_CLS_ACT | |
332 | switch (result) { | |
333 | case TC_ACT_QUEUED: | |
334 | case TC_ACT_STOLEN: | |
e25ea21f | 335 | case TC_ACT_TRAP: |
13d2a1d2 | 336 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
f3ae608e | 337 | /* fall through */ |
13d2a1d2 PM |
338 | case TC_ACT_SHOT: |
339 | return NULL; | |
340 | } | |
341 | #endif | |
342 | cl = (struct drr_class *)res.class; | |
343 | if (cl == NULL) | |
344 | cl = drr_find_class(sch, res.classid); | |
345 | return cl; | |
346 | } | |
347 | return NULL; | |
348 | } | |
349 | ||
520ac30f ED |
350 | static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
351 | struct sk_buff **to_free) | |
13d2a1d2 | 352 | { |
f6bab199 | 353 | unsigned int len = qdisc_pkt_len(skb); |
13d2a1d2 PM |
354 | struct drr_sched *q = qdisc_priv(sch); |
355 | struct drr_class *cl; | |
f54ba779 | 356 | int err = 0; |
37d9cf1a | 357 | bool first; |
13d2a1d2 PM |
358 | |
359 | cl = drr_classify(skb, sch, &err); | |
360 | if (cl == NULL) { | |
361 | if (err & __NET_XMIT_BYPASS) | |
25331d6c | 362 | qdisc_qstats_drop(sch); |
520ac30f | 363 | __qdisc_drop(skb, to_free); |
13d2a1d2 PM |
364 | return err; |
365 | } | |
366 | ||
37d9cf1a | 367 | first = !cl->qdisc->q.qlen; |
520ac30f | 368 | err = qdisc_enqueue(skb, cl->qdisc, to_free); |
13d2a1d2 PM |
369 | if (unlikely(err != NET_XMIT_SUCCESS)) { |
370 | if (net_xmit_drop_count(err)) { | |
371 | cl->qstats.drops++; | |
25331d6c | 372 | qdisc_qstats_drop(sch); |
13d2a1d2 PM |
373 | } |
374 | return err; | |
375 | } | |
376 | ||
37d9cf1a | 377 | if (first) { |
13d2a1d2 PM |
378 | list_add_tail(&cl->alist, &q->active); |
379 | cl->deficit = cl->quantum; | |
380 | } | |
381 | ||
f6bab199 | 382 | sch->qstats.backlog += len; |
13d2a1d2 PM |
383 | sch->q.qlen++; |
384 | return err; | |
385 | } | |
386 | ||
387 | static struct sk_buff *drr_dequeue(struct Qdisc *sch) | |
388 | { | |
389 | struct drr_sched *q = qdisc_priv(sch); | |
390 | struct drr_class *cl; | |
391 | struct sk_buff *skb; | |
392 | unsigned int len; | |
393 | ||
3f0947c3 PM |
394 | if (list_empty(&q->active)) |
395 | goto out; | |
396 | while (1) { | |
13d2a1d2 PM |
397 | cl = list_first_entry(&q->active, struct drr_class, alist); |
398 | skb = cl->qdisc->ops->peek(cl->qdisc); | |
6e765a00 FW |
399 | if (skb == NULL) { |
400 | qdisc_warn_nonwc(__func__, cl->qdisc); | |
3f0947c3 | 401 | goto out; |
6e765a00 | 402 | } |
13d2a1d2 PM |
403 | |
404 | len = qdisc_pkt_len(skb); | |
405 | if (len <= cl->deficit) { | |
406 | cl->deficit -= len; | |
407 | skb = qdisc_dequeue_peeked(cl->qdisc); | |
df3eb6cd BH |
408 | if (unlikely(skb == NULL)) |
409 | goto out; | |
13d2a1d2 PM |
410 | if (cl->qdisc->q.qlen == 0) |
411 | list_del(&cl->alist); | |
2dd875ff ED |
412 | |
413 | bstats_update(&cl->bstats, skb); | |
9190b3b3 | 414 | qdisc_bstats_update(sch, skb); |
6a73b571 | 415 | qdisc_qstats_backlog_dec(sch, skb); |
13d2a1d2 PM |
416 | sch->q.qlen--; |
417 | return skb; | |
418 | } | |
419 | ||
420 | cl->deficit += cl->quantum; | |
13d2a1d2 PM |
421 | list_move_tail(&cl->alist, &q->active); |
422 | } | |
3f0947c3 | 423 | out: |
13d2a1d2 PM |
424 | return NULL; |
425 | } | |
426 | ||
e63d7dfd AA |
427 | static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt, |
428 | struct netlink_ext_ack *extack) | |
13d2a1d2 PM |
429 | { |
430 | struct drr_sched *q = qdisc_priv(sch); | |
431 | int err; | |
432 | ||
8d1a77f9 | 433 | err = tcf_block_get(&q->block, &q->filter_list, sch, extack); |
6529eaba JP |
434 | if (err) |
435 | return err; | |
13d2a1d2 PM |
436 | err = qdisc_class_hash_init(&q->clhash); |
437 | if (err < 0) | |
438 | return err; | |
439 | INIT_LIST_HEAD(&q->active); | |
440 | return 0; | |
441 | } | |
442 | ||
443 | static void drr_reset_qdisc(struct Qdisc *sch) | |
444 | { | |
445 | struct drr_sched *q = qdisc_priv(sch); | |
446 | struct drr_class *cl; | |
13d2a1d2 PM |
447 | unsigned int i; |
448 | ||
449 | for (i = 0; i < q->clhash.hashsize; i++) { | |
b67bfe0d | 450 | hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { |
13d2a1d2 PM |
451 | if (cl->qdisc->q.qlen) |
452 | list_del(&cl->alist); | |
453 | qdisc_reset(cl->qdisc); | |
454 | } | |
455 | } | |
6a73b571 | 456 | sch->qstats.backlog = 0; |
13d2a1d2 PM |
457 | sch->q.qlen = 0; |
458 | } | |
459 | ||
460 | static void drr_destroy_qdisc(struct Qdisc *sch) | |
461 | { | |
462 | struct drr_sched *q = qdisc_priv(sch); | |
463 | struct drr_class *cl; | |
b67bfe0d | 464 | struct hlist_node *next; |
13d2a1d2 PM |
465 | unsigned int i; |
466 | ||
6529eaba | 467 | tcf_block_put(q->block); |
13d2a1d2 PM |
468 | |
469 | for (i = 0; i < q->clhash.hashsize; i++) { | |
b67bfe0d | 470 | hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], |
13d2a1d2 PM |
471 | common.hnode) |
472 | drr_destroy_class(sch, cl); | |
473 | } | |
474 | qdisc_class_hash_destroy(&q->clhash); | |
475 | } | |
476 | ||
477 | static const struct Qdisc_class_ops drr_class_ops = { | |
478 | .change = drr_change_class, | |
479 | .delete = drr_delete_class, | |
143976ce | 480 | .find = drr_search_class, |
6529eaba | 481 | .tcf_block = drr_tcf_block, |
13d2a1d2 PM |
482 | .bind_tcf = drr_bind_tcf, |
483 | .unbind_tcf = drr_unbind_tcf, | |
484 | .graft = drr_graft_class, | |
485 | .leaf = drr_class_leaf, | |
486 | .qlen_notify = drr_qlen_notify, | |
487 | .dump = drr_dump_class, | |
488 | .dump_stats = drr_dump_class_stats, | |
489 | .walk = drr_walk, | |
490 | }; | |
491 | ||
492 | static struct Qdisc_ops drr_qdisc_ops __read_mostly = { | |
493 | .cl_ops = &drr_class_ops, | |
494 | .id = "drr", | |
495 | .priv_size = sizeof(struct drr_sched), | |
496 | .enqueue = drr_enqueue, | |
497 | .dequeue = drr_dequeue, | |
498 | .peek = qdisc_peek_dequeued, | |
13d2a1d2 PM |
499 | .init = drr_init_qdisc, |
500 | .reset = drr_reset_qdisc, | |
501 | .destroy = drr_destroy_qdisc, | |
502 | .owner = THIS_MODULE, | |
503 | }; | |
504 | ||
505 | static int __init drr_init(void) | |
506 | { | |
507 | return register_qdisc(&drr_qdisc_ops); | |
508 | } | |
509 | ||
510 | static void __exit drr_exit(void) | |
511 | { | |
512 | unregister_qdisc(&drr_qdisc_ops); | |
513 | } | |
514 | ||
515 | module_init(drr_init); | |
516 | module_exit(drr_exit); | |
517 | MODULE_LICENSE("GPL"); |