]> git.ipfire.org Git - thirdparty/kernel/stable.git/blob - net/sched/sch_prio.c
net: sched: introduce tcf block infractructure
[thirdparty/kernel/stable.git] / net / sched / sch_prio.c
1 /*
2 * net/sched/sch_prio.c Simple 3-band priority "scheduler".
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 * Fixes: 19990609: J Hadi Salim <hadi@nortelnetworks.com>:
11 * Init -- EINVAL when opt undefined
12 */
13
14 #include <linux/module.h>
15 #include <linux/slab.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/string.h>
19 #include <linux/errno.h>
20 #include <linux/skbuff.h>
21 #include <net/netlink.h>
22 #include <net/pkt_sched.h>
23 #include <net/pkt_cls.h>
24
25 struct prio_sched_data {
26 int bands;
27 struct tcf_proto __rcu *filter_list;
28 struct tcf_block *block;
29 u8 prio2band[TC_PRIO_MAX+1];
30 struct Qdisc *queues[TCQ_PRIO_BANDS];
31 };
32
33
34 static struct Qdisc *
35 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
36 {
37 struct prio_sched_data *q = qdisc_priv(sch);
38 u32 band = skb->priority;
39 struct tcf_result res;
40 struct tcf_proto *fl;
41 int err;
42
43 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
44 if (TC_H_MAJ(skb->priority) != sch->handle) {
45 fl = rcu_dereference_bh(q->filter_list);
46 err = tcf_classify(skb, fl, &res, false);
47 #ifdef CONFIG_NET_CLS_ACT
48 switch (err) {
49 case TC_ACT_STOLEN:
50 case TC_ACT_QUEUED:
51 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
52 case TC_ACT_SHOT:
53 return NULL;
54 }
55 #endif
56 if (!fl || err < 0) {
57 if (TC_H_MAJ(band))
58 band = 0;
59 return q->queues[q->prio2band[band & TC_PRIO_MAX]];
60 }
61 band = res.classid;
62 }
63 band = TC_H_MIN(band) - 1;
64 if (band >= q->bands)
65 return q->queues[q->prio2band[0]];
66
67 return q->queues[band];
68 }
69
70 static int
71 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free)
72 {
73 struct Qdisc *qdisc;
74 int ret;
75
76 qdisc = prio_classify(skb, sch, &ret);
77 #ifdef CONFIG_NET_CLS_ACT
78 if (qdisc == NULL) {
79
80 if (ret & __NET_XMIT_BYPASS)
81 qdisc_qstats_drop(sch);
82 kfree_skb(skb);
83 return ret;
84 }
85 #endif
86
87 ret = qdisc_enqueue(skb, qdisc, to_free);
88 if (ret == NET_XMIT_SUCCESS) {
89 qdisc_qstats_backlog_inc(sch, skb);
90 sch->q.qlen++;
91 return NET_XMIT_SUCCESS;
92 }
93 if (net_xmit_drop_count(ret))
94 qdisc_qstats_drop(sch);
95 return ret;
96 }
97
98 static struct sk_buff *prio_peek(struct Qdisc *sch)
99 {
100 struct prio_sched_data *q = qdisc_priv(sch);
101 int prio;
102
103 for (prio = 0; prio < q->bands; prio++) {
104 struct Qdisc *qdisc = q->queues[prio];
105 struct sk_buff *skb = qdisc->ops->peek(qdisc);
106 if (skb)
107 return skb;
108 }
109 return NULL;
110 }
111
112 static struct sk_buff *prio_dequeue(struct Qdisc *sch)
113 {
114 struct prio_sched_data *q = qdisc_priv(sch);
115 int prio;
116
117 for (prio = 0; prio < q->bands; prio++) {
118 struct Qdisc *qdisc = q->queues[prio];
119 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
120 if (skb) {
121 qdisc_bstats_update(sch, skb);
122 qdisc_qstats_backlog_dec(sch, skb);
123 sch->q.qlen--;
124 return skb;
125 }
126 }
127 return NULL;
128
129 }
130
131 static void
132 prio_reset(struct Qdisc *sch)
133 {
134 int prio;
135 struct prio_sched_data *q = qdisc_priv(sch);
136
137 for (prio = 0; prio < q->bands; prio++)
138 qdisc_reset(q->queues[prio]);
139 sch->qstats.backlog = 0;
140 sch->q.qlen = 0;
141 }
142
143 static void
144 prio_destroy(struct Qdisc *sch)
145 {
146 int prio;
147 struct prio_sched_data *q = qdisc_priv(sch);
148
149 tcf_block_put(q->block);
150 for (prio = 0; prio < q->bands; prio++)
151 qdisc_destroy(q->queues[prio]);
152 }
153
154 static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
155 {
156 struct prio_sched_data *q = qdisc_priv(sch);
157 struct Qdisc *queues[TCQ_PRIO_BANDS];
158 int oldbands = q->bands, i;
159 struct tc_prio_qopt *qopt;
160
161 if (nla_len(opt) < sizeof(*qopt))
162 return -EINVAL;
163 qopt = nla_data(opt);
164
165 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
166 return -EINVAL;
167
168 for (i = 0; i <= TC_PRIO_MAX; i++) {
169 if (qopt->priomap[i] >= qopt->bands)
170 return -EINVAL;
171 }
172
173 /* Before commit, make sure we can allocate all new qdiscs */
174 for (i = oldbands; i < qopt->bands; i++) {
175 queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
176 TC_H_MAKE(sch->handle, i + 1));
177 if (!queues[i]) {
178 while (i > oldbands)
179 qdisc_destroy(queues[--i]);
180 return -ENOMEM;
181 }
182 }
183
184 sch_tree_lock(sch);
185 q->bands = qopt->bands;
186 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
187
188 for (i = q->bands; i < oldbands; i++) {
189 struct Qdisc *child = q->queues[i];
190
191 qdisc_tree_reduce_backlog(child, child->q.qlen,
192 child->qstats.backlog);
193 qdisc_destroy(child);
194 }
195
196 for (i = oldbands; i < q->bands; i++) {
197 q->queues[i] = queues[i];
198 if (q->queues[i] != &noop_qdisc)
199 qdisc_hash_add(q->queues[i], true);
200 }
201
202 sch_tree_unlock(sch);
203 return 0;
204 }
205
206 static int prio_init(struct Qdisc *sch, struct nlattr *opt)
207 {
208 struct prio_sched_data *q = qdisc_priv(sch);
209 int err;
210
211 if (!opt)
212 return -EINVAL;
213
214 err = tcf_block_get(&q->block, &q->filter_list);
215 if (err)
216 return err;
217
218 return prio_tune(sch, opt);
219 }
220
221 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
222 {
223 struct prio_sched_data *q = qdisc_priv(sch);
224 unsigned char *b = skb_tail_pointer(skb);
225 struct tc_prio_qopt opt;
226
227 opt.bands = q->bands;
228 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
229
230 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
231 goto nla_put_failure;
232
233 return skb->len;
234
235 nla_put_failure:
236 nlmsg_trim(skb, b);
237 return -1;
238 }
239
240 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
241 struct Qdisc **old)
242 {
243 struct prio_sched_data *q = qdisc_priv(sch);
244 unsigned long band = arg - 1;
245
246 if (new == NULL)
247 new = &noop_qdisc;
248
249 *old = qdisc_replace(sch, new, &q->queues[band]);
250 return 0;
251 }
252
253 static struct Qdisc *
254 prio_leaf(struct Qdisc *sch, unsigned long arg)
255 {
256 struct prio_sched_data *q = qdisc_priv(sch);
257 unsigned long band = arg - 1;
258
259 return q->queues[band];
260 }
261
262 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
263 {
264 struct prio_sched_data *q = qdisc_priv(sch);
265 unsigned long band = TC_H_MIN(classid);
266
267 if (band - 1 >= q->bands)
268 return 0;
269 return band;
270 }
271
272 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
273 {
274 return prio_get(sch, classid);
275 }
276
277
278 static void prio_put(struct Qdisc *q, unsigned long cl)
279 {
280 }
281
282 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
283 struct tcmsg *tcm)
284 {
285 struct prio_sched_data *q = qdisc_priv(sch);
286
287 tcm->tcm_handle |= TC_H_MIN(cl);
288 tcm->tcm_info = q->queues[cl-1]->handle;
289 return 0;
290 }
291
292 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
293 struct gnet_dump *d)
294 {
295 struct prio_sched_data *q = qdisc_priv(sch);
296 struct Qdisc *cl_q;
297
298 cl_q = q->queues[cl - 1];
299 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
300 d, NULL, &cl_q->bstats) < 0 ||
301 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
302 return -1;
303
304 return 0;
305 }
306
307 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
308 {
309 struct prio_sched_data *q = qdisc_priv(sch);
310 int prio;
311
312 if (arg->stop)
313 return;
314
315 for (prio = 0; prio < q->bands; prio++) {
316 if (arg->count < arg->skip) {
317 arg->count++;
318 continue;
319 }
320 if (arg->fn(sch, prio + 1, arg) < 0) {
321 arg->stop = 1;
322 break;
323 }
324 arg->count++;
325 }
326 }
327
328 static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl)
329 {
330 struct prio_sched_data *q = qdisc_priv(sch);
331
332 if (cl)
333 return NULL;
334 return q->block;
335 }
336
337 static const struct Qdisc_class_ops prio_class_ops = {
338 .graft = prio_graft,
339 .leaf = prio_leaf,
340 .get = prio_get,
341 .put = prio_put,
342 .walk = prio_walk,
343 .tcf_block = prio_tcf_block,
344 .bind_tcf = prio_bind,
345 .unbind_tcf = prio_put,
346 .dump = prio_dump_class,
347 .dump_stats = prio_dump_class_stats,
348 };
349
350 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
351 .next = NULL,
352 .cl_ops = &prio_class_ops,
353 .id = "prio",
354 .priv_size = sizeof(struct prio_sched_data),
355 .enqueue = prio_enqueue,
356 .dequeue = prio_dequeue,
357 .peek = prio_peek,
358 .init = prio_init,
359 .reset = prio_reset,
360 .destroy = prio_destroy,
361 .change = prio_tune,
362 .dump = prio_dump,
363 .owner = THIS_MODULE,
364 };
365
366 static int __init prio_module_init(void)
367 {
368 return register_qdisc(&prio_qdisc_ops);
369 }
370
371 static void __exit prio_module_exit(void)
372 {
373 unregister_qdisc(&prio_qdisc_ops);
374 }
375
376 module_init(prio_module_init)
377 module_exit(prio_module_exit)
378
379 MODULE_LICENSE("GPL");