Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[sfrench/cifs-2.6.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                 case TC_ACT_TRAP:
52                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
53                         /* fall through */
54                 case TC_ACT_SHOT:
55                         return NULL;
56                 }
57 #endif
58                 if (!fl || err < 0) {
59                         if (TC_H_MAJ(band))
60                                 band = 0;
61                         return q->queues[q->prio2band[band & TC_PRIO_MAX]];
62                 }
63                 band = res.classid;
64         }
65         band = TC_H_MIN(band) - 1;
66         if (band >= q->bands)
67                 return q->queues[q->prio2band[0]];
68
69         return q->queues[band];
70 }
71
72 static int
73 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free)
74 {
75         struct Qdisc *qdisc;
76         int ret;
77
78         qdisc = prio_classify(skb, sch, &ret);
79 #ifdef CONFIG_NET_CLS_ACT
80         if (qdisc == NULL) {
81
82                 if (ret & __NET_XMIT_BYPASS)
83                         qdisc_qstats_drop(sch);
84                 __qdisc_drop(skb, to_free);
85                 return ret;
86         }
87 #endif
88
89         ret = qdisc_enqueue(skb, qdisc, to_free);
90         if (ret == NET_XMIT_SUCCESS) {
91                 qdisc_qstats_backlog_inc(sch, skb);
92                 sch->q.qlen++;
93                 return NET_XMIT_SUCCESS;
94         }
95         if (net_xmit_drop_count(ret))
96                 qdisc_qstats_drop(sch);
97         return ret;
98 }
99
100 static struct sk_buff *prio_peek(struct Qdisc *sch)
101 {
102         struct prio_sched_data *q = qdisc_priv(sch);
103         int prio;
104
105         for (prio = 0; prio < q->bands; prio++) {
106                 struct Qdisc *qdisc = q->queues[prio];
107                 struct sk_buff *skb = qdisc->ops->peek(qdisc);
108                 if (skb)
109                         return skb;
110         }
111         return NULL;
112 }
113
114 static struct sk_buff *prio_dequeue(struct Qdisc *sch)
115 {
116         struct prio_sched_data *q = qdisc_priv(sch);
117         int prio;
118
119         for (prio = 0; prio < q->bands; prio++) {
120                 struct Qdisc *qdisc = q->queues[prio];
121                 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
122                 if (skb) {
123                         qdisc_bstats_update(sch, skb);
124                         qdisc_qstats_backlog_dec(sch, skb);
125                         sch->q.qlen--;
126                         return skb;
127                 }
128         }
129         return NULL;
130
131 }
132
133 static void
134 prio_reset(struct Qdisc *sch)
135 {
136         int prio;
137         struct prio_sched_data *q = qdisc_priv(sch);
138
139         for (prio = 0; prio < q->bands; prio++)
140                 qdisc_reset(q->queues[prio]);
141         sch->qstats.backlog = 0;
142         sch->q.qlen = 0;
143 }
144
145 static int prio_offload(struct Qdisc *sch, struct tc_prio_qopt *qopt)
146 {
147         struct net_device *dev = qdisc_dev(sch);
148         struct tc_prio_qopt_offload opt = {
149                 .handle = sch->handle,
150                 .parent = sch->parent,
151         };
152
153         if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
154                 return -EOPNOTSUPP;
155
156         if (qopt) {
157                 opt.command = TC_PRIO_REPLACE;
158                 opt.replace_params.bands = qopt->bands;
159                 memcpy(&opt.replace_params.priomap, qopt->priomap,
160                        TC_PRIO_MAX + 1);
161                 opt.replace_params.qstats = &sch->qstats;
162         } else {
163                 opt.command = TC_PRIO_DESTROY;
164         }
165
166         return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_PRIO, &opt);
167 }
168
169 static void
170 prio_destroy(struct Qdisc *sch)
171 {
172         int prio;
173         struct prio_sched_data *q = qdisc_priv(sch);
174
175         tcf_block_put(q->block);
176         prio_offload(sch, NULL);
177         for (prio = 0; prio < q->bands; prio++)
178                 qdisc_put(q->queues[prio]);
179 }
180
181 static int prio_tune(struct Qdisc *sch, struct nlattr *opt,
182                      struct netlink_ext_ack *extack)
183 {
184         struct prio_sched_data *q = qdisc_priv(sch);
185         struct Qdisc *queues[TCQ_PRIO_BANDS];
186         int oldbands = q->bands, i;
187         struct tc_prio_qopt *qopt;
188
189         if (nla_len(opt) < sizeof(*qopt))
190                 return -EINVAL;
191         qopt = nla_data(opt);
192
193         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
194                 return -EINVAL;
195
196         for (i = 0; i <= TC_PRIO_MAX; i++) {
197                 if (qopt->priomap[i] >= qopt->bands)
198                         return -EINVAL;
199         }
200
201         /* Before commit, make sure we can allocate all new qdiscs */
202         for (i = oldbands; i < qopt->bands; i++) {
203                 queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
204                                               TC_H_MAKE(sch->handle, i + 1),
205                                               extack);
206                 if (!queues[i]) {
207                         while (i > oldbands)
208                                 qdisc_put(queues[--i]);
209                         return -ENOMEM;
210                 }
211         }
212
213         prio_offload(sch, qopt);
214         sch_tree_lock(sch);
215         q->bands = qopt->bands;
216         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
217
218         for (i = q->bands; i < oldbands; i++) {
219                 struct Qdisc *child = q->queues[i];
220
221                 qdisc_tree_reduce_backlog(child, child->q.qlen,
222                                           child->qstats.backlog);
223         }
224
225         for (i = oldbands; i < q->bands; i++) {
226                 q->queues[i] = queues[i];
227                 if (q->queues[i] != &noop_qdisc)
228                         qdisc_hash_add(q->queues[i], true);
229         }
230
231         sch_tree_unlock(sch);
232
233         for (i = q->bands; i < oldbands; i++)
234                 qdisc_put(q->queues[i]);
235         return 0;
236 }
237
238 static int prio_init(struct Qdisc *sch, struct nlattr *opt,
239                      struct netlink_ext_ack *extack)
240 {
241         struct prio_sched_data *q = qdisc_priv(sch);
242         int err;
243
244         if (!opt)
245                 return -EINVAL;
246
247         err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
248         if (err)
249                 return err;
250
251         return prio_tune(sch, opt, extack);
252 }
253
254 static int prio_dump_offload(struct Qdisc *sch)
255 {
256         struct tc_prio_qopt_offload hw_stats = {
257                 .command = TC_PRIO_STATS,
258                 .handle = sch->handle,
259                 .parent = sch->parent,
260                 {
261                         .stats = {
262                                 .bstats = &sch->bstats,
263                                 .qstats = &sch->qstats,
264                         },
265                 },
266         };
267
268         return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_PRIO, &hw_stats);
269 }
270
271 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
272 {
273         struct prio_sched_data *q = qdisc_priv(sch);
274         unsigned char *b = skb_tail_pointer(skb);
275         struct tc_prio_qopt opt;
276         int err;
277
278         opt.bands = q->bands;
279         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
280
281         err = prio_dump_offload(sch);
282         if (err)
283                 goto nla_put_failure;
284
285         if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
286                 goto nla_put_failure;
287
288         return skb->len;
289
290 nla_put_failure:
291         nlmsg_trim(skb, b);
292         return -1;
293 }
294
295 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
296                       struct Qdisc **old, struct netlink_ext_ack *extack)
297 {
298         struct prio_sched_data *q = qdisc_priv(sch);
299         struct tc_prio_qopt_offload graft_offload;
300         unsigned long band = arg - 1;
301
302         if (new == NULL)
303                 new = &noop_qdisc;
304
305         *old = qdisc_replace(sch, new, &q->queues[band]);
306
307         graft_offload.handle = sch->handle;
308         graft_offload.parent = sch->parent;
309         graft_offload.graft_params.band = band;
310         graft_offload.graft_params.child_handle = new->handle;
311         graft_offload.command = TC_PRIO_GRAFT;
312
313         qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, *old,
314                                    TC_SETUP_QDISC_PRIO, &graft_offload,
315                                    extack);
316         return 0;
317 }
318
319 static struct Qdisc *
320 prio_leaf(struct Qdisc *sch, unsigned long arg)
321 {
322         struct prio_sched_data *q = qdisc_priv(sch);
323         unsigned long band = arg - 1;
324
325         return q->queues[band];
326 }
327
328 static unsigned long prio_find(struct Qdisc *sch, u32 classid)
329 {
330         struct prio_sched_data *q = qdisc_priv(sch);
331         unsigned long band = TC_H_MIN(classid);
332
333         if (band - 1 >= q->bands)
334                 return 0;
335         return band;
336 }
337
338 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
339 {
340         return prio_find(sch, classid);
341 }
342
343
344 static void prio_unbind(struct Qdisc *q, unsigned long cl)
345 {
346 }
347
348 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
349                            struct tcmsg *tcm)
350 {
351         struct prio_sched_data *q = qdisc_priv(sch);
352
353         tcm->tcm_handle |= TC_H_MIN(cl);
354         tcm->tcm_info = q->queues[cl-1]->handle;
355         return 0;
356 }
357
358 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
359                                  struct gnet_dump *d)
360 {
361         struct prio_sched_data *q = qdisc_priv(sch);
362         struct Qdisc *cl_q;
363
364         cl_q = q->queues[cl - 1];
365         if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
366                                   d, NULL, &cl_q->bstats) < 0 ||
367             gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
368                 return -1;
369
370         return 0;
371 }
372
373 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
374 {
375         struct prio_sched_data *q = qdisc_priv(sch);
376         int prio;
377
378         if (arg->stop)
379                 return;
380
381         for (prio = 0; prio < q->bands; prio++) {
382                 if (arg->count < arg->skip) {
383                         arg->count++;
384                         continue;
385                 }
386                 if (arg->fn(sch, prio + 1, arg) < 0) {
387                         arg->stop = 1;
388                         break;
389                 }
390                 arg->count++;
391         }
392 }
393
394 static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl,
395                                         struct netlink_ext_ack *extack)
396 {
397         struct prio_sched_data *q = qdisc_priv(sch);
398
399         if (cl)
400                 return NULL;
401         return q->block;
402 }
403
404 static const struct Qdisc_class_ops prio_class_ops = {
405         .graft          =       prio_graft,
406         .leaf           =       prio_leaf,
407         .find           =       prio_find,
408         .walk           =       prio_walk,
409         .tcf_block      =       prio_tcf_block,
410         .bind_tcf       =       prio_bind,
411         .unbind_tcf     =       prio_unbind,
412         .dump           =       prio_dump_class,
413         .dump_stats     =       prio_dump_class_stats,
414 };
415
416 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
417         .next           =       NULL,
418         .cl_ops         =       &prio_class_ops,
419         .id             =       "prio",
420         .priv_size      =       sizeof(struct prio_sched_data),
421         .enqueue        =       prio_enqueue,
422         .dequeue        =       prio_dequeue,
423         .peek           =       prio_peek,
424         .init           =       prio_init,
425         .reset          =       prio_reset,
426         .destroy        =       prio_destroy,
427         .change         =       prio_tune,
428         .dump           =       prio_dump,
429         .owner          =       THIS_MODULE,
430 };
431
432 static int __init prio_module_init(void)
433 {
434         return register_qdisc(&prio_qdisc_ops);
435 }
436
437 static void __exit prio_module_exit(void)
438 {
439         unregister_qdisc(&prio_qdisc_ops);
440 }
441
442 module_init(prio_module_init)
443 module_exit(prio_module_exit)
444
445 MODULE_LICENSE("GPL");