]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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> | |
10297b99 | 10 | * Fixes: 19990609: J Hadi Salim <hadi@nortelnetworks.com>: |
1da177e4 LT |
11 | * Init -- EINVAL when opt undefined |
12 | */ | |
13 | ||
1da177e4 | 14 | #include <linux/module.h> |
5a0e3ad6 | 15 | #include <linux/slab.h> |
1da177e4 LT |
16 | #include <linux/types.h> |
17 | #include <linux/kernel.h> | |
1da177e4 | 18 | #include <linux/string.h> |
1da177e4 | 19 | #include <linux/errno.h> |
1da177e4 | 20 | #include <linux/skbuff.h> |
dc5fc579 | 21 | #include <net/netlink.h> |
1da177e4 | 22 | #include <net/pkt_sched.h> |
cf1facda | 23 | #include <net/pkt_cls.h> |
1da177e4 | 24 | |
cc7ec456 | 25 | struct prio_sched_data { |
1da177e4 | 26 | int bands; |
25d8c0d5 | 27 | struct tcf_proto __rcu *filter_list; |
6529eaba | 28 | struct tcf_block *block; |
1da177e4 LT |
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; | |
25d8c0d5 | 40 | struct tcf_proto *fl; |
bdba91ec | 41 | int err; |
1da177e4 | 42 | |
c27f339a | 43 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; |
1da177e4 | 44 | if (TC_H_MAJ(skb->priority) != sch->handle) { |
25d8c0d5 | 45 | fl = rcu_dereference_bh(q->filter_list); |
87d83093 | 46 | err = tcf_classify(skb, fl, &res, false); |
1da177e4 | 47 | #ifdef CONFIG_NET_CLS_ACT |
dbaaa07a | 48 | switch (err) { |
1da177e4 LT |
49 | case TC_ACT_STOLEN: |
50 | case TC_ACT_QUEUED: | |
e25ea21f | 51 | case TC_ACT_TRAP: |
378a2f09 | 52 | *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; |
1da177e4 LT |
53 | case TC_ACT_SHOT: |
54 | return NULL; | |
3ff50b79 | 55 | } |
1da177e4 | 56 | #endif |
25d8c0d5 | 57 | if (!fl || err < 0) { |
1da177e4 LT |
58 | if (TC_H_MAJ(band)) |
59 | band = 0; | |
cc7ec456 | 60 | return q->queues[q->prio2band[band & TC_PRIO_MAX]]; |
1da177e4 LT |
61 | } |
62 | band = res.classid; | |
63 | } | |
64 | band = TC_H_MIN(band) - 1; | |
3e5c2d3b | 65 | if (band >= q->bands) |
1d8ae3fd DM |
66 | return q->queues[q->prio2band[0]]; |
67 | ||
1da177e4 LT |
68 | return q->queues[band]; |
69 | } | |
70 | ||
71 | static int | |
520ac30f | 72 | prio_enqueue(struct sk_buff *skb, struct Qdisc *sch, struct sk_buff **to_free) |
1da177e4 LT |
73 | { |
74 | struct Qdisc *qdisc; | |
75 | int ret; | |
76 | ||
77 | qdisc = prio_classify(skb, sch, &ret); | |
78 | #ifdef CONFIG_NET_CLS_ACT | |
79 | if (qdisc == NULL) { | |
29f1df6c | 80 | |
c27f339a | 81 | if (ret & __NET_XMIT_BYPASS) |
25331d6c | 82 | qdisc_qstats_drop(sch); |
1da177e4 LT |
83 | kfree_skb(skb); |
84 | return ret; | |
85 | } | |
86 | #endif | |
87 | ||
520ac30f | 88 | ret = qdisc_enqueue(skb, qdisc, to_free); |
5f86173b | 89 | if (ret == NET_XMIT_SUCCESS) { |
6529d75a | 90 | qdisc_qstats_backlog_inc(sch, skb); |
1da177e4 LT |
91 | sch->q.qlen++; |
92 | return NET_XMIT_SUCCESS; | |
93 | } | |
378a2f09 | 94 | if (net_xmit_drop_count(ret)) |
25331d6c | 95 | qdisc_qstats_drop(sch); |
10297b99 | 96 | return ret; |
1da177e4 LT |
97 | } |
98 | ||
48a8f519 PM |
99 | static struct sk_buff *prio_peek(struct Qdisc *sch) |
100 | { | |
101 | struct prio_sched_data *q = qdisc_priv(sch); | |
102 | int prio; | |
103 | ||
104 | for (prio = 0; prio < q->bands; prio++) { | |
105 | struct Qdisc *qdisc = q->queues[prio]; | |
106 | struct sk_buff *skb = qdisc->ops->peek(qdisc); | |
107 | if (skb) | |
108 | return skb; | |
109 | } | |
110 | return NULL; | |
111 | } | |
1da177e4 | 112 | |
cc7ec456 | 113 | static struct sk_buff *prio_dequeue(struct Qdisc *sch) |
1da177e4 | 114 | { |
1da177e4 LT |
115 | struct prio_sched_data *q = qdisc_priv(sch); |
116 | int prio; | |
1da177e4 LT |
117 | |
118 | for (prio = 0; prio < q->bands; prio++) { | |
1d8ae3fd | 119 | struct Qdisc *qdisc = q->queues[prio]; |
3557619f | 120 | struct sk_buff *skb = qdisc_dequeue_peeked(qdisc); |
1d8ae3fd | 121 | if (skb) { |
9190b3b3 | 122 | qdisc_bstats_update(sch, skb); |
6529d75a | 123 | qdisc_qstats_backlog_dec(sch, skb); |
1d8ae3fd DM |
124 | sch->q.qlen--; |
125 | return skb; | |
1da177e4 LT |
126 | } |
127 | } | |
128 | return NULL; | |
129 | ||
130 | } | |
131 | ||
1da177e4 | 132 | static void |
cc7ec456 | 133 | prio_reset(struct Qdisc *sch) |
1da177e4 LT |
134 | { |
135 | int prio; | |
136 | struct prio_sched_data *q = qdisc_priv(sch); | |
137 | ||
cc7ec456 | 138 | for (prio = 0; prio < q->bands; prio++) |
1da177e4 | 139 | qdisc_reset(q->queues[prio]); |
6529d75a | 140 | sch->qstats.backlog = 0; |
1da177e4 LT |
141 | sch->q.qlen = 0; |
142 | } | |
143 | ||
144 | static void | |
cc7ec456 | 145 | prio_destroy(struct Qdisc *sch) |
1da177e4 LT |
146 | { |
147 | int prio; | |
148 | struct prio_sched_data *q = qdisc_priv(sch); | |
1da177e4 | 149 | |
6529eaba | 150 | tcf_block_put(q->block); |
cc7ec456 | 151 | for (prio = 0; prio < q->bands; prio++) |
1da177e4 LT |
152 | qdisc_destroy(q->queues[prio]); |
153 | } | |
154 | ||
1e90474c | 155 | static int prio_tune(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
156 | { |
157 | struct prio_sched_data *q = qdisc_priv(sch); | |
3d7c8257 ED |
158 | struct Qdisc *queues[TCQ_PRIO_BANDS]; |
159 | int oldbands = q->bands, i; | |
d62733c8 | 160 | struct tc_prio_qopt *qopt; |
1da177e4 | 161 | |
1d8ae3fd DM |
162 | if (nla_len(opt) < sizeof(*qopt)) |
163 | return -EINVAL; | |
164 | qopt = nla_data(opt); | |
d62733c8 | 165 | |
1d8ae3fd | 166 | if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2) |
1da177e4 LT |
167 | return -EINVAL; |
168 | ||
cc7ec456 | 169 | for (i = 0; i <= TC_PRIO_MAX; i++) { |
1d8ae3fd | 170 | if (qopt->priomap[i] >= qopt->bands) |
1da177e4 LT |
171 | return -EINVAL; |
172 | } | |
173 | ||
3d7c8257 ED |
174 | /* Before commit, make sure we can allocate all new qdiscs */ |
175 | for (i = oldbands; i < qopt->bands; i++) { | |
176 | queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, | |
177 | TC_H_MAKE(sch->handle, i + 1)); | |
178 | if (!queues[i]) { | |
179 | while (i > oldbands) | |
180 | qdisc_destroy(queues[--i]); | |
181 | return -ENOMEM; | |
182 | } | |
183 | } | |
184 | ||
1da177e4 | 185 | sch_tree_lock(sch); |
1d8ae3fd | 186 | q->bands = qopt->bands; |
1da177e4 LT |
187 | memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1); |
188 | ||
3d7c8257 | 189 | for (i = q->bands; i < oldbands; i++) { |
b94c8afc | 190 | struct Qdisc *child = q->queues[i]; |
1da177e4 | 191 | |
3d7c8257 ED |
192 | qdisc_tree_reduce_backlog(child, child->q.qlen, |
193 | child->qstats.backlog); | |
194 | qdisc_destroy(child); | |
1da177e4 | 195 | } |
cbdf4511 | 196 | |
49b49971 | 197 | for (i = oldbands; i < q->bands; i++) { |
3d7c8257 | 198 | q->queues[i] = queues[i]; |
49b49971 JK |
199 | if (q->queues[i] != &noop_qdisc) |
200 | qdisc_hash_add(q->queues[i], true); | |
201 | } | |
cbdf4511 | 202 | |
3d7c8257 | 203 | sch_tree_unlock(sch); |
1da177e4 LT |
204 | return 0; |
205 | } | |
206 | ||
1e90474c | 207 | static int prio_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 | 208 | { |
6529eaba JP |
209 | struct prio_sched_data *q = qdisc_priv(sch); |
210 | int err; | |
211 | ||
3d7c8257 | 212 | if (!opt) |
1da177e4 | 213 | return -EINVAL; |
1da177e4 | 214 | |
6529eaba JP |
215 | err = tcf_block_get(&q->block, &q->filter_list); |
216 | if (err) | |
217 | return err; | |
218 | ||
3d7c8257 | 219 | return prio_tune(sch, opt); |
1da177e4 LT |
220 | } |
221 | ||
222 | static int prio_dump(struct Qdisc *sch, struct sk_buff *skb) | |
223 | { | |
224 | struct prio_sched_data *q = qdisc_priv(sch); | |
27a884dc | 225 | unsigned char *b = skb_tail_pointer(skb); |
1da177e4 LT |
226 | struct tc_prio_qopt opt; |
227 | ||
228 | opt.bands = q->bands; | |
cc7ec456 | 229 | memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1); |
d62733c8 | 230 | |
1b34ec43 DM |
231 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
232 | goto nla_put_failure; | |
d62733c8 | 233 | |
1da177e4 LT |
234 | return skb->len; |
235 | ||
1e90474c | 236 | nla_put_failure: |
dc5fc579 | 237 | nlmsg_trim(skb, b); |
1da177e4 LT |
238 | return -1; |
239 | } | |
240 | ||
241 | static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
242 | struct Qdisc **old) | |
243 | { | |
244 | struct prio_sched_data *q = qdisc_priv(sch); | |
245 | unsigned long band = arg - 1; | |
246 | ||
1da177e4 LT |
247 | if (new == NULL) |
248 | new = &noop_qdisc; | |
249 | ||
86a7996c | 250 | *old = qdisc_replace(sch, new, &q->queues[band]); |
1da177e4 LT |
251 | return 0; |
252 | } | |
253 | ||
254 | static struct Qdisc * | |
255 | prio_leaf(struct Qdisc *sch, unsigned long arg) | |
256 | { | |
257 | struct prio_sched_data *q = qdisc_priv(sch); | |
258 | unsigned long band = arg - 1; | |
259 | ||
1da177e4 LT |
260 | return q->queues[band]; |
261 | } | |
262 | ||
263 | static unsigned long prio_get(struct Qdisc *sch, u32 classid) | |
264 | { | |
265 | struct prio_sched_data *q = qdisc_priv(sch); | |
266 | unsigned long band = TC_H_MIN(classid); | |
267 | ||
268 | if (band - 1 >= q->bands) | |
269 | return 0; | |
270 | return band; | |
271 | } | |
272 | ||
273 | static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid) | |
274 | { | |
275 | return prio_get(sch, classid); | |
276 | } | |
277 | ||
278 | ||
279 | static void prio_put(struct Qdisc *q, unsigned long cl) | |
280 | { | |
1da177e4 LT |
281 | } |
282 | ||
1da177e4 LT |
283 | static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb, |
284 | struct tcmsg *tcm) | |
285 | { | |
286 | struct prio_sched_data *q = qdisc_priv(sch); | |
287 | ||
1da177e4 | 288 | tcm->tcm_handle |= TC_H_MIN(cl); |
5b9a9ccf | 289 | tcm->tcm_info = q->queues[cl-1]->handle; |
1da177e4 LT |
290 | return 0; |
291 | } | |
292 | ||
2cf6c36c JP |
293 | static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl, |
294 | struct gnet_dump *d) | |
295 | { | |
296 | struct prio_sched_data *q = qdisc_priv(sch); | |
297 | struct Qdisc *cl_q; | |
298 | ||
299 | cl_q = q->queues[cl - 1]; | |
edb09eb1 ED |
300 | if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), |
301 | d, NULL, &cl_q->bstats) < 0 || | |
b0ab6f92 | 302 | gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0) |
2cf6c36c JP |
303 | return -1; |
304 | ||
305 | return 0; | |
306 | } | |
307 | ||
1da177e4 LT |
308 | static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg) |
309 | { | |
310 | struct prio_sched_data *q = qdisc_priv(sch); | |
311 | int prio; | |
312 | ||
313 | if (arg->stop) | |
314 | return; | |
315 | ||
316 | for (prio = 0; prio < q->bands; prio++) { | |
317 | if (arg->count < arg->skip) { | |
318 | arg->count++; | |
319 | continue; | |
320 | } | |
cc7ec456 | 321 | if (arg->fn(sch, prio + 1, arg) < 0) { |
1da177e4 LT |
322 | arg->stop = 1; |
323 | break; | |
324 | } | |
325 | arg->count++; | |
326 | } | |
327 | } | |
328 | ||
6529eaba | 329 | static struct tcf_block *prio_tcf_block(struct Qdisc *sch, unsigned long cl) |
1da177e4 LT |
330 | { |
331 | struct prio_sched_data *q = qdisc_priv(sch); | |
332 | ||
333 | if (cl) | |
334 | return NULL; | |
6529eaba | 335 | return q->block; |
1da177e4 LT |
336 | } |
337 | ||
20fea08b | 338 | static const struct Qdisc_class_ops prio_class_ops = { |
1da177e4 LT |
339 | .graft = prio_graft, |
340 | .leaf = prio_leaf, | |
341 | .get = prio_get, | |
342 | .put = prio_put, | |
1da177e4 | 343 | .walk = prio_walk, |
6529eaba | 344 | .tcf_block = prio_tcf_block, |
1da177e4 LT |
345 | .bind_tcf = prio_bind, |
346 | .unbind_tcf = prio_put, | |
347 | .dump = prio_dump_class, | |
2cf6c36c | 348 | .dump_stats = prio_dump_class_stats, |
1da177e4 LT |
349 | }; |
350 | ||
20fea08b | 351 | static struct Qdisc_ops prio_qdisc_ops __read_mostly = { |
1da177e4 LT |
352 | .next = NULL, |
353 | .cl_ops = &prio_class_ops, | |
354 | .id = "prio", | |
355 | .priv_size = sizeof(struct prio_sched_data), | |
356 | .enqueue = prio_enqueue, | |
357 | .dequeue = prio_dequeue, | |
48a8f519 | 358 | .peek = prio_peek, |
1da177e4 LT |
359 | .init = prio_init, |
360 | .reset = prio_reset, | |
361 | .destroy = prio_destroy, | |
362 | .change = prio_tune, | |
363 | .dump = prio_dump, | |
364 | .owner = THIS_MODULE, | |
365 | }; | |
366 | ||
367 | static int __init prio_module_init(void) | |
368 | { | |
1d8ae3fd | 369 | return register_qdisc(&prio_qdisc_ops); |
1da177e4 LT |
370 | } |
371 | ||
10297b99 | 372 | static void __exit prio_module_exit(void) |
1da177e4 LT |
373 | { |
374 | unregister_qdisc(&prio_qdisc_ops); | |
375 | } | |
376 | ||
377 | module_init(prio_module_init) | |
378 | module_exit(prio_module_exit) | |
379 | ||
380 | MODULE_LICENSE("GPL"); |