pkt_sched: Add multiqueue handling to qdisc_graft().
[linux-2.6/linux-loongson.git] / net / sched / sch_dsmark.c
blob3aafbd17393abb8c86372cef999b893a9bb72b7e
1 /* net/sched/sch_dsmark.c - Differentiated Services field marker */
3 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
6 #include <linux/module.h>
7 #include <linux/init.h>
8 #include <linux/types.h>
9 #include <linux/string.h>
10 #include <linux/errno.h>
11 #include <linux/skbuff.h>
12 #include <linux/rtnetlink.h>
13 #include <linux/bitops.h>
14 #include <net/pkt_sched.h>
15 #include <net/dsfield.h>
16 #include <net/inet_ecn.h>
17 #include <asm/byteorder.h>
20 * classid class marking
21 * ------- ----- -------
22 * n/a 0 n/a
23 * x:0 1 use entry [0]
24 * ... ... ...
25 * x:y y>0 y+1 use entry [y]
26 * ... ... ...
27 * x:indices-1 indices use entry [indices-1]
28 * ... ... ...
29 * x:y y+1 use entry [y & (indices-1)]
30 * ... ... ...
31 * 0xffff 0x10000 use entry [indices-1]
35 #define NO_DEFAULT_INDEX (1 << 16)
37 struct dsmark_qdisc_data {
38 struct Qdisc *q;
39 struct tcf_proto *filter_list;
40 u8 *mask; /* "owns" the array */
41 u8 *value;
42 u16 indices;
43 u32 default_index; /* index range is 0...0xffff */
44 int set_tc_index;
47 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
49 return (index <= p->indices && index > 0);
52 /* ------------------------- Class/flow operations ------------------------- */
54 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
55 struct Qdisc *new, struct Qdisc **old)
57 struct dsmark_qdisc_data *p = qdisc_priv(sch);
59 pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
60 sch, p, new, old);
62 if (new == NULL) {
63 new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
64 &pfifo_qdisc_ops,
65 sch->handle);
66 if (new == NULL)
67 new = &noop_qdisc;
70 sch_tree_lock(sch);
71 *old = xchg(&p->q, new);
72 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
73 qdisc_reset(*old);
74 sch_tree_unlock(sch);
76 return 0;
79 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
81 struct dsmark_qdisc_data *p = qdisc_priv(sch);
82 return p->q;
85 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
87 pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
88 sch, qdisc_priv(sch), classid);
90 return TC_H_MIN(classid) + 1;
93 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
94 unsigned long parent, u32 classid)
96 return dsmark_get(sch, classid);
99 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
103 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
104 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
105 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
106 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
107 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
108 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
111 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
112 struct nlattr **tca, unsigned long *arg)
114 struct dsmark_qdisc_data *p = qdisc_priv(sch);
115 struct nlattr *opt = tca[TCA_OPTIONS];
116 struct nlattr *tb[TCA_DSMARK_MAX + 1];
117 int err = -EINVAL;
118 u8 mask = 0;
120 pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
121 "arg 0x%lx\n", sch, p, classid, parent, *arg);
123 if (!dsmark_valid_index(p, *arg)) {
124 err = -ENOENT;
125 goto errout;
128 if (!opt)
129 goto errout;
131 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
132 if (err < 0)
133 goto errout;
135 if (tb[TCA_DSMARK_MASK])
136 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
138 if (tb[TCA_DSMARK_VALUE])
139 p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
141 if (tb[TCA_DSMARK_MASK])
142 p->mask[*arg-1] = mask;
144 err = 0;
146 errout:
147 return err;
150 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
152 struct dsmark_qdisc_data *p = qdisc_priv(sch);
154 if (!dsmark_valid_index(p, arg))
155 return -EINVAL;
157 p->mask[arg-1] = 0xff;
158 p->value[arg-1] = 0;
160 return 0;
163 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
165 struct dsmark_qdisc_data *p = qdisc_priv(sch);
166 int i;
168 pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
170 if (walker->stop)
171 return;
173 for (i = 0; i < p->indices; i++) {
174 if (p->mask[i] == 0xff && !p->value[i])
175 goto ignore;
176 if (walker->count >= walker->skip) {
177 if (walker->fn(sch, i+1, walker) < 0) {
178 walker->stop = 1;
179 break;
182 ignore:
183 walker->count++;
187 static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
188 unsigned long cl)
190 struct dsmark_qdisc_data *p = qdisc_priv(sch);
191 return &p->filter_list;
194 /* --------------------------- Qdisc operations ---------------------------- */
196 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
198 struct dsmark_qdisc_data *p = qdisc_priv(sch);
199 int err;
201 pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
203 if (p->set_tc_index) {
204 switch (skb->protocol) {
205 case __constant_htons(ETH_P_IP):
206 if (skb_cow_head(skb, sizeof(struct iphdr)))
207 goto drop;
209 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
210 & ~INET_ECN_MASK;
211 break;
213 case __constant_htons(ETH_P_IPV6):
214 if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
215 goto drop;
217 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
218 & ~INET_ECN_MASK;
219 break;
220 default:
221 skb->tc_index = 0;
222 break;
226 if (TC_H_MAJ(skb->priority) == sch->handle)
227 skb->tc_index = TC_H_MIN(skb->priority);
228 else {
229 struct tcf_result res;
230 int result = tc_classify(skb, p->filter_list, &res);
232 pr_debug("result %d class 0x%04x\n", result, res.classid);
234 switch (result) {
235 #ifdef CONFIG_NET_CLS_ACT
236 case TC_ACT_QUEUED:
237 case TC_ACT_STOLEN:
238 kfree_skb(skb);
239 return NET_XMIT_SUCCESS;
241 case TC_ACT_SHOT:
242 goto drop;
243 #endif
244 case TC_ACT_OK:
245 skb->tc_index = TC_H_MIN(res.classid);
246 break;
248 default:
249 if (p->default_index != NO_DEFAULT_INDEX)
250 skb->tc_index = p->default_index;
251 break;
255 err = p->q->enqueue(skb, p->q);
256 if (err != NET_XMIT_SUCCESS) {
257 sch->qstats.drops++;
258 return err;
261 sch->bstats.bytes += skb->len;
262 sch->bstats.packets++;
263 sch->q.qlen++;
265 return NET_XMIT_SUCCESS;
267 drop:
268 kfree_skb(skb);
269 sch->qstats.drops++;
270 return NET_XMIT_BYPASS;
273 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
275 struct dsmark_qdisc_data *p = qdisc_priv(sch);
276 struct sk_buff *skb;
277 u32 index;
279 pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
281 skb = p->q->ops->dequeue(p->q);
282 if (skb == NULL)
283 return NULL;
285 sch->q.qlen--;
287 index = skb->tc_index & (p->indices - 1);
288 pr_debug("index %d->%d\n", skb->tc_index, index);
290 switch (skb->protocol) {
291 case __constant_htons(ETH_P_IP):
292 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
293 p->value[index]);
294 break;
295 case __constant_htons(ETH_P_IPV6):
296 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
297 p->value[index]);
298 break;
299 default:
301 * Only complain if a change was actually attempted.
302 * This way, we can send non-IP traffic through dsmark
303 * and don't need yet another qdisc as a bypass.
305 if (p->mask[index] != 0xff || p->value[index])
306 printk(KERN_WARNING
307 "dsmark_dequeue: unsupported protocol %d\n",
308 ntohs(skb->protocol));
309 break;
312 return skb;
315 static int dsmark_requeue(struct sk_buff *skb, struct Qdisc *sch)
317 struct dsmark_qdisc_data *p = qdisc_priv(sch);
318 int err;
320 pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
322 err = p->q->ops->requeue(skb, p->q);
323 if (err != NET_XMIT_SUCCESS) {
324 sch->qstats.drops++;
325 return err;
328 sch->q.qlen++;
329 sch->qstats.requeues++;
331 return NET_XMIT_SUCCESS;
334 static unsigned int dsmark_drop(struct Qdisc *sch)
336 struct dsmark_qdisc_data *p = qdisc_priv(sch);
337 unsigned int len;
339 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
341 if (p->q->ops->drop == NULL)
342 return 0;
344 len = p->q->ops->drop(p->q);
345 if (len)
346 sch->q.qlen--;
348 return len;
351 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
353 struct dsmark_qdisc_data *p = qdisc_priv(sch);
354 struct nlattr *tb[TCA_DSMARK_MAX + 1];
355 int err = -EINVAL;
356 u32 default_index = NO_DEFAULT_INDEX;
357 u16 indices;
358 u8 *mask;
360 pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
362 if (!opt)
363 goto errout;
365 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
366 if (err < 0)
367 goto errout;
369 err = -EINVAL;
370 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
372 if (hweight32(indices) != 1)
373 goto errout;
375 if (tb[TCA_DSMARK_DEFAULT_INDEX])
376 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
378 mask = kmalloc(indices * 2, GFP_KERNEL);
379 if (mask == NULL) {
380 err = -ENOMEM;
381 goto errout;
384 p->mask = mask;
385 memset(p->mask, 0xff, indices);
387 p->value = p->mask + indices;
388 memset(p->value, 0, indices);
390 p->indices = indices;
391 p->default_index = default_index;
392 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
394 p->q = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
395 &pfifo_qdisc_ops, sch->handle);
396 if (p->q == NULL)
397 p->q = &noop_qdisc;
399 pr_debug("dsmark_init: qdisc %p\n", p->q);
401 err = 0;
402 errout:
403 return err;
406 static void dsmark_reset(struct Qdisc *sch)
408 struct dsmark_qdisc_data *p = qdisc_priv(sch);
410 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
411 qdisc_reset(p->q);
412 sch->q.qlen = 0;
415 static void dsmark_destroy(struct Qdisc *sch)
417 struct dsmark_qdisc_data *p = qdisc_priv(sch);
419 pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
421 tcf_destroy_chain(&p->filter_list);
422 qdisc_destroy(p->q);
423 kfree(p->mask);
426 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
427 struct sk_buff *skb, struct tcmsg *tcm)
429 struct dsmark_qdisc_data *p = qdisc_priv(sch);
430 struct nlattr *opts = NULL;
432 pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
434 if (!dsmark_valid_index(p, cl))
435 return -EINVAL;
437 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
438 tcm->tcm_info = p->q->handle;
440 opts = nla_nest_start(skb, TCA_OPTIONS);
441 if (opts == NULL)
442 goto nla_put_failure;
443 NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
444 NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
446 return nla_nest_end(skb, opts);
448 nla_put_failure:
449 nla_nest_cancel(skb, opts);
450 return -EMSGSIZE;
453 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
455 struct dsmark_qdisc_data *p = qdisc_priv(sch);
456 struct nlattr *opts = NULL;
458 opts = nla_nest_start(skb, TCA_OPTIONS);
459 if (opts == NULL)
460 goto nla_put_failure;
461 NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
463 if (p->default_index != NO_DEFAULT_INDEX)
464 NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
466 if (p->set_tc_index)
467 NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
469 return nla_nest_end(skb, opts);
471 nla_put_failure:
472 nla_nest_cancel(skb, opts);
473 return -EMSGSIZE;
476 static const struct Qdisc_class_ops dsmark_class_ops = {
477 .graft = dsmark_graft,
478 .leaf = dsmark_leaf,
479 .get = dsmark_get,
480 .put = dsmark_put,
481 .change = dsmark_change,
482 .delete = dsmark_delete,
483 .walk = dsmark_walk,
484 .tcf_chain = dsmark_find_tcf,
485 .bind_tcf = dsmark_bind_filter,
486 .unbind_tcf = dsmark_put,
487 .dump = dsmark_dump_class,
490 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
491 .next = NULL,
492 .cl_ops = &dsmark_class_ops,
493 .id = "dsmark",
494 .priv_size = sizeof(struct dsmark_qdisc_data),
495 .enqueue = dsmark_enqueue,
496 .dequeue = dsmark_dequeue,
497 .requeue = dsmark_requeue,
498 .drop = dsmark_drop,
499 .init = dsmark_init,
500 .reset = dsmark_reset,
501 .destroy = dsmark_destroy,
502 .change = NULL,
503 .dump = dsmark_dump,
504 .owner = THIS_MODULE,
507 static int __init dsmark_module_init(void)
509 return register_qdisc(&dsmark_qdisc_ops);
512 static void __exit dsmark_module_exit(void)
514 unregister_qdisc(&dsmark_qdisc_ops);
517 module_init(dsmark_module_init)
518 module_exit(dsmark_module_exit)
520 MODULE_LICENSE("GPL");