initial commit with v2.6.9
[linux-2.6.9-moxart.git] / net / sched / sch_dsmark.c
blob28b61f0f87a7d1db4f402b8f4c300b78565feba9
1 /* net/sched/sch_dsmark.c - Differentiated Services field marker */
3 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
6 #include <linux/config.h>
7 #include <linux/module.h>
8 #include <linux/init.h>
9 #include <linux/types.h>
10 #include <linux/string.h>
11 #include <linux/errno.h>
12 #include <linux/skbuff.h>
13 #include <linux/netdevice.h> /* for pkt_sched */
14 #include <linux/rtnetlink.h>
15 #include <net/pkt_sched.h>
16 #include <net/dsfield.h>
17 #include <asm/byteorder.h>
20 #if 1 /* control */
21 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
22 #else
23 #define DPRINTK(format,args...)
24 #endif
26 #if 0 /* data */
27 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
28 #else
29 #define D2PRINTK(format,args...)
30 #endif
33 #define PRIV(sch) qdisc_priv(sch)
37 * classid class marking
38 * ------- ----- -------
39 * n/a 0 n/a
40 * x:0 1 use entry [0]
41 * ... ... ...
42 * x:y y>0 y+1 use entry [y]
43 * ... ... ...
44 * x:indices-1 indices use entry [indices-1]
45 * ... ... ...
46 * x:y y+1 use entry [y & (indices-1)]
47 * ... ... ...
48 * 0xffff 0x10000 use entry [indices-1]
52 #define NO_DEFAULT_INDEX (1 << 16)
54 struct dsmark_qdisc_data {
55 struct Qdisc *q;
56 struct tcf_proto *filter_list;
57 __u8 *mask; /* "owns" the array */
58 __u8 *value;
59 __u16 indices;
60 __u32 default_index; /* index range is 0...0xffff */
61 int set_tc_index;
65 /* ------------------------- Class/flow operations ------------------------- */
68 static int dsmark_graft(struct Qdisc *sch,unsigned long arg,
69 struct Qdisc *new,struct Qdisc **old)
71 struct dsmark_qdisc_data *p = PRIV(sch);
73 DPRINTK("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",sch,p,new,
74 old);
75 if (!new)
76 new = &noop_qdisc;
77 sch_tree_lock(sch);
78 *old = xchg(&p->q,new);
79 if (*old)
80 qdisc_reset(*old);
81 sch->q.qlen = 0;
82 sch_tree_unlock(sch); /* @@@ move up ? */
83 return 0;
87 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
89 struct dsmark_qdisc_data *p = PRIV(sch);
91 return p->q;
95 static unsigned long dsmark_get(struct Qdisc *sch,u32 classid)
97 struct dsmark_qdisc_data *p __attribute__((unused)) = PRIV(sch);
99 DPRINTK("dsmark_get(sch %p,[qdisc %p],classid %x)\n",sch,p,classid);
100 return TC_H_MIN(classid)+1;
104 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
105 unsigned long parent, u32 classid)
107 return dsmark_get(sch,classid);
111 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
116 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
117 struct rtattr **tca, unsigned long *arg)
119 struct dsmark_qdisc_data *p = PRIV(sch);
120 struct rtattr *opt = tca[TCA_OPTIONS-1];
121 struct rtattr *tb[TCA_DSMARK_MAX];
123 DPRINTK("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
124 "arg 0x%lx\n",sch,p,classid,parent,*arg);
125 if (*arg > p->indices)
126 return -ENOENT;
127 if (!opt || rtattr_parse(tb, TCA_DSMARK_MAX, RTA_DATA(opt),
128 RTA_PAYLOAD(opt)))
129 return -EINVAL;
130 if (tb[TCA_DSMARK_MASK-1]) {
131 if (!RTA_PAYLOAD(tb[TCA_DSMARK_MASK-1]))
132 return -EINVAL;
133 p->mask[*arg-1] = *(__u8 *) RTA_DATA(tb[TCA_DSMARK_MASK-1]);
135 if (tb[TCA_DSMARK_VALUE-1]) {
136 if (!RTA_PAYLOAD(tb[TCA_DSMARK_VALUE-1]))
137 return -EINVAL;
138 p->value[*arg-1] = *(__u8 *) RTA_DATA(tb[TCA_DSMARK_VALUE-1]);
140 return 0;
144 static int dsmark_delete(struct Qdisc *sch,unsigned long arg)
146 struct dsmark_qdisc_data *p = PRIV(sch);
148 if (!arg || arg > p->indices)
149 return -EINVAL;
150 p->mask[arg-1] = 0xff;
151 p->value[arg-1] = 0;
152 return 0;
156 static void dsmark_walk(struct Qdisc *sch,struct qdisc_walker *walker)
158 struct dsmark_qdisc_data *p = PRIV(sch);
159 int i;
161 DPRINTK("dsmark_walk(sch %p,[qdisc %p],walker %p)\n",sch,p,walker);
162 if (walker->stop)
163 return;
164 for (i = 0; i < p->indices; i++) {
165 if (p->mask[i] == 0xff && !p->value[i])
166 continue;
167 if (walker->count >= walker->skip) {
168 if (walker->fn(sch, i+1, walker) < 0) {
169 walker->stop = 1;
170 break;
173 walker->count++;
178 static struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,unsigned long cl)
180 struct dsmark_qdisc_data *p = PRIV(sch);
182 return &p->filter_list;
186 /* --------------------------- Qdisc operations ---------------------------- */
189 static int dsmark_enqueue(struct sk_buff *skb,struct Qdisc *sch)
191 struct dsmark_qdisc_data *p = PRIV(sch);
192 struct tcf_result res;
193 int result;
194 int ret = NET_XMIT_POLICED;
196 D2PRINTK("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
197 if (p->set_tc_index) {
198 /* FIXME: Safe with non-linear skbs? --RR */
199 switch (skb->protocol) {
200 case __constant_htons(ETH_P_IP):
201 skb->tc_index = ipv4_get_dsfield(skb->nh.iph);
202 break;
203 case __constant_htons(ETH_P_IPV6):
204 skb->tc_index = ipv6_get_dsfield(skb->nh.ipv6h);
205 break;
206 default:
207 skb->tc_index = 0;
208 break;
211 result = TC_POLICE_OK; /* be nice to gcc */
212 if (TC_H_MAJ(skb->priority) == sch->handle) {
213 skb->tc_index = TC_H_MIN(skb->priority);
214 } else {
215 result = tc_classify(skb,p->filter_list,&res);
216 D2PRINTK("result %d class 0x%04x\n",result,res.classid);
217 switch (result) {
218 #ifdef CONFIG_NET_CLS_POLICE
219 case TC_POLICE_SHOT:
220 kfree_skb(skb);
221 break;
222 #if 0
223 case TC_POLICE_RECLASSIFY:
224 /* FIXME: what to do here ??? */
225 #endif
226 #endif
227 case TC_POLICE_OK:
228 skb->tc_index = TC_H_MIN(res.classid);
229 break;
230 case TC_POLICE_UNSPEC:
231 /* fall through */
232 default:
233 if (p->default_index != NO_DEFAULT_INDEX)
234 skb->tc_index = p->default_index;
235 break;
238 if (
239 #ifdef CONFIG_NET_CLS_POLICE
240 result == TC_POLICE_SHOT ||
241 #endif
243 ((ret = p->q->enqueue(skb,p->q)) != 0)) {
244 sch->stats.drops++;
245 return ret;
247 sch->stats.bytes += skb->len;
248 sch->stats.packets++;
249 sch->q.qlen++;
250 return ret;
254 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
256 struct dsmark_qdisc_data *p = PRIV(sch);
257 struct sk_buff *skb;
258 int index;
260 D2PRINTK("dsmark_dequeue(sch %p,[qdisc %p])\n",sch,p);
261 skb = p->q->ops->dequeue(p->q);
262 if (!skb)
263 return NULL;
264 sch->q.qlen--;
265 index = skb->tc_index & (p->indices-1);
266 D2PRINTK("index %d->%d\n",skb->tc_index,index);
267 switch (skb->protocol) {
268 case __constant_htons(ETH_P_IP):
269 ipv4_change_dsfield(skb->nh.iph,
270 p->mask[index],p->value[index]);
271 break;
272 case __constant_htons(ETH_P_IPV6):
273 ipv6_change_dsfield(skb->nh.ipv6h,
274 p->mask[index],p->value[index]);
275 break;
276 default:
278 * Only complain if a change was actually attempted.
279 * This way, we can send non-IP traffic through dsmark
280 * and don't need yet another qdisc as a bypass.
282 if (p->mask[index] != 0xff || p->value[index])
283 printk(KERN_WARNING "dsmark_dequeue: "
284 "unsupported protocol %d\n",
285 htons(skb->protocol));
286 break;
288 return skb;
292 static int dsmark_requeue(struct sk_buff *skb,struct Qdisc *sch)
294 int ret;
295 struct dsmark_qdisc_data *p = PRIV(sch);
297 D2PRINTK("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
298 if ((ret = p->q->ops->requeue(skb, p->q)) == 0) {
299 sch->q.qlen++;
300 return 0;
302 sch->stats.drops++;
303 return ret;
307 static unsigned int dsmark_drop(struct Qdisc *sch)
309 struct dsmark_qdisc_data *p = PRIV(sch);
310 unsigned int len;
312 DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n",sch,p);
313 if (!p->q->ops->drop)
314 return 0;
315 if (!(len = p->q->ops->drop(p->q)))
316 return 0;
317 sch->q.qlen--;
318 return len;
322 int dsmark_init(struct Qdisc *sch,struct rtattr *opt)
324 struct dsmark_qdisc_data *p = PRIV(sch);
325 struct rtattr *tb[TCA_DSMARK_MAX];
326 __u16 tmp;
328 DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt);
329 if (!opt ||
330 rtattr_parse(tb,TCA_DSMARK_MAX,RTA_DATA(opt),RTA_PAYLOAD(opt)) < 0 ||
331 !tb[TCA_DSMARK_INDICES-1] ||
332 RTA_PAYLOAD(tb[TCA_DSMARK_INDICES-1]) < sizeof(__u16))
333 return -EINVAL;
334 p->indices = *(__u16 *) RTA_DATA(tb[TCA_DSMARK_INDICES-1]);
335 if (!p->indices)
336 return -EINVAL;
337 for (tmp = p->indices; tmp != 1; tmp >>= 1) {
338 if (tmp & 1)
339 return -EINVAL;
341 p->default_index = NO_DEFAULT_INDEX;
342 if (tb[TCA_DSMARK_DEFAULT_INDEX-1]) {
343 if (RTA_PAYLOAD(tb[TCA_DSMARK_DEFAULT_INDEX-1]) < sizeof(__u16))
344 return -EINVAL;
345 p->default_index =
346 *(__u16 *) RTA_DATA(tb[TCA_DSMARK_DEFAULT_INDEX-1]);
348 p->set_tc_index = !!tb[TCA_DSMARK_SET_TC_INDEX-1];
349 p->mask = kmalloc(p->indices*2,GFP_KERNEL);
350 if (!p->mask)
351 return -ENOMEM;
352 p->value = p->mask+p->indices;
353 memset(p->mask,0xff,p->indices);
354 memset(p->value,0,p->indices);
355 if (!(p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops)))
356 p->q = &noop_qdisc;
357 DPRINTK("dsmark_init: qdisc %p\n",&p->q);
358 return 0;
362 static void dsmark_reset(struct Qdisc *sch)
364 struct dsmark_qdisc_data *p = PRIV(sch);
366 DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n",sch,p);
367 qdisc_reset(p->q);
368 sch->q.qlen = 0;
372 static void dsmark_destroy(struct Qdisc *sch)
374 struct dsmark_qdisc_data *p = PRIV(sch);
375 struct tcf_proto *tp;
377 DPRINTK("dsmark_destroy(sch %p,[qdisc %p])\n",sch,p);
378 while (p->filter_list) {
379 tp = p->filter_list;
380 p->filter_list = tp->next;
381 tcf_destroy(tp);
383 qdisc_destroy(p->q);
384 kfree(p->mask);
388 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
389 struct sk_buff *skb, struct tcmsg *tcm)
391 struct dsmark_qdisc_data *p = PRIV(sch);
392 unsigned char *b = skb->tail;
393 struct rtattr *rta;
395 DPRINTK("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n",sch,p,cl);
396 if (!cl || cl > p->indices)
397 return -EINVAL;
398 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle),cl-1);
399 rta = (struct rtattr *) b;
400 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
401 RTA_PUT(skb,TCA_DSMARK_MASK,1,&p->mask[cl-1]);
402 RTA_PUT(skb,TCA_DSMARK_VALUE,1,&p->value[cl-1]);
403 rta->rta_len = skb->tail-b;
404 return skb->len;
406 rtattr_failure:
407 skb_trim(skb,b-skb->data);
408 return -1;
411 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
413 struct dsmark_qdisc_data *p = PRIV(sch);
414 unsigned char *b = skb->tail;
415 struct rtattr *rta;
417 rta = (struct rtattr *) b;
418 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
419 RTA_PUT(skb,TCA_DSMARK_INDICES,sizeof(__u16),&p->indices);
420 if (p->default_index != NO_DEFAULT_INDEX) {
421 __u16 tmp = p->default_index;
423 RTA_PUT(skb,TCA_DSMARK_DEFAULT_INDEX, sizeof(__u16), &tmp);
425 if (p->set_tc_index)
426 RTA_PUT(skb, TCA_DSMARK_SET_TC_INDEX, 0, NULL);
427 rta->rta_len = skb->tail-b;
428 return skb->len;
430 rtattr_failure:
431 skb_trim(skb,b-skb->data);
432 return -1;
435 static struct Qdisc_class_ops dsmark_class_ops = {
436 .graft = dsmark_graft,
437 .leaf = dsmark_leaf,
438 .get = dsmark_get,
439 .put = dsmark_put,
440 .change = dsmark_change,
441 .delete = dsmark_delete,
442 .walk = dsmark_walk,
443 .tcf_chain = dsmark_find_tcf,
444 .bind_tcf = dsmark_bind_filter,
445 .unbind_tcf = dsmark_put,
446 .dump = dsmark_dump_class,
449 static struct Qdisc_ops dsmark_qdisc_ops = {
450 .next = NULL,
451 .cl_ops = &dsmark_class_ops,
452 .id = "dsmark",
453 .priv_size = sizeof(struct dsmark_qdisc_data),
454 .enqueue = dsmark_enqueue,
455 .dequeue = dsmark_dequeue,
456 .requeue = dsmark_requeue,
457 .drop = dsmark_drop,
458 .init = dsmark_init,
459 .reset = dsmark_reset,
460 .destroy = dsmark_destroy,
461 .change = NULL,
462 .dump = dsmark_dump,
463 .owner = THIS_MODULE,
466 static int __init dsmark_module_init(void)
468 return register_qdisc(&dsmark_qdisc_ops);
470 static void __exit dsmark_module_exit(void)
472 unregister_qdisc(&dsmark_qdisc_ops);
474 module_init(dsmark_module_init)
475 module_exit(dsmark_module_exit)
476 MODULE_LICENSE("GPL");