File: /usr/src/linux/net/sched/sch_dsmark.c
1 /* net/sched/sch_dsmark.c - Differentiated Services field marker */
2
3 /* Written 1998-2000 by Werner Almesberger, EPFL ICA */
4
5
6 #include <linux/config.h>
7 #include <linux/module.h>
8 #include <linux/types.h>
9 #include <linux/string.h>
10 #include <linux/errno.h>
11 #include <linux/skbuff.h>
12 #include <linux/netdevice.h> /* for pkt_sched */
13 #include <linux/rtnetlink.h>
14 #include <net/pkt_sched.h>
15 #include <net/dsfield.h>
16 #include <asm/byteorder.h>
17
18
19 #if 1 /* control */
20 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
21 #else
22 #define DPRINTK(format,args...)
23 #endif
24
25 #if 0 /* data */
26 #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
27 #else
28 #define D2PRINTK(format,args...)
29 #endif
30
31
32 #define PRIV(sch) ((struct dsmark_qdisc_data *) (sch)->data)
33
34
35 /*
36 * classid class marking
37 * ------- ----- -------
38 * n/a 0 n/a
39 * x:0 1 use entry [0]
40 * ... ... ...
41 * x:y y>0 y+1 use entry [y]
42 * ... ... ...
43 * x:indices-1 indices use entry [indices-1]
44 * ... ... ...
45 * x:y y+1 use entry [y & (indices-1)]
46 * ... ... ...
47 * 0xffff 0x10000 use entry [indices-1]
48 */
49
50
51 #define NO_DEFAULT_INDEX (1 << 16)
52
53 struct dsmark_qdisc_data {
54 struct Qdisc *q;
55 struct tcf_proto *filter_list;
56 __u8 *mask; /* "owns" the array */
57 __u8 *value;
58 __u16 indices;
59 __u32 default_index; /* index range is 0...0xffff */
60 int set_tc_index;
61 };
62
63
64 /* ------------------------- Class/flow operations ------------------------- */
65
66
67 static int dsmark_graft(struct Qdisc *sch,unsigned long arg,
68 struct Qdisc *new,struct Qdisc **old)
69 {
70 struct dsmark_qdisc_data *p = PRIV(sch);
71
72 DPRINTK("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",sch,p,new,
73 old);
74 if (!new)
75 new = &noop_qdisc;
76 sch_tree_lock(sch);
77 *old = xchg(&p->q,new);
78 if (*old)
79 qdisc_reset(*old);
80 sch_tree_unlock(sch); /* @@@ move up ? */
81 return 0;
82 }
83
84
85 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
86 {
87 struct dsmark_qdisc_data *p = PRIV(sch);
88
89 return p->q;
90 }
91
92
93 static unsigned long dsmark_get(struct Qdisc *sch,u32 classid)
94 {
95 struct dsmark_qdisc_data *p __attribute__((unused)) = PRIV(sch);
96
97 DPRINTK("dsmark_get(sch %p,[qdisc %p],classid %x)\n",sch,p,classid);
98 return TC_H_MIN(classid)+1;
99 }
100
101
102 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
103 unsigned long parent, u32 classid)
104 {
105 return dsmark_get(sch,classid);
106 }
107
108
109 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
110 {
111 }
112
113
114 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
115 struct rtattr **tca, unsigned long *arg)
116 {
117 struct dsmark_qdisc_data *p = PRIV(sch);
118 struct rtattr *opt = tca[TCA_OPTIONS-1];
119 struct rtattr *tb[TCA_DSMARK_MAX];
120
121 DPRINTK("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
122 "arg 0x%lx\n",sch,p,classid,parent,*arg);
123 if (*arg > p->indices)
124 return -ENOENT;
125 if (!opt || rtattr_parse(tb, TCA_DSMARK_MAX, RTA_DATA(opt),
126 RTA_PAYLOAD(opt)))
127 return -EINVAL;
128 if (tb[TCA_DSMARK_MASK-1]) {
129 if (!RTA_PAYLOAD(tb[TCA_DSMARK_MASK-1]))
130 return -EINVAL;
131 p->mask[*arg-1] = *(__u8 *) RTA_DATA(tb[TCA_DSMARK_MASK-1]);
132 }
133 if (tb[TCA_DSMARK_VALUE-1]) {
134 if (!RTA_PAYLOAD(tb[TCA_DSMARK_VALUE-1]))
135 return -EINVAL;
136 p->value[*arg-1] = *(__u8 *) RTA_DATA(tb[TCA_DSMARK_VALUE-1]);
137 }
138 return 0;
139 }
140
141
142 static int dsmark_delete(struct Qdisc *sch,unsigned long arg)
143 {
144 struct dsmark_qdisc_data *p = PRIV(sch);
145
146 if (!arg || arg > p->indices)
147 return -EINVAL;
148 p->mask[arg-1] = 0xff;
149 p->value[arg-1] = 0;
150 return 0;
151 }
152
153
154 static void dsmark_walk(struct Qdisc *sch,struct qdisc_walker *walker)
155 {
156 struct dsmark_qdisc_data *p = PRIV(sch);
157 int i;
158
159 DPRINTK("dsmark_walk(sch %p,[qdisc %p],walker %p)\n",sch,p,walker);
160 if (walker->stop)
161 return;
162 for (i = 0; i < p->indices; i++) {
163 if (p->mask[i] == 0xff && !p->value[i])
164 continue;
165 if (walker->count >= walker->skip) {
166 if (walker->fn(sch, i+1, walker) < 0) {
167 walker->stop = 1;
168 break;
169 }
170 }
171 walker->count++;
172 }
173 }
174
175
176 static struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,unsigned long cl)
177 {
178 struct dsmark_qdisc_data *p = PRIV(sch);
179
180 return &p->filter_list;
181 }
182
183
184 /* --------------------------- Qdisc operations ---------------------------- */
185
186
187 static int dsmark_enqueue(struct sk_buff *skb,struct Qdisc *sch)
188 {
189 struct dsmark_qdisc_data *p = PRIV(sch);
190 struct tcf_result res;
191 int result;
192 int ret = NET_XMIT_POLICED;
193
194 D2PRINTK("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
195 if (p->set_tc_index) {
196 switch (skb->protocol) {
197 case __constant_htons(ETH_P_IP):
198 skb->tc_index = ipv4_get_dsfield(skb->nh.iph);
199 break;
200 case __constant_htons(ETH_P_IPV6):
201 skb->tc_index = ipv6_get_dsfield(skb->nh.ipv6h);
202 break;
203 default:
204 skb->tc_index = 0;
205 break;
206 };
207 }
208 result = TC_POLICE_OK; /* be nice to gcc */
209 if (TC_H_MAJ(skb->priority) == sch->handle) {
210 skb->tc_index = TC_H_MIN(skb->priority);
211 } else {
212 result = tc_classify(skb,p->filter_list,&res);
213 D2PRINTK("result %d class 0x%04x\n",result,res.classid);
214 switch (result) {
215 #ifdef CONFIG_NET_CLS_POLICE
216 case TC_POLICE_SHOT:
217 kfree_skb(skb);
218 break;
219 #if 0
220 case TC_POLICE_RECLASSIFY:
221 /* FIXME: what to do here ??? */
222 #endif
223 #endif
224 case TC_POLICE_OK:
225 skb->tc_index = TC_H_MIN(res.classid);
226 break;
227 case TC_POLICE_UNSPEC:
228 /* fall through */
229 default:
230 if (p->default_index != NO_DEFAULT_INDEX)
231 skb->tc_index = p->default_index;
232 break;
233 };
234 }
235 if (
236 #ifdef CONFIG_NET_CLS_POLICE
237 result == TC_POLICE_SHOT ||
238 #endif
239
240 ((ret = p->q->enqueue(skb,p->q)) != 0)) {
241 sch->stats.drops++;
242 return ret;
243 }
244 sch->stats.bytes += skb->len;
245 sch->stats.packets++;
246 sch->q.qlen++;
247 return ret;
248 }
249
250
251 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
252 {
253 struct dsmark_qdisc_data *p = PRIV(sch);
254 struct sk_buff *skb;
255 int index;
256
257 D2PRINTK("dsmark_dequeue(sch %p,[qdisc %p])\n",sch,p);
258 skb = p->q->ops->dequeue(p->q);
259 if (!skb)
260 return NULL;
261 sch->q.qlen--;
262 index = skb->tc_index & (p->indices-1);
263 D2PRINTK("index %d->%d\n",skb->tc_index,index);
264 switch (skb->protocol) {
265 case __constant_htons(ETH_P_IP):
266 ipv4_change_dsfield(skb->nh.iph,
267 p->mask[index],p->value[index]);
268 break;
269 case __constant_htons(ETH_P_IPV6):
270 ipv6_change_dsfield(skb->nh.ipv6h,
271 p->mask[index],p->value[index]);
272 break;
273 default:
274 /*
275 * Only complain if a change was actually attempted.
276 * This way, we can send non-IP traffic through dsmark
277 * and don't need yet another qdisc as a bypass.
278 */
279 if (p->mask[index] != 0xff || p->value[index])
280 printk(KERN_WARNING "dsmark_dequeue: "
281 "unsupported protocol %d\n",
282 htons(skb->protocol));
283 break;
284 };
285 return skb;
286 }
287
288
289 static int dsmark_requeue(struct sk_buff *skb,struct Qdisc *sch)
290 {
291 int ret;
292 struct dsmark_qdisc_data *p = PRIV(sch);
293
294 D2PRINTK("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
295 if ((ret = p->q->ops->requeue(skb, p->q)) == 0) {
296 sch->q.qlen++;
297 return 0;
298 }
299 sch->stats.drops++;
300 return ret;
301 }
302
303
304 static int dsmark_drop(struct Qdisc *sch)
305 {
306 struct dsmark_qdisc_data *p = PRIV(sch);
307
308 DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n",sch,p);
309 if (!p->q->ops->drop)
310 return 0;
311 if (!p->q->ops->drop(p->q))
312 return 0;
313 sch->q.qlen--;
314 return 1;
315 }
316
317
318 int dsmark_init(struct Qdisc *sch,struct rtattr *opt)
319 {
320 struct dsmark_qdisc_data *p = PRIV(sch);
321 struct rtattr *tb[TCA_DSMARK_MAX];
322 __u16 tmp;
323
324 DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt);
325 if (rtattr_parse(tb,TCA_DSMARK_MAX,RTA_DATA(opt),RTA_PAYLOAD(opt)) < 0 ||
326 !tb[TCA_DSMARK_INDICES-1] ||
327 RTA_PAYLOAD(tb[TCA_DSMARK_INDICES-1]) < sizeof(__u16))
328 return -EINVAL;
329 memset(p,0,sizeof(*p));
330 p->filter_list = NULL;
331 p->indices = *(__u16 *) RTA_DATA(tb[TCA_DSMARK_INDICES-1]);
332 if (!p->indices)
333 return -EINVAL;
334 for (tmp = p->indices; tmp != 1; tmp >>= 1) {
335 if (tmp & 1)
336 return -EINVAL;
337 }
338 p->default_index = NO_DEFAULT_INDEX;
339 if (tb[TCA_DSMARK_DEFAULT_INDEX-1]) {
340 if (RTA_PAYLOAD(tb[TCA_DSMARK_DEFAULT_INDEX-1]) < sizeof(__u16))
341 return -EINVAL;
342 p->default_index =
343 *(__u16 *) RTA_DATA(tb[TCA_DSMARK_DEFAULT_INDEX-1]);
344 }
345 p->set_tc_index = !!tb[TCA_DSMARK_SET_TC_INDEX-1];
346 p->mask = kmalloc(p->indices*2,GFP_KERNEL);
347 if (!p->mask)
348 return -ENOMEM;
349 p->value = p->mask+p->indices;
350 memset(p->mask,0xff,p->indices);
351 memset(p->value,0,p->indices);
352 if (!(p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops)))
353 p->q = &noop_qdisc;
354 DPRINTK("dsmark_init: qdisc %p\n",&p->q);
355 MOD_INC_USE_COUNT;
356 return 0;
357 }
358
359
360 static void dsmark_reset(struct Qdisc *sch)
361 {
362 struct dsmark_qdisc_data *p = PRIV(sch);
363
364 DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n",sch,p);
365 qdisc_reset(p->q);
366 sch->q.qlen = 0;
367 }
368
369
370 static void dsmark_destroy(struct Qdisc *sch)
371 {
372 struct dsmark_qdisc_data *p = PRIV(sch);
373 struct tcf_proto *tp;
374
375 DPRINTK("dsmark_destroy(sch %p,[qdisc %p])\n",sch,p);
376 while (p->filter_list) {
377 tp = p->filter_list;
378 p->filter_list = tp->next;
379 tp->ops->destroy(tp);
380 }
381 qdisc_destroy(p->q);
382 p->q = &noop_qdisc;
383 kfree(p->mask);
384 MOD_DEC_USE_COUNT;
385 }
386
387
388 #ifdef CONFIG_RTNETLINK
389
390 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
391 struct sk_buff *skb, struct tcmsg *tcm)
392 {
393 struct dsmark_qdisc_data *p = PRIV(sch);
394 unsigned char *b = skb->tail;
395 struct rtattr *rta;
396
397 DPRINTK("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n",sch,p,cl);
398 if (!cl || cl > p->indices)
399 return -EINVAL;
400 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle),cl-1);
401 rta = (struct rtattr *) b;
402 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
403 RTA_PUT(skb,TCA_DSMARK_MASK,1,&p->mask[cl-1]);
404 RTA_PUT(skb,TCA_DSMARK_VALUE,1,&p->value[cl-1]);
405 rta->rta_len = skb->tail-b;
406 return skb->len;
407
408 rtattr_failure:
409 skb_trim(skb,b-skb->data);
410 return -1;
411 }
412
413 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
414 {
415 struct dsmark_qdisc_data *p = PRIV(sch);
416 unsigned char *b = skb->tail;
417 struct rtattr *rta;
418
419 rta = (struct rtattr *) b;
420 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
421 RTA_PUT(skb,TCA_DSMARK_INDICES,sizeof(__u16),&p->indices);
422 if (p->default_index != NO_DEFAULT_INDEX) {
423 __u16 tmp = p->default_index;
424
425 RTA_PUT(skb,TCA_DSMARK_DEFAULT_INDEX, sizeof(__u16), &tmp);
426 }
427 if (p->set_tc_index)
428 RTA_PUT(skb, TCA_DSMARK_SET_TC_INDEX, 0, NULL);
429 rta->rta_len = skb->tail-b;
430 return skb->len;
431
432 rtattr_failure:
433 skb_trim(skb,b-skb->data);
434 return -1;
435 }
436
437 #endif
438
439
440 static struct Qdisc_class_ops dsmark_class_ops =
441 {
442 dsmark_graft, /* graft */
443 dsmark_leaf, /* leaf */
444 dsmark_get, /* get */
445 dsmark_put, /* put */
446 dsmark_change, /* change */
447 dsmark_delete, /* delete */
448 dsmark_walk, /* walk */
449
450 dsmark_find_tcf, /* tcf_chain */
451 dsmark_bind_filter, /* bind_tcf */
452 dsmark_put, /* unbind_tcf */
453
454 #ifdef CONFIG_RTNETLINK
455 dsmark_dump_class, /* dump */
456 #endif
457 };
458
459 struct Qdisc_ops dsmark_qdisc_ops =
460 {
461 NULL, /* next */
462 &dsmark_class_ops, /* cl_ops */
463 "dsmark",
464 sizeof(struct dsmark_qdisc_data),
465
466 dsmark_enqueue, /* enqueue */
467 dsmark_dequeue, /* dequeue */
468 dsmark_requeue, /* requeue */
469 dsmark_drop, /* drop */
470
471 dsmark_init, /* init */
472 dsmark_reset, /* reset */
473 dsmark_destroy, /* destroy */
474 NULL, /* change */
475
476 #ifdef CONFIG_RTNETLINK
477 dsmark_dump /* dump */
478 #endif
479 };
480
481 #ifdef MODULE
482 int init_module(void)
483 {
484 return register_qdisc(&dsmark_qdisc_ops);
485 }
486
487
488 void cleanup_module(void)
489 {
490 unregister_qdisc(&dsmark_qdisc_ops);
491 }
492 #endif
493