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->q.qlen = 0;
81 sch_tree_unlock(sch); /* @@@ move up ? */
82 return 0;
83 }
84
85
86 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
87 {
88 struct dsmark_qdisc_data *p = PRIV(sch);
89
90 return p->q;
91 }
92
93
94 static unsigned long dsmark_get(struct Qdisc *sch,u32 classid)
95 {
96 struct dsmark_qdisc_data *p __attribute__((unused)) = PRIV(sch);
97
98 DPRINTK("dsmark_get(sch %p,[qdisc %p],classid %x)\n",sch,p,classid);
99 return TC_H_MIN(classid)+1;
100 }
101
102
103 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
104 unsigned long parent, u32 classid)
105 {
106 return dsmark_get(sch,classid);
107 }
108
109
110 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
111 {
112 }
113
114
115 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
116 struct rtattr **tca, unsigned long *arg)
117 {
118 struct dsmark_qdisc_data *p = PRIV(sch);
119 struct rtattr *opt = tca[TCA_OPTIONS-1];
120 struct rtattr *tb[TCA_DSMARK_MAX];
121
122 DPRINTK("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
123 "arg 0x%lx\n",sch,p,classid,parent,*arg);
124 if (*arg > p->indices)
125 return -ENOENT;
126 if (!opt || rtattr_parse(tb, TCA_DSMARK_MAX, RTA_DATA(opt),
127 RTA_PAYLOAD(opt)))
128 return -EINVAL;
129 if (tb[TCA_DSMARK_MASK-1]) {
130 if (!RTA_PAYLOAD(tb[TCA_DSMARK_MASK-1]))
131 return -EINVAL;
132 p->mask[*arg-1] = *(__u8 *) RTA_DATA(tb[TCA_DSMARK_MASK-1]);
133 }
134 if (tb[TCA_DSMARK_VALUE-1]) {
135 if (!RTA_PAYLOAD(tb[TCA_DSMARK_VALUE-1]))
136 return -EINVAL;
137 p->value[*arg-1] = *(__u8 *) RTA_DATA(tb[TCA_DSMARK_VALUE-1]);
138 }
139 return 0;
140 }
141
142
143 static int dsmark_delete(struct Qdisc *sch,unsigned long arg)
144 {
145 struct dsmark_qdisc_data *p = PRIV(sch);
146
147 if (!arg || arg > p->indices)
148 return -EINVAL;
149 p->mask[arg-1] = 0xff;
150 p->value[arg-1] = 0;
151 return 0;
152 }
153
154
155 static void dsmark_walk(struct Qdisc *sch,struct qdisc_walker *walker)
156 {
157 struct dsmark_qdisc_data *p = PRIV(sch);
158 int i;
159
160 DPRINTK("dsmark_walk(sch %p,[qdisc %p],walker %p)\n",sch,p,walker);
161 if (walker->stop)
162 return;
163 for (i = 0; i < p->indices; i++) {
164 if (p->mask[i] == 0xff && !p->value[i])
165 continue;
166 if (walker->count >= walker->skip) {
167 if (walker->fn(sch, i+1, walker) < 0) {
168 walker->stop = 1;
169 break;
170 }
171 }
172 walker->count++;
173 }
174 }
175
176
177 static struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,unsigned long cl)
178 {
179 struct dsmark_qdisc_data *p = PRIV(sch);
180
181 return &p->filter_list;
182 }
183
184
185 /* --------------------------- Qdisc operations ---------------------------- */
186
187
188 static int dsmark_enqueue(struct sk_buff *skb,struct Qdisc *sch)
189 {
190 struct dsmark_qdisc_data *p = PRIV(sch);
191 struct tcf_result res;
192 int result;
193 int ret = NET_XMIT_POLICED;
194
195 D2PRINTK("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
196 if (p->set_tc_index) {
197 /* FIXME: Safe with non-linear skbs? --RR */
198 switch (skb->protocol) {
199 case __constant_htons(ETH_P_IP):
200 skb->tc_index = ipv4_get_dsfield(skb->nh.iph);
201 break;
202 case __constant_htons(ETH_P_IPV6):
203 skb->tc_index = ipv6_get_dsfield(skb->nh.ipv6h);
204 break;
205 default:
206 skb->tc_index = 0;
207 break;
208 };
209 }
210 result = TC_POLICE_OK; /* be nice to gcc */
211 if (TC_H_MAJ(skb->priority) == sch->handle) {
212 skb->tc_index = TC_H_MIN(skb->priority);
213 } else {
214 result = tc_classify(skb,p->filter_list,&res);
215 D2PRINTK("result %d class 0x%04x\n",result,res.classid);
216 switch (result) {
217 #ifdef CONFIG_NET_CLS_POLICE
218 case TC_POLICE_SHOT:
219 kfree_skb(skb);
220 break;
221 #if 0
222 case TC_POLICE_RECLASSIFY:
223 /* FIXME: what to do here ??? */
224 #endif
225 #endif
226 case TC_POLICE_OK:
227 skb->tc_index = TC_H_MIN(res.classid);
228 break;
229 case TC_POLICE_UNSPEC:
230 /* fall through */
231 default:
232 if (p->default_index != NO_DEFAULT_INDEX)
233 skb->tc_index = p->default_index;
234 break;
235 };
236 }
237 if (
238 #ifdef CONFIG_NET_CLS_POLICE
239 result == TC_POLICE_SHOT ||
240 #endif
241
242 ((ret = p->q->enqueue(skb,p->q)) != 0)) {
243 sch->stats.drops++;
244 return ret;
245 }
246 sch->stats.bytes += skb->len;
247 sch->stats.packets++;
248 sch->q.qlen++;
249 return ret;
250 }
251
252
253 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
254 {
255 struct dsmark_qdisc_data *p = PRIV(sch);
256 struct sk_buff *skb;
257 int index;
258
259 D2PRINTK("dsmark_dequeue(sch %p,[qdisc %p])\n",sch,p);
260 skb = p->q->ops->dequeue(p->q);
261 if (!skb)
262 return NULL;
263 sch->q.qlen--;
264 index = skb->tc_index & (p->indices-1);
265 D2PRINTK("index %d->%d\n",skb->tc_index,index);
266 switch (skb->protocol) {
267 case __constant_htons(ETH_P_IP):
268 ipv4_change_dsfield(skb->nh.iph,
269 p->mask[index],p->value[index]);
270 break;
271 case __constant_htons(ETH_P_IPV6):
272 ipv6_change_dsfield(skb->nh.ipv6h,
273 p->mask[index],p->value[index]);
274 break;
275 default:
276 /*
277 * Only complain if a change was actually attempted.
278 * This way, we can send non-IP traffic through dsmark
279 * and don't need yet another qdisc as a bypass.
280 */
281 if (p->mask[index] != 0xff || p->value[index])
282 printk(KERN_WARNING "dsmark_dequeue: "
283 "unsupported protocol %d\n",
284 htons(skb->protocol));
285 break;
286 };
287 return skb;
288 }
289
290
291 static int dsmark_requeue(struct sk_buff *skb,struct Qdisc *sch)
292 {
293 int ret;
294 struct dsmark_qdisc_data *p = PRIV(sch);
295
296 D2PRINTK("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p);
297 if ((ret = p->q->ops->requeue(skb, p->q)) == 0) {
298 sch->q.qlen++;
299 return 0;
300 }
301 sch->stats.drops++;
302 return ret;
303 }
304
305
306 static unsigned int dsmark_drop(struct Qdisc *sch)
307 {
308 struct dsmark_qdisc_data *p = PRIV(sch);
309 unsigned int len;
310
311 DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n",sch,p);
312 if (!p->q->ops->drop)
313 return 0;
314 if (!(len = p->q->ops->drop(p->q)))
315 return 0;
316 sch->q.qlen--;
317 return len;
318 }
319
320
321 int dsmark_init(struct Qdisc *sch,struct rtattr *opt)
322 {
323 struct dsmark_qdisc_data *p = PRIV(sch);
324 struct rtattr *tb[TCA_DSMARK_MAX];
325 __u16 tmp;
326
327 DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n",sch,p,opt);
328 if (rtattr_parse(tb,TCA_DSMARK_MAX,RTA_DATA(opt),RTA_PAYLOAD(opt)) < 0 ||
329 !tb[TCA_DSMARK_INDICES-1] ||
330 RTA_PAYLOAD(tb[TCA_DSMARK_INDICES-1]) < sizeof(__u16))
331 return -EINVAL;
332 memset(p,0,sizeof(*p));
333 p->filter_list = NULL;
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;
340 }
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]);
347 }
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;
359 }
360
361
362 static void dsmark_reset(struct Qdisc *sch)
363 {
364 struct dsmark_qdisc_data *p = PRIV(sch);
365
366 DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n",sch,p);
367 qdisc_reset(p->q);
368 sch->q.qlen = 0;
369 }
370
371
372 static void dsmark_destroy(struct Qdisc *sch)
373 {
374 struct dsmark_qdisc_data *p = PRIV(sch);
375 struct tcf_proto *tp;
376
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 tp->ops->destroy(tp);
382 }
383 qdisc_destroy(p->q);
384 p->q = &noop_qdisc;
385 kfree(p->mask);
386 }
387
388
389 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
390 struct sk_buff *skb, struct tcmsg *tcm)
391 {
392 struct dsmark_qdisc_data *p = PRIV(sch);
393 unsigned char *b = skb->tail;
394 struct rtattr *rta;
395
396 DPRINTK("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n",sch,p,cl);
397 if (!cl || cl > p->indices)
398 return -EINVAL;
399 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle),cl-1);
400 rta = (struct rtattr *) b;
401 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
402 RTA_PUT(skb,TCA_DSMARK_MASK,1,&p->mask[cl-1]);
403 RTA_PUT(skb,TCA_DSMARK_VALUE,1,&p->value[cl-1]);
404 rta->rta_len = skb->tail-b;
405 return skb->len;
406
407 rtattr_failure:
408 skb_trim(skb,b-skb->data);
409 return -1;
410 }
411
412 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
413 {
414 struct dsmark_qdisc_data *p = PRIV(sch);
415 unsigned char *b = skb->tail;
416 struct rtattr *rta;
417
418 rta = (struct rtattr *) b;
419 RTA_PUT(skb,TCA_OPTIONS,0,NULL);
420 RTA_PUT(skb,TCA_DSMARK_INDICES,sizeof(__u16),&p->indices);
421 if (p->default_index != NO_DEFAULT_INDEX) {
422 __u16 tmp = p->default_index;
423
424 RTA_PUT(skb,TCA_DSMARK_DEFAULT_INDEX, sizeof(__u16), &tmp);
425 }
426 if (p->set_tc_index)
427 RTA_PUT(skb, TCA_DSMARK_SET_TC_INDEX, 0, NULL);
428 rta->rta_len = skb->tail-b;
429 return skb->len;
430
431 rtattr_failure:
432 skb_trim(skb,b-skb->data);
433 return -1;
434 }
435
436 static struct Qdisc_class_ops dsmark_class_ops = {
437 .graft = dsmark_graft,
438 .leaf = dsmark_leaf,
439 .get = dsmark_get,
440 .put = dsmark_put,
441 .change = dsmark_change,
442 .delete = dsmark_delete,
443 .walk = dsmark_walk,
444 .tcf_chain = dsmark_find_tcf,
445 .bind_tcf = dsmark_bind_filter,
446 .unbind_tcf = dsmark_put,
447 .dump = dsmark_dump_class,
448 };
449
450 struct Qdisc_ops dsmark_qdisc_ops = {
451 .next = NULL,
452 .cl_ops = &dsmark_class_ops,
453 .id = "dsmark",
454 .priv_size = sizeof(struct dsmark_qdisc_data),
455 .enqueue = dsmark_enqueue,
456 .dequeue = dsmark_dequeue,
457 .requeue = dsmark_requeue,
458 .drop = dsmark_drop,
459 .init = dsmark_init,
460 .reset = dsmark_reset,
461 .destroy = dsmark_destroy,
462 .change = NULL,
463 .dump = dsmark_dump,
464 .owner = THIS_MODULE,
465 };
466
467 #ifdef MODULE
468 int init_module(void)
469 {
470 return register_qdisc(&dsmark_qdisc_ops);
471 }
472
473
474 void cleanup_module(void)
475 {
476 unregister_qdisc(&dsmark_qdisc_ops);
477 }
478 #endif
479 MODULE_LICENSE("GPL");
480
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.