Commit | Line | Data |
---|---|---|
1da177e4 LT |
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/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 <net/inet_ecn.h> | |
18 | #include <asm/byteorder.h> | |
19 | ||
20 | ||
08e9cd1f | 21 | #if 0 /* control */ |
1da177e4 LT |
22 | #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args) |
23 | #else | |
24 | #define DPRINTK(format,args...) | |
25 | #endif | |
26 | ||
27 | #if 0 /* data */ | |
28 | #define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args) | |
29 | #else | |
30 | #define D2PRINTK(format,args...) | |
31 | #endif | |
32 | ||
33 | ||
34 | #define PRIV(sch) qdisc_priv(sch) | |
35 | ||
36 | ||
37 | /* | |
38 | * classid class marking | |
39 | * ------- ----- ------- | |
40 | * n/a 0 n/a | |
41 | * x:0 1 use entry [0] | |
42 | * ... ... ... | |
43 | * x:y y>0 y+1 use entry [y] | |
44 | * ... ... ... | |
45 | * x:indices-1 indices use entry [indices-1] | |
46 | * ... ... ... | |
47 | * x:y y+1 use entry [y & (indices-1)] | |
48 | * ... ... ... | |
49 | * 0xffff 0x10000 use entry [indices-1] | |
50 | */ | |
51 | ||
52 | ||
53 | #define NO_DEFAULT_INDEX (1 << 16) | |
54 | ||
55 | struct dsmark_qdisc_data { | |
56 | struct Qdisc *q; | |
57 | struct tcf_proto *filter_list; | |
58 | __u8 *mask; /* "owns" the array */ | |
59 | __u8 *value; | |
60 | __u16 indices; | |
61 | __u32 default_index; /* index range is 0...0xffff */ | |
62 | int set_tc_index; | |
63 | }; | |
64 | ||
758cc43c TG |
65 | static inline int dsmark_valid_indices(u16 indices) |
66 | { | |
67 | while (indices != 1) { | |
68 | if (indices & 1) | |
69 | return 0; | |
70 | indices >>= 1; | |
71 | } | |
72 | ||
73 | return 1; | |
74 | } | |
75 | ||
76 | static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index) | |
77 | { | |
78 | return (index <= p->indices && index > 0); | |
79 | } | |
1da177e4 LT |
80 | |
81 | /* ------------------------- Class/flow operations ------------------------- */ | |
82 | ||
83 | ||
84 | static int dsmark_graft(struct Qdisc *sch,unsigned long arg, | |
85 | struct Qdisc *new,struct Qdisc **old) | |
86 | { | |
87 | struct dsmark_qdisc_data *p = PRIV(sch); | |
88 | ||
89 | DPRINTK("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",sch,p,new, | |
90 | old); | |
486b53e5 TG |
91 | |
92 | if (new == NULL) { | |
93 | new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops); | |
94 | if (new == NULL) | |
95 | new = &noop_qdisc; | |
96 | } | |
97 | ||
1da177e4 LT |
98 | sch_tree_lock(sch); |
99 | *old = xchg(&p->q,new); | |
100 | if (*old) | |
101 | qdisc_reset(*old); | |
102 | sch->q.qlen = 0; | |
103 | sch_tree_unlock(sch); /* @@@ move up ? */ | |
104 | return 0; | |
105 | } | |
106 | ||
107 | ||
108 | static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg) | |
109 | { | |
110 | struct dsmark_qdisc_data *p = PRIV(sch); | |
111 | ||
112 | return p->q; | |
113 | } | |
114 | ||
115 | ||
116 | static unsigned long dsmark_get(struct Qdisc *sch,u32 classid) | |
117 | { | |
118 | struct dsmark_qdisc_data *p __attribute__((unused)) = PRIV(sch); | |
119 | ||
120 | DPRINTK("dsmark_get(sch %p,[qdisc %p],classid %x)\n",sch,p,classid); | |
121 | return TC_H_MIN(classid)+1; | |
122 | } | |
123 | ||
124 | ||
125 | static unsigned long dsmark_bind_filter(struct Qdisc *sch, | |
126 | unsigned long parent, u32 classid) | |
127 | { | |
128 | return dsmark_get(sch,classid); | |
129 | } | |
130 | ||
131 | ||
132 | static void dsmark_put(struct Qdisc *sch, unsigned long cl) | |
133 | { | |
134 | } | |
135 | ||
136 | ||
137 | static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent, | |
758cc43c | 138 | struct rtattr **tca, unsigned long *arg) |
1da177e4 LT |
139 | { |
140 | struct dsmark_qdisc_data *p = PRIV(sch); | |
141 | struct rtattr *opt = tca[TCA_OPTIONS-1]; | |
142 | struct rtattr *tb[TCA_DSMARK_MAX]; | |
758cc43c TG |
143 | int err = -EINVAL; |
144 | u8 mask = 0; | |
1da177e4 LT |
145 | |
146 | DPRINTK("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x)," | |
758cc43c TG |
147 | "arg 0x%lx\n", sch, p, classid, parent, *arg); |
148 | ||
149 | if (!dsmark_valid_index(p, *arg)) { | |
150 | err = -ENOENT; | |
151 | goto rtattr_failure; | |
1da177e4 | 152 | } |
1da177e4 | 153 | |
758cc43c TG |
154 | if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt)) |
155 | goto rtattr_failure; | |
156 | ||
157 | if (tb[TCA_DSMARK_MASK-1]) | |
158 | mask = RTA_GET_U8(tb[TCA_DSMARK_MASK-1]); | |
159 | ||
160 | if (tb[TCA_DSMARK_VALUE-1]) | |
161 | p->value[*arg-1] = RTA_GET_U8(tb[TCA_DSMARK_VALUE-1]); | |
162 | ||
163 | if (tb[TCA_DSMARK_MASK-1]) | |
164 | p->mask[*arg-1] = mask; | |
165 | ||
166 | err = 0; | |
167 | ||
168 | rtattr_failure: | |
169 | return err; | |
170 | } | |
1da177e4 LT |
171 | |
172 | static int dsmark_delete(struct Qdisc *sch,unsigned long arg) | |
173 | { | |
174 | struct dsmark_qdisc_data *p = PRIV(sch); | |
175 | ||
176 | if (!arg || arg > p->indices) | |
177 | return -EINVAL; | |
178 | p->mask[arg-1] = 0xff; | |
179 | p->value[arg-1] = 0; | |
180 | return 0; | |
181 | } | |
182 | ||
183 | ||
184 | static void dsmark_walk(struct Qdisc *sch,struct qdisc_walker *walker) | |
185 | { | |
186 | struct dsmark_qdisc_data *p = PRIV(sch); | |
187 | int i; | |
188 | ||
189 | DPRINTK("dsmark_walk(sch %p,[qdisc %p],walker %p)\n",sch,p,walker); | |
190 | if (walker->stop) | |
191 | return; | |
192 | for (i = 0; i < p->indices; i++) { | |
193 | if (p->mask[i] == 0xff && !p->value[i]) | |
0451eb07 | 194 | goto ignore; |
1da177e4 LT |
195 | if (walker->count >= walker->skip) { |
196 | if (walker->fn(sch, i+1, walker) < 0) { | |
197 | walker->stop = 1; | |
198 | break; | |
199 | } | |
200 | } | |
0451eb07 TG |
201 | ignore: |
202 | walker->count++; | |
1da177e4 LT |
203 | } |
204 | } | |
205 | ||
206 | ||
207 | static struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,unsigned long cl) | |
208 | { | |
209 | struct dsmark_qdisc_data *p = PRIV(sch); | |
210 | ||
211 | return &p->filter_list; | |
212 | } | |
213 | ||
214 | ||
215 | /* --------------------------- Qdisc operations ---------------------------- */ | |
216 | ||
217 | ||
218 | static int dsmark_enqueue(struct sk_buff *skb,struct Qdisc *sch) | |
219 | { | |
220 | struct dsmark_qdisc_data *p = PRIV(sch); | |
221 | struct tcf_result res; | |
222 | int result; | |
223 | int ret = NET_XMIT_POLICED; | |
224 | ||
225 | D2PRINTK("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p); | |
226 | if (p->set_tc_index) { | |
227 | /* FIXME: Safe with non-linear skbs? --RR */ | |
228 | switch (skb->protocol) { | |
229 | case __constant_htons(ETH_P_IP): | |
230 | skb->tc_index = ipv4_get_dsfield(skb->nh.iph) | |
231 | & ~INET_ECN_MASK; | |
232 | break; | |
233 | case __constant_htons(ETH_P_IPV6): | |
234 | skb->tc_index = ipv6_get_dsfield(skb->nh.ipv6h) | |
235 | & ~INET_ECN_MASK; | |
236 | break; | |
237 | default: | |
238 | skb->tc_index = 0; | |
239 | break; | |
240 | }; | |
241 | } | |
242 | result = TC_POLICE_OK; /* be nice to gcc */ | |
243 | if (TC_H_MAJ(skb->priority) == sch->handle) { | |
244 | skb->tc_index = TC_H_MIN(skb->priority); | |
245 | } else { | |
246 | result = tc_classify(skb,p->filter_list,&res); | |
247 | D2PRINTK("result %d class 0x%04x\n",result,res.classid); | |
248 | switch (result) { | |
249 | #ifdef CONFIG_NET_CLS_POLICE | |
250 | case TC_POLICE_SHOT: | |
251 | kfree_skb(skb); | |
252 | break; | |
253 | #if 0 | |
254 | case TC_POLICE_RECLASSIFY: | |
255 | /* FIXME: what to do here ??? */ | |
256 | #endif | |
257 | #endif | |
258 | case TC_POLICE_OK: | |
259 | skb->tc_index = TC_H_MIN(res.classid); | |
260 | break; | |
261 | case TC_POLICE_UNSPEC: | |
262 | /* fall through */ | |
263 | default: | |
264 | if (p->default_index != NO_DEFAULT_INDEX) | |
265 | skb->tc_index = p->default_index; | |
266 | break; | |
267 | }; | |
268 | } | |
269 | if ( | |
270 | #ifdef CONFIG_NET_CLS_POLICE | |
271 | result == TC_POLICE_SHOT || | |
272 | #endif | |
273 | ||
274 | ((ret = p->q->enqueue(skb,p->q)) != 0)) { | |
275 | sch->qstats.drops++; | |
276 | return ret; | |
277 | } | |
278 | sch->bstats.bytes += skb->len; | |
279 | sch->bstats.packets++; | |
280 | sch->q.qlen++; | |
281 | return ret; | |
282 | } | |
283 | ||
284 | ||
285 | static struct sk_buff *dsmark_dequeue(struct Qdisc *sch) | |
286 | { | |
287 | struct dsmark_qdisc_data *p = PRIV(sch); | |
288 | struct sk_buff *skb; | |
289 | int index; | |
290 | ||
291 | D2PRINTK("dsmark_dequeue(sch %p,[qdisc %p])\n",sch,p); | |
292 | skb = p->q->ops->dequeue(p->q); | |
293 | if (!skb) | |
294 | return NULL; | |
295 | sch->q.qlen--; | |
296 | index = skb->tc_index & (p->indices-1); | |
297 | D2PRINTK("index %d->%d\n",skb->tc_index,index); | |
298 | switch (skb->protocol) { | |
299 | case __constant_htons(ETH_P_IP): | |
300 | ipv4_change_dsfield(skb->nh.iph, | |
301 | p->mask[index],p->value[index]); | |
302 | break; | |
303 | case __constant_htons(ETH_P_IPV6): | |
304 | ipv6_change_dsfield(skb->nh.ipv6h, | |
305 | p->mask[index],p->value[index]); | |
306 | break; | |
307 | default: | |
308 | /* | |
309 | * Only complain if a change was actually attempted. | |
310 | * This way, we can send non-IP traffic through dsmark | |
311 | * and don't need yet another qdisc as a bypass. | |
312 | */ | |
313 | if (p->mask[index] != 0xff || p->value[index]) | |
314 | printk(KERN_WARNING "dsmark_dequeue: " | |
315 | "unsupported protocol %d\n", | |
316 | htons(skb->protocol)); | |
317 | break; | |
318 | }; | |
319 | return skb; | |
320 | } | |
321 | ||
322 | ||
323 | static int dsmark_requeue(struct sk_buff *skb,struct Qdisc *sch) | |
324 | { | |
325 | int ret; | |
326 | struct dsmark_qdisc_data *p = PRIV(sch); | |
327 | ||
328 | D2PRINTK("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n",skb,sch,p); | |
329 | if ((ret = p->q->ops->requeue(skb, p->q)) == 0) { | |
330 | sch->q.qlen++; | |
331 | sch->qstats.requeues++; | |
332 | return 0; | |
333 | } | |
334 | sch->qstats.drops++; | |
335 | return ret; | |
336 | } | |
337 | ||
338 | ||
339 | static unsigned int dsmark_drop(struct Qdisc *sch) | |
340 | { | |
341 | struct dsmark_qdisc_data *p = PRIV(sch); | |
342 | unsigned int len; | |
343 | ||
344 | DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n",sch,p); | |
345 | if (!p->q->ops->drop) | |
346 | return 0; | |
347 | if (!(len = p->q->ops->drop(p->q))) | |
348 | return 0; | |
349 | sch->q.qlen--; | |
350 | return len; | |
351 | } | |
352 | ||
353 | ||
758cc43c | 354 | static int dsmark_init(struct Qdisc *sch, struct rtattr *opt) |
1da177e4 LT |
355 | { |
356 | struct dsmark_qdisc_data *p = PRIV(sch); | |
357 | struct rtattr *tb[TCA_DSMARK_MAX]; | |
758cc43c TG |
358 | int err = -EINVAL; |
359 | u32 default_index = NO_DEFAULT_INDEX; | |
360 | u16 indices; | |
361 | u8 *mask; | |
362 | ||
363 | DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt); | |
364 | ||
365 | if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt) < 0) | |
366 | goto errout; | |
367 | ||
368 | indices = RTA_GET_U16(tb[TCA_DSMARK_INDICES-1]); | |
369 | if (!indices || !dsmark_valid_indices(indices)) | |
370 | goto errout; | |
371 | ||
372 | if (tb[TCA_DSMARK_DEFAULT_INDEX-1]) | |
373 | default_index = RTA_GET_U16(tb[TCA_DSMARK_DEFAULT_INDEX-1]); | |
374 | ||
375 | mask = kmalloc(indices * 2, GFP_KERNEL); | |
376 | if (mask == NULL) { | |
377 | err = -ENOMEM; | |
378 | goto errout; | |
1da177e4 | 379 | } |
758cc43c TG |
380 | |
381 | p->mask = mask; | |
382 | memset(p->mask, 0xff, indices); | |
383 | ||
384 | p->value = p->mask + indices; | |
385 | memset(p->value, 0, indices); | |
386 | ||
387 | p->indices = indices; | |
388 | p->default_index = default_index; | |
389 | p->set_tc_index = RTA_GET_FLAG(tb[TCA_DSMARK_SET_TC_INDEX-1]); | |
390 | ||
391 | p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops); | |
392 | if (p->q == NULL) | |
1da177e4 | 393 | p->q = &noop_qdisc; |
758cc43c TG |
394 | |
395 | DPRINTK("dsmark_init: qdisc %p\n", p->q); | |
396 | ||
397 | err = 0; | |
398 | errout: | |
399 | rtattr_failure: | |
400 | return err; | |
1da177e4 LT |
401 | } |
402 | ||
403 | ||
404 | static void dsmark_reset(struct Qdisc *sch) | |
405 | { | |
406 | struct dsmark_qdisc_data *p = PRIV(sch); | |
407 | ||
408 | DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n",sch,p); | |
409 | qdisc_reset(p->q); | |
410 | sch->q.qlen = 0; | |
411 | } | |
412 | ||
413 | ||
414 | static void dsmark_destroy(struct Qdisc *sch) | |
415 | { | |
416 | struct dsmark_qdisc_data *p = PRIV(sch); | |
417 | struct tcf_proto *tp; | |
418 | ||
419 | DPRINTK("dsmark_destroy(sch %p,[qdisc %p])\n",sch,p); | |
420 | while (p->filter_list) { | |
421 | tp = p->filter_list; | |
422 | p->filter_list = tp->next; | |
423 | tcf_destroy(tp); | |
424 | } | |
425 | qdisc_destroy(p->q); | |
426 | kfree(p->mask); | |
427 | } | |
428 | ||
429 | ||
430 | static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl, | |
02f23f09 | 431 | struct sk_buff *skb, struct tcmsg *tcm) |
1da177e4 LT |
432 | { |
433 | struct dsmark_qdisc_data *p = PRIV(sch); | |
02f23f09 | 434 | struct rtattr *opts = NULL; |
1da177e4 | 435 | |
02f23f09 TG |
436 | DPRINTK("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl); |
437 | ||
438 | if (!dsmark_valid_index(p, cl)) | |
1da177e4 | 439 | return -EINVAL; |
02f23f09 TG |
440 | |
441 | tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1); | |
442 | ||
443 | opts = RTA_NEST(skb, TCA_OPTIONS); | |
444 | RTA_PUT_U8(skb,TCA_DSMARK_MASK, p->mask[cl-1]); | |
445 | RTA_PUT_U8(skb,TCA_DSMARK_VALUE, p->value[cl-1]); | |
446 | ||
447 | return RTA_NEST_END(skb, opts); | |
1da177e4 LT |
448 | |
449 | rtattr_failure: | |
02f23f09 | 450 | return RTA_NEST_CANCEL(skb, opts); |
1da177e4 LT |
451 | } |
452 | ||
453 | static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb) | |
454 | { | |
455 | struct dsmark_qdisc_data *p = PRIV(sch); | |
02f23f09 | 456 | struct rtattr *opts = NULL; |
1da177e4 | 457 | |
02f23f09 TG |
458 | opts = RTA_NEST(skb, TCA_OPTIONS); |
459 | RTA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices); | |
460 | ||
461 | if (p->default_index != NO_DEFAULT_INDEX) | |
462 | RTA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index); | |
1da177e4 | 463 | |
1da177e4 | 464 | if (p->set_tc_index) |
02f23f09 TG |
465 | RTA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX); |
466 | ||
467 | return RTA_NEST_END(skb, opts); | |
1da177e4 LT |
468 | |
469 | rtattr_failure: | |
02f23f09 | 470 | return RTA_NEST_CANCEL(skb, opts); |
1da177e4 LT |
471 | } |
472 | ||
473 | static struct Qdisc_class_ops dsmark_class_ops = { | |
474 | .graft = dsmark_graft, | |
475 | .leaf = dsmark_leaf, | |
476 | .get = dsmark_get, | |
477 | .put = dsmark_put, | |
478 | .change = dsmark_change, | |
479 | .delete = dsmark_delete, | |
480 | .walk = dsmark_walk, | |
481 | .tcf_chain = dsmark_find_tcf, | |
482 | .bind_tcf = dsmark_bind_filter, | |
483 | .unbind_tcf = dsmark_put, | |
484 | .dump = dsmark_dump_class, | |
485 | }; | |
486 | ||
487 | static struct Qdisc_ops dsmark_qdisc_ops = { | |
488 | .next = NULL, | |
489 | .cl_ops = &dsmark_class_ops, | |
490 | .id = "dsmark", | |
491 | .priv_size = sizeof(struct dsmark_qdisc_data), | |
492 | .enqueue = dsmark_enqueue, | |
493 | .dequeue = dsmark_dequeue, | |
494 | .requeue = dsmark_requeue, | |
495 | .drop = dsmark_drop, | |
496 | .init = dsmark_init, | |
497 | .reset = dsmark_reset, | |
498 | .destroy = dsmark_destroy, | |
499 | .change = NULL, | |
500 | .dump = dsmark_dump, | |
501 | .owner = THIS_MODULE, | |
502 | }; | |
503 | ||
504 | static int __init dsmark_module_init(void) | |
505 | { | |
506 | return register_qdisc(&dsmark_qdisc_ops); | |
507 | } | |
508 | static void __exit dsmark_module_exit(void) | |
509 | { | |
510 | unregister_qdisc(&dsmark_qdisc_ops); | |
511 | } | |
512 | module_init(dsmark_module_init) | |
513 | module_exit(dsmark_module_exit) | |
514 | MODULE_LICENSE("GPL"); |