Linux 2.6.26
[linux-2.6/kmemtrace.git] / net / sched / sch_dsmark.c
blobc4c1317cd47db91c56310db02bf933f3b521e722
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(sch->dev, &pfifo_qdisc_ops,
64 sch->handle);
65 if (new == NULL)
66 new = &noop_qdisc;
69 sch_tree_lock(sch);
70 *old = xchg(&p->q, new);
71 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
72 qdisc_reset(*old);
73 sch_tree_unlock(sch);
75 return 0;
78 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
80 struct dsmark_qdisc_data *p = qdisc_priv(sch);
81 return p->q;
84 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
86 pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
87 sch, qdisc_priv(sch), classid);
89 return TC_H_MIN(classid) + 1;
92 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
93 unsigned long parent, u32 classid)
95 return dsmark_get(sch, classid);
98 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
102 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
103 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
104 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
105 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
106 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
107 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
110 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
111 struct nlattr **tca, unsigned long *arg)
113 struct dsmark_qdisc_data *p = qdisc_priv(sch);
114 struct nlattr *opt = tca[TCA_OPTIONS];
115 struct nlattr *tb[TCA_DSMARK_MAX + 1];
116 int err = -EINVAL;
117 u8 mask = 0;
119 pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
120 "arg 0x%lx\n", sch, p, classid, parent, *arg);
122 if (!dsmark_valid_index(p, *arg)) {
123 err = -ENOENT;
124 goto errout;
127 if (!opt)
128 goto errout;
130 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
131 if (err < 0)
132 goto errout;
134 if (tb[TCA_DSMARK_MASK])
135 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
137 if (tb[TCA_DSMARK_VALUE])
138 p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
140 if (tb[TCA_DSMARK_MASK])
141 p->mask[*arg-1] = mask;
143 err = 0;
145 errout:
146 return err;
149 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
151 struct dsmark_qdisc_data *p = qdisc_priv(sch);
153 if (!dsmark_valid_index(p, arg))
154 return -EINVAL;
156 p->mask[arg-1] = 0xff;
157 p->value[arg-1] = 0;
159 return 0;
162 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
164 struct dsmark_qdisc_data *p = qdisc_priv(sch);
165 int i;
167 pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
169 if (walker->stop)
170 return;
172 for (i = 0; i < p->indices; i++) {
173 if (p->mask[i] == 0xff && !p->value[i])
174 goto ignore;
175 if (walker->count >= walker->skip) {
176 if (walker->fn(sch, i+1, walker) < 0) {
177 walker->stop = 1;
178 break;
181 ignore:
182 walker->count++;
186 static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
187 unsigned long cl)
189 struct dsmark_qdisc_data *p = qdisc_priv(sch);
190 return &p->filter_list;
193 /* --------------------------- Qdisc operations ---------------------------- */
195 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
197 struct dsmark_qdisc_data *p = qdisc_priv(sch);
198 int err;
200 pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
202 if (p->set_tc_index) {
203 switch (skb->protocol) {
204 case __constant_htons(ETH_P_IP):
205 if (skb_cow_head(skb, sizeof(struct iphdr)))
206 goto drop;
208 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
209 & ~INET_ECN_MASK;
210 break;
212 case __constant_htons(ETH_P_IPV6):
213 if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
214 goto drop;
216 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
217 & ~INET_ECN_MASK;
218 break;
219 default:
220 skb->tc_index = 0;
221 break;
225 if (TC_H_MAJ(skb->priority) == sch->handle)
226 skb->tc_index = TC_H_MIN(skb->priority);
227 else {
228 struct tcf_result res;
229 int result = tc_classify(skb, p->filter_list, &res);
231 pr_debug("result %d class 0x%04x\n", result, res.classid);
233 switch (result) {
234 #ifdef CONFIG_NET_CLS_ACT
235 case TC_ACT_QUEUED:
236 case TC_ACT_STOLEN:
237 kfree_skb(skb);
238 return NET_XMIT_SUCCESS;
240 case TC_ACT_SHOT:
241 goto drop;
242 #endif
243 case TC_ACT_OK:
244 skb->tc_index = TC_H_MIN(res.classid);
245 break;
247 default:
248 if (p->default_index != NO_DEFAULT_INDEX)
249 skb->tc_index = p->default_index;
250 break;
254 err = p->q->enqueue(skb, p->q);
255 if (err != NET_XMIT_SUCCESS) {
256 sch->qstats.drops++;
257 return err;
260 sch->bstats.bytes += skb->len;
261 sch->bstats.packets++;
262 sch->q.qlen++;
264 return NET_XMIT_SUCCESS;
266 drop:
267 kfree_skb(skb);
268 sch->qstats.drops++;
269 return NET_XMIT_BYPASS;
272 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
274 struct dsmark_qdisc_data *p = qdisc_priv(sch);
275 struct sk_buff *skb;
276 u32 index;
278 pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
280 skb = p->q->ops->dequeue(p->q);
281 if (skb == NULL)
282 return NULL;
284 sch->q.qlen--;
286 index = skb->tc_index & (p->indices - 1);
287 pr_debug("index %d->%d\n", skb->tc_index, index);
289 switch (skb->protocol) {
290 case __constant_htons(ETH_P_IP):
291 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
292 p->value[index]);
293 break;
294 case __constant_htons(ETH_P_IPV6):
295 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
296 p->value[index]);
297 break;
298 default:
300 * Only complain if a change was actually attempted.
301 * This way, we can send non-IP traffic through dsmark
302 * and don't need yet another qdisc as a bypass.
304 if (p->mask[index] != 0xff || p->value[index])
305 printk(KERN_WARNING
306 "dsmark_dequeue: unsupported protocol %d\n",
307 ntohs(skb->protocol));
308 break;
311 return skb;
314 static int dsmark_requeue(struct sk_buff *skb, struct Qdisc *sch)
316 struct dsmark_qdisc_data *p = qdisc_priv(sch);
317 int err;
319 pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
321 err = p->q->ops->requeue(skb, p->q);
322 if (err != NET_XMIT_SUCCESS) {
323 sch->qstats.drops++;
324 return err;
327 sch->q.qlen++;
328 sch->qstats.requeues++;
330 return NET_XMIT_SUCCESS;
333 static unsigned int dsmark_drop(struct Qdisc *sch)
335 struct dsmark_qdisc_data *p = qdisc_priv(sch);
336 unsigned int len;
338 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
340 if (p->q->ops->drop == NULL)
341 return 0;
343 len = p->q->ops->drop(p->q);
344 if (len)
345 sch->q.qlen--;
347 return len;
350 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
352 struct dsmark_qdisc_data *p = qdisc_priv(sch);
353 struct nlattr *tb[TCA_DSMARK_MAX + 1];
354 int err = -EINVAL;
355 u32 default_index = NO_DEFAULT_INDEX;
356 u16 indices;
357 u8 *mask;
359 pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
361 if (!opt)
362 goto errout;
364 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
365 if (err < 0)
366 goto errout;
368 err = -EINVAL;
369 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
371 if (hweight32(indices) != 1)
372 goto errout;
374 if (tb[TCA_DSMARK_DEFAULT_INDEX])
375 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
377 mask = kmalloc(indices * 2, GFP_KERNEL);
378 if (mask == NULL) {
379 err = -ENOMEM;
380 goto errout;
383 p->mask = mask;
384 memset(p->mask, 0xff, indices);
386 p->value = p->mask + indices;
387 memset(p->value, 0, indices);
389 p->indices = indices;
390 p->default_index = default_index;
391 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
393 p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
394 if (p->q == NULL)
395 p->q = &noop_qdisc;
397 pr_debug("dsmark_init: qdisc %p\n", p->q);
399 err = 0;
400 errout:
401 return err;
404 static void dsmark_reset(struct Qdisc *sch)
406 struct dsmark_qdisc_data *p = qdisc_priv(sch);
408 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
409 qdisc_reset(p->q);
410 sch->q.qlen = 0;
413 static void dsmark_destroy(struct Qdisc *sch)
415 struct dsmark_qdisc_data *p = qdisc_priv(sch);
417 pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
419 tcf_destroy_chain(&p->filter_list);
420 qdisc_destroy(p->q);
421 kfree(p->mask);
424 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
425 struct sk_buff *skb, struct tcmsg *tcm)
427 struct dsmark_qdisc_data *p = qdisc_priv(sch);
428 struct nlattr *opts = NULL;
430 pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
432 if (!dsmark_valid_index(p, cl))
433 return -EINVAL;
435 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
436 tcm->tcm_info = p->q->handle;
438 opts = nla_nest_start(skb, TCA_OPTIONS);
439 if (opts == NULL)
440 goto nla_put_failure;
441 NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
442 NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
444 return nla_nest_end(skb, opts);
446 nla_put_failure:
447 nla_nest_cancel(skb, opts);
448 return -EMSGSIZE;
451 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
453 struct dsmark_qdisc_data *p = qdisc_priv(sch);
454 struct nlattr *opts = NULL;
456 opts = nla_nest_start(skb, TCA_OPTIONS);
457 if (opts == NULL)
458 goto nla_put_failure;
459 NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
461 if (p->default_index != NO_DEFAULT_INDEX)
462 NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
464 if (p->set_tc_index)
465 NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
467 return nla_nest_end(skb, opts);
469 nla_put_failure:
470 nla_nest_cancel(skb, opts);
471 return -EMSGSIZE;
474 static const struct Qdisc_class_ops dsmark_class_ops = {
475 .graft = dsmark_graft,
476 .leaf = dsmark_leaf,
477 .get = dsmark_get,
478 .put = dsmark_put,
479 .change = dsmark_change,
480 .delete = dsmark_delete,
481 .walk = dsmark_walk,
482 .tcf_chain = dsmark_find_tcf,
483 .bind_tcf = dsmark_bind_filter,
484 .unbind_tcf = dsmark_put,
485 .dump = dsmark_dump_class,
488 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
489 .next = NULL,
490 .cl_ops = &dsmark_class_ops,
491 .id = "dsmark",
492 .priv_size = sizeof(struct dsmark_qdisc_data),
493 .enqueue = dsmark_enqueue,
494 .dequeue = dsmark_dequeue,
495 .requeue = dsmark_requeue,
496 .drop = dsmark_drop,
497 .init = dsmark_init,
498 .reset = dsmark_reset,
499 .destroy = dsmark_destroy,
500 .change = NULL,
501 .dump = dsmark_dump,
502 .owner = THIS_MODULE,
505 static int __init dsmark_module_init(void)
507 return register_qdisc(&dsmark_qdisc_ops);
510 static void __exit dsmark_module_exit(void)
512 unregister_qdisc(&dsmark_qdisc_ops);
515 module_init(dsmark_module_init)
516 module_exit(dsmark_module_exit)
518 MODULE_LICENSE("GPL");