Merge branch 'akpm'
[linux-2.6/next.git] / net / sched / sch_dsmark.c
blob2c790204d042e3ac5ec4f2b322b8a0429628dbb5
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/slab.h>
9 #include <linux/types.h>
10 #include <linux/string.h>
11 #include <linux/errno.h>
12 #include <linux/skbuff.h>
13 #include <linux/rtnetlink.h>
14 #include <linux/bitops.h>
15 #include <net/pkt_sched.h>
16 #include <net/dsfield.h>
17 #include <net/inet_ecn.h>
18 #include <asm/byteorder.h>
21 * classid class marking
22 * ------- ----- -------
23 * n/a 0 n/a
24 * x:0 1 use entry [0]
25 * ... ... ...
26 * x:y y>0 y+1 use entry [y]
27 * ... ... ...
28 * x:indices-1 indices use entry [indices-1]
29 * ... ... ...
30 * x:y y+1 use entry [y & (indices-1)]
31 * ... ... ...
32 * 0xffff 0x10000 use entry [indices-1]
36 #define NO_DEFAULT_INDEX (1 << 16)
38 struct dsmark_qdisc_data {
39 struct Qdisc *q;
40 struct tcf_proto *filter_list;
41 u8 *mask; /* "owns" the array */
42 u8 *value;
43 u16 indices;
44 u32 default_index; /* index range is 0...0xffff */
45 int set_tc_index;
48 static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
50 return (index <= p->indices && index > 0);
53 /* ------------------------- Class/flow operations ------------------------- */
55 static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
56 struct Qdisc *new, struct Qdisc **old)
58 struct dsmark_qdisc_data *p = qdisc_priv(sch);
60 pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
61 sch, p, new, old);
63 if (new == NULL) {
64 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
65 sch->handle);
66 if (new == NULL)
67 new = &noop_qdisc;
70 sch_tree_lock(sch);
71 *old = p->q;
72 p->q = new;
73 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
74 qdisc_reset(*old);
75 sch_tree_unlock(sch);
77 return 0;
80 static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
82 struct dsmark_qdisc_data *p = qdisc_priv(sch);
83 return p->q;
86 static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
88 pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
89 sch, qdisc_priv(sch), classid);
91 return TC_H_MIN(classid) + 1;
94 static unsigned long dsmark_bind_filter(struct Qdisc *sch,
95 unsigned long parent, u32 classid)
97 return dsmark_get(sch, classid);
100 static void dsmark_put(struct Qdisc *sch, unsigned long cl)
104 static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
105 [TCA_DSMARK_INDICES] = { .type = NLA_U16 },
106 [TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
107 [TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
108 [TCA_DSMARK_MASK] = { .type = NLA_U8 },
109 [TCA_DSMARK_VALUE] = { .type = NLA_U8 },
112 static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
113 struct nlattr **tca, unsigned long *arg)
115 struct dsmark_qdisc_data *p = qdisc_priv(sch);
116 struct nlattr *opt = tca[TCA_OPTIONS];
117 struct nlattr *tb[TCA_DSMARK_MAX + 1];
118 int err = -EINVAL;
119 u8 mask = 0;
121 pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
122 "arg 0x%lx\n", sch, p, classid, parent, *arg);
124 if (!dsmark_valid_index(p, *arg)) {
125 err = -ENOENT;
126 goto errout;
129 if (!opt)
130 goto errout;
132 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
133 if (err < 0)
134 goto errout;
136 if (tb[TCA_DSMARK_MASK])
137 mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
139 if (tb[TCA_DSMARK_VALUE])
140 p->value[*arg - 1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
142 if (tb[TCA_DSMARK_MASK])
143 p->mask[*arg - 1] = mask;
145 err = 0;
147 errout:
148 return err;
151 static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
153 struct dsmark_qdisc_data *p = qdisc_priv(sch);
155 if (!dsmark_valid_index(p, arg))
156 return -EINVAL;
158 p->mask[arg - 1] = 0xff;
159 p->value[arg - 1] = 0;
161 return 0;
164 static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
166 struct dsmark_qdisc_data *p = qdisc_priv(sch);
167 int i;
169 pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
171 if (walker->stop)
172 return;
174 for (i = 0; i < p->indices; i++) {
175 if (p->mask[i] == 0xff && !p->value[i])
176 goto ignore;
177 if (walker->count >= walker->skip) {
178 if (walker->fn(sch, i + 1, walker) < 0) {
179 walker->stop = 1;
180 break;
183 ignore:
184 walker->count++;
188 static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
189 unsigned long cl)
191 struct dsmark_qdisc_data *p = qdisc_priv(sch);
192 return &p->filter_list;
195 /* --------------------------- Qdisc operations ---------------------------- */
197 static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
199 struct dsmark_qdisc_data *p = qdisc_priv(sch);
200 int err;
202 pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
204 if (p->set_tc_index) {
205 switch (skb->protocol) {
206 case htons(ETH_P_IP):
207 if (skb_cow_head(skb, sizeof(struct iphdr)))
208 goto drop;
210 skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
211 & ~INET_ECN_MASK;
212 break;
214 case htons(ETH_P_IPV6):
215 if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
216 goto drop;
218 skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
219 & ~INET_ECN_MASK;
220 break;
221 default:
222 skb->tc_index = 0;
223 break;
227 if (TC_H_MAJ(skb->priority) == sch->handle)
228 skb->tc_index = TC_H_MIN(skb->priority);
229 else {
230 struct tcf_result res;
231 int result = tc_classify(skb, p->filter_list, &res);
233 pr_debug("result %d class 0x%04x\n", result, res.classid);
235 switch (result) {
236 #ifdef CONFIG_NET_CLS_ACT
237 case TC_ACT_QUEUED:
238 case TC_ACT_STOLEN:
239 kfree_skb(skb);
240 return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
242 case TC_ACT_SHOT:
243 goto drop;
244 #endif
245 case TC_ACT_OK:
246 skb->tc_index = TC_H_MIN(res.classid);
247 break;
249 default:
250 if (p->default_index != NO_DEFAULT_INDEX)
251 skb->tc_index = p->default_index;
252 break;
256 err = qdisc_enqueue(skb, p->q);
257 if (err != NET_XMIT_SUCCESS) {
258 if (net_xmit_drop_count(err))
259 sch->qstats.drops++;
260 return err;
263 sch->q.qlen++;
265 return NET_XMIT_SUCCESS;
267 drop:
268 kfree_skb(skb);
269 sch->qstats.drops++;
270 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
273 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
275 struct dsmark_qdisc_data *p = qdisc_priv(sch);
276 struct sk_buff *skb;
277 u32 index;
279 pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
281 skb = p->q->ops->dequeue(p->q);
282 if (skb == NULL)
283 return NULL;
285 qdisc_bstats_update(sch, skb);
286 sch->q.qlen--;
288 index = skb->tc_index & (p->indices - 1);
289 pr_debug("index %d->%d\n", skb->tc_index, index);
291 switch (skb->protocol) {
292 case htons(ETH_P_IP):
293 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
294 p->value[index]);
295 break;
296 case htons(ETH_P_IPV6):
297 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
298 p->value[index]);
299 break;
300 default:
302 * Only complain if a change was actually attempted.
303 * This way, we can send non-IP traffic through dsmark
304 * and don't need yet another qdisc as a bypass.
306 if (p->mask[index] != 0xff || p->value[index])
307 pr_warning("dsmark_dequeue: unsupported protocol %d\n",
308 ntohs(skb->protocol));
309 break;
312 return skb;
315 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
317 struct dsmark_qdisc_data *p = qdisc_priv(sch);
319 pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", sch, p);
321 return p->q->ops->peek(p->q);
324 static unsigned int dsmark_drop(struct Qdisc *sch)
326 struct dsmark_qdisc_data *p = qdisc_priv(sch);
327 unsigned int len;
329 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
331 if (p->q->ops->drop == NULL)
332 return 0;
334 len = p->q->ops->drop(p->q);
335 if (len)
336 sch->q.qlen--;
338 return len;
341 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
343 struct dsmark_qdisc_data *p = qdisc_priv(sch);
344 struct nlattr *tb[TCA_DSMARK_MAX + 1];
345 int err = -EINVAL;
346 u32 default_index = NO_DEFAULT_INDEX;
347 u16 indices;
348 u8 *mask;
350 pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
352 if (!opt)
353 goto errout;
355 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
356 if (err < 0)
357 goto errout;
359 err = -EINVAL;
360 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
362 if (hweight32(indices) != 1)
363 goto errout;
365 if (tb[TCA_DSMARK_DEFAULT_INDEX])
366 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
368 mask = kmalloc(indices * 2, GFP_KERNEL);
369 if (mask == NULL) {
370 err = -ENOMEM;
371 goto errout;
374 p->mask = mask;
375 memset(p->mask, 0xff, indices);
377 p->value = p->mask + indices;
378 memset(p->value, 0, indices);
380 p->indices = indices;
381 p->default_index = default_index;
382 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
384 p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
385 if (p->q == NULL)
386 p->q = &noop_qdisc;
388 pr_debug("dsmark_init: qdisc %p\n", p->q);
390 err = 0;
391 errout:
392 return err;
395 static void dsmark_reset(struct Qdisc *sch)
397 struct dsmark_qdisc_data *p = qdisc_priv(sch);
399 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
400 qdisc_reset(p->q);
401 sch->q.qlen = 0;
404 static void dsmark_destroy(struct Qdisc *sch)
406 struct dsmark_qdisc_data *p = qdisc_priv(sch);
408 pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
410 tcf_destroy_chain(&p->filter_list);
411 qdisc_destroy(p->q);
412 kfree(p->mask);
415 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
416 struct sk_buff *skb, struct tcmsg *tcm)
418 struct dsmark_qdisc_data *p = qdisc_priv(sch);
419 struct nlattr *opts = NULL;
421 pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
423 if (!dsmark_valid_index(p, cl))
424 return -EINVAL;
426 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
427 tcm->tcm_info = p->q->handle;
429 opts = nla_nest_start(skb, TCA_OPTIONS);
430 if (opts == NULL)
431 goto nla_put_failure;
432 NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl - 1]);
433 NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]);
435 return nla_nest_end(skb, opts);
437 nla_put_failure:
438 nla_nest_cancel(skb, opts);
439 return -EMSGSIZE;
442 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
444 struct dsmark_qdisc_data *p = qdisc_priv(sch);
445 struct nlattr *opts = NULL;
447 opts = nla_nest_start(skb, TCA_OPTIONS);
448 if (opts == NULL)
449 goto nla_put_failure;
450 NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
452 if (p->default_index != NO_DEFAULT_INDEX)
453 NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
455 if (p->set_tc_index)
456 NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
458 return nla_nest_end(skb, opts);
460 nla_put_failure:
461 nla_nest_cancel(skb, opts);
462 return -EMSGSIZE;
465 static const struct Qdisc_class_ops dsmark_class_ops = {
466 .graft = dsmark_graft,
467 .leaf = dsmark_leaf,
468 .get = dsmark_get,
469 .put = dsmark_put,
470 .change = dsmark_change,
471 .delete = dsmark_delete,
472 .walk = dsmark_walk,
473 .tcf_chain = dsmark_find_tcf,
474 .bind_tcf = dsmark_bind_filter,
475 .unbind_tcf = dsmark_put,
476 .dump = dsmark_dump_class,
479 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
480 .next = NULL,
481 .cl_ops = &dsmark_class_ops,
482 .id = "dsmark",
483 .priv_size = sizeof(struct dsmark_qdisc_data),
484 .enqueue = dsmark_enqueue,
485 .dequeue = dsmark_dequeue,
486 .peek = dsmark_peek,
487 .drop = dsmark_drop,
488 .init = dsmark_init,
489 .reset = dsmark_reset,
490 .destroy = dsmark_destroy,
491 .change = NULL,
492 .dump = dsmark_dump,
493 .owner = THIS_MODULE,
496 static int __init dsmark_module_init(void)
498 return register_qdisc(&dsmark_qdisc_ops);
501 static void __exit dsmark_module_exit(void)
503 unregister_qdisc(&dsmark_qdisc_ops);
506 module_init(dsmark_module_init)
507 module_exit(dsmark_module_exit)
509 MODULE_LICENSE("GPL");