NFS: Use local disk inode cache
[linux-2.6/verdex.git] / net / sched / sch_dsmark.c
blobd303daa45d49e54f34bc11d698d309253c5e77b2
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(qdisc_dev(sch), sch->dev_queue,
64 &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->bstats.bytes += qdisc_pkt_len(skb);
264 sch->bstats.packets++;
265 sch->q.qlen++;
267 return NET_XMIT_SUCCESS;
269 drop:
270 kfree_skb(skb);
271 sch->qstats.drops++;
272 return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
275 static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
277 struct dsmark_qdisc_data *p = qdisc_priv(sch);
278 struct sk_buff *skb;
279 u32 index;
281 pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
283 skb = p->q->ops->dequeue(p->q);
284 if (skb == NULL)
285 return NULL;
287 sch->q.qlen--;
289 index = skb->tc_index & (p->indices - 1);
290 pr_debug("index %d->%d\n", skb->tc_index, index);
292 switch (skb->protocol) {
293 case htons(ETH_P_IP):
294 ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
295 p->value[index]);
296 break;
297 case htons(ETH_P_IPV6):
298 ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
299 p->value[index]);
300 break;
301 default:
303 * Only complain if a change was actually attempted.
304 * This way, we can send non-IP traffic through dsmark
305 * and don't need yet another qdisc as a bypass.
307 if (p->mask[index] != 0xff || p->value[index])
308 printk(KERN_WARNING
309 "dsmark_dequeue: unsupported protocol %d\n",
310 ntohs(skb->protocol));
311 break;
314 return skb;
317 static struct sk_buff *dsmark_peek(struct Qdisc *sch)
319 struct dsmark_qdisc_data *p = qdisc_priv(sch);
321 pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", sch, p);
323 return p->q->ops->peek(p->q);
326 static unsigned int dsmark_drop(struct Qdisc *sch)
328 struct dsmark_qdisc_data *p = qdisc_priv(sch);
329 unsigned int len;
331 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
333 if (p->q->ops->drop == NULL)
334 return 0;
336 len = p->q->ops->drop(p->q);
337 if (len)
338 sch->q.qlen--;
340 return len;
343 static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
345 struct dsmark_qdisc_data *p = qdisc_priv(sch);
346 struct nlattr *tb[TCA_DSMARK_MAX + 1];
347 int err = -EINVAL;
348 u32 default_index = NO_DEFAULT_INDEX;
349 u16 indices;
350 u8 *mask;
352 pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
354 if (!opt)
355 goto errout;
357 err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
358 if (err < 0)
359 goto errout;
361 err = -EINVAL;
362 indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
364 if (hweight32(indices) != 1)
365 goto errout;
367 if (tb[TCA_DSMARK_DEFAULT_INDEX])
368 default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
370 mask = kmalloc(indices * 2, GFP_KERNEL);
371 if (mask == NULL) {
372 err = -ENOMEM;
373 goto errout;
376 p->mask = mask;
377 memset(p->mask, 0xff, indices);
379 p->value = p->mask + indices;
380 memset(p->value, 0, indices);
382 p->indices = indices;
383 p->default_index = default_index;
384 p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
386 p->q = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
387 &pfifo_qdisc_ops, sch->handle);
388 if (p->q == NULL)
389 p->q = &noop_qdisc;
391 pr_debug("dsmark_init: qdisc %p\n", p->q);
393 err = 0;
394 errout:
395 return err;
398 static void dsmark_reset(struct Qdisc *sch)
400 struct dsmark_qdisc_data *p = qdisc_priv(sch);
402 pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
403 qdisc_reset(p->q);
404 sch->q.qlen = 0;
407 static void dsmark_destroy(struct Qdisc *sch)
409 struct dsmark_qdisc_data *p = qdisc_priv(sch);
411 pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
413 tcf_destroy_chain(&p->filter_list);
414 qdisc_destroy(p->q);
415 kfree(p->mask);
418 static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
419 struct sk_buff *skb, struct tcmsg *tcm)
421 struct dsmark_qdisc_data *p = qdisc_priv(sch);
422 struct nlattr *opts = NULL;
424 pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
426 if (!dsmark_valid_index(p, cl))
427 return -EINVAL;
429 tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
430 tcm->tcm_info = p->q->handle;
432 opts = nla_nest_start(skb, TCA_OPTIONS);
433 if (opts == NULL)
434 goto nla_put_failure;
435 NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
436 NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
438 return nla_nest_end(skb, opts);
440 nla_put_failure:
441 nla_nest_cancel(skb, opts);
442 return -EMSGSIZE;
445 static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
447 struct dsmark_qdisc_data *p = qdisc_priv(sch);
448 struct nlattr *opts = NULL;
450 opts = nla_nest_start(skb, TCA_OPTIONS);
451 if (opts == NULL)
452 goto nla_put_failure;
453 NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
455 if (p->default_index != NO_DEFAULT_INDEX)
456 NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
458 if (p->set_tc_index)
459 NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
461 return nla_nest_end(skb, opts);
463 nla_put_failure:
464 nla_nest_cancel(skb, opts);
465 return -EMSGSIZE;
468 static const struct Qdisc_class_ops dsmark_class_ops = {
469 .graft = dsmark_graft,
470 .leaf = dsmark_leaf,
471 .get = dsmark_get,
472 .put = dsmark_put,
473 .change = dsmark_change,
474 .delete = dsmark_delete,
475 .walk = dsmark_walk,
476 .tcf_chain = dsmark_find_tcf,
477 .bind_tcf = dsmark_bind_filter,
478 .unbind_tcf = dsmark_put,
479 .dump = dsmark_dump_class,
482 static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
483 .next = NULL,
484 .cl_ops = &dsmark_class_ops,
485 .id = "dsmark",
486 .priv_size = sizeof(struct dsmark_qdisc_data),
487 .enqueue = dsmark_enqueue,
488 .dequeue = dsmark_dequeue,
489 .peek = dsmark_peek,
490 .drop = dsmark_drop,
491 .init = dsmark_init,
492 .reset = dsmark_reset,
493 .destroy = dsmark_destroy,
494 .change = NULL,
495 .dump = dsmark_dump,
496 .owner = THIS_MODULE,
499 static int __init dsmark_module_init(void)
501 return register_qdisc(&dsmark_qdisc_ops);
504 static void __exit dsmark_module_exit(void)
506 unregister_qdisc(&dsmark_qdisc_ops);
509 module_init(dsmark_module_init)
510 module_exit(dsmark_module_exit)
512 MODULE_LICENSE("GPL");