/net/sched/sch_dsmark.c

  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/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>
 19
 20/*
 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]
 33 */
 34
 35
 36#define NO_DEFAULT_INDEX	(1 << 16)
 37
 38struct 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;
 46};
 47
 48static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
 49{
 50	return (index <= p->indices && index > 0);
 51}
 52
 53/* ------------------------- Class/flow operations ------------------------- */
 54
 55static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
 56			struct Qdisc *new, struct Qdisc **old)
 57{
 58	struct dsmark_qdisc_data *p = qdisc_priv(sch);
 59
 60	pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
 61		sch, p, new, old);
 62
 63	if (new == NULL) {
 64		new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
 65					&pfifo_qdisc_ops,
 66					sch->handle);
 67		if (new == NULL)
 68			new = &noop_qdisc;
 69	}
 70
 71	sch_tree_lock(sch);
 72	*old = p->q;
 73	p->q = new;
 74	qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
 75	qdisc_reset(*old);
 76	sch_tree_unlock(sch);
 77
 78	return 0;
 79}
 80
 81static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
 82{
 83	struct dsmark_qdisc_data *p = qdisc_priv(sch);
 84	return p->q;
 85}
 86
 87static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
 88{
 89	pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
 90		sch, qdisc_priv(sch), classid);
 91
 92	return TC_H_MIN(classid) + 1;
 93}
 94
 95static unsigned long dsmark_bind_filter(struct Qdisc *sch,
 96					unsigned long parent, u32 classid)
 97{
 98	return dsmark_get(sch, classid);
 99}
100
101static void dsmark_put(struct Qdisc *sch, unsigned long cl)
102{
103}
104
105static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
106	[TCA_DSMARK_INDICES]		= { .type = NLA_U16 },
107	[TCA_DSMARK_DEFAULT_INDEX]	= { .type = NLA_U16 },
108	[TCA_DSMARK_SET_TC_INDEX]	= { .type = NLA_FLAG },
109	[TCA_DSMARK_MASK]		= { .type = NLA_U8 },
110	[TCA_DSMARK_VALUE]		= { .type = NLA_U8 },
111};
112
113static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
114			 struct nlattr **tca, unsigned long *arg)
115{
116	struct dsmark_qdisc_data *p = qdisc_priv(sch);
117	struct nlattr *opt = tca[TCA_OPTIONS];
118	struct nlattr *tb[TCA_DSMARK_MAX + 1];
119	int err = -EINVAL;
120	u8 mask = 0;
121
122	pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
123		"arg 0x%lx\n", sch, p, classid, parent, *arg);
124
125	if (!dsmark_valid_index(p, *arg)) {
126		err = -ENOENT;
127		goto errout;
128	}
129
130	if (!opt)
131		goto errout;
132
133	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
134	if (err < 0)
135		goto errout;
136
137	if (tb[TCA_DSMARK_MASK])
138		mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
139
140	if (tb[TCA_DSMARK_VALUE])
141		p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
142
143	if (tb[TCA_DSMARK_MASK])
144		p->mask[*arg-1] = mask;
145
146	err = 0;
147
148errout:
149	return err;
150}
151
152static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
153{
154	struct dsmark_qdisc_data *p = qdisc_priv(sch);
155
156	if (!dsmark_valid_index(p, arg))
157		return -EINVAL;
158
159	p->mask[arg-1] = 0xff;
160	p->value[arg-1] = 0;
161
162	return 0;
163}
164
165static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
166{
167	struct dsmark_qdisc_data *p = qdisc_priv(sch);
168	int i;
169
170	pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
171
172	if (walker->stop)
173		return;
174
175	for (i = 0; i < p->indices; i++) {
176		if (p->mask[i] == 0xff && !p->value[i])
177			goto ignore;
178		if (walker->count >= walker->skip) {
179			if (walker->fn(sch, i+1, walker) < 0) {
180				walker->stop = 1;
181				break;
182			}
183		}
184ignore:
185		walker->count++;
186	}
187}
188
189static inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
190						 unsigned long cl)
191{
192	struct dsmark_qdisc_data *p = qdisc_priv(sch);
193	return &p->filter_list;
194}
195
196/* --------------------------- Qdisc operations ---------------------------- */
197
198static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch)
199{
200	struct dsmark_qdisc_data *p = qdisc_priv(sch);
201	int err;
202
203	pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
204
205	if (p->set_tc_index) {
206		switch (skb->protocol) {
207		case htons(ETH_P_IP):
208			if (skb_cow_head(skb, sizeof(struct iphdr)))
209				goto drop;
210
211			skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
212				& ~INET_ECN_MASK;
213			break;
214
215		case htons(ETH_P_IPV6):
216			if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
217				goto drop;
218
219			skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
220				& ~INET_ECN_MASK;
221			break;
222		default:
223			skb->tc_index = 0;
224			break;
225		}
226	}
227
228	if (TC_H_MAJ(skb->priority) == sch->handle)
229		skb->tc_index = TC_H_MIN(skb->priority);
230	else {
231		struct tcf_result res;
232		int result = tc_classify(skb, p->filter_list, &res);
233
234		pr_debug("result %d class 0x%04x\n", result, res.classid);
235
236		switch (result) {
237#ifdef CONFIG_NET_CLS_ACT
238		case TC_ACT_QUEUED:
239		case TC_ACT_STOLEN:
240			kfree_skb(skb);
241			return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
242
243		case TC_ACT_SHOT:
244			goto drop;
245#endif
246		case TC_ACT_OK:
247			skb->tc_index = TC_H_MIN(res.classid);
248			break;
249
250		default:
251			if (p->default_index != NO_DEFAULT_INDEX)
252				skb->tc_index = p->default_index;
253			break;
254		}
255	}
256
257	err = qdisc_enqueue(skb, p->q);
258	if (err != NET_XMIT_SUCCESS) {
259		if (net_xmit_drop_count(err))
260			sch->qstats.drops++;
261		return err;
262	}
263
264	sch->bstats.bytes += qdisc_pkt_len(skb);
265	sch->bstats.packets++;
266	sch->q.qlen++;
267
268	return NET_XMIT_SUCCESS;
269
270drop:
271	kfree_skb(skb);
272	sch->qstats.drops++;
273	return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
274}
275
276static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
277{
278	struct dsmark_qdisc_data *p = qdisc_priv(sch);
279	struct sk_buff *skb;
280	u32 index;
281
282	pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
283
284	skb = p->q->ops->dequeue(p->q);
285	if (skb == NULL)
286		return NULL;
287
288	sch->q.qlen--;
289
290	index = skb->tc_index & (p->indices - 1);
291	pr_debug("index %d->%d\n", skb->tc_index, index);
292
293	switch (skb->protocol) {
294	case htons(ETH_P_IP):
295		ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
296				    p->value[index]);
297			break;
298	case htons(ETH_P_IPV6):
299		ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
300				    p->value[index]);
301			break;
302	default:
303		/*
304		 * Only complain if a change was actually attempted.
305		 * This way, we can send non-IP traffic through dsmark
306		 * and don't need yet another qdisc as a bypass.
307		 */
308		if (p->mask[index] != 0xff || p->value[index])
309			printk(KERN_WARNING
310			       "dsmark_dequeue: unsupported protocol %d\n",
311			       ntohs(skb->protocol));
312		break;
313	}
314
315	return skb;
316}
317
318static struct sk_buff *dsmark_peek(struct Qdisc *sch)
319{
320	struct dsmark_qdisc_data *p = qdisc_priv(sch);
321
322	pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", sch, p);
323
324	return p->q->ops->peek(p->q);
325}
326
327static unsigned int dsmark_drop(struct Qdisc *sch)
328{
329	struct dsmark_qdisc_data *p = qdisc_priv(sch);
330	unsigned int len;
331
332	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
333
334	if (p->q->ops->drop == NULL)
335		return 0;
336
337	len = p->q->ops->drop(p->q);
338	if (len)
339		sch->q.qlen--;
340
341	return len;
342}
343
344static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
345{
346	struct dsmark_qdisc_data *p = qdisc_priv(sch);
347	struct nlattr *tb[TCA_DSMARK_MAX + 1];
348	int err = -EINVAL;
349	u32 default_index = NO_DEFAULT_INDEX;
350	u16 indices;
351	u8 *mask;
352
353	pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
354
355	if (!opt)
356		goto errout;
357
358	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
359	if (err < 0)
360		goto errout;
361
362	err = -EINVAL;
363	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
364
365	if (hweight32(indices) != 1)
366		goto errout;
367
368	if (tb[TCA_DSMARK_DEFAULT_INDEX])
369		default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
370
371	mask = kmalloc(indices * 2, GFP_KERNEL);
372	if (mask == NULL) {
373		err = -ENOMEM;
374		goto errout;
375	}
376
377	p->mask = mask;
378	memset(p->mask, 0xff, indices);
379
380	p->value = p->mask + indices;
381	memset(p->value, 0, indices);
382
383	p->indices = indices;
384	p->default_index = default_index;
385	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
386
387	p->q = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
388				 &pfifo_qdisc_ops, sch->handle);
389	if (p->q == NULL)
390		p->q = &noop_qdisc;
391
392	pr_debug("dsmark_init: qdisc %p\n", p->q);
393
394	err = 0;
395errout:
396	return err;
397}
398
399static void dsmark_reset(struct Qdisc *sch)
400{
401	struct dsmark_qdisc_data *p = qdisc_priv(sch);
402
403	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
404	qdisc_reset(p->q);
405	sch->q.qlen = 0;
406}
407
408static void dsmark_destroy(struct Qdisc *sch)
409{
410	struct dsmark_qdisc_data *p = qdisc_priv(sch);
411
412	pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
413
414	tcf_destroy_chain(&p->filter_list);
415	qdisc_destroy(p->q);
416	kfree(p->mask);
417}
418
419static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
420			     struct sk_buff *skb, struct tcmsg *tcm)
421{
422	struct dsmark_qdisc_data *p = qdisc_priv(sch);
423	struct nlattr *opts = NULL;
424
425	pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
426
427	if (!dsmark_valid_index(p, cl))
428		return -EINVAL;
429
430	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
431	tcm->tcm_info = p->q->handle;
432
433	opts = nla_nest_start(skb, TCA_OPTIONS);
434	if (opts == NULL)
435		goto nla_put_failure;
436	NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
437	NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
438
439	return nla_nest_end(skb, opts);
440
441nla_put_failure:
442	nla_nest_cancel(skb, opts);
443	return -EMSGSIZE;
444}
445
446static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
447{
448	struct dsmark_qdisc_data *p = qdisc_priv(sch);
449	struct nlattr *opts = NULL;
450
451	opts = nla_nest_start(skb, TCA_OPTIONS);
452	if (opts == NULL)
453		goto nla_put_failure;
454	NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
455
456	if (p->default_index != NO_DEFAULT_INDEX)
457		NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
458
459	if (p->set_tc_index)
460		NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
461
462	return nla_nest_end(skb, opts);
463
464nla_put_failure:
465	nla_nest_cancel(skb, opts);
466	return -EMSGSIZE;
467}
468
469static const struct Qdisc_class_ops dsmark_class_ops = {
470	.graft		=	dsmark_graft,
471	.leaf		=	dsmark_leaf,
472	.get		=	dsmark_get,
473	.put		=	dsmark_put,
474	.change		=	dsmark_change,
475	.delete		=	dsmark_delete,
476	.walk		=	dsmark_walk,
477	.tcf_chain	=	dsmark_find_tcf,
478	.bind_tcf	=	dsmark_bind_filter,
479	.unbind_tcf	=	dsmark_put,
480	.dump		=	dsmark_dump_class,
481};
482
483static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
484	.next		=	NULL,
485	.cl_ops		=	&dsmark_class_ops,
486	.id		=	"dsmark",
487	.priv_size	=	sizeof(struct dsmark_qdisc_data),
488	.enqueue	=	dsmark_enqueue,
489	.dequeue	=	dsmark_dequeue,
490	.peek		=	dsmark_peek,
491	.drop		=	dsmark_drop,
492	.init		=	dsmark_init,
493	.reset		=	dsmark_reset,
494	.destroy	=	dsmark_destroy,
495	.change		=	NULL,
496	.dump		=	dsmark_dump,
497	.owner		=	THIS_MODULE,
498};
499
500static int __init dsmark_module_init(void)
501{
502	return register_qdisc(&dsmark_qdisc_ops);
503}
504
505static void __exit dsmark_module_exit(void)
506{
507	unregister_qdisc(&dsmark_qdisc_ops);
508}
509
510module_init(dsmark_module_init)
511module_exit(dsmark_module_exit)
512
513MODULE_LICENSE("GPL");