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 __rcu	*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("%s(sch %p,[qdisc %p],new %p,old %p)\n",
61		 __func__, sch, p, new, old);
62
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;
68	}
69
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);
76
77	return 0;
78}
79
80static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
81{
82	struct dsmark_qdisc_data *p = qdisc_priv(sch);
83	return p->q;
84}
85
86static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
87{
88	pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
89		 __func__, sch, qdisc_priv(sch), classid);
90
91	return TC_H_MIN(classid) + 1;
92}
93
94static unsigned long dsmark_bind_filter(struct Qdisc *sch,
95					unsigned long parent, u32 classid)
96{
97	return dsmark_get(sch, classid);
98}
99
100static void dsmark_put(struct Qdisc *sch, unsigned long cl)
101{
102}
103
104static 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 },
110};
111
112static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
113			 struct nlattr **tca, unsigned long *arg)
114{
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;
120
121	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
122		 __func__, sch, p, classid, parent, *arg);
123
124	if (!dsmark_valid_index(p, *arg)) {
125		err = -ENOENT;
126		goto errout;
127	}
128
129	if (!opt)
130		goto errout;
131
132	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
133	if (err < 0)
134		goto errout;
135
136	if (tb[TCA_DSMARK_MASK])
137		mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
138
139	if (tb[TCA_DSMARK_VALUE])
140		p->value[*arg - 1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
141
142	if (tb[TCA_DSMARK_MASK])
143		p->mask[*arg - 1] = mask;
144
145	err = 0;
146
147errout:
148	return err;
149}
150
151static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
152{
153	struct dsmark_qdisc_data *p = qdisc_priv(sch);
154
155	if (!dsmark_valid_index(p, arg))
156		return -EINVAL;
157
158	p->mask[arg - 1] = 0xff;
159	p->value[arg - 1] = 0;
160
161	return 0;
162}
163
164static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
165{
166	struct dsmark_qdisc_data *p = qdisc_priv(sch);
167	int i;
168
169	pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
170		 __func__, 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 __rcu **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("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
204
205	if (p->set_tc_index) {
206		switch (tc_skb_protocol(skb)) {
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		struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
233		int result = tc_classify(skb, fl, &res);
234
235		pr_debug("result %d class 0x%04x\n", result, res.classid);
236
237		switch (result) {
238#ifdef CONFIG_NET_CLS_ACT
239		case TC_ACT_QUEUED:
240		case TC_ACT_STOLEN:
241			kfree_skb(skb);
242			return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
243
244		case TC_ACT_SHOT:
245			goto drop;
246#endif
247		case TC_ACT_OK:
248			skb->tc_index = TC_H_MIN(res.classid);
249			break;
250
251		default:
252			if (p->default_index != NO_DEFAULT_INDEX)
253				skb->tc_index = p->default_index;
254			break;
255		}
256	}
257
258	err = qdisc_enqueue(skb, p->q);
259	if (err != NET_XMIT_SUCCESS) {
260		if (net_xmit_drop_count(err))
261			qdisc_qstats_drop(sch);
262		return err;
263	}
264
265	sch->q.qlen++;
266
267	return NET_XMIT_SUCCESS;
268
269drop:
270	qdisc_drop(skb, sch);
271	return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
272}
273
274static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
275{
276	struct dsmark_qdisc_data *p = qdisc_priv(sch);
277	struct sk_buff *skb;
278	u32 index;
279
280	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
281
282	skb = p->q->ops->dequeue(p->q);
283	if (skb == NULL)
284		return NULL;
285
286	qdisc_bstats_update(sch, skb);
287	sch->q.qlen--;
288
289	index = skb->tc_index & (p->indices - 1);
290	pr_debug("index %d->%d\n", skb->tc_index, index);
291
292	switch (tc_skb_protocol(skb)) {
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:
302		/*
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.
306		 */
307		if (p->mask[index] != 0xff || p->value[index])
308			pr_warn("%s: unsupported protocol %d\n",
309				__func__, ntohs(tc_skb_protocol(skb)));
310		break;
311	}
312
313	return skb;
314}
315
316static struct sk_buff *dsmark_peek(struct Qdisc *sch)
317{
318	struct dsmark_qdisc_data *p = qdisc_priv(sch);
319
320	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
321
322	return p->q->ops->peek(p->q);
323}
324
325static unsigned int dsmark_drop(struct Qdisc *sch)
326{
327	struct dsmark_qdisc_data *p = qdisc_priv(sch);
328	unsigned int len;
329
330	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
331
332	if (p->q->ops->drop == NULL)
333		return 0;
334
335	len = p->q->ops->drop(p->q);
336	if (len)
337		sch->q.qlen--;
338
339	return len;
340}
341
342static int dsmark_init(struct Qdisc *sch, struct nlattr *opt)
343{
344	struct dsmark_qdisc_data *p = qdisc_priv(sch);
345	struct nlattr *tb[TCA_DSMARK_MAX + 1];
346	int err = -EINVAL;
347	u32 default_index = NO_DEFAULT_INDEX;
348	u16 indices;
349	u8 *mask;
350
351	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
352
353	if (!opt)
354		goto errout;
355
356	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
357	if (err < 0)
358		goto errout;
359
360	err = -EINVAL;
361	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
362
363	if (hweight32(indices) != 1)
364		goto errout;
365
366	if (tb[TCA_DSMARK_DEFAULT_INDEX])
367		default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
368
369	mask = kmalloc(indices * 2, GFP_KERNEL);
370	if (mask == NULL) {
371		err = -ENOMEM;
372		goto errout;
373	}
374
375	p->mask = mask;
376	memset(p->mask, 0xff, indices);
377
378	p->value = p->mask + indices;
379	memset(p->value, 0, indices);
380
381	p->indices = indices;
382	p->default_index = default_index;
383	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
384
385	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
386	if (p->q == NULL)
387		p->q = &noop_qdisc;
388
389	pr_debug("%s: qdisc %p\n", __func__, p->q);
390
391	err = 0;
392errout:
393	return err;
394}
395
396static void dsmark_reset(struct Qdisc *sch)
397{
398	struct dsmark_qdisc_data *p = qdisc_priv(sch);
399
400	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
401	qdisc_reset(p->q);
402	sch->q.qlen = 0;
403}
404
405static void dsmark_destroy(struct Qdisc *sch)
406{
407	struct dsmark_qdisc_data *p = qdisc_priv(sch);
408
409	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
410
411	tcf_destroy_chain(&p->filter_list);
412	qdisc_destroy(p->q);
413	kfree(p->mask);
414}
415
416static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
417			     struct sk_buff *skb, struct tcmsg *tcm)
418{
419	struct dsmark_qdisc_data *p = qdisc_priv(sch);
420	struct nlattr *opts = NULL;
421
422	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
423
424	if (!dsmark_valid_index(p, cl))
425		return -EINVAL;
426
427	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
428	tcm->tcm_info = p->q->handle;
429
430	opts = nla_nest_start(skb, TCA_OPTIONS);
431	if (opts == NULL)
432		goto nla_put_failure;
433	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mask[cl - 1]) ||
434	    nla_put_u8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]))
435		goto nla_put_failure;
436
437	return nla_nest_end(skb, opts);
438
439nla_put_failure:
440	nla_nest_cancel(skb, opts);
441	return -EMSGSIZE;
442}
443
444static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
445{
446	struct dsmark_qdisc_data *p = qdisc_priv(sch);
447	struct nlattr *opts = NULL;
448
449	opts = nla_nest_start(skb, TCA_OPTIONS);
450	if (opts == NULL)
451		goto nla_put_failure;
452	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
453		goto nla_put_failure;
454
455	if (p->default_index != NO_DEFAULT_INDEX &&
456	    nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
457		goto nla_put_failure;
458
459	if (p->set_tc_index &&
460	    nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
461		goto nla_put_failure;
462
463	return nla_nest_end(skb, opts);
464
465nla_put_failure:
466	nla_nest_cancel(skb, opts);
467	return -EMSGSIZE;
468}
469
470static const struct Qdisc_class_ops dsmark_class_ops = {
471	.graft		=	dsmark_graft,
472	.leaf		=	dsmark_leaf,
473	.get		=	dsmark_get,
474	.put		=	dsmark_put,
475	.change		=	dsmark_change,
476	.delete		=	dsmark_delete,
477	.walk		=	dsmark_walk,
478	.tcf_chain	=	dsmark_find_tcf,
479	.bind_tcf	=	dsmark_bind_filter,
480	.unbind_tcf	=	dsmark_put,
481	.dump		=	dsmark_dump_class,
482};
483
484static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
485	.next		=	NULL,
486	.cl_ops		=	&dsmark_class_ops,
487	.id		=	"dsmark",
488	.priv_size	=	sizeof(struct dsmark_qdisc_data),
489	.enqueue	=	dsmark_enqueue,
490	.dequeue	=	dsmark_dequeue,
491	.peek		=	dsmark_peek,
492	.drop		=	dsmark_drop,
493	.init		=	dsmark_init,
494	.reset		=	dsmark_reset,
495	.destroy	=	dsmark_destroy,
496	.change		=	NULL,
497	.dump		=	dsmark_dump,
498	.owner		=	THIS_MODULE,
499};
500
501static int __init dsmark_module_init(void)
502{
503	return register_qdisc(&dsmark_qdisc_ops);
504}
505
506static void __exit dsmark_module_exit(void)
507{
508	unregister_qdisc(&dsmark_qdisc_ops);
509}
510
511module_init(dsmark_module_init)
512module_exit(dsmark_module_exit)
513
514MODULE_LICENSE("GPL");
515