[deliverable/linux.git] / net / sched / sch_dsmark.c

/* net/sched/sch_dsmark.c - Differentiated Services field marker */

/* Written 1998-2000 by Werner Almesberger, EPFL ICA */


#include <linux/module.h>
#include <linux/init.h>
#include <linux/types.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/skbuff.h>
#include <linux/rtnetlink.h>
#include <net/pkt_sched.h>
#include <net/dsfield.h>
#include <net/inet_ecn.h>
#include <asm/byteorder.h>


#if 0 /* control */
#define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
#else
#define DPRINTK(format,args...)
#endif

#if 0 /* data */
#define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
#else
#define D2PRINTK(format,args...)
#endif


#define PRIV(sch) ((struct dsmark_qdisc_data *) qdisc_priv(sch))


/*
 * classid	class		marking
 * -------	-----		-------
 *   n/a	  0		n/a
 *   x:0	  1		use entry [0]
 *   ...	 ...		...
 *   x:y y>0	 y+1		use entry [y]
 *   ...	 ...		...
 * x:indices-1	indices		use entry [indices-1]
 *   ...	 ...		...
 *   x:y	 y+1		use entry [y & (indices-1)]
 *   ...	 ...		...
 * 0xffff	0x10000		use entry [indices-1]
 */


#define NO_DEFAULT_INDEX	(1 << 16)

struct dsmark_qdisc_data {
	struct Qdisc		*q;
	struct tcf_proto	*filter_list;
	u8			*mask;	/* "owns" the array */
	u8			*value;
	u16			indices;
	u32			default_index;	/* index range is 0...0xffff */
	int			set_tc_index;
};

static inline int dsmark_valid_indices(u16 indices)
{
	while (indices != 1) {
		if (indices & 1)
			return 0;
		indices >>= 1;
	}

	return 1;
}

static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
{
	return (index <= p->indices && index > 0);
}

/* ------------------------- Class/flow operations ------------------------- */

static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
			struct Qdisc *new, struct Qdisc **old)
{
	struct dsmark_qdisc_data *p = PRIV(sch);

	DPRINTK("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
		sch, p, new, old);

	if (new == NULL) {
		new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
					sch->handle);
		if (new == NULL)
			new = &noop_qdisc;
	}

	sch_tree_lock(sch);
	*old = xchg(&p->q, new);
	qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
	qdisc_reset(*old);
	sch_tree_unlock(sch);

	return 0;
}

static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
{
	return PRIV(sch)->q;
}

static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
{
	DPRINTK("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
		sch, PRIV(sch), classid);

	return TC_H_MIN(classid) + 1;
}

static unsigned long dsmark_bind_filter(struct Qdisc *sch,
					unsigned long parent, u32 classid)
{
	return dsmark_get(sch, classid);
}

static void dsmark_put(struct Qdisc *sch, unsigned long cl)
{
}

static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
			 struct rtattr **tca, unsigned long *arg)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	struct rtattr *opt = tca[TCA_OPTIONS-1];
	struct rtattr *tb[TCA_DSMARK_MAX];
	int err = -EINVAL;
	u8 mask = 0;

	DPRINTK("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
		"arg 0x%lx\n", sch, p, classid, parent, *arg);

	if (!dsmark_valid_index(p, *arg)) {
		err = -ENOENT;
		goto rtattr_failure;
	}

	if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt))
		goto rtattr_failure;

	if (tb[TCA_DSMARK_MASK-1])
		mask = RTA_GET_U8(tb[TCA_DSMARK_MASK-1]);

	if (tb[TCA_DSMARK_VALUE-1])
		p->value[*arg-1] = RTA_GET_U8(tb[TCA_DSMARK_VALUE-1]);

	if (tb[TCA_DSMARK_MASK-1])
		p->mask[*arg-1] = mask;

	err = 0;

rtattr_failure:
	return err;
}

static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
{
	struct dsmark_qdisc_data *p = PRIV(sch);

	if (!dsmark_valid_index(p, arg))
		return -EINVAL;

	p->mask[arg-1] = 0xff;
	p->value[arg-1] = 0;

	return 0;
}

static void dsmark_walk(struct Qdisc *sch,struct qdisc_walker *walker)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	int i;

	DPRINTK("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);

	if (walker->stop)
		return;

	for (i = 0; i < p->indices; i++) {
		if (p->mask[i] == 0xff && !p->value[i])
			goto ignore;
		if (walker->count >= walker->skip) {
			if (walker->fn(sch, i+1, walker) < 0) {
				walker->stop = 1;
				break;
			}
		}
ignore:
		walker->count++;
	}
}

static struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,unsigned long cl)
{
	return &PRIV(sch)->filter_list;
}

/* --------------------------- Qdisc operations ---------------------------- */

static int dsmark_enqueue(struct sk_buff *skb,struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	int err;

	D2PRINTK("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);

	if (p->set_tc_index) {
		/* FIXME: Safe with non-linear skbs? --RR */
		switch (skb->protocol) {
			case __constant_htons(ETH_P_IP):
				skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
					& ~INET_ECN_MASK;
				break;
			case __constant_htons(ETH_P_IPV6):
				skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
					& ~INET_ECN_MASK;
				break;
			default:
				skb->tc_index = 0;
				break;
		}
	}

	if (TC_H_MAJ(skb->priority) == sch->handle)
		skb->tc_index = TC_H_MIN(skb->priority);
	else {
		struct tcf_result res;
		int result = tc_classify(skb, p->filter_list, &res);

		D2PRINTK("result %d class 0x%04x\n", result, res.classid);

		switch (result) {
#ifdef CONFIG_NET_CLS_POLICE
			case TC_POLICE_SHOT:
				kfree_skb(skb);
				sch->qstats.drops++;
				return NET_XMIT_POLICED;
#if 0
			case TC_POLICE_RECLASSIFY:
				/* FIXME: what to do here ??? */
#endif
#endif
			case TC_POLICE_OK:
				skb->tc_index = TC_H_MIN(res.classid);
				break;
			case TC_POLICE_UNSPEC:
				/* fall through */
			default:
				if (p->default_index != NO_DEFAULT_INDEX)
					skb->tc_index = p->default_index;
				break;
		}
	}

	err = p->q->enqueue(skb,p->q);
	if (err != NET_XMIT_SUCCESS) {
		sch->qstats.drops++;
		return err;
	}

	sch->bstats.bytes += skb->len;
	sch->bstats.packets++;
	sch->q.qlen++;

	return NET_XMIT_SUCCESS;
}

static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	struct sk_buff *skb;
	u32 index;

	D2PRINTK("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);

	skb = p->q->ops->dequeue(p->q);
	if (skb == NULL)
		return NULL;

	sch->q.qlen--;

	index = skb->tc_index & (p->indices - 1);
	D2PRINTK("index %d->%d\n", skb->tc_index, index);

	switch (skb->protocol) {
		case __constant_htons(ETH_P_IP):
			ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
					    p->value[index]);
			break;
		case __constant_htons(ETH_P_IPV6):
			ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
					    p->value[index]);
			break;
		default:
			/*
			 * Only complain if a change was actually attempted.
			 * This way, we can send non-IP traffic through dsmark
			 * and don't need yet another qdisc as a bypass.
			 */
			if (p->mask[index] != 0xff || p->value[index])
				printk(KERN_WARNING "dsmark_dequeue: "
				       "unsupported protocol %d\n",
				       ntohs(skb->protocol));
			break;
	}

	return skb;
}

static int dsmark_requeue(struct sk_buff *skb,struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	int err;

	D2PRINTK("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);

	err = p->q->ops->requeue(skb, p->q);
	if (err != NET_XMIT_SUCCESS) {
		sch->qstats.drops++;
		return err;
	}

	sch->q.qlen++;
	sch->qstats.requeues++;

	return NET_XMIT_SUCCESS;
}

static unsigned int dsmark_drop(struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	unsigned int len;

	DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);

	if (p->q->ops->drop == NULL)
		return 0;

	len = p->q->ops->drop(p->q);
	if (len)
		sch->q.qlen--;

	return len;
}

static int dsmark_init(struct Qdisc *sch, struct rtattr *opt)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	struct rtattr *tb[TCA_DSMARK_MAX];
	int err = -EINVAL;
	u32 default_index = NO_DEFAULT_INDEX;
	u16 indices;
	u8 *mask;

	DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);

	if (!opt || rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt) < 0)
		goto errout;

	indices = RTA_GET_U16(tb[TCA_DSMARK_INDICES-1]);
	if (!indices || !dsmark_valid_indices(indices))
		goto errout;

	if (tb[TCA_DSMARK_DEFAULT_INDEX-1])
		default_index = RTA_GET_U16(tb[TCA_DSMARK_DEFAULT_INDEX-1]);

	mask = kmalloc(indices * 2, GFP_KERNEL);
	if (mask == NULL) {
		err = -ENOMEM;
		goto errout;
	}

	p->mask = mask;
	memset(p->mask, 0xff, indices);

	p->value = p->mask + indices;
	memset(p->value, 0, indices);

	p->indices = indices;
	p->default_index = default_index;
	p->set_tc_index = RTA_GET_FLAG(tb[TCA_DSMARK_SET_TC_INDEX-1]);

	p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
	if (p->q == NULL)
		p->q = &noop_qdisc;

	DPRINTK("dsmark_init: qdisc %p\n", p->q);

	err = 0;
errout:
rtattr_failure:
	return err;
}

static void dsmark_reset(struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = PRIV(sch);

	DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
	qdisc_reset(p->q);
	sch->q.qlen = 0;
}

static void dsmark_destroy(struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = PRIV(sch);

	DPRINTK("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);

	tcf_destroy_chain(p->filter_list);
	qdisc_destroy(p->q);
	kfree(p->mask);
}

static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
			     struct sk_buff *skb, struct tcmsg *tcm)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	struct rtattr *opts = NULL;

	DPRINTK("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);

	if (!dsmark_valid_index(p, cl))
		return -EINVAL;

	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
	tcm->tcm_info = p->q->handle;

	opts = RTA_NEST(skb, TCA_OPTIONS);
	RTA_PUT_U8(skb,TCA_DSMARK_MASK, p->mask[cl-1]);
	RTA_PUT_U8(skb,TCA_DSMARK_VALUE, p->value[cl-1]);

	return RTA_NEST_END(skb, opts);

rtattr_failure:
	return RTA_NEST_CANCEL(skb, opts);
}

static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
{
	struct dsmark_qdisc_data *p = PRIV(sch);
	struct rtattr *opts = NULL;

	opts = RTA_NEST(skb, TCA_OPTIONS);
	RTA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);

	if (p->default_index != NO_DEFAULT_INDEX)
		RTA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);

	if (p->set_tc_index)
		RTA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);

	return RTA_NEST_END(skb, opts);

rtattr_failure:
	return RTA_NEST_CANCEL(skb, opts);
}

static struct Qdisc_class_ops dsmark_class_ops = {
	.graft		=	dsmark_graft,
	.leaf		=	dsmark_leaf,
	.get		=	dsmark_get,
	.put		=	dsmark_put,
	.change		=	dsmark_change,
	.delete		=	dsmark_delete,
	.walk		=	dsmark_walk,
	.tcf_chain	=	dsmark_find_tcf,
	.bind_tcf	=	dsmark_bind_filter,
	.unbind_tcf	=	dsmark_put,
	.dump		=	dsmark_dump_class,
};

static struct Qdisc_ops dsmark_qdisc_ops = {
	.next		=	NULL,
	.cl_ops		=	&dsmark_class_ops,
	.id		=	"dsmark",
	.priv_size	=	sizeof(struct dsmark_qdisc_data),
	.enqueue	=	dsmark_enqueue,
	.dequeue	=	dsmark_dequeue,
	.requeue	=	dsmark_requeue,
	.drop		=	dsmark_drop,
	.init		=	dsmark_init,
	.reset		=	dsmark_reset,
	.destroy	=	dsmark_destroy,
	.change		=	NULL,
	.dump		=	dsmark_dump,
	.owner		=	THIS_MODULE,
};

static int __init dsmark_module_init(void)
{
	return register_qdisc(&dsmark_qdisc_ops);
}

static void __exit dsmark_module_exit(void)
{
	unregister_qdisc(&dsmark_qdisc_ops);
}

module_init(dsmark_module_init)
module_exit(dsmark_module_exit)

MODULE_LICENSE("GPL");
Commit	Line	Data
1da177e4 LT	1	/* net/sched/sch_dsmark.c - Differentiated Services field marker */
	2
	3	/* Written 1998-2000 by Werner Almesberger, EPFL ICA */
	4
	5
1da177e4 LT	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>
1da177e4 LT	12	#include <linux/rtnetlink.h>
	13	#include <net/pkt_sched.h>
	14	#include <net/dsfield.h>
	15	#include <net/inet_ecn.h>
	16	#include <asm/byteorder.h>
	17
	18
08e9cd1f	19	#if 0 /* control */
1da177e4 LT	20	#define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
	21	#else
	22	#define DPRINTK(format,args...)
	23	#endif
	24
	25	#if 0 /* data */
	26	#define D2PRINTK(format,args...) printk(KERN_DEBUG format,##args)
	27	#else
	28	#define D2PRINTK(format,args...)
	29	#endif
	30
	31
af0d1141	32	#define PRIV(sch) ((struct dsmark_qdisc_data *) qdisc_priv(sch))
1da177e4 LT	33
	34
	35	/*
	36	* classid class marking
	37	* ------- ----- -------
	38	* n/a 0 n/a
	39	* x:0 1 use entry [0]
	40	* ... ... ...
	41	* x:y y>0 y+1 use entry [y]
	42	* ... ... ...
	43	* x:indices-1 indices use entry [indices-1]
	44	* ... ... ...
	45	* x:y y+1 use entry [y & (indices-1)]
	46	* ... ... ...
	47	* 0xffff 0x10000 use entry [indices-1]
	48	*/
	49
	50
	51	#define NO_DEFAULT_INDEX (1 << 16)
	52
	53	struct dsmark_qdisc_data {
	54	struct Qdisc *q;
	55	struct tcf_proto *filter_list;
af0d1141 TG	56	u8 mask; / "owns" the array */
	57	u8 *value;
	58	u16 indices;
	59	u32 default_index; /* index range is 0...0xffff */
1da177e4 LT	60	int set_tc_index;
	61	};
	62
758cc43c TG	63	static inline int dsmark_valid_indices(u16 indices)
	64	{
	65	while (indices != 1) {
	66	if (indices & 1)
	67	return 0;
	68	indices >>= 1;
	69	}
10297b99	70
758cc43c TG	71	return 1;
	72	}
	73
	74	static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
	75	{
	76	return (index <= p->indices && index > 0);
	77	}
1da177e4 LT	78
	79	/* ------------------------- Class/flow operations ------------------------- */
	80
af0d1141 TG	81	static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
af0d1141 TG	82	struct Qdisc new, struct Qdisc *old)
1da177e4 LT	83	{
	84	struct dsmark_qdisc_data *p = PRIV(sch);
	85
af0d1141 TG	86	DPRINTK("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
af0d1141 TG	87	sch, p, new, old);
486b53e5 TG	88
486b53e5 TG	89	if (new == NULL) {
9f9afec4 PM	90	new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
9f9afec4 PM	91	sch->handle);
486b53e5 TG	92	if (new == NULL)
	93	new = &noop_qdisc;
	94	}
	95
1da177e4	96	sch_tree_lock(sch);
af0d1141	97	*old = xchg(&p->q, new);
5e50da01	98	qdisc_tree_decrease_qlen(old, (old)->q.qlen);
af0d1141	99	qdisc_reset(*old);
af0d1141 TG	100	sch_tree_unlock(sch);
af0d1141 TG	101
10297b99	102	return 0;
1da177e4 LT	103	}
1da177e4 LT	104
1da177e4 LT	105	static struct Qdisc dsmark_leaf(struct Qdisc sch, unsigned long arg)
1da177e4 LT	106	{
af0d1141	107	return PRIV(sch)->q;
1da177e4 LT	108	}
1da177e4 LT	109
af0d1141	110	static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
1da177e4	111	{
af0d1141 TG	112	DPRINTK("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
af0d1141 TG	113	sch, PRIV(sch), classid);
1da177e4	114
af0d1141	115	return TC_H_MIN(classid) + 1;
1da177e4 LT	116	}
1da177e4 LT	117
1da177e4	118	static unsigned long dsmark_bind_filter(struct Qdisc *sch,
af0d1141	119	unsigned long parent, u32 classid)
1da177e4	120	{
af0d1141	121	return dsmark_get(sch, classid);
1da177e4 LT	122	}
1da177e4 LT	123
1da177e4 LT	124	static void dsmark_put(struct Qdisc *sch, unsigned long cl)
	125	{
	126	}
	127
1da177e4	128	static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
758cc43c	129	struct rtattr *tca, unsigned long arg)
1da177e4 LT	130	{
	131	struct dsmark_qdisc_data *p = PRIV(sch);
	132	struct rtattr *opt = tca[TCA_OPTIONS-1];
	133	struct rtattr *tb[TCA_DSMARK_MAX];
758cc43c TG	134	int err = -EINVAL;
758cc43c TG	135	u8 mask = 0;
1da177e4 LT	136
1da177e4 LT	137	DPRINTK("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
758cc43c TG	138	"arg 0x%lx\n", sch, p, classid, parent, *arg);
	139
	140	if (!dsmark_valid_index(p, *arg)) {
	141	err = -ENOENT;
	142	goto rtattr_failure;
1da177e4	143	}
1da177e4	144
758cc43c TG	145	if (!opt \|\| rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt))
	146	goto rtattr_failure;
	147
	148	if (tb[TCA_DSMARK_MASK-1])
	149	mask = RTA_GET_U8(tb[TCA_DSMARK_MASK-1]);
	150
	151	if (tb[TCA_DSMARK_VALUE-1])
	152	p->value[*arg-1] = RTA_GET_U8(tb[TCA_DSMARK_VALUE-1]);
10297b99	153
758cc43c TG	154	if (tb[TCA_DSMARK_MASK-1])
	155	p->mask[*arg-1] = mask;
	156
	157	err = 0;
	158
	159	rtattr_failure:
	160	return err;
	161	}
1da177e4	162
af0d1141	163	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4 LT	164	{
	165	struct dsmark_qdisc_data *p = PRIV(sch);
	166
af0d1141	167	if (!dsmark_valid_index(p, arg))
1da177e4	168	return -EINVAL;
10297b99	169
1da177e4 LT	170	p->mask[arg-1] = 0xff;
1da177e4 LT	171	p->value[arg-1] = 0;
af0d1141	172
1da177e4 LT	173	return 0;
	174	}
	175
1da177e4 LT	176	static void dsmark_walk(struct Qdisc sch,struct qdisc_walker walker)
	177	{
	178	struct dsmark_qdisc_data *p = PRIV(sch);
	179	int i;
	180
af0d1141 TG	181	DPRINTK("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
af0d1141 TG	182
1da177e4 LT	183	if (walker->stop)
1da177e4 LT	184	return;
af0d1141	185
1da177e4 LT	186	for (i = 0; i < p->indices; i++) {
1da177e4 LT	187	if (p->mask[i] == 0xff && !p->value[i])
0451eb07	188	goto ignore;
1da177e4 LT	189	if (walker->count >= walker->skip) {
	190	if (walker->fn(sch, i+1, walker) < 0) {
	191	walker->stop = 1;
	192	break;
	193	}
	194	}
10297b99	195	ignore:
0451eb07	196	walker->count++;
10297b99	197	}
1da177e4 LT	198	}
1da177e4 LT	199
1da177e4 LT	200	static struct tcf_proto *dsmark_find_tcf(struct Qdisc sch,unsigned long cl)
1da177e4 LT	201	{
af0d1141	202	return &PRIV(sch)->filter_list;
1da177e4 LT	203	}
1da177e4 LT	204
1da177e4 LT	205	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	206
1da177e4 LT	207	static int dsmark_enqueue(struct sk_buff skb,struct Qdisc sch)
	208	{
	209	struct dsmark_qdisc_data *p = PRIV(sch);
af0d1141 TG	210	int err;
	211
	212	D2PRINTK("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
1da177e4	213
1da177e4 LT	214	if (p->set_tc_index) {
	215	/* FIXME: Safe with non-linear skbs? --RR */
	216	switch (skb->protocol) {
	217	case __constant_htons(ETH_P_IP):
eddc9ec5	218	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
1da177e4 LT	219	& ~INET_ECN_MASK;
	220	break;
	221	case __constant_htons(ETH_P_IPV6):
0660e03f	222	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
1da177e4 LT	223	& ~INET_ECN_MASK;
	224	break;
	225	default:
	226	skb->tc_index = 0;
	227	break;
3ff50b79	228	}
1da177e4	229	}
af0d1141 TG	230
af0d1141 TG	231	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	232	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	233	else {
	234	struct tcf_result res;
	235	int result = tc_classify(skb, p->filter_list, &res);
	236
	237	D2PRINTK("result %d class 0x%04x\n", result, res.classid);
	238
1da177e4 LT	239	switch (result) {
	240	#ifdef CONFIG_NET_CLS_POLICE
	241	case TC_POLICE_SHOT:
	242	kfree_skb(skb);
af0d1141 TG	243	sch->qstats.drops++;
af0d1141 TG	244	return NET_XMIT_POLICED;
1da177e4 LT	245	#if 0
	246	case TC_POLICE_RECLASSIFY:
	247	/* FIXME: what to do here ??? */
	248	#endif
	249	#endif
	250	case TC_POLICE_OK:
	251	skb->tc_index = TC_H_MIN(res.classid);
	252	break;
	253	case TC_POLICE_UNSPEC:
	254	/* fall through */
	255	default:
	256	if (p->default_index != NO_DEFAULT_INDEX)
	257	skb->tc_index = p->default_index;
	258	break;
3ff50b79	259	}
1da177e4	260	}
1da177e4	261
af0d1141 TG	262	err = p->q->enqueue(skb,p->q);
af0d1141 TG	263	if (err != NET_XMIT_SUCCESS) {
1da177e4	264	sch->qstats.drops++;
af0d1141	265	return err;
1da177e4	266	}
af0d1141	267
1da177e4 LT	268	sch->bstats.bytes += skb->len;
	269	sch->bstats.packets++;
	270	sch->q.qlen++;
1da177e4	271
af0d1141 TG	272	return NET_XMIT_SUCCESS;
af0d1141 TG	273	}
1da177e4 LT	274
	275	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	276	{
	277	struct dsmark_qdisc_data *p = PRIV(sch);
	278	struct sk_buff *skb;
af0d1141 TG	279	u32 index;
	280
	281	D2PRINTK("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
1da177e4	282
1da177e4	283	skb = p->q->ops->dequeue(p->q);
af0d1141	284	if (skb == NULL)
1da177e4	285	return NULL;
af0d1141	286
1da177e4	287	sch->q.qlen--;
af0d1141 TG	288
	289	index = skb->tc_index & (p->indices - 1);
	290	D2PRINTK("index %d->%d\n", skb->tc_index, index);
	291
1da177e4 LT	292	switch (skb->protocol) {
1da177e4 LT	293	case __constant_htons(ETH_P_IP):
eddc9ec5	294	ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
af0d1141	295	p->value[index]);
1da177e4 LT	296	break;
1da177e4 LT	297	case __constant_htons(ETH_P_IPV6):
0660e03f	298	ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
af0d1141	299	p->value[index]);
1da177e4 LT	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	printk(KERN_WARNING "dsmark_dequeue: "
	309	"unsupported protocol %d\n",
66c6f529	310	ntohs(skb->protocol));
1da177e4	311	break;
3ff50b79	312	}
af0d1141	313
1da177e4 LT	314	return skb;
	315	}
	316
1da177e4 LT	317	static int dsmark_requeue(struct sk_buff skb,struct Qdisc sch)
1da177e4 LT	318	{
1da177e4	319	struct dsmark_qdisc_data *p = PRIV(sch);
af0d1141	320	int err;
1da177e4	321
af0d1141 TG	322	D2PRINTK("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
	323
	324	err = p->q->ops->requeue(skb, p->q);
	325	if (err != NET_XMIT_SUCCESS) {
	326	sch->qstats.drops++;
	327	return err;
1da177e4	328	}
1da177e4	329
af0d1141 TG	330	sch->q.qlen++;
	331	sch->qstats.requeues++;
	332
	333	return NET_XMIT_SUCCESS;
	334	}
1da177e4 LT	335
	336	static unsigned int dsmark_drop(struct Qdisc *sch)
	337	{
	338	struct dsmark_qdisc_data *p = PRIV(sch);
	339	unsigned int len;
10297b99	340
af0d1141 TG	341	DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
	342
	343	if (p->q->ops->drop == NULL)
1da177e4	344	return 0;
af0d1141 TG	345
	346	len = p->q->ops->drop(p->q);
	347	if (len)
	348	sch->q.qlen--;
	349
1da177e4 LT	350	return len;
	351	}
	352
758cc43c	353	static int dsmark_init(struct Qdisc sch, struct rtattr opt)
1da177e4 LT	354	{
	355	struct dsmark_qdisc_data *p = PRIV(sch);
	356	struct rtattr *tb[TCA_DSMARK_MAX];
758cc43c TG	357	int err = -EINVAL;
	358	u32 default_index = NO_DEFAULT_INDEX;
	359	u16 indices;
	360	u8 *mask;
	361
	362	DPRINTK("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
	363
	364	if (!opt \|\| rtattr_parse_nested(tb, TCA_DSMARK_MAX, opt) < 0)
	365	goto errout;
	366
	367	indices = RTA_GET_U16(tb[TCA_DSMARK_INDICES-1]);
	368	if (!indices \|\| !dsmark_valid_indices(indices))
	369	goto errout;
	370
	371	if (tb[TCA_DSMARK_DEFAULT_INDEX-1])
	372	default_index = RTA_GET_U16(tb[TCA_DSMARK_DEFAULT_INDEX-1]);
	373
	374	mask = kmalloc(indices * 2, GFP_KERNEL);
	375	if (mask == NULL) {
	376	err = -ENOMEM;
	377	goto errout;
1da177e4	378	}
758cc43c TG	379
	380	p->mask = mask;
	381	memset(p->mask, 0xff, indices);
	382
	383	p->value = p->mask + indices;
	384	memset(p->value, 0, indices);
	385
	386	p->indices = indices;
	387	p->default_index = default_index;
	388	p->set_tc_index = RTA_GET_FLAG(tb[TCA_DSMARK_SET_TC_INDEX-1]);
	389
9f9afec4	390	p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
758cc43c	391	if (p->q == NULL)
1da177e4	392	p->q = &noop_qdisc;
758cc43c TG	393
	394	DPRINTK("dsmark_init: qdisc %p\n", p->q);
	395
	396	err = 0;
	397	errout:
	398	rtattr_failure:
	399	return err;
1da177e4 LT	400	}
1da177e4 LT	401
1da177e4 LT	402	static void dsmark_reset(struct Qdisc *sch)
	403	{
	404	struct dsmark_qdisc_data *p = PRIV(sch);
	405
af0d1141	406	DPRINTK("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
1da177e4 LT	407	qdisc_reset(p->q);
	408	sch->q.qlen = 0;
	409	}
	410
1da177e4 LT	411	static void dsmark_destroy(struct Qdisc *sch)
	412	{
	413	struct dsmark_qdisc_data *p = PRIV(sch);
1da177e4	414
af0d1141 TG	415	DPRINTK("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
af0d1141 TG	416
a48b5a61	417	tcf_destroy_chain(p->filter_list);
1da177e4 LT	418	qdisc_destroy(p->q);
	419	kfree(p->mask);
	420	}
	421
1da177e4	422	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	423	struct sk_buff skb, struct tcmsg tcm)
1da177e4 LT	424	{
1da177e4 LT	425	struct dsmark_qdisc_data *p = PRIV(sch);
02f23f09	426	struct rtattr *opts = NULL;
1da177e4	427
02f23f09 TG	428	DPRINTK("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
	429
	430	if (!dsmark_valid_index(p, cl))
1da177e4	431	return -EINVAL;
02f23f09 TG	432
02f23f09 TG	433	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
cdc7f8e3	434	tcm->tcm_info = p->q->handle;
02f23f09 TG	435
	436	opts = RTA_NEST(skb, TCA_OPTIONS);
	437	RTA_PUT_U8(skb,TCA_DSMARK_MASK, p->mask[cl-1]);
	438	RTA_PUT_U8(skb,TCA_DSMARK_VALUE, p->value[cl-1]);
	439
	440	return RTA_NEST_END(skb, opts);
1da177e4 LT	441
1da177e4 LT	442	rtattr_failure:
02f23f09	443	return RTA_NEST_CANCEL(skb, opts);
1da177e4 LT	444	}
	445
	446	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	447	{
	448	struct dsmark_qdisc_data *p = PRIV(sch);
02f23f09	449	struct rtattr *opts = NULL;
1da177e4	450
02f23f09 TG	451	opts = RTA_NEST(skb, TCA_OPTIONS);
	452	RTA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
	453
	454	if (p->default_index != NO_DEFAULT_INDEX)
	455	RTA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
1da177e4	456
1da177e4	457	if (p->set_tc_index)
02f23f09 TG	458	RTA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
	459
	460	return RTA_NEST_END(skb, opts);
1da177e4 LT	461
1da177e4 LT	462	rtattr_failure:
02f23f09	463	return RTA_NEST_CANCEL(skb, opts);
1da177e4 LT	464	}
	465
	466	static struct Qdisc_class_ops dsmark_class_ops = {
	467	.graft = dsmark_graft,
	468	.leaf = dsmark_leaf,
	469	.get = dsmark_get,
	470	.put = dsmark_put,
	471	.change = dsmark_change,
	472	.delete = dsmark_delete,
	473	.walk = dsmark_walk,
	474	.tcf_chain = dsmark_find_tcf,
	475	.bind_tcf = dsmark_bind_filter,
	476	.unbind_tcf = dsmark_put,
	477	.dump = dsmark_dump_class,
	478	};
	479
	480	static struct Qdisc_ops dsmark_qdisc_ops = {
	481	.next = NULL,
	482	.cl_ops = &dsmark_class_ops,
	483	.id = "dsmark",
	484	.priv_size = sizeof(struct dsmark_qdisc_data),
	485	.enqueue = dsmark_enqueue,
	486	.dequeue = dsmark_dequeue,
	487	.requeue = dsmark_requeue,
	488	.drop = dsmark_drop,
	489	.init = dsmark_init,
	490	.reset = dsmark_reset,
	491	.destroy = dsmark_destroy,
	492	.change = NULL,
	493	.dump = dsmark_dump,
	494	.owner = THIS_MODULE,
	495	};
	496
	497	static int __init dsmark_module_init(void)
	498	{
	499	return register_qdisc(&dsmark_qdisc_ops);
	500	}
af0d1141	501
10297b99	502	static void __exit dsmark_module_exit(void)
1da177e4 LT	503	{
	504	unregister_qdisc(&dsmark_qdisc_ops);
	505	}
af0d1141	506
1da177e4 LT	507	module_init(dsmark_module_init)
1da177e4 LT	508	module_exit(dsmark_module_exit)
af0d1141	509
1da177e4	510	MODULE_LICENSE("GPL");