[net-next-2.6.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/slab.h>
#include <linux/types.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/skbuff.h>
#include <linux/rtnetlink.h>
#include <linux/bitops.h>
#include <net/pkt_sched.h>
#include <net/dsfield.h>
#include <net/inet_ecn.h>
#include <asm/byteorder.h>

/*
 * 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_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 = qdisc_priv(sch);

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

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

	sch_tree_lock(sch);
	*old = p->q;
	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)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	return p->q;
}

static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
{
	pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
		sch, qdisc_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 const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
	[TCA_DSMARK_INDICES]		= { .type = NLA_U16 },
	[TCA_DSMARK_DEFAULT_INDEX]	= { .type = NLA_U16 },
	[TCA_DSMARK_SET_TC_INDEX]	= { .type = NLA_FLAG },
	[TCA_DSMARK_MASK]		= { .type = NLA_U8 },
	[TCA_DSMARK_VALUE]		= { .type = NLA_U8 },
};

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

	pr_debug("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 errout;
	}

	if (!opt)
		goto errout;

	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
	if (err < 0)
		goto errout;

	if (tb[TCA_DSMARK_MASK])
		mask = nla_get_u8(tb[TCA_DSMARK_MASK]);

	if (tb[TCA_DSMARK_VALUE])
		p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);

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

	err = 0;

errout:
	return err;
}

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

	pr_debug("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 inline struct tcf_proto **dsmark_find_tcf(struct Qdisc *sch,
						 unsigned long cl)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	return &p->filter_list;
}

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

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

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

	if (p->set_tc_index) {
		switch (skb->protocol) {
		case htons(ETH_P_IP):
			if (skb_cow_head(skb, sizeof(struct iphdr)))
				goto drop;

			skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
				& ~INET_ECN_MASK;
			break;

		case htons(ETH_P_IPV6):
			if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
				goto drop;

			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);

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

		switch (result) {
#ifdef CONFIG_NET_CLS_ACT
		case TC_ACT_QUEUED:
		case TC_ACT_STOLEN:
			kfree_skb(skb);
			return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;

		case TC_ACT_SHOT:
			goto drop;
#endif
		case TC_ACT_OK:
			skb->tc_index = TC_H_MIN(res.classid);
			break;

		default:
			if (p->default_index != NO_DEFAULT_INDEX)
				skb->tc_index = p->default_index;
			break;
		}
	}

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

	sch->bstats.bytes += qdisc_pkt_len(skb);
	sch->bstats.packets++;
	sch->q.qlen++;

	return NET_XMIT_SUCCESS;

drop:
	kfree_skb(skb);
	sch->qstats.drops++;
	return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
}

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

	pr_debug("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);
	pr_debug("index %d->%d\n", skb->tc_index, index);

	switch (skb->protocol) {
	case htons(ETH_P_IP):
		ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
				    p->value[index]);
			break;
	case 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 struct sk_buff *dsmark_peek(struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);

	pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", sch, p);

	return p->q->ops->peek(p->q);
}

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

	pr_debug("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 nlattr *opt)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	struct nlattr *tb[TCA_DSMARK_MAX + 1];
	int err = -EINVAL;
	u32 default_index = NO_DEFAULT_INDEX;
	u16 indices;
	u8 *mask;

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

	if (!opt)
		goto errout;

	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
	if (err < 0)
		goto errout;

	err = -EINVAL;
	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);

	if (hweight32(indices) != 1)
		goto errout;

	if (tb[TCA_DSMARK_DEFAULT_INDEX])
		default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);

	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 = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);

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

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

	err = 0;
errout:
	return err;
}

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

	pr_debug("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 = qdisc_priv(sch);

	pr_debug("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 = qdisc_priv(sch);
	struct nlattr *opts = NULL;

	pr_debug("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 = nla_nest_start(skb, TCA_OPTIONS);
	if (opts == NULL)
		goto nla_put_failure;
	NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
	NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);

	return nla_nest_end(skb, opts);

nla_put_failure:
	nla_nest_cancel(skb, opts);
	return -EMSGSIZE;
}

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

	opts = nla_nest_start(skb, TCA_OPTIONS);
	if (opts == NULL)
		goto nla_put_failure;
	NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);

	if (p->default_index != NO_DEFAULT_INDEX)
		NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);

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

	return nla_nest_end(skb, opts);

nla_put_failure:
	nla_nest_cancel(skb, opts);
	return -EMSGSIZE;
}

static const 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 __read_mostly = {
	.next		=	NULL,
	.cl_ops		=	&dsmark_class_ops,
	.id		=	"dsmark",
	.priv_size	=	sizeof(struct dsmark_qdisc_data),
	.enqueue	=	dsmark_enqueue,
	.dequeue	=	dsmark_dequeue,
	.peek		=	dsmark_peek,
	.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>
1da177e4 LT	7	#include <linux/init.h>
5a0e3ad6	8	#include <linux/slab.h>
1da177e4 LT	9	#include <linux/types.h>
	10	#include <linux/string.h>
	11	#include <linux/errno.h>
	12	#include <linux/skbuff.h>
1da177e4	13	#include <linux/rtnetlink.h>
5b0ac72b	14	#include <linux/bitops.h>
1da177e4 LT	15	#include <net/pkt_sched.h>
	16	#include <net/dsfield.h>
	17	#include <net/inet_ecn.h>
	18	#include <asm/byteorder.h>
	19
1da177e4 LT	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
	38	struct dsmark_qdisc_data {
	39	struct Qdisc *q;
	40	struct tcf_proto *filter_list;
af0d1141 TG	41	u8 mask; / "owns" the array */
	42	u8 *value;
	43	u16 indices;
	44	u32 default_index; /* index range is 0...0xffff */
1da177e4 LT	45	int set_tc_index;
	46	};
	47
758cc43c TG	48	static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
	49	{
	50	return (index <= p->indices && index > 0);
	51	}
1da177e4 LT	52
	53	/* ------------------------- Class/flow operations ------------------------- */
	54
af0d1141 TG	55	static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
af0d1141 TG	56	struct Qdisc new, struct Qdisc *old)
1da177e4	57	{
81da99ed	58	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	59
81da99ed	60	pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
af0d1141	61	sch, p, new, old);
486b53e5 TG	62
486b53e5 TG	63	if (new == NULL) {
5ce2d488	64	new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd	65	&pfifo_qdisc_ops,
9f9afec4	66	sch->handle);
486b53e5 TG	67	if (new == NULL)
	68	new = &noop_qdisc;
	69	}
	70
1da177e4	71	sch_tree_lock(sch);
b94c8afc PM	72	*old = p->q;
b94c8afc PM	73	p->q = new;
5e50da01	74	qdisc_tree_decrease_qlen(old, (old)->q.qlen);
af0d1141	75	qdisc_reset(*old);
af0d1141 TG	76	sch_tree_unlock(sch);
af0d1141 TG	77
10297b99	78	return 0;
1da177e4 LT	79	}
1da177e4 LT	80
1da177e4 LT	81	static struct Qdisc dsmark_leaf(struct Qdisc sch, unsigned long arg)
1da177e4 LT	82	{
81da99ed SH	83	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	84	return p->q;
1da177e4 LT	85	}
1da177e4 LT	86
af0d1141	87	static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
1da177e4	88	{
81da99ed SH	89	pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
81da99ed SH	90	sch, qdisc_priv(sch), classid);
1da177e4	91
af0d1141	92	return TC_H_MIN(classid) + 1;
1da177e4 LT	93	}
1da177e4 LT	94
1da177e4	95	static unsigned long dsmark_bind_filter(struct Qdisc *sch,
af0d1141	96	unsigned long parent, u32 classid)
1da177e4	97	{
af0d1141	98	return dsmark_get(sch, classid);
1da177e4 LT	99	}
1da177e4 LT	100
1da177e4 LT	101	static void dsmark_put(struct Qdisc *sch, unsigned long cl)
	102	{
	103	}
	104
27a3421e PM	105	static 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
1da177e4	113	static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
1e90474c	114	struct nlattr *tca, unsigned long arg)
1da177e4	115	{
81da99ed	116	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c PM	117	struct nlattr *opt = tca[TCA_OPTIONS];
1e90474c PM	118	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c TG	119	int err = -EINVAL;
758cc43c TG	120	u8 mask = 0;
1da177e4	121
81da99ed	122	pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
758cc43c TG	123	"arg 0x%lx\n", sch, p, classid, parent, *arg);
	124
	125	if (!dsmark_valid_index(p, *arg)) {
	126	err = -ENOENT;
1e90474c	127	goto errout;
1da177e4	128	}
1da177e4	129
cee63723	130	if (!opt)
1e90474c	131	goto errout;
758cc43c	132
27a3421e	133	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	134	if (err < 0)
27a3421e	135	goto errout;
cee63723	136
27a3421e	137	if (tb[TCA_DSMARK_MASK])
1e90474c	138	mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
27a3421e PM	139
27a3421e PM	140	if (tb[TCA_DSMARK_VALUE])
1e90474c	141	p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99	142
1e90474c	143	if (tb[TCA_DSMARK_MASK])
758cc43c TG	144	p->mask[*arg-1] = mask;
	145
	146	err = 0;
	147
1e90474c	148	errout:
758cc43c TG	149	return err;
758cc43c TG	150	}
1da177e4	151
af0d1141	152	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4	153	{
81da99ed	154	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	155
af0d1141	156	if (!dsmark_valid_index(p, arg))
1da177e4	157	return -EINVAL;
10297b99	158
1da177e4 LT	159	p->mask[arg-1] = 0xff;
1da177e4 LT	160	p->value[arg-1] = 0;
af0d1141	161
1da177e4 LT	162	return 0;
	163	}
	164
9d127fbd	165	static void dsmark_walk(struct Qdisc sch, struct qdisc_walker walker)
1da177e4	166	{
81da99ed	167	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 LT	168	int i;
1da177e4 LT	169
81da99ed	170	pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
af0d1141	171
1da177e4 LT	172	if (walker->stop)
1da177e4 LT	173	return;
af0d1141	174
1da177e4 LT	175	for (i = 0; i < p->indices; i++) {
1da177e4 LT	176	if (p->mask[i] == 0xff && !p->value[i])
0451eb07	177	goto ignore;
1da177e4 LT	178	if (walker->count >= walker->skip) {
	179	if (walker->fn(sch, i+1, walker) < 0) {
	180	walker->stop = 1;
	181	break;
	182	}
	183	}
10297b99	184	ignore:
0451eb07	185	walker->count++;
10297b99	186	}
1da177e4 LT	187	}
1da177e4 LT	188
9d127fbd SH	189	static inline struct tcf_proto *dsmark_find_tcf(struct Qdisc sch,
9d127fbd SH	190	unsigned long cl)
1da177e4	191	{
81da99ed SH	192	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	193	return &p->filter_list;
1da177e4 LT	194	}
1da177e4 LT	195
1da177e4 LT	196	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	197
9d127fbd	198	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	199	{
81da99ed	200	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	201	int err;
af0d1141 TG	202
81da99ed	203	pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
1da177e4	204
1da177e4	205	if (p->set_tc_index) {
1da177e4	206	switch (skb->protocol) {
60678040	207	case htons(ETH_P_IP):
9d127fbd SH	208	if (skb_cow_head(skb, sizeof(struct iphdr)))
9d127fbd SH	209	goto drop;
4c30719f	210
9d127fbd SH	211	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	212	& ~INET_ECN_MASK;
	213	break;
4c30719f	214
60678040	215	case htons(ETH_P_IPV6):
9d127fbd SH	216	if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
9d127fbd SH	217	goto drop;
4c30719f	218
9d127fbd SH	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;
3ff50b79	225	}
1da177e4	226	}
af0d1141 TG	227
af0d1141 TG	228	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	229	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	230	else {
	231	struct tcf_result res;
	232	int result = tc_classify(skb, p->filter_list, &res);
	233
81da99ed	234	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	235
1da177e4	236	switch (result) {
f6853e2d PM	237	#ifdef CONFIG_NET_CLS_ACT
	238	case TC_ACT_QUEUED:
	239	case TC_ACT_STOLEN:
	240	kfree_skb(skb);
378a2f09	241	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	242
f6853e2d	243	case TC_ACT_SHOT:
4c30719f	244	goto drop;
1da177e4	245	#endif
c3bc7cff	246	case TC_ACT_OK:
f6853e2d PM	247	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	248	break;
4c30719f	249
f6853e2d PM	250	default:
	251	if (p->default_index != NO_DEFAULT_INDEX)
	252	skb->tc_index = p->default_index;
	253	break;
3ff50b79	254	}
1da177e4	255	}
1da177e4	256
5f86173b	257	err = qdisc_enqueue(skb, p->q);
af0d1141	258	if (err != NET_XMIT_SUCCESS) {
378a2f09 JP	259	if (net_xmit_drop_count(err))
378a2f09 JP	260	sch->qstats.drops++;
af0d1141	261	return err;
1da177e4	262	}
af0d1141	263
0abf77e5	264	sch->bstats.bytes += qdisc_pkt_len(skb);
1da177e4 LT	265	sch->bstats.packets++;
1da177e4 LT	266	sch->q.qlen++;
1da177e4	267
af0d1141	268	return NET_XMIT_SUCCESS;
4c30719f SH	269
	270	drop:
	271	kfree_skb(skb);
	272	sch->qstats.drops++;
c27f339a	273	return NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
af0d1141	274	}
1da177e4 LT	275
	276	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	277	{
81da99ed	278	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	279	struct sk_buff *skb;
af0d1141 TG	280	u32 index;
af0d1141 TG	281
81da99ed	282	pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
1da177e4	283
1da177e4	284	skb = p->q->ops->dequeue(p->q);
af0d1141	285	if (skb == NULL)
1da177e4	286	return NULL;
af0d1141	287
1da177e4	288	sch->q.qlen--;
af0d1141 TG	289
af0d1141 TG	290	index = skb->tc_index & (p->indices - 1);
81da99ed	291	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	292
1da177e4	293	switch (skb->protocol) {
60678040	294	case htons(ETH_P_IP):
9d127fbd SH	295	ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
9d127fbd SH	296	p->value[index]);
1da177e4	297	break;
60678040	298	case htons(ETH_P_IPV6):
9d127fbd SH	299	ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
9d127fbd SH	300	p->value[index]);
1da177e4	301	break;
9d127fbd SH	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;
3ff50b79	313	}
af0d1141	314
1da177e4 LT	315	return skb;
	316	}
	317
8e3af978 JP	318	static 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
1da177e4 LT	327	static unsigned int dsmark_drop(struct Qdisc *sch)
1da177e4 LT	328	{
81da99ed	329	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	330	unsigned int len;
10297b99	331
81da99ed	332	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
af0d1141 TG	333
af0d1141 TG	334	if (p->q->ops->drop == NULL)
1da177e4	335	return 0;
af0d1141 TG	336
	337	len = p->q->ops->drop(p->q);
	338	if (len)
	339	sch->q.qlen--;
	340
1da177e4 LT	341	return len;
	342	}
	343
1e90474c	344	static int dsmark_init(struct Qdisc sch, struct nlattr opt)
1da177e4	345	{
81da99ed	346	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	347	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c TG	348	int err = -EINVAL;
	349	u32 default_index = NO_DEFAULT_INDEX;
	350	u16 indices;
	351	u8 *mask;
	352
81da99ed	353	pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
758cc43c	354
cee63723 PM	355	if (!opt)
	356	goto errout;
	357
27a3421e	358	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	359	if (err < 0)
758cc43c TG	360	goto errout;
758cc43c TG	361
cee63723	362	err = -EINVAL;
1e90474c	363	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	364
5b0ac72b DM	365	if (hweight32(indices) != 1)
758cc43c TG	366	goto errout;
758cc43c TG	367
27a3421e	368	if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c	369	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c TG	370
	371	mask = kmalloc(indices * 2, GFP_KERNEL);
	372	if (mask == NULL) {
	373	err = -ENOMEM;
	374	goto errout;
1da177e4	375	}
758cc43c TG	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;
1e90474c	385	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	386
5ce2d488	387	p->q = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd	388	&pfifo_qdisc_ops, sch->handle);
758cc43c	389	if (p->q == NULL)
1da177e4	390	p->q = &noop_qdisc;
758cc43c	391
81da99ed	392	pr_debug("dsmark_init: qdisc %p\n", p->q);
758cc43c TG	393
	394	err = 0;
	395	errout:
758cc43c	396	return err;
1da177e4 LT	397	}
1da177e4 LT	398
1da177e4 LT	399	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	400	{
81da99ed	401	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	402
81da99ed	403	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
1da177e4 LT	404	qdisc_reset(p->q);
	405	sch->q.qlen = 0;
	406	}
	407
1da177e4 LT	408	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	409	{
81da99ed	410	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	411
81da99ed	412	pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
af0d1141	413
ff31ab56	414	tcf_destroy_chain(&p->filter_list);
1da177e4 LT	415	qdisc_destroy(p->q);
	416	kfree(p->mask);
	417	}
	418
1da177e4	419	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	420	struct sk_buff skb, struct tcmsg tcm)
1da177e4	421	{
81da99ed	422	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	423	struct nlattr *opts = NULL;
1da177e4	424
81da99ed	425	pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
02f23f09 TG	426
02f23f09 TG	427	if (!dsmark_valid_index(p, cl))
1da177e4	428	return -EINVAL;
02f23f09 TG	429
02f23f09 TG	430	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
cdc7f8e3	431	tcm->tcm_info = p->q->handle;
02f23f09	432
1e90474c PM	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]);
02f23f09	438
1e90474c	439	return nla_nest_end(skb, opts);
1da177e4	440
1e90474c	441	nla_put_failure:
bc3ed28c TG	442	nla_nest_cancel(skb, opts);
bc3ed28c TG	443	return -EMSGSIZE;
1da177e4 LT	444	}
	445
	446	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	447	{
81da99ed	448	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	449	struct nlattr *opts = NULL;
1da177e4	450
1e90474c PM	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);
02f23f09 TG	455
02f23f09 TG	456	if (p->default_index != NO_DEFAULT_INDEX)
1e90474c	457	NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
1da177e4	458
1da177e4	459	if (p->set_tc_index)
1e90474c	460	NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
02f23f09	461
1e90474c	462	return nla_nest_end(skb, opts);
1da177e4	463
1e90474c	464	nla_put_failure:
bc3ed28c TG	465	nla_nest_cancel(skb, opts);
bc3ed28c TG	466	return -EMSGSIZE;
1da177e4 LT	467	}
1da177e4 LT	468
20fea08b	469	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	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
20fea08b	483	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	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,
8e3af978	490	.peek = dsmark_peek,
1da177e4 LT	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
	500	static int __init dsmark_module_init(void)
	501	{
	502	return register_qdisc(&dsmark_qdisc_ops);
	503	}
af0d1141	504
10297b99	505	static void __exit dsmark_module_exit(void)
1da177e4 LT	506	{
	507	unregister_qdisc(&dsmark_qdisc_ops);
	508	}
af0d1141	509
1da177e4 LT	510	module_init(dsmark_module_init)
1da177e4 LT	511	module_exit(dsmark_module_exit)
af0d1141	512
1da177e4	513	MODULE_LICENSE("GPL");