[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/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 mask_value {
	u8			mask;
	u8			value;
};

struct dsmark_qdisc_data {
	struct Qdisc		*q;
	struct tcf_proto __rcu	*filter_list;
	struct mask_value	*mv;
	u16			indices;
	u8			set_tc_index;
	u32			default_index;	/* index range is 0...0xffff */
#define DSMARK_EMBEDDED_SZ	16
	struct mask_value	embedded[DSMARK_EMBEDDED_SZ];
};

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("%s(sch %p,[qdisc %p],new %p,old %p)\n",
		 __func__, sch, p, new, old);

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

	*old = qdisc_replace(sch, new, &p->q);
	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("%s(sch %p,[qdisc %p],classid %x)\n",
		 __func__, 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;

	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
		 __func__, 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_VALUE])
		p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);

	if (tb[TCA_DSMARK_MASK])
		p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_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->mv[arg - 1].mask = 0xff;
	p->mv[arg - 1].value = 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("%s(sch %p,[qdisc %p],walker %p)\n",
		 __func__, sch, p, walker);

	if (walker->stop)
		return;

	for (i = 0; i < p->indices; i++) {
		if (p->mv[i].mask == 0xff && !p->mv[i].value)
			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 __rcu **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("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);

	if (p->set_tc_index) {
		switch (tc_skb_protocol(skb)) {
		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;
		struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
		int result = tc_classify(skb, fl, &res, false);

		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))
			qdisc_qstats_drop(sch);
		return err;
	}

	sch->q.qlen++;

	return NET_XMIT_SUCCESS;

drop:
	qdisc_drop(skb, sch);
	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("%s(sch %p,[qdisc %p])\n", __func__, sch, p);

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

	qdisc_bstats_update(sch, skb);
	sch->q.qlen--;

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

	switch (tc_skb_protocol(skb)) {
	case htons(ETH_P_IP):
		ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
				    p->mv[index].value);
			break;
	case htons(ETH_P_IPV6):
		ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
				    p->mv[index].value);
			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->mv[index].mask != 0xff || p->mv[index].value)
			pr_warn("%s: unsupported protocol %d\n",
				__func__, ntohs(tc_skb_protocol(skb)));
		break;
	}

	return skb;
}

static struct sk_buff *dsmark_peek(struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);

	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, 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("%s(sch %p,[qdisc %p])\n", __func__, 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;
	int i;

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

	if (indices <= DSMARK_EMBEDDED_SZ)
		p->mv = p->embedded;
	else
		p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
	if (!p->mv) {
		err = -ENOMEM;
		goto errout;
	}
	for (i = 0; i < indices; i++) {
		p->mv[i].mask = 0xff;
		p->mv[i].value = 0;
	}
	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(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
	if (p->q == NULL)
		p->q = &noop_qdisc;

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

	err = 0;
errout:
	return err;
}

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

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

	tcf_destroy_chain(&p->filter_list);
	qdisc_destroy(p->q);
	if (p->mv != p->embedded)
		kfree(p->mv);
}

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("%s(sch %p,[qdisc %p],class %ld\n", __func__, 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;
	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
	    nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
		goto nla_put_failure;

	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;
	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
		goto nla_put_failure;

	if (p->default_index != NO_DEFAULT_INDEX &&
	    nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
		goto nla_put_failure;

	if (p->set_tc_index &&
	    nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
		goto nla_put_failure;

	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
47bbbb30 ED	38	struct mask_value {
	39	u8 mask;
	40	u8 value;
	41	};
	42
1da177e4 LT	43	struct dsmark_qdisc_data {
1da177e4 LT	44	struct Qdisc *q;
25d8c0d5	45	struct tcf_proto __rcu *filter_list;
47bbbb30	46	struct mask_value *mv;
af0d1141	47	u16 indices;
47bbbb30	48	u8 set_tc_index;
af0d1141	49	u32 default_index; /* index range is 0...0xffff */
47bbbb30 ED	50	#define DSMARK_EMBEDDED_SZ 16
47bbbb30 ED	51	struct mask_value embedded[DSMARK_EMBEDDED_SZ];
1da177e4 LT	52	};
1da177e4 LT	53
758cc43c TG	54	static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
758cc43c TG	55	{
17569fae	56	return index <= p->indices && index > 0;
758cc43c	57	}
1da177e4 LT	58
	59	/* ------------------------- Class/flow operations ------------------------- */
	60
af0d1141 TG	61	static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
af0d1141 TG	62	struct Qdisc new, struct Qdisc *old)
1da177e4	63	{
81da99ed	64	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	65
c76f2a2c YY	66	pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
c76f2a2c YY	67	__func__, sch, p, new, old);
486b53e5 TG	68
486b53e5 TG	69	if (new == NULL) {
3511c913	70	new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
9f9afec4	71	sch->handle);
486b53e5 TG	72	if (new == NULL)
	73	new = &noop_qdisc;
	74	}
	75
86a7996c	76	*old = qdisc_replace(sch, new, &p->q);
10297b99	77	return 0;
1da177e4 LT	78	}
1da177e4 LT	79
1da177e4 LT	80	static struct Qdisc dsmark_leaf(struct Qdisc sch, unsigned long arg)
1da177e4 LT	81	{
81da99ed SH	82	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	83	return p->q;
1da177e4 LT	84	}
1da177e4 LT	85
af0d1141	86	static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
1da177e4	87	{
c76f2a2c YY	88	pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
c76f2a2c YY	89	__func__, sch, qdisc_priv(sch), classid);
1da177e4	90
af0d1141	91	return TC_H_MIN(classid) + 1;
1da177e4 LT	92	}
1da177e4 LT	93
1da177e4	94	static unsigned long dsmark_bind_filter(struct Qdisc *sch,
af0d1141	95	unsigned long parent, u32 classid)
1da177e4	96	{
af0d1141	97	return dsmark_get(sch, classid);
1da177e4 LT	98	}
1da177e4 LT	99
1da177e4 LT	100	static void dsmark_put(struct Qdisc *sch, unsigned long cl)
	101	{
	102	}
	103
27a3421e PM	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 },
	110	};
	111
1da177e4	112	static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
1e90474c	113	struct nlattr *tca, unsigned long arg)
1da177e4	114	{
81da99ed	115	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c PM	116	struct nlattr *opt = tca[TCA_OPTIONS];
1e90474c PM	117	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c	118	int err = -EINVAL;
1da177e4	119
c76f2a2c YY	120	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
c76f2a2c YY	121	__func__, sch, p, classid, parent, *arg);
758cc43c TG	122
	123	if (!dsmark_valid_index(p, *arg)) {
	124	err = -ENOENT;
1e90474c	125	goto errout;
1da177e4	126	}
1da177e4	127
cee63723	128	if (!opt)
1e90474c	129	goto errout;
758cc43c	130
27a3421e	131	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	132	if (err < 0)
27a3421e	133	goto errout;
cee63723	134
27a3421e	135	if (tb[TCA_DSMARK_VALUE])
47bbbb30	136	p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99	137
1e90474c	138	if (tb[TCA_DSMARK_MASK])
47bbbb30	139	p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
758cc43c TG	140
	141	err = 0;
	142
1e90474c	143	errout:
758cc43c TG	144	return err;
758cc43c TG	145	}
1da177e4	146
af0d1141	147	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4	148	{
81da99ed	149	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	150
af0d1141	151	if (!dsmark_valid_index(p, arg))
1da177e4	152	return -EINVAL;
10297b99	153
47bbbb30 ED	154	p->mv[arg - 1].mask = 0xff;
47bbbb30 ED	155	p->mv[arg - 1].value = 0;
af0d1141	156
1da177e4 LT	157	return 0;
	158	}
	159
9d127fbd	160	static void dsmark_walk(struct Qdisc sch, struct qdisc_walker walker)
1da177e4	161	{
81da99ed	162	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 LT	163	int i;
1da177e4 LT	164
c76f2a2c YY	165	pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
c76f2a2c YY	166	__func__, sch, p, walker);
af0d1141	167
1da177e4 LT	168	if (walker->stop)
1da177e4 LT	169	return;
af0d1141	170
1da177e4	171	for (i = 0; i < p->indices; i++) {
47bbbb30	172	if (p->mv[i].mask == 0xff && !p->mv[i].value)
0451eb07	173	goto ignore;
1da177e4	174	if (walker->count >= walker->skip) {
cc7ec456	175	if (walker->fn(sch, i + 1, walker) < 0) {
1da177e4 LT	176	walker->stop = 1;
	177	break;
	178	}
	179	}
10297b99	180	ignore:
0451eb07	181	walker->count++;
10297b99	182	}
1da177e4 LT	183	}
1da177e4 LT	184
25d8c0d5 JF	185	static inline struct tcf_proto __rcu *dsmark_find_tcf(struct Qdisc sch,
25d8c0d5 JF	186	unsigned long cl)
1da177e4	187	{
81da99ed SH	188	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	189	return &p->filter_list;
1da177e4 LT	190	}
1da177e4 LT	191
1da177e4 LT	192	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	193
9d127fbd	194	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	195	{
81da99ed	196	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	197	int err;
af0d1141 TG	198
c76f2a2c	199	pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
1da177e4	200
1da177e4	201	if (p->set_tc_index) {
d8b9605d	202	switch (tc_skb_protocol(skb)) {
60678040	203	case htons(ETH_P_IP):
9d127fbd SH	204	if (skb_cow_head(skb, sizeof(struct iphdr)))
9d127fbd SH	205	goto drop;
4c30719f	206
9d127fbd SH	207	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	208	& ~INET_ECN_MASK;
	209	break;
4c30719f	210
60678040	211	case htons(ETH_P_IPV6):
9d127fbd SH	212	if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
9d127fbd SH	213	goto drop;
4c30719f	214
9d127fbd SH	215	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	216	& ~INET_ECN_MASK;
	217	break;
	218	default:
	219	skb->tc_index = 0;
	220	break;
3ff50b79	221	}
1da177e4	222	}
af0d1141 TG	223
af0d1141 TG	224	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	225	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	226	else {
af0d1141 TG	227	struct tcf_result res;
25d8c0d5	228	struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
3b3ae880	229	int result = tc_classify(skb, fl, &res, false);
af0d1141	230
81da99ed	231	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	232
1da177e4	233	switch (result) {
f6853e2d PM	234	#ifdef CONFIG_NET_CLS_ACT
	235	case TC_ACT_QUEUED:
	236	case TC_ACT_STOLEN:
	237	kfree_skb(skb);
378a2f09	238	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	239
f6853e2d	240	case TC_ACT_SHOT:
4c30719f	241	goto drop;
1da177e4	242	#endif
c3bc7cff	243	case TC_ACT_OK:
f6853e2d PM	244	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	245	break;
4c30719f	246
f6853e2d PM	247	default:
	248	if (p->default_index != NO_DEFAULT_INDEX)
	249	skb->tc_index = p->default_index;
	250	break;
3ff50b79	251	}
1da177e4	252	}
1da177e4	253
5f86173b	254	err = qdisc_enqueue(skb, p->q);
af0d1141	255	if (err != NET_XMIT_SUCCESS) {
378a2f09	256	if (net_xmit_drop_count(err))
25331d6c	257	qdisc_qstats_drop(sch);
af0d1141	258	return err;
1da177e4	259	}
af0d1141	260
1da177e4	261	sch->q.qlen++;
1da177e4	262
af0d1141	263	return NET_XMIT_SUCCESS;
4c30719f SH	264
4c30719f SH	265	drop:
17045755	266	qdisc_drop(skb, sch);
c27f339a	267	return NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
af0d1141	268	}
1da177e4 LT	269
	270	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	271	{
81da99ed	272	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	273	struct sk_buff *skb;
af0d1141 TG	274	u32 index;
af0d1141 TG	275
c76f2a2c	276	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	277
1da177e4	278	skb = p->q->ops->dequeue(p->q);
af0d1141	279	if (skb == NULL)
1da177e4	280	return NULL;
af0d1141	281
9190b3b3	282	qdisc_bstats_update(sch, skb);
1da177e4	283	sch->q.qlen--;
af0d1141 TG	284
af0d1141 TG	285	index = skb->tc_index & (p->indices - 1);
81da99ed	286	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	287
d8b9605d	288	switch (tc_skb_protocol(skb)) {
60678040	289	case htons(ETH_P_IP):
47bbbb30 ED	290	ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
47bbbb30 ED	291	p->mv[index].value);
1da177e4	292	break;
60678040	293	case htons(ETH_P_IPV6):
47bbbb30 ED	294	ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
47bbbb30 ED	295	p->mv[index].value);
1da177e4	296	break;
9d127fbd SH	297	default:
	298	/*
	299	* Only complain if a change was actually attempted.
	300	* This way, we can send non-IP traffic through dsmark
	301	* and don't need yet another qdisc as a bypass.
	302	*/
47bbbb30	303	if (p->mv[index].mask != 0xff \|\| p->mv[index].value)
c76f2a2c	304	pr_warn("%s: unsupported protocol %d\n",
d8b9605d	305	__func__, ntohs(tc_skb_protocol(skb)));
9d127fbd	306	break;
3ff50b79	307	}
af0d1141	308
1da177e4 LT	309	return skb;
	310	}
	311
8e3af978 JP	312	static struct sk_buff dsmark_peek(struct Qdisc sch)
	313	{
	314	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	315
c76f2a2c	316	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
8e3af978 JP	317
	318	return p->q->ops->peek(p->q);
	319	}
	320
1da177e4 LT	321	static unsigned int dsmark_drop(struct Qdisc *sch)
1da177e4 LT	322	{
81da99ed	323	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	324	unsigned int len;
10297b99	325
c76f2a2c	326	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141 TG	327
af0d1141 TG	328	if (p->q->ops->drop == NULL)
1da177e4	329	return 0;
af0d1141 TG	330
	331	len = p->q->ops->drop(p->q);
	332	if (len)
	333	sch->q.qlen--;
	334
1da177e4 LT	335	return len;
	336	}
	337
1e90474c	338	static int dsmark_init(struct Qdisc sch, struct nlattr opt)
1da177e4	339	{
81da99ed	340	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	341	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c TG	342	int err = -EINVAL;
	343	u32 default_index = NO_DEFAULT_INDEX;
	344	u16 indices;
47bbbb30	345	int i;
758cc43c	346
c76f2a2c	347	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
758cc43c	348
cee63723 PM	349	if (!opt)
	350	goto errout;
	351
27a3421e	352	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	353	if (err < 0)
758cc43c TG	354	goto errout;
758cc43c TG	355
cee63723	356	err = -EINVAL;
1e90474c	357	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	358
5b0ac72b DM	359	if (hweight32(indices) != 1)
758cc43c TG	360	goto errout;
758cc43c TG	361
27a3421e	362	if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c	363	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c	364
47bbbb30 ED	365	if (indices <= DSMARK_EMBEDDED_SZ)
	366	p->mv = p->embedded;
	367	else
	368	p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
	369	if (!p->mv) {
758cc43c TG	370	err = -ENOMEM;
758cc43c TG	371	goto errout;
1da177e4	372	}
47bbbb30 ED	373	for (i = 0; i < indices; i++) {
	374	p->mv[i].mask = 0xff;
	375	p->mv[i].value = 0;
	376	}
758cc43c TG	377	p->indices = indices;
758cc43c TG	378	p->default_index = default_index;
1e90474c	379	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	380
3511c913	381	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
758cc43c	382	if (p->q == NULL)
1da177e4	383	p->q = &noop_qdisc;
758cc43c	384
c76f2a2c	385	pr_debug("%s: qdisc %p\n", __func__, p->q);
758cc43c TG	386
	387	err = 0;
	388	errout:
758cc43c	389	return err;
1da177e4 LT	390	}
1da177e4 LT	391
1da177e4 LT	392	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	393	{
81da99ed	394	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	395
c76f2a2c	396	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4 LT	397	qdisc_reset(p->q);
	398	sch->q.qlen = 0;
	399	}
	400
1da177e4 LT	401	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	402	{
81da99ed	403	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	404
c76f2a2c	405	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141	406
ff31ab56	407	tcf_destroy_chain(&p->filter_list);
1da177e4	408	qdisc_destroy(p->q);
47bbbb30 ED	409	if (p->mv != p->embedded)
47bbbb30 ED	410	kfree(p->mv);
1da177e4 LT	411	}
1da177e4 LT	412
1da177e4	413	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	414	struct sk_buff skb, struct tcmsg tcm)
1da177e4	415	{
81da99ed	416	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	417	struct nlattr *opts = NULL;
1da177e4	418
c76f2a2c	419	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
02f23f09 TG	420
02f23f09 TG	421	if (!dsmark_valid_index(p, cl))
1da177e4	422	return -EINVAL;
02f23f09	423
cc7ec456	424	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
cdc7f8e3	425	tcm->tcm_info = p->q->handle;
02f23f09	426
1e90474c PM	427	opts = nla_nest_start(skb, TCA_OPTIONS);
	428	if (opts == NULL)
	429	goto nla_put_failure;
47bbbb30 ED	430	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) \|\|
47bbbb30 ED	431	nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
1b34ec43	432	goto nla_put_failure;
02f23f09	433
1e90474c	434	return nla_nest_end(skb, opts);
1da177e4	435
1e90474c	436	nla_put_failure:
bc3ed28c TG	437	nla_nest_cancel(skb, opts);
bc3ed28c TG	438	return -EMSGSIZE;
1da177e4 LT	439	}
	440
	441	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	442	{
81da99ed	443	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	444	struct nlattr *opts = NULL;
1da177e4	445
1e90474c PM	446	opts = nla_nest_start(skb, TCA_OPTIONS);
	447	if (opts == NULL)
	448	goto nla_put_failure;
1b34ec43 DM	449	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
1b34ec43 DM	450	goto nla_put_failure;
02f23f09	451
1b34ec43 DM	452	if (p->default_index != NO_DEFAULT_INDEX &&
	453	nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
	454	goto nla_put_failure;
1da177e4	455
1b34ec43 DM	456	if (p->set_tc_index &&
	457	nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
	458	goto nla_put_failure;
02f23f09	459
1e90474c	460	return nla_nest_end(skb, opts);
1da177e4	461
1e90474c	462	nla_put_failure:
bc3ed28c TG	463	nla_nest_cancel(skb, opts);
bc3ed28c TG	464	return -EMSGSIZE;
1da177e4 LT	465	}
1da177e4 LT	466
20fea08b	467	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	468	.graft = dsmark_graft,
	469	.leaf = dsmark_leaf,
	470	.get = dsmark_get,
	471	.put = dsmark_put,
	472	.change = dsmark_change,
	473	.delete = dsmark_delete,
	474	.walk = dsmark_walk,
	475	.tcf_chain = dsmark_find_tcf,
	476	.bind_tcf = dsmark_bind_filter,
	477	.unbind_tcf = dsmark_put,
	478	.dump = dsmark_dump_class,
	479	};
	480
20fea08b	481	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	482	.next = NULL,
	483	.cl_ops = &dsmark_class_ops,
	484	.id = "dsmark",
	485	.priv_size = sizeof(struct dsmark_qdisc_data),
	486	.enqueue = dsmark_enqueue,
	487	.dequeue = dsmark_dequeue,
8e3af978	488	.peek = dsmark_peek,
1da177e4 LT	489	.drop = dsmark_drop,
	490	.init = dsmark_init,
	491	.reset = dsmark_reset,
	492	.destroy = dsmark_destroy,
	493	.change = NULL,
	494	.dump = dsmark_dump,
	495	.owner = THIS_MODULE,
	496	};
	497
	498	static int __init dsmark_module_init(void)
	499	{
	500	return register_qdisc(&dsmark_qdisc_ops);
	501	}
af0d1141	502
10297b99	503	static void __exit dsmark_module_exit(void)
1da177e4 LT	504	{
	505	unregister_qdisc(&dsmark_qdisc_ops);
	506	}
af0d1141	507
1da177e4 LT	508	module_init(dsmark_module_init)
1da177e4 LT	509	module_exit(dsmark_module_exit)
af0d1141	510
1da177e4	511	MODULE_LICENSE("GPL");