[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;
	}

	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("%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
1da177e4	76	sch_tree_lock(sch);
b94c8afc PM	77	*old = p->q;
b94c8afc PM	78	p->q = new;
5e50da01	79	qdisc_tree_decrease_qlen(old, (old)->q.qlen);
af0d1141	80	qdisc_reset(*old);
af0d1141 TG	81	sch_tree_unlock(sch);
af0d1141 TG	82
10297b99	83	return 0;
1da177e4 LT	84	}
1da177e4 LT	85
1da177e4 LT	86	static struct Qdisc dsmark_leaf(struct Qdisc sch, unsigned long arg)
1da177e4 LT	87	{
81da99ed SH	88	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	89	return p->q;
1da177e4 LT	90	}
1da177e4 LT	91
af0d1141	92	static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
1da177e4	93	{
c76f2a2c YY	94	pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
c76f2a2c YY	95	__func__, sch, qdisc_priv(sch), classid);
1da177e4	96
af0d1141	97	return TC_H_MIN(classid) + 1;
1da177e4 LT	98	}
1da177e4 LT	99
1da177e4	100	static unsigned long dsmark_bind_filter(struct Qdisc *sch,
af0d1141	101	unsigned long parent, u32 classid)
1da177e4	102	{
af0d1141	103	return dsmark_get(sch, classid);
1da177e4 LT	104	}
1da177e4 LT	105
1da177e4 LT	106	static void dsmark_put(struct Qdisc *sch, unsigned long cl)
	107	{
	108	}
	109
27a3421e PM	110	static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
	111	[TCA_DSMARK_INDICES] = { .type = NLA_U16 },
	112	[TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
	113	[TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
	114	[TCA_DSMARK_MASK] = { .type = NLA_U8 },
	115	[TCA_DSMARK_VALUE] = { .type = NLA_U8 },
	116	};
	117
1da177e4	118	static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
1e90474c	119	struct nlattr *tca, unsigned long arg)
1da177e4	120	{
81da99ed	121	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c PM	122	struct nlattr *opt = tca[TCA_OPTIONS];
1e90474c PM	123	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c	124	int err = -EINVAL;
1da177e4	125
c76f2a2c YY	126	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
c76f2a2c YY	127	__func__, sch, p, classid, parent, *arg);
758cc43c TG	128
	129	if (!dsmark_valid_index(p, *arg)) {
	130	err = -ENOENT;
1e90474c	131	goto errout;
1da177e4	132	}
1da177e4	133
cee63723	134	if (!opt)
1e90474c	135	goto errout;
758cc43c	136
27a3421e	137	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	138	if (err < 0)
27a3421e	139	goto errout;
cee63723	140
27a3421e	141	if (tb[TCA_DSMARK_VALUE])
47bbbb30	142	p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99	143
1e90474c	144	if (tb[TCA_DSMARK_MASK])
47bbbb30	145	p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
758cc43c TG	146
	147	err = 0;
	148
1e90474c	149	errout:
758cc43c TG	150	return err;
758cc43c TG	151	}
1da177e4	152
af0d1141	153	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4	154	{
81da99ed	155	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	156
af0d1141	157	if (!dsmark_valid_index(p, arg))
1da177e4	158	return -EINVAL;
10297b99	159
47bbbb30 ED	160	p->mv[arg - 1].mask = 0xff;
47bbbb30 ED	161	p->mv[arg - 1].value = 0;
af0d1141	162
1da177e4 LT	163	return 0;
	164	}
	165
9d127fbd	166	static void dsmark_walk(struct Qdisc sch, struct qdisc_walker walker)
1da177e4	167	{
81da99ed	168	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 LT	169	int i;
1da177e4 LT	170
c76f2a2c YY	171	pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
c76f2a2c YY	172	__func__, sch, p, walker);
af0d1141	173
1da177e4 LT	174	if (walker->stop)
1da177e4 LT	175	return;
af0d1141	176
1da177e4	177	for (i = 0; i < p->indices; i++) {
47bbbb30	178	if (p->mv[i].mask == 0xff && !p->mv[i].value)
0451eb07	179	goto ignore;
1da177e4	180	if (walker->count >= walker->skip) {
cc7ec456	181	if (walker->fn(sch, i + 1, walker) < 0) {
1da177e4 LT	182	walker->stop = 1;
	183	break;
	184	}
	185	}
10297b99	186	ignore:
0451eb07	187	walker->count++;
10297b99	188	}
1da177e4 LT	189	}
1da177e4 LT	190
25d8c0d5 JF	191	static inline struct tcf_proto __rcu *dsmark_find_tcf(struct Qdisc sch,
25d8c0d5 JF	192	unsigned long cl)
1da177e4	193	{
81da99ed SH	194	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	195	return &p->filter_list;
1da177e4 LT	196	}
1da177e4 LT	197
1da177e4 LT	198	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	199
9d127fbd	200	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	201	{
81da99ed	202	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	203	int err;
af0d1141 TG	204
c76f2a2c	205	pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
1da177e4	206
1da177e4	207	if (p->set_tc_index) {
d8b9605d	208	switch (tc_skb_protocol(skb)) {
60678040	209	case htons(ETH_P_IP):
9d127fbd SH	210	if (skb_cow_head(skb, sizeof(struct iphdr)))
9d127fbd SH	211	goto drop;
4c30719f	212
9d127fbd SH	213	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	214	& ~INET_ECN_MASK;
	215	break;
4c30719f	216
60678040	217	case htons(ETH_P_IPV6):
9d127fbd SH	218	if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
9d127fbd SH	219	goto drop;
4c30719f	220
9d127fbd SH	221	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	222	& ~INET_ECN_MASK;
	223	break;
	224	default:
	225	skb->tc_index = 0;
	226	break;
3ff50b79	227	}
1da177e4	228	}
af0d1141 TG	229
af0d1141 TG	230	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	231	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	232	else {
af0d1141 TG	233	struct tcf_result res;
25d8c0d5	234	struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
3b3ae880	235	int result = tc_classify(skb, fl, &res, false);
af0d1141	236
81da99ed	237	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	238
1da177e4	239	switch (result) {
f6853e2d PM	240	#ifdef CONFIG_NET_CLS_ACT
	241	case TC_ACT_QUEUED:
	242	case TC_ACT_STOLEN:
	243	kfree_skb(skb);
378a2f09	244	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	245
f6853e2d	246	case TC_ACT_SHOT:
4c30719f	247	goto drop;
1da177e4	248	#endif
c3bc7cff	249	case TC_ACT_OK:
f6853e2d PM	250	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	251	break;
4c30719f	252
f6853e2d PM	253	default:
	254	if (p->default_index != NO_DEFAULT_INDEX)
	255	skb->tc_index = p->default_index;
	256	break;
3ff50b79	257	}
1da177e4	258	}
1da177e4	259
5f86173b	260	err = qdisc_enqueue(skb, p->q);
af0d1141	261	if (err != NET_XMIT_SUCCESS) {
378a2f09	262	if (net_xmit_drop_count(err))
25331d6c	263	qdisc_qstats_drop(sch);
af0d1141	264	return err;
1da177e4	265	}
af0d1141	266
1da177e4	267	sch->q.qlen++;
1da177e4	268
af0d1141	269	return NET_XMIT_SUCCESS;
4c30719f SH	270
4c30719f SH	271	drop:
17045755	272	qdisc_drop(skb, sch);
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
c76f2a2c	282	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	283
1da177e4	284	skb = p->q->ops->dequeue(p->q);
af0d1141	285	if (skb == NULL)
1da177e4	286	return NULL;
af0d1141	287
9190b3b3	288	qdisc_bstats_update(sch, skb);
1da177e4	289	sch->q.qlen--;
af0d1141 TG	290
af0d1141 TG	291	index = skb->tc_index & (p->indices - 1);
81da99ed	292	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	293
d8b9605d	294	switch (tc_skb_protocol(skb)) {
60678040	295	case htons(ETH_P_IP):
47bbbb30 ED	296	ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
47bbbb30 ED	297	p->mv[index].value);
1da177e4	298	break;
60678040	299	case htons(ETH_P_IPV6):
47bbbb30 ED	300	ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
47bbbb30 ED	301	p->mv[index].value);
1da177e4	302	break;
9d127fbd SH	303	default:
	304	/*
	305	* Only complain if a change was actually attempted.
	306	* This way, we can send non-IP traffic through dsmark
	307	* and don't need yet another qdisc as a bypass.
	308	*/
47bbbb30	309	if (p->mv[index].mask != 0xff \|\| p->mv[index].value)
c76f2a2c	310	pr_warn("%s: unsupported protocol %d\n",
d8b9605d	311	__func__, ntohs(tc_skb_protocol(skb)));
9d127fbd	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
c76f2a2c	322	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
8e3af978 JP	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
c76f2a2c	332	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, 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;
47bbbb30	351	int i;
758cc43c	352
c76f2a2c	353	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, 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	370
47bbbb30 ED	371	if (indices <= DSMARK_EMBEDDED_SZ)
	372	p->mv = p->embedded;
	373	else
	374	p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
	375	if (!p->mv) {
758cc43c TG	376	err = -ENOMEM;
758cc43c TG	377	goto errout;
1da177e4	378	}
47bbbb30 ED	379	for (i = 0; i < indices; i++) {
	380	p->mv[i].mask = 0xff;
	381	p->mv[i].value = 0;
	382	}
758cc43c TG	383	p->indices = indices;
758cc43c TG	384	p->default_index = default_index;
1e90474c	385	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	386
3511c913	387	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
758cc43c	388	if (p->q == NULL)
1da177e4	389	p->q = &noop_qdisc;
758cc43c	390
c76f2a2c	391	pr_debug("%s: qdisc %p\n", __func__, p->q);
758cc43c TG	392
	393	err = 0;
	394	errout:
758cc43c	395	return err;
1da177e4 LT	396	}
1da177e4 LT	397
1da177e4 LT	398	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	399	{
81da99ed	400	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	401
c76f2a2c	402	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4 LT	403	qdisc_reset(p->q);
	404	sch->q.qlen = 0;
	405	}
	406
1da177e4 LT	407	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	408	{
81da99ed	409	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	410
c76f2a2c	411	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141	412
ff31ab56	413	tcf_destroy_chain(&p->filter_list);
1da177e4	414	qdisc_destroy(p->q);
47bbbb30 ED	415	if (p->mv != p->embedded)
47bbbb30 ED	416	kfree(p->mv);
1da177e4 LT	417	}
1da177e4 LT	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
c76f2a2c	425	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
02f23f09 TG	426
02f23f09 TG	427	if (!dsmark_valid_index(p, cl))
1da177e4	428	return -EINVAL;
02f23f09	429
cc7ec456	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;
47bbbb30 ED	436	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) \|\|
47bbbb30 ED	437	nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
1b34ec43	438	goto nla_put_failure;
02f23f09	439
1e90474c	440	return nla_nest_end(skb, opts);
1da177e4	441
1e90474c	442	nla_put_failure:
bc3ed28c TG	443	nla_nest_cancel(skb, opts);
bc3ed28c TG	444	return -EMSGSIZE;
1da177e4 LT	445	}
	446
	447	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	448	{
81da99ed	449	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	450	struct nlattr *opts = NULL;
1da177e4	451
1e90474c PM	452	opts = nla_nest_start(skb, TCA_OPTIONS);
	453	if (opts == NULL)
	454	goto nla_put_failure;
1b34ec43 DM	455	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
1b34ec43 DM	456	goto nla_put_failure;
02f23f09	457
1b34ec43 DM	458	if (p->default_index != NO_DEFAULT_INDEX &&
	459	nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
	460	goto nla_put_failure;
1da177e4	461
1b34ec43 DM	462	if (p->set_tc_index &&
	463	nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
	464	goto nla_put_failure;
02f23f09	465
1e90474c	466	return nla_nest_end(skb, opts);
1da177e4	467
1e90474c	468	nla_put_failure:
bc3ed28c TG	469	nla_nest_cancel(skb, opts);
bc3ed28c TG	470	return -EMSGSIZE;
1da177e4 LT	471	}
1da177e4 LT	472
20fea08b	473	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	474	.graft = dsmark_graft,
	475	.leaf = dsmark_leaf,
	476	.get = dsmark_get,
	477	.put = dsmark_put,
	478	.change = dsmark_change,
	479	.delete = dsmark_delete,
	480	.walk = dsmark_walk,
	481	.tcf_chain = dsmark_find_tcf,
	482	.bind_tcf = dsmark_bind_filter,
	483	.unbind_tcf = dsmark_put,
	484	.dump = dsmark_dump_class,
	485	};
	486
20fea08b	487	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	488	.next = NULL,
	489	.cl_ops = &dsmark_class_ops,
	490	.id = "dsmark",
	491	.priv_size = sizeof(struct dsmark_qdisc_data),
	492	.enqueue = dsmark_enqueue,
	493	.dequeue = dsmark_dequeue,
8e3af978	494	.peek = dsmark_peek,
1da177e4 LT	495	.drop = dsmark_drop,
	496	.init = dsmark_init,
	497	.reset = dsmark_reset,
	498	.destroy = dsmark_destroy,
	499	.change = NULL,
	500	.dump = dsmark_dump,
	501	.owner = THIS_MODULE,
	502	};
	503
	504	static int __init dsmark_module_init(void)
	505	{
	506	return register_qdisc(&dsmark_qdisc_ops);
	507	}
af0d1141	508
10297b99	509	static void __exit dsmark_module_exit(void)
1da177e4 LT	510	{
	511	unregister_qdisc(&dsmark_qdisc_ops);
	512	}
af0d1141	513
1da177e4 LT	514	module_init(dsmark_module_init)
1da177e4 LT	515	module_exit(dsmark_module_exit)
af0d1141	516
1da177e4	517	MODULE_LICENSE("GPL");