[linux-2.6-block.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 sk_buff **to_free)
{
	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:
			__qdisc_drop(skb, to_free);
			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, to_free);
	if (err != NET_XMIT_SUCCESS) {
		if (net_xmit_drop_count(err))
			qdisc_qstats_drop(sch);
		return err;
	}

	qdisc_qstats_backlog_inc(sch, skb);
	sch->q.qlen++;

	return NET_XMIT_SUCCESS;

drop:
	qdisc_drop(skb, sch, to_free);
	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 = qdisc_dequeue_peeked(p->q);
	if (skb == NULL)
		return NULL;

	qdisc_bstats_update(sch, skb);
	qdisc_qstats_backlog_dec(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 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->qstats.backlog = 0;
	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,
	.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
520ac30f ED	194	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch,
520ac30f ED	195	struct sk_buff **to_free)
1da177e4	196	{
81da99ed	197	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	198	int err;
af0d1141 TG	199
c76f2a2c	200	pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
1da177e4	201
1da177e4	202	if (p->set_tc_index) {
d8b9605d	203	switch (tc_skb_protocol(skb)) {
60678040	204	case htons(ETH_P_IP):
9d127fbd SH	205	if (skb_cow_head(skb, sizeof(struct iphdr)))
9d127fbd SH	206	goto drop;
4c30719f	207
9d127fbd SH	208	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	209	& ~INET_ECN_MASK;
	210	break;
4c30719f	211
60678040	212	case htons(ETH_P_IPV6):
9d127fbd SH	213	if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
9d127fbd SH	214	goto drop;
4c30719f	215
9d127fbd SH	216	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	217	& ~INET_ECN_MASK;
	218	break;
	219	default:
	220	skb->tc_index = 0;
	221	break;
3ff50b79	222	}
1da177e4	223	}
af0d1141 TG	224
af0d1141 TG	225	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	226	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	227	else {
af0d1141 TG	228	struct tcf_result res;
25d8c0d5	229	struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
3b3ae880	230	int result = tc_classify(skb, fl, &res, false);
af0d1141	231
81da99ed	232	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	233
1da177e4	234	switch (result) {
f6853e2d PM	235	#ifdef CONFIG_NET_CLS_ACT
	236	case TC_ACT_QUEUED:
	237	case TC_ACT_STOLEN:
520ac30f	238	__qdisc_drop(skb, to_free);
378a2f09	239	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	240
f6853e2d	241	case TC_ACT_SHOT:
4c30719f	242	goto drop;
1da177e4	243	#endif
c3bc7cff	244	case TC_ACT_OK:
f6853e2d PM	245	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	246	break;
4c30719f	247
f6853e2d PM	248	default:
	249	if (p->default_index != NO_DEFAULT_INDEX)
	250	skb->tc_index = p->default_index;
	251	break;
3ff50b79	252	}
1da177e4	253	}
1da177e4	254
520ac30f	255	err = qdisc_enqueue(skb, p->q, to_free);
af0d1141	256	if (err != NET_XMIT_SUCCESS) {
378a2f09	257	if (net_xmit_drop_count(err))
25331d6c	258	qdisc_qstats_drop(sch);
af0d1141	259	return err;
1da177e4	260	}
af0d1141	261
bdf17661	262	qdisc_qstats_backlog_inc(sch, skb);
1da177e4	263	sch->q.qlen++;
1da177e4	264
af0d1141	265	return NET_XMIT_SUCCESS;
4c30719f SH	266
4c30719f SH	267	drop:
520ac30f	268	qdisc_drop(skb, sch, to_free);
c27f339a	269	return NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
af0d1141	270	}
1da177e4 LT	271
	272	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	273	{
81da99ed	274	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	275	struct sk_buff *skb;
af0d1141 TG	276	u32 index;
af0d1141 TG	277
c76f2a2c	278	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	279
f8b33d8e	280	skb = qdisc_dequeue_peeked(p->q);
af0d1141	281	if (skb == NULL)
1da177e4	282	return NULL;
af0d1141	283
9190b3b3	284	qdisc_bstats_update(sch, skb);
bdf17661	285	qdisc_qstats_backlog_dec(sch, skb);
1da177e4	286	sch->q.qlen--;
af0d1141 TG	287
af0d1141 TG	288	index = skb->tc_index & (p->indices - 1);
81da99ed	289	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	290
d8b9605d	291	switch (tc_skb_protocol(skb)) {
60678040	292	case htons(ETH_P_IP):
47bbbb30 ED	293	ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
47bbbb30 ED	294	p->mv[index].value);
1da177e4	295	break;
60678040	296	case htons(ETH_P_IPV6):
47bbbb30 ED	297	ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
47bbbb30 ED	298	p->mv[index].value);
1da177e4	299	break;
9d127fbd SH	300	default:
	301	/*
	302	* Only complain if a change was actually attempted.
	303	* This way, we can send non-IP traffic through dsmark
	304	* and don't need yet another qdisc as a bypass.
	305	*/
47bbbb30	306	if (p->mv[index].mask != 0xff \|\| p->mv[index].value)
c76f2a2c	307	pr_warn("%s: unsupported protocol %d\n",
d8b9605d	308	__func__, ntohs(tc_skb_protocol(skb)));
9d127fbd	309	break;
3ff50b79	310	}
af0d1141	311
1da177e4 LT	312	return skb;
	313	}
	314
8e3af978 JP	315	static struct sk_buff dsmark_peek(struct Qdisc sch)
	316	{
	317	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	318
c76f2a2c	319	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
8e3af978 JP	320
	321	return p->q->ops->peek(p->q);
	322	}
	323
1e90474c	324	static int dsmark_init(struct Qdisc sch, struct nlattr opt)
1da177e4	325	{
81da99ed	326	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	327	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c TG	328	int err = -EINVAL;
	329	u32 default_index = NO_DEFAULT_INDEX;
	330	u16 indices;
47bbbb30	331	int i;
758cc43c	332
c76f2a2c	333	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
758cc43c	334
cee63723 PM	335	if (!opt)
	336	goto errout;
	337
27a3421e	338	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	339	if (err < 0)
758cc43c TG	340	goto errout;
758cc43c TG	341
cee63723	342	err = -EINVAL;
1e90474c	343	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	344
5b0ac72b DM	345	if (hweight32(indices) != 1)
758cc43c TG	346	goto errout;
758cc43c TG	347
27a3421e	348	if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c	349	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c	350
47bbbb30 ED	351	if (indices <= DSMARK_EMBEDDED_SZ)
	352	p->mv = p->embedded;
	353	else
	354	p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
	355	if (!p->mv) {
758cc43c TG	356	err = -ENOMEM;
758cc43c TG	357	goto errout;
1da177e4	358	}
47bbbb30 ED	359	for (i = 0; i < indices; i++) {
	360	p->mv[i].mask = 0xff;
	361	p->mv[i].value = 0;
	362	}
758cc43c TG	363	p->indices = indices;
758cc43c TG	364	p->default_index = default_index;
1e90474c	365	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	366
3511c913	367	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
758cc43c	368	if (p->q == NULL)
1da177e4	369	p->q = &noop_qdisc;
758cc43c	370
c76f2a2c	371	pr_debug("%s: qdisc %p\n", __func__, p->q);
758cc43c TG	372
	373	err = 0;
	374	errout:
758cc43c	375	return err;
1da177e4 LT	376	}
1da177e4 LT	377
1da177e4 LT	378	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	379	{
81da99ed	380	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	381
c76f2a2c	382	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	383	qdisc_reset(p->q);
bdf17661	384	sch->qstats.backlog = 0;
1da177e4 LT	385	sch->q.qlen = 0;
	386	}
	387
1da177e4 LT	388	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	389	{
81da99ed	390	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	391
c76f2a2c	392	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141	393
ff31ab56	394	tcf_destroy_chain(&p->filter_list);
1da177e4	395	qdisc_destroy(p->q);
47bbbb30 ED	396	if (p->mv != p->embedded)
47bbbb30 ED	397	kfree(p->mv);
1da177e4 LT	398	}
1da177e4 LT	399
1da177e4	400	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	401	struct sk_buff skb, struct tcmsg tcm)
1da177e4	402	{
81da99ed	403	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	404	struct nlattr *opts = NULL;
1da177e4	405
c76f2a2c	406	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
02f23f09 TG	407
02f23f09 TG	408	if (!dsmark_valid_index(p, cl))
1da177e4	409	return -EINVAL;
02f23f09	410
cc7ec456	411	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
cdc7f8e3	412	tcm->tcm_info = p->q->handle;
02f23f09	413
1e90474c PM	414	opts = nla_nest_start(skb, TCA_OPTIONS);
	415	if (opts == NULL)
	416	goto nla_put_failure;
47bbbb30 ED	417	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) \|\|
47bbbb30 ED	418	nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
1b34ec43	419	goto nla_put_failure;
02f23f09	420
1e90474c	421	return nla_nest_end(skb, opts);
1da177e4	422
1e90474c	423	nla_put_failure:
bc3ed28c TG	424	nla_nest_cancel(skb, opts);
bc3ed28c TG	425	return -EMSGSIZE;
1da177e4 LT	426	}
	427
	428	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	429	{
81da99ed	430	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	431	struct nlattr *opts = NULL;
1da177e4	432
1e90474c PM	433	opts = nla_nest_start(skb, TCA_OPTIONS);
	434	if (opts == NULL)
	435	goto nla_put_failure;
1b34ec43 DM	436	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
1b34ec43 DM	437	goto nla_put_failure;
02f23f09	438
1b34ec43 DM	439	if (p->default_index != NO_DEFAULT_INDEX &&
	440	nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
	441	goto nla_put_failure;
1da177e4	442
1b34ec43 DM	443	if (p->set_tc_index &&
	444	nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
	445	goto nla_put_failure;
02f23f09	446
1e90474c	447	return nla_nest_end(skb, opts);
1da177e4	448
1e90474c	449	nla_put_failure:
bc3ed28c TG	450	nla_nest_cancel(skb, opts);
bc3ed28c TG	451	return -EMSGSIZE;
1da177e4 LT	452	}
1da177e4 LT	453
20fea08b	454	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	455	.graft = dsmark_graft,
	456	.leaf = dsmark_leaf,
	457	.get = dsmark_get,
	458	.put = dsmark_put,
	459	.change = dsmark_change,
	460	.delete = dsmark_delete,
	461	.walk = dsmark_walk,
	462	.tcf_chain = dsmark_find_tcf,
	463	.bind_tcf = dsmark_bind_filter,
	464	.unbind_tcf = dsmark_put,
	465	.dump = dsmark_dump_class,
	466	};
	467
20fea08b	468	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	469	.next = NULL,
	470	.cl_ops = &dsmark_class_ops,
	471	.id = "dsmark",
	472	.priv_size = sizeof(struct dsmark_qdisc_data),
	473	.enqueue = dsmark_enqueue,
	474	.dequeue = dsmark_dequeue,
8e3af978	475	.peek = dsmark_peek,
1da177e4 LT	476	.init = dsmark_init,
	477	.reset = dsmark_reset,
	478	.destroy = dsmark_destroy,
	479	.change = NULL,
	480	.dump = dsmark_dump,
	481	.owner = THIS_MODULE,
	482	};
	483
	484	static int __init dsmark_module_init(void)
	485	{
	486	return register_qdisc(&dsmark_qdisc_ops);
	487	}
af0d1141	488
10297b99	489	static void __exit dsmark_module_exit(void)
1da177e4 LT	490	{
	491	unregister_qdisc(&dsmark_qdisc_ops);
	492	}
af0d1141	493
1da177e4 LT	494	module_init(dsmark_module_init)
1da177e4 LT	495	module_exit(dsmark_module_exit)
af0d1141	496
1da177e4	497	MODULE_LICENSE("GPL");