[people/arne_f/kernel.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/pkt_cls.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 tcf_block	*block;
	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_find(struct Qdisc *sch, u32 classid)
{
	return TC_H_MIN(classid) + 1;
}

static unsigned long dsmark_bind_filter(struct Qdisc *sch,
					unsigned long parent, u32 classid)
{
	pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
		 __func__, sch, qdisc_priv(sch), classid);

	return dsmark_find(sch, classid);
}

static void dsmark_unbind_filter(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, NULL);
	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 struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);

	return p->block;
}

/* --------------------------- 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) {
		int wlen = skb_network_offset(skb);

		switch (tc_skb_protocol(skb)) {
		case htons(ETH_P_IP):
			wlen += sizeof(struct iphdr);
			if (!pskb_may_pull(skb, wlen) ||
			    skb_try_make_writable(skb, wlen))
				goto drop;

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

		case htons(ETH_P_IPV6):
			wlen += sizeof(struct ipv6hdr);
			if (!pskb_may_pull(skb, wlen) ||
			    skb_try_make_writable(skb, wlen))
				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 = tcf_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:
		case TC_ACT_TRAP:
			__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 = tcf_block_get(&p->block, &p->filter_list);
	if (err)
		return err;

	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
	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;
	else
		qdisc_hash_add(p->q, true);

	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_block_put(p->block);
	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,
	.find		=	dsmark_find,
	.change		=	dsmark_change,
	.delete		=	dsmark_delete,
	.walk		=	dsmark_walk,
	.tcf_block	=	dsmark_tcf_block,
	.bind_tcf	=	dsmark_bind_filter,
	.unbind_tcf	=	dsmark_unbind_filter,
	.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	15	#include <net/pkt_sched.h>
cf1facda	16	#include <net/pkt_cls.h>
1da177e4 LT	17	#include <net/dsfield.h>
	18	#include <net/inet_ecn.h>
	19	#include <asm/byteorder.h>
	20
1da177e4 LT	21	/*
	22	* classid class marking
	23	* ------- ----- -------
	24	* n/a 0 n/a
	25	* x:0 1 use entry [0]
	26	* ... ... ...
	27	* x:y y>0 y+1 use entry [y]
	28	* ... ... ...
	29	* x:indices-1 indices use entry [indices-1]
	30	* ... ... ...
	31	* x:y y+1 use entry [y & (indices-1)]
	32	* ... ... ...
	33	* 0xffff 0x10000 use entry [indices-1]
	34	*/
	35
	36
	37	#define NO_DEFAULT_INDEX (1 << 16)
	38
47bbbb30 ED	39	struct mask_value {
	40	u8 mask;
	41	u8 value;
	42	};
	43
1da177e4 LT	44	struct dsmark_qdisc_data {
1da177e4 LT	45	struct Qdisc *q;
25d8c0d5	46	struct tcf_proto __rcu *filter_list;
6529eaba	47	struct tcf_block *block;
47bbbb30	48	struct mask_value *mv;
af0d1141	49	u16 indices;
47bbbb30	50	u8 set_tc_index;
af0d1141	51	u32 default_index; /* index range is 0...0xffff */
47bbbb30 ED	52	#define DSMARK_EMBEDDED_SZ 16
47bbbb30 ED	53	struct mask_value embedded[DSMARK_EMBEDDED_SZ];
1da177e4 LT	54	};
1da177e4 LT	55
758cc43c TG	56	static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
758cc43c TG	57	{
17569fae	58	return index <= p->indices && index > 0;
758cc43c	59	}
1da177e4 LT	60
	61	/* ------------------------- Class/flow operations ------------------------- */
	62
af0d1141 TG	63	static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
af0d1141 TG	64	struct Qdisc new, struct Qdisc *old)
1da177e4	65	{
81da99ed	66	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	67
c76f2a2c YY	68	pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
c76f2a2c YY	69	__func__, sch, p, new, old);
486b53e5 TG	70
486b53e5 TG	71	if (new == NULL) {
3511c913	72	new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
9f9afec4	73	sch->handle);
486b53e5 TG	74	if (new == NULL)
	75	new = &noop_qdisc;
	76	}
	77
86a7996c	78	*old = qdisc_replace(sch, new, &p->q);
10297b99	79	return 0;
1da177e4 LT	80	}
1da177e4 LT	81
1da177e4 LT	82	static struct Qdisc dsmark_leaf(struct Qdisc sch, unsigned long arg)
1da177e4 LT	83	{
81da99ed SH	84	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	85	return p->q;
1da177e4 LT	86	}
1da177e4 LT	87
143976ce	88	static unsigned long dsmark_find(struct Qdisc *sch, u32 classid)
1da177e4	89	{
af0d1141	90	return TC_H_MIN(classid) + 1;
1da177e4 LT	91	}
1da177e4 LT	92
1da177e4	93	static unsigned long dsmark_bind_filter(struct Qdisc *sch,
af0d1141	94	unsigned long parent, u32 classid)
1da177e4	95	{
143976ce WC	96	pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
	97	__func__, sch, qdisc_priv(sch), classid);
	98
	99	return dsmark_find(sch, classid);
1da177e4 LT	100	}
1da177e4 LT	101
143976ce	102	static void dsmark_unbind_filter(struct Qdisc *sch, unsigned long cl)
1da177e4 LT	103	{
	104	}
	105
27a3421e PM	106	static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
	107	[TCA_DSMARK_INDICES] = { .type = NLA_U16 },
	108	[TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
	109	[TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
	110	[TCA_DSMARK_MASK] = { .type = NLA_U8 },
	111	[TCA_DSMARK_VALUE] = { .type = NLA_U8 },
	112	};
	113
1da177e4	114	static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
1e90474c	115	struct nlattr *tca, unsigned long arg)
1da177e4	116	{
81da99ed	117	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c PM	118	struct nlattr *opt = tca[TCA_OPTIONS];
1e90474c PM	119	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c	120	int err = -EINVAL;
1da177e4	121
c76f2a2c YY	122	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
c76f2a2c YY	123	__func__, sch, p, classid, parent, *arg);
758cc43c TG	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
fceb6435	133	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
cee63723	134	if (err < 0)
27a3421e	135	goto errout;
cee63723	136
27a3421e	137	if (tb[TCA_DSMARK_VALUE])
47bbbb30	138	p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99	139
1e90474c	140	if (tb[TCA_DSMARK_MASK])
47bbbb30	141	p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
758cc43c TG	142
	143	err = 0;
	144
1e90474c	145	errout:
758cc43c TG	146	return err;
758cc43c TG	147	}
1da177e4	148
af0d1141	149	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4	150	{
81da99ed	151	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	152
af0d1141	153	if (!dsmark_valid_index(p, arg))
1da177e4	154	return -EINVAL;
10297b99	155
47bbbb30 ED	156	p->mv[arg - 1].mask = 0xff;
47bbbb30 ED	157	p->mv[arg - 1].value = 0;
af0d1141	158
1da177e4 LT	159	return 0;
	160	}
	161
9d127fbd	162	static void dsmark_walk(struct Qdisc sch, struct qdisc_walker walker)
1da177e4	163	{
81da99ed	164	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 LT	165	int i;
1da177e4 LT	166
c76f2a2c YY	167	pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
c76f2a2c YY	168	__func__, sch, p, walker);
af0d1141	169
1da177e4 LT	170	if (walker->stop)
1da177e4 LT	171	return;
af0d1141	172
1da177e4	173	for (i = 0; i < p->indices; i++) {
47bbbb30	174	if (p->mv[i].mask == 0xff && !p->mv[i].value)
0451eb07	175	goto ignore;
1da177e4	176	if (walker->count >= walker->skip) {
cc7ec456	177	if (walker->fn(sch, i + 1, walker) < 0) {
1da177e4 LT	178	walker->stop = 1;
	179	break;
	180	}
	181	}
10297b99	182	ignore:
0451eb07	183	walker->count++;
10297b99	184	}
1da177e4 LT	185	}
1da177e4 LT	186
6529eaba	187	static struct tcf_block dsmark_tcf_block(struct Qdisc sch, unsigned long cl)
1da177e4	188	{
81da99ed	189	struct dsmark_qdisc_data *p = qdisc_priv(sch);
6529eaba JP	190
6529eaba JP	191	return p->block;
1da177e4 LT	192	}
1da177e4 LT	193
1da177e4 LT	194	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	195
520ac30f ED	196	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch,
520ac30f ED	197	struct sk_buff **to_free)
1da177e4	198	{
81da99ed	199	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	200	int err;
af0d1141 TG	201
c76f2a2c	202	pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
1da177e4	203
1da177e4	204	if (p->set_tc_index) {
aea92fb2 ED	205	int wlen = skb_network_offset(skb);
aea92fb2 ED	206
d8b9605d	207	switch (tc_skb_protocol(skb)) {
60678040	208	case htons(ETH_P_IP):
aea92fb2 ED	209	wlen += sizeof(struct iphdr);
	210	if (!pskb_may_pull(skb, wlen) \|\|
	211	skb_try_make_writable(skb, wlen))
9d127fbd	212	goto drop;
4c30719f	213
9d127fbd SH	214	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	215	& ~INET_ECN_MASK;
	216	break;
4c30719f	217
60678040	218	case htons(ETH_P_IPV6):
aea92fb2 ED	219	wlen += sizeof(struct ipv6hdr);
	220	if (!pskb_may_pull(skb, wlen) \|\|
	221	skb_try_make_writable(skb, wlen))
9d127fbd	222	goto drop;
4c30719f	223
9d127fbd SH	224	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	225	& ~INET_ECN_MASK;
	226	break;
	227	default:
	228	skb->tc_index = 0;
	229	break;
3ff50b79	230	}
1da177e4	231	}
af0d1141 TG	232
af0d1141 TG	233	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	234	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	235	else {
af0d1141 TG	236	struct tcf_result res;
25d8c0d5	237	struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
87d83093	238	int result = tcf_classify(skb, fl, &res, false);
af0d1141	239
81da99ed	240	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	241
1da177e4	242	switch (result) {
f6853e2d PM	243	#ifdef CONFIG_NET_CLS_ACT
	244	case TC_ACT_QUEUED:
	245	case TC_ACT_STOLEN:
e25ea21f	246	case TC_ACT_TRAP:
520ac30f	247	__qdisc_drop(skb, to_free);
378a2f09	248	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	249
f6853e2d	250	case TC_ACT_SHOT:
4c30719f	251	goto drop;
1da177e4	252	#endif
c3bc7cff	253	case TC_ACT_OK:
f6853e2d PM	254	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	255	break;
4c30719f	256
f6853e2d PM	257	default:
	258	if (p->default_index != NO_DEFAULT_INDEX)
	259	skb->tc_index = p->default_index;
	260	break;
3ff50b79	261	}
1da177e4	262	}
1da177e4	263
520ac30f	264	err = qdisc_enqueue(skb, p->q, to_free);
af0d1141	265	if (err != NET_XMIT_SUCCESS) {
378a2f09	266	if (net_xmit_drop_count(err))
25331d6c	267	qdisc_qstats_drop(sch);
af0d1141	268	return err;
1da177e4	269	}
af0d1141	270
bdf17661	271	qdisc_qstats_backlog_inc(sch, skb);
1da177e4	272	sch->q.qlen++;
1da177e4	273
af0d1141	274	return NET_XMIT_SUCCESS;
4c30719f SH	275
4c30719f SH	276	drop:
520ac30f	277	qdisc_drop(skb, sch, to_free);
c27f339a	278	return NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
af0d1141	279	}
1da177e4 LT	280
	281	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	282	{
81da99ed	283	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	284	struct sk_buff *skb;
af0d1141 TG	285	u32 index;
af0d1141 TG	286
c76f2a2c	287	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	288
f8b33d8e	289	skb = qdisc_dequeue_peeked(p->q);
af0d1141	290	if (skb == NULL)
1da177e4	291	return NULL;
af0d1141	292
9190b3b3	293	qdisc_bstats_update(sch, skb);
bdf17661	294	qdisc_qstats_backlog_dec(sch, skb);
1da177e4	295	sch->q.qlen--;
af0d1141 TG	296
af0d1141 TG	297	index = skb->tc_index & (p->indices - 1);
81da99ed	298	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	299
d8b9605d	300	switch (tc_skb_protocol(skb)) {
60678040	301	case htons(ETH_P_IP):
47bbbb30 ED	302	ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
47bbbb30 ED	303	p->mv[index].value);
1da177e4	304	break;
60678040	305	case htons(ETH_P_IPV6):
47bbbb30 ED	306	ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
47bbbb30 ED	307	p->mv[index].value);
1da177e4	308	break;
9d127fbd SH	309	default:
	310	/*
	311	* Only complain if a change was actually attempted.
	312	* This way, we can send non-IP traffic through dsmark
	313	* and don't need yet another qdisc as a bypass.
	314	*/
47bbbb30	315	if (p->mv[index].mask != 0xff \|\| p->mv[index].value)
c76f2a2c	316	pr_warn("%s: unsupported protocol %d\n",
d8b9605d	317	__func__, ntohs(tc_skb_protocol(skb)));
9d127fbd	318	break;
3ff50b79	319	}
af0d1141	320
1da177e4 LT	321	return skb;
	322	}
	323
8e3af978 JP	324	static struct sk_buff dsmark_peek(struct Qdisc sch)
	325	{
	326	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	327
c76f2a2c	328	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
8e3af978 JP	329
	330	return p->q->ops->peek(p->q);
	331	}
	332
1e90474c	333	static int dsmark_init(struct Qdisc sch, struct nlattr opt)
1da177e4	334	{
81da99ed	335	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	336	struct nlattr *tb[TCA_DSMARK_MAX + 1];
9d4f97f9	337	int err = -EINVAL;
758cc43c TG	338	u32 default_index = NO_DEFAULT_INDEX;
758cc43c TG	339	u16 indices;
47bbbb30	340	int i;
758cc43c	341
c76f2a2c	342	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
758cc43c	343
cee63723 PM	344	if (!opt)
	345	goto errout;
	346
6529eaba JP	347	err = tcf_block_get(&p->block, &p->filter_list);
	348	if (err)
	349	return err;
	350
fceb6435	351	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
cee63723	352	if (err < 0)
758cc43c TG	353	goto errout;
758cc43c TG	354
cee63723	355	err = -EINVAL;
1e90474c	356	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	357
5b0ac72b DM	358	if (hweight32(indices) != 1)
758cc43c TG	359	goto errout;
758cc43c TG	360
27a3421e	361	if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c	362	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c	363
47bbbb30 ED	364	if (indices <= DSMARK_EMBEDDED_SZ)
	365	p->mv = p->embedded;
	366	else
	367	p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
	368	if (!p->mv) {
758cc43c TG	369	err = -ENOMEM;
758cc43c TG	370	goto errout;
1da177e4	371	}
47bbbb30 ED	372	for (i = 0; i < indices; i++) {
	373	p->mv[i].mask = 0xff;
	374	p->mv[i].value = 0;
	375	}
758cc43c TG	376	p->indices = indices;
758cc43c TG	377	p->default_index = default_index;
1e90474c	378	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	379
3511c913	380	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
758cc43c	381	if (p->q == NULL)
1da177e4	382	p->q = &noop_qdisc;
49b49971 JK	383	else
49b49971 JK	384	qdisc_hash_add(p->q, true);
758cc43c	385
c76f2a2c	386	pr_debug("%s: qdisc %p\n", __func__, p->q);
758cc43c TG	387
	388	err = 0;
	389	errout:
758cc43c	390	return err;
1da177e4 LT	391	}
1da177e4 LT	392
1da177e4 LT	393	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	394	{
81da99ed	395	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	396
c76f2a2c	397	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	398	qdisc_reset(p->q);
bdf17661	399	sch->qstats.backlog = 0;
1da177e4 LT	400	sch->q.qlen = 0;
	401	}
	402
1da177e4 LT	403	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	404	{
81da99ed	405	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	406
c76f2a2c	407	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141	408
6529eaba	409	tcf_block_put(p->block);
1da177e4	410	qdisc_destroy(p->q);
47bbbb30 ED	411	if (p->mv != p->embedded)
47bbbb30 ED	412	kfree(p->mv);
1da177e4 LT	413	}
1da177e4 LT	414
1da177e4	415	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	416	struct sk_buff skb, struct tcmsg tcm)
1da177e4	417	{
81da99ed	418	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	419	struct nlattr *opts = NULL;
1da177e4	420
c76f2a2c	421	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
02f23f09 TG	422
02f23f09 TG	423	if (!dsmark_valid_index(p, cl))
1da177e4	424	return -EINVAL;
02f23f09	425
cc7ec456	426	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
cdc7f8e3	427	tcm->tcm_info = p->q->handle;
02f23f09	428
1e90474c PM	429	opts = nla_nest_start(skb, TCA_OPTIONS);
	430	if (opts == NULL)
	431	goto nla_put_failure;
47bbbb30 ED	432	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) \|\|
47bbbb30 ED	433	nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
1b34ec43	434	goto nla_put_failure;
02f23f09	435
1e90474c	436	return nla_nest_end(skb, opts);
1da177e4	437
1e90474c	438	nla_put_failure:
bc3ed28c TG	439	nla_nest_cancel(skb, opts);
bc3ed28c TG	440	return -EMSGSIZE;
1da177e4 LT	441	}
	442
	443	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	444	{
81da99ed	445	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	446	struct nlattr *opts = NULL;
1da177e4	447
1e90474c PM	448	opts = nla_nest_start(skb, TCA_OPTIONS);
	449	if (opts == NULL)
	450	goto nla_put_failure;
1b34ec43 DM	451	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
1b34ec43 DM	452	goto nla_put_failure;
02f23f09	453
1b34ec43 DM	454	if (p->default_index != NO_DEFAULT_INDEX &&
	455	nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
	456	goto nla_put_failure;
1da177e4	457
1b34ec43 DM	458	if (p->set_tc_index &&
	459	nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
	460	goto nla_put_failure;
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,
1da177e4 LT	471	.leaf = dsmark_leaf,
143976ce	472	.find = dsmark_find,
1da177e4 LT	473	.change = dsmark_change,
	474	.delete = dsmark_delete,
	475	.walk = dsmark_walk,
6529eaba	476	.tcf_block = dsmark_tcf_block,
1da177e4	477	.bind_tcf = dsmark_bind_filter,
143976ce	478	.unbind_tcf = dsmark_unbind_filter,
1da177e4 LT	479	.dump = dsmark_dump_class,
	480	};
	481
20fea08b	482	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	483	.next = NULL,
	484	.cl_ops = &dsmark_class_ops,
	485	.id = "dsmark",
	486	.priv_size = sizeof(struct dsmark_qdisc_data),
	487	.enqueue = dsmark_enqueue,
	488	.dequeue = dsmark_dequeue,
8e3af978	489	.peek = dsmark_peek,
1da177e4 LT	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");