[mirror_ubuntu-zesty-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/dsfield.h>
#include <net/inet_ecn.h>
#include <asm/byteorder.h>

/*
 * classid	class		marking
 * -------	-----		-------
 *   n/a	  0		n/a
 *   x:0	  1		use entry [0]
 *   ...	 ...		...
 *   x:y y>0	 y+1		use entry [y]
 *   ...	 ...		...
 * x:indices-1	indices		use entry [indices-1]
 *   ...	 ...		...
 *   x:y	 y+1		use entry [y & (indices-1)]
 *   ...	 ...		...
 * 0xffff	0x10000		use entry [indices-1]
 */


#define NO_DEFAULT_INDEX	(1 << 16)

struct dsmark_qdisc_data {
	struct Qdisc		*q;
	struct tcf_proto __rcu	*filter_list;
	u8			*mask;	/* "owns" the array */
	u8			*value;
	u16			indices;
	u32			default_index;	/* index range is 0...0xffff */
	int			set_tc_index;
};

static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
{
	return index <= p->indices && index > 0;
}

/* ------------------------- Class/flow operations ------------------------- */

static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
			struct Qdisc *new, struct Qdisc **old)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);

	pr_debug("%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;
	u8 mask = 0;

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

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

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

	err = 0;

errout:
	return err;
}

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

	if (!dsmark_valid_index(p, arg))
		return -EINVAL;

	p->mask[arg - 1] = 0xff;
	p->value[arg - 1] = 0;

	return 0;
}

static void dsmark_walk(struct Qdisc *sch, struct qdisc_walker *walker)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	int i;

	pr_debug("%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->mask[i] == 0xff && !p->value[i])
			goto ignore;
		if (walker->count >= walker->skip) {
			if (walker->fn(sch, i + 1, walker) < 0) {
				walker->stop = 1;
				break;
			}
		}
ignore:
		walker->count++;
	}
}

static inline struct tcf_proto __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->mask[index],
				    p->value[index]);
			break;
	case htons(ETH_P_IPV6):
		ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
				    p->value[index]);
			break;
	default:
		/*
		 * Only complain if a change was actually attempted.
		 * This way, we can send non-IP traffic through dsmark
		 * and don't need yet another qdisc as a bypass.
		 */
		if (p->mask[index] != 0xff || p->value[index])
			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;
	u8 *mask;

	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]);

	mask = kmalloc(indices * 2, GFP_KERNEL);
	if (mask == NULL) {
		err = -ENOMEM;
		goto errout;
	}

	p->mask = mask;
	memset(p->mask, 0xff, indices);

	p->value = p->mask + indices;
	memset(p->value, 0, indices);

	p->indices = indices;
	p->default_index = default_index;
	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);

	p->q = qdisc_create_dflt(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);
	kfree(p->mask);
}

static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
			     struct sk_buff *skb, struct tcmsg *tcm)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	struct nlattr *opts = NULL;

	pr_debug("%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->mask[cl - 1]) ||
	    nla_put_u8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]))
		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
	38	struct dsmark_qdisc_data {
	39	struct Qdisc *q;
25d8c0d5	40	struct tcf_proto __rcu *filter_list;
af0d1141 TG	41	u8 mask; / "owns" the array */
	42	u8 *value;
	43	u16 indices;
	44	u32 default_index; /* index range is 0...0xffff */
1da177e4 LT	45	int set_tc_index;
	46	};
	47
758cc43c TG	48	static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
758cc43c TG	49	{
17569fae	50	return index <= p->indices && index > 0;
758cc43c	51	}
1da177e4 LT	52
	53	/* ------------------------- Class/flow operations ------------------------- */
	54
af0d1141 TG	55	static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
af0d1141 TG	56	struct Qdisc new, struct Qdisc *old)
1da177e4	57	{
81da99ed	58	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	59
c76f2a2c YY	60	pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
c76f2a2c YY	61	__func__, sch, p, new, old);
486b53e5 TG	62
486b53e5 TG	63	if (new == NULL) {
3511c913	64	new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
9f9afec4	65	sch->handle);
486b53e5 TG	66	if (new == NULL)
	67	new = &noop_qdisc;
	68	}
	69
1da177e4	70	sch_tree_lock(sch);
b94c8afc PM	71	*old = p->q;
b94c8afc PM	72	p->q = new;
5e50da01	73	qdisc_tree_decrease_qlen(old, (old)->q.qlen);
af0d1141	74	qdisc_reset(*old);
af0d1141 TG	75	sch_tree_unlock(sch);
af0d1141 TG	76
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 TG	118	int err = -EINVAL;
758cc43c TG	119	u8 mask = 0;
1da177e4	120
c76f2a2c YY	121	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
c76f2a2c YY	122	__func__, sch, p, classid, parent, *arg);
758cc43c TG	123
	124	if (!dsmark_valid_index(p, *arg)) {
	125	err = -ENOENT;
1e90474c	126	goto errout;
1da177e4	127	}
1da177e4	128
cee63723	129	if (!opt)
1e90474c	130	goto errout;
758cc43c	131
27a3421e	132	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	133	if (err < 0)
27a3421e	134	goto errout;
cee63723	135
27a3421e	136	if (tb[TCA_DSMARK_MASK])
1e90474c	137	mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
27a3421e PM	138
27a3421e PM	139	if (tb[TCA_DSMARK_VALUE])
cc7ec456	140	p->value[*arg - 1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99	141
1e90474c	142	if (tb[TCA_DSMARK_MASK])
cc7ec456	143	p->mask[*arg - 1] = mask;
758cc43c TG	144
	145	err = 0;
	146
1e90474c	147	errout:
758cc43c TG	148	return err;
758cc43c TG	149	}
1da177e4	150
af0d1141	151	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4	152	{
81da99ed	153	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	154
af0d1141	155	if (!dsmark_valid_index(p, arg))
1da177e4	156	return -EINVAL;
10297b99	157
cc7ec456 ED	158	p->mask[arg - 1] = 0xff;
cc7ec456 ED	159	p->value[arg - 1] = 0;
af0d1141	160
1da177e4 LT	161	return 0;
	162	}
	163
9d127fbd	164	static void dsmark_walk(struct Qdisc sch, struct qdisc_walker walker)
1da177e4	165	{
81da99ed	166	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 LT	167	int i;
1da177e4 LT	168
c76f2a2c YY	169	pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
c76f2a2c YY	170	__func__, sch, p, walker);
af0d1141	171
1da177e4 LT	172	if (walker->stop)
1da177e4 LT	173	return;
af0d1141	174
1da177e4 LT	175	for (i = 0; i < p->indices; i++) {
1da177e4 LT	176	if (p->mask[i] == 0xff && !p->value[i])
0451eb07	177	goto ignore;
1da177e4	178	if (walker->count >= walker->skip) {
cc7ec456	179	if (walker->fn(sch, i + 1, walker) < 0) {
1da177e4 LT	180	walker->stop = 1;
	181	break;
	182	}
	183	}
10297b99	184	ignore:
0451eb07	185	walker->count++;
10297b99	186	}
1da177e4 LT	187	}
1da177e4 LT	188
25d8c0d5 JF	189	static inline struct tcf_proto __rcu *dsmark_find_tcf(struct Qdisc sch,
25d8c0d5 JF	190	unsigned long cl)
1da177e4	191	{
81da99ed SH	192	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	193	return &p->filter_list;
1da177e4 LT	194	}
1da177e4 LT	195
1da177e4 LT	196	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	197
9d127fbd	198	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	199	{
81da99ed	200	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	201	int err;
af0d1141 TG	202
c76f2a2c	203	pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
1da177e4	204
1da177e4	205	if (p->set_tc_index) {
d8b9605d	206	switch (tc_skb_protocol(skb)) {
60678040	207	case htons(ETH_P_IP):
9d127fbd SH	208	if (skb_cow_head(skb, sizeof(struct iphdr)))
9d127fbd SH	209	goto drop;
4c30719f	210
9d127fbd SH	211	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	212	& ~INET_ECN_MASK;
	213	break;
4c30719f	214
60678040	215	case htons(ETH_P_IPV6):
9d127fbd SH	216	if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
9d127fbd SH	217	goto drop;
4c30719f	218
9d127fbd SH	219	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	220	& ~INET_ECN_MASK;
	221	break;
	222	default:
	223	skb->tc_index = 0;
	224	break;
3ff50b79	225	}
1da177e4	226	}
af0d1141 TG	227
af0d1141 TG	228	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	229	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	230	else {
af0d1141 TG	231	struct tcf_result res;
25d8c0d5	232	struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
3b3ae880	233	int result = tc_classify(skb, fl, &res, false);
af0d1141	234
81da99ed	235	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	236
1da177e4	237	switch (result) {
f6853e2d PM	238	#ifdef CONFIG_NET_CLS_ACT
	239	case TC_ACT_QUEUED:
	240	case TC_ACT_STOLEN:
	241	kfree_skb(skb);
378a2f09	242	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	243
f6853e2d	244	case TC_ACT_SHOT:
4c30719f	245	goto drop;
1da177e4	246	#endif
c3bc7cff	247	case TC_ACT_OK:
f6853e2d PM	248	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	249	break;
4c30719f	250
f6853e2d PM	251	default:
	252	if (p->default_index != NO_DEFAULT_INDEX)
	253	skb->tc_index = p->default_index;
	254	break;
3ff50b79	255	}
1da177e4	256	}
1da177e4	257
5f86173b	258	err = qdisc_enqueue(skb, p->q);
af0d1141	259	if (err != NET_XMIT_SUCCESS) {
378a2f09	260	if (net_xmit_drop_count(err))
25331d6c	261	qdisc_qstats_drop(sch);
af0d1141	262	return err;
1da177e4	263	}
af0d1141	264
1da177e4	265	sch->q.qlen++;
1da177e4	266
af0d1141	267	return NET_XMIT_SUCCESS;
4c30719f SH	268
4c30719f SH	269	drop:
17045755	270	qdisc_drop(skb, sch);
c27f339a	271	return NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
af0d1141	272	}
1da177e4 LT	273
	274	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	275	{
81da99ed	276	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	277	struct sk_buff *skb;
af0d1141 TG	278	u32 index;
af0d1141 TG	279
c76f2a2c	280	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	281
1da177e4	282	skb = p->q->ops->dequeue(p->q);
af0d1141	283	if (skb == NULL)
1da177e4	284	return NULL;
af0d1141	285
9190b3b3	286	qdisc_bstats_update(sch, skb);
1da177e4	287	sch->q.qlen--;
af0d1141 TG	288
af0d1141 TG	289	index = skb->tc_index & (p->indices - 1);
81da99ed	290	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	291
d8b9605d	292	switch (tc_skb_protocol(skb)) {
60678040	293	case htons(ETH_P_IP):
9d127fbd SH	294	ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
9d127fbd SH	295	p->value[index]);
1da177e4	296	break;
60678040	297	case htons(ETH_P_IPV6):
9d127fbd SH	298	ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
9d127fbd SH	299	p->value[index]);
1da177e4	300	break;
9d127fbd SH	301	default:
	302	/*
	303	* Only complain if a change was actually attempted.
	304	* This way, we can send non-IP traffic through dsmark
	305	* and don't need yet another qdisc as a bypass.
	306	*/
	307	if (p->mask[index] != 0xff \|\| p->value[index])
c76f2a2c	308	pr_warn("%s: unsupported protocol %d\n",
d8b9605d	309	__func__, ntohs(tc_skb_protocol(skb)));
9d127fbd	310	break;
3ff50b79	311	}
af0d1141	312
1da177e4 LT	313	return skb;
	314	}
	315
8e3af978 JP	316	static struct sk_buff dsmark_peek(struct Qdisc sch)
	317	{
	318	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	319
c76f2a2c	320	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
8e3af978 JP	321
	322	return p->q->ops->peek(p->q);
	323	}
	324
1da177e4 LT	325	static unsigned int dsmark_drop(struct Qdisc *sch)
1da177e4 LT	326	{
81da99ed	327	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	328	unsigned int len;
10297b99	329
c76f2a2c	330	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141 TG	331
af0d1141 TG	332	if (p->q->ops->drop == NULL)
1da177e4	333	return 0;
af0d1141 TG	334
	335	len = p->q->ops->drop(p->q);
	336	if (len)
	337	sch->q.qlen--;
	338
1da177e4 LT	339	return len;
	340	}
	341
1e90474c	342	static int dsmark_init(struct Qdisc sch, struct nlattr opt)
1da177e4	343	{
81da99ed	344	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	345	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c TG	346	int err = -EINVAL;
	347	u32 default_index = NO_DEFAULT_INDEX;
	348	u16 indices;
	349	u8 *mask;
	350
c76f2a2c	351	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
758cc43c	352
cee63723 PM	353	if (!opt)
	354	goto errout;
	355
27a3421e	356	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	357	if (err < 0)
758cc43c TG	358	goto errout;
758cc43c TG	359
cee63723	360	err = -EINVAL;
1e90474c	361	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	362
5b0ac72b DM	363	if (hweight32(indices) != 1)
758cc43c TG	364	goto errout;
758cc43c TG	365
27a3421e	366	if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c	367	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c TG	368
	369	mask = kmalloc(indices * 2, GFP_KERNEL);
	370	if (mask == NULL) {
	371	err = -ENOMEM;
	372	goto errout;
1da177e4	373	}
758cc43c TG	374
	375	p->mask = mask;
	376	memset(p->mask, 0xff, indices);
	377
	378	p->value = p->mask + indices;
	379	memset(p->value, 0, indices);
	380
	381	p->indices = indices;
	382	p->default_index = default_index;
1e90474c	383	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	384
3511c913	385	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle);
758cc43c	386	if (p->q == NULL)
1da177e4	387	p->q = &noop_qdisc;
758cc43c	388
c76f2a2c	389	pr_debug("%s: qdisc %p\n", __func__, p->q);
758cc43c TG	390
	391	err = 0;
	392	errout:
758cc43c	393	return err;
1da177e4 LT	394	}
1da177e4 LT	395
1da177e4 LT	396	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	397	{
81da99ed	398	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	399
c76f2a2c	400	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4 LT	401	qdisc_reset(p->q);
	402	sch->q.qlen = 0;
	403	}
	404
1da177e4 LT	405	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	406	{
81da99ed	407	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	408
c76f2a2c	409	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141	410
ff31ab56	411	tcf_destroy_chain(&p->filter_list);
1da177e4 LT	412	qdisc_destroy(p->q);
	413	kfree(p->mask);
	414	}
	415
1da177e4	416	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	417	struct sk_buff skb, struct tcmsg tcm)
1da177e4	418	{
81da99ed	419	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	420	struct nlattr *opts = NULL;
1da177e4	421
c76f2a2c	422	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
02f23f09 TG	423
02f23f09 TG	424	if (!dsmark_valid_index(p, cl))
1da177e4	425	return -EINVAL;
02f23f09	426
cc7ec456	427	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
cdc7f8e3	428	tcm->tcm_info = p->q->handle;
02f23f09	429
1e90474c PM	430	opts = nla_nest_start(skb, TCA_OPTIONS);
	431	if (opts == NULL)
	432	goto nla_put_failure;
1b34ec43 DM	433	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mask[cl - 1]) \|\|
	434	nla_put_u8(skb, TCA_DSMARK_VALUE, p->value[cl - 1]))
	435	goto nla_put_failure;
02f23f09	436
1e90474c	437	return nla_nest_end(skb, opts);
1da177e4	438
1e90474c	439	nla_put_failure:
bc3ed28c TG	440	nla_nest_cancel(skb, opts);
bc3ed28c TG	441	return -EMSGSIZE;
1da177e4 LT	442	}
	443
	444	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	445	{
81da99ed	446	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	447	struct nlattr *opts = NULL;
1da177e4	448
1e90474c PM	449	opts = nla_nest_start(skb, TCA_OPTIONS);
	450	if (opts == NULL)
	451	goto nla_put_failure;
1b34ec43 DM	452	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
1b34ec43 DM	453	goto nla_put_failure;
02f23f09	454
1b34ec43 DM	455	if (p->default_index != NO_DEFAULT_INDEX &&
	456	nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
	457	goto nla_put_failure;
1da177e4	458
1b34ec43 DM	459	if (p->set_tc_index &&
	460	nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
	461	goto nla_put_failure;
02f23f09	462
1e90474c	463	return nla_nest_end(skb, opts);
1da177e4	464
1e90474c	465	nla_put_failure:
bc3ed28c TG	466	nla_nest_cancel(skb, opts);
bc3ed28c TG	467	return -EMSGSIZE;
1da177e4 LT	468	}
1da177e4 LT	469
20fea08b	470	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	471	.graft = dsmark_graft,
	472	.leaf = dsmark_leaf,
	473	.get = dsmark_get,
	474	.put = dsmark_put,
	475	.change = dsmark_change,
	476	.delete = dsmark_delete,
	477	.walk = dsmark_walk,
	478	.tcf_chain = dsmark_find_tcf,
	479	.bind_tcf = dsmark_bind_filter,
	480	.unbind_tcf = dsmark_put,
	481	.dump = dsmark_dump_class,
	482	};
	483
20fea08b	484	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	485	.next = NULL,
	486	.cl_ops = &dsmark_class_ops,
	487	.id = "dsmark",
	488	.priv_size = sizeof(struct dsmark_qdisc_data),
	489	.enqueue = dsmark_enqueue,
	490	.dequeue = dsmark_dequeue,
8e3af978	491	.peek = dsmark_peek,
1da177e4 LT	492	.drop = dsmark_drop,
	493	.init = dsmark_init,
	494	.reset = dsmark_reset,
	495	.destroy = dsmark_destroy,
	496	.change = NULL,
	497	.dump = dsmark_dump,
	498	.owner = THIS_MODULE,
	499	};
	500
	501	static int __init dsmark_module_init(void)
	502	{
	503	return register_qdisc(&dsmark_qdisc_ops);
	504	}
af0d1141	505
10297b99	506	static void __exit dsmark_module_exit(void)
1da177e4 LT	507	{
	508	unregister_qdisc(&dsmark_qdisc_ops);
	509	}
af0d1141	510
1da177e4 LT	511	module_init(dsmark_module_init)
1da177e4 LT	512	module_exit(dsmark_module_exit)
af0d1141	513
1da177e4	514	MODULE_LICENSE("GPL");