[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/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	*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("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
		sch, p, new, old);

	if (new == NULL) {
		new = qdisc_create_dflt(qdisc_dev(sch), 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("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
		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("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
		"arg 0x%lx\n", 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("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", 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 **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("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);

	if (p->set_tc_index) {
		switch (skb->protocol) {
		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;
		int result = tc_classify(skb, p->filter_list, &res);

		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))
			sch->qstats.drops++;
		return err;
	}

	sch->bstats.bytes += qdisc_pkt_len(skb);
	sch->bstats.packets++;
	sch->q.qlen++;

	return NET_XMIT_SUCCESS;

drop:
	kfree_skb(skb);
	sch->qstats.drops++;
	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("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);

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

	sch->q.qlen--;

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

	switch (skb->protocol) {
	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])
			printk(KERN_WARNING
			       "dsmark_dequeue: unsupported protocol %d\n",
			       ntohs(skb->protocol));
		break;
	}

	return skb;
}

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

	pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", 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("dsmark_reset(sch %p,[qdisc %p])\n", 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("dsmark_init(sch %p,[qdisc %p],opt %p)\n", 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(qdisc_dev(sch), sch->dev_queue,
				 &pfifo_qdisc_ops, sch->handle);
	if (p->q == NULL)
		p->q = &noop_qdisc;

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

	err = 0;
errout:
	return err;
}

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

	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", 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("dsmark_destroy(sch %p,[qdisc %p])\n", 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("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", 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;
	NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
	NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);

	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;
	NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);

	if (p->default_index != NO_DEFAULT_INDEX)
		NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);

	if (p->set_tc_index)
		NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);

	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>
	7	#include <linux/init.h>
	8	#include <linux/types.h>
	9	#include <linux/string.h>
	10	#include <linux/errno.h>
	11	#include <linux/skbuff.h>
1da177e4	12	#include <linux/rtnetlink.h>
5b0ac72b	13	#include <linux/bitops.h>
1da177e4 LT	14	#include <net/pkt_sched.h>
	15	#include <net/dsfield.h>
	16	#include <net/inet_ecn.h>
	17	#include <asm/byteorder.h>
	18
1da177e4 LT	19	/*
	20	* classid class marking
	21	* ------- ----- -------
	22	* n/a 0 n/a
	23	* x:0 1 use entry [0]
	24	* ... ... ...
	25	* x:y y>0 y+1 use entry [y]
	26	* ... ... ...
	27	* x:indices-1 indices use entry [indices-1]
	28	* ... ... ...
	29	* x:y y+1 use entry [y & (indices-1)]
	30	* ... ... ...
	31	* 0xffff 0x10000 use entry [indices-1]
	32	*/
	33
	34
	35	#define NO_DEFAULT_INDEX (1 << 16)
	36
	37	struct dsmark_qdisc_data {
	38	struct Qdisc *q;
	39	struct tcf_proto *filter_list;
af0d1141 TG	40	u8 mask; / "owns" the array */
	41	u8 *value;
	42	u16 indices;
	43	u32 default_index; /* index range is 0...0xffff */
1da177e4 LT	44	int set_tc_index;
	45	};
	46
758cc43c TG	47	static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
	48	{
	49	return (index <= p->indices && index > 0);
	50	}
1da177e4 LT	51
	52	/* ------------------------- Class/flow operations ------------------------- */
	53
af0d1141 TG	54	static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
af0d1141 TG	55	struct Qdisc new, struct Qdisc *old)
1da177e4	56	{
81da99ed	57	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	58
81da99ed	59	pr_debug("dsmark_graft(sch %p,[qdisc %p],new %p,old %p)\n",
af0d1141	60	sch, p, new, old);
486b53e5 TG	61
486b53e5 TG	62	if (new == NULL) {
5ce2d488	63	new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd	64	&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	{
81da99ed SH	88	pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
81da99ed SH	89	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
81da99ed	121	pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
758cc43c TG	122	"arg 0x%lx\n", sch, p, classid, parent, *arg);
	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])
1e90474c	140	p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99	141
1e90474c	142	if (tb[TCA_DSMARK_MASK])
758cc43c TG	143	p->mask[*arg-1] = mask;
	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
1da177e4 LT	158	p->mask[arg-1] = 0xff;
1da177e4 LT	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
81da99ed	169	pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
af0d1141	170
1da177e4 LT	171	if (walker->stop)
1da177e4 LT	172	return;
af0d1141	173
1da177e4 LT	174	for (i = 0; i < p->indices; i++) {
1da177e4 LT	175	if (p->mask[i] == 0xff && !p->value[i])
0451eb07	176	goto ignore;
1da177e4 LT	177	if (walker->count >= walker->skip) {
	178	if (walker->fn(sch, i+1, walker) < 0) {
	179	walker->stop = 1;
	180	break;
	181	}
	182	}
10297b99	183	ignore:
0451eb07	184	walker->count++;
10297b99	185	}
1da177e4 LT	186	}
1da177e4 LT	187
9d127fbd SH	188	static inline struct tcf_proto *dsmark_find_tcf(struct Qdisc sch,
9d127fbd SH	189	unsigned long cl)
1da177e4	190	{
81da99ed SH	191	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	192	return &p->filter_list;
1da177e4 LT	193	}
1da177e4 LT	194
1da177e4 LT	195	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	196
9d127fbd	197	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	198	{
81da99ed	199	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	200	int err;
af0d1141 TG	201
81da99ed	202	pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
1da177e4	203
1da177e4	204	if (p->set_tc_index) {
1da177e4	205	switch (skb->protocol) {
60678040	206	case htons(ETH_P_IP):
9d127fbd SH	207	if (skb_cow_head(skb, sizeof(struct iphdr)))
9d127fbd SH	208	goto drop;
4c30719f	209
9d127fbd SH	210	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	211	& ~INET_ECN_MASK;
	212	break;
4c30719f	213
60678040	214	case htons(ETH_P_IPV6):
9d127fbd SH	215	if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
9d127fbd SH	216	goto drop;
4c30719f	217
9d127fbd SH	218	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	219	& ~INET_ECN_MASK;
	220	break;
	221	default:
	222	skb->tc_index = 0;
	223	break;
3ff50b79	224	}
1da177e4	225	}
af0d1141 TG	226
af0d1141 TG	227	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	228	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	229	else {
	230	struct tcf_result res;
	231	int result = tc_classify(skb, p->filter_list, &res);
	232
81da99ed	233	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	234
1da177e4	235	switch (result) {
f6853e2d PM	236	#ifdef CONFIG_NET_CLS_ACT
	237	case TC_ACT_QUEUED:
	238	case TC_ACT_STOLEN:
	239	kfree_skb(skb);
378a2f09	240	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	241
f6853e2d	242	case TC_ACT_SHOT:
4c30719f	243	goto drop;
1da177e4	244	#endif
c3bc7cff	245	case TC_ACT_OK:
f6853e2d PM	246	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	247	break;
4c30719f	248
f6853e2d PM	249	default:
	250	if (p->default_index != NO_DEFAULT_INDEX)
	251	skb->tc_index = p->default_index;
	252	break;
3ff50b79	253	}
1da177e4	254	}
1da177e4	255
5f86173b	256	err = qdisc_enqueue(skb, p->q);
af0d1141	257	if (err != NET_XMIT_SUCCESS) {
378a2f09 JP	258	if (net_xmit_drop_count(err))
378a2f09 JP	259	sch->qstats.drops++;
af0d1141	260	return err;
1da177e4	261	}
af0d1141	262
0abf77e5	263	sch->bstats.bytes += qdisc_pkt_len(skb);
1da177e4 LT	264	sch->bstats.packets++;
1da177e4 LT	265	sch->q.qlen++;
1da177e4	266
af0d1141	267	return NET_XMIT_SUCCESS;
4c30719f SH	268
	269	drop:
	270	kfree_skb(skb);
	271	sch->qstats.drops++;
c27f339a	272	return NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
af0d1141	273	}
1da177e4 LT	274
	275	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	276	{
81da99ed	277	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	278	struct sk_buff *skb;
af0d1141 TG	279	u32 index;
af0d1141 TG	280
81da99ed	281	pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
1da177e4	282
1da177e4	283	skb = p->q->ops->dequeue(p->q);
af0d1141	284	if (skb == NULL)
1da177e4	285	return NULL;
af0d1141	286
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
1da177e4	292	switch (skb->protocol) {
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])
	308	printk(KERN_WARNING
	309	"dsmark_dequeue: unsupported protocol %d\n",
	310	ntohs(skb->protocol));
	311	break;
3ff50b79	312	}
af0d1141	313
1da177e4 LT	314	return skb;
	315	}
	316
8e3af978 JP	317	static struct sk_buff dsmark_peek(struct Qdisc sch)
	318	{
	319	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	320
	321	pr_debug("dsmark_peek(sch %p,[qdisc %p])\n", sch, p);
	322
	323	return p->q->ops->peek(p->q);
	324	}
	325
1da177e4 LT	326	static unsigned int dsmark_drop(struct Qdisc *sch)
1da177e4 LT	327	{
81da99ed	328	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	329	unsigned int len;
10297b99	330
81da99ed	331	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
af0d1141 TG	332
af0d1141 TG	333	if (p->q->ops->drop == NULL)
1da177e4	334	return 0;
af0d1141 TG	335
	336	len = p->q->ops->drop(p->q);
	337	if (len)
	338	sch->q.qlen--;
	339
1da177e4 LT	340	return len;
	341	}
	342
1e90474c	343	static int dsmark_init(struct Qdisc sch, struct nlattr opt)
1da177e4	344	{
81da99ed	345	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	346	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c TG	347	int err = -EINVAL;
	348	u32 default_index = NO_DEFAULT_INDEX;
	349	u16 indices;
	350	u8 *mask;
	351
81da99ed	352	pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
758cc43c	353
cee63723 PM	354	if (!opt)
	355	goto errout;
	356
27a3421e	357	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy);
cee63723	358	if (err < 0)
758cc43c TG	359	goto errout;
758cc43c TG	360
cee63723	361	err = -EINVAL;
1e90474c	362	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	363
5b0ac72b DM	364	if (hweight32(indices) != 1)
758cc43c TG	365	goto errout;
758cc43c TG	366
27a3421e	367	if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c	368	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c TG	369
	370	mask = kmalloc(indices * 2, GFP_KERNEL);
	371	if (mask == NULL) {
	372	err = -ENOMEM;
	373	goto errout;
1da177e4	374	}
758cc43c TG	375
	376	p->mask = mask;
	377	memset(p->mask, 0xff, indices);
	378
	379	p->value = p->mask + indices;
	380	memset(p->value, 0, indices);
	381
	382	p->indices = indices;
	383	p->default_index = default_index;
1e90474c	384	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	385
5ce2d488	386	p->q = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd	387	&pfifo_qdisc_ops, sch->handle);
758cc43c	388	if (p->q == NULL)
1da177e4	389	p->q = &noop_qdisc;
758cc43c	390
81da99ed	391	pr_debug("dsmark_init: qdisc %p\n", 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
81da99ed	402	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", 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
81da99ed	411	pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
af0d1141	412
ff31ab56	413	tcf_destroy_chain(&p->filter_list);
1da177e4 LT	414	qdisc_destroy(p->q);
	415	kfree(p->mask);
	416	}
	417
1da177e4	418	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	419	struct sk_buff skb, struct tcmsg tcm)
1da177e4	420	{
81da99ed	421	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	422	struct nlattr *opts = NULL;
1da177e4	423
81da99ed	424	pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
02f23f09 TG	425
02f23f09 TG	426	if (!dsmark_valid_index(p, cl))
1da177e4	427	return -EINVAL;
02f23f09 TG	428
02f23f09 TG	429	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
cdc7f8e3	430	tcm->tcm_info = p->q->handle;
02f23f09	431
1e90474c PM	432	opts = nla_nest_start(skb, TCA_OPTIONS);
	433	if (opts == NULL)
	434	goto nla_put_failure;
	435	NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
	436	NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
02f23f09	437
1e90474c	438	return nla_nest_end(skb, opts);
1da177e4	439
1e90474c	440	nla_put_failure:
bc3ed28c TG	441	nla_nest_cancel(skb, opts);
bc3ed28c TG	442	return -EMSGSIZE;
1da177e4 LT	443	}
	444
	445	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	446	{
81da99ed	447	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	448	struct nlattr *opts = NULL;
1da177e4	449
1e90474c PM	450	opts = nla_nest_start(skb, TCA_OPTIONS);
	451	if (opts == NULL)
	452	goto nla_put_failure;
	453	NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
02f23f09 TG	454
02f23f09 TG	455	if (p->default_index != NO_DEFAULT_INDEX)
1e90474c	456	NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
1da177e4	457
1da177e4	458	if (p->set_tc_index)
1e90474c	459	NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
02f23f09	460
1e90474c	461	return nla_nest_end(skb, opts);
1da177e4	462
1e90474c	463	nla_put_failure:
bc3ed28c TG	464	nla_nest_cancel(skb, opts);
bc3ed28c TG	465	return -EMSGSIZE;
1da177e4 LT	466	}
1da177e4 LT	467
20fea08b	468	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	469	.graft = dsmark_graft,
	470	.leaf = dsmark_leaf,
	471	.get = dsmark_get,
	472	.put = dsmark_put,
	473	.change = dsmark_change,
	474	.delete = dsmark_delete,
	475	.walk = dsmark_walk,
	476	.tcf_chain = dsmark_find_tcf,
	477	.bind_tcf = dsmark_bind_filter,
	478	.unbind_tcf = dsmark_put,
	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	.drop = dsmark_drop,
	491	.init = dsmark_init,
	492	.reset = dsmark_reset,
	493	.destroy = dsmark_destroy,
	494	.change = NULL,
	495	.dump = dsmark_dump,
	496	.owner = THIS_MODULE,
	497	};
	498
	499	static int __init dsmark_module_init(void)
	500	{
	501	return register_qdisc(&dsmark_qdisc_ops);
	502	}
af0d1141	503
10297b99	504	static void __exit dsmark_module_exit(void)
1da177e4 LT	505	{
	506	unregister_qdisc(&dsmark_qdisc_ops);
	507	}
af0d1141	508
1da177e4 LT	509	module_init(dsmark_module_init)
1da177e4 LT	510	module_exit(dsmark_module_exit)
af0d1141	511
1da177e4	512	MODULE_LICENSE("GPL");