[deliverable/linux.git] / net / sched / sch_dsmark.c

/* net/sched/sch_dsmark.c - Differentiated Services field marker */

/* Written 1998-2000 by Werner Almesberger, EPFL ICA */


#include <linux/module.h>
#include <linux/init.h>
#include <linux/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(sch->dev, &pfifo_qdisc_ops,
					sch->handle);
		if (new == NULL)
			new = &noop_qdisc;
	}

	sch_tree_lock(sch);
	*old = xchg(&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 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 || nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL))
		goto errout;

	if (tb[TCA_DSMARK_MASK]) {
		if (nla_len(tb[TCA_DSMARK_MASK]) < sizeof(u8))
			goto errout;
		mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
	}
	if (tb[TCA_DSMARK_VALUE]) {
		if (nla_len(tb[TCA_DSMARK_VALUE]) < sizeof(u8))
			goto errout;
		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 __constant_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 __constant_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;

		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 = p->q->enqueue(skb, p->q);
	if (err != NET_XMIT_SUCCESS) {
		sch->qstats.drops++;
		return err;
	}

	sch->bstats.bytes += skb->len;
	sch->bstats.packets++;
	sch->q.qlen++;

	return NET_XMIT_SUCCESS;

drop:
	kfree_skb(skb);
	sch->qstats.drops++;
	return 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 __constant_htons(ETH_P_IP):
		ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
				    p->value[index]);
			break;
	case __constant_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 int dsmark_requeue(struct sk_buff *skb, struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	int err;

	pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);

	err = p->q->ops->requeue(skb, p->q);
	if (err != NET_XMIT_SUCCESS) {
		sch->qstats.drops++;
		return err;
	}

	sch->q.qlen++;
	sch->qstats.requeues++;

	return NET_XMIT_SUCCESS;
}

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 || nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL) < 0)
		goto errout;

	if (nla_len(tb[TCA_DSMARK_INDICES]) < sizeof(u16))
		goto errout;
	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);

	if (hweight32(indices) != 1)
		goto errout;

	if (tb[TCA_DSMARK_DEFAULT_INDEX]) {
		if (nla_len(tb[TCA_DSMARK_DEFAULT_INDEX]) < sizeof(u16))
			goto errout;
		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, &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:
	return nla_nest_cancel(skb, opts);
}

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:
	return nla_nest_cancel(skb, opts);
}

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,
	.requeue	=	dsmark_requeue,
	.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) {
9f9afec4 PM	63	new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
9f9afec4 PM	64	sch->handle);
486b53e5 TG	65	if (new == NULL)
	66	new = &noop_qdisc;
	67	}
	68
1da177e4	69	sch_tree_lock(sch);
af0d1141	70	*old = xchg(&p->q, new);
5e50da01	71	qdisc_tree_decrease_qlen(old, (old)->q.qlen);
af0d1141	72	qdisc_reset(*old);
af0d1141 TG	73	sch_tree_unlock(sch);
af0d1141 TG	74
10297b99	75	return 0;
1da177e4 LT	76	}
1da177e4 LT	77
1da177e4 LT	78	static struct Qdisc dsmark_leaf(struct Qdisc sch, unsigned long arg)
1da177e4 LT	79	{
81da99ed SH	80	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	81	return p->q;
1da177e4 LT	82	}
1da177e4 LT	83
af0d1141	84	static unsigned long dsmark_get(struct Qdisc *sch, u32 classid)
1da177e4	85	{
81da99ed SH	86	pr_debug("dsmark_get(sch %p,[qdisc %p],classid %x)\n",
81da99ed SH	87	sch, qdisc_priv(sch), classid);
1da177e4	88
af0d1141	89	return TC_H_MIN(classid) + 1;
1da177e4 LT	90	}
1da177e4 LT	91
1da177e4	92	static unsigned long dsmark_bind_filter(struct Qdisc *sch,
af0d1141	93	unsigned long parent, u32 classid)
1da177e4	94	{
af0d1141	95	return dsmark_get(sch, classid);
1da177e4 LT	96	}
1da177e4 LT	97
1da177e4 LT	98	static void dsmark_put(struct Qdisc *sch, unsigned long cl)
	99	{
	100	}
	101
1da177e4	102	static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
1e90474c	103	struct nlattr *tca, unsigned long arg)
1da177e4	104	{
81da99ed	105	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c PM	106	struct nlattr *opt = tca[TCA_OPTIONS];
1e90474c PM	107	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c TG	108	int err = -EINVAL;
758cc43c TG	109	u8 mask = 0;
1da177e4	110
81da99ed	111	pr_debug("dsmark_change(sch %p,[qdisc %p],classid %x,parent %x),"
758cc43c TG	112	"arg 0x%lx\n", sch, p, classid, parent, *arg);
	113
	114	if (!dsmark_valid_index(p, *arg)) {
	115	err = -ENOENT;
1e90474c	116	goto errout;
1da177e4	117	}
1da177e4	118
1e90474c PM	119	if (!opt \|\| nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL))
1e90474c PM	120	goto errout;
758cc43c	121
1e90474c PM	122	if (tb[TCA_DSMARK_MASK]) {
	123	if (nla_len(tb[TCA_DSMARK_MASK]) < sizeof(u8))
	124	goto errout;
	125	mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
	126	}
	127	if (tb[TCA_DSMARK_VALUE]) {
	128	if (nla_len(tb[TCA_DSMARK_VALUE]) < sizeof(u8))
	129	goto errout;
	130	p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
	131	}
10297b99	132
1e90474c	133	if (tb[TCA_DSMARK_MASK])
758cc43c TG	134	p->mask[*arg-1] = mask;
	135
	136	err = 0;
	137
1e90474c	138	errout:
758cc43c TG	139	return err;
758cc43c TG	140	}
1da177e4	141
af0d1141	142	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4	143	{
81da99ed	144	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	145
af0d1141	146	if (!dsmark_valid_index(p, arg))
1da177e4	147	return -EINVAL;
10297b99	148
1da177e4 LT	149	p->mask[arg-1] = 0xff;
1da177e4 LT	150	p->value[arg-1] = 0;
af0d1141	151
1da177e4 LT	152	return 0;
	153	}
	154
9d127fbd	155	static void dsmark_walk(struct Qdisc sch, struct qdisc_walker walker)
1da177e4	156	{
81da99ed	157	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 LT	158	int i;
1da177e4 LT	159
81da99ed	160	pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
af0d1141	161
1da177e4 LT	162	if (walker->stop)
1da177e4 LT	163	return;
af0d1141	164
1da177e4 LT	165	for (i = 0; i < p->indices; i++) {
1da177e4 LT	166	if (p->mask[i] == 0xff && !p->value[i])
0451eb07	167	goto ignore;
1da177e4 LT	168	if (walker->count >= walker->skip) {
	169	if (walker->fn(sch, i+1, walker) < 0) {
	170	walker->stop = 1;
	171	break;
	172	}
	173	}
10297b99	174	ignore:
0451eb07	175	walker->count++;
10297b99	176	}
1da177e4 LT	177	}
1da177e4 LT	178
9d127fbd SH	179	static inline struct tcf_proto *dsmark_find_tcf(struct Qdisc sch,
9d127fbd SH	180	unsigned long cl)
1da177e4	181	{
81da99ed SH	182	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	183	return &p->filter_list;
1da177e4 LT	184	}
1da177e4 LT	185
1da177e4 LT	186	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	187
9d127fbd	188	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	189	{
81da99ed	190	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	191	int err;
af0d1141 TG	192
81da99ed	193	pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
1da177e4	194
1da177e4	195	if (p->set_tc_index) {
1da177e4	196	switch (skb->protocol) {
9d127fbd SH	197	case __constant_htons(ETH_P_IP):
	198	if (skb_cow_head(skb, sizeof(struct iphdr)))
	199	goto drop;
4c30719f	200
9d127fbd SH	201	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	202	& ~INET_ECN_MASK;
	203	break;
4c30719f	204
9d127fbd SH	205	case __constant_htons(ETH_P_IPV6):
	206	if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
	207	goto drop;
4c30719f	208
9d127fbd SH	209	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	210	& ~INET_ECN_MASK;
	211	break;
	212	default:
	213	skb->tc_index = 0;
	214	break;
3ff50b79	215	}
1da177e4	216	}
af0d1141 TG	217
af0d1141 TG	218	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	219	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	220	else {
	221	struct tcf_result res;
	222	int result = tc_classify(skb, p->filter_list, &res);
	223
81da99ed	224	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	225
1da177e4	226	switch (result) {
f6853e2d PM	227	#ifdef CONFIG_NET_CLS_ACT
	228	case TC_ACT_QUEUED:
	229	case TC_ACT_STOLEN:
	230	kfree_skb(skb);
	231	return NET_XMIT_SUCCESS;
4c30719f	232
f6853e2d	233	case TC_ACT_SHOT:
4c30719f	234	goto drop;
1da177e4	235	#endif
c3bc7cff	236	case TC_ACT_OK:
f6853e2d PM	237	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	238	break;
4c30719f	239
f6853e2d PM	240	default:
	241	if (p->default_index != NO_DEFAULT_INDEX)
	242	skb->tc_index = p->default_index;
	243	break;
3ff50b79	244	}
1da177e4	245	}
1da177e4	246
9d127fbd	247	err = p->q->enqueue(skb, p->q);
af0d1141	248	if (err != NET_XMIT_SUCCESS) {
1da177e4	249	sch->qstats.drops++;
af0d1141	250	return err;
1da177e4	251	}
af0d1141	252
1da177e4 LT	253	sch->bstats.bytes += skb->len;
	254	sch->bstats.packets++;
	255	sch->q.qlen++;
1da177e4	256
af0d1141	257	return NET_XMIT_SUCCESS;
4c30719f SH	258
	259	drop:
	260	kfree_skb(skb);
	261	sch->qstats.drops++;
	262	return NET_XMIT_BYPASS;
af0d1141	263	}
1da177e4 LT	264
	265	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	266	{
81da99ed	267	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	268	struct sk_buff *skb;
af0d1141 TG	269	u32 index;
af0d1141 TG	270
81da99ed	271	pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
1da177e4	272
1da177e4	273	skb = p->q->ops->dequeue(p->q);
af0d1141	274	if (skb == NULL)
1da177e4	275	return NULL;
af0d1141	276
1da177e4	277	sch->q.qlen--;
af0d1141 TG	278
af0d1141 TG	279	index = skb->tc_index & (p->indices - 1);
81da99ed	280	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	281
1da177e4	282	switch (skb->protocol) {
9d127fbd SH	283	case __constant_htons(ETH_P_IP):
	284	ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
	285	p->value[index]);
1da177e4	286	break;
9d127fbd SH	287	case __constant_htons(ETH_P_IPV6):
	288	ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
	289	p->value[index]);
1da177e4	290	break;
9d127fbd SH	291	default:
	292	/*
	293	* Only complain if a change was actually attempted.
	294	* This way, we can send non-IP traffic through dsmark
	295	* and don't need yet another qdisc as a bypass.
	296	*/
	297	if (p->mask[index] != 0xff \|\| p->value[index])
	298	printk(KERN_WARNING
	299	"dsmark_dequeue: unsupported protocol %d\n",
	300	ntohs(skb->protocol));
	301	break;
3ff50b79	302	}
af0d1141	303
1da177e4 LT	304	return skb;
	305	}
	306
9d127fbd	307	static int dsmark_requeue(struct sk_buff skb, struct Qdisc sch)
1da177e4	308	{
81da99ed	309	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141	310	int err;
1da177e4	311
81da99ed	312	pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
af0d1141 TG	313
	314	err = p->q->ops->requeue(skb, p->q);
	315	if (err != NET_XMIT_SUCCESS) {
	316	sch->qstats.drops++;
	317	return err;
1da177e4	318	}
1da177e4	319
af0d1141 TG	320	sch->q.qlen++;
	321	sch->qstats.requeues++;
	322
	323	return NET_XMIT_SUCCESS;
	324	}
1da177e4 LT	325
	326	static unsigned int dsmark_drop(struct Qdisc *sch)
	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
1e90474c	354	if (!opt \|\| nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL) < 0)
758cc43c TG	355	goto errout;
758cc43c TG	356
1e90474c PM	357	if (nla_len(tb[TCA_DSMARK_INDICES]) < sizeof(u16))
	358	goto errout;
	359	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	360
5b0ac72b DM	361	if (hweight32(indices) != 1)
758cc43c TG	362	goto errout;
758cc43c TG	363
1e90474c PM	364	if (tb[TCA_DSMARK_DEFAULT_INDEX]) {
	365	if (nla_len(tb[TCA_DSMARK_DEFAULT_INDEX]) < sizeof(u16))
	366	goto errout;
	367	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
	368	}
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
9f9afec4	386	p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
758cc43c	387	if (p->q == NULL)
1da177e4	388	p->q = &noop_qdisc;
758cc43c	389
81da99ed	390	pr_debug("dsmark_init: qdisc %p\n", p->q);
758cc43c TG	391
	392	err = 0;
	393	errout:
758cc43c	394	return err;
1da177e4 LT	395	}
1da177e4 LT	396
1da177e4 LT	397	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	398	{
81da99ed	399	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	400
81da99ed	401	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
1da177e4 LT	402	qdisc_reset(p->q);
	403	sch->q.qlen = 0;
	404	}
	405
1da177e4 LT	406	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	407	{
81da99ed	408	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	409
81da99ed	410	pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
af0d1141	411
a48b5a61	412	tcf_destroy_chain(p->filter_list);
1da177e4 LT	413	qdisc_destroy(p->q);
	414	kfree(p->mask);
	415	}
	416
1da177e4	417	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	418	struct sk_buff skb, struct tcmsg tcm)
1da177e4	419	{
81da99ed	420	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	421	struct nlattr *opts = NULL;
1da177e4	422
81da99ed	423	pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
02f23f09 TG	424
02f23f09 TG	425	if (!dsmark_valid_index(p, cl))
1da177e4	426	return -EINVAL;
02f23f09 TG	427
02f23f09 TG	428	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
cdc7f8e3	429	tcm->tcm_info = p->q->handle;
02f23f09	430
1e90474c PM	431	opts = nla_nest_start(skb, TCA_OPTIONS);
	432	if (opts == NULL)
	433	goto nla_put_failure;
	434	NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
	435	NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
02f23f09	436
1e90474c	437	return nla_nest_end(skb, opts);
1da177e4	438
1e90474c PM	439	nla_put_failure:
1e90474c PM	440	return nla_nest_cancel(skb, opts);
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;
	451	NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
02f23f09 TG	452
02f23f09 TG	453	if (p->default_index != NO_DEFAULT_INDEX)
1e90474c	454	NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
1da177e4	455
1da177e4	456	if (p->set_tc_index)
1e90474c	457	NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
02f23f09	458
1e90474c	459	return nla_nest_end(skb, opts);
1da177e4	460
1e90474c PM	461	nla_put_failure:
1e90474c PM	462	return nla_nest_cancel(skb, opts);
1da177e4 LT	463	}
1da177e4 LT	464
20fea08b	465	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	466	.graft = dsmark_graft,
	467	.leaf = dsmark_leaf,
	468	.get = dsmark_get,
	469	.put = dsmark_put,
	470	.change = dsmark_change,
	471	.delete = dsmark_delete,
	472	.walk = dsmark_walk,
	473	.tcf_chain = dsmark_find_tcf,
	474	.bind_tcf = dsmark_bind_filter,
	475	.unbind_tcf = dsmark_put,
	476	.dump = dsmark_dump_class,
	477	};
	478
20fea08b	479	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	480	.next = NULL,
	481	.cl_ops = &dsmark_class_ops,
	482	.id = "dsmark",
	483	.priv_size = sizeof(struct dsmark_qdisc_data),
	484	.enqueue = dsmark_enqueue,
	485	.dequeue = dsmark_dequeue,
	486	.requeue = dsmark_requeue,
	487	.drop = dsmark_drop,
	488	.init = dsmark_init,
	489	.reset = dsmark_reset,
	490	.destroy = dsmark_destroy,
	491	.change = NULL,
	492	.dump = dsmark_dump,
	493	.owner = THIS_MODULE,
	494	};
	495
	496	static int __init dsmark_module_init(void)
	497	{
	498	return register_qdisc(&dsmark_qdisc_ops);
	499	}
af0d1141	500
10297b99	501	static void __exit dsmark_module_exit(void)
1da177e4 LT	502	{
	503	unregister_qdisc(&dsmark_qdisc_ops);
	504	}
af0d1141	505
1da177e4 LT	506	module_init(dsmark_module_init)
1da177e4 LT	507	module_exit(dsmark_module_exit)
af0d1141	508
1da177e4	509	MODULE_LICENSE("GPL");