[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(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)
		goto errout;

	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL);
	if (err < 0)
		return err;

	err = -EINVAL;
	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)
		goto errout;

	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL);
	if (err < 0)
		goto errout;

	err = -EINVAL;
	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
cee63723	119	if (!opt)
1e90474c	120	goto errout;
758cc43c	121
cee63723 PM	122	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL);
	123	if (err < 0)
	124	return err;
	125
	126	err = -EINVAL;
1e90474c PM	127	if (tb[TCA_DSMARK_MASK]) {
	128	if (nla_len(tb[TCA_DSMARK_MASK]) < sizeof(u8))
	129	goto errout;
	130	mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
	131	}
	132	if (tb[TCA_DSMARK_VALUE]) {
	133	if (nla_len(tb[TCA_DSMARK_VALUE]) < sizeof(u8))
	134	goto errout;
	135	p->value[*arg-1] = nla_get_u8(tb[TCA_DSMARK_VALUE]);
	136	}
10297b99	137
1e90474c	138	if (tb[TCA_DSMARK_MASK])
758cc43c TG	139	p->mask[*arg-1] = mask;
	140
	141	err = 0;
	142
1e90474c	143	errout:
758cc43c TG	144	return err;
758cc43c TG	145	}
1da177e4	146
af0d1141	147	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4	148	{
81da99ed	149	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	150
af0d1141	151	if (!dsmark_valid_index(p, arg))
1da177e4	152	return -EINVAL;
10297b99	153
1da177e4 LT	154	p->mask[arg-1] = 0xff;
1da177e4 LT	155	p->value[arg-1] = 0;
af0d1141	156
1da177e4 LT	157	return 0;
	158	}
	159
9d127fbd	160	static void dsmark_walk(struct Qdisc sch, struct qdisc_walker walker)
1da177e4	161	{
81da99ed	162	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 LT	163	int i;
1da177e4 LT	164
81da99ed	165	pr_debug("dsmark_walk(sch %p,[qdisc %p],walker %p)\n", sch, p, walker);
af0d1141	166
1da177e4 LT	167	if (walker->stop)
1da177e4 LT	168	return;
af0d1141	169
1da177e4 LT	170	for (i = 0; i < p->indices; i++) {
1da177e4 LT	171	if (p->mask[i] == 0xff && !p->value[i])
0451eb07	172	goto ignore;
1da177e4 LT	173	if (walker->count >= walker->skip) {
	174	if (walker->fn(sch, i+1, walker) < 0) {
	175	walker->stop = 1;
	176	break;
	177	}
	178	}
10297b99	179	ignore:
0451eb07	180	walker->count++;
10297b99	181	}
1da177e4 LT	182	}
1da177e4 LT	183
9d127fbd SH	184	static inline struct tcf_proto *dsmark_find_tcf(struct Qdisc sch,
9d127fbd SH	185	unsigned long cl)
1da177e4	186	{
81da99ed SH	187	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	188	return &p->filter_list;
1da177e4 LT	189	}
1da177e4 LT	190
1da177e4 LT	191	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	192
9d127fbd	193	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch)
1da177e4	194	{
81da99ed	195	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	196	int err;
af0d1141 TG	197
81da99ed	198	pr_debug("dsmark_enqueue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
1da177e4	199
1da177e4	200	if (p->set_tc_index) {
1da177e4	201	switch (skb->protocol) {
9d127fbd SH	202	case __constant_htons(ETH_P_IP):
	203	if (skb_cow_head(skb, sizeof(struct iphdr)))
	204	goto drop;
4c30719f	205
9d127fbd SH	206	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	207	& ~INET_ECN_MASK;
	208	break;
4c30719f	209
9d127fbd SH	210	case __constant_htons(ETH_P_IPV6):
	211	if (skb_cow_head(skb, sizeof(struct ipv6hdr)))
	212	goto drop;
4c30719f	213
9d127fbd SH	214	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	215	& ~INET_ECN_MASK;
	216	break;
	217	default:
	218	skb->tc_index = 0;
	219	break;
3ff50b79	220	}
1da177e4	221	}
af0d1141 TG	222
af0d1141 TG	223	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	224	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	225	else {
	226	struct tcf_result res;
	227	int result = tc_classify(skb, p->filter_list, &res);
	228
81da99ed	229	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	230
1da177e4	231	switch (result) {
f6853e2d PM	232	#ifdef CONFIG_NET_CLS_ACT
	233	case TC_ACT_QUEUED:
	234	case TC_ACT_STOLEN:
	235	kfree_skb(skb);
	236	return NET_XMIT_SUCCESS;
4c30719f	237
f6853e2d	238	case TC_ACT_SHOT:
4c30719f	239	goto drop;
1da177e4	240	#endif
c3bc7cff	241	case TC_ACT_OK:
f6853e2d PM	242	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	243	break;
4c30719f	244
f6853e2d PM	245	default:
	246	if (p->default_index != NO_DEFAULT_INDEX)
	247	skb->tc_index = p->default_index;
	248	break;
3ff50b79	249	}
1da177e4	250	}
1da177e4	251
9d127fbd	252	err = p->q->enqueue(skb, p->q);
af0d1141	253	if (err != NET_XMIT_SUCCESS) {
1da177e4	254	sch->qstats.drops++;
af0d1141	255	return err;
1da177e4	256	}
af0d1141	257
1da177e4 LT	258	sch->bstats.bytes += skb->len;
	259	sch->bstats.packets++;
	260	sch->q.qlen++;
1da177e4	261
af0d1141	262	return NET_XMIT_SUCCESS;
4c30719f SH	263
	264	drop:
	265	kfree_skb(skb);
	266	sch->qstats.drops++;
	267	return NET_XMIT_BYPASS;
af0d1141	268	}
1da177e4 LT	269
	270	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	271	{
81da99ed	272	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	273	struct sk_buff *skb;
af0d1141 TG	274	u32 index;
af0d1141 TG	275
81da99ed	276	pr_debug("dsmark_dequeue(sch %p,[qdisc %p])\n", sch, p);
1da177e4	277
1da177e4	278	skb = p->q->ops->dequeue(p->q);
af0d1141	279	if (skb == NULL)
1da177e4	280	return NULL;
af0d1141	281
1da177e4	282	sch->q.qlen--;
af0d1141 TG	283
af0d1141 TG	284	index = skb->tc_index & (p->indices - 1);
81da99ed	285	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	286
1da177e4	287	switch (skb->protocol) {
9d127fbd SH	288	case __constant_htons(ETH_P_IP):
	289	ipv4_change_dsfield(ip_hdr(skb), p->mask[index],
	290	p->value[index]);
1da177e4	291	break;
9d127fbd SH	292	case __constant_htons(ETH_P_IPV6):
	293	ipv6_change_dsfield(ipv6_hdr(skb), p->mask[index],
	294	p->value[index]);
1da177e4	295	break;
9d127fbd SH	296	default:
	297	/*
	298	* Only complain if a change was actually attempted.
	299	* This way, we can send non-IP traffic through dsmark
	300	* and don't need yet another qdisc as a bypass.
	301	*/
	302	if (p->mask[index] != 0xff \|\| p->value[index])
	303	printk(KERN_WARNING
	304	"dsmark_dequeue: unsupported protocol %d\n",
	305	ntohs(skb->protocol));
	306	break;
3ff50b79	307	}
af0d1141	308
1da177e4 LT	309	return skb;
	310	}
	311
9d127fbd	312	static int dsmark_requeue(struct sk_buff skb, struct Qdisc sch)
1da177e4	313	{
81da99ed	314	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141	315	int err;
1da177e4	316
81da99ed	317	pr_debug("dsmark_requeue(skb %p,sch %p,[qdisc %p])\n", skb, sch, p);
af0d1141 TG	318
	319	err = p->q->ops->requeue(skb, p->q);
	320	if (err != NET_XMIT_SUCCESS) {
	321	sch->qstats.drops++;
	322	return err;
1da177e4	323	}
1da177e4	324
af0d1141 TG	325	sch->q.qlen++;
	326	sch->qstats.requeues++;
	327
	328	return NET_XMIT_SUCCESS;
	329	}
1da177e4 LT	330
	331	static unsigned int dsmark_drop(struct Qdisc *sch)
	332	{
81da99ed	333	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	334	unsigned int len;
10297b99	335
81da99ed	336	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
af0d1141 TG	337
af0d1141 TG	338	if (p->q->ops->drop == NULL)
1da177e4	339	return 0;
af0d1141 TG	340
	341	len = p->q->ops->drop(p->q);
	342	if (len)
	343	sch->q.qlen--;
	344
1da177e4 LT	345	return len;
	346	}
	347
1e90474c	348	static int dsmark_init(struct Qdisc sch, struct nlattr opt)
1da177e4	349	{
81da99ed	350	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	351	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c TG	352	int err = -EINVAL;
	353	u32 default_index = NO_DEFAULT_INDEX;
	354	u16 indices;
	355	u8 *mask;
	356
81da99ed	357	pr_debug("dsmark_init(sch %p,[qdisc %p],opt %p)\n", sch, p, opt);
758cc43c	358
cee63723 PM	359	if (!opt)
	360	goto errout;
	361
	362	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, NULL);
	363	if (err < 0)
758cc43c TG	364	goto errout;
758cc43c TG	365
cee63723	366	err = -EINVAL;
1e90474c PM	367	if (nla_len(tb[TCA_DSMARK_INDICES]) < sizeof(u16))
	368	goto errout;
	369	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	370
5b0ac72b DM	371	if (hweight32(indices) != 1)
758cc43c TG	372	goto errout;
758cc43c TG	373
1e90474c PM	374	if (tb[TCA_DSMARK_DEFAULT_INDEX]) {
	375	if (nla_len(tb[TCA_DSMARK_DEFAULT_INDEX]) < sizeof(u16))
	376	goto errout;
	377	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
	378	}
758cc43c TG	379
	380	mask = kmalloc(indices * 2, GFP_KERNEL);
	381	if (mask == NULL) {
	382	err = -ENOMEM;
	383	goto errout;
1da177e4	384	}
758cc43c TG	385
	386	p->mask = mask;
	387	memset(p->mask, 0xff, indices);
	388
	389	p->value = p->mask + indices;
	390	memset(p->value, 0, indices);
	391
	392	p->indices = indices;
	393	p->default_index = default_index;
1e90474c	394	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	395
9f9afec4	396	p->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops, sch->handle);
758cc43c	397	if (p->q == NULL)
1da177e4	398	p->q = &noop_qdisc;
758cc43c	399
81da99ed	400	pr_debug("dsmark_init: qdisc %p\n", p->q);
758cc43c TG	401
	402	err = 0;
	403	errout:
758cc43c	404	return err;
1da177e4 LT	405	}
1da177e4 LT	406
1da177e4 LT	407	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	408	{
81da99ed	409	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	410
81da99ed	411	pr_debug("dsmark_reset(sch %p,[qdisc %p])\n", sch, p);
1da177e4 LT	412	qdisc_reset(p->q);
	413	sch->q.qlen = 0;
	414	}
	415
1da177e4 LT	416	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	417	{
81da99ed	418	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	419
81da99ed	420	pr_debug("dsmark_destroy(sch %p,[qdisc %p])\n", sch, p);
af0d1141	421
a48b5a61	422	tcf_destroy_chain(p->filter_list);
1da177e4 LT	423	qdisc_destroy(p->q);
	424	kfree(p->mask);
	425	}
	426
1da177e4	427	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	428	struct sk_buff skb, struct tcmsg tcm)
1da177e4	429	{
81da99ed	430	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	431	struct nlattr *opts = NULL;
1da177e4	432
81da99ed	433	pr_debug("dsmark_dump_class(sch %p,[qdisc %p],class %ld\n", sch, p, cl);
02f23f09 TG	434
02f23f09 TG	435	if (!dsmark_valid_index(p, cl))
1da177e4	436	return -EINVAL;
02f23f09 TG	437
02f23f09 TG	438	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl-1);
cdc7f8e3	439	tcm->tcm_info = p->q->handle;
02f23f09	440
1e90474c PM	441	opts = nla_nest_start(skb, TCA_OPTIONS);
	442	if (opts == NULL)
	443	goto nla_put_failure;
	444	NLA_PUT_U8(skb, TCA_DSMARK_MASK, p->mask[cl-1]);
	445	NLA_PUT_U8(skb, TCA_DSMARK_VALUE, p->value[cl-1]);
02f23f09	446
1e90474c	447	return nla_nest_end(skb, opts);
1da177e4	448
1e90474c PM	449	nla_put_failure:
1e90474c PM	450	return nla_nest_cancel(skb, opts);
1da177e4 LT	451	}
	452
	453	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	454	{
81da99ed	455	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	456	struct nlattr *opts = NULL;
1da177e4	457
1e90474c PM	458	opts = nla_nest_start(skb, TCA_OPTIONS);
	459	if (opts == NULL)
	460	goto nla_put_failure;
	461	NLA_PUT_U16(skb, TCA_DSMARK_INDICES, p->indices);
02f23f09 TG	462
02f23f09 TG	463	if (p->default_index != NO_DEFAULT_INDEX)
1e90474c	464	NLA_PUT_U16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index);
1da177e4	465
1da177e4	466	if (p->set_tc_index)
1e90474c	467	NLA_PUT_FLAG(skb, TCA_DSMARK_SET_TC_INDEX);
02f23f09	468
1e90474c	469	return nla_nest_end(skb, opts);
1da177e4	470
1e90474c PM	471	nla_put_failure:
1e90474c PM	472	return nla_nest_cancel(skb, opts);
1da177e4 LT	473	}
1da177e4 LT	474
20fea08b	475	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	476	.graft = dsmark_graft,
	477	.leaf = dsmark_leaf,
	478	.get = dsmark_get,
	479	.put = dsmark_put,
	480	.change = dsmark_change,
	481	.delete = dsmark_delete,
	482	.walk = dsmark_walk,
	483	.tcf_chain = dsmark_find_tcf,
	484	.bind_tcf = dsmark_bind_filter,
	485	.unbind_tcf = dsmark_put,
	486	.dump = dsmark_dump_class,
	487	};
	488
20fea08b	489	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	490	.next = NULL,
	491	.cl_ops = &dsmark_class_ops,
	492	.id = "dsmark",
	493	.priv_size = sizeof(struct dsmark_qdisc_data),
	494	.enqueue = dsmark_enqueue,
	495	.dequeue = dsmark_dequeue,
	496	.requeue = dsmark_requeue,
	497	.drop = dsmark_drop,
	498	.init = dsmark_init,
	499	.reset = dsmark_reset,
	500	.destroy = dsmark_destroy,
	501	.change = NULL,
	502	.dump = dsmark_dump,
	503	.owner = THIS_MODULE,
	504	};
	505
	506	static int __init dsmark_module_init(void)
	507	{
	508	return register_qdisc(&dsmark_qdisc_ops);
	509	}
af0d1141	510
10297b99	511	static void __exit dsmark_module_exit(void)
1da177e4 LT	512	{
	513	unregister_qdisc(&dsmark_qdisc_ops);
	514	}
af0d1141	515
1da177e4 LT	516	module_init(dsmark_module_init)
1da177e4 LT	517	module_exit(dsmark_module_exit)
af0d1141	518
1da177e4	519	MODULE_LICENSE("GPL");