[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/slab.h>
#include <linux/types.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/skbuff.h>
#include <linux/rtnetlink.h>
#include <linux/bitops.h>
#include <net/pkt_sched.h>
#include <net/pkt_cls.h>
#include <net/dsfield.h>
#include <net/inet_ecn.h>
#include <asm/byteorder.h>

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


#define NO_DEFAULT_INDEX	(1 << 16)

struct mask_value {
	u8			mask;
	u8			value;
};

struct dsmark_qdisc_data {
	struct Qdisc		*q;
	struct tcf_proto __rcu	*filter_list;
	struct tcf_block	*block;
	struct mask_value	*mv;
	u16			indices;
	u8			set_tc_index;
	u32			default_index;	/* index range is 0...0xffff */
#define DSMARK_EMBEDDED_SZ	16
	struct mask_value	embedded[DSMARK_EMBEDDED_SZ];
};

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

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

static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
			struct Qdisc *new, struct Qdisc **old,
			struct netlink_ext_ack *extack)
{
	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, NULL);
		if (new == NULL)
			new = &noop_qdisc;
	}

	*old = qdisc_replace(sch, new, &p->q);
	return 0;
}

static struct Qdisc *dsmark_leaf(struct Qdisc *sch, unsigned long arg)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	return p->q;
}

static unsigned long dsmark_find(struct Qdisc *sch, u32 classid)
{
	return TC_H_MIN(classid) + 1;
}

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

	return dsmark_find(sch, classid);
}

static void dsmark_unbind_filter(struct Qdisc *sch, unsigned long cl)
{
}

static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
	[TCA_DSMARK_INDICES]		= { .type = NLA_U16 },
	[TCA_DSMARK_DEFAULT_INDEX]	= { .type = NLA_U16 },
	[TCA_DSMARK_SET_TC_INDEX]	= { .type = NLA_FLAG },
	[TCA_DSMARK_MASK]		= { .type = NLA_U8 },
	[TCA_DSMARK_VALUE]		= { .type = NLA_U8 },
};

static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
			 struct nlattr **tca, unsigned long *arg,
			 struct netlink_ext_ack *extack)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	struct nlattr *opt = tca[TCA_OPTIONS];
	struct nlattr *tb[TCA_DSMARK_MAX + 1];
	int err = -EINVAL;

	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
		 __func__, sch, p, classid, parent, *arg);

	if (!dsmark_valid_index(p, *arg)) {
		err = -ENOENT;
		goto errout;
	}

	if (!opt)
		goto errout;

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

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

	if (tb[TCA_DSMARK_MASK])
		p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);

	err = 0;

errout:
	return err;
}

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

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

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

	return 0;
}

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

	pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
		 __func__, sch, p, walker);

	if (walker->stop)
		return;

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

static struct tcf_block *dsmark_tcf_block(struct Qdisc *sch, unsigned long cl,
					  struct netlink_ext_ack *extack)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);

	return p->block;
}

/* --------------------------- Qdisc operations ---------------------------- */

static int dsmark_enqueue(struct sk_buff *skb, struct Qdisc *sch,
			  struct sk_buff **to_free)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	int err;

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

	if (p->set_tc_index) {
		int wlen = skb_network_offset(skb);

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

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

		case htons(ETH_P_IPV6):
			wlen += sizeof(struct ipv6hdr);
			if (!pskb_may_pull(skb, wlen) ||
			    skb_try_make_writable(skb, wlen))
				goto drop;

			skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
				& ~INET_ECN_MASK;
			break;
		default:
			skb->tc_index = 0;
			break;
		}
	}

	if (TC_H_MAJ(skb->priority) == sch->handle)
		skb->tc_index = TC_H_MIN(skb->priority);
	else {
		struct tcf_result res;
		struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
		int result = tcf_classify(skb, fl, &res, false);

		pr_debug("result %d class 0x%04x\n", result, res.classid);

		switch (result) {
#ifdef CONFIG_NET_CLS_ACT
		case TC_ACT_QUEUED:
		case TC_ACT_STOLEN:
		case TC_ACT_TRAP:
			__qdisc_drop(skb, to_free);
			return NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;

		case TC_ACT_SHOT:
			goto drop;
#endif
		case TC_ACT_OK:
			skb->tc_index = TC_H_MIN(res.classid);
			break;

		default:
			if (p->default_index != NO_DEFAULT_INDEX)
				skb->tc_index = p->default_index;
			break;
		}
	}

	err = qdisc_enqueue(skb, p->q, to_free);
	if (err != NET_XMIT_SUCCESS) {
		if (net_xmit_drop_count(err))
			qdisc_qstats_drop(sch);
		return err;
	}

	qdisc_qstats_backlog_inc(sch, skb);
	sch->q.qlen++;

	return NET_XMIT_SUCCESS;

drop:
	qdisc_drop(skb, sch, to_free);
	return NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
}

static struct sk_buff *dsmark_dequeue(struct Qdisc *sch)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	struct sk_buff *skb;
	u32 index;

	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);

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

	qdisc_bstats_update(sch, skb);
	qdisc_qstats_backlog_dec(sch, skb);
	sch->q.qlen--;

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

	switch (tc_skb_protocol(skb)) {
	case htons(ETH_P_IP):
		ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
				    p->mv[index].value);
			break;
	case htons(ETH_P_IPV6):
		ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
				    p->mv[index].value);
			break;
	default:
		/*
		 * Only complain if a change was actually attempted.
		 * This way, we can send non-IP traffic through dsmark
		 * and don't need yet another qdisc as a bypass.
		 */
		if (p->mv[index].mask != 0xff || p->mv[index].value)
			pr_warn("%s: unsupported protocol %d\n",
				__func__, ntohs(tc_skb_protocol(skb)));
		break;
	}

	return skb;
}

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

	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);

	return p->q->ops->peek(p->q);
}

static int dsmark_init(struct Qdisc *sch, struct nlattr *opt,
		       struct netlink_ext_ack *extack)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	struct nlattr *tb[TCA_DSMARK_MAX + 1];
	int err = -EINVAL;
	u32 default_index = NO_DEFAULT_INDEX;
	u16 indices;
	int i;

	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);

	if (!opt)
		goto errout;

	err = tcf_block_get(&p->block, &p->filter_list, sch, extack);
	if (err)
		return err;

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

	err = -EINVAL;
	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);

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

	if (tb[TCA_DSMARK_DEFAULT_INDEX])
		default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);

	if (indices <= DSMARK_EMBEDDED_SZ)
		p->mv = p->embedded;
	else
		p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
	if (!p->mv) {
		err = -ENOMEM;
		goto errout;
	}
	for (i = 0; i < indices; i++) {
		p->mv[i].mask = 0xff;
		p->mv[i].value = 0;
	}
	p->indices = indices;
	p->default_index = default_index;
	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);

	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle,
				 NULL);
	if (p->q == NULL)
		p->q = &noop_qdisc;
	else
		qdisc_hash_add(p->q, true);

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

	err = 0;
errout:
	return err;
}

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

	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
	qdisc_reset(p->q);
	sch->qstats.backlog = 0;
	sch->q.qlen = 0;
}

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

	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);

	tcf_block_put(p->block);
	qdisc_put(p->q);
	if (p->mv != p->embedded)
		kfree(p->mv);
}

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

	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);

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

	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
	tcm->tcm_info = p->q->handle;

	opts = nla_nest_start(skb, TCA_OPTIONS);
	if (opts == NULL)
		goto nla_put_failure;
	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) ||
	    nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
		goto nla_put_failure;

	return nla_nest_end(skb, opts);

nla_put_failure:
	nla_nest_cancel(skb, opts);
	return -EMSGSIZE;
}

static int dsmark_dump(struct Qdisc *sch, struct sk_buff *skb)
{
	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	struct nlattr *opts = NULL;

	opts = nla_nest_start(skb, TCA_OPTIONS);
	if (opts == NULL)
		goto nla_put_failure;
	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
		goto nla_put_failure;

	if (p->default_index != NO_DEFAULT_INDEX &&
	    nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
		goto nla_put_failure;

	if (p->set_tc_index &&
	    nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
		goto nla_put_failure;

	return nla_nest_end(skb, opts);

nla_put_failure:
	nla_nest_cancel(skb, opts);
	return -EMSGSIZE;
}

static const struct Qdisc_class_ops dsmark_class_ops = {
	.graft		=	dsmark_graft,
	.leaf		=	dsmark_leaf,
	.find		=	dsmark_find,
	.change		=	dsmark_change,
	.delete		=	dsmark_delete,
	.walk		=	dsmark_walk,
	.tcf_block	=	dsmark_tcf_block,
	.bind_tcf	=	dsmark_bind_filter,
	.unbind_tcf	=	dsmark_unbind_filter,
	.dump		=	dsmark_dump_class,
};

static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
	.next		=	NULL,
	.cl_ops		=	&dsmark_class_ops,
	.id		=	"dsmark",
	.priv_size	=	sizeof(struct dsmark_qdisc_data),
	.enqueue	=	dsmark_enqueue,
	.dequeue	=	dsmark_dequeue,
	.peek		=	dsmark_peek,
	.init		=	dsmark_init,
	.reset		=	dsmark_reset,
	.destroy	=	dsmark_destroy,
	.change		=	NULL,
	.dump		=	dsmark_dump,
	.owner		=	THIS_MODULE,
};

static int __init dsmark_module_init(void)
{
	return register_qdisc(&dsmark_qdisc_ops);
}

static void __exit dsmark_module_exit(void)
{
	unregister_qdisc(&dsmark_qdisc_ops);
}

module_init(dsmark_module_init)
module_exit(dsmark_module_exit)

MODULE_LICENSE("GPL");
Commit	Line	Data
1da177e4 LT	1	/* net/sched/sch_dsmark.c - Differentiated Services field marker */
	2
	3	/* Written 1998-2000 by Werner Almesberger, EPFL ICA */
	4
	5
1da177e4 LT	6	#include <linux/module.h>
1da177e4 LT	7	#include <linux/init.h>
5a0e3ad6	8	#include <linux/slab.h>
1da177e4 LT	9	#include <linux/types.h>
	10	#include <linux/string.h>
	11	#include <linux/errno.h>
	12	#include <linux/skbuff.h>
1da177e4	13	#include <linux/rtnetlink.h>
5b0ac72b	14	#include <linux/bitops.h>
1da177e4	15	#include <net/pkt_sched.h>
cf1facda	16	#include <net/pkt_cls.h>
1da177e4 LT	17	#include <net/dsfield.h>
	18	#include <net/inet_ecn.h>
	19	#include <asm/byteorder.h>
	20
1da177e4 LT	21	/*
	22	* classid class marking
	23	* ------- ----- -------
	24	* n/a 0 n/a
	25	* x:0 1 use entry [0]
	26	* ... ... ...
	27	* x:y y>0 y+1 use entry [y]
	28	* ... ... ...
	29	* x:indices-1 indices use entry [indices-1]
	30	* ... ... ...
	31	* x:y y+1 use entry [y & (indices-1)]
	32	* ... ... ...
	33	* 0xffff 0x10000 use entry [indices-1]
	34	*/
	35
	36
	37	#define NO_DEFAULT_INDEX (1 << 16)
	38
47bbbb30 ED	39	struct mask_value {
	40	u8 mask;
	41	u8 value;
	42	};
	43
1da177e4 LT	44	struct dsmark_qdisc_data {
1da177e4 LT	45	struct Qdisc *q;
25d8c0d5	46	struct tcf_proto __rcu *filter_list;
6529eaba	47	struct tcf_block *block;
47bbbb30	48	struct mask_value *mv;
af0d1141	49	u16 indices;
47bbbb30	50	u8 set_tc_index;
af0d1141	51	u32 default_index; /* index range is 0...0xffff */
47bbbb30 ED	52	#define DSMARK_EMBEDDED_SZ 16
47bbbb30 ED	53	struct mask_value embedded[DSMARK_EMBEDDED_SZ];
1da177e4 LT	54	};
1da177e4 LT	55
758cc43c TG	56	static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
758cc43c TG	57	{
17569fae	58	return index <= p->indices && index > 0;
758cc43c	59	}
1da177e4 LT	60
	61	/* ------------------------- Class/flow operations ------------------------- */
	62
af0d1141	63	static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
653d6fd6 AA	64	struct Qdisc new, struct Qdisc *old,
653d6fd6 AA	65	struct netlink_ext_ack *extack)
1da177e4	66	{
81da99ed	67	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	68
c76f2a2c YY	69	pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
c76f2a2c YY	70	__func__, sch, p, new, old);
486b53e5 TG	71
486b53e5 TG	72	if (new == NULL) {
3511c913	73	new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
a38a9882	74	sch->handle, NULL);
486b53e5 TG	75	if (new == NULL)
	76	new = &noop_qdisc;
	77	}
	78
86a7996c	79	*old = qdisc_replace(sch, new, &p->q);
10297b99	80	return 0;
1da177e4 LT	81	}
1da177e4 LT	82
1da177e4 LT	83	static struct Qdisc dsmark_leaf(struct Qdisc sch, unsigned long arg)
1da177e4 LT	84	{
81da99ed SH	85	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	86	return p->q;
1da177e4 LT	87	}
1da177e4 LT	88
143976ce	89	static unsigned long dsmark_find(struct Qdisc *sch, u32 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	{
143976ce WC	97	pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
	98	__func__, sch, qdisc_priv(sch), classid);
	99
	100	return dsmark_find(sch, classid);
1da177e4 LT	101	}
1da177e4 LT	102
143976ce	103	static void dsmark_unbind_filter(struct Qdisc *sch, unsigned long cl)
1da177e4 LT	104	{
	105	}
	106
27a3421e PM	107	static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
	108	[TCA_DSMARK_INDICES] = { .type = NLA_U16 },
	109	[TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
	110	[TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
	111	[TCA_DSMARK_MASK] = { .type = NLA_U8 },
	112	[TCA_DSMARK_VALUE] = { .type = NLA_U8 },
	113	};
	114
1da177e4	115	static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
793d81d6 AA	116	struct nlattr *tca, unsigned long arg,
793d81d6 AA	117	struct netlink_ext_ack *extack)
1da177e4	118	{
81da99ed	119	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c PM	120	struct nlattr *opt = tca[TCA_OPTIONS];
1e90474c PM	121	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c	122	int err = -EINVAL;
1da177e4	123
c76f2a2c YY	124	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
c76f2a2c YY	125	__func__, sch, p, classid, parent, *arg);
758cc43c TG	126
	127	if (!dsmark_valid_index(p, *arg)) {
	128	err = -ENOENT;
1e90474c	129	goto errout;
1da177e4	130	}
1da177e4	131
cee63723	132	if (!opt)
1e90474c	133	goto errout;
758cc43c	134
fceb6435	135	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
cee63723	136	if (err < 0)
27a3421e	137	goto errout;
cee63723	138
27a3421e	139	if (tb[TCA_DSMARK_VALUE])
47bbbb30	140	p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99	141
1e90474c	142	if (tb[TCA_DSMARK_MASK])
47bbbb30	143	p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_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
47bbbb30 ED	158	p->mv[arg - 1].mask = 0xff;
47bbbb30 ED	159	p->mv[arg - 1].value = 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	175	for (i = 0; i < p->indices; i++) {
47bbbb30	176	if (p->mv[i].mask == 0xff && !p->mv[i].value)
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
cbaacc4e AA	189	static struct tcf_block dsmark_tcf_block(struct Qdisc sch, unsigned long cl,
cbaacc4e AA	190	struct netlink_ext_ack *extack)
1da177e4	191	{
81da99ed	192	struct dsmark_qdisc_data *p = qdisc_priv(sch);
6529eaba JP	193
6529eaba JP	194	return p->block;
1da177e4 LT	195	}
1da177e4 LT	196
1da177e4 LT	197	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	198
520ac30f ED	199	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch,
520ac30f ED	200	struct sk_buff **to_free)
1da177e4	201	{
81da99ed	202	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	203	int err;
af0d1141 TG	204
c76f2a2c	205	pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
1da177e4	206
1da177e4	207	if (p->set_tc_index) {
aea92fb2 ED	208	int wlen = skb_network_offset(skb);
aea92fb2 ED	209
d8b9605d	210	switch (tc_skb_protocol(skb)) {
60678040	211	case htons(ETH_P_IP):
aea92fb2 ED	212	wlen += sizeof(struct iphdr);
	213	if (!pskb_may_pull(skb, wlen) \|\|
	214	skb_try_make_writable(skb, wlen))
9d127fbd	215	goto drop;
4c30719f	216
9d127fbd SH	217	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	218	& ~INET_ECN_MASK;
	219	break;
4c30719f	220
60678040	221	case htons(ETH_P_IPV6):
aea92fb2 ED	222	wlen += sizeof(struct ipv6hdr);
	223	if (!pskb_may_pull(skb, wlen) \|\|
	224	skb_try_make_writable(skb, wlen))
9d127fbd	225	goto drop;
4c30719f	226
9d127fbd SH	227	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	228	& ~INET_ECN_MASK;
	229	break;
	230	default:
	231	skb->tc_index = 0;
	232	break;
3ff50b79	233	}
1da177e4	234	}
af0d1141 TG	235
af0d1141 TG	236	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	237	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	238	else {
af0d1141 TG	239	struct tcf_result res;
25d8c0d5	240	struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
87d83093	241	int result = tcf_classify(skb, fl, &res, false);
af0d1141	242
81da99ed	243	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	244
1da177e4	245	switch (result) {
f6853e2d PM	246	#ifdef CONFIG_NET_CLS_ACT
	247	case TC_ACT_QUEUED:
	248	case TC_ACT_STOLEN:
e25ea21f	249	case TC_ACT_TRAP:
520ac30f	250	__qdisc_drop(skb, to_free);
378a2f09	251	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	252
f6853e2d	253	case TC_ACT_SHOT:
4c30719f	254	goto drop;
1da177e4	255	#endif
c3bc7cff	256	case TC_ACT_OK:
f6853e2d PM	257	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	258	break;
4c30719f	259
f6853e2d PM	260	default:
	261	if (p->default_index != NO_DEFAULT_INDEX)
	262	skb->tc_index = p->default_index;
	263	break;
3ff50b79	264	}
1da177e4	265	}
1da177e4	266
520ac30f	267	err = qdisc_enqueue(skb, p->q, to_free);
af0d1141	268	if (err != NET_XMIT_SUCCESS) {
378a2f09	269	if (net_xmit_drop_count(err))
25331d6c	270	qdisc_qstats_drop(sch);
af0d1141	271	return err;
1da177e4	272	}
af0d1141	273
bdf17661	274	qdisc_qstats_backlog_inc(sch, skb);
1da177e4	275	sch->q.qlen++;
1da177e4	276
af0d1141	277	return NET_XMIT_SUCCESS;
4c30719f SH	278
4c30719f SH	279	drop:
520ac30f	280	qdisc_drop(skb, sch, to_free);
c27f339a	281	return NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
af0d1141	282	}
1da177e4 LT	283
	284	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	285	{
81da99ed	286	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	287	struct sk_buff *skb;
af0d1141 TG	288	u32 index;
af0d1141 TG	289
c76f2a2c	290	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	291
f8b33d8e	292	skb = qdisc_dequeue_peeked(p->q);
af0d1141	293	if (skb == NULL)
1da177e4	294	return NULL;
af0d1141	295
9190b3b3	296	qdisc_bstats_update(sch, skb);
bdf17661	297	qdisc_qstats_backlog_dec(sch, skb);
1da177e4	298	sch->q.qlen--;
af0d1141 TG	299
af0d1141 TG	300	index = skb->tc_index & (p->indices - 1);
81da99ed	301	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	302
d8b9605d	303	switch (tc_skb_protocol(skb)) {
60678040	304	case htons(ETH_P_IP):
47bbbb30 ED	305	ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
47bbbb30 ED	306	p->mv[index].value);
1da177e4	307	break;
60678040	308	case htons(ETH_P_IPV6):
47bbbb30 ED	309	ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
47bbbb30 ED	310	p->mv[index].value);
1da177e4	311	break;
9d127fbd SH	312	default:
	313	/*
	314	* Only complain if a change was actually attempted.
	315	* This way, we can send non-IP traffic through dsmark
	316	* and don't need yet another qdisc as a bypass.
	317	*/
47bbbb30	318	if (p->mv[index].mask != 0xff \|\| p->mv[index].value)
c76f2a2c	319	pr_warn("%s: unsupported protocol %d\n",
d8b9605d	320	__func__, ntohs(tc_skb_protocol(skb)));
9d127fbd	321	break;
3ff50b79	322	}
af0d1141	323
1da177e4 LT	324	return skb;
	325	}
	326
8e3af978 JP	327	static struct sk_buff dsmark_peek(struct Qdisc sch)
	328	{
	329	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	330
c76f2a2c	331	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
8e3af978 JP	332
	333	return p->q->ops->peek(p->q);
	334	}
	335
e63d7dfd AA	336	static int dsmark_init(struct Qdisc sch, struct nlattr opt,
e63d7dfd AA	337	struct netlink_ext_ack *extack)
1da177e4	338	{
81da99ed	339	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	340	struct nlattr *tb[TCA_DSMARK_MAX + 1];
9d4f97f9	341	int err = -EINVAL;
758cc43c TG	342	u32 default_index = NO_DEFAULT_INDEX;
758cc43c TG	343	u16 indices;
47bbbb30	344	int i;
758cc43c	345
c76f2a2c	346	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
758cc43c	347
cee63723 PM	348	if (!opt)
	349	goto errout;
	350
8d1a77f9	351	err = tcf_block_get(&p->block, &p->filter_list, sch, extack);
6529eaba JP	352	if (err)
	353	return err;
	354
fceb6435	355	err = nla_parse_nested(tb, TCA_DSMARK_MAX, opt, dsmark_policy, NULL);
cee63723	356	if (err < 0)
758cc43c TG	357	goto errout;
758cc43c TG	358
cee63723	359	err = -EINVAL;
1e90474c	360	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	361
5b0ac72b DM	362	if (hweight32(indices) != 1)
758cc43c TG	363	goto errout;
758cc43c TG	364
27a3421e	365	if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c	366	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c	367
47bbbb30 ED	368	if (indices <= DSMARK_EMBEDDED_SZ)
	369	p->mv = p->embedded;
	370	else
	371	p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
	372	if (!p->mv) {
758cc43c TG	373	err = -ENOMEM;
758cc43c TG	374	goto errout;
1da177e4	375	}
47bbbb30 ED	376	for (i = 0; i < indices; i++) {
	377	p->mv[i].mask = 0xff;
	378	p->mv[i].value = 0;
	379	}
758cc43c TG	380	p->indices = indices;
758cc43c TG	381	p->default_index = default_index;
1e90474c	382	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	383
a38a9882 AA	384	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle,
a38a9882 AA	385	NULL);
758cc43c	386	if (p->q == NULL)
1da177e4	387	p->q = &noop_qdisc;
49b49971 JK	388	else
49b49971 JK	389	qdisc_hash_add(p->q, true);
758cc43c	390
c76f2a2c	391	pr_debug("%s: qdisc %p\n", __func__, 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
c76f2a2c	402	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	403	qdisc_reset(p->q);
bdf17661	404	sch->qstats.backlog = 0;
1da177e4 LT	405	sch->q.qlen = 0;
	406	}
	407
1da177e4 LT	408	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	409	{
81da99ed	410	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	411
c76f2a2c	412	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141	413
6529eaba	414	tcf_block_put(p->block);
86bd446b	415	qdisc_put(p->q);
47bbbb30 ED	416	if (p->mv != p->embedded)
47bbbb30 ED	417	kfree(p->mv);
1da177e4 LT	418	}
1da177e4 LT	419
1da177e4	420	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	421	struct sk_buff skb, struct tcmsg tcm)
1da177e4	422	{
81da99ed	423	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	424	struct nlattr *opts = NULL;
1da177e4	425
c76f2a2c	426	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
02f23f09 TG	427
02f23f09 TG	428	if (!dsmark_valid_index(p, cl))
1da177e4	429	return -EINVAL;
02f23f09	430
cc7ec456	431	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
cdc7f8e3	432	tcm->tcm_info = p->q->handle;
02f23f09	433
1e90474c PM	434	opts = nla_nest_start(skb, TCA_OPTIONS);
	435	if (opts == NULL)
	436	goto nla_put_failure;
47bbbb30 ED	437	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) \|\|
47bbbb30 ED	438	nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
1b34ec43	439	goto nla_put_failure;
02f23f09	440
1e90474c	441	return nla_nest_end(skb, opts);
1da177e4	442
1e90474c	443	nla_put_failure:
bc3ed28c TG	444	nla_nest_cancel(skb, opts);
bc3ed28c TG	445	return -EMSGSIZE;
1da177e4 LT	446	}
	447
	448	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	449	{
81da99ed	450	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	451	struct nlattr *opts = NULL;
1da177e4	452
1e90474c PM	453	opts = nla_nest_start(skb, TCA_OPTIONS);
	454	if (opts == NULL)
	455	goto nla_put_failure;
1b34ec43 DM	456	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
1b34ec43 DM	457	goto nla_put_failure;
02f23f09	458
1b34ec43 DM	459	if (p->default_index != NO_DEFAULT_INDEX &&
	460	nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
	461	goto nla_put_failure;
1da177e4	462
1b34ec43 DM	463	if (p->set_tc_index &&
	464	nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
	465	goto nla_put_failure;
02f23f09	466
1e90474c	467	return nla_nest_end(skb, opts);
1da177e4	468
1e90474c	469	nla_put_failure:
bc3ed28c TG	470	nla_nest_cancel(skb, opts);
bc3ed28c TG	471	return -EMSGSIZE;
1da177e4 LT	472	}
1da177e4 LT	473
20fea08b	474	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	475	.graft = dsmark_graft,
1da177e4 LT	476	.leaf = dsmark_leaf,
143976ce	477	.find = dsmark_find,
1da177e4 LT	478	.change = dsmark_change,
	479	.delete = dsmark_delete,
	480	.walk = dsmark_walk,
6529eaba	481	.tcf_block = dsmark_tcf_block,
1da177e4	482	.bind_tcf = dsmark_bind_filter,
143976ce	483	.unbind_tcf = dsmark_unbind_filter,
1da177e4 LT	484	.dump = dsmark_dump_class,
	485	};
	486
20fea08b	487	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	488	.next = NULL,
	489	.cl_ops = &dsmark_class_ops,
	490	.id = "dsmark",
	491	.priv_size = sizeof(struct dsmark_qdisc_data),
	492	.enqueue = dsmark_enqueue,
	493	.dequeue = dsmark_dequeue,
8e3af978	494	.peek = dsmark_peek,
1da177e4 LT	495	.init = dsmark_init,
	496	.reset = dsmark_reset,
	497	.destroy = dsmark_destroy,
	498	.change = NULL,
	499	.dump = dsmark_dump,
	500	.owner = THIS_MODULE,
	501	};
	502
	503	static int __init dsmark_module_init(void)
	504	{
	505	return register_qdisc(&dsmark_qdisc_ops);
	506	}
af0d1141	507
10297b99	508	static void __exit dsmark_module_exit(void)
1da177e4 LT	509	{
	510	unregister_qdisc(&dsmark_qdisc_ops);
	511	}
af0d1141	512
1da177e4 LT	513	module_init(dsmark_module_init)
1da177e4 LT	514	module_exit(dsmark_module_exit)
af0d1141	515
1da177e4	516	MODULE_LICENSE("GPL");