[linux-2.6-block.git] / net / sched / sch_dsmark.c

// SPDX-License-Identifier: GPL-2.0-only
/* 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_deprecated(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)
{
	unsigned int len = qdisc_pkt_len(skb);
	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;
	}

	sch->qstats.backlog += len;
	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_deprecated(tb, TCA_DSMARK_MAX, opt,
					  dsmark_policy, NULL);
	if (err < 0)
		goto errout;

	err = -EINVAL;
	if (!tb[TCA_DSMARK_INDICES])
		goto errout;
	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_noflag(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_noflag(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
09c434b8	1	// SPDX-License-Identifier: GPL-2.0-only
1da177e4 LT	2	/* net/sched/sch_dsmark.c - Differentiated Services field marker */
	3
	4	/* Written 1998-2000 by Werner Almesberger, EPFL ICA */
	5
	6
1da177e4 LT	7	#include <linux/module.h>
1da177e4 LT	8	#include <linux/init.h>
5a0e3ad6	9	#include <linux/slab.h>
1da177e4 LT	10	#include <linux/types.h>
	11	#include <linux/string.h>
	12	#include <linux/errno.h>
	13	#include <linux/skbuff.h>
1da177e4	14	#include <linux/rtnetlink.h>
5b0ac72b	15	#include <linux/bitops.h>
1da177e4	16	#include <net/pkt_sched.h>
cf1facda	17	#include <net/pkt_cls.h>
1da177e4 LT	18	#include <net/dsfield.h>
	19	#include <net/inet_ecn.h>
	20	#include <asm/byteorder.h>
	21
1da177e4 LT	22	/*
	23	* classid class marking
	24	* ------- ----- -------
	25	* n/a 0 n/a
	26	* x:0 1 use entry [0]
	27	* ... ... ...
	28	* x:y y>0 y+1 use entry [y]
	29	* ... ... ...
	30	* x:indices-1 indices use entry [indices-1]
	31	* ... ... ...
	32	* x:y y+1 use entry [y & (indices-1)]
	33	* ... ... ...
	34	* 0xffff 0x10000 use entry [indices-1]
	35	*/
	36
	37
	38	#define NO_DEFAULT_INDEX (1 << 16)
	39
47bbbb30 ED	40	struct mask_value {
	41	u8 mask;
	42	u8 value;
	43	};
	44
1da177e4 LT	45	struct dsmark_qdisc_data {
1da177e4 LT	46	struct Qdisc *q;
25d8c0d5	47	struct tcf_proto __rcu *filter_list;
6529eaba	48	struct tcf_block *block;
47bbbb30	49	struct mask_value *mv;
af0d1141	50	u16 indices;
47bbbb30	51	u8 set_tc_index;
af0d1141	52	u32 default_index; /* index range is 0...0xffff */
47bbbb30 ED	53	#define DSMARK_EMBEDDED_SZ 16
47bbbb30 ED	54	struct mask_value embedded[DSMARK_EMBEDDED_SZ];
1da177e4 LT	55	};
1da177e4 LT	56
758cc43c TG	57	static inline int dsmark_valid_index(struct dsmark_qdisc_data *p, u16 index)
758cc43c TG	58	{
17569fae	59	return index <= p->indices && index > 0;
758cc43c	60	}
1da177e4 LT	61
	62	/* ------------------------- Class/flow operations ------------------------- */
	63
af0d1141	64	static int dsmark_graft(struct Qdisc *sch, unsigned long arg,
653d6fd6 AA	65	struct Qdisc new, struct Qdisc *old,
653d6fd6 AA	66	struct netlink_ext_ack *extack)
1da177e4	67	{
81da99ed	68	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	69
c76f2a2c YY	70	pr_debug("%s(sch %p,[qdisc %p],new %p,old %p)\n",
c76f2a2c YY	71	__func__, sch, p, new, old);
486b53e5 TG	72
486b53e5 TG	73	if (new == NULL) {
3511c913	74	new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
a38a9882	75	sch->handle, NULL);
486b53e5 TG	76	if (new == NULL)
	77	new = &noop_qdisc;
	78	}
	79
86a7996c	80	*old = qdisc_replace(sch, new, &p->q);
10297b99	81	return 0;
1da177e4 LT	82	}
1da177e4 LT	83
1da177e4 LT	84	static struct Qdisc dsmark_leaf(struct Qdisc sch, unsigned long arg)
1da177e4 LT	85	{
81da99ed SH	86	struct dsmark_qdisc_data *p = qdisc_priv(sch);
81da99ed SH	87	return p->q;
1da177e4 LT	88	}
1da177e4 LT	89
143976ce	90	static unsigned long dsmark_find(struct Qdisc *sch, u32 classid)
1da177e4	91	{
af0d1141	92	return TC_H_MIN(classid) + 1;
1da177e4 LT	93	}
1da177e4 LT	94
1da177e4	95	static unsigned long dsmark_bind_filter(struct Qdisc *sch,
af0d1141	96	unsigned long parent, u32 classid)
1da177e4	97	{
143976ce WC	98	pr_debug("%s(sch %p,[qdisc %p],classid %x)\n",
	99	__func__, sch, qdisc_priv(sch), classid);
	100
	101	return dsmark_find(sch, classid);
1da177e4 LT	102	}
1da177e4 LT	103
143976ce	104	static void dsmark_unbind_filter(struct Qdisc *sch, unsigned long cl)
1da177e4 LT	105	{
	106	}
	107
27a3421e PM	108	static const struct nla_policy dsmark_policy[TCA_DSMARK_MAX + 1] = {
	109	[TCA_DSMARK_INDICES] = { .type = NLA_U16 },
	110	[TCA_DSMARK_DEFAULT_INDEX] = { .type = NLA_U16 },
	111	[TCA_DSMARK_SET_TC_INDEX] = { .type = NLA_FLAG },
	112	[TCA_DSMARK_MASK] = { .type = NLA_U8 },
	113	[TCA_DSMARK_VALUE] = { .type = NLA_U8 },
	114	};
	115
1da177e4	116	static int dsmark_change(struct Qdisc *sch, u32 classid, u32 parent,
793d81d6 AA	117	struct nlattr *tca, unsigned long arg,
793d81d6 AA	118	struct netlink_ext_ack *extack)
1da177e4	119	{
81da99ed	120	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c PM	121	struct nlattr *opt = tca[TCA_OPTIONS];
1e90474c PM	122	struct nlattr *tb[TCA_DSMARK_MAX + 1];
758cc43c	123	int err = -EINVAL;
1da177e4	124
c76f2a2c YY	125	pr_debug("%s(sch %p,[qdisc %p],classid %x,parent %x), arg 0x%lx\n",
c76f2a2c YY	126	__func__, sch, p, classid, parent, *arg);
758cc43c TG	127
	128	if (!dsmark_valid_index(p, *arg)) {
	129	err = -ENOENT;
1e90474c	130	goto errout;
1da177e4	131	}
1da177e4	132
cee63723	133	if (!opt)
1e90474c	134	goto errout;
758cc43c	135
8cb08174 JB	136	err = nla_parse_nested_deprecated(tb, TCA_DSMARK_MAX, opt,
8cb08174 JB	137	dsmark_policy, NULL);
cee63723	138	if (err < 0)
27a3421e	139	goto errout;
cee63723	140
27a3421e	141	if (tb[TCA_DSMARK_VALUE])
47bbbb30	142	p->mv[*arg - 1].value = nla_get_u8(tb[TCA_DSMARK_VALUE]);
10297b99	143
1e90474c	144	if (tb[TCA_DSMARK_MASK])
47bbbb30	145	p->mv[*arg - 1].mask = nla_get_u8(tb[TCA_DSMARK_MASK]);
758cc43c TG	146
	147	err = 0;
	148
1e90474c	149	errout:
758cc43c TG	150	return err;
758cc43c TG	151	}
1da177e4	152
af0d1141	153	static int dsmark_delete(struct Qdisc *sch, unsigned long arg)
1da177e4	154	{
81da99ed	155	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	156
af0d1141	157	if (!dsmark_valid_index(p, arg))
1da177e4	158	return -EINVAL;
10297b99	159
47bbbb30 ED	160	p->mv[arg - 1].mask = 0xff;
47bbbb30 ED	161	p->mv[arg - 1].value = 0;
af0d1141	162
1da177e4 LT	163	return 0;
	164	}
	165
9d127fbd	166	static void dsmark_walk(struct Qdisc sch, struct qdisc_walker walker)
1da177e4	167	{
81da99ed	168	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4 LT	169	int i;
1da177e4 LT	170
c76f2a2c YY	171	pr_debug("%s(sch %p,[qdisc %p],walker %p)\n",
c76f2a2c YY	172	__func__, sch, p, walker);
af0d1141	173
1da177e4 LT	174	if (walker->stop)
1da177e4 LT	175	return;
af0d1141	176
1da177e4	177	for (i = 0; i < p->indices; i++) {
47bbbb30	178	if (p->mv[i].mask == 0xff && !p->mv[i].value)
0451eb07	179	goto ignore;
1da177e4	180	if (walker->count >= walker->skip) {
cc7ec456	181	if (walker->fn(sch, i + 1, walker) < 0) {
1da177e4 LT	182	walker->stop = 1;
	183	break;
	184	}
	185	}
10297b99	186	ignore:
0451eb07	187	walker->count++;
10297b99	188	}
1da177e4 LT	189	}
1da177e4 LT	190
cbaacc4e AA	191	static struct tcf_block dsmark_tcf_block(struct Qdisc sch, unsigned long cl,
cbaacc4e AA	192	struct netlink_ext_ack *extack)
1da177e4	193	{
81da99ed	194	struct dsmark_qdisc_data *p = qdisc_priv(sch);
6529eaba JP	195
6529eaba JP	196	return p->block;
1da177e4 LT	197	}
1da177e4 LT	198
1da177e4 LT	199	/* --------------------------- Qdisc operations ---------------------------- */
1da177e4 LT	200
520ac30f ED	201	static int dsmark_enqueue(struct sk_buff skb, struct Qdisc sch,
520ac30f ED	202	struct sk_buff **to_free)
1da177e4	203	{
f6bab199	204	unsigned int len = qdisc_pkt_len(skb);
81da99ed	205	struct dsmark_qdisc_data *p = qdisc_priv(sch);
af0d1141 TG	206	int err;
af0d1141 TG	207
c76f2a2c	208	pr_debug("%s(skb %p,sch %p,[qdisc %p])\n", __func__, skb, sch, p);
1da177e4	209
1da177e4	210	if (p->set_tc_index) {
aea92fb2 ED	211	int wlen = skb_network_offset(skb);
aea92fb2 ED	212
d8b9605d	213	switch (tc_skb_protocol(skb)) {
60678040	214	case htons(ETH_P_IP):
aea92fb2 ED	215	wlen += sizeof(struct iphdr);
	216	if (!pskb_may_pull(skb, wlen) \|\|
	217	skb_try_make_writable(skb, wlen))
9d127fbd	218	goto drop;
4c30719f	219
9d127fbd SH	220	skb->tc_index = ipv4_get_dsfield(ip_hdr(skb))
	221	& ~INET_ECN_MASK;
	222	break;
4c30719f	223
60678040	224	case htons(ETH_P_IPV6):
aea92fb2 ED	225	wlen += sizeof(struct ipv6hdr);
	226	if (!pskb_may_pull(skb, wlen) \|\|
	227	skb_try_make_writable(skb, wlen))
9d127fbd	228	goto drop;
4c30719f	229
9d127fbd SH	230	skb->tc_index = ipv6_get_dsfield(ipv6_hdr(skb))
	231	& ~INET_ECN_MASK;
	232	break;
	233	default:
	234	skb->tc_index = 0;
	235	break;
3ff50b79	236	}
1da177e4	237	}
af0d1141 TG	238
af0d1141 TG	239	if (TC_H_MAJ(skb->priority) == sch->handle)
1da177e4	240	skb->tc_index = TC_H_MIN(skb->priority);
af0d1141 TG	241	else {
af0d1141 TG	242	struct tcf_result res;
25d8c0d5	243	struct tcf_proto *fl = rcu_dereference_bh(p->filter_list);
87d83093	244	int result = tcf_classify(skb, fl, &res, false);
af0d1141	245
81da99ed	246	pr_debug("result %d class 0x%04x\n", result, res.classid);
af0d1141	247
1da177e4	248	switch (result) {
f6853e2d PM	249	#ifdef CONFIG_NET_CLS_ACT
	250	case TC_ACT_QUEUED:
	251	case TC_ACT_STOLEN:
e25ea21f	252	case TC_ACT_TRAP:
520ac30f	253	__qdisc_drop(skb, to_free);
378a2f09	254	return NET_XMIT_SUCCESS \| __NET_XMIT_STOLEN;
4c30719f	255
f6853e2d	256	case TC_ACT_SHOT:
4c30719f	257	goto drop;
1da177e4	258	#endif
c3bc7cff	259	case TC_ACT_OK:
f6853e2d PM	260	skb->tc_index = TC_H_MIN(res.classid);
f6853e2d PM	261	break;
4c30719f	262
f6853e2d PM	263	default:
	264	if (p->default_index != NO_DEFAULT_INDEX)
	265	skb->tc_index = p->default_index;
	266	break;
3ff50b79	267	}
1da177e4	268	}
1da177e4	269
520ac30f	270	err = qdisc_enqueue(skb, p->q, to_free);
af0d1141	271	if (err != NET_XMIT_SUCCESS) {
378a2f09	272	if (net_xmit_drop_count(err))
25331d6c	273	qdisc_qstats_drop(sch);
af0d1141	274	return err;
1da177e4	275	}
af0d1141	276
f6bab199	277	sch->qstats.backlog += len;
1da177e4	278	sch->q.qlen++;
1da177e4	279
af0d1141	280	return NET_XMIT_SUCCESS;
4c30719f SH	281
4c30719f SH	282	drop:
520ac30f	283	qdisc_drop(skb, sch, to_free);
c27f339a	284	return NET_XMIT_SUCCESS \| __NET_XMIT_BYPASS;
af0d1141	285	}
1da177e4 LT	286
	287	static struct sk_buff dsmark_dequeue(struct Qdisc sch)
	288	{
81da99ed	289	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	290	struct sk_buff *skb;
af0d1141 TG	291	u32 index;
af0d1141 TG	292
c76f2a2c	293	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	294
f8b33d8e	295	skb = qdisc_dequeue_peeked(p->q);
af0d1141	296	if (skb == NULL)
1da177e4	297	return NULL;
af0d1141	298
9190b3b3	299	qdisc_bstats_update(sch, skb);
bdf17661	300	qdisc_qstats_backlog_dec(sch, skb);
1da177e4	301	sch->q.qlen--;
af0d1141 TG	302
af0d1141 TG	303	index = skb->tc_index & (p->indices - 1);
81da99ed	304	pr_debug("index %d->%d\n", skb->tc_index, index);
af0d1141	305
d8b9605d	306	switch (tc_skb_protocol(skb)) {
60678040	307	case htons(ETH_P_IP):
47bbbb30 ED	308	ipv4_change_dsfield(ip_hdr(skb), p->mv[index].mask,
47bbbb30 ED	309	p->mv[index].value);
1da177e4	310	break;
60678040	311	case htons(ETH_P_IPV6):
47bbbb30 ED	312	ipv6_change_dsfield(ipv6_hdr(skb), p->mv[index].mask,
47bbbb30 ED	313	p->mv[index].value);
1da177e4	314	break;
9d127fbd SH	315	default:
	316	/*
	317	* Only complain if a change was actually attempted.
	318	* This way, we can send non-IP traffic through dsmark
	319	* and don't need yet another qdisc as a bypass.
	320	*/
47bbbb30	321	if (p->mv[index].mask != 0xff \|\| p->mv[index].value)
c76f2a2c	322	pr_warn("%s: unsupported protocol %d\n",
d8b9605d	323	__func__, ntohs(tc_skb_protocol(skb)));
9d127fbd	324	break;
3ff50b79	325	}
af0d1141	326
1da177e4 LT	327	return skb;
	328	}
	329
8e3af978 JP	330	static struct sk_buff dsmark_peek(struct Qdisc sch)
	331	{
	332	struct dsmark_qdisc_data *p = qdisc_priv(sch);
	333
c76f2a2c	334	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
8e3af978 JP	335
	336	return p->q->ops->peek(p->q);
	337	}
	338
e63d7dfd AA	339	static int dsmark_init(struct Qdisc sch, struct nlattr opt,
e63d7dfd AA	340	struct netlink_ext_ack *extack)
1da177e4	341	{
81da99ed	342	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	343	struct nlattr *tb[TCA_DSMARK_MAX + 1];
9d4f97f9	344	int err = -EINVAL;
758cc43c TG	345	u32 default_index = NO_DEFAULT_INDEX;
758cc43c TG	346	u16 indices;
47bbbb30	347	int i;
758cc43c	348
c76f2a2c	349	pr_debug("%s(sch %p,[qdisc %p],opt %p)\n", __func__, sch, p, opt);
758cc43c	350
cee63723 PM	351	if (!opt)
	352	goto errout;
	353
8d1a77f9	354	err = tcf_block_get(&p->block, &p->filter_list, sch, extack);
6529eaba JP	355	if (err)
	356	return err;
	357
8cb08174 JB	358	err = nla_parse_nested_deprecated(tb, TCA_DSMARK_MAX, opt,
8cb08174 JB	359	dsmark_policy, NULL);
cee63723	360	if (err < 0)
758cc43c TG	361	goto errout;
758cc43c TG	362
cee63723	363	err = -EINVAL;
474f0813 ED	364	if (!tb[TCA_DSMARK_INDICES])
474f0813 ED	365	goto errout;
1e90474c	366	indices = nla_get_u16(tb[TCA_DSMARK_INDICES]);
5b0ac72b DM	367
5b0ac72b DM	368	if (hweight32(indices) != 1)
758cc43c TG	369	goto errout;
758cc43c TG	370
27a3421e	371	if (tb[TCA_DSMARK_DEFAULT_INDEX])
1e90474c	372	default_index = nla_get_u16(tb[TCA_DSMARK_DEFAULT_INDEX]);
758cc43c	373
47bbbb30 ED	374	if (indices <= DSMARK_EMBEDDED_SZ)
	375	p->mv = p->embedded;
	376	else
	377	p->mv = kmalloc_array(indices, sizeof(*p->mv), GFP_KERNEL);
	378	if (!p->mv) {
758cc43c TG	379	err = -ENOMEM;
758cc43c TG	380	goto errout;
1da177e4	381	}
47bbbb30 ED	382	for (i = 0; i < indices; i++) {
	383	p->mv[i].mask = 0xff;
	384	p->mv[i].value = 0;
	385	}
758cc43c TG	386	p->indices = indices;
758cc43c TG	387	p->default_index = default_index;
1e90474c	388	p->set_tc_index = nla_get_flag(tb[TCA_DSMARK_SET_TC_INDEX]);
758cc43c	389
a38a9882 AA	390	p->q = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, sch->handle,
a38a9882 AA	391	NULL);
758cc43c	392	if (p->q == NULL)
1da177e4	393	p->q = &noop_qdisc;
49b49971 JK	394	else
49b49971 JK	395	qdisc_hash_add(p->q, true);
758cc43c	396
c76f2a2c	397	pr_debug("%s: qdisc %p\n", __func__, p->q);
758cc43c TG	398
	399	err = 0;
	400	errout:
758cc43c	401	return err;
1da177e4 LT	402	}
1da177e4 LT	403
1da177e4 LT	404	static void dsmark_reset(struct Qdisc *sch)
1da177e4 LT	405	{
81da99ed	406	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	407
c76f2a2c	408	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
1da177e4	409	qdisc_reset(p->q);
bdf17661	410	sch->qstats.backlog = 0;
1da177e4 LT	411	sch->q.qlen = 0;
	412	}
	413
1da177e4 LT	414	static void dsmark_destroy(struct Qdisc *sch)
1da177e4 LT	415	{
81da99ed	416	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1da177e4	417
c76f2a2c	418	pr_debug("%s(sch %p,[qdisc %p])\n", __func__, sch, p);
af0d1141	419
6529eaba	420	tcf_block_put(p->block);
86bd446b	421	qdisc_put(p->q);
47bbbb30 ED	422	if (p->mv != p->embedded)
47bbbb30 ED	423	kfree(p->mv);
1da177e4 LT	424	}
1da177e4 LT	425
1da177e4	426	static int dsmark_dump_class(struct Qdisc *sch, unsigned long cl,
02f23f09	427	struct sk_buff skb, struct tcmsg tcm)
1da177e4	428	{
81da99ed	429	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	430	struct nlattr *opts = NULL;
1da177e4	431
c76f2a2c	432	pr_debug("%s(sch %p,[qdisc %p],class %ld\n", __func__, sch, p, cl);
02f23f09 TG	433
02f23f09 TG	434	if (!dsmark_valid_index(p, cl))
1da177e4	435	return -EINVAL;
02f23f09	436
cc7ec456	437	tcm->tcm_handle = TC_H_MAKE(TC_H_MAJ(sch->handle), cl - 1);
cdc7f8e3	438	tcm->tcm_info = p->q->handle;
02f23f09	439
ae0be8de	440	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
1e90474c PM	441	if (opts == NULL)
1e90474c PM	442	goto nla_put_failure;
47bbbb30 ED	443	if (nla_put_u8(skb, TCA_DSMARK_MASK, p->mv[cl - 1].mask) \|\|
47bbbb30 ED	444	nla_put_u8(skb, TCA_DSMARK_VALUE, p->mv[cl - 1].value))
1b34ec43	445	goto nla_put_failure;
02f23f09	446
1e90474c	447	return nla_nest_end(skb, opts);
1da177e4	448
1e90474c	449	nla_put_failure:
bc3ed28c TG	450	nla_nest_cancel(skb, opts);
bc3ed28c TG	451	return -EMSGSIZE;
1da177e4 LT	452	}
	453
	454	static int dsmark_dump(struct Qdisc sch, struct sk_buff skb)
	455	{
81da99ed	456	struct dsmark_qdisc_data *p = qdisc_priv(sch);
1e90474c	457	struct nlattr *opts = NULL;
1da177e4	458
ae0be8de	459	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
1e90474c PM	460	if (opts == NULL)
1e90474c PM	461	goto nla_put_failure;
1b34ec43 DM	462	if (nla_put_u16(skb, TCA_DSMARK_INDICES, p->indices))
1b34ec43 DM	463	goto nla_put_failure;
02f23f09	464
1b34ec43 DM	465	if (p->default_index != NO_DEFAULT_INDEX &&
	466	nla_put_u16(skb, TCA_DSMARK_DEFAULT_INDEX, p->default_index))
	467	goto nla_put_failure;
1da177e4	468
1b34ec43 DM	469	if (p->set_tc_index &&
	470	nla_put_flag(skb, TCA_DSMARK_SET_TC_INDEX))
	471	goto nla_put_failure;
02f23f09	472
1e90474c	473	return nla_nest_end(skb, opts);
1da177e4	474
1e90474c	475	nla_put_failure:
bc3ed28c TG	476	nla_nest_cancel(skb, opts);
bc3ed28c TG	477	return -EMSGSIZE;
1da177e4 LT	478	}
1da177e4 LT	479
20fea08b	480	static const struct Qdisc_class_ops dsmark_class_ops = {
1da177e4 LT	481	.graft = dsmark_graft,
1da177e4 LT	482	.leaf = dsmark_leaf,
143976ce	483	.find = dsmark_find,
1da177e4 LT	484	.change = dsmark_change,
	485	.delete = dsmark_delete,
	486	.walk = dsmark_walk,
6529eaba	487	.tcf_block = dsmark_tcf_block,
1da177e4	488	.bind_tcf = dsmark_bind_filter,
143976ce	489	.unbind_tcf = dsmark_unbind_filter,
1da177e4 LT	490	.dump = dsmark_dump_class,
	491	};
	492
20fea08b	493	static struct Qdisc_ops dsmark_qdisc_ops __read_mostly = {
1da177e4 LT	494	.next = NULL,
	495	.cl_ops = &dsmark_class_ops,
	496	.id = "dsmark",
	497	.priv_size = sizeof(struct dsmark_qdisc_data),
	498	.enqueue = dsmark_enqueue,
	499	.dequeue = dsmark_dequeue,
8e3af978	500	.peek = dsmark_peek,
1da177e4 LT	501	.init = dsmark_init,
	502	.reset = dsmark_reset,
	503	.destroy = dsmark_destroy,
	504	.change = NULL,
	505	.dump = dsmark_dump,
	506	.owner = THIS_MODULE,
	507	};
	508
	509	static int __init dsmark_module_init(void)
	510	{
	511	return register_qdisc(&dsmark_qdisc_ops);
	512	}
af0d1141	513
10297b99	514	static void __exit dsmark_module_exit(void)
1da177e4 LT	515	{
	516	unregister_qdisc(&dsmark_qdisc_ops);
	517	}
af0d1141	518
1da177e4 LT	519	module_init(dsmark_module_init)
1da177e4 LT	520	module_exit(dsmark_module_exit)
af0d1141	521
1da177e4	522	MODULE_LICENSE("GPL");