[linux-block.git] / net / psample / psample.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * net/psample/psample.c - Netlink channel for packet sampling
 * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com>
 */

#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/skbuff.h>
#include <linux/module.h>
#include <linux/timekeeping.h>
#include <net/net_namespace.h>
#include <net/sock.h>
#include <net/netlink.h>
#include <net/genetlink.h>
#include <net/psample.h>
#include <linux/spinlock.h>
#include <net/ip_tunnels.h>
#include <net/dst_metadata.h>

#define PSAMPLE_MAX_PACKET_SIZE 0xffff

static LIST_HEAD(psample_groups_list);
static DEFINE_SPINLOCK(psample_groups_lock);

/* multicast groups */
enum psample_nl_multicast_groups {
	PSAMPLE_NL_MCGRP_CONFIG,
	PSAMPLE_NL_MCGRP_SAMPLE,
};

static const struct genl_multicast_group psample_nl_mcgrps[] = {
	[PSAMPLE_NL_MCGRP_CONFIG] = { .name = PSAMPLE_NL_MCGRP_CONFIG_NAME },
	[PSAMPLE_NL_MCGRP_SAMPLE] = { .name = PSAMPLE_NL_MCGRP_SAMPLE_NAME },
};

static struct genl_family psample_nl_family __ro_after_init;

static int psample_group_nl_fill(struct sk_buff *msg,
				 struct psample_group *group,
				 enum psample_command cmd, u32 portid, u32 seq,
				 int flags)
{
	void *hdr;
	int ret;

	hdr = genlmsg_put(msg, portid, seq, &psample_nl_family, flags, cmd);
	if (!hdr)
		return -EMSGSIZE;

	ret = nla_put_u32(msg, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
	if (ret < 0)
		goto error;

	ret = nla_put_u32(msg, PSAMPLE_ATTR_GROUP_REFCOUNT, group->refcount);
	if (ret < 0)
		goto error;

	ret = nla_put_u32(msg, PSAMPLE_ATTR_GROUP_SEQ, group->seq);
	if (ret < 0)
		goto error;

	genlmsg_end(msg, hdr);
	return 0;

error:
	genlmsg_cancel(msg, hdr);
	return -EMSGSIZE;
}

static int psample_nl_cmd_get_group_dumpit(struct sk_buff *msg,
					   struct netlink_callback *cb)
{
	struct psample_group *group;
	int start = cb->args[0];
	int idx = 0;
	int err;

	spin_lock_bh(&psample_groups_lock);
	list_for_each_entry(group, &psample_groups_list, list) {
		if (!net_eq(group->net, sock_net(msg->sk)))
			continue;
		if (idx < start) {
			idx++;
			continue;
		}
		err = psample_group_nl_fill(msg, group, PSAMPLE_CMD_NEW_GROUP,
					    NETLINK_CB(cb->skb).portid,
					    cb->nlh->nlmsg_seq, NLM_F_MULTI);
		if (err)
			break;
		idx++;
	}

	spin_unlock_bh(&psample_groups_lock);
	cb->args[0] = idx;
	return msg->len;
}

static const struct genl_small_ops psample_nl_ops[] = {
	{
		.cmd = PSAMPLE_CMD_GET_GROUP,
		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
		.dumpit = psample_nl_cmd_get_group_dumpit,
		/* can be retrieved by unprivileged users */
	}
};

static struct genl_family psample_nl_family __ro_after_init = {
	.name		= PSAMPLE_GENL_NAME,
	.version	= PSAMPLE_GENL_VERSION,
	.maxattr	= PSAMPLE_ATTR_MAX,
	.netnsok	= true,
	.module		= THIS_MODULE,
	.mcgrps		= psample_nl_mcgrps,
	.small_ops	= psample_nl_ops,
	.n_small_ops	= ARRAY_SIZE(psample_nl_ops),
	.n_mcgrps	= ARRAY_SIZE(psample_nl_mcgrps),
};

static void psample_group_notify(struct psample_group *group,
				 enum psample_command cmd)
{
	struct sk_buff *msg;
	int err;

	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
	if (!msg)
		return;

	err = psample_group_nl_fill(msg, group, cmd, 0, 0, NLM_F_MULTI);
	if (!err)
		genlmsg_multicast_netns(&psample_nl_family, group->net, msg, 0,
					PSAMPLE_NL_MCGRP_CONFIG, GFP_ATOMIC);
	else
		nlmsg_free(msg);
}

static struct psample_group *psample_group_create(struct net *net,
						  u32 group_num)
{
	struct psample_group *group;

	group = kzalloc(sizeof(*group), GFP_ATOMIC);
	if (!group)
		return NULL;

	group->net = net;
	group->group_num = group_num;
	list_add_tail(&group->list, &psample_groups_list);

	psample_group_notify(group, PSAMPLE_CMD_NEW_GROUP);
	return group;
}

static void psample_group_destroy(struct psample_group *group)
{
	psample_group_notify(group, PSAMPLE_CMD_DEL_GROUP);
	list_del(&group->list);
	kfree_rcu(group, rcu);
}

static struct psample_group *
psample_group_lookup(struct net *net, u32 group_num)
{
	struct psample_group *group;

	list_for_each_entry(group, &psample_groups_list, list)
		if ((group->group_num == group_num) && (group->net == net))
			return group;
	return NULL;
}

struct psample_group *psample_group_get(struct net *net, u32 group_num)
{
	struct psample_group *group;

	spin_lock_bh(&psample_groups_lock);

	group = psample_group_lookup(net, group_num);
	if (!group) {
		group = psample_group_create(net, group_num);
		if (!group)
			goto out;
	}
	group->refcount++;

out:
	spin_unlock_bh(&psample_groups_lock);
	return group;
}
EXPORT_SYMBOL_GPL(psample_group_get);

void psample_group_take(struct psample_group *group)
{
	spin_lock_bh(&psample_groups_lock);
	group->refcount++;
	spin_unlock_bh(&psample_groups_lock);
}
EXPORT_SYMBOL_GPL(psample_group_take);

void psample_group_put(struct psample_group *group)
{
	spin_lock_bh(&psample_groups_lock);

	if (--group->refcount == 0)
		psample_group_destroy(group);

	spin_unlock_bh(&psample_groups_lock);
}
EXPORT_SYMBOL_GPL(psample_group_put);

#ifdef CONFIG_INET
static int __psample_ip_tun_to_nlattr(struct sk_buff *skb,
			      struct ip_tunnel_info *tun_info)
{
	unsigned short tun_proto = ip_tunnel_info_af(tun_info);
	const void *tun_opts = ip_tunnel_info_opts(tun_info);
	const struct ip_tunnel_key *tun_key = &tun_info->key;
	int tun_opts_len = tun_info->options_len;

	if (tun_key->tun_flags & TUNNEL_KEY &&
	    nla_put_be64(skb, PSAMPLE_TUNNEL_KEY_ATTR_ID, tun_key->tun_id,
			 PSAMPLE_TUNNEL_KEY_ATTR_PAD))
		return -EMSGSIZE;

	if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE &&
	    nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE))
		return -EMSGSIZE;

	switch (tun_proto) {
	case AF_INET:
		if (tun_key->u.ipv4.src &&
		    nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_SRC,
				    tun_key->u.ipv4.src))
			return -EMSGSIZE;
		if (tun_key->u.ipv4.dst &&
		    nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_DST,
				    tun_key->u.ipv4.dst))
			return -EMSGSIZE;
		break;
	case AF_INET6:
		if (!ipv6_addr_any(&tun_key->u.ipv6.src) &&
		    nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_SRC,
				     &tun_key->u.ipv6.src))
			return -EMSGSIZE;
		if (!ipv6_addr_any(&tun_key->u.ipv6.dst) &&
		    nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_DST,
				     &tun_key->u.ipv6.dst))
			return -EMSGSIZE;
		break;
	}
	if (tun_key->tos &&
	    nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TOS, tun_key->tos))
		return -EMSGSIZE;
	if (nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TTL, tun_key->ttl))
		return -EMSGSIZE;
	if ((tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) &&
	    nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
		return -EMSGSIZE;
	if ((tun_key->tun_flags & TUNNEL_CSUM) &&
	    nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_CSUM))
		return -EMSGSIZE;
	if (tun_key->tp_src &&
	    nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_SRC, tun_key->tp_src))
		return -EMSGSIZE;
	if (tun_key->tp_dst &&
	    nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_DST, tun_key->tp_dst))
		return -EMSGSIZE;
	if ((tun_key->tun_flags & TUNNEL_OAM) &&
	    nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_OAM))
		return -EMSGSIZE;
	if (tun_opts_len) {
		if (tun_key->tun_flags & TUNNEL_GENEVE_OPT &&
		    nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_GENEVE_OPTS,
			    tun_opts_len, tun_opts))
			return -EMSGSIZE;
		else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT &&
			 nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
				 tun_opts_len, tun_opts))
			return -EMSGSIZE;
	}

	return 0;
}

static int psample_ip_tun_to_nlattr(struct sk_buff *skb,
			    struct ip_tunnel_info *tun_info)
{
	struct nlattr *nla;
	int err;

	nla = nla_nest_start_noflag(skb, PSAMPLE_ATTR_TUNNEL);
	if (!nla)
		return -EMSGSIZE;

	err = __psample_ip_tun_to_nlattr(skb, tun_info);
	if (err) {
		nla_nest_cancel(skb, nla);
		return err;
	}

	nla_nest_end(skb, nla);

	return 0;
}

static int psample_tunnel_meta_len(struct ip_tunnel_info *tun_info)
{
	unsigned short tun_proto = ip_tunnel_info_af(tun_info);
	const struct ip_tunnel_key *tun_key = &tun_info->key;
	int tun_opts_len = tun_info->options_len;
	int sum = nla_total_size(0);	/* PSAMPLE_ATTR_TUNNEL */

	if (tun_key->tun_flags & TUNNEL_KEY)
		sum += nla_total_size_64bit(sizeof(u64));

	if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE)
		sum += nla_total_size(0);

	switch (tun_proto) {
	case AF_INET:
		if (tun_key->u.ipv4.src)
			sum += nla_total_size(sizeof(u32));
		if (tun_key->u.ipv4.dst)
			sum += nla_total_size(sizeof(u32));
		break;
	case AF_INET6:
		if (!ipv6_addr_any(&tun_key->u.ipv6.src))
			sum += nla_total_size(sizeof(struct in6_addr));
		if (!ipv6_addr_any(&tun_key->u.ipv6.dst))
			sum += nla_total_size(sizeof(struct in6_addr));
		break;
	}
	if (tun_key->tos)
		sum += nla_total_size(sizeof(u8));
	sum += nla_total_size(sizeof(u8));	/* TTL */
	if (tun_key->tun_flags & TUNNEL_DONT_FRAGMENT)
		sum += nla_total_size(0);
	if (tun_key->tun_flags & TUNNEL_CSUM)
		sum += nla_total_size(0);
	if (tun_key->tp_src)
		sum += nla_total_size(sizeof(u16));
	if (tun_key->tp_dst)
		sum += nla_total_size(sizeof(u16));
	if (tun_key->tun_flags & TUNNEL_OAM)
		sum += nla_total_size(0);
	if (tun_opts_len) {
		if (tun_key->tun_flags & TUNNEL_GENEVE_OPT)
			sum += nla_total_size(tun_opts_len);
		else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT)
			sum += nla_total_size(tun_opts_len);
	}

	return sum;
}
#endif

void psample_sample_packet(struct psample_group *group, struct sk_buff *skb,
			   u32 sample_rate, const struct psample_metadata *md)
{
	ktime_t tstamp = ktime_get_real();
	int out_ifindex = md->out_ifindex;
	int in_ifindex = md->in_ifindex;
	u32 trunc_size = md->trunc_size;
#ifdef CONFIG_INET
	struct ip_tunnel_info *tun_info;
#endif
	struct sk_buff *nl_skb;
	int data_len;
	int meta_len;
	void *data;
	int ret;

	meta_len = (in_ifindex ? nla_total_size(sizeof(u16)) : 0) +
		   (out_ifindex ? nla_total_size(sizeof(u16)) : 0) +
		   (md->out_tc_valid ? nla_total_size(sizeof(u16)) : 0) +
		   (md->out_tc_occ_valid ? nla_total_size_64bit(sizeof(u64)) : 0) +
		   (md->latency_valid ? nla_total_size_64bit(sizeof(u64)) : 0) +
		   nla_total_size(sizeof(u32)) +	/* sample_rate */
		   nla_total_size(sizeof(u32)) +	/* orig_size */
		   nla_total_size(sizeof(u32)) +	/* group_num */
		   nla_total_size(sizeof(u32)) +	/* seq */
		   nla_total_size_64bit(sizeof(u64)) +	/* timestamp */
		   nla_total_size(sizeof(u16));		/* protocol */

#ifdef CONFIG_INET
	tun_info = skb_tunnel_info(skb);
	if (tun_info)
		meta_len += psample_tunnel_meta_len(tun_info);
#endif

	data_len = min(skb->len, trunc_size);
	if (meta_len + nla_total_size(data_len) > PSAMPLE_MAX_PACKET_SIZE)
		data_len = PSAMPLE_MAX_PACKET_SIZE - meta_len - NLA_HDRLEN
			    - NLA_ALIGNTO;

	nl_skb = genlmsg_new(meta_len + nla_total_size(data_len), GFP_ATOMIC);
	if (unlikely(!nl_skb))
		return;

	data = genlmsg_put(nl_skb, 0, 0, &psample_nl_family, 0,
			   PSAMPLE_CMD_SAMPLE);
	if (unlikely(!data))
		goto error;

	if (in_ifindex) {
		ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_IIFINDEX, in_ifindex);
		if (unlikely(ret < 0))
			goto error;
	}

	if (out_ifindex) {
		ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OIFINDEX, out_ifindex);
		if (unlikely(ret < 0))
			goto error;
	}

	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_RATE, sample_rate);
	if (unlikely(ret < 0))
		goto error;

	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_ORIGSIZE, skb->len);
	if (unlikely(ret < 0))
		goto error;

	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
	if (unlikely(ret < 0))
		goto error;

	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_GROUP_SEQ, group->seq++);
	if (unlikely(ret < 0))
		goto error;

	if (md->out_tc_valid) {
		ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OUT_TC, md->out_tc);
		if (unlikely(ret < 0))
			goto error;
	}

	if (md->out_tc_occ_valid) {
		ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_OUT_TC_OCC,
					md->out_tc_occ, PSAMPLE_ATTR_PAD);
		if (unlikely(ret < 0))
			goto error;
	}

	if (md->latency_valid) {
		ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_LATENCY,
					md->latency, PSAMPLE_ATTR_PAD);
		if (unlikely(ret < 0))
			goto error;
	}

	ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_TIMESTAMP,
				ktime_to_ns(tstamp), PSAMPLE_ATTR_PAD);
	if (unlikely(ret < 0))
		goto error;

	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_PROTO,
			  be16_to_cpu(skb->protocol));
	if (unlikely(ret < 0))
		goto error;

	if (data_len) {
		int nla_len = nla_total_size(data_len);
		struct nlattr *nla;

		nla = skb_put(nl_skb, nla_len);
		nla->nla_type = PSAMPLE_ATTR_DATA;
		nla->nla_len = nla_attr_size(data_len);

		if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
			goto error;
	}

#ifdef CONFIG_INET
	if (tun_info) {
		ret = psample_ip_tun_to_nlattr(nl_skb, tun_info);
		if (unlikely(ret < 0))
			goto error;
	}
#endif

	genlmsg_end(nl_skb, data);
	genlmsg_multicast_netns(&psample_nl_family, group->net, nl_skb, 0,
				PSAMPLE_NL_MCGRP_SAMPLE, GFP_ATOMIC);

	return;
error:
	pr_err_ratelimited("Could not create psample log message\n");
	nlmsg_free(nl_skb);
}
EXPORT_SYMBOL_GPL(psample_sample_packet);

static int __init psample_module_init(void)
{
	return genl_register_family(&psample_nl_family);
}

static void __exit psample_module_exit(void)
{
	genl_unregister_family(&psample_nl_family);
}

module_init(psample_module_init);
module_exit(psample_module_exit);

MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
MODULE_DESCRIPTION("netlink channel for packet sampling");
MODULE_LICENSE("GPL v2");
Commit	Line	Data
d2912cb1	1	// SPDX-License-Identifier: GPL-2.0-only
6ae0a628 YG	2	/*
	3	* net/psample/psample.c - Netlink channel for packet sampling
	4	* Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com>
6ae0a628 YG	5	*/
	6
	7	#include <linux/types.h>
	8	#include <linux/kernel.h>
	9	#include <linux/skbuff.h>
	10	#include <linux/module.h>
07e1a580	11	#include <linux/timekeeping.h>
6ae0a628 YG	12	#include <net/net_namespace.h>
	13	#include <net/sock.h>
	14	#include <net/netlink.h>
	15	#include <net/genetlink.h>
	16	#include <net/psample.h>
	17	#include <linux/spinlock.h>
d8bed686 CM	18	#include <net/ip_tunnels.h>
d8bed686 CM	19	#include <net/dst_metadata.h>
6ae0a628 YG	20
	21	#define PSAMPLE_MAX_PACKET_SIZE 0xffff
	22
	23	static LIST_HEAD(psample_groups_list);
	24	static DEFINE_SPINLOCK(psample_groups_lock);
	25
	26	/* multicast groups */
	27	enum psample_nl_multicast_groups {
	28	PSAMPLE_NL_MCGRP_CONFIG,
	29	PSAMPLE_NL_MCGRP_SAMPLE,
	30	};
	31
	32	static const struct genl_multicast_group psample_nl_mcgrps[] = {
	33	[PSAMPLE_NL_MCGRP_CONFIG] = { .name = PSAMPLE_NL_MCGRP_CONFIG_NAME },
	34	[PSAMPLE_NL_MCGRP_SAMPLE] = { .name = PSAMPLE_NL_MCGRP_SAMPLE_NAME },
	35	};
	36
	37	static struct genl_family psample_nl_family __ro_after_init;
	38
	39	static int psample_group_nl_fill(struct sk_buff *msg,
	40	struct psample_group *group,
	41	enum psample_command cmd, u32 portid, u32 seq,
	42	int flags)
	43	{
	44	void *hdr;
	45	int ret;
	46
	47	hdr = genlmsg_put(msg, portid, seq, &psample_nl_family, flags, cmd);
	48	if (!hdr)
	49	return -EMSGSIZE;
	50
	51	ret = nla_put_u32(msg, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
	52	if (ret < 0)
	53	goto error;
	54
	55	ret = nla_put_u32(msg, PSAMPLE_ATTR_GROUP_REFCOUNT, group->refcount);
	56	if (ret < 0)
	57	goto error;
	58
	59	ret = nla_put_u32(msg, PSAMPLE_ATTR_GROUP_SEQ, group->seq);
	60	if (ret < 0)
	61	goto error;
	62
	63	genlmsg_end(msg, hdr);
	64	return 0;
	65
	66	error:
	67	genlmsg_cancel(msg, hdr);
	68	return -EMSGSIZE;
	69	}
	70
	71	static int psample_nl_cmd_get_group_dumpit(struct sk_buff *msg,
	72	struct netlink_callback *cb)
	73	{
	74	struct psample_group *group;
	75	int start = cb->args[0];
	76	int idx = 0;
	77	int err;
	78
4a5da47d	79	spin_lock_bh(&psample_groups_lock);
6ae0a628 YG	80	list_for_each_entry(group, &psample_groups_list, list) {
	81	if (!net_eq(group->net, sock_net(msg->sk)))
	82	continue;
	83	if (idx < start) {
	84	idx++;
	85	continue;
	86	}
	87	err = psample_group_nl_fill(msg, group, PSAMPLE_CMD_NEW_GROUP,
	88	NETLINK_CB(cb->skb).portid,
	89	cb->nlh->nlmsg_seq, NLM_F_MULTI);
	90	if (err)
	91	break;
	92	idx++;
	93	}
	94
4a5da47d	95	spin_unlock_bh(&psample_groups_lock);
6ae0a628 YG	96	cb->args[0] = idx;
	97	return msg->len;
	98	}
	99
66a9b928	100	static const struct genl_small_ops psample_nl_ops[] = {
6ae0a628 YG	101	{
6ae0a628 YG	102	.cmd = PSAMPLE_CMD_GET_GROUP,
ef6243ac	103	.validate = GENL_DONT_VALIDATE_STRICT \| GENL_DONT_VALIDATE_DUMP,
6ae0a628 YG	104	.dumpit = psample_nl_cmd_get_group_dumpit,
	105	/* can be retrieved by unprivileged users */
	106	}
	107	};
	108
	109	static struct genl_family psample_nl_family __ro_after_init = {
	110	.name = PSAMPLE_GENL_NAME,
	111	.version = PSAMPLE_GENL_VERSION,
	112	.maxattr = PSAMPLE_ATTR_MAX,
	113	.netnsok = true,
	114	.module = THIS_MODULE,
	115	.mcgrps = psample_nl_mcgrps,
66a9b928 JK	116	.small_ops = psample_nl_ops,
66a9b928 JK	117	.n_small_ops = ARRAY_SIZE(psample_nl_ops),
6ae0a628 YG	118	.n_mcgrps = ARRAY_SIZE(psample_nl_mcgrps),
	119	};
	120
	121	static void psample_group_notify(struct psample_group *group,
	122	enum psample_command cmd)
	123	{
	124	struct sk_buff *msg;
	125	int err;
	126
	127	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
	128	if (!msg)
	129	return;
	130
	131	err = psample_group_nl_fill(msg, group, cmd, 0, 0, NLM_F_MULTI);
	132	if (!err)
	133	genlmsg_multicast_netns(&psample_nl_family, group->net, msg, 0,
	134	PSAMPLE_NL_MCGRP_CONFIG, GFP_ATOMIC);
	135	else
	136	nlmsg_free(msg);
	137	}
	138
	139	static struct psample_group psample_group_create(struct net net,
	140	u32 group_num)
	141	{
	142	struct psample_group *group;
	143
	144	group = kzalloc(sizeof(*group), GFP_ATOMIC);
	145	if (!group)
	146	return NULL;
	147
	148	group->net = net;
	149	group->group_num = group_num;
	150	list_add_tail(&group->list, &psample_groups_list);
	151
	152	psample_group_notify(group, PSAMPLE_CMD_NEW_GROUP);
	153	return group;
	154	}
	155
	156	static void psample_group_destroy(struct psample_group *group)
	157	{
	158	psample_group_notify(group, PSAMPLE_CMD_DEL_GROUP);
	159	list_del(&group->list);
dbf47a2a	160	kfree_rcu(group, rcu);
6ae0a628 YG	161	}
	162
	163	static struct psample_group *
	164	psample_group_lookup(struct net *net, u32 group_num)
	165	{
	166	struct psample_group *group;
	167
	168	list_for_each_entry(group, &psample_groups_list, list)
	169	if ((group->group_num == group_num) && (group->net == net))
	170	return group;
	171	return NULL;
	172	}
	173
	174	struct psample_group psample_group_get(struct net net, u32 group_num)
	175	{
	176	struct psample_group *group;
	177
4a5da47d	178	spin_lock_bh(&psample_groups_lock);
6ae0a628 YG	179
	180	group = psample_group_lookup(net, group_num);
	181	if (!group) {
	182	group = psample_group_create(net, group_num);
	183	if (!group)
	184	goto out;
	185	}
	186	group->refcount++;
	187
	188	out:
4a5da47d	189	spin_unlock_bh(&psample_groups_lock);
6ae0a628 YG	190	return group;
	191	}
	192	EXPORT_SYMBOL_GPL(psample_group_get);
	193
4a5da47d VB	194	void psample_group_take(struct psample_group *group)
	195	{
	196	spin_lock_bh(&psample_groups_lock);
	197	group->refcount++;
	198	spin_unlock_bh(&psample_groups_lock);
	199	}
	200	EXPORT_SYMBOL_GPL(psample_group_take);
	201
6ae0a628 YG	202	void psample_group_put(struct psample_group *group)
6ae0a628 YG	203	{
4a5da47d	204	spin_lock_bh(&psample_groups_lock);
6ae0a628 YG	205
	206	if (--group->refcount == 0)
	207	psample_group_destroy(group);
	208
4a5da47d	209	spin_unlock_bh(&psample_groups_lock);
6ae0a628 YG	210	}
	211	EXPORT_SYMBOL_GPL(psample_group_put);
	212
07a7f308	213	#ifdef CONFIG_INET
d8bed686 CM	214	static int __psample_ip_tun_to_nlattr(struct sk_buff *skb,
	215	struct ip_tunnel_info *tun_info)
	216	{
	217	unsigned short tun_proto = ip_tunnel_info_af(tun_info);
	218	const void *tun_opts = ip_tunnel_info_opts(tun_info);
	219	const struct ip_tunnel_key *tun_key = &tun_info->key;
	220	int tun_opts_len = tun_info->options_len;
	221
	222	if (tun_key->tun_flags & TUNNEL_KEY &&
	223	nla_put_be64(skb, PSAMPLE_TUNNEL_KEY_ATTR_ID, tun_key->tun_id,
	224	PSAMPLE_TUNNEL_KEY_ATTR_PAD))
	225	return -EMSGSIZE;
	226
	227	if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE &&
	228	nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE))
	229	return -EMSGSIZE;
	230
	231	switch (tun_proto) {
	232	case AF_INET:
	233	if (tun_key->u.ipv4.src &&
	234	nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_SRC,
	235	tun_key->u.ipv4.src))
	236	return -EMSGSIZE;
	237	if (tun_key->u.ipv4.dst &&
	238	nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_DST,
	239	tun_key->u.ipv4.dst))
	240	return -EMSGSIZE;
	241	break;
	242	case AF_INET6:
	243	if (!ipv6_addr_any(&tun_key->u.ipv6.src) &&
	244	nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_SRC,
	245	&tun_key->u.ipv6.src))
	246	return -EMSGSIZE;
	247	if (!ipv6_addr_any(&tun_key->u.ipv6.dst) &&
	248	nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_DST,
	249	&tun_key->u.ipv6.dst))
	250	return -EMSGSIZE;
	251	break;
	252	}
	253	if (tun_key->tos &&
	254	nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TOS, tun_key->tos))
	255	return -EMSGSIZE;
	256	if (nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TTL, tun_key->ttl))
	257	return -EMSGSIZE;
	258	if ((tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) &&
	259	nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
	260	return -EMSGSIZE;
	261	if ((tun_key->tun_flags & TUNNEL_CSUM) &&
	262	nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_CSUM))
	263	return -EMSGSIZE;
	264	if (tun_key->tp_src &&
	265	nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_SRC, tun_key->tp_src))
	266	return -EMSGSIZE;
	267	if (tun_key->tp_dst &&
	268	nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_DST, tun_key->tp_dst))
	269	return -EMSGSIZE;
	270	if ((tun_key->tun_flags & TUNNEL_OAM) &&
	271	nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_OAM))
	272	return -EMSGSIZE;
	273	if (tun_opts_len) {
	274	if (tun_key->tun_flags & TUNNEL_GENEVE_OPT &&
	275	nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_GENEVE_OPTS,
	276	tun_opts_len, tun_opts))
	277	return -EMSGSIZE;
278	else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT &&
279	nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
280	tun_opts_len, tun_opts))
281	return -EMSGSIZE;
282	}
283
284	return 0;
285	}
286
287	static int psample_ip_tun_to_nlattr(struct sk_buff *skb,
288	struct ip_tunnel_info *tun_info)
289	{
290	struct nlattr *nla;
291	int err;
292
293	nla = nla_nest_start_noflag(skb, PSAMPLE_ATTR_TUNNEL);
294	if (!nla)
295	return -EMSGSIZE;
296
297	err = __psample_ip_tun_to_nlattr(skb, tun_info);
298	if (err) {
299	nla_nest_cancel(skb, nla);
300	return err;
301	}
302
303	nla_nest_end(skb, nla);
304
305	return 0;
306	}
307
308	static int psample_tunnel_meta_len(struct ip_tunnel_info *tun_info)
309	{
310	unsigned short tun_proto = ip_tunnel_info_af(tun_info);
311	const struct ip_tunnel_key *tun_key = &tun_info->key;
312	int tun_opts_len = tun_info->options_len;
a93dcaad	313	int sum = nla_total_size(0); /* PSAMPLE_ATTR_TUNNEL */
d8bed686 CM	314
d8bed686 CM	315	if (tun_key->tun_flags & TUNNEL_KEY)
a93dcaad	316	sum += nla_total_size_64bit(sizeof(u64));
d8bed686 CM	317
	318	if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE)
	319	sum += nla_total_size(0);
	320
	321	switch (tun_proto) {
	322	case AF_INET:
	323	if (tun_key->u.ipv4.src)
	324	sum += nla_total_size(sizeof(u32));
	325	if (tun_key->u.ipv4.dst)
	326	sum += nla_total_size(sizeof(u32));
	327	break;
	328	case AF_INET6:
	329	if (!ipv6_addr_any(&tun_key->u.ipv6.src))
	330	sum += nla_total_size(sizeof(struct in6_addr));
	331	if (!ipv6_addr_any(&tun_key->u.ipv6.dst))
	332	sum += nla_total_size(sizeof(struct in6_addr));
	333	break;
	334	}
	335	if (tun_key->tos)
	336	sum += nla_total_size(sizeof(u8));
	337	sum += nla_total_size(sizeof(u8)); /* TTL */
	338	if (tun_key->tun_flags & TUNNEL_DONT_FRAGMENT)
	339	sum += nla_total_size(0);
	340	if (tun_key->tun_flags & TUNNEL_CSUM)
	341	sum += nla_total_size(0);
	342	if (tun_key->tp_src)
	343	sum += nla_total_size(sizeof(u16));
	344	if (tun_key->tp_dst)
	345	sum += nla_total_size(sizeof(u16));
	346	if (tun_key->tun_flags & TUNNEL_OAM)
	347	sum += nla_total_size(0);
	348	if (tun_opts_len) {
	349	if (tun_key->tun_flags & TUNNEL_GENEVE_OPT)
	350	sum += nla_total_size(tun_opts_len);
	351	else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT)
	352	sum += nla_total_size(tun_opts_len);
	353	}
	354
	355	return sum;
	356	}
07a7f308	357	#endif
d8bed686	358
6ae0a628	359	void psample_sample_packet(struct psample_group group, struct sk_buff skb,
a03e99d3	360	u32 sample_rate, const struct psample_metadata *md)
6ae0a628	361	{
07e1a580	362	ktime_t tstamp = ktime_get_real();
a03e99d3 IS	363	int out_ifindex = md->out_ifindex;
	364	int in_ifindex = md->in_ifindex;
	365	u32 trunc_size = md->trunc_size;
07a7f308	366	#ifdef CONFIG_INET
d8bed686	367	struct ip_tunnel_info *tun_info;
07a7f308	368	#endif
6ae0a628 YG	369	struct sk_buff *nl_skb;
	370	int data_len;
	371	int meta_len;
	372	void *data;
	373	int ret;
	374
	375	meta_len = (in_ifindex ? nla_total_size(sizeof(u16)) : 0) +
	376	(out_ifindex ? nla_total_size(sizeof(u16)) : 0) +
07e1a580 IS	377	(md->out_tc_valid ? nla_total_size(sizeof(u16)) : 0) +
	378	(md->out_tc_occ_valid ? nla_total_size_64bit(sizeof(u64)) : 0) +
	379	(md->latency_valid ? nla_total_size_64bit(sizeof(u64)) : 0) +
6ae0a628 YG	380	nla_total_size(sizeof(u32)) + /* sample_rate */
	381	nla_total_size(sizeof(u32)) + /* orig_size */
	382	nla_total_size(sizeof(u32)) + /* group_num */
07e1a580 IS	383	nla_total_size(sizeof(u32)) + /* seq */
	384	nla_total_size_64bit(sizeof(u64)) + /* timestamp */
	385	nla_total_size(sizeof(u16)); /* protocol */
6ae0a628	386
07a7f308	387	#ifdef CONFIG_INET
d8bed686 CM	388	tun_info = skb_tunnel_info(skb);
	389	if (tun_info)
	390	meta_len += psample_tunnel_meta_len(tun_info);
07a7f308	391	#endif
d8bed686	392
6ae0a628 YG	393	data_len = min(skb->len, trunc_size);
	394	if (meta_len + nla_total_size(data_len) > PSAMPLE_MAX_PACKET_SIZE)
	395	data_len = PSAMPLE_MAX_PACKET_SIZE - meta_len - NLA_HDRLEN
	396	- NLA_ALIGNTO;
	397
7eb9d767	398	nl_skb = genlmsg_new(meta_len + nla_total_size(data_len), GFP_ATOMIC);
6ae0a628 YG	399	if (unlikely(!nl_skb))
	400	return;
	401
	402	data = genlmsg_put(nl_skb, 0, 0, &psample_nl_family, 0,
	403	PSAMPLE_CMD_SAMPLE);
	404	if (unlikely(!data))
	405	goto error;
	406
	407	if (in_ifindex) {
	408	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_IIFINDEX, in_ifindex);
	409	if (unlikely(ret < 0))
	410	goto error;
	411	}
	412
	413	if (out_ifindex) {
	414	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OIFINDEX, out_ifindex);
	415	if (unlikely(ret < 0))
	416	goto error;
	417	}
	418
	419	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_RATE, sample_rate);
	420	if (unlikely(ret < 0))
	421	goto error;
	422
	423	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_ORIGSIZE, skb->len);
	424	if (unlikely(ret < 0))
	425	goto error;
	426
	427	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
	428	if (unlikely(ret < 0))
	429	goto error;
	430
	431	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_GROUP_SEQ, group->seq++);
	432	if (unlikely(ret < 0))
	433	goto error;
	434
07e1a580 IS	435	if (md->out_tc_valid) {
	436	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OUT_TC, md->out_tc);
	437	if (unlikely(ret < 0))
	438	goto error;
	439	}
	440
	441	if (md->out_tc_occ_valid) {
	442	ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_OUT_TC_OCC,
	443	md->out_tc_occ, PSAMPLE_ATTR_PAD);
	444	if (unlikely(ret < 0))
	445	goto error;
	446	}
	447
	448	if (md->latency_valid) {
	449	ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_LATENCY,
	450	md->latency, PSAMPLE_ATTR_PAD);
	451	if (unlikely(ret < 0))
	452	goto error;
	453	}
	454
	455	ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_TIMESTAMP,
	456	ktime_to_ns(tstamp), PSAMPLE_ATTR_PAD);
	457	if (unlikely(ret < 0))
	458	goto error;
	459
	460	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_PROTO,
	461	be16_to_cpu(skb->protocol));
	462	if (unlikely(ret < 0))
	463	goto error;
	464
6ae0a628 YG	465	if (data_len) {
	466	int nla_len = nla_total_size(data_len);
	467	struct nlattr *nla;
	468
4df864c1	469	nla = skb_put(nl_skb, nla_len);
6ae0a628 YG	470	nla->nla_type = PSAMPLE_ATTR_DATA;
	471	nla->nla_len = nla_attr_size(data_len);
	472
	473	if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
	474	goto error;
	475	}
	476
07a7f308	477	#ifdef CONFIG_INET
d8bed686 CM	478	if (tun_info) {
	479	ret = psample_ip_tun_to_nlattr(nl_skb, tun_info);
	480	if (unlikely(ret < 0))
	481	goto error;
	482	}
07a7f308	483	#endif
d8bed686	484
6ae0a628 YG	485	genlmsg_end(nl_skb, data);
	486	genlmsg_multicast_netns(&psample_nl_family, group->net, nl_skb, 0,
	487	PSAMPLE_NL_MCGRP_SAMPLE, GFP_ATOMIC);
	488
	489	return;
	490	error:
	491	pr_err_ratelimited("Could not create psample log message\n");
	492	nlmsg_free(nl_skb);
	493	}
	494	EXPORT_SYMBOL_GPL(psample_sample_packet);
	495
	496	static int __init psample_module_init(void)
	497	{
	498	return genl_register_family(&psample_nl_family);
	499	}
	500
	501	static void __exit psample_module_exit(void)
	502	{
	503	genl_unregister_family(&psample_nl_family);
	504	}
	505
	506	module_init(psample_module_init);
	507	module_exit(psample_module_exit);
	508
f1fd20c3	509	MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
6ae0a628 YG	510	MODULE_DESCRIPTION("netlink channel for packet sampling");
6ae0a628 YG	511	MODULE_LICENSE("GPL v2");