[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),
	.resv_start_op	= PSAMPLE_CMD_GET_GROUP + 1,
	.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),
9c5d03d3	118	.resv_start_op = PSAMPLE_CMD_GET_GROUP + 1,
6ae0a628 YG	119	.n_mcgrps = ARRAY_SIZE(psample_nl_mcgrps),
	120	};
	121
	122	static void psample_group_notify(struct psample_group *group,
	123	enum psample_command cmd)
	124	{
	125	struct sk_buff *msg;
	126	int err;
	127
	128	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
	129	if (!msg)
	130	return;
	131
	132	err = psample_group_nl_fill(msg, group, cmd, 0, 0, NLM_F_MULTI);
	133	if (!err)
	134	genlmsg_multicast_netns(&psample_nl_family, group->net, msg, 0,
	135	PSAMPLE_NL_MCGRP_CONFIG, GFP_ATOMIC);
	136	else
	137	nlmsg_free(msg);
	138	}
	139
	140	static struct psample_group psample_group_create(struct net net,
	141	u32 group_num)
	142	{
	143	struct psample_group *group;
	144
	145	group = kzalloc(sizeof(*group), GFP_ATOMIC);
	146	if (!group)
	147	return NULL;
	148
	149	group->net = net;
	150	group->group_num = group_num;
	151	list_add_tail(&group->list, &psample_groups_list);
	152
	153	psample_group_notify(group, PSAMPLE_CMD_NEW_GROUP);
	154	return group;
	155	}
	156
	157	static void psample_group_destroy(struct psample_group *group)
	158	{
	159	psample_group_notify(group, PSAMPLE_CMD_DEL_GROUP);
	160	list_del(&group->list);
dbf47a2a	161	kfree_rcu(group, rcu);
6ae0a628 YG	162	}
	163
	164	static struct psample_group *
	165	psample_group_lookup(struct net *net, u32 group_num)
	166	{
	167	struct psample_group *group;
	168
	169	list_for_each_entry(group, &psample_groups_list, list)
	170	if ((group->group_num == group_num) && (group->net == net))
	171	return group;
	172	return NULL;
	173	}
	174
	175	struct psample_group psample_group_get(struct net net, u32 group_num)
	176	{
	177	struct psample_group *group;
	178
4a5da47d	179	spin_lock_bh(&psample_groups_lock);
6ae0a628 YG	180
	181	group = psample_group_lookup(net, group_num);
	182	if (!group) {
	183	group = psample_group_create(net, group_num);
	184	if (!group)
	185	goto out;
	186	}
	187	group->refcount++;
	188
	189	out:
4a5da47d	190	spin_unlock_bh(&psample_groups_lock);
6ae0a628 YG	191	return group;
	192	}
	193	EXPORT_SYMBOL_GPL(psample_group_get);
	194
4a5da47d VB	195	void psample_group_take(struct psample_group *group)
	196	{
	197	spin_lock_bh(&psample_groups_lock);
	198	group->refcount++;
	199	spin_unlock_bh(&psample_groups_lock);
	200	}
	201	EXPORT_SYMBOL_GPL(psample_group_take);
	202
6ae0a628 YG	203	void psample_group_put(struct psample_group *group)
6ae0a628 YG	204	{
4a5da47d	205	spin_lock_bh(&psample_groups_lock);
6ae0a628 YG	206
	207	if (--group->refcount == 0)
	208	psample_group_destroy(group);
	209
4a5da47d	210	spin_unlock_bh(&psample_groups_lock);
6ae0a628 YG	211	}
	212	EXPORT_SYMBOL_GPL(psample_group_put);
	213
07a7f308	214	#ifdef CONFIG_INET
d8bed686 CM	215	static int __psample_ip_tun_to_nlattr(struct sk_buff *skb,
	216	struct ip_tunnel_info *tun_info)
	217	{
	218	unsigned short tun_proto = ip_tunnel_info_af(tun_info);
	219	const void *tun_opts = ip_tunnel_info_opts(tun_info);
	220	const struct ip_tunnel_key *tun_key = &tun_info->key;
	221	int tun_opts_len = tun_info->options_len;
	222
	223	if (tun_key->tun_flags & TUNNEL_KEY &&
	224	nla_put_be64(skb, PSAMPLE_TUNNEL_KEY_ATTR_ID, tun_key->tun_id,
	225	PSAMPLE_TUNNEL_KEY_ATTR_PAD))
	226	return -EMSGSIZE;
	227
	228	if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE &&
	229	nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_INFO_BRIDGE))
	230	return -EMSGSIZE;
	231
	232	switch (tun_proto) {
	233	case AF_INET:
	234	if (tun_key->u.ipv4.src &&
	235	nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_SRC,
	236	tun_key->u.ipv4.src))
	237	return -EMSGSIZE;
	238	if (tun_key->u.ipv4.dst &&
	239	nla_put_in_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV4_DST,
	240	tun_key->u.ipv4.dst))
	241	return -EMSGSIZE;
	242	break;
	243	case AF_INET6:
	244	if (!ipv6_addr_any(&tun_key->u.ipv6.src) &&
	245	nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_SRC,
	246	&tun_key->u.ipv6.src))
	247	return -EMSGSIZE;
	248	if (!ipv6_addr_any(&tun_key->u.ipv6.dst) &&
	249	nla_put_in6_addr(skb, PSAMPLE_TUNNEL_KEY_ATTR_IPV6_DST,
	250	&tun_key->u.ipv6.dst))
	251	return -EMSGSIZE;
	252	break;
	253	}
	254	if (tun_key->tos &&
	255	nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TOS, tun_key->tos))
	256	return -EMSGSIZE;
	257	if (nla_put_u8(skb, PSAMPLE_TUNNEL_KEY_ATTR_TTL, tun_key->ttl))
	258	return -EMSGSIZE;
	259	if ((tun_key->tun_flags & TUNNEL_DONT_FRAGMENT) &&
	260	nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_DONT_FRAGMENT))
	261	return -EMSGSIZE;
	262	if ((tun_key->tun_flags & TUNNEL_CSUM) &&
	263	nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_CSUM))
	264	return -EMSGSIZE;
	265	if (tun_key->tp_src &&
	266	nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_SRC, tun_key->tp_src))
	267	return -EMSGSIZE;
	268	if (tun_key->tp_dst &&
	269	nla_put_be16(skb, PSAMPLE_TUNNEL_KEY_ATTR_TP_DST, tun_key->tp_dst))
	270	return -EMSGSIZE;
	271	if ((tun_key->tun_flags & TUNNEL_OAM) &&
	272	nla_put_flag(skb, PSAMPLE_TUNNEL_KEY_ATTR_OAM))
	273	return -EMSGSIZE;
	274	if (tun_opts_len) {
	275	if (tun_key->tun_flags & TUNNEL_GENEVE_OPT &&
	276	nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_GENEVE_OPTS,
	277	tun_opts_len, tun_opts))
	278	return -EMSGSIZE;
279	else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT &&
280	nla_put(skb, PSAMPLE_TUNNEL_KEY_ATTR_ERSPAN_OPTS,
281	tun_opts_len, tun_opts))
282	return -EMSGSIZE;
283	}
284
285	return 0;
286	}
287
288	static int psample_ip_tun_to_nlattr(struct sk_buff *skb,
289	struct ip_tunnel_info *tun_info)
290	{
291	struct nlattr *nla;
292	int err;
293
294	nla = nla_nest_start_noflag(skb, PSAMPLE_ATTR_TUNNEL);
295	if (!nla)
296	return -EMSGSIZE;
297
298	err = __psample_ip_tun_to_nlattr(skb, tun_info);
299	if (err) {
300	nla_nest_cancel(skb, nla);
301	return err;
302	}
303
304	nla_nest_end(skb, nla);
305
306	return 0;
307	}
308
309	static int psample_tunnel_meta_len(struct ip_tunnel_info *tun_info)
310	{
311	unsigned short tun_proto = ip_tunnel_info_af(tun_info);
312	const struct ip_tunnel_key *tun_key = &tun_info->key;
313	int tun_opts_len = tun_info->options_len;
a93dcaad	314	int sum = nla_total_size(0); /* PSAMPLE_ATTR_TUNNEL */
d8bed686 CM	315
d8bed686 CM	316	if (tun_key->tun_flags & TUNNEL_KEY)
a93dcaad	317	sum += nla_total_size_64bit(sizeof(u64));
d8bed686 CM	318
	319	if (tun_info->mode & IP_TUNNEL_INFO_BRIDGE)
	320	sum += nla_total_size(0);
	321
	322	switch (tun_proto) {
	323	case AF_INET:
	324	if (tun_key->u.ipv4.src)
	325	sum += nla_total_size(sizeof(u32));
	326	if (tun_key->u.ipv4.dst)
	327	sum += nla_total_size(sizeof(u32));
	328	break;
	329	case AF_INET6:
	330	if (!ipv6_addr_any(&tun_key->u.ipv6.src))
	331	sum += nla_total_size(sizeof(struct in6_addr));
	332	if (!ipv6_addr_any(&tun_key->u.ipv6.dst))
	333	sum += nla_total_size(sizeof(struct in6_addr));
	334	break;
	335	}
	336	if (tun_key->tos)
	337	sum += nla_total_size(sizeof(u8));
	338	sum += nla_total_size(sizeof(u8)); /* TTL */
	339	if (tun_key->tun_flags & TUNNEL_DONT_FRAGMENT)
	340	sum += nla_total_size(0);
	341	if (tun_key->tun_flags & TUNNEL_CSUM)
	342	sum += nla_total_size(0);
	343	if (tun_key->tp_src)
	344	sum += nla_total_size(sizeof(u16));
	345	if (tun_key->tp_dst)
	346	sum += nla_total_size(sizeof(u16));
	347	if (tun_key->tun_flags & TUNNEL_OAM)
	348	sum += nla_total_size(0);
	349	if (tun_opts_len) {
	350	if (tun_key->tun_flags & TUNNEL_GENEVE_OPT)
	351	sum += nla_total_size(tun_opts_len);
	352	else if (tun_key->tun_flags & TUNNEL_ERSPAN_OPT)
	353	sum += nla_total_size(tun_opts_len);
	354	}
	355
	356	return sum;
	357	}
07a7f308	358	#endif
d8bed686	359
6ae0a628	360	void psample_sample_packet(struct psample_group group, struct sk_buff skb,
a03e99d3	361	u32 sample_rate, const struct psample_metadata *md)
6ae0a628	362	{
07e1a580	363	ktime_t tstamp = ktime_get_real();
a03e99d3 IS	364	int out_ifindex = md->out_ifindex;
	365	int in_ifindex = md->in_ifindex;
	366	u32 trunc_size = md->trunc_size;
07a7f308	367	#ifdef CONFIG_INET
d8bed686	368	struct ip_tunnel_info *tun_info;
07a7f308	369	#endif
6ae0a628 YG	370	struct sk_buff *nl_skb;
	371	int data_len;
	372	int meta_len;
	373	void *data;
	374	int ret;
	375
	376	meta_len = (in_ifindex ? nla_total_size(sizeof(u16)) : 0) +
	377	(out_ifindex ? nla_total_size(sizeof(u16)) : 0) +
07e1a580 IS	378	(md->out_tc_valid ? nla_total_size(sizeof(u16)) : 0) +
	379	(md->out_tc_occ_valid ? nla_total_size_64bit(sizeof(u64)) : 0) +
	380	(md->latency_valid ? nla_total_size_64bit(sizeof(u64)) : 0) +
6ae0a628 YG	381	nla_total_size(sizeof(u32)) + /* sample_rate */
	382	nla_total_size(sizeof(u32)) + /* orig_size */
	383	nla_total_size(sizeof(u32)) + /* group_num */
07e1a580 IS	384	nla_total_size(sizeof(u32)) + /* seq */
	385	nla_total_size_64bit(sizeof(u64)) + /* timestamp */
	386	nla_total_size(sizeof(u16)); /* protocol */
6ae0a628	387
07a7f308	388	#ifdef CONFIG_INET
d8bed686 CM	389	tun_info = skb_tunnel_info(skb);
	390	if (tun_info)
	391	meta_len += psample_tunnel_meta_len(tun_info);
07a7f308	392	#endif
d8bed686	393
6ae0a628 YG	394	data_len = min(skb->len, trunc_size);
	395	if (meta_len + nla_total_size(data_len) > PSAMPLE_MAX_PACKET_SIZE)
	396	data_len = PSAMPLE_MAX_PACKET_SIZE - meta_len - NLA_HDRLEN
	397	- NLA_ALIGNTO;
	398
7eb9d767	399	nl_skb = genlmsg_new(meta_len + nla_total_size(data_len), GFP_ATOMIC);
6ae0a628 YG	400	if (unlikely(!nl_skb))
	401	return;
	402
	403	data = genlmsg_put(nl_skb, 0, 0, &psample_nl_family, 0,
	404	PSAMPLE_CMD_SAMPLE);
	405	if (unlikely(!data))
	406	goto error;
	407
	408	if (in_ifindex) {
	409	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_IIFINDEX, in_ifindex);
	410	if (unlikely(ret < 0))
	411	goto error;
	412	}
	413
	414	if (out_ifindex) {
	415	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OIFINDEX, out_ifindex);
	416	if (unlikely(ret < 0))
	417	goto error;
	418	}
	419
	420	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_RATE, sample_rate);
	421	if (unlikely(ret < 0))
	422	goto error;
	423
	424	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_ORIGSIZE, skb->len);
	425	if (unlikely(ret < 0))
	426	goto error;
	427
	428	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
	429	if (unlikely(ret < 0))
	430	goto error;
	431
	432	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_GROUP_SEQ, group->seq++);
	433	if (unlikely(ret < 0))
	434	goto error;
	435
07e1a580 IS	436	if (md->out_tc_valid) {
	437	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OUT_TC, md->out_tc);
	438	if (unlikely(ret < 0))
	439	goto error;
	440	}
	441
	442	if (md->out_tc_occ_valid) {
	443	ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_OUT_TC_OCC,
	444	md->out_tc_occ, PSAMPLE_ATTR_PAD);
	445	if (unlikely(ret < 0))
	446	goto error;
	447	}
	448
	449	if (md->latency_valid) {
	450	ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_LATENCY,
	451	md->latency, PSAMPLE_ATTR_PAD);
	452	if (unlikely(ret < 0))
	453	goto error;
	454	}
	455
	456	ret = nla_put_u64_64bit(nl_skb, PSAMPLE_ATTR_TIMESTAMP,
	457	ktime_to_ns(tstamp), PSAMPLE_ATTR_PAD);
	458	if (unlikely(ret < 0))
	459	goto error;
	460
	461	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_PROTO,
	462	be16_to_cpu(skb->protocol));
	463	if (unlikely(ret < 0))
	464	goto error;
	465
6ae0a628 YG	466	if (data_len) {
	467	int nla_len = nla_total_size(data_len);
	468	struct nlattr *nla;
	469
4df864c1	470	nla = skb_put(nl_skb, nla_len);
6ae0a628 YG	471	nla->nla_type = PSAMPLE_ATTR_DATA;
	472	nla->nla_len = nla_attr_size(data_len);
	473
	474	if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
	475	goto error;
	476	}
	477
07a7f308	478	#ifdef CONFIG_INET
d8bed686 CM	479	if (tun_info) {
	480	ret = psample_ip_tun_to_nlattr(nl_skb, tun_info);
	481	if (unlikely(ret < 0))
	482	goto error;
	483	}
07a7f308	484	#endif
d8bed686	485
6ae0a628 YG	486	genlmsg_end(nl_skb, data);
	487	genlmsg_multicast_netns(&psample_nl_family, group->net, nl_skb, 0,
	488	PSAMPLE_NL_MCGRP_SAMPLE, GFP_ATOMIC);
	489
	490	return;
	491	error:
	492	pr_err_ratelimited("Could not create psample log message\n");
	493	nlmsg_free(nl_skb);
	494	}
	495	EXPORT_SYMBOL_GPL(psample_sample_packet);
	496
	497	static int __init psample_module_init(void)
	498	{
	499	return genl_register_family(&psample_nl_family);
	500	}
	501
	502	static void __exit psample_module_exit(void)
	503	{
	504	genl_unregister_family(&psample_nl_family);
	505	}
	506
	507	module_init(psample_module_init);
	508	module_exit(psample_module_exit);
	509
f1fd20c3	510	MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
6ae0a628 YG	511	MODULE_DESCRIPTION("netlink channel for packet sampling");
6ae0a628 YG	512	MODULE_LICENSE("GPL v2");