[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 <net/net_namespace.h>
#include <net/sock.h>
#include <net/netlink.h>
#include <net/genetlink.h>
#include <net/psample.h>
#include <linux/spinlock.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_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,
	.ops		= psample_nl_ops,
	.n_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);

void psample_sample_packet(struct psample_group *group, struct sk_buff *skb,
			   u32 trunc_size, int in_ifindex, int out_ifindex,
			   u32 sample_rate)
{
	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) +
		   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 */

	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 + 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 (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;
	}

	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>
	11	#include <net/net_namespace.h>
	12	#include <net/sock.h>
	13	#include <net/netlink.h>
	14	#include <net/genetlink.h>
	15	#include <net/psample.h>
	16	#include <linux/spinlock.h>
	17
	18	#define PSAMPLE_MAX_PACKET_SIZE 0xffff
	19
	20	static LIST_HEAD(psample_groups_list);
	21	static DEFINE_SPINLOCK(psample_groups_lock);
	22
	23	/* multicast groups */
	24	enum psample_nl_multicast_groups {
	25	PSAMPLE_NL_MCGRP_CONFIG,
	26	PSAMPLE_NL_MCGRP_SAMPLE,
	27	};
	28
	29	static const struct genl_multicast_group psample_nl_mcgrps[] = {
	30	[PSAMPLE_NL_MCGRP_CONFIG] = { .name = PSAMPLE_NL_MCGRP_CONFIG_NAME },
	31	[PSAMPLE_NL_MCGRP_SAMPLE] = { .name = PSAMPLE_NL_MCGRP_SAMPLE_NAME },
	32	};
	33
	34	static struct genl_family psample_nl_family __ro_after_init;
	35
	36	static int psample_group_nl_fill(struct sk_buff *msg,
	37	struct psample_group *group,
	38	enum psample_command cmd, u32 portid, u32 seq,
	39	int flags)
	40	{
	41	void *hdr;
	42	int ret;
	43
	44	hdr = genlmsg_put(msg, portid, seq, &psample_nl_family, flags, cmd);
	45	if (!hdr)
	46	return -EMSGSIZE;
	47
	48	ret = nla_put_u32(msg, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
	49	if (ret < 0)
	50	goto error;
	51
	52	ret = nla_put_u32(msg, PSAMPLE_ATTR_GROUP_REFCOUNT, group->refcount);
	53	if (ret < 0)
	54	goto error;
	55
	56	ret = nla_put_u32(msg, PSAMPLE_ATTR_GROUP_SEQ, group->seq);
	57	if (ret < 0)
	58	goto error;
	59
	60	genlmsg_end(msg, hdr);
	61	return 0;
	62
	63	error:
	64	genlmsg_cancel(msg, hdr);
	65	return -EMSGSIZE;
	66	}
	67
	68	static int psample_nl_cmd_get_group_dumpit(struct sk_buff *msg,
69	struct netlink_callback *cb)
70	{
71	struct psample_group *group;
72	int start = cb->args[0];
73	int idx = 0;
74	int err;
75
4a5da47d	76	spin_lock_bh(&psample_groups_lock);
6ae0a628 YG	77	list_for_each_entry(group, &psample_groups_list, list) {
	78	if (!net_eq(group->net, sock_net(msg->sk)))
	79	continue;
	80	if (idx < start) {
	81	idx++;
	82	continue;
	83	}
	84	err = psample_group_nl_fill(msg, group, PSAMPLE_CMD_NEW_GROUP,
	85	NETLINK_CB(cb->skb).portid,
	86	cb->nlh->nlmsg_seq, NLM_F_MULTI);
	87	if (err)
	88	break;
	89	idx++;
	90	}
	91
4a5da47d	92	spin_unlock_bh(&psample_groups_lock);
6ae0a628 YG	93	cb->args[0] = idx;
	94	return msg->len;
	95	}
	96
	97	static const struct genl_ops psample_nl_ops[] = {
	98	{
	99	.cmd = PSAMPLE_CMD_GET_GROUP,
ef6243ac	100	.validate = GENL_DONT_VALIDATE_STRICT \| GENL_DONT_VALIDATE_DUMP,
6ae0a628 YG	101	.dumpit = psample_nl_cmd_get_group_dumpit,
	102	/* can be retrieved by unprivileged users */
	103	}
	104	};
	105
	106	static struct genl_family psample_nl_family __ro_after_init = {
	107	.name = PSAMPLE_GENL_NAME,
	108	.version = PSAMPLE_GENL_VERSION,
	109	.maxattr = PSAMPLE_ATTR_MAX,
	110	.netnsok = true,
	111	.module = THIS_MODULE,
	112	.mcgrps = psample_nl_mcgrps,
	113	.ops = psample_nl_ops,
	114	.n_ops = ARRAY_SIZE(psample_nl_ops),
	115	.n_mcgrps = ARRAY_SIZE(psample_nl_mcgrps),
	116	};
	117
	118	static void psample_group_notify(struct psample_group *group,
	119	enum psample_command cmd)
	120	{
	121	struct sk_buff *msg;
	122	int err;
	123
	124	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
	125	if (!msg)
	126	return;
	127
	128	err = psample_group_nl_fill(msg, group, cmd, 0, 0, NLM_F_MULTI);
	129	if (!err)
	130	genlmsg_multicast_netns(&psample_nl_family, group->net, msg, 0,
	131	PSAMPLE_NL_MCGRP_CONFIG, GFP_ATOMIC);
	132	else
	133	nlmsg_free(msg);
	134	}
	135
	136	static struct psample_group psample_group_create(struct net net,
	137	u32 group_num)
	138	{
	139	struct psample_group *group;
	140
	141	group = kzalloc(sizeof(*group), GFP_ATOMIC);
	142	if (!group)
	143	return NULL;
	144
	145	group->net = net;
	146	group->group_num = group_num;
	147	list_add_tail(&group->list, &psample_groups_list);
	148
	149	psample_group_notify(group, PSAMPLE_CMD_NEW_GROUP);
	150	return group;
	151	}
	152
	153	static void psample_group_destroy(struct psample_group *group)
	154	{
	155	psample_group_notify(group, PSAMPLE_CMD_DEL_GROUP);
	156	list_del(&group->list);
dbf47a2a	157	kfree_rcu(group, rcu);
6ae0a628 YG	158	}
	159
	160	static struct psample_group *
	161	psample_group_lookup(struct net *net, u32 group_num)
	162	{
	163	struct psample_group *group;
	164
	165	list_for_each_entry(group, &psample_groups_list, list)
	166	if ((group->group_num == group_num) && (group->net == net))
	167	return group;
	168	return NULL;
	169	}
	170
	171	struct psample_group psample_group_get(struct net net, u32 group_num)
	172	{
	173	struct psample_group *group;
	174
4a5da47d	175	spin_lock_bh(&psample_groups_lock);
6ae0a628 YG	176
	177	group = psample_group_lookup(net, group_num);
	178	if (!group) {
	179	group = psample_group_create(net, group_num);
	180	if (!group)
	181	goto out;
	182	}
	183	group->refcount++;
	184
	185	out:
4a5da47d	186	spin_unlock_bh(&psample_groups_lock);
6ae0a628 YG	187	return group;
	188	}
	189	EXPORT_SYMBOL_GPL(psample_group_get);
	190
4a5da47d VB	191	void psample_group_take(struct psample_group *group)
	192	{
	193	spin_lock_bh(&psample_groups_lock);
	194	group->refcount++;
	195	spin_unlock_bh(&psample_groups_lock);
	196	}
	197	EXPORT_SYMBOL_GPL(psample_group_take);
	198
6ae0a628 YG	199	void psample_group_put(struct psample_group *group)
6ae0a628 YG	200	{
4a5da47d	201	spin_lock_bh(&psample_groups_lock);
6ae0a628 YG	202
	203	if (--group->refcount == 0)
	204	psample_group_destroy(group);
	205
4a5da47d	206	spin_unlock_bh(&psample_groups_lock);
6ae0a628 YG	207	}
	208	EXPORT_SYMBOL_GPL(psample_group_put);
	209
	210	void psample_sample_packet(struct psample_group group, struct sk_buff skb,
	211	u32 trunc_size, int in_ifindex, int out_ifindex,
	212	u32 sample_rate)
	213	{
	214	struct sk_buff *nl_skb;
	215	int data_len;
	216	int meta_len;
	217	void *data;
	218	int ret;
	219
	220	meta_len = (in_ifindex ? nla_total_size(sizeof(u16)) : 0) +
	221	(out_ifindex ? nla_total_size(sizeof(u16)) : 0) +
	222	nla_total_size(sizeof(u32)) + /* sample_rate */
	223	nla_total_size(sizeof(u32)) + /* orig_size */
	224	nla_total_size(sizeof(u32)) + /* group_num */
	225	nla_total_size(sizeof(u32)); /* seq */
	226
	227	data_len = min(skb->len, trunc_size);
	228	if (meta_len + nla_total_size(data_len) > PSAMPLE_MAX_PACKET_SIZE)
	229	data_len = PSAMPLE_MAX_PACKET_SIZE - meta_len - NLA_HDRLEN
	230	- NLA_ALIGNTO;
	231
	232	nl_skb = genlmsg_new(meta_len + data_len, GFP_ATOMIC);
	233	if (unlikely(!nl_skb))
	234	return;
	235
	236	data = genlmsg_put(nl_skb, 0, 0, &psample_nl_family, 0,
	237	PSAMPLE_CMD_SAMPLE);
	238	if (unlikely(!data))
	239	goto error;
	240
	241	if (in_ifindex) {
	242	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_IIFINDEX, in_ifindex);
	243	if (unlikely(ret < 0))
	244	goto error;
	245	}
	246
	247	if (out_ifindex) {
	248	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OIFINDEX, out_ifindex);
	249	if (unlikely(ret < 0))
	250	goto error;
	251	}
	252
	253	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_RATE, sample_rate);
	254	if (unlikely(ret < 0))
	255	goto error;
	256
	257	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_ORIGSIZE, skb->len);
	258	if (unlikely(ret < 0))
	259	goto error;
	260
	261	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
	262	if (unlikely(ret < 0))
	263	goto error;
	264
	265	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_GROUP_SEQ, group->seq++);
	266	if (unlikely(ret < 0))
	267	goto error;
	268
	269	if (data_len) {
	270	int nla_len = nla_total_size(data_len);
271	struct nlattr *nla;
272
4df864c1	273	nla = skb_put(nl_skb, nla_len);
6ae0a628 YG	274	nla->nla_type = PSAMPLE_ATTR_DATA;
	275	nla->nla_len = nla_attr_size(data_len);
	276
	277	if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
	278	goto error;
	279	}
	280
	281	genlmsg_end(nl_skb, data);
	282	genlmsg_multicast_netns(&psample_nl_family, group->net, nl_skb, 0,
	283	PSAMPLE_NL_MCGRP_SAMPLE, GFP_ATOMIC);
	284
	285	return;
	286	error:
	287	pr_err_ratelimited("Could not create psample log message\n");
	288	nlmsg_free(nl_skb);
	289	}
	290	EXPORT_SYMBOL_GPL(psample_sample_packet);
	291
	292	static int __init psample_module_init(void)
	293	{
	294	return genl_register_family(&psample_nl_family);
	295	}
	296
	297	static void __exit psample_module_exit(void)
	298	{
	299	genl_unregister_family(&psample_nl_family);
	300	}
	301
	302	module_init(psample_module_init);
	303	module_exit(psample_module_exit);
	304
f1fd20c3	305	MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
6ae0a628 YG	306	MODULE_DESCRIPTION("netlink channel for packet sampling");
6ae0a628 YG	307	MODULE_LICENSE("GPL v2");