[mirror_ubuntu-artful-kernel.git] / net / psample / psample.c

/*
 * net/psample/psample.c - Netlink channel for packet sampling
 * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */

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

static const struct genl_ops psample_nl_ops[] = {
	{
		.cmd = PSAMPLE_CMD_GET_GROUP,
		.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(group);
}

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(&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(&psample_groups_lock);
	return group;
}
EXPORT_SYMBOL_GPL(psample_group_get);

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

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

	spin_unlock(&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 <yotamg@mellanox.com>");
MODULE_DESCRIPTION("netlink channel for packet sampling");
MODULE_LICENSE("GPL v2");
Commit	Line	Data
6ae0a628 YG	1	/*
	2	* net/psample/psample.c - Netlink channel for packet sampling
	3	* Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com>
	4	*
	5	* This program is free software; you can redistribute it and/or modify
	6	* it under the terms of the GNU General Public License version 2 as
	7	* published by the Free Software Foundation.
	8	*/
	9
	10	#include <linux/types.h>
	11	#include <linux/kernel.h>
	12	#include <linux/skbuff.h>
	13	#include <linux/module.h>
	14	#include <net/net_namespace.h>
	15	#include <net/sock.h>
	16	#include <net/netlink.h>
	17	#include <net/genetlink.h>
	18	#include <net/psample.h>
	19	#include <linux/spinlock.h>
	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
79	spin_lock(&psample_groups_lock);
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
95	spin_unlock(&psample_groups_lock);
96	cb->args[0] = idx;
97	return msg->len;
98	}
99
100	static const struct genl_ops psample_nl_ops[] = {
101	{
102	.cmd = PSAMPLE_CMD_GET_GROUP,
103	.dumpit = psample_nl_cmd_get_group_dumpit,
104	/* can be retrieved by unprivileged users */
105	}
106	};
107
108	static struct genl_family psample_nl_family __ro_after_init = {
109	.name = PSAMPLE_GENL_NAME,
110	.version = PSAMPLE_GENL_VERSION,
111	.maxattr = PSAMPLE_ATTR_MAX,
112	.netnsok = true,
113	.module = THIS_MODULE,
114	.mcgrps = psample_nl_mcgrps,
115	.ops = psample_nl_ops,
116	.n_ops = ARRAY_SIZE(psample_nl_ops),
117	.n_mcgrps = ARRAY_SIZE(psample_nl_mcgrps),
118	};
119
120	static void psample_group_notify(struct psample_group *group,
121	enum psample_command cmd)
122	{
123	struct sk_buff *msg;
124	int err;
125
126	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
127	if (!msg)
128	return;
129
130	err = psample_group_nl_fill(msg, group, cmd, 0, 0, NLM_F_MULTI);
131	if (!err)
132	genlmsg_multicast_netns(&psample_nl_family, group->net, msg, 0,
133	PSAMPLE_NL_MCGRP_CONFIG, GFP_ATOMIC);
134	else
135	nlmsg_free(msg);
136	}
137
138	static struct psample_group psample_group_create(struct net net,
139	u32 group_num)
140	{
141	struct psample_group *group;
142
143	group = kzalloc(sizeof(*group), GFP_ATOMIC);
144	if (!group)
145	return NULL;
146
147	group->net = net;
148	group->group_num = group_num;
149	list_add_tail(&group->list, &psample_groups_list);
150
151	psample_group_notify(group, PSAMPLE_CMD_NEW_GROUP);
152	return group;
153	}
154
155	static void psample_group_destroy(struct psample_group *group)
156	{
157	psample_group_notify(group, PSAMPLE_CMD_DEL_GROUP);
158	list_del(&group->list);
159	kfree(group);
160	}
161
162	static struct psample_group *
163	psample_group_lookup(struct net *net, u32 group_num)
164	{
165	struct psample_group *group;
166
167	list_for_each_entry(group, &psample_groups_list, list)
168	if ((group->group_num == group_num) && (group->net == net))
169	return group;
170	return NULL;
171	}
172
173	struct psample_group psample_group_get(struct net net, u32 group_num)
174	{
175	struct psample_group *group;
176
177	spin_lock(&psample_groups_lock);
178
179	group = psample_group_lookup(net, group_num);
180	if (!group) {
181	group = psample_group_create(net, group_num);
182	if (!group)
183	goto out;
184	}
185	group->refcount++;
186
187	out:
188	spin_unlock(&psample_groups_lock);
189	return group;
190	}
191	EXPORT_SYMBOL_GPL(psample_group_get);
192
193	void psample_group_put(struct psample_group *group)
194	{
195	spin_lock(&psample_groups_lock);
196
197	if (--group->refcount == 0)
198	psample_group_destroy(group);
199
200	spin_unlock(&psample_groups_lock);
201	}
202	EXPORT_SYMBOL_GPL(psample_group_put);
203
204	void psample_sample_packet(struct psample_group group, struct sk_buff skb,
205	u32 trunc_size, int in_ifindex, int out_ifindex,
206	u32 sample_rate)
207	{
208	struct sk_buff *nl_skb;
209	int data_len;
210	int meta_len;
211	void *data;
212	int ret;
213
214	meta_len = (in_ifindex ? nla_total_size(sizeof(u16)) : 0) +
215	(out_ifindex ? nla_total_size(sizeof(u16)) : 0) +
216	nla_total_size(sizeof(u32)) + /* sample_rate */
217	nla_total_size(sizeof(u32)) + /* orig_size */
218	nla_total_size(sizeof(u32)) + /* group_num */
219	nla_total_size(sizeof(u32)); /* seq */
220
221	data_len = min(skb->len, trunc_size);
222	if (meta_len + nla_total_size(data_len) > PSAMPLE_MAX_PACKET_SIZE)
223	data_len = PSAMPLE_MAX_PACKET_SIZE - meta_len - NLA_HDRLEN
224	- NLA_ALIGNTO;
225
226	nl_skb = genlmsg_new(meta_len + data_len, GFP_ATOMIC);
227	if (unlikely(!nl_skb))
228	return;
229
230	data = genlmsg_put(nl_skb, 0, 0, &psample_nl_family, 0,
231	PSAMPLE_CMD_SAMPLE);
232	if (unlikely(!data))
233	goto error;
234
235	if (in_ifindex) {
236	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_IIFINDEX, in_ifindex);
237	if (unlikely(ret < 0))
238	goto error;
239	}
240
241	if (out_ifindex) {
242	ret = nla_put_u16(nl_skb, PSAMPLE_ATTR_OIFINDEX, out_ifindex);
243	if (unlikely(ret < 0))
244	goto error;
245	}
246
247	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_RATE, sample_rate);
248	if (unlikely(ret < 0))
249	goto error;
250
251	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_ORIGSIZE, skb->len);
252	if (unlikely(ret < 0))
253	goto error;
254
255	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_SAMPLE_GROUP, group->group_num);
256	if (unlikely(ret < 0))
257	goto error;
258
259	ret = nla_put_u32(nl_skb, PSAMPLE_ATTR_GROUP_SEQ, group->seq++);
260	if (unlikely(ret < 0))
261	goto error;
262
263	if (data_len) {
264	int nla_len = nla_total_size(data_len);
265	struct nlattr *nla;
266
4df864c1	267	nla = skb_put(nl_skb, nla_len);
6ae0a628 YG	268	nla->nla_type = PSAMPLE_ATTR_DATA;
	269	nla->nla_len = nla_attr_size(data_len);
	270
	271	if (skb_copy_bits(skb, 0, nla_data(nla), data_len))
	272	goto error;
	273	}
	274
	275	genlmsg_end(nl_skb, data);
	276	genlmsg_multicast_netns(&psample_nl_family, group->net, nl_skb, 0,
	277	PSAMPLE_NL_MCGRP_SAMPLE, GFP_ATOMIC);
	278
	279	return;
	280	error:
	281	pr_err_ratelimited("Could not create psample log message\n");
	282	nlmsg_free(nl_skb);
	283	}
	284	EXPORT_SYMBOL_GPL(psample_sample_packet);
	285
	286	static int __init psample_module_init(void)
	287	{
	288	return genl_register_family(&psample_nl_family);
	289	}
	290
	291	static void __exit psample_module_exit(void)
	292	{
	293	genl_unregister_family(&psample_nl_family);
	294	}
	295
	296	module_init(psample_module_init);
	297	module_exit(psample_module_exit);
	298
	299	MODULE_AUTHOR("Yotam Gigi <yotamg@mellanox.com>");
	300	MODULE_DESCRIPTION("netlink channel for packet sampling");
	301	MODULE_LICENSE("GPL v2");