1/*
  2 * net/sched/act_sample.c - Packet sampling tc action
  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/string.h>
 13#include <linux/errno.h>
 14#include <linux/skbuff.h>
 15#include <linux/rtnetlink.h>
 16#include <linux/module.h>
 17#include <linux/init.h>
 18#include <linux/gfp.h>
 19#include <net/net_namespace.h>
 20#include <net/netlink.h>
 21#include <net/pkt_sched.h>
 22#include <linux/tc_act/tc_sample.h>
 23#include <net/tc_act/tc_sample.h>
 24#include <net/psample.h>
 25
 26#include <linux/if_arp.h>
 27
 28static unsigned int sample_net_id;
 29static struct tc_action_ops act_sample_ops;
 30
 31static const struct nla_policy sample_policy[TCA_SAMPLE_MAX + 1] = {
 32	[TCA_SAMPLE_PARMS]		= { .len = sizeof(struct tc_sample) },
 33	[TCA_SAMPLE_RATE]		= { .type = NLA_U32 },
 34	[TCA_SAMPLE_TRUNC_SIZE]		= { .type = NLA_U32 },
 35	[TCA_SAMPLE_PSAMPLE_GROUP]	= { .type = NLA_U32 },
 36};
 37
 38static int tcf_sample_init(struct net *net, struct nlattr *nla,
 39			   struct nlattr *est, struct tc_action **a, int ovr,
 40			   int bind, struct netlink_ext_ack *extack)
 41{
 42	struct tc_action_net *tn = net_generic(net, sample_net_id);
 43	struct nlattr *tb[TCA_SAMPLE_MAX + 1];
 44	struct psample_group *psample_group;
 45	struct tc_sample *parm;
 46	struct tcf_sample *s;
 47	bool exists = false;
 48	int ret;
 49
 50	if (!nla)
 51		return -EINVAL;
 52	ret = nla_parse_nested(tb, TCA_SAMPLE_MAX, nla, sample_policy, NULL);
 53	if (ret < 0)
 54		return ret;
 55	if (!tb[TCA_SAMPLE_PARMS] || !tb[TCA_SAMPLE_RATE] ||
 56	    !tb[TCA_SAMPLE_PSAMPLE_GROUP])
 57		return -EINVAL;
 58
 59	parm = nla_data(tb[TCA_SAMPLE_PARMS]);
 60
 61	exists = tcf_idr_check(tn, parm->index, a, bind);
 62	if (exists && bind)
 63		return 0;
 64
 65	if (!exists) {
 66		ret = tcf_idr_create(tn, parm->index, est, a,
 67				     &act_sample_ops, bind, false);
 68		if (ret)
 69			return ret;
 70		ret = ACT_P_CREATED;
 71	} else {
 72		tcf_idr_release(*a, bind);
 73		if (!ovr)
 74			return -EEXIST;
 75	}
 76	s = to_sample(*a);
 77
 78	s->tcf_action = parm->action;
 79	s->rate = nla_get_u32(tb[TCA_SAMPLE_RATE]);
 80	s->psample_group_num = nla_get_u32(tb[TCA_SAMPLE_PSAMPLE_GROUP]);
 81	psample_group = psample_group_get(net, s->psample_group_num);
 82	if (!psample_group) {
 83		if (ret == ACT_P_CREATED)
 84			tcf_idr_release(*a, bind);
 85		return -ENOMEM;
 86	}
 87	RCU_INIT_POINTER(s->psample_group, psample_group);
 88
 89	if (tb[TCA_SAMPLE_TRUNC_SIZE]) {
 90		s->truncate = true;
 91		s->trunc_size = nla_get_u32(tb[TCA_SAMPLE_TRUNC_SIZE]);
 92	}
 93
 94	if (ret == ACT_P_CREATED)
 95		tcf_idr_insert(tn, *a);
 96	return ret;
 97}
 98
 99static void tcf_sample_cleanup(struct tc_action *a)
100{
101	struct tcf_sample *s = to_sample(a);
102	struct psample_group *psample_group;
103
104	psample_group = rtnl_dereference(s->psample_group);
105	RCU_INIT_POINTER(s->psample_group, NULL);
106	if (psample_group)
107		psample_group_put(psample_group);
108}
109
110static bool tcf_sample_dev_ok_push(struct net_device *dev)
111{
112	switch (dev->type) {
113	case ARPHRD_TUNNEL:
114	case ARPHRD_TUNNEL6:
115	case ARPHRD_SIT:
116	case ARPHRD_IPGRE:
117	case ARPHRD_VOID:
118	case ARPHRD_NONE:
119		return false;
120	default:
121		return true;
122	}
123}
124
125static int tcf_sample_act(struct sk_buff *skb, const struct tc_action *a,
126			  struct tcf_result *res)
127{
128	struct tcf_sample *s = to_sample(a);
129	struct psample_group *psample_group;
130	int retval;
131	int size;
132	int iif;
133	int oif;
134
135	tcf_lastuse_update(&s->tcf_tm);
136	bstats_cpu_update(this_cpu_ptr(s->common.cpu_bstats), skb);
137	retval = READ_ONCE(s->tcf_action);
138
139	rcu_read_lock();
140	psample_group = rcu_dereference(s->psample_group);
141
142	/* randomly sample packets according to rate */
143	if (psample_group && (prandom_u32() % s->rate == 0)) {
144		if (!skb_at_tc_ingress(skb)) {
145			iif = skb->skb_iif;
146			oif = skb->dev->ifindex;
147		} else {
148			iif = skb->dev->ifindex;
149			oif = 0;
150		}
151
152		/* on ingress, the mac header gets popped, so push it back */
153		if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
154			skb_push(skb, skb->mac_len);
155
156		size = s->truncate ? s->trunc_size : skb->len;
157		psample_sample_packet(psample_group, skb, size, iif, oif,
158				      s->rate);
159
160		if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
161			skb_pull(skb, skb->mac_len);
162	}
163
164	rcu_read_unlock();
165	return retval;
166}
167
168static int tcf_sample_dump(struct sk_buff *skb, struct tc_action *a,
169			   int bind, int ref)
170{
171	unsigned char *b = skb_tail_pointer(skb);
172	struct tcf_sample *s = to_sample(a);
173	struct tc_sample opt = {
174		.index      = s->tcf_index,
175		.action     = s->tcf_action,
176		.refcnt     = s->tcf_refcnt - ref,
177		.bindcnt    = s->tcf_bindcnt - bind,
178	};
179	struct tcf_t t;
180
181	if (nla_put(skb, TCA_SAMPLE_PARMS, sizeof(opt), &opt))
182		goto nla_put_failure;
183
184	tcf_tm_dump(&t, &s->tcf_tm);
185	if (nla_put_64bit(skb, TCA_SAMPLE_TM, sizeof(t), &t, TCA_SAMPLE_PAD))
186		goto nla_put_failure;
187
188	if (nla_put_u32(skb, TCA_SAMPLE_RATE, s->rate))
189		goto nla_put_failure;
190
191	if (s->truncate)
192		if (nla_put_u32(skb, TCA_SAMPLE_TRUNC_SIZE, s->trunc_size))
193			goto nla_put_failure;
194
195	if (nla_put_u32(skb, TCA_SAMPLE_PSAMPLE_GROUP, s->psample_group_num))
196		goto nla_put_failure;
197	return skb->len;
198
199nla_put_failure:
200	nlmsg_trim(skb, b);
201	return -1;
202}
203
204static int tcf_sample_walker(struct net *net, struct sk_buff *skb,
205			     struct netlink_callback *cb, int type,
206			     const struct tc_action_ops *ops,
207			     struct netlink_ext_ack *extack)
208{
209	struct tc_action_net *tn = net_generic(net, sample_net_id);
210
211	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
212}
213
214static int tcf_sample_search(struct net *net, struct tc_action **a, u32 index,
215			     struct netlink_ext_ack *extack)
216{
217	struct tc_action_net *tn = net_generic(net, sample_net_id);
218
219	return tcf_idr_search(tn, a, index);
220}
221
222static struct tc_action_ops act_sample_ops = {
223	.kind	  = "sample",
224	.type	  = TCA_ACT_SAMPLE,
225	.owner	  = THIS_MODULE,
226	.act	  = tcf_sample_act,
227	.dump	  = tcf_sample_dump,
228	.init	  = tcf_sample_init,
229	.cleanup  = tcf_sample_cleanup,
230	.walk	  = tcf_sample_walker,
231	.lookup	  = tcf_sample_search,
232	.size	  = sizeof(struct tcf_sample),
233};
234
235static __net_init int sample_init_net(struct net *net)
236{
237	struct tc_action_net *tn = net_generic(net, sample_net_id);
238
239	return tc_action_net_init(tn, &act_sample_ops);
240}
241
242static void __net_exit sample_exit_net(struct list_head *net_list)
243{
244	tc_action_net_exit(net_list, sample_net_id);
245}
246
247static struct pernet_operations sample_net_ops = {
248	.init = sample_init_net,
249	.exit_batch = sample_exit_net,
250	.id   = &sample_net_id,
251	.size = sizeof(struct tc_action_net),
252};
253
254static int __init sample_init_module(void)
255{
256	return tcf_register_action(&act_sample_ops, &sample_net_ops);
257}
258
259static void __exit sample_cleanup_module(void)
260{
261	tcf_unregister_action(&act_sample_ops, &sample_net_ops);
262}
263
264module_init(sample_init_module);
265module_exit(sample_cleanup_module);
266
267MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
268MODULE_DESCRIPTION("Packet sampling action");
269MODULE_LICENSE("GPL v2");