xref: /OK3568_Linux_fs/kernel/net/sched/act_sample.c (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1*4882a593Smuzhiyun // SPDX-License-Identifier: GPL-2.0-only
2*4882a593Smuzhiyun /*
3*4882a593Smuzhiyun  * net/sched/act_sample.c - Packet sampling tc action
4*4882a593Smuzhiyun  * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com>
5*4882a593Smuzhiyun  */
6*4882a593Smuzhiyun 
7*4882a593Smuzhiyun #include <linux/types.h>
8*4882a593Smuzhiyun #include <linux/kernel.h>
9*4882a593Smuzhiyun #include <linux/string.h>
10*4882a593Smuzhiyun #include <linux/errno.h>
11*4882a593Smuzhiyun #include <linux/skbuff.h>
12*4882a593Smuzhiyun #include <linux/rtnetlink.h>
13*4882a593Smuzhiyun #include <linux/module.h>
14*4882a593Smuzhiyun #include <linux/init.h>
15*4882a593Smuzhiyun #include <linux/gfp.h>
16*4882a593Smuzhiyun #include <net/net_namespace.h>
17*4882a593Smuzhiyun #include <net/netlink.h>
18*4882a593Smuzhiyun #include <net/pkt_sched.h>
19*4882a593Smuzhiyun #include <linux/tc_act/tc_sample.h>
20*4882a593Smuzhiyun #include <net/tc_act/tc_sample.h>
21*4882a593Smuzhiyun #include <net/psample.h>
22*4882a593Smuzhiyun #include <net/pkt_cls.h>
23*4882a593Smuzhiyun 
24*4882a593Smuzhiyun #include <linux/if_arp.h>
25*4882a593Smuzhiyun 
26*4882a593Smuzhiyun static unsigned int sample_net_id;
27*4882a593Smuzhiyun static struct tc_action_ops act_sample_ops;
28*4882a593Smuzhiyun 
29*4882a593Smuzhiyun static const struct nla_policy sample_policy[TCA_SAMPLE_MAX + 1] = {
30*4882a593Smuzhiyun 	[TCA_SAMPLE_PARMS]		= { .len = sizeof(struct tc_sample) },
31*4882a593Smuzhiyun 	[TCA_SAMPLE_RATE]		= { .type = NLA_U32 },
32*4882a593Smuzhiyun 	[TCA_SAMPLE_TRUNC_SIZE]		= { .type = NLA_U32 },
33*4882a593Smuzhiyun 	[TCA_SAMPLE_PSAMPLE_GROUP]	= { .type = NLA_U32 },
34*4882a593Smuzhiyun };
35*4882a593Smuzhiyun 
tcf_sample_init(struct net * net,struct nlattr * nla,struct nlattr * est,struct tc_action ** a,int ovr,int bind,bool rtnl_held,struct tcf_proto * tp,u32 flags,struct netlink_ext_ack * extack)36*4882a593Smuzhiyun static int tcf_sample_init(struct net *net, struct nlattr *nla,
37*4882a593Smuzhiyun 			   struct nlattr *est, struct tc_action **a, int ovr,
38*4882a593Smuzhiyun 			   int bind, bool rtnl_held, struct tcf_proto *tp,
39*4882a593Smuzhiyun 			   u32 flags, struct netlink_ext_ack *extack)
40*4882a593Smuzhiyun {
41*4882a593Smuzhiyun 	struct tc_action_net *tn = net_generic(net, sample_net_id);
42*4882a593Smuzhiyun 	struct nlattr *tb[TCA_SAMPLE_MAX + 1];
43*4882a593Smuzhiyun 	struct psample_group *psample_group;
44*4882a593Smuzhiyun 	u32 psample_group_num, rate, index;
45*4882a593Smuzhiyun 	struct tcf_chain *goto_ch = NULL;
46*4882a593Smuzhiyun 	struct tc_sample *parm;
47*4882a593Smuzhiyun 	struct tcf_sample *s;
48*4882a593Smuzhiyun 	bool exists = false;
49*4882a593Smuzhiyun 	int ret, err;
50*4882a593Smuzhiyun 
51*4882a593Smuzhiyun 	if (!nla)
52*4882a593Smuzhiyun 		return -EINVAL;
53*4882a593Smuzhiyun 	ret = nla_parse_nested_deprecated(tb, TCA_SAMPLE_MAX, nla,
54*4882a593Smuzhiyun 					  sample_policy, NULL);
55*4882a593Smuzhiyun 	if (ret < 0)
56*4882a593Smuzhiyun 		return ret;
57*4882a593Smuzhiyun 	if (!tb[TCA_SAMPLE_PARMS] || !tb[TCA_SAMPLE_RATE] ||
58*4882a593Smuzhiyun 	    !tb[TCA_SAMPLE_PSAMPLE_GROUP])
59*4882a593Smuzhiyun 		return -EINVAL;
60*4882a593Smuzhiyun 
61*4882a593Smuzhiyun 	parm = nla_data(tb[TCA_SAMPLE_PARMS]);
62*4882a593Smuzhiyun 	index = parm->index;
63*4882a593Smuzhiyun 	err = tcf_idr_check_alloc(tn, &index, a, bind);
64*4882a593Smuzhiyun 	if (err < 0)
65*4882a593Smuzhiyun 		return err;
66*4882a593Smuzhiyun 	exists = err;
67*4882a593Smuzhiyun 	if (exists && bind)
68*4882a593Smuzhiyun 		return 0;
69*4882a593Smuzhiyun 
70*4882a593Smuzhiyun 	if (!exists) {
71*4882a593Smuzhiyun 		ret = tcf_idr_create(tn, index, est, a,
72*4882a593Smuzhiyun 				     &act_sample_ops, bind, true, 0);
73*4882a593Smuzhiyun 		if (ret) {
74*4882a593Smuzhiyun 			tcf_idr_cleanup(tn, index);
75*4882a593Smuzhiyun 			return ret;
76*4882a593Smuzhiyun 		}
77*4882a593Smuzhiyun 		ret = ACT_P_CREATED;
78*4882a593Smuzhiyun 	} else if (!ovr) {
79*4882a593Smuzhiyun 		tcf_idr_release(*a, bind);
80*4882a593Smuzhiyun 		return -EEXIST;
81*4882a593Smuzhiyun 	}
82*4882a593Smuzhiyun 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
83*4882a593Smuzhiyun 	if (err < 0)
84*4882a593Smuzhiyun 		goto release_idr;
85*4882a593Smuzhiyun 
86*4882a593Smuzhiyun 	rate = nla_get_u32(tb[TCA_SAMPLE_RATE]);
87*4882a593Smuzhiyun 	if (!rate) {
88*4882a593Smuzhiyun 		NL_SET_ERR_MSG(extack, "invalid sample rate");
89*4882a593Smuzhiyun 		err = -EINVAL;
90*4882a593Smuzhiyun 		goto put_chain;
91*4882a593Smuzhiyun 	}
92*4882a593Smuzhiyun 	psample_group_num = nla_get_u32(tb[TCA_SAMPLE_PSAMPLE_GROUP]);
93*4882a593Smuzhiyun 	psample_group = psample_group_get(net, psample_group_num);
94*4882a593Smuzhiyun 	if (!psample_group) {
95*4882a593Smuzhiyun 		err = -ENOMEM;
96*4882a593Smuzhiyun 		goto put_chain;
97*4882a593Smuzhiyun 	}
98*4882a593Smuzhiyun 
99*4882a593Smuzhiyun 	s = to_sample(*a);
100*4882a593Smuzhiyun 
101*4882a593Smuzhiyun 	spin_lock_bh(&s->tcf_lock);
102*4882a593Smuzhiyun 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
103*4882a593Smuzhiyun 	s->rate = rate;
104*4882a593Smuzhiyun 	s->psample_group_num = psample_group_num;
105*4882a593Smuzhiyun 	psample_group = rcu_replace_pointer(s->psample_group, psample_group,
106*4882a593Smuzhiyun 					    lockdep_is_held(&s->tcf_lock));
107*4882a593Smuzhiyun 
108*4882a593Smuzhiyun 	if (tb[TCA_SAMPLE_TRUNC_SIZE]) {
109*4882a593Smuzhiyun 		s->truncate = true;
110*4882a593Smuzhiyun 		s->trunc_size = nla_get_u32(tb[TCA_SAMPLE_TRUNC_SIZE]);
111*4882a593Smuzhiyun 	}
112*4882a593Smuzhiyun 	spin_unlock_bh(&s->tcf_lock);
113*4882a593Smuzhiyun 
114*4882a593Smuzhiyun 	if (psample_group)
115*4882a593Smuzhiyun 		psample_group_put(psample_group);
116*4882a593Smuzhiyun 	if (goto_ch)
117*4882a593Smuzhiyun 		tcf_chain_put_by_act(goto_ch);
118*4882a593Smuzhiyun 
119*4882a593Smuzhiyun 	return ret;
120*4882a593Smuzhiyun put_chain:
121*4882a593Smuzhiyun 	if (goto_ch)
122*4882a593Smuzhiyun 		tcf_chain_put_by_act(goto_ch);
123*4882a593Smuzhiyun release_idr:
124*4882a593Smuzhiyun 	tcf_idr_release(*a, bind);
125*4882a593Smuzhiyun 	return err;
126*4882a593Smuzhiyun }
127*4882a593Smuzhiyun 
tcf_sample_cleanup(struct tc_action * a)128*4882a593Smuzhiyun static void tcf_sample_cleanup(struct tc_action *a)
129*4882a593Smuzhiyun {
130*4882a593Smuzhiyun 	struct tcf_sample *s = to_sample(a);
131*4882a593Smuzhiyun 	struct psample_group *psample_group;
132*4882a593Smuzhiyun 
133*4882a593Smuzhiyun 	/* last reference to action, no need to lock */
134*4882a593Smuzhiyun 	psample_group = rcu_dereference_protected(s->psample_group, 1);
135*4882a593Smuzhiyun 	RCU_INIT_POINTER(s->psample_group, NULL);
136*4882a593Smuzhiyun 	if (psample_group)
137*4882a593Smuzhiyun 		psample_group_put(psample_group);
138*4882a593Smuzhiyun }
139*4882a593Smuzhiyun 
tcf_sample_dev_ok_push(struct net_device * dev)140*4882a593Smuzhiyun static bool tcf_sample_dev_ok_push(struct net_device *dev)
141*4882a593Smuzhiyun {
142*4882a593Smuzhiyun 	switch (dev->type) {
143*4882a593Smuzhiyun 	case ARPHRD_TUNNEL:
144*4882a593Smuzhiyun 	case ARPHRD_TUNNEL6:
145*4882a593Smuzhiyun 	case ARPHRD_SIT:
146*4882a593Smuzhiyun 	case ARPHRD_IPGRE:
147*4882a593Smuzhiyun 	case ARPHRD_IP6GRE:
148*4882a593Smuzhiyun 	case ARPHRD_VOID:
149*4882a593Smuzhiyun 	case ARPHRD_NONE:
150*4882a593Smuzhiyun 		return false;
151*4882a593Smuzhiyun 	default:
152*4882a593Smuzhiyun 		return true;
153*4882a593Smuzhiyun 	}
154*4882a593Smuzhiyun }
155*4882a593Smuzhiyun 
tcf_sample_act(struct sk_buff * skb,const struct tc_action * a,struct tcf_result * res)156*4882a593Smuzhiyun static int tcf_sample_act(struct sk_buff *skb, const struct tc_action *a,
157*4882a593Smuzhiyun 			  struct tcf_result *res)
158*4882a593Smuzhiyun {
159*4882a593Smuzhiyun 	struct tcf_sample *s = to_sample(a);
160*4882a593Smuzhiyun 	struct psample_group *psample_group;
161*4882a593Smuzhiyun 	int retval;
162*4882a593Smuzhiyun 	int size;
163*4882a593Smuzhiyun 	int iif;
164*4882a593Smuzhiyun 	int oif;
165*4882a593Smuzhiyun 
166*4882a593Smuzhiyun 	tcf_lastuse_update(&s->tcf_tm);
167*4882a593Smuzhiyun 	bstats_cpu_update(this_cpu_ptr(s->common.cpu_bstats), skb);
168*4882a593Smuzhiyun 	retval = READ_ONCE(s->tcf_action);
169*4882a593Smuzhiyun 
170*4882a593Smuzhiyun 	psample_group = rcu_dereference_bh(s->psample_group);
171*4882a593Smuzhiyun 
172*4882a593Smuzhiyun 	/* randomly sample packets according to rate */
173*4882a593Smuzhiyun 	if (psample_group && (prandom_u32() % s->rate == 0)) {
174*4882a593Smuzhiyun 		if (!skb_at_tc_ingress(skb)) {
175*4882a593Smuzhiyun 			iif = skb->skb_iif;
176*4882a593Smuzhiyun 			oif = skb->dev->ifindex;
177*4882a593Smuzhiyun 		} else {
178*4882a593Smuzhiyun 			iif = skb->dev->ifindex;
179*4882a593Smuzhiyun 			oif = 0;
180*4882a593Smuzhiyun 		}
181*4882a593Smuzhiyun 
182*4882a593Smuzhiyun 		/* on ingress, the mac header gets popped, so push it back */
183*4882a593Smuzhiyun 		if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
184*4882a593Smuzhiyun 			skb_push(skb, skb->mac_len);
185*4882a593Smuzhiyun 
186*4882a593Smuzhiyun 		size = s->truncate ? s->trunc_size : skb->len;
187*4882a593Smuzhiyun 		psample_sample_packet(psample_group, skb, size, iif, oif,
188*4882a593Smuzhiyun 				      s->rate);
189*4882a593Smuzhiyun 
190*4882a593Smuzhiyun 		if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
191*4882a593Smuzhiyun 			skb_pull(skb, skb->mac_len);
192*4882a593Smuzhiyun 	}
193*4882a593Smuzhiyun 
194*4882a593Smuzhiyun 	return retval;
195*4882a593Smuzhiyun }
196*4882a593Smuzhiyun 
tcf_sample_dump(struct sk_buff * skb,struct tc_action * a,int bind,int ref)197*4882a593Smuzhiyun static int tcf_sample_dump(struct sk_buff *skb, struct tc_action *a,
198*4882a593Smuzhiyun 			   int bind, int ref)
199*4882a593Smuzhiyun {
200*4882a593Smuzhiyun 	unsigned char *b = skb_tail_pointer(skb);
201*4882a593Smuzhiyun 	struct tcf_sample *s = to_sample(a);
202*4882a593Smuzhiyun 	struct tc_sample opt = {
203*4882a593Smuzhiyun 		.index      = s->tcf_index,
204*4882a593Smuzhiyun 		.refcnt     = refcount_read(&s->tcf_refcnt) - ref,
205*4882a593Smuzhiyun 		.bindcnt    = atomic_read(&s->tcf_bindcnt) - bind,
206*4882a593Smuzhiyun 	};
207*4882a593Smuzhiyun 	struct tcf_t t;
208*4882a593Smuzhiyun 
209*4882a593Smuzhiyun 	spin_lock_bh(&s->tcf_lock);
210*4882a593Smuzhiyun 	opt.action = s->tcf_action;
211*4882a593Smuzhiyun 	if (nla_put(skb, TCA_SAMPLE_PARMS, sizeof(opt), &opt))
212*4882a593Smuzhiyun 		goto nla_put_failure;
213*4882a593Smuzhiyun 
214*4882a593Smuzhiyun 	tcf_tm_dump(&t, &s->tcf_tm);
215*4882a593Smuzhiyun 	if (nla_put_64bit(skb, TCA_SAMPLE_TM, sizeof(t), &t, TCA_SAMPLE_PAD))
216*4882a593Smuzhiyun 		goto nla_put_failure;
217*4882a593Smuzhiyun 
218*4882a593Smuzhiyun 	if (nla_put_u32(skb, TCA_SAMPLE_RATE, s->rate))
219*4882a593Smuzhiyun 		goto nla_put_failure;
220*4882a593Smuzhiyun 
221*4882a593Smuzhiyun 	if (s->truncate)
222*4882a593Smuzhiyun 		if (nla_put_u32(skb, TCA_SAMPLE_TRUNC_SIZE, s->trunc_size))
223*4882a593Smuzhiyun 			goto nla_put_failure;
224*4882a593Smuzhiyun 
225*4882a593Smuzhiyun 	if (nla_put_u32(skb, TCA_SAMPLE_PSAMPLE_GROUP, s->psample_group_num))
226*4882a593Smuzhiyun 		goto nla_put_failure;
227*4882a593Smuzhiyun 	spin_unlock_bh(&s->tcf_lock);
228*4882a593Smuzhiyun 
229*4882a593Smuzhiyun 	return skb->len;
230*4882a593Smuzhiyun 
231*4882a593Smuzhiyun nla_put_failure:
232*4882a593Smuzhiyun 	spin_unlock_bh(&s->tcf_lock);
233*4882a593Smuzhiyun 	nlmsg_trim(skb, b);
234*4882a593Smuzhiyun 	return -1;
235*4882a593Smuzhiyun }
236*4882a593Smuzhiyun 
tcf_sample_walker(struct net * net,struct sk_buff * skb,struct netlink_callback * cb,int type,const struct tc_action_ops * ops,struct netlink_ext_ack * extack)237*4882a593Smuzhiyun static int tcf_sample_walker(struct net *net, struct sk_buff *skb,
238*4882a593Smuzhiyun 			     struct netlink_callback *cb, int type,
239*4882a593Smuzhiyun 			     const struct tc_action_ops *ops,
240*4882a593Smuzhiyun 			     struct netlink_ext_ack *extack)
241*4882a593Smuzhiyun {
242*4882a593Smuzhiyun 	struct tc_action_net *tn = net_generic(net, sample_net_id);
243*4882a593Smuzhiyun 
244*4882a593Smuzhiyun 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
245*4882a593Smuzhiyun }
246*4882a593Smuzhiyun 
tcf_sample_search(struct net * net,struct tc_action ** a,u32 index)247*4882a593Smuzhiyun static int tcf_sample_search(struct net *net, struct tc_action **a, u32 index)
248*4882a593Smuzhiyun {
249*4882a593Smuzhiyun 	struct tc_action_net *tn = net_generic(net, sample_net_id);
250*4882a593Smuzhiyun 
251*4882a593Smuzhiyun 	return tcf_idr_search(tn, a, index);
252*4882a593Smuzhiyun }
253*4882a593Smuzhiyun 
tcf_psample_group_put(void * priv)254*4882a593Smuzhiyun static void tcf_psample_group_put(void *priv)
255*4882a593Smuzhiyun {
256*4882a593Smuzhiyun 	struct psample_group *group = priv;
257*4882a593Smuzhiyun 
258*4882a593Smuzhiyun 	psample_group_put(group);
259*4882a593Smuzhiyun }
260*4882a593Smuzhiyun 
261*4882a593Smuzhiyun static struct psample_group *
tcf_sample_get_group(const struct tc_action * a,tc_action_priv_destructor * destructor)262*4882a593Smuzhiyun tcf_sample_get_group(const struct tc_action *a,
263*4882a593Smuzhiyun 		     tc_action_priv_destructor *destructor)
264*4882a593Smuzhiyun {
265*4882a593Smuzhiyun 	struct tcf_sample *s = to_sample(a);
266*4882a593Smuzhiyun 	struct psample_group *group;
267*4882a593Smuzhiyun 
268*4882a593Smuzhiyun 	group = rcu_dereference_protected(s->psample_group,
269*4882a593Smuzhiyun 					  lockdep_is_held(&s->tcf_lock));
270*4882a593Smuzhiyun 	if (group) {
271*4882a593Smuzhiyun 		psample_group_take(group);
272*4882a593Smuzhiyun 		*destructor = tcf_psample_group_put;
273*4882a593Smuzhiyun 	}
274*4882a593Smuzhiyun 
275*4882a593Smuzhiyun 	return group;
276*4882a593Smuzhiyun }
277*4882a593Smuzhiyun 
278*4882a593Smuzhiyun static struct tc_action_ops act_sample_ops = {
279*4882a593Smuzhiyun 	.kind	  = "sample",
280*4882a593Smuzhiyun 	.id	  = TCA_ID_SAMPLE,
281*4882a593Smuzhiyun 	.owner	  = THIS_MODULE,
282*4882a593Smuzhiyun 	.act	  = tcf_sample_act,
283*4882a593Smuzhiyun 	.dump	  = tcf_sample_dump,
284*4882a593Smuzhiyun 	.init	  = tcf_sample_init,
285*4882a593Smuzhiyun 	.cleanup  = tcf_sample_cleanup,
286*4882a593Smuzhiyun 	.walk	  = tcf_sample_walker,
287*4882a593Smuzhiyun 	.lookup	  = tcf_sample_search,
288*4882a593Smuzhiyun 	.get_psample_group = tcf_sample_get_group,
289*4882a593Smuzhiyun 	.size	  = sizeof(struct tcf_sample),
290*4882a593Smuzhiyun };
291*4882a593Smuzhiyun 
sample_init_net(struct net * net)292*4882a593Smuzhiyun static __net_init int sample_init_net(struct net *net)
293*4882a593Smuzhiyun {
294*4882a593Smuzhiyun 	struct tc_action_net *tn = net_generic(net, sample_net_id);
295*4882a593Smuzhiyun 
296*4882a593Smuzhiyun 	return tc_action_net_init(net, tn, &act_sample_ops);
297*4882a593Smuzhiyun }
298*4882a593Smuzhiyun 
sample_exit_net(struct list_head * net_list)299*4882a593Smuzhiyun static void __net_exit sample_exit_net(struct list_head *net_list)
300*4882a593Smuzhiyun {
301*4882a593Smuzhiyun 	tc_action_net_exit(net_list, sample_net_id);
302*4882a593Smuzhiyun }
303*4882a593Smuzhiyun 
304*4882a593Smuzhiyun static struct pernet_operations sample_net_ops = {
305*4882a593Smuzhiyun 	.init = sample_init_net,
306*4882a593Smuzhiyun 	.exit_batch = sample_exit_net,
307*4882a593Smuzhiyun 	.id   = &sample_net_id,
308*4882a593Smuzhiyun 	.size = sizeof(struct tc_action_net),
309*4882a593Smuzhiyun };
310*4882a593Smuzhiyun 
sample_init_module(void)311*4882a593Smuzhiyun static int __init sample_init_module(void)
312*4882a593Smuzhiyun {
313*4882a593Smuzhiyun 	return tcf_register_action(&act_sample_ops, &sample_net_ops);
314*4882a593Smuzhiyun }
315*4882a593Smuzhiyun 
sample_cleanup_module(void)316*4882a593Smuzhiyun static void __exit sample_cleanup_module(void)
317*4882a593Smuzhiyun {
318*4882a593Smuzhiyun 	tcf_unregister_action(&act_sample_ops, &sample_net_ops);
319*4882a593Smuzhiyun }
320*4882a593Smuzhiyun 
321*4882a593Smuzhiyun module_init(sample_init_module);
322*4882a593Smuzhiyun module_exit(sample_cleanup_module);
323*4882a593Smuzhiyun 
324*4882a593Smuzhiyun MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
325*4882a593Smuzhiyun MODULE_DESCRIPTION("Packet sampling action");
326*4882a593Smuzhiyun MODULE_LICENSE("GPL v2");
327