1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * net/sched/act_sample.c - Packet sampling tc action
4 * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com>
5 */
6
7 #include <linux/types.h>
8 #include <linux/kernel.h>
9 #include <linux/string.h>
10 #include <linux/errno.h>
11 #include <linux/skbuff.h>
12 #include <linux/rtnetlink.h>
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/gfp.h>
16 #include <net/net_namespace.h>
17 #include <net/netlink.h>
18 #include <net/pkt_sched.h>
19 #include <linux/tc_act/tc_sample.h>
20 #include <net/tc_act/tc_sample.h>
21 #include <net/psample.h>
22 #include <net/pkt_cls.h>
23
24 #include <linux/if_arp.h>
25
26 static unsigned int sample_net_id;
27 static struct tc_action_ops act_sample_ops;
28
29 static const struct nla_policy sample_policy[TCA_SAMPLE_MAX + 1] = {
30 [TCA_SAMPLE_PARMS] = { .len = sizeof(struct tc_sample) },
31 [TCA_SAMPLE_RATE] = { .type = NLA_U32 },
32 [TCA_SAMPLE_TRUNC_SIZE] = { .type = NLA_U32 },
33 [TCA_SAMPLE_PSAMPLE_GROUP] = { .type = NLA_U32 },
34 };
35
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 static int tcf_sample_init(struct net *net, struct nlattr *nla,
37 struct nlattr *est, struct tc_action **a, int ovr,
38 int bind, bool rtnl_held, struct tcf_proto *tp,
39 u32 flags, struct netlink_ext_ack *extack)
40 {
41 struct tc_action_net *tn = net_generic(net, sample_net_id);
42 struct nlattr *tb[TCA_SAMPLE_MAX + 1];
43 struct psample_group *psample_group;
44 u32 psample_group_num, rate, index;
45 struct tcf_chain *goto_ch = NULL;
46 struct tc_sample *parm;
47 struct tcf_sample *s;
48 bool exists = false;
49 int ret, err;
50
51 if (!nla)
52 return -EINVAL;
53 ret = nla_parse_nested_deprecated(tb, TCA_SAMPLE_MAX, nla,
54 sample_policy, NULL);
55 if (ret < 0)
56 return ret;
57
58 if (!tb[TCA_SAMPLE_PARMS])
59 return -EINVAL;
60
61 parm = nla_data(tb[TCA_SAMPLE_PARMS]);
62 index = parm->index;
63 err = tcf_idr_check_alloc(tn, &index, a, bind);
64 if (err < 0)
65 return err;
66 exists = err;
67 if (exists && bind)
68 return 0;
69
70 if (!exists) {
71 ret = tcf_idr_create(tn, index, est, a,
72 &act_sample_ops, bind, true, flags);
73 if (ret) {
74 tcf_idr_cleanup(tn, index);
75 return ret;
76 }
77 ret = ACT_P_CREATED;
78 } else if (!ovr) {
79 tcf_idr_release(*a, bind);
80 return -EEXIST;
81 }
82
83 if (!tb[TCA_SAMPLE_RATE] || !tb[TCA_SAMPLE_PSAMPLE_GROUP]) {
84 NL_SET_ERR_MSG(extack, "sample rate and group are required");
85 err = -EINVAL;
86 goto release_idr;
87 }
88
89 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
90 if (err < 0)
91 goto release_idr;
92
93 rate = nla_get_u32(tb[TCA_SAMPLE_RATE]);
94 if (!rate) {
95 NL_SET_ERR_MSG(extack, "invalid sample rate");
96 err = -EINVAL;
97 goto put_chain;
98 }
99 psample_group_num = nla_get_u32(tb[TCA_SAMPLE_PSAMPLE_GROUP]);
100 psample_group = psample_group_get(net, psample_group_num);
101 if (!psample_group) {
102 err = -ENOMEM;
103 goto put_chain;
104 }
105
106 s = to_sample(*a);
107
108 spin_lock_bh(&s->tcf_lock);
109 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
110 s->rate = rate;
111 s->psample_group_num = psample_group_num;
112 psample_group = rcu_replace_pointer(s->psample_group, psample_group,
113 lockdep_is_held(&s->tcf_lock));
114
115 if (tb[TCA_SAMPLE_TRUNC_SIZE]) {
116 s->truncate = true;
117 s->trunc_size = nla_get_u32(tb[TCA_SAMPLE_TRUNC_SIZE]);
118 }
119 spin_unlock_bh(&s->tcf_lock);
120
121 if (psample_group)
122 psample_group_put(psample_group);
123 if (goto_ch)
124 tcf_chain_put_by_act(goto_ch);
125
126 return ret;
127 put_chain:
128 if (goto_ch)
129 tcf_chain_put_by_act(goto_ch);
130 release_idr:
131 tcf_idr_release(*a, bind);
132 return err;
133 }
134
tcf_sample_cleanup(struct tc_action * a)135 static void tcf_sample_cleanup(struct tc_action *a)
136 {
137 struct tcf_sample *s = to_sample(a);
138 struct psample_group *psample_group;
139
140 /* last reference to action, no need to lock */
141 psample_group = rcu_dereference_protected(s->psample_group, 1);
142 RCU_INIT_POINTER(s->psample_group, NULL);
143 if (psample_group)
144 psample_group_put(psample_group);
145 }
146
tcf_sample_dev_ok_push(struct net_device * dev)147 static bool tcf_sample_dev_ok_push(struct net_device *dev)
148 {
149 switch (dev->type) {
150 case ARPHRD_TUNNEL:
151 case ARPHRD_TUNNEL6:
152 case ARPHRD_SIT:
153 case ARPHRD_IPGRE:
154 case ARPHRD_IP6GRE:
155 case ARPHRD_VOID:
156 case ARPHRD_NONE:
157 return false;
158 default:
159 return true;
160 }
161 }
162
tcf_sample_act(struct sk_buff * skb,const struct tc_action * a,struct tcf_result * res)163 static int tcf_sample_act(struct sk_buff *skb, const struct tc_action *a,
164 struct tcf_result *res)
165 {
166 struct tcf_sample *s = to_sample(a);
167 struct psample_group *psample_group;
168 int retval;
169 int size;
170 int iif;
171 int oif;
172
173 tcf_lastuse_update(&s->tcf_tm);
174 bstats_cpu_update(this_cpu_ptr(s->common.cpu_bstats), skb);
175 retval = READ_ONCE(s->tcf_action);
176
177 psample_group = rcu_dereference_bh(s->psample_group);
178
179 /* randomly sample packets according to rate */
180 if (psample_group && (prandom_u32() % s->rate == 0)) {
181 if (!skb_at_tc_ingress(skb)) {
182 iif = skb->skb_iif;
183 oif = skb->dev->ifindex;
184 } else {
185 iif = skb->dev->ifindex;
186 oif = 0;
187 }
188
189 /* on ingress, the mac header gets popped, so push it back */
190 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
191 skb_push(skb, skb->mac_len);
192
193 size = s->truncate ? s->trunc_size : skb->len;
194 psample_sample_packet(psample_group, skb, size, iif, oif,
195 s->rate);
196
197 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
198 skb_pull(skb, skb->mac_len);
199 }
200
201 return retval;
202 }
203
tcf_sample_dump(struct sk_buff * skb,struct tc_action * a,int bind,int ref)204 static int tcf_sample_dump(struct sk_buff *skb, struct tc_action *a,
205 int bind, int ref)
206 {
207 unsigned char *b = skb_tail_pointer(skb);
208 struct tcf_sample *s = to_sample(a);
209 struct tc_sample opt = {
210 .index = s->tcf_index,
211 .refcnt = refcount_read(&s->tcf_refcnt) - ref,
212 .bindcnt = atomic_read(&s->tcf_bindcnt) - bind,
213 };
214 struct tcf_t t;
215
216 spin_lock_bh(&s->tcf_lock);
217 opt.action = s->tcf_action;
218 if (nla_put(skb, TCA_SAMPLE_PARMS, sizeof(opt), &opt))
219 goto nla_put_failure;
220
221 tcf_tm_dump(&t, &s->tcf_tm);
222 if (nla_put_64bit(skb, TCA_SAMPLE_TM, sizeof(t), &t, TCA_SAMPLE_PAD))
223 goto nla_put_failure;
224
225 if (nla_put_u32(skb, TCA_SAMPLE_RATE, s->rate))
226 goto nla_put_failure;
227
228 if (s->truncate)
229 if (nla_put_u32(skb, TCA_SAMPLE_TRUNC_SIZE, s->trunc_size))
230 goto nla_put_failure;
231
232 if (nla_put_u32(skb, TCA_SAMPLE_PSAMPLE_GROUP, s->psample_group_num))
233 goto nla_put_failure;
234 spin_unlock_bh(&s->tcf_lock);
235
236 return skb->len;
237
238 nla_put_failure:
239 spin_unlock_bh(&s->tcf_lock);
240 nlmsg_trim(skb, b);
241 return -1;
242 }
243
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)244 static int tcf_sample_walker(struct net *net, struct sk_buff *skb,
245 struct netlink_callback *cb, int type,
246 const struct tc_action_ops *ops,
247 struct netlink_ext_ack *extack)
248 {
249 struct tc_action_net *tn = net_generic(net, sample_net_id);
250
251 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
252 }
253
tcf_sample_search(struct net * net,struct tc_action ** a,u32 index)254 static int tcf_sample_search(struct net *net, struct tc_action **a, u32 index)
255 {
256 struct tc_action_net *tn = net_generic(net, sample_net_id);
257
258 return tcf_idr_search(tn, a, index);
259 }
260
tcf_psample_group_put(void * priv)261 static void tcf_psample_group_put(void *priv)
262 {
263 struct psample_group *group = priv;
264
265 psample_group_put(group);
266 }
267
268 static struct psample_group *
tcf_sample_get_group(const struct tc_action * a,tc_action_priv_destructor * destructor)269 tcf_sample_get_group(const struct tc_action *a,
270 tc_action_priv_destructor *destructor)
271 {
272 struct tcf_sample *s = to_sample(a);
273 struct psample_group *group;
274
275 group = rcu_dereference_protected(s->psample_group,
276 lockdep_is_held(&s->tcf_lock));
277 if (group) {
278 psample_group_take(group);
279 *destructor = tcf_psample_group_put;
280 }
281
282 return group;
283 }
284
285 static struct tc_action_ops act_sample_ops = {
286 .kind = "sample",
287 .id = TCA_ID_SAMPLE,
288 .owner = THIS_MODULE,
289 .act = tcf_sample_act,
290 .dump = tcf_sample_dump,
291 .init = tcf_sample_init,
292 .cleanup = tcf_sample_cleanup,
293 .walk = tcf_sample_walker,
294 .lookup = tcf_sample_search,
295 .get_psample_group = tcf_sample_get_group,
296 .size = sizeof(struct tcf_sample),
297 };
298
sample_init_net(struct net * net)299 static __net_init int sample_init_net(struct net *net)
300 {
301 struct tc_action_net *tn = net_generic(net, sample_net_id);
302
303 return tc_action_net_init(net, tn, &act_sample_ops);
304 }
305
sample_exit_net(struct list_head * net_list)306 static void __net_exit sample_exit_net(struct list_head *net_list)
307 {
308 tc_action_net_exit(net_list, sample_net_id);
309 }
310
311 static struct pernet_operations sample_net_ops = {
312 .init = sample_init_net,
313 .exit_batch = sample_exit_net,
314 .id = &sample_net_id,
315 .size = sizeof(struct tc_action_net),
316 };
317
sample_init_module(void)318 static int __init sample_init_module(void)
319 {
320 return tcf_register_action(&act_sample_ops, &sample_net_ops);
321 }
322
sample_cleanup_module(void)323 static void __exit sample_cleanup_module(void)
324 {
325 tcf_unregister_action(&act_sample_ops, &sample_net_ops);
326 }
327
328 module_init(sample_init_module);
329 module_exit(sample_cleanup_module);
330
331 MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
332 MODULE_DESCRIPTION("Packet sampling action");
333 MODULE_LICENSE("GPL v2");
334