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 #include <net/tc_wrapper.h>
24
25 #include <linux/if_arp.h>
26
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
36 static int tcf_sample_init(struct net *net, struct nlattr *nla,
37 struct nlattr *est, struct tc_action **a,
38 struct tcf_proto *tp,
39 u32 flags, struct netlink_ext_ack *extack)
40 {
41 struct tc_action_net *tn = net_generic(net, act_sample_ops.net_id);
42 bool bind = flags & TCA_ACT_FLAGS_BIND;
43 struct nlattr *tb[TCA_SAMPLE_MAX + 1];
44 struct psample_group *psample_group;
45 u32 psample_group_num, rate, index;
46 struct tcf_chain *goto_ch = NULL;
47 struct tc_sample *parm;
48 struct tcf_sample *s;
49 bool exists = false;
50 int ret, err;
51
52 if (!nla)
53 return -EINVAL;
54 ret = nla_parse_nested_deprecated(tb, TCA_SAMPLE_MAX, nla,
55 sample_policy, NULL);
56 if (ret < 0)
57 return ret;
58
59 if (!tb[TCA_SAMPLE_PARMS])
60 return -EINVAL;
61
62 parm = nla_data(tb[TCA_SAMPLE_PARMS]);
63 index = parm->index;
64 err = tcf_idr_check_alloc(tn, &index, a, bind);
65 if (err < 0)
66 return err;
67 exists = err;
68 if (exists && bind)
69 return ACT_P_BOUND;
70
71 if (!exists) {
72 ret = tcf_idr_create(tn, index, est, a,
73 &act_sample_ops, bind, true, flags);
74 if (ret) {
75 tcf_idr_cleanup(tn, index);
76 return ret;
77 }
78 ret = ACT_P_CREATED;
79 } else if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
80 tcf_idr_release(*a, bind);
81 return -EEXIST;
82 }
83
84 if (!tb[TCA_SAMPLE_RATE] || !tb[TCA_SAMPLE_PSAMPLE_GROUP]) {
85 NL_SET_ERR_MSG(extack, "sample rate and group are required");
86 err = -EINVAL;
87 goto release_idr;
88 }
89
90 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
91 if (err < 0)
92 goto release_idr;
93
94 rate = nla_get_u32(tb[TCA_SAMPLE_RATE]);
95 if (!rate) {
96 NL_SET_ERR_MSG(extack, "invalid sample rate");
97 err = -EINVAL;
98 goto put_chain;
99 }
100 psample_group_num = nla_get_u32(tb[TCA_SAMPLE_PSAMPLE_GROUP]);
101 psample_group = psample_group_get(net, psample_group_num);
102 if (!psample_group) {
103 err = -ENOMEM;
104 goto put_chain;
105 }
106
107 s = to_sample(*a);
108
109 spin_lock_bh(&s->tcf_lock);
110 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
111 s->rate = rate;
112 s->psample_group_num = psample_group_num;
113 psample_group = rcu_replace_pointer(s->psample_group, psample_group,
114 lockdep_is_held(&s->tcf_lock));
115
116 if (tb[TCA_SAMPLE_TRUNC_SIZE]) {
117 s->truncate = true;
118 s->trunc_size = nla_get_u32(tb[TCA_SAMPLE_TRUNC_SIZE]);
119 }
120 spin_unlock_bh(&s->tcf_lock);
121
122 if (psample_group)
123 psample_group_put(psample_group);
124 if (goto_ch)
125 tcf_chain_put_by_act(goto_ch);
126
127 return ret;
128 put_chain:
129 if (goto_ch)
130 tcf_chain_put_by_act(goto_ch);
131 release_idr:
132 tcf_idr_release(*a, bind);
133 return err;
134 }
135
136 static void tcf_sample_cleanup(struct tc_action *a)
137 {
138 struct tcf_sample *s = to_sample(a);
139 struct psample_group *psample_group;
140
141 /* last reference to action, no need to lock */
142 psample_group = rcu_dereference_protected(s->psample_group, 1);
143 RCU_INIT_POINTER(s->psample_group, NULL);
144 if (psample_group)
145 psample_group_put(psample_group);
146 }
147
148 static bool tcf_sample_dev_ok_push(struct net_device *dev)
149 {
150 switch (dev->type) {
151 case ARPHRD_TUNNEL:
152 case ARPHRD_TUNNEL6:
153 case ARPHRD_SIT:
154 case ARPHRD_IPGRE:
155 case ARPHRD_IP6GRE:
156 case ARPHRD_VOID:
157 case ARPHRD_NONE:
158 return false;
159 default:
160 return true;
161 }
162 }
163
164 TC_INDIRECT_SCOPE int tcf_sample_act(struct sk_buff *skb,
165 const struct tc_action *a,
166 struct tcf_result *res)
167 {
168 struct tcf_sample *s = to_sample(a);
169 struct psample_group *psample_group;
170 u8 cookie_data[TC_COOKIE_MAX_SIZE];
171 struct psample_metadata md = {};
172 struct tc_cookie *user_cookie;
173 int retval;
174
175 tcf_lastuse_update(&s->tcf_tm);
176 bstats_update(this_cpu_ptr(s->common.cpu_bstats), skb);
177 retval = READ_ONCE(s->tcf_action);
178
179 psample_group = rcu_dereference_bh(s->psample_group);
180
181 /* randomly sample packets according to rate */
182 if (psample_group && (get_random_u32_below(s->rate) == 0)) {
183 if (!skb_at_tc_ingress(skb)) {
184 md.in_ifindex = skb->skb_iif;
185 md.out_ifindex = skb->dev->ifindex;
186 } else {
187 md.in_ifindex = skb->dev->ifindex;
188 }
189
190 /* on ingress, the mac header gets popped, so push it back */
191 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
192 skb_push(skb, skb->mac_len);
193
194 rcu_read_lock();
195 user_cookie = rcu_dereference(a->user_cookie);
196 if (user_cookie) {
197 memcpy(cookie_data, user_cookie->data,
198 user_cookie->len);
199 md.user_cookie = cookie_data;
200 md.user_cookie_len = user_cookie->len;
201 }
202 rcu_read_unlock();
203
204 md.trunc_size = s->truncate ? s->trunc_size : skb->len;
205 psample_sample_packet(psample_group, skb, s->rate, &md);
206
207 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
208 skb_pull(skb, skb->mac_len);
209 }
210
211 return retval;
212 }
213
214 static void tcf_sample_stats_update(struct tc_action *a, u64 bytes, u64 packets,
215 u64 drops, u64 lastuse, bool hw)
216 {
217 struct tcf_sample *s = to_sample(a);
218 struct tcf_t *tm = &s->tcf_tm;
219
220 tcf_action_update_stats(a, bytes, packets, drops, hw);
221 tm->lastuse = max_t(u64, tm->lastuse, lastuse);
222 }
223
224 static int tcf_sample_dump(struct sk_buff *skb, struct tc_action *a,
225 int bind, int ref)
226 {
227 unsigned char *b = skb_tail_pointer(skb);
228 struct tcf_sample *s = to_sample(a);
229 struct tc_sample opt = {
230 .index = s->tcf_index,
231 .refcnt = refcount_read(&s->tcf_refcnt) - ref,
232 .bindcnt = atomic_read(&s->tcf_bindcnt) - bind,
233 };
234 struct tcf_t t;
235
236 spin_lock_bh(&s->tcf_lock);
237 opt.action = s->tcf_action;
238 if (nla_put(skb, TCA_SAMPLE_PARMS, sizeof(opt), &opt))
239 goto nla_put_failure;
240
241 tcf_tm_dump(&t, &s->tcf_tm);
242 if (nla_put_64bit(skb, TCA_SAMPLE_TM, sizeof(t), &t, TCA_SAMPLE_PAD))
243 goto nla_put_failure;
244
245 if (nla_put_u32(skb, TCA_SAMPLE_RATE, s->rate))
246 goto nla_put_failure;
247
248 if (s->truncate)
249 if (nla_put_u32(skb, TCA_SAMPLE_TRUNC_SIZE, s->trunc_size))
250 goto nla_put_failure;
251
252 if (nla_put_u32(skb, TCA_SAMPLE_PSAMPLE_GROUP, s->psample_group_num))
253 goto nla_put_failure;
254 spin_unlock_bh(&s->tcf_lock);
255
256 return skb->len;
257
258 nla_put_failure:
259 spin_unlock_bh(&s->tcf_lock);
260 nlmsg_trim(skb, b);
261 return -1;
262 }
263
264 static void tcf_psample_group_put(void *priv)
265 {
266 struct psample_group *group = priv;
267
268 psample_group_put(group);
269 }
270
271 static struct psample_group *
272 tcf_sample_get_group(const struct tc_action *a,
273 tc_action_priv_destructor *destructor)
274 {
275 struct tcf_sample *s = to_sample(a);
276 struct psample_group *group;
277
278 group = rcu_dereference_protected(s->psample_group,
279 lockdep_is_held(&s->tcf_lock));
280 if (group) {
281 psample_group_take(group);
282 *destructor = tcf_psample_group_put;
283 }
284
285 return group;
286 }
287
288 static void tcf_offload_sample_get_group(struct flow_action_entry *entry,
289 const struct tc_action *act)
290 {
291 entry->sample.psample_group =
292 act->ops->get_psample_group(act, &entry->destructor);
293 entry->destructor_priv = entry->sample.psample_group;
294 }
295
296 static int tcf_sample_offload_act_setup(struct tc_action *act, void *entry_data,
297 u32 *index_inc, bool bind,
298 struct netlink_ext_ack *extack)
299 {
300 if (bind) {
301 struct flow_action_entry *entry = entry_data;
302
303 entry->id = FLOW_ACTION_SAMPLE;
304 entry->sample.trunc_size = tcf_sample_trunc_size(act);
305 entry->sample.truncate = tcf_sample_truncate(act);
306 entry->sample.rate = tcf_sample_rate(act);
307 tcf_offload_sample_get_group(entry, act);
308 *index_inc = 1;
309 } else {
310 struct flow_offload_action *fl_action = entry_data;
311
312 fl_action->id = FLOW_ACTION_SAMPLE;
313 }
314
315 return 0;
316 }
317
318 static struct tc_action_ops act_sample_ops = {
319 .kind = "sample",
320 .id = TCA_ID_SAMPLE,
321 .owner = THIS_MODULE,
322 .act = tcf_sample_act,
323 .stats_update = tcf_sample_stats_update,
324 .dump = tcf_sample_dump,
325 .init = tcf_sample_init,
326 .cleanup = tcf_sample_cleanup,
327 .get_psample_group = tcf_sample_get_group,
328 .offload_act_setup = tcf_sample_offload_act_setup,
329 .size = sizeof(struct tcf_sample),
330 };
331 MODULE_ALIAS_NET_ACT("sample");
332
333 static __net_init int sample_init_net(struct net *net)
334 {
335 struct tc_action_net *tn = net_generic(net, act_sample_ops.net_id);
336
337 return tc_action_net_init(net, tn, &act_sample_ops);
338 }
339
340 static void __net_exit sample_exit_net(struct list_head *net_list)
341 {
342 tc_action_net_exit(net_list, act_sample_ops.net_id);
343 }
344
345 static struct pernet_operations sample_net_ops = {
346 .init = sample_init_net,
347 .exit_batch = sample_exit_net,
348 .id = &act_sample_ops.net_id,
349 .size = sizeof(struct tc_action_net),
350 };
351
352 static int __init sample_init_module(void)
353 {
354 return tcf_register_action(&act_sample_ops, &sample_net_ops);
355 }
356
357 static void __exit sample_cleanup_module(void)
358 {
359 tcf_unregister_action(&act_sample_ops, &sample_net_ops);
360 }
361
362 module_init(sample_init_module);
363 module_exit(sample_cleanup_module);
364
365 MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
366 MODULE_DESCRIPTION("Packet sampling action");
367 MODULE_LICENSE("GPL v2");
368
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.