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,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,
38 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 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 0;
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
tcf_sample_cleanup(struct tc_action * a)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
tcf_sample_dev_ok_push(struct net_device * dev)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
tcf_sample_act(struct sk_buff * skb,const struct tc_action * a,struct tcf_result * res)164 static int tcf_sample_act(struct sk_buff *skb, const struct tc_action *a,
165 struct tcf_result *res)
166 {
167 struct tcf_sample *s = to_sample(a);
168 struct psample_group *psample_group;
169 struct psample_metadata md = {};
170 int retval;
171
172 tcf_lastuse_update(&s->tcf_tm);
173 bstats_cpu_update(this_cpu_ptr(s->common.cpu_bstats), skb);
174 retval = READ_ONCE(s->tcf_action);
175
176 psample_group = rcu_dereference_bh(s->psample_group);
177
178 /* randomly sample packets according to rate */
179 if (psample_group && (prandom_u32() % s->rate == 0)) {
180 if (!skb_at_tc_ingress(skb)) {
181 md.in_ifindex = skb->skb_iif;
182 md.out_ifindex = skb->dev->ifindex;
183 } else {
184 md.in_ifindex = skb->dev->ifindex;
185 }
186
187 /* on ingress, the mac header gets popped, so push it back */
188 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
189 skb_push(skb, skb->mac_len);
190
191 md.trunc_size = s->truncate ? s->trunc_size : skb->len;
192 psample_sample_packet(psample_group, skb, s->rate, &md);
193
194 if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev))
195 skb_pull(skb, skb->mac_len);
196 }
197
198 return retval;
199 }
200
tcf_sample_stats_update(struct tc_action * a,u64 bytes,u64 packets,u64 drops,u64 lastuse,bool hw)201 static void tcf_sample_stats_update(struct tc_action *a, u64 bytes, u64 packets,
202 u64 drops, u64 lastuse, bool hw)
203 {
204 struct tcf_sample *s = to_sample(a);
205 struct tcf_t *tm = &s->tcf_tm;
206
207 tcf_action_update_stats(a, bytes, packets, drops, hw);
208 tm->lastuse = max_t(u64, tm->lastuse, lastuse);
209 }
210
tcf_sample_dump(struct sk_buff * skb,struct tc_action * a,int bind,int ref)211 static int tcf_sample_dump(struct sk_buff *skb, struct tc_action *a,
212 int bind, int ref)
213 {
214 unsigned char *b = skb_tail_pointer(skb);
215 struct tcf_sample *s = to_sample(a);
216 struct tc_sample opt = {
217 .index = s->tcf_index,
218 .refcnt = refcount_read(&s->tcf_refcnt) - ref,
219 .bindcnt = atomic_read(&s->tcf_bindcnt) - bind,
220 };
221 struct tcf_t t;
222
223 spin_lock_bh(&s->tcf_lock);
224 opt.action = s->tcf_action;
225 if (nla_put(skb, TCA_SAMPLE_PARMS, sizeof(opt), &opt))
226 goto nla_put_failure;
227
228 tcf_tm_dump(&t, &s->tcf_tm);
229 if (nla_put_64bit(skb, TCA_SAMPLE_TM, sizeof(t), &t, TCA_SAMPLE_PAD))
230 goto nla_put_failure;
231
232 if (nla_put_u32(skb, TCA_SAMPLE_RATE, s->rate))
233 goto nla_put_failure;
234
235 if (s->truncate)
236 if (nla_put_u32(skb, TCA_SAMPLE_TRUNC_SIZE, s->trunc_size))
237 goto nla_put_failure;
238
239 if (nla_put_u32(skb, TCA_SAMPLE_PSAMPLE_GROUP, s->psample_group_num))
240 goto nla_put_failure;
241 spin_unlock_bh(&s->tcf_lock);
242
243 return skb->len;
244
245 nla_put_failure:
246 spin_unlock_bh(&s->tcf_lock);
247 nlmsg_trim(skb, b);
248 return -1;
249 }
250
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)251 static int tcf_sample_walker(struct net *net, struct sk_buff *skb,
252 struct netlink_callback *cb, int type,
253 const struct tc_action_ops *ops,
254 struct netlink_ext_ack *extack)
255 {
256 struct tc_action_net *tn = net_generic(net, sample_net_id);
257
258 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
259 }
260
tcf_sample_search(struct net * net,struct tc_action ** a,u32 index)261 static int tcf_sample_search(struct net *net, struct tc_action **a, u32 index)
262 {
263 struct tc_action_net *tn = net_generic(net, sample_net_id);
264
265 return tcf_idr_search(tn, a, index);
266 }
267
tcf_psample_group_put(void * priv)268 static void tcf_psample_group_put(void *priv)
269 {
270 struct psample_group *group = priv;
271
272 psample_group_put(group);
273 }
274
275 static struct psample_group *
tcf_sample_get_group(const struct tc_action * a,tc_action_priv_destructor * destructor)276 tcf_sample_get_group(const struct tc_action *a,
277 tc_action_priv_destructor *destructor)
278 {
279 struct tcf_sample *s = to_sample(a);
280 struct psample_group *group;
281
282 group = rcu_dereference_protected(s->psample_group,
283 lockdep_is_held(&s->tcf_lock));
284 if (group) {
285 psample_group_take(group);
286 *destructor = tcf_psample_group_put;
287 }
288
289 return group;
290 }
291
292 static struct tc_action_ops act_sample_ops = {
293 .kind = "sample",
294 .id = TCA_ID_SAMPLE,
295 .owner = THIS_MODULE,
296 .act = tcf_sample_act,
297 .stats_update = tcf_sample_stats_update,
298 .dump = tcf_sample_dump,
299 .init = tcf_sample_init,
300 .cleanup = tcf_sample_cleanup,
301 .walk = tcf_sample_walker,
302 .lookup = tcf_sample_search,
303 .get_psample_group = tcf_sample_get_group,
304 .size = sizeof(struct tcf_sample),
305 };
306
sample_init_net(struct net * net)307 static __net_init int sample_init_net(struct net *net)
308 {
309 struct tc_action_net *tn = net_generic(net, sample_net_id);
310
311 return tc_action_net_init(net, tn, &act_sample_ops);
312 }
313
sample_exit_net(struct list_head * net_list)314 static void __net_exit sample_exit_net(struct list_head *net_list)
315 {
316 tc_action_net_exit(net_list, sample_net_id);
317 }
318
319 static struct pernet_operations sample_net_ops = {
320 .init = sample_init_net,
321 .exit_batch = sample_exit_net,
322 .id = &sample_net_id,
323 .size = sizeof(struct tc_action_net),
324 };
325
sample_init_module(void)326 static int __init sample_init_module(void)
327 {
328 return tcf_register_action(&act_sample_ops, &sample_net_ops);
329 }
330
sample_cleanup_module(void)331 static void __exit sample_cleanup_module(void)
332 {
333 tcf_unregister_action(&act_sample_ops, &sample_net_ops);
334 }
335
336 module_init(sample_init_module);
337 module_exit(sample_cleanup_module);
338
339 MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>");
340 MODULE_DESCRIPTION("Packet sampling action");
341 MODULE_LICENSE("GPL v2");
342