Commit | Line | Data |
---|---|---|
d2912cb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
5c5670fa YG |
2 | /* |
3 | * net/sched/act_sample.c - Packet sampling tc action | |
4 | * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com> | |
5c5670fa YG |
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> | |
e8c87c64 | 22 | #include <net/pkt_cls.h> |
5c5670fa YG |
23 | |
24 | #include <linux/if_arp.h> | |
25 | ||
5c5670fa YG |
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 | ||
36 | static int tcf_sample_init(struct net *net, struct nlattr *nla, | |
37 | struct nlattr *est, struct tc_action **a, int ovr, | |
85d0966f | 38 | int bind, bool rtnl_held, struct tcf_proto *tp, |
abbb0d33 | 39 | u32 flags, struct netlink_ext_ack *extack) |
5c5670fa YG |
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; | |
7be8ef2c | 44 | u32 psample_group_num, rate, index; |
e8c87c64 | 45 | struct tcf_chain *goto_ch = NULL; |
5c5670fa YG |
46 | struct tc_sample *parm; |
47 | struct tcf_sample *s; | |
48 | bool exists = false; | |
0190c1d4 | 49 | int ret, err; |
5c5670fa YG |
50 | |
51 | if (!nla) | |
52 | return -EINVAL; | |
8cb08174 JB |
53 | ret = nla_parse_nested_deprecated(tb, TCA_SAMPLE_MAX, nla, |
54 | sample_policy, NULL); | |
5c5670fa YG |
55 | if (ret < 0) |
56 | return ret; | |
57 | if (!tb[TCA_SAMPLE_PARMS] || !tb[TCA_SAMPLE_RATE] || | |
58 | !tb[TCA_SAMPLE_PSAMPLE_GROUP]) | |
59 | return -EINVAL; | |
60 | ||
61 | parm = nla_data(tb[TCA_SAMPLE_PARMS]); | |
7be8ef2c DL |
62 | index = parm->index; |
63 | err = tcf_idr_check_alloc(tn, &index, a, bind); | |
0190c1d4 VB |
64 | if (err < 0) |
65 | return err; | |
66 | exists = err; | |
5c5670fa YG |
67 | if (exists && bind) |
68 | return 0; | |
69 | ||
70 | if (!exists) { | |
7be8ef2c | 71 | ret = tcf_idr_create(tn, index, est, a, |
e3822678 | 72 | &act_sample_ops, bind, true, 0); |
0190c1d4 | 73 | if (ret) { |
7be8ef2c | 74 | tcf_idr_cleanup(tn, index); |
5c5670fa | 75 | return ret; |
0190c1d4 | 76 | } |
5c5670fa | 77 | ret = ACT_P_CREATED; |
4e8ddd7f | 78 | } else if (!ovr) { |
65a206c0 | 79 | tcf_idr_release(*a, bind); |
4e8ddd7f | 80 | return -EEXIST; |
5c5670fa | 81 | } |
e8c87c64 DC |
82 | err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); |
83 | if (err < 0) | |
84 | goto release_idr; | |
5c5670fa | 85 | |
fae27081 DC |
86 | rate = nla_get_u32(tb[TCA_SAMPLE_RATE]); |
87 | if (!rate) { | |
88 | NL_SET_ERR_MSG(extack, "invalid sample rate"); | |
89 | err = -EINVAL; | |
90 | goto put_chain; | |
91 | } | |
653cd284 VB |
92 | psample_group_num = nla_get_u32(tb[TCA_SAMPLE_PSAMPLE_GROUP]); |
93 | psample_group = psample_group_get(net, psample_group_num); | |
cadb9c9f | 94 | if (!psample_group) { |
e8c87c64 DC |
95 | err = -ENOMEM; |
96 | goto put_chain; | |
cadb9c9f | 97 | } |
653cd284 VB |
98 | |
99 | s = to_sample(*a); | |
100 | ||
101 | spin_lock_bh(&s->tcf_lock); | |
e8c87c64 | 102 | goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch); |
fae27081 | 103 | s->rate = rate; |
653cd284 | 104 | s->psample_group_num = psample_group_num; |
445d3749 PM |
105 | psample_group = rcu_replace_pointer(s->psample_group, psample_group, |
106 | lockdep_is_held(&s->tcf_lock)); | |
5c5670fa YG |
107 | |
108 | if (tb[TCA_SAMPLE_TRUNC_SIZE]) { | |
109 | s->truncate = true; | |
110 | s->trunc_size = nla_get_u32(tb[TCA_SAMPLE_TRUNC_SIZE]); | |
111 | } | |
653cd284 | 112 | spin_unlock_bh(&s->tcf_lock); |
dbf47a2a VB |
113 | |
114 | if (psample_group) | |
115 | psample_group_put(psample_group); | |
e8c87c64 DC |
116 | if (goto_ch) |
117 | tcf_chain_put_by_act(goto_ch); | |
5c5670fa YG |
118 | |
119 | if (ret == ACT_P_CREATED) | |
65a206c0 | 120 | tcf_idr_insert(tn, *a); |
5c5670fa | 121 | return ret; |
e8c87c64 DC |
122 | put_chain: |
123 | if (goto_ch) | |
124 | tcf_chain_put_by_act(goto_ch); | |
125 | release_idr: | |
126 | tcf_idr_release(*a, bind); | |
127 | return err; | |
5c5670fa YG |
128 | } |
129 | ||
9a63b255 | 130 | static void tcf_sample_cleanup(struct tc_action *a) |
5c5670fa | 131 | { |
90a6ec85 | 132 | struct tcf_sample *s = to_sample(a); |
5c5670fa YG |
133 | struct psample_group *psample_group; |
134 | ||
d7728495 VB |
135 | /* last reference to action, no need to lock */ |
136 | psample_group = rcu_dereference_protected(s->psample_group, 1); | |
5c5670fa | 137 | RCU_INIT_POINTER(s->psample_group, NULL); |
1f110e7c DC |
138 | if (psample_group) |
139 | psample_group_put(psample_group); | |
5c5670fa YG |
140 | } |
141 | ||
5c5670fa YG |
142 | static bool tcf_sample_dev_ok_push(struct net_device *dev) |
143 | { | |
144 | switch (dev->type) { | |
145 | case ARPHRD_TUNNEL: | |
146 | case ARPHRD_TUNNEL6: | |
147 | case ARPHRD_SIT: | |
148 | case ARPHRD_IPGRE: | |
92974a1d | 149 | case ARPHRD_IP6GRE: |
5c5670fa YG |
150 | case ARPHRD_VOID: |
151 | case ARPHRD_NONE: | |
152 | return false; | |
153 | default: | |
154 | return true; | |
155 | } | |
156 | } | |
157 | ||
158 | static int tcf_sample_act(struct sk_buff *skb, const struct tc_action *a, | |
159 | struct tcf_result *res) | |
160 | { | |
161 | struct tcf_sample *s = to_sample(a); | |
162 | struct psample_group *psample_group; | |
163 | int retval; | |
164 | int size; | |
165 | int iif; | |
166 | int oif; | |
167 | ||
168 | tcf_lastuse_update(&s->tcf_tm); | |
169 | bstats_cpu_update(this_cpu_ptr(s->common.cpu_bstats), skb); | |
170 | retval = READ_ONCE(s->tcf_action); | |
171 | ||
7fd4b288 | 172 | psample_group = rcu_dereference_bh(s->psample_group); |
5c5670fa YG |
173 | |
174 | /* randomly sample packets according to rate */ | |
175 | if (psample_group && (prandom_u32() % s->rate == 0)) { | |
176 | if (!skb_at_tc_ingress(skb)) { | |
177 | iif = skb->skb_iif; | |
178 | oif = skb->dev->ifindex; | |
179 | } else { | |
180 | iif = skb->dev->ifindex; | |
181 | oif = 0; | |
182 | } | |
183 | ||
184 | /* on ingress, the mac header gets popped, so push it back */ | |
185 | if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev)) | |
186 | skb_push(skb, skb->mac_len); | |
187 | ||
188 | size = s->truncate ? s->trunc_size : skb->len; | |
189 | psample_sample_packet(psample_group, skb, size, iif, oif, | |
190 | s->rate); | |
191 | ||
192 | if (skb_at_tc_ingress(skb) && tcf_sample_dev_ok_push(skb->dev)) | |
193 | skb_pull(skb, skb->mac_len); | |
194 | } | |
195 | ||
5c5670fa YG |
196 | return retval; |
197 | } | |
198 | ||
199 | static int tcf_sample_dump(struct sk_buff *skb, struct tc_action *a, | |
200 | int bind, int ref) | |
201 | { | |
202 | unsigned char *b = skb_tail_pointer(skb); | |
203 | struct tcf_sample *s = to_sample(a); | |
204 | struct tc_sample opt = { | |
205 | .index = s->tcf_index, | |
036bb443 VB |
206 | .refcnt = refcount_read(&s->tcf_refcnt) - ref, |
207 | .bindcnt = atomic_read(&s->tcf_bindcnt) - bind, | |
5c5670fa YG |
208 | }; |
209 | struct tcf_t t; | |
210 | ||
653cd284 | 211 | spin_lock_bh(&s->tcf_lock); |
d7728495 | 212 | opt.action = s->tcf_action; |
5c5670fa YG |
213 | if (nla_put(skb, TCA_SAMPLE_PARMS, sizeof(opt), &opt)) |
214 | goto nla_put_failure; | |
215 | ||
216 | tcf_tm_dump(&t, &s->tcf_tm); | |
217 | if (nla_put_64bit(skb, TCA_SAMPLE_TM, sizeof(t), &t, TCA_SAMPLE_PAD)) | |
218 | goto nla_put_failure; | |
219 | ||
220 | if (nla_put_u32(skb, TCA_SAMPLE_RATE, s->rate)) | |
221 | goto nla_put_failure; | |
222 | ||
223 | if (s->truncate) | |
224 | if (nla_put_u32(skb, TCA_SAMPLE_TRUNC_SIZE, s->trunc_size)) | |
225 | goto nla_put_failure; | |
226 | ||
227 | if (nla_put_u32(skb, TCA_SAMPLE_PSAMPLE_GROUP, s->psample_group_num)) | |
228 | goto nla_put_failure; | |
653cd284 | 229 | spin_unlock_bh(&s->tcf_lock); |
d7728495 | 230 | |
5c5670fa YG |
231 | return skb->len; |
232 | ||
233 | nla_put_failure: | |
653cd284 | 234 | spin_unlock_bh(&s->tcf_lock); |
5c5670fa YG |
235 | nlmsg_trim(skb, b); |
236 | return -1; | |
237 | } | |
238 | ||
239 | static int tcf_sample_walker(struct net *net, struct sk_buff *skb, | |
240 | struct netlink_callback *cb, int type, | |
41780105 AA |
241 | const struct tc_action_ops *ops, |
242 | struct netlink_ext_ack *extack) | |
5c5670fa YG |
243 | { |
244 | struct tc_action_net *tn = net_generic(net, sample_net_id); | |
245 | ||
b3620145 | 246 | return tcf_generic_walker(tn, skb, cb, type, ops, extack); |
5c5670fa YG |
247 | } |
248 | ||
f061b48c | 249 | static int tcf_sample_search(struct net *net, struct tc_action **a, u32 index) |
5c5670fa YG |
250 | { |
251 | struct tc_action_net *tn = net_generic(net, sample_net_id); | |
252 | ||
65a206c0 | 253 | return tcf_idr_search(tn, a, index); |
5c5670fa YG |
254 | } |
255 | ||
4a5da47d VB |
256 | static void tcf_psample_group_put(void *priv) |
257 | { | |
258 | struct psample_group *group = priv; | |
259 | ||
260 | psample_group_put(group); | |
261 | } | |
262 | ||
263 | static struct psample_group * | |
264 | tcf_sample_get_group(const struct tc_action *a, | |
265 | tc_action_priv_destructor *destructor) | |
266 | { | |
267 | struct tcf_sample *s = to_sample(a); | |
268 | struct psample_group *group; | |
269 | ||
270 | spin_lock_bh(&s->tcf_lock); | |
271 | group = rcu_dereference_protected(s->psample_group, | |
272 | lockdep_is_held(&s->tcf_lock)); | |
273 | if (group) { | |
274 | psample_group_take(group); | |
275 | *destructor = tcf_psample_group_put; | |
276 | } | |
277 | spin_unlock_bh(&s->tcf_lock); | |
278 | ||
279 | return group; | |
280 | } | |
281 | ||
5c5670fa YG |
282 | static struct tc_action_ops act_sample_ops = { |
283 | .kind = "sample", | |
eddd2cf1 | 284 | .id = TCA_ID_SAMPLE, |
5c5670fa YG |
285 | .owner = THIS_MODULE, |
286 | .act = tcf_sample_act, | |
287 | .dump = tcf_sample_dump, | |
288 | .init = tcf_sample_init, | |
289 | .cleanup = tcf_sample_cleanup, | |
290 | .walk = tcf_sample_walker, | |
291 | .lookup = tcf_sample_search, | |
4a5da47d | 292 | .get_psample_group = tcf_sample_get_group, |
5c5670fa YG |
293 | .size = sizeof(struct tcf_sample), |
294 | }; | |
295 | ||
296 | static __net_init int sample_init_net(struct net *net) | |
297 | { | |
298 | struct tc_action_net *tn = net_generic(net, sample_net_id); | |
299 | ||
981471bd | 300 | return tc_action_net_init(net, tn, &act_sample_ops); |
5c5670fa YG |
301 | } |
302 | ||
039af9c6 | 303 | static void __net_exit sample_exit_net(struct list_head *net_list) |
5c5670fa | 304 | { |
039af9c6 | 305 | tc_action_net_exit(net_list, sample_net_id); |
5c5670fa YG |
306 | } |
307 | ||
308 | static struct pernet_operations sample_net_ops = { | |
309 | .init = sample_init_net, | |
039af9c6 | 310 | .exit_batch = sample_exit_net, |
5c5670fa YG |
311 | .id = &sample_net_id, |
312 | .size = sizeof(struct tc_action_net), | |
313 | }; | |
314 | ||
315 | static int __init sample_init_module(void) | |
316 | { | |
317 | return tcf_register_action(&act_sample_ops, &sample_net_ops); | |
318 | } | |
319 | ||
320 | static void __exit sample_cleanup_module(void) | |
321 | { | |
322 | tcf_unregister_action(&act_sample_ops, &sample_net_ops); | |
323 | } | |
324 | ||
325 | module_init(sample_init_module); | |
326 | module_exit(sample_cleanup_module); | |
327 | ||
f1fd20c3 | 328 | MODULE_AUTHOR("Yotam Gigi <yotam.gi@gmail.com>"); |
5c5670fa YG |
329 | MODULE_DESCRIPTION("Packet sampling action"); |
330 | MODULE_LICENSE("GPL v2"); |