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