cls_cgroup.c revision 676f7c8f84d15e94065841529016da5ab92e901b 
1/*
2 * net/sched/cls_cgroup.c	Control Group Classifier
3 *
4 *		This program is free software; you can redistribute it and/or
5 *		modify it under the terms of the GNU General Public License
6 *		as published by the Free Software Foundation; either version
7 *		2 of the License, or (at your option) any later version.
8 *
9 * Authors:	Thomas Graf <tgraf@suug.ch>
10 */
11
12#include <linux/module.h>
13#include <linux/slab.h>
14#include <linux/types.h>
15#include <linux/string.h>
16#include <linux/errno.h>
17#include <linux/skbuff.h>
18#include <linux/cgroup.h>
19#include <linux/rcupdate.h>
20#include <net/rtnetlink.h>
21#include <net/pkt_cls.h>
22#include <net/sock.h>
23#include <net/cls_cgroup.h>
24
25static inline struct cgroup_cls_state *cgrp_cls_state(struct cgroup *cgrp)
26{
27	return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
28			    struct cgroup_cls_state, css);
29}
30
31static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
32{
33	return container_of(task_subsys_state(p, net_cls_subsys_id),
34			    struct cgroup_cls_state, css);
35}
36
37static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp)
38{
39	struct cgroup_cls_state *cs;
40
41	cs = kzalloc(sizeof(*cs), GFP_KERNEL);
42	if (!cs)
43		return ERR_PTR(-ENOMEM);
44
45	if (cgrp->parent)
46		cs->classid = cgrp_cls_state(cgrp->parent)->classid;
47
48	return &cs->css;
49}
50
51static void cgrp_destroy(struct cgroup *cgrp)
52{
53	kfree(cgrp_cls_state(cgrp));
54}
55
56static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
57{
58	return cgrp_cls_state(cgrp)->classid;
59}
60
61static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
62{
63	cgrp_cls_state(cgrp)->classid = (u32) value;
64	return 0;
65}
66
67static struct cftype ss_files[] = {
68	{
69		.name = "classid",
70		.read_u64 = read_classid,
71		.write_u64 = write_classid,
72	},
73};
74
75static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
76{
77	return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files));
78}
79
80struct cgroup_subsys net_cls_subsys = {
81	.name		= "net_cls",
82	.create		= cgrp_create,
83	.destroy	= cgrp_destroy,
84	.populate	= cgrp_populate,
85#ifdef CONFIG_NET_CLS_CGROUP
86	.subsys_id	= net_cls_subsys_id,
87#endif
88	.module		= THIS_MODULE,
89};
90
91struct cls_cgroup_head {
92	u32			handle;
93	struct tcf_exts		exts;
94	struct tcf_ematch_tree	ematches;
95};
96
97static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
98			       struct tcf_result *res)
99{
100	struct cls_cgroup_head *head = tp->root;
101	u32 classid;
102
103	rcu_read_lock();
104	classid = task_cls_state(current)->classid;
105	rcu_read_unlock();
106
107	/*
108	 * Due to the nature of the classifier it is required to ignore all
109	 * packets originating from softirq context as accessing `current'
110	 * would lead to false results.
111	 *
112	 * This test assumes that all callers of dev_queue_xmit() explicitely
113	 * disable bh. Knowing this, it is possible to detect softirq based
114	 * calls by looking at the number of nested bh disable calls because
115	 * softirqs always disables bh.
116	 */
117	if (in_serving_softirq()) {
118		/* If there is an sk_classid we'll use that. */
119		if (!skb->sk)
120			return -1;
121		classid = skb->sk->sk_classid;
122	}
123
124	if (!classid)
125		return -1;
126
127	if (!tcf_em_tree_match(skb, &head->ematches, NULL))
128		return -1;
129
130	res->classid = classid;
131	res->class = 0;
132	return tcf_exts_exec(skb, &head->exts, res);
133}
134
135static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
136{
137	return 0UL;
138}
139
140static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
141{
142}
143
144static int cls_cgroup_init(struct tcf_proto *tp)
145{
146	return 0;
147}
148
149static const struct tcf_ext_map cgroup_ext_map = {
150	.action = TCA_CGROUP_ACT,
151	.police = TCA_CGROUP_POLICE,
152};
153
154static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
155	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
156};
157
158static int cls_cgroup_change(struct tcf_proto *tp, unsigned long base,
159			     u32 handle, struct nlattr **tca,
160			     unsigned long *arg)
161{
162	struct nlattr *tb[TCA_CGROUP_MAX + 1];
163	struct cls_cgroup_head *head = tp->root;
164	struct tcf_ematch_tree t;
165	struct tcf_exts e;
166	int err;
167
168	if (!tca[TCA_OPTIONS])
169		return -EINVAL;
170
171	if (head == NULL) {
172		if (!handle)
173			return -EINVAL;
174
175		head = kzalloc(sizeof(*head), GFP_KERNEL);
176		if (head == NULL)
177			return -ENOBUFS;
178
179		head->handle = handle;
180
181		tcf_tree_lock(tp);
182		tp->root = head;
183		tcf_tree_unlock(tp);
184	}
185
186	if (handle != head->handle)
187		return -ENOENT;
188
189	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
190			       cgroup_policy);
191	if (err < 0)
192		return err;
193
194	err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
195	if (err < 0)
196		return err;
197
198	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
199	if (err < 0)
200		return err;
201
202	tcf_exts_change(tp, &head->exts, &e);
203	tcf_em_tree_change(tp, &head->ematches, &t);
204
205	return 0;
206}
207
208static void cls_cgroup_destroy(struct tcf_proto *tp)
209{
210	struct cls_cgroup_head *head = tp->root;
211
212	if (head) {
213		tcf_exts_destroy(tp, &head->exts);
214		tcf_em_tree_destroy(tp, &head->ematches);
215		kfree(head);
216	}
217}
218
219static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
220{
221	return -EOPNOTSUPP;
222}
223
224static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
225{
226	struct cls_cgroup_head *head = tp->root;
227
228	if (arg->count < arg->skip)
229		goto skip;
230
231	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
232		arg->stop = 1;
233		return;
234	}
235skip:
236	arg->count++;
237}
238
239static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
240			   struct sk_buff *skb, struct tcmsg *t)
241{
242	struct cls_cgroup_head *head = tp->root;
243	unsigned char *b = skb_tail_pointer(skb);
244	struct nlattr *nest;
245
246	t->tcm_handle = head->handle;
247
248	nest = nla_nest_start(skb, TCA_OPTIONS);
249	if (nest == NULL)
250		goto nla_put_failure;
251
252	if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
253	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
254		goto nla_put_failure;
255
256	nla_nest_end(skb, nest);
257
258	if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
259		goto nla_put_failure;
260
261	return skb->len;
262
263nla_put_failure:
264	nlmsg_trim(skb, b);
265	return -1;
266}
267
268static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
269	.kind		=	"cgroup",
270	.init		=	cls_cgroup_init,
271	.change		=	cls_cgroup_change,
272	.classify	=	cls_cgroup_classify,
273	.destroy	=	cls_cgroup_destroy,
274	.get		=	cls_cgroup_get,
275	.put		=	cls_cgroup_put,
276	.delete		=	cls_cgroup_delete,
277	.walk		=	cls_cgroup_walk,
278	.dump		=	cls_cgroup_dump,
279	.owner		=	THIS_MODULE,
280};
281
282static int __init init_cgroup_cls(void)
283{
284	int ret;
285
286	ret = cgroup_load_subsys(&net_cls_subsys);
287	if (ret)
288		goto out;
289
290#ifndef CONFIG_NET_CLS_CGROUP
291	/* We can't use rcu_assign_pointer because this is an int. */
292	smp_wmb();
293	net_cls_subsys_id = net_cls_subsys.subsys_id;
294#endif
295
296	ret = register_tcf_proto_ops(&cls_cgroup_ops);
297	if (ret)
298		cgroup_unload_subsys(&net_cls_subsys);
299
300out:
301	return ret;
302}
303
304static void __exit exit_cgroup_cls(void)
305{
306	unregister_tcf_proto_ops(&cls_cgroup_ops);
307
308#ifndef CONFIG_NET_CLS_CGROUP
309	net_cls_subsys_id = -1;
310	synchronize_rcu();
311#endif
312
313	cgroup_unload_subsys(&net_cls_subsys);
314}
315
316module_init(init_cgroup_cls);
317module_exit(exit_cgroup_cls);
318MODULE_LICENSE("GPL");
319