cls_cgroup.c revision 5a0e3ad6af8660be21ca98a971cd00f331318c05 
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 <net/rtnetlink.h>
20#include <net/pkt_cls.h>
21
22struct cgroup_cls_state
23{
24	struct cgroup_subsys_state css;
25	u32 classid;
26};
27
28static struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss,
29					       struct cgroup *cgrp);
30static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp);
31static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp);
32
33struct cgroup_subsys net_cls_subsys = {
34	.name		= "net_cls",
35	.create		= cgrp_create,
36	.destroy	= cgrp_destroy,
37	.populate	= cgrp_populate,
38#ifdef CONFIG_NET_CLS_CGROUP
39	.subsys_id	= net_cls_subsys_id,
40#else
41#define net_cls_subsys_id net_cls_subsys.subsys_id
42#endif
43	.module		= THIS_MODULE,
44};
45
46
47static inline struct cgroup_cls_state *cgrp_cls_state(struct cgroup *cgrp)
48{
49	return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
50			    struct cgroup_cls_state, css);
51}
52
53static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
54{
55	return container_of(task_subsys_state(p, net_cls_subsys_id),
56			    struct cgroup_cls_state, css);
57}
58
59static struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss,
60						 struct cgroup *cgrp)
61{
62	struct cgroup_cls_state *cs;
63
64	if (!(cs = kzalloc(sizeof(*cs), GFP_KERNEL)))
65		return ERR_PTR(-ENOMEM);
66
67	if (cgrp->parent)
68		cs->classid = cgrp_cls_state(cgrp->parent)->classid;
69
70	return &cs->css;
71}
72
73static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
74{
75	kfree(cgrp_cls_state(cgrp));
76}
77
78static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
79{
80	return cgrp_cls_state(cgrp)->classid;
81}
82
83static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
84{
85	cgrp_cls_state(cgrp)->classid = (u32) value;
86	return 0;
87}
88
89static struct cftype ss_files[] = {
90	{
91		.name = "classid",
92		.read_u64 = read_classid,
93		.write_u64 = write_classid,
94	},
95};
96
97static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
98{
99	return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files));
100}
101
102struct cls_cgroup_head
103{
104	u32			handle;
105	struct tcf_exts		exts;
106	struct tcf_ematch_tree	ematches;
107};
108
109static int cls_cgroup_classify(struct sk_buff *skb, struct tcf_proto *tp,
110			       struct tcf_result *res)
111{
112	struct cls_cgroup_head *head = tp->root;
113	u32 classid;
114
115	/*
116	 * Due to the nature of the classifier it is required to ignore all
117	 * packets originating from softirq context as accessing `current'
118	 * would lead to false results.
119	 *
120	 * This test assumes that all callers of dev_queue_xmit() explicitely
121	 * disable bh. Knowing this, it is possible to detect softirq based
122	 * calls by looking at the number of nested bh disable calls because
123	 * softirqs always disables bh.
124	 */
125	if (softirq_count() != SOFTIRQ_OFFSET)
126		return -1;
127
128	rcu_read_lock();
129	classid = task_cls_state(current)->classid;
130	rcu_read_unlock();
131
132	if (!classid)
133		return -1;
134
135	if (!tcf_em_tree_match(skb, &head->ematches, NULL))
136		return -1;
137
138	res->classid = classid;
139	res->class = 0;
140	return tcf_exts_exec(skb, &head->exts, res);
141}
142
143static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
144{
145	return 0UL;
146}
147
148static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
149{
150}
151
152static int cls_cgroup_init(struct tcf_proto *tp)
153{
154	return 0;
155}
156
157static const struct tcf_ext_map cgroup_ext_map = {
158	.action = TCA_CGROUP_ACT,
159	.police = TCA_CGROUP_POLICE,
160};
161
162static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
163	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
164};
165
166static int cls_cgroup_change(struct tcf_proto *tp, unsigned long base,
167			     u32 handle, struct nlattr **tca,
168			     unsigned long *arg)
169{
170	struct nlattr *tb[TCA_CGROUP_MAX+1];
171	struct cls_cgroup_head *head = tp->root;
172	struct tcf_ematch_tree t;
173	struct tcf_exts e;
174	int err;
175
176	if (!tca[TCA_OPTIONS])
177		return -EINVAL;
178
179	if (head == NULL) {
180		if (!handle)
181			return -EINVAL;
182
183		head = kzalloc(sizeof(*head), GFP_KERNEL);
184		if (head == NULL)
185			return -ENOBUFS;
186
187		head->handle = handle;
188
189		tcf_tree_lock(tp);
190		tp->root = head;
191		tcf_tree_unlock(tp);
192	}
193
194	if (handle != head->handle)
195		return -ENOENT;
196
197	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
198			       cgroup_policy);
199	if (err < 0)
200		return err;
201
202	err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
203	if (err < 0)
204		return err;
205
206	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
207	if (err < 0)
208		return err;
209
210	tcf_exts_change(tp, &head->exts, &e);
211	tcf_em_tree_change(tp, &head->ematches, &t);
212
213	return 0;
214}
215
216static void cls_cgroup_destroy(struct tcf_proto *tp)
217{
218	struct cls_cgroup_head *head = tp->root;
219
220	if (head) {
221		tcf_exts_destroy(tp, &head->exts);
222		tcf_em_tree_destroy(tp, &head->ematches);
223		kfree(head);
224	}
225}
226
227static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
228{
229	return -EOPNOTSUPP;
230}
231
232static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
233{
234	struct cls_cgroup_head *head = tp->root;
235
236	if (arg->count < arg->skip)
237		goto skip;
238
239	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
240		arg->stop = 1;
241		return;
242	}
243skip:
244	arg->count++;
245}
246
247static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
248			   struct sk_buff *skb, struct tcmsg *t)
249{
250	struct cls_cgroup_head *head = tp->root;
251	unsigned char *b = skb_tail_pointer(skb);
252	struct nlattr *nest;
253
254	t->tcm_handle = head->handle;
255
256	nest = nla_nest_start(skb, TCA_OPTIONS);
257	if (nest == NULL)
258		goto nla_put_failure;
259
260	if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
261	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
262		goto nla_put_failure;
263
264	nla_nest_end(skb, nest);
265
266	if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
267		goto nla_put_failure;
268
269	return skb->len;
270
271nla_put_failure:
272	nlmsg_trim(skb, b);
273	return -1;
274}
275
276static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
277	.kind		=	"cgroup",
278	.init		=	cls_cgroup_init,
279	.change		=	cls_cgroup_change,
280	.classify	=	cls_cgroup_classify,
281	.destroy	=	cls_cgroup_destroy,
282	.get		=	cls_cgroup_get,
283	.put		=	cls_cgroup_put,
284	.delete		=	cls_cgroup_delete,
285	.walk		=	cls_cgroup_walk,
286	.dump		=	cls_cgroup_dump,
287	.owner		=	THIS_MODULE,
288};
289
290static int __init init_cgroup_cls(void)
291{
292	int ret = register_tcf_proto_ops(&cls_cgroup_ops);
293	if (ret)
294		return ret;
295	ret = cgroup_load_subsys(&net_cls_subsys);
296	if (ret)
297		unregister_tcf_proto_ops(&cls_cgroup_ops);
298	return ret;
299}
300
301static void __exit exit_cgroup_cls(void)
302{
303	unregister_tcf_proto_ops(&cls_cgroup_ops);
304	cgroup_unload_subsys(&net_cls_subsys);
305}
306
307module_init(init_cgroup_cls);
308module_exit(exit_cgroup_cls);
309MODULE_LICENSE("GPL");
310