cls_cgroup.c revision 8e8ba85417366afd2361e315c6ba5949d3eff56f 
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/types.h>
14#include <linux/string.h>
15#include <linux/errno.h>
16#include <linux/skbuff.h>
17#include <linux/cgroup.h>
18#include <net/rtnetlink.h>
19#include <net/pkt_cls.h>
20
21struct cgroup_cls_state
22{
23	struct cgroup_subsys_state css;
24	u32 classid;
25};
26
27static inline struct cgroup_cls_state *cgrp_cls_state(struct cgroup *cgrp)
28{
29	return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
30			    struct cgroup_cls_state, css);
31}
32
33static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
34{
35	return container_of(task_subsys_state(p, net_cls_subsys_id),
36			    struct cgroup_cls_state, css);
37}
38
39static struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss,
40						 struct cgroup *cgrp)
41{
42	struct cgroup_cls_state *cs;
43
44	if (!(cs = kzalloc(sizeof(*cs), GFP_KERNEL)))
45		return ERR_PTR(-ENOMEM);
46
47	if (cgrp->parent)
48		cs->classid = cgrp_cls_state(cgrp->parent)->classid;
49
50	return &cs->css;
51}
52
53static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
54{
55	kfree(cgrp_cls_state(cgrp));
56}
57
58static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
59{
60	return cgrp_cls_state(cgrp)->classid;
61}
62
63static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
64{
65	if (!cgroup_lock_live_group(cgrp))
66		return -ENODEV;
67
68	cgrp_cls_state(cgrp)->classid = (u32) value;
69
70	cgroup_unlock();
71
72	return 0;
73}
74
75static struct cftype ss_files[] = {
76	{
77		.name = "classid",
78		.read_u64 = read_classid,
79		.write_u64 = write_classid,
80	},
81};
82
83static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
84{
85	return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files));
86}
87
88struct cgroup_subsys net_cls_subsys = {
89	.name		= "net_cls",
90	.create		= cgrp_create,
91	.destroy	= cgrp_destroy,
92	.populate	= cgrp_populate,
93	.subsys_id	= net_cls_subsys_id,
94};
95
96struct cls_cgroup_head
97{
98	u32			handle;
99	struct tcf_exts		exts;
100	struct tcf_ematch_tree	ematches;
101};
102
103static int cls_cgroup_classify(struct sk_buff *skb, struct tcf_proto *tp,
104			       struct tcf_result *res)
105{
106	struct cls_cgroup_head *head = tp->root;
107	struct cgroup_cls_state *cs;
108	int ret = 0;
109
110	/*
111	 * Due to the nature of the classifier it is required to ignore all
112	 * packets originating from softirq context as accessing `current'
113	 * would lead to false results.
114	 *
115	 * This test assumes that all callers of dev_queue_xmit() explicitely
116	 * disable bh. Knowing this, it is possible to detect softirq based
117	 * calls by looking at the number of nested bh disable calls because
118	 * softirqs always disables bh.
119	 */
120	if (softirq_count() != SOFTIRQ_OFFSET)
121		return -1;
122
123	rcu_read_lock();
124	cs = task_cls_state(current);
125	if (cs->classid && tcf_em_tree_match(skb, &head->ematches, NULL)) {
126		res->classid = cs->classid;
127		res->class = 0;
128		ret = tcf_exts_exec(skb, &head->exts, res);
129	} else
130		ret = -1;
131
132	rcu_read_unlock();
133
134	return ret;
135}
136
137static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
138{
139	return 0UL;
140}
141
142static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
143{
144}
145
146static int cls_cgroup_init(struct tcf_proto *tp)
147{
148	return 0;
149}
150
151static const struct tcf_ext_map cgroup_ext_map = {
152	.action = TCA_CGROUP_ACT,
153	.police = TCA_CGROUP_POLICE,
154};
155
156static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
157	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
158};
159
160static int cls_cgroup_change(struct tcf_proto *tp, unsigned long base,
161			     u32 handle, struct nlattr **tca,
162			     unsigned long *arg)
163{
164	struct nlattr *tb[TCA_CGROUP_MAX+1];
165	struct cls_cgroup_head *head = tp->root;
166	struct tcf_ematch_tree t;
167	struct tcf_exts e;
168	int err;
169
170	if (head == NULL) {
171		if (!handle)
172			return -EINVAL;
173
174		head = kzalloc(sizeof(*head), GFP_KERNEL);
175		if (head == NULL)
176			return -ENOBUFS;
177
178		head->handle = handle;
179
180		tcf_tree_lock(tp);
181		tp->root = head;
182		tcf_tree_unlock(tp);
183	}
184
185	if (handle != head->handle)
186		return -ENOENT;
187
188	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
189			       cgroup_policy);
190	if (err < 0)
191		return err;
192
193	err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
194	if (err < 0)
195		return err;
196
197	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
198	if (err < 0)
199		return err;
200
201	tcf_exts_change(tp, &head->exts, &e);
202	tcf_em_tree_change(tp, &head->ematches, &t);
203
204	return 0;
205}
206
207static void cls_cgroup_destroy(struct tcf_proto *tp)
208{
209	struct cls_cgroup_head *head = tp->root;
210
211	if (head) {
212		tcf_exts_destroy(tp, &head->exts);
213		tcf_em_tree_destroy(tp, &head->ematches);
214		kfree(head);
215	}
216}
217
218static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
219{
220	return -EOPNOTSUPP;
221}
222
223static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
224{
225	struct cls_cgroup_head *head = tp->root;
226
227	if (arg->count < arg->skip)
228		goto skip;
229
230	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
231		arg->stop = 1;
232		return;
233	}
234skip:
235	arg->count++;
236}
237
238static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
239			   struct sk_buff *skb, struct tcmsg *t)
240{
241	struct cls_cgroup_head *head = tp->root;
242	unsigned char *b = skb_tail_pointer(skb);
243	struct nlattr *nest;
244
245	t->tcm_handle = head->handle;
246
247	nest = nla_nest_start(skb, TCA_OPTIONS);
248	if (nest == NULL)
249		goto nla_put_failure;
250
251	if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
252	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
253		goto nla_put_failure;
254
255	nla_nest_end(skb, nest);
256
257	if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
258		goto nla_put_failure;
259
260	return skb->len;
261
262nla_put_failure:
263	nlmsg_trim(skb, b);
264	return -1;
265}
266
267static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
268	.kind		=	"cgroup",
269	.init		=	cls_cgroup_init,
270	.change		=	cls_cgroup_change,
271	.classify	=	cls_cgroup_classify,
272	.destroy	=	cls_cgroup_destroy,
273	.get		=	cls_cgroup_get,
274	.put		=	cls_cgroup_put,
275	.delete		=	cls_cgroup_delete,
276	.walk		=	cls_cgroup_walk,
277	.dump		=	cls_cgroup_dump,
278	.owner		=	THIS_MODULE,
279};
280
281static int __init init_cgroup_cls(void)
282{
283	return register_tcf_proto_ops(&cls_cgroup_ops);
284}
285
286static void __exit exit_cgroup_cls(void)
287{
288	unregister_tcf_proto_ops(&cls_cgroup_ops);
289}
290
291module_init(init_cgroup_cls);
292module_exit(exit_cgroup_cls);
293MODULE_LICENSE("GPL");
294