cls_cgroup.c revision c00b2c9e79466d61979cd21af526cc6d5d0ee04f 
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 struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss,
26					       struct cgroup *cgrp);
27static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp);
28static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp);
29
30struct cgroup_subsys net_cls_subsys = {
31	.name		= "net_cls",
32	.create		= cgrp_create,
33	.destroy	= cgrp_destroy,
34	.populate	= cgrp_populate,
35#ifdef CONFIG_NET_CLS_CGROUP
36	.subsys_id	= net_cls_subsys_id,
37#endif
38	.module		= THIS_MODULE,
39};
40
41
42static inline struct cgroup_cls_state *cgrp_cls_state(struct cgroup *cgrp)
43{
44	return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
45			    struct cgroup_cls_state, css);
46}
47
48static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
49{
50	return container_of(task_subsys_state(p, net_cls_subsys_id),
51			    struct cgroup_cls_state, css);
52}
53
54static struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss,
55						 struct cgroup *cgrp)
56{
57	struct cgroup_cls_state *cs;
58
59	if (!(cs = kzalloc(sizeof(*cs), GFP_KERNEL)))
60		return ERR_PTR(-ENOMEM);
61
62	if (cgrp->parent)
63		cs->classid = cgrp_cls_state(cgrp->parent)->classid;
64
65	return &cs->css;
66}
67
68static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
69{
70	kfree(cgrp_cls_state(cgrp));
71}
72
73static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
74{
75	return cgrp_cls_state(cgrp)->classid;
76}
77
78static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
79{
80	cgrp_cls_state(cgrp)->classid = (u32) value;
81	return 0;
82}
83
84static struct cftype ss_files[] = {
85	{
86		.name = "classid",
87		.read_u64 = read_classid,
88		.write_u64 = write_classid,
89	},
90};
91
92static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
93{
94	return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files));
95}
96
97struct cls_cgroup_head
98{
99	u32			handle;
100	struct tcf_exts		exts;
101	struct tcf_ematch_tree	ematches;
102};
103
104static int cls_cgroup_classify(struct sk_buff *skb, struct tcf_proto *tp,
105			       struct tcf_result *res)
106{
107	struct cls_cgroup_head *head = tp->root;
108	u32 classid;
109
110	rcu_read_lock();
111	classid = task_cls_state(current)->classid;
112	rcu_read_unlock();
113
114	/*
115	 * Due to the nature of the classifier it is required to ignore all
116	 * packets originating from softirq context as accessing `current'
117	 * would lead to false results.
118	 *
119	 * This test assumes that all callers of dev_queue_xmit() explicitely
120	 * disable bh. Knowing this, it is possible to detect softirq based
121	 * calls by looking at the number of nested bh disable calls because
122	 * softirqs always disables bh.
123	 */
124	if (in_serving_softirq()) {
125		/* If there is an sk_classid we'll use that. */
126		if (!skb->sk)
127			return -1;
128		classid = skb->sk->sk_classid;
129	}
130
131	if (!classid)
132		return -1;
133
134	if (!tcf_em_tree_match(skb, &head->ematches, NULL))
135		return -1;
136
137	res->classid = classid;
138	res->class = 0;
139	return tcf_exts_exec(skb, &head->exts, res);
140}
141
142static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
143{
144	return 0UL;
145}
146
147static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
148{
149}
150
151static int cls_cgroup_init(struct tcf_proto *tp)
152{
153	return 0;
154}
155
156static const struct tcf_ext_map cgroup_ext_map = {
157	.action = TCA_CGROUP_ACT,
158	.police = TCA_CGROUP_POLICE,
159};
160
161static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
162	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
163};
164
165static int cls_cgroup_change(struct tcf_proto *tp, unsigned long base,
166			     u32 handle, struct nlattr **tca,
167			     unsigned long *arg)
168{
169	struct nlattr *tb[TCA_CGROUP_MAX+1];
170	struct cls_cgroup_head *head = tp->root;
171	struct tcf_ematch_tree t;
172	struct tcf_exts e;
173	int err;
174
175	if (!tca[TCA_OPTIONS])
176		return -EINVAL;
177
178	if (head == NULL) {
179		if (!handle)
180			return -EINVAL;
181
182		head = kzalloc(sizeof(*head), GFP_KERNEL);
183		if (head == NULL)
184			return -ENOBUFS;
185
186		head->handle = handle;
187
188		tcf_tree_lock(tp);
189		tp->root = head;
190		tcf_tree_unlock(tp);
191	}
192
193	if (handle != head->handle)
194		return -ENOENT;
195
196	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
197			       cgroup_policy);
198	if (err < 0)
199		return err;
200
201	err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
202	if (err < 0)
203		return err;
204
205	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
206	if (err < 0)
207		return err;
208
209	tcf_exts_change(tp, &head->exts, &e);
210	tcf_em_tree_change(tp, &head->ematches, &t);
211
212	return 0;
213}
214
215static void cls_cgroup_destroy(struct tcf_proto *tp)
216{
217	struct cls_cgroup_head *head = tp->root;
218
219	if (head) {
220		tcf_exts_destroy(tp, &head->exts);
221		tcf_em_tree_destroy(tp, &head->ematches);
222		kfree(head);
223	}
224}
225
226static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
227{
228	return -EOPNOTSUPP;
229}
230
231static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
232{
233	struct cls_cgroup_head *head = tp->root;
234
235	if (arg->count < arg->skip)
236		goto skip;
237
238	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
239		arg->stop = 1;
240		return;
241	}
242skip:
243	arg->count++;
244}
245
246static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
247			   struct sk_buff *skb, struct tcmsg *t)
248{
249	struct cls_cgroup_head *head = tp->root;
250	unsigned char *b = skb_tail_pointer(skb);
251	struct nlattr *nest;
252
253	t->tcm_handle = head->handle;
254
255	nest = nla_nest_start(skb, TCA_OPTIONS);
256	if (nest == NULL)
257		goto nla_put_failure;
258
259	if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
260	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
261		goto nla_put_failure;
262
263	nla_nest_end(skb, nest);
264
265	if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
266		goto nla_put_failure;
267
268	return skb->len;
269
270nla_put_failure:
271	nlmsg_trim(skb, b);
272	return -1;
273}
274
275static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
276	.kind		=	"cgroup",
277	.init		=	cls_cgroup_init,
278	.change		=	cls_cgroup_change,
279	.classify	=	cls_cgroup_classify,
280	.destroy	=	cls_cgroup_destroy,
281	.get		=	cls_cgroup_get,
282	.put		=	cls_cgroup_put,
283	.delete		=	cls_cgroup_delete,
284	.walk		=	cls_cgroup_walk,
285	.dump		=	cls_cgroup_dump,
286	.owner		=	THIS_MODULE,
287};
288
289static int __init init_cgroup_cls(void)
290{
291	int ret;
292
293	ret = cgroup_load_subsys(&net_cls_subsys);
294	if (ret)
295		goto out;
296
297#ifndef CONFIG_NET_CLS_CGROUP
298	/* We can't use rcu_assign_pointer because this is an int. */
299	smp_wmb();
300	net_cls_subsys_id = net_cls_subsys.subsys_id;
301#endif
302
303	ret = register_tcf_proto_ops(&cls_cgroup_ops);
304	if (ret)
305		cgroup_unload_subsys(&net_cls_subsys);
306
307out:
308	return ret;
309}
310
311static void __exit exit_cgroup_cls(void)
312{
313	unregister_tcf_proto_ops(&cls_cgroup_ops);
314
315#ifndef CONFIG_NET_CLS_CGROUP
316	net_cls_subsys_id = -1;
317	synchronize_rcu();
318#endif
319
320	cgroup_unload_subsys(&net_cls_subsys);
321}
322
323module_init(init_cgroup_cls);
324module_exit(exit_cgroup_cls);
325MODULE_LICENSE("GPL");
326