cls_cgroup.c revision 4baf6e33251b37f111e21289f8ee71fe4cce236e 
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	{ }	/* terminate */
74};
75
76struct cgroup_subsys net_cls_subsys = {
77	.name		= "net_cls",
78	.create		= cgrp_create,
79	.destroy	= cgrp_destroy,
80#ifdef CONFIG_NET_CLS_CGROUP
81	.subsys_id	= net_cls_subsys_id,
82#endif
83	.base_cftypes	= ss_files,
84	.module		= THIS_MODULE,
85};
86
87struct cls_cgroup_head {
88	u32			handle;
89	struct tcf_exts		exts;
90	struct tcf_ematch_tree	ematches;
91};
92
93static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
94			       struct tcf_result *res)
95{
96	struct cls_cgroup_head *head = tp->root;
97	u32 classid;
98
99	rcu_read_lock();
100	classid = task_cls_state(current)->classid;
101	rcu_read_unlock();
102
103	/*
104	 * Due to the nature of the classifier it is required to ignore all
105	 * packets originating from softirq context as accessing `current'
106	 * would lead to false results.
107	 *
108	 * This test assumes that all callers of dev_queue_xmit() explicitely
109	 * disable bh. Knowing this, it is possible to detect softirq based
110	 * calls by looking at the number of nested bh disable calls because
111	 * softirqs always disables bh.
112	 */
113	if (in_serving_softirq()) {
114		/* If there is an sk_classid we'll use that. */
115		if (!skb->sk)
116			return -1;
117		classid = skb->sk->sk_classid;
118	}
119
120	if (!classid)
121		return -1;
122
123	if (!tcf_em_tree_match(skb, &head->ematches, NULL))
124		return -1;
125
126	res->classid = classid;
127	res->class = 0;
128	return tcf_exts_exec(skb, &head->exts, res);
129}
130
131static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
132{
133	return 0UL;
134}
135
136static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
137{
138}
139
140static int cls_cgroup_init(struct tcf_proto *tp)
141{
142	return 0;
143}
144
145static const struct tcf_ext_map cgroup_ext_map = {
146	.action = TCA_CGROUP_ACT,
147	.police = TCA_CGROUP_POLICE,
148};
149
150static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
151	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
152};
153
154static int cls_cgroup_change(struct tcf_proto *tp, unsigned long base,
155			     u32 handle, struct nlattr **tca,
156			     unsigned long *arg)
157{
158	struct nlattr *tb[TCA_CGROUP_MAX + 1];
159	struct cls_cgroup_head *head = tp->root;
160	struct tcf_ematch_tree t;
161	struct tcf_exts e;
162	int err;
163
164	if (!tca[TCA_OPTIONS])
165		return -EINVAL;
166
167	if (head == NULL) {
168		if (!handle)
169			return -EINVAL;
170
171		head = kzalloc(sizeof(*head), GFP_KERNEL);
172		if (head == NULL)
173			return -ENOBUFS;
174
175		head->handle = handle;
176
177		tcf_tree_lock(tp);
178		tp->root = head;
179		tcf_tree_unlock(tp);
180	}
181
182	if (handle != head->handle)
183		return -ENOENT;
184
185	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
186			       cgroup_policy);
187	if (err < 0)
188		return err;
189
190	err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
191	if (err < 0)
192		return err;
193
194	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
195	if (err < 0)
196		return err;
197
198	tcf_exts_change(tp, &head->exts, &e);
199	tcf_em_tree_change(tp, &head->ematches, &t);
200
201	return 0;
202}
203
204static void cls_cgroup_destroy(struct tcf_proto *tp)
205{
206	struct cls_cgroup_head *head = tp->root;
207
208	if (head) {
209		tcf_exts_destroy(tp, &head->exts);
210		tcf_em_tree_destroy(tp, &head->ematches);
211		kfree(head);
212	}
213}
214
215static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
216{
217	return -EOPNOTSUPP;
218}
219
220static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
221{
222	struct cls_cgroup_head *head = tp->root;
223
224	if (arg->count < arg->skip)
225		goto skip;
226
227	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
228		arg->stop = 1;
229		return;
230	}
231skip:
232	arg->count++;
233}
234
235static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
236			   struct sk_buff *skb, struct tcmsg *t)
237{
238	struct cls_cgroup_head *head = tp->root;
239	unsigned char *b = skb_tail_pointer(skb);
240	struct nlattr *nest;
241
242	t->tcm_handle = head->handle;
243
244	nest = nla_nest_start(skb, TCA_OPTIONS);
245	if (nest == NULL)
246		goto nla_put_failure;
247
248	if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
249	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
250		goto nla_put_failure;
251
252	nla_nest_end(skb, nest);
253
254	if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
255		goto nla_put_failure;
256
257	return skb->len;
258
259nla_put_failure:
260	nlmsg_trim(skb, b);
261	return -1;
262}
263
264static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
265	.kind		=	"cgroup",
266	.init		=	cls_cgroup_init,
267	.change		=	cls_cgroup_change,
268	.classify	=	cls_cgroup_classify,
269	.destroy	=	cls_cgroup_destroy,
270	.get		=	cls_cgroup_get,
271	.put		=	cls_cgroup_put,
272	.delete		=	cls_cgroup_delete,
273	.walk		=	cls_cgroup_walk,
274	.dump		=	cls_cgroup_dump,
275	.owner		=	THIS_MODULE,
276};
277
278static int __init init_cgroup_cls(void)
279{
280	int ret;
281
282	ret = cgroup_load_subsys(&net_cls_subsys);
283	if (ret)
284		goto out;
285
286#ifndef CONFIG_NET_CLS_CGROUP
287	/* We can't use rcu_assign_pointer because this is an int. */
288	smp_wmb();
289	net_cls_subsys_id = net_cls_subsys.subsys_id;
290#endif
291
292	ret = register_tcf_proto_ops(&cls_cgroup_ops);
293	if (ret)
294		cgroup_unload_subsys(&net_cls_subsys);
295
296out:
297	return ret;
298}
299
300static void __exit exit_cgroup_cls(void)
301{
302	unregister_tcf_proto_ops(&cls_cgroup_ops);
303
304#ifndef CONFIG_NET_CLS_CGROUP
305	net_cls_subsys_id = -1;
306	synchronize_rcu();
307#endif
308
309	cgroup_unload_subsys(&net_cls_subsys);
310}
311
312module_init(init_cgroup_cls);
313module_exit(exit_cgroup_cls);
314MODULE_LICENSE("GPL");
315