cls_cgroup.c revision 75e1056f5c57050415b64cb761a3acc35d91f013 
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#else
38#define net_cls_subsys_id net_cls_subsys.subsys_id
39#endif
40	.module		= THIS_MODULE,
41};
42
43
44static inline struct cgroup_cls_state *cgrp_cls_state(struct cgroup *cgrp)
45{
46	return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
47			    struct cgroup_cls_state, css);
48}
49
50static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
51{
52	return container_of(task_subsys_state(p, net_cls_subsys_id),
53			    struct cgroup_cls_state, css);
54}
55
56static struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss,
57						 struct cgroup *cgrp)
58{
59	struct cgroup_cls_state *cs;
60
61	if (!(cs = kzalloc(sizeof(*cs), GFP_KERNEL)))
62		return ERR_PTR(-ENOMEM);
63
64	if (cgrp->parent)
65		cs->classid = cgrp_cls_state(cgrp->parent)->classid;
66
67	return &cs->css;
68}
69
70static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
71{
72	kfree(cgrp_cls_state(cgrp));
73}
74
75static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
76{
77	return cgrp_cls_state(cgrp)->classid;
78}
79
80static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
81{
82	cgrp_cls_state(cgrp)->classid = (u32) value;
83	return 0;
84}
85
86static struct cftype ss_files[] = {
87	{
88		.name = "classid",
89		.read_u64 = read_classid,
90		.write_u64 = write_classid,
91	},
92};
93
94static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
95{
96	return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files));
97}
98
99struct cls_cgroup_head
100{
101	u32			handle;
102	struct tcf_exts		exts;
103	struct tcf_ematch_tree	ematches;
104};
105
106static int cls_cgroup_classify(struct sk_buff *skb, struct tcf_proto *tp,
107			       struct tcf_result *res)
108{
109	struct cls_cgroup_head *head = tp->root;
110	u32 classid;
111
112	rcu_read_lock();
113	classid = task_cls_state(current)->classid;
114	rcu_read_unlock();
115
116	/*
117	 * Due to the nature of the classifier it is required to ignore all
118	 * packets originating from softirq context as accessing `current'
119	 * would lead to false results.
120	 *
121	 * This test assumes that all callers of dev_queue_xmit() explicitely
122	 * disable bh. Knowing this, it is possible to detect softirq based
123	 * calls by looking at the number of nested bh disable calls because
124	 * softirqs always disables bh.
125	 */
126	if (in_serving_softirq()) {
127		/* If there is an sk_classid we'll use that. */
128		if (!skb->sk)
129			return -1;
130		classid = skb->sk->sk_classid;
131	}
132
133	if (!classid)
134		return -1;
135
136	if (!tcf_em_tree_match(skb, &head->ematches, NULL))
137		return -1;
138
139	res->classid = classid;
140	res->class = 0;
141	return tcf_exts_exec(skb, &head->exts, res);
142}
143
144static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
145{
146	return 0UL;
147}
148
149static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
150{
151}
152
153static int cls_cgroup_init(struct tcf_proto *tp)
154{
155	return 0;
156}
157
158static const struct tcf_ext_map cgroup_ext_map = {
159	.action = TCA_CGROUP_ACT,
160	.police = TCA_CGROUP_POLICE,
161};
162
163static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
164	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
165};
166
167static int cls_cgroup_change(struct tcf_proto *tp, unsigned long base,
168			     u32 handle, struct nlattr **tca,
169			     unsigned long *arg)
170{
171	struct nlattr *tb[TCA_CGROUP_MAX+1];
172	struct cls_cgroup_head *head = tp->root;
173	struct tcf_ematch_tree t;
174	struct tcf_exts e;
175	int err;
176
177	if (!tca[TCA_OPTIONS])
178		return -EINVAL;
179
180	if (head == NULL) {
181		if (!handle)
182			return -EINVAL;
183
184		head = kzalloc(sizeof(*head), GFP_KERNEL);
185		if (head == NULL)
186			return -ENOBUFS;
187
188		head->handle = handle;
189
190		tcf_tree_lock(tp);
191		tp->root = head;
192		tcf_tree_unlock(tp);
193	}
194
195	if (handle != head->handle)
196		return -ENOENT;
197
198	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
199			       cgroup_policy);
200	if (err < 0)
201		return err;
202
203	err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
204	if (err < 0)
205		return err;
206
207	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
208	if (err < 0)
209		return err;
210
211	tcf_exts_change(tp, &head->exts, &e);
212	tcf_em_tree_change(tp, &head->ematches, &t);
213
214	return 0;
215}
216
217static void cls_cgroup_destroy(struct tcf_proto *tp)
218{
219	struct cls_cgroup_head *head = tp->root;
220
221	if (head) {
222		tcf_exts_destroy(tp, &head->exts);
223		tcf_em_tree_destroy(tp, &head->ematches);
224		kfree(head);
225	}
226}
227
228static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
229{
230	return -EOPNOTSUPP;
231}
232
233static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
234{
235	struct cls_cgroup_head *head = tp->root;
236
237	if (arg->count < arg->skip)
238		goto skip;
239
240	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
241		arg->stop = 1;
242		return;
243	}
244skip:
245	arg->count++;
246}
247
248static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
249			   struct sk_buff *skb, struct tcmsg *t)
250{
251	struct cls_cgroup_head *head = tp->root;
252	unsigned char *b = skb_tail_pointer(skb);
253	struct nlattr *nest;
254
255	t->tcm_handle = head->handle;
256
257	nest = nla_nest_start(skb, TCA_OPTIONS);
258	if (nest == NULL)
259		goto nla_put_failure;
260
261	if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
262	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
263		goto nla_put_failure;
264
265	nla_nest_end(skb, nest);
266
267	if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
268		goto nla_put_failure;
269
270	return skb->len;
271
272nla_put_failure:
273	nlmsg_trim(skb, b);
274	return -1;
275}
276
277static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
278	.kind		=	"cgroup",
279	.init		=	cls_cgroup_init,
280	.change		=	cls_cgroup_change,
281	.classify	=	cls_cgroup_classify,
282	.destroy	=	cls_cgroup_destroy,
283	.get		=	cls_cgroup_get,
284	.put		=	cls_cgroup_put,
285	.delete		=	cls_cgroup_delete,
286	.walk		=	cls_cgroup_walk,
287	.dump		=	cls_cgroup_dump,
288	.owner		=	THIS_MODULE,
289};
290
291static int __init init_cgroup_cls(void)
292{
293	int ret;
294
295	ret = cgroup_load_subsys(&net_cls_subsys);
296	if (ret)
297		goto out;
298
299#ifndef CONFIG_NET_CLS_CGROUP
300	/* We can't use rcu_assign_pointer because this is an int. */
301	smp_wmb();
302	net_cls_subsys_id = net_cls_subsys.subsys_id;
303#endif
304
305	ret = register_tcf_proto_ops(&cls_cgroup_ops);
306	if (ret)
307		cgroup_unload_subsys(&net_cls_subsys);
308
309out:
310	return ret;
311}
312
313static void __exit exit_cgroup_cls(void)
314{
315	unregister_tcf_proto_ops(&cls_cgroup_ops);
316
317#ifndef CONFIG_NET_CLS_CGROUP
318	net_cls_subsys_id = -1;
319	synchronize_rcu();
320#endif
321
322	cgroup_unload_subsys(&net_cls_subsys);
323}
324
325module_init(init_cgroup_cls);
326module_exit(exit_cgroup_cls);
327MODULE_LICENSE("GPL");
328