cls_cgroup.c revision 92fb97487a7e41b222c1417cabd1d1ab7cc3a48c 
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_css_alloc(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_css_free(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	.css_alloc	= cgrp_css_alloc,
79	.css_free	= cgrp_css_free,
80	.subsys_id	= net_cls_subsys_id,
81	.base_cftypes	= ss_files,
82	.module		= THIS_MODULE,
83
84	/*
85	 * While net_cls cgroup has the rudimentary hierarchy support of
86	 * inheriting the parent's classid on cgroup creation, it doesn't
87	 * properly propagates config changes in ancestors to their
88	 * descendents.  A child should follow the parent's configuration
89	 * but be allowed to override it.  Fix it and remove the following.
90	 */
91	.broken_hierarchy = true,
92};
93
94struct cls_cgroup_head {
95	u32			handle;
96	struct tcf_exts		exts;
97	struct tcf_ematch_tree	ematches;
98};
99
100static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
101			       struct tcf_result *res)
102{
103	struct cls_cgroup_head *head = tp->root;
104	u32 classid;
105
106	rcu_read_lock();
107	classid = task_cls_state(current)->classid;
108	rcu_read_unlock();
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 (in_serving_softirq()) {
121		/* If there is an sk_classid we'll use that. */
122		if (!skb->sk)
123			return -1;
124		classid = skb->sk->sk_classid;
125	}
126
127	if (!classid)
128		return -1;
129
130	if (!tcf_em_tree_match(skb, &head->ematches, NULL))
131		return -1;
132
133	res->classid = classid;
134	res->class = 0;
135	return tcf_exts_exec(skb, &head->exts, res);
136}
137
138static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
139{
140	return 0UL;
141}
142
143static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
144{
145}
146
147static int cls_cgroup_init(struct tcf_proto *tp)
148{
149	return 0;
150}
151
152static const struct tcf_ext_map cgroup_ext_map = {
153	.action = TCA_CGROUP_ACT,
154	.police = TCA_CGROUP_POLICE,
155};
156
157static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
158	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
159};
160
161static int cls_cgroup_change(struct sk_buff *in_skb,
162			     struct tcf_proto *tp, unsigned long base,
163			     u32 handle, struct nlattr **tca,
164			     unsigned long *arg)
165{
166	struct nlattr *tb[TCA_CGROUP_MAX + 1];
167	struct cls_cgroup_head *head = tp->root;
168	struct tcf_ematch_tree t;
169	struct tcf_exts e;
170	int err;
171
172	if (!tca[TCA_OPTIONS])
173		return -EINVAL;
174
175	if (head == NULL) {
176		if (!handle)
177			return -EINVAL;
178
179		head = kzalloc(sizeof(*head), GFP_KERNEL);
180		if (head == NULL)
181			return -ENOBUFS;
182
183		head->handle = handle;
184
185		tcf_tree_lock(tp);
186		tp->root = head;
187		tcf_tree_unlock(tp);
188	}
189
190	if (handle != head->handle)
191		return -ENOENT;
192
193	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
194			       cgroup_policy);
195	if (err < 0)
196		return err;
197
198	err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
199	if (err < 0)
200		return err;
201
202	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
203	if (err < 0)
204		return err;
205
206	tcf_exts_change(tp, &head->exts, &e);
207	tcf_em_tree_change(tp, &head->ematches, &t);
208
209	return 0;
210}
211
212static void cls_cgroup_destroy(struct tcf_proto *tp)
213{
214	struct cls_cgroup_head *head = tp->root;
215
216	if (head) {
217		tcf_exts_destroy(tp, &head->exts);
218		tcf_em_tree_destroy(tp, &head->ematches);
219		kfree(head);
220	}
221}
222
223static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
224{
225	return -EOPNOTSUPP;
226}
227
228static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
229{
230	struct cls_cgroup_head *head = tp->root;
231
232	if (arg->count < arg->skip)
233		goto skip;
234
235	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
236		arg->stop = 1;
237		return;
238	}
239skip:
240	arg->count++;
241}
242
243static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
244			   struct sk_buff *skb, struct tcmsg *t)
245{
246	struct cls_cgroup_head *head = tp->root;
247	unsigned char *b = skb_tail_pointer(skb);
248	struct nlattr *nest;
249
250	t->tcm_handle = head->handle;
251
252	nest = nla_nest_start(skb, TCA_OPTIONS);
253	if (nest == NULL)
254		goto nla_put_failure;
255
256	if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
257	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
258		goto nla_put_failure;
259
260	nla_nest_end(skb, nest);
261
262	if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
263		goto nla_put_failure;
264
265	return skb->len;
266
267nla_put_failure:
268	nlmsg_trim(skb, b);
269	return -1;
270}
271
272static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
273	.kind		=	"cgroup",
274	.init		=	cls_cgroup_init,
275	.change		=	cls_cgroup_change,
276	.classify	=	cls_cgroup_classify,
277	.destroy	=	cls_cgroup_destroy,
278	.get		=	cls_cgroup_get,
279	.put		=	cls_cgroup_put,
280	.delete		=	cls_cgroup_delete,
281	.walk		=	cls_cgroup_walk,
282	.dump		=	cls_cgroup_dump,
283	.owner		=	THIS_MODULE,
284};
285
286static int __init init_cgroup_cls(void)
287{
288	int ret;
289
290	ret = cgroup_load_subsys(&net_cls_subsys);
291	if (ret)
292		goto out;
293
294	ret = register_tcf_proto_ops(&cls_cgroup_ops);
295	if (ret)
296		cgroup_unload_subsys(&net_cls_subsys);
297
298out:
299	return ret;
300}
301
302static void __exit exit_cgroup_cls(void)
303{
304	unregister_tcf_proto_ops(&cls_cgroup_ops);
305
306	cgroup_unload_subsys(&net_cls_subsys);
307}
308
309module_init(init_cgroup_cls);
310module_exit(exit_cgroup_cls);
311MODULE_LICENSE("GPL");
312