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/skbuff.h>
15#include <linux/rcupdate.h>
16#include <net/rtnetlink.h>
17#include <net/pkt_cls.h>
18#include <net/sock.h>
19#include <net/cls_cgroup.h>
20
21struct cls_cgroup_head {
22	u32			handle;
23	struct tcf_exts		exts;
24	struct tcf_ematch_tree	ematches;
25	struct tcf_proto	*tp;
26	struct rcu_head		rcu;
27};
28
29static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
30			       struct tcf_result *res)
31{
32	struct cls_cgroup_head *head = rcu_dereference_bh(tp->root);
33	u32 classid;
34
35	classid = task_cls_state(current)->classid;
36
37	/*
38	 * Due to the nature of the classifier it is required to ignore all
39	 * packets originating from softirq context as accessing `current'
40	 * would lead to false results.
41	 *
42	 * This test assumes that all callers of dev_queue_xmit() explicitely
43	 * disable bh. Knowing this, it is possible to detect softirq based
44	 * calls by looking at the number of nested bh disable calls because
45	 * softirqs always disables bh.
46	 */
47	if (in_serving_softirq()) {
48		/* If there is an sk_classid we'll use that. */
49		if (!skb->sk)
50			return -1;
51		classid = skb->sk->sk_classid;
52	}
53
54	if (!classid)
55		return -1;
56
57	if (!tcf_em_tree_match(skb, &head->ematches, NULL))
58		return -1;
59
60	res->classid = classid;
61	res->class = 0;
62	return tcf_exts_exec(skb, &head->exts, res);
63}
64
65static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
66{
67	return 0UL;
68}
69
70static int cls_cgroup_init(struct tcf_proto *tp)
71{
72	return 0;
73}
74
75static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
76	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
77};
78
79static void cls_cgroup_destroy_rcu(struct rcu_head *root)
80{
81	struct cls_cgroup_head *head = container_of(root,
82						    struct cls_cgroup_head,
83						    rcu);
84
85	tcf_exts_destroy(&head->exts);
86	tcf_em_tree_destroy(&head->ematches);
87	kfree(head);
88}
89
90static int cls_cgroup_change(struct net *net, struct sk_buff *in_skb,
91			     struct tcf_proto *tp, unsigned long base,
92			     u32 handle, struct nlattr **tca,
93			     unsigned long *arg, bool ovr)
94{
95	struct nlattr *tb[TCA_CGROUP_MAX + 1];
96	struct cls_cgroup_head *head = rtnl_dereference(tp->root);
97	struct cls_cgroup_head *new;
98	struct tcf_ematch_tree t;
99	struct tcf_exts e;
100	int err;
101
102	if (!tca[TCA_OPTIONS])
103		return -EINVAL;
104
105	if (!head && !handle)
106		return -EINVAL;
107
108	if (head && handle != head->handle)
109		return -ENOENT;
110
111	new = kzalloc(sizeof(*head), GFP_KERNEL);
112	if (!new)
113		return -ENOBUFS;
114
115	tcf_exts_init(&new->exts, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
116	new->handle = handle;
117	new->tp = tp;
118	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
119			       cgroup_policy);
120	if (err < 0)
121		goto errout;
122
123	tcf_exts_init(&e, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
124	err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
125	if (err < 0)
126		goto errout;
127
128	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
129	if (err < 0) {
130		tcf_exts_destroy(&e);
131		goto errout;
132	}
133
134	tcf_exts_change(tp, &new->exts, &e);
135	tcf_em_tree_change(tp, &new->ematches, &t);
136
137	rcu_assign_pointer(tp->root, new);
138	if (head)
139		call_rcu(&head->rcu, cls_cgroup_destroy_rcu);
140	return 0;
141errout:
142	kfree(new);
143	return err;
144}
145
146static bool cls_cgroup_destroy(struct tcf_proto *tp, bool force)
147{
148	struct cls_cgroup_head *head = rtnl_dereference(tp->root);
149
150	if (!force)
151		return false;
152
153	if (head) {
154		RCU_INIT_POINTER(tp->root, NULL);
155		call_rcu(&head->rcu, cls_cgroup_destroy_rcu);
156	}
157	return true;
158}
159
160static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
161{
162	return -EOPNOTSUPP;
163}
164
165static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
166{
167	struct cls_cgroup_head *head = rtnl_dereference(tp->root);
168
169	if (arg->count < arg->skip)
170		goto skip;
171
172	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
173		arg->stop = 1;
174		return;
175	}
176skip:
177	arg->count++;
178}
179
180static int cls_cgroup_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
181			   struct sk_buff *skb, struct tcmsg *t)
182{
183	struct cls_cgroup_head *head = rtnl_dereference(tp->root);
184	struct nlattr *nest;
185
186	t->tcm_handle = head->handle;
187
188	nest = nla_nest_start(skb, TCA_OPTIONS);
189	if (nest == NULL)
190		goto nla_put_failure;
191
192	if (tcf_exts_dump(skb, &head->exts) < 0 ||
193	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
194		goto nla_put_failure;
195
196	nla_nest_end(skb, nest);
197
198	if (tcf_exts_dump_stats(skb, &head->exts) < 0)
199		goto nla_put_failure;
200
201	return skb->len;
202
203nla_put_failure:
204	nla_nest_cancel(skb, nest);
205	return -1;
206}
207
208static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
209	.kind		=	"cgroup",
210	.init		=	cls_cgroup_init,
211	.change		=	cls_cgroup_change,
212	.classify	=	cls_cgroup_classify,
213	.destroy	=	cls_cgroup_destroy,
214	.get		=	cls_cgroup_get,
215	.delete		=	cls_cgroup_delete,
216	.walk		=	cls_cgroup_walk,
217	.dump		=	cls_cgroup_dump,
218	.owner		=	THIS_MODULE,
219};
220
221static int __init init_cgroup_cls(void)
222{
223	return register_tcf_proto_ops(&cls_cgroup_ops);
224}
225
226static void __exit exit_cgroup_cls(void)
227{
228	unregister_tcf_proto_ops(&cls_cgroup_ops);
229}
230
231module_init(init_cgroup_cls);
232module_exit(exit_cgroup_cls);
233MODULE_LICENSE("GPL");
234