blob: 53ada2c0e41ce2ccc2debf5e1e1687ccb5aedd24 [file] [log] [blame]
Thomas Graff4009232008-11-07 22:56:00 -08001/*
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/types.h>
14#include <linux/string.h>
15#include <linux/errno.h>
16#include <linux/skbuff.h>
17#include <linux/cgroup.h>
18#include <net/rtnetlink.h>
19#include <net/pkt_cls.h>
20
21struct cgroup_cls_state
22{
23 struct cgroup_subsys_state css;
24 u32 classid;
25};
26
27static inline struct cgroup_cls_state *net_cls_state(struct cgroup *cgrp)
28{
29 return (struct cgroup_cls_state *)
30 cgroup_subsys_state(cgrp, net_cls_subsys_id);
31}
32
33static struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss,
34 struct cgroup *cgrp)
35{
36 struct cgroup_cls_state *cs;
37
38 if (!(cs = kzalloc(sizeof(*cs), GFP_KERNEL)))
39 return ERR_PTR(-ENOMEM);
40
41 if (cgrp->parent)
42 cs->classid = net_cls_state(cgrp->parent)->classid;
43
44 return &cs->css;
45}
46
47static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
48{
49 kfree(ss);
50}
51
52static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
53{
54 return net_cls_state(cgrp)->classid;
55}
56
57static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
58{
59 if (!cgroup_lock_live_group(cgrp))
60 return -ENODEV;
61
62 net_cls_state(cgrp)->classid = (u32) value;
63
64 cgroup_unlock();
65
66 return 0;
67}
68
69static struct cftype ss_files[] = {
70 {
71 .name = "classid",
72 .read_u64 = read_classid,
73 .write_u64 = write_classid,
74 },
75};
76
77static int cgrp_populate(struct cgroup_subsys *ss, struct cgroup *cgrp)
78{
79 return cgroup_add_files(cgrp, ss, ss_files, ARRAY_SIZE(ss_files));
80}
81
82struct cgroup_subsys net_cls_subsys = {
83 .name = "net_cls",
84 .create = cgrp_create,
85 .destroy = cgrp_destroy,
86 .populate = cgrp_populate,
87 .subsys_id = net_cls_subsys_id,
88};
89
90struct cls_cgroup_head
91{
92 u32 handle;
93 struct tcf_exts exts;
94 struct tcf_ematch_tree ematches;
95};
96
97static int cls_cgroup_classify(struct sk_buff *skb, struct tcf_proto *tp,
98 struct tcf_result *res)
99{
100 struct cls_cgroup_head *head = tp->root;
101 struct cgroup_cls_state *cs;
102 int ret = 0;
103
104 /*
105 * Due to the nature of the classifier it is required to ignore all
106 * packets originating from softirq context as accessing `current'
107 * would lead to false results.
108 *
109 * This test assumes that all callers of dev_queue_xmit() explicitely
110 * disable bh. Knowing this, it is possible to detect softirq based
111 * calls by looking at the number of nested bh disable calls because
112 * softirqs always disables bh.
113 */
114 if (softirq_count() != SOFTIRQ_OFFSET)
115 return -1;
116
117 rcu_read_lock();
118 cs = (struct cgroup_cls_state *) task_subsys_state(current,
119 net_cls_subsys_id);
120 if (cs->classid && tcf_em_tree_match(skb, &head->ematches, NULL)) {
121 res->classid = cs->classid;
122 res->class = 0;
123 ret = tcf_exts_exec(skb, &head->exts, res);
124 } else
125 ret = -1;
126
127 rcu_read_unlock();
128
129 return ret;
130}
131
132static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
133{
134 return 0UL;
135}
136
137static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
138{
139}
140
141static int cls_cgroup_init(struct tcf_proto *tp)
142{
143 return 0;
144}
145
146static const struct tcf_ext_map cgroup_ext_map = {
147 .action = TCA_CGROUP_ACT,
148 .police = TCA_CGROUP_POLICE,
149};
150
151static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
152 [TCA_CGROUP_EMATCHES] = { .type = NLA_NESTED },
153};
154
155static int cls_cgroup_change(struct tcf_proto *tp, unsigned long base,
156 u32 handle, struct nlattr **tca,
157 unsigned long *arg)
158{
159 struct nlattr *tb[TCA_CGROUP_MAX+1];
160 struct cls_cgroup_head *head = tp->root;
161 struct tcf_ematch_tree t;
162 struct tcf_exts e;
163 int err;
164
165 if (head == NULL) {
166 if (!handle)
167 return -EINVAL;
168
169 head = kzalloc(sizeof(*head), GFP_KERNEL);
170 if (head == NULL)
171 return -ENOBUFS;
172
173 head->handle = handle;
174
175 tcf_tree_lock(tp);
176 tp->root = head;
177 tcf_tree_unlock(tp);
178 }
179
180 if (handle != head->handle)
181 return -ENOENT;
182
183 err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
184 cgroup_policy);
185 if (err < 0)
186 return err;
187
188 err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
189 if (err < 0)
190 return err;
191
192 err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
193 if (err < 0)
194 return err;
195
196 tcf_exts_change(tp, &head->exts, &e);
197 tcf_em_tree_change(tp, &head->ematches, &t);
198
199 return 0;
200}
201
202static void cls_cgroup_destroy(struct tcf_proto *tp)
203{
204 struct cls_cgroup_head *head;
205
206 head = (struct cls_cgroup_head *)xchg(&tp->root, NULL);
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 return register_tcf_proto_ops(&cls_cgroup_ops);
281}
282
283static void __exit exit_cgroup_cls(void)
284{
285 unregister_tcf_proto_ops(&cls_cgroup_ops);
286}
287
288module_init(init_cgroup_cls);
289module_exit(exit_cgroup_cls);
290MODULE_LICENSE("GPL");