Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6.git] / net / sched / cls_cgroup.c
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/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
21 struct cgroup_cls_state
22 {
23         struct cgroup_subsys_state css;
24         u32 classid;
25 };
26
27 static inline struct cgroup_cls_state *cgrp_cls_state(struct cgroup *cgrp)
28 {
29         return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
30                             struct cgroup_cls_state, css);
31 }
32
33 static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
34 {
35         return container_of(task_subsys_state(p, net_cls_subsys_id),
36                             struct cgroup_cls_state, css);
37 }
38
39 static struct cgroup_subsys_state *cgrp_create(struct cgroup_subsys *ss,
40                                                  struct cgroup *cgrp)
41 {
42         struct cgroup_cls_state *cs;
43
44         if (!(cs = kzalloc(sizeof(*cs), GFP_KERNEL)))
45                 return ERR_PTR(-ENOMEM);
46
47         if (cgrp->parent)
48                 cs->classid = cgrp_cls_state(cgrp->parent)->classid;
49
50         return &cs->css;
51 }
52
53 static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
54 {
55         kfree(cgrp_cls_state(cgrp));
56 }
57
58 static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
59 {
60         return cgrp_cls_state(cgrp)->classid;
61 }
62
63 static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
64 {
65         cgrp_cls_state(cgrp)->classid = (u32) value;
66         return 0;
67 }
68
69 static struct cftype ss_files[] = {
70         {
71                 .name = "classid",
72                 .read_u64 = read_classid,
73                 .write_u64 = write_classid,
74         },
75 };
76
77 static 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
82 struct 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
90 struct cls_cgroup_head
91 {
92         u32                     handle;
93         struct tcf_exts         exts;
94         struct tcf_ematch_tree  ematches;
95 };
96
97 static 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         u32 classid;
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 (softirq_count() != SOFTIRQ_OFFSET)
114                 return -1;
115
116         rcu_read_lock();
117         classid = task_cls_state(current)->classid;
118         rcu_read_unlock();
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
131 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
132 {
133         return 0UL;
134 }
135
136 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
137 {
138 }
139
140 static int cls_cgroup_init(struct tcf_proto *tp)
141 {
142         return 0;
143 }
144
145 static const struct tcf_ext_map cgroup_ext_map = {
146         .action = TCA_CGROUP_ACT,
147         .police = TCA_CGROUP_POLICE,
148 };
149
150 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
151         [TCA_CGROUP_EMATCHES]   = { .type = NLA_NESTED },
152 };
153
154 static 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 (head == NULL) {
165                 if (!handle)
166                         return -EINVAL;
167
168                 head = kzalloc(sizeof(*head), GFP_KERNEL);
169                 if (head == NULL)
170                         return -ENOBUFS;
171
172                 head->handle = handle;
173
174                 tcf_tree_lock(tp);
175                 tp->root = head;
176                 tcf_tree_unlock(tp);
177         }
178
179         if (handle != head->handle)
180                 return -ENOENT;
181
182         err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
183                                cgroup_policy);
184         if (err < 0)
185                 return err;
186
187         err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
188         if (err < 0)
189                 return err;
190
191         err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
192         if (err < 0)
193                 return err;
194
195         tcf_exts_change(tp, &head->exts, &e);
196         tcf_em_tree_change(tp, &head->ematches, &t);
197
198         return 0;
199 }
200
201 static void cls_cgroup_destroy(struct tcf_proto *tp)
202 {
203         struct cls_cgroup_head *head = tp->root;
204
205         if (head) {
206                 tcf_exts_destroy(tp, &head->exts);
207                 tcf_em_tree_destroy(tp, &head->ematches);
208                 kfree(head);
209         }
210 }
211
212 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
213 {
214         return -EOPNOTSUPP;
215 }
216
217 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
218 {
219         struct cls_cgroup_head *head = tp->root;
220
221         if (arg->count < arg->skip)
222                 goto skip;
223
224         if (arg->fn(tp, (unsigned long) head, arg) < 0) {
225                 arg->stop = 1;
226                 return;
227         }
228 skip:
229         arg->count++;
230 }
231
232 static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
233                            struct sk_buff *skb, struct tcmsg *t)
234 {
235         struct cls_cgroup_head *head = tp->root;
236         unsigned char *b = skb_tail_pointer(skb);
237         struct nlattr *nest;
238
239         t->tcm_handle = head->handle;
240
241         nest = nla_nest_start(skb, TCA_OPTIONS);
242         if (nest == NULL)
243                 goto nla_put_failure;
244
245         if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
246             tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
247                 goto nla_put_failure;
248
249         nla_nest_end(skb, nest);
250
251         if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
252                 goto nla_put_failure;
253
254         return skb->len;
255
256 nla_put_failure:
257         nlmsg_trim(skb, b);
258         return -1;
259 }
260
261 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
262         .kind           =       "cgroup",
263         .init           =       cls_cgroup_init,
264         .change         =       cls_cgroup_change,
265         .classify       =       cls_cgroup_classify,
266         .destroy        =       cls_cgroup_destroy,
267         .get            =       cls_cgroup_get,
268         .put            =       cls_cgroup_put,
269         .delete         =       cls_cgroup_delete,
270         .walk           =       cls_cgroup_walk,
271         .dump           =       cls_cgroup_dump,
272         .owner          =       THIS_MODULE,
273 };
274
275 static int __init init_cgroup_cls(void)
276 {
277         return register_tcf_proto_ops(&cls_cgroup_ops);
278 }
279
280 static void __exit exit_cgroup_cls(void)
281 {
282         unregister_tcf_proto_ops(&cls_cgroup_ops);
283 }
284
285 module_init(init_cgroup_cls);
286 module_exit(exit_cgroup_cls);
287 MODULE_LICENSE("GPL");