cls_cgroup: fix an oops when removing a cgroup
[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 *net_cls_state(struct cgroup *cgrp)
28 {
29         return (struct cgroup_cls_state *)
30                 cgroup_subsys_state(cgrp, net_cls_subsys_id);
31 }
32
33 static 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
47 static void cgrp_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
48 {
49         kfree(net_cls_state(cgrp));
50 }
51
52 static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
53 {
54         return net_cls_state(cgrp)->classid;
55 }
56
57 static 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
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         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
132 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
133 {
134         return 0UL;
135 }
136
137 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
138 {
139 }
140
141 static int cls_cgroup_init(struct tcf_proto *tp)
142 {
143         return 0;
144 }
145
146 static const struct tcf_ext_map cgroup_ext_map = {
147         .action = TCA_CGROUP_ACT,
148         .police = TCA_CGROUP_POLICE,
149 };
150
151 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
152         [TCA_CGROUP_EMATCHES]   = { .type = NLA_NESTED },
153 };
154
155 static 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
202 static void cls_cgroup_destroy(struct tcf_proto *tp)
203 {
204         struct cls_cgroup_head *head = tp->root;
205
206         if (head) {
207                 tcf_exts_destroy(tp, &head->exts);
208                 tcf_em_tree_destroy(tp, &head->ematches);
209                 kfree(head);
210         }
211 }
212
213 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
214 {
215         return -EOPNOTSUPP;
216 }
217
218 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
219 {
220         struct cls_cgroup_head *head = tp->root;
221
222         if (arg->count < arg->skip)
223                 goto skip;
224
225         if (arg->fn(tp, (unsigned long) head, arg) < 0) {
226                 arg->stop = 1;
227                 return;
228         }
229 skip:
230         arg->count++;
231 }
232
233 static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
234                            struct sk_buff *skb, struct tcmsg *t)
235 {
236         struct cls_cgroup_head *head = tp->root;
237         unsigned char *b = skb_tail_pointer(skb);
238         struct nlattr *nest;
239
240         t->tcm_handle = head->handle;
241
242         nest = nla_nest_start(skb, TCA_OPTIONS);
243         if (nest == NULL)
244                 goto nla_put_failure;
245
246         if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
247             tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
248                 goto nla_put_failure;
249
250         nla_nest_end(skb, nest);
251
252         if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
253                 goto nla_put_failure;
254
255         return skb->len;
256
257 nla_put_failure:
258         nlmsg_trim(skb, b);
259         return -1;
260 }
261
262 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
263         .kind           =       "cgroup",
264         .init           =       cls_cgroup_init,
265         .change         =       cls_cgroup_change,
266         .classify       =       cls_cgroup_classify,
267         .destroy        =       cls_cgroup_destroy,
268         .get            =       cls_cgroup_get,
269         .put            =       cls_cgroup_put,
270         .delete         =       cls_cgroup_delete,
271         .walk           =       cls_cgroup_walk,
272         .dump           =       cls_cgroup_dump,
273         .owner          =       THIS_MODULE,
274 };
275
276 static int __init init_cgroup_cls(void)
277 {
278         return register_tcf_proto_ops(&cls_cgroup_ops);
279 }
280
281 static void __exit exit_cgroup_cls(void)
282 {
283         unregister_tcf_proto_ops(&cls_cgroup_ops);
284 }
285
286 module_init(init_cgroup_cls);
287 module_exit(exit_cgroup_cls);
288 MODULE_LICENSE("GPL");