Merge tag 'cleanup' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
[linux-2.6-microblaze.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/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
25 static 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
31 static 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
37 static 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         return &cs->css;
45 }
46
47 static int cgrp_css_online(struct cgroup *cgrp)
48 {
49         if (cgrp->parent)
50                 cgrp_cls_state(cgrp)->classid =
51                         cgrp_cls_state(cgrp->parent)->classid;
52         return 0;
53 }
54
55 static void cgrp_css_free(struct cgroup *cgrp)
56 {
57         kfree(cgrp_cls_state(cgrp));
58 }
59
60 static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
61 {
62         return cgrp_cls_state(cgrp)->classid;
63 }
64
65 static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
66 {
67         cgrp_cls_state(cgrp)->classid = (u32) value;
68         return 0;
69 }
70
71 static struct cftype ss_files[] = {
72         {
73                 .name = "classid",
74                 .read_u64 = read_classid,
75                 .write_u64 = write_classid,
76         },
77         { }     /* terminate */
78 };
79
80 struct cgroup_subsys net_cls_subsys = {
81         .name           = "net_cls",
82         .css_alloc      = cgrp_css_alloc,
83         .css_online     = cgrp_css_online,
84         .css_free       = cgrp_css_free,
85         .subsys_id      = net_cls_subsys_id,
86         .base_cftypes   = ss_files,
87         .module         = THIS_MODULE,
88 };
89
90 struct cls_cgroup_head {
91         u32                     handle;
92         struct tcf_exts         exts;
93         struct tcf_ematch_tree  ematches;
94 };
95
96 static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
97                                struct tcf_result *res)
98 {
99         struct cls_cgroup_head *head = tp->root;
100         u32 classid;
101
102         rcu_read_lock();
103         classid = task_cls_state(current)->classid;
104         rcu_read_unlock();
105
106         /*
107          * Due to the nature of the classifier it is required to ignore all
108          * packets originating from softirq context as accessing `current'
109          * would lead to false results.
110          *
111          * This test assumes that all callers of dev_queue_xmit() explicitely
112          * disable bh. Knowing this, it is possible to detect softirq based
113          * calls by looking at the number of nested bh disable calls because
114          * softirqs always disables bh.
115          */
116         if (in_serving_softirq()) {
117                 /* If there is an sk_classid we'll use that. */
118                 if (!skb->sk)
119                         return -1;
120                 classid = skb->sk->sk_classid;
121         }
122
123         if (!classid)
124                 return -1;
125
126         if (!tcf_em_tree_match(skb, &head->ematches, NULL))
127                 return -1;
128
129         res->classid = classid;
130         res->class = 0;
131         return tcf_exts_exec(skb, &head->exts, res);
132 }
133
134 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
135 {
136         return 0UL;
137 }
138
139 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
140 {
141 }
142
143 static int cls_cgroup_init(struct tcf_proto *tp)
144 {
145         return 0;
146 }
147
148 static const struct tcf_ext_map cgroup_ext_map = {
149         .action = TCA_CGROUP_ACT,
150         .police = TCA_CGROUP_POLICE,
151 };
152
153 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
154         [TCA_CGROUP_EMATCHES]   = { .type = NLA_NESTED },
155 };
156
157 static int cls_cgroup_change(struct sk_buff *in_skb,
158                              struct tcf_proto *tp, unsigned long base,
159                              u32 handle, struct nlattr **tca,
160                              unsigned long *arg)
161 {
162         struct nlattr *tb[TCA_CGROUP_MAX + 1];
163         struct cls_cgroup_head *head = tp->root;
164         struct tcf_ematch_tree t;
165         struct tcf_exts e;
166         int err;
167
168         if (!tca[TCA_OPTIONS])
169                 return -EINVAL;
170
171         if (head == NULL) {
172                 if (!handle)
173                         return -EINVAL;
174
175                 head = kzalloc(sizeof(*head), GFP_KERNEL);
176                 if (head == NULL)
177                         return -ENOBUFS;
178
179                 head->handle = handle;
180
181                 tcf_tree_lock(tp);
182                 tp->root = head;
183                 tcf_tree_unlock(tp);
184         }
185
186         if (handle != head->handle)
187                 return -ENOENT;
188
189         err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
190                                cgroup_policy);
191         if (err < 0)
192                 return err;
193
194         err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
195         if (err < 0)
196                 return err;
197
198         err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
199         if (err < 0)
200                 return err;
201
202         tcf_exts_change(tp, &head->exts, &e);
203         tcf_em_tree_change(tp, &head->ematches, &t);
204
205         return 0;
206 }
207
208 static void cls_cgroup_destroy(struct tcf_proto *tp)
209 {
210         struct cls_cgroup_head *head = tp->root;
211
212         if (head) {
213                 tcf_exts_destroy(tp, &head->exts);
214                 tcf_em_tree_destroy(tp, &head->ematches);
215                 kfree(head);
216         }
217 }
218
219 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
220 {
221         return -EOPNOTSUPP;
222 }
223
224 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
225 {
226         struct cls_cgroup_head *head = tp->root;
227
228         if (arg->count < arg->skip)
229                 goto skip;
230
231         if (arg->fn(tp, (unsigned long) head, arg) < 0) {
232                 arg->stop = 1;
233                 return;
234         }
235 skip:
236         arg->count++;
237 }
238
239 static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
240                            struct sk_buff *skb, struct tcmsg *t)
241 {
242         struct cls_cgroup_head *head = tp->root;
243         unsigned char *b = skb_tail_pointer(skb);
244         struct nlattr *nest;
245
246         t->tcm_handle = head->handle;
247
248         nest = nla_nest_start(skb, TCA_OPTIONS);
249         if (nest == NULL)
250                 goto nla_put_failure;
251
252         if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
253             tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
254                 goto nla_put_failure;
255
256         nla_nest_end(skb, nest);
257
258         if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
259                 goto nla_put_failure;
260
261         return skb->len;
262
263 nla_put_failure:
264         nlmsg_trim(skb, b);
265         return -1;
266 }
267
268 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
269         .kind           =       "cgroup",
270         .init           =       cls_cgroup_init,
271         .change         =       cls_cgroup_change,
272         .classify       =       cls_cgroup_classify,
273         .destroy        =       cls_cgroup_destroy,
274         .get            =       cls_cgroup_get,
275         .put            =       cls_cgroup_put,
276         .delete         =       cls_cgroup_delete,
277         .walk           =       cls_cgroup_walk,
278         .dump           =       cls_cgroup_dump,
279         .owner          =       THIS_MODULE,
280 };
281
282 static int __init init_cgroup_cls(void)
283 {
284         int ret;
285
286         ret = cgroup_load_subsys(&net_cls_subsys);
287         if (ret)
288                 goto out;
289
290         ret = register_tcf_proto_ops(&cls_cgroup_ops);
291         if (ret)
292                 cgroup_unload_subsys(&net_cls_subsys);
293
294 out:
295         return ret;
296 }
297
298 static void __exit exit_cgroup_cls(void)
299 {
300         unregister_tcf_proto_ops(&cls_cgroup_ops);
301
302         cgroup_unload_subsys(&net_cls_subsys);
303 }
304
305 module_init(init_cgroup_cls);
306 module_exit(exit_cgroup_cls);
307 MODULE_LICENSE("GPL");