net_sched: cls: refactor out struct tcf_ext_map
[pandora-kernel.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 <linux/fdtable.h>
21 #include <net/rtnetlink.h>
22 #include <net/pkt_cls.h>
23 #include <net/sock.h>
24 #include <net/cls_cgroup.h>
25
26 static inline struct cgroup_cls_state *css_cls_state(struct cgroup_subsys_state *css)
27 {
28         return css ? container_of(css, struct cgroup_cls_state, css) : NULL;
29 }
30
31 static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
32 {
33         return css_cls_state(task_css(p, net_cls_subsys_id));
34 }
35
36 static struct cgroup_subsys_state *
37 cgrp_css_alloc(struct cgroup_subsys_state *parent_css)
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_subsys_state *css)
48 {
49         struct cgroup_cls_state *cs = css_cls_state(css);
50         struct cgroup_cls_state *parent = css_cls_state(css_parent(css));
51
52         if (parent)
53                 cs->classid = parent->classid;
54         return 0;
55 }
56
57 static void cgrp_css_free(struct cgroup_subsys_state *css)
58 {
59         kfree(css_cls_state(css));
60 }
61
62 static int update_classid(const void *v, struct file *file, unsigned n)
63 {
64         int err;
65         struct socket *sock = sock_from_file(file, &err);
66         if (sock)
67                 sock->sk->sk_classid = (u32)(unsigned long)v;
68         return 0;
69 }
70
71 static void cgrp_attach(struct cgroup_subsys_state *css,
72                         struct cgroup_taskset *tset)
73 {
74         struct task_struct *p;
75         struct cgroup_cls_state *cs = css_cls_state(css);
76         void *v = (void *)(unsigned long)cs->classid;
77
78         cgroup_taskset_for_each(p, css, tset) {
79                 task_lock(p);
80                 iterate_fd(p->files, 0, update_classid, v);
81                 task_unlock(p);
82         }
83 }
84
85 static u64 read_classid(struct cgroup_subsys_state *css, struct cftype *cft)
86 {
87         return css_cls_state(css)->classid;
88 }
89
90 static int write_classid(struct cgroup_subsys_state *css, struct cftype *cft,
91                          u64 value)
92 {
93         css_cls_state(css)->classid = (u32) value;
94         return 0;
95 }
96
97 static struct cftype ss_files[] = {
98         {
99                 .name = "classid",
100                 .read_u64 = read_classid,
101                 .write_u64 = write_classid,
102         },
103         { }     /* terminate */
104 };
105
106 struct cgroup_subsys net_cls_subsys = {
107         .name           = "net_cls",
108         .css_alloc      = cgrp_css_alloc,
109         .css_online     = cgrp_css_online,
110         .css_free       = cgrp_css_free,
111         .attach         = cgrp_attach,
112         .subsys_id      = net_cls_subsys_id,
113         .base_cftypes   = ss_files,
114         .module         = THIS_MODULE,
115 };
116
117 struct cls_cgroup_head {
118         u32                     handle;
119         struct tcf_exts         exts;
120         struct tcf_ematch_tree  ematches;
121 };
122
123 static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
124                                struct tcf_result *res)
125 {
126         struct cls_cgroup_head *head = tp->root;
127         u32 classid;
128
129         rcu_read_lock();
130         classid = task_cls_state(current)->classid;
131         rcu_read_unlock();
132
133         /*
134          * Due to the nature of the classifier it is required to ignore all
135          * packets originating from softirq context as accessing `current'
136          * would lead to false results.
137          *
138          * This test assumes that all callers of dev_queue_xmit() explicitely
139          * disable bh. Knowing this, it is possible to detect softirq based
140          * calls by looking at the number of nested bh disable calls because
141          * softirqs always disables bh.
142          */
143         if (in_serving_softirq()) {
144                 /* If there is an sk_classid we'll use that. */
145                 if (!skb->sk)
146                         return -1;
147                 classid = skb->sk->sk_classid;
148         }
149
150         if (!classid)
151                 return -1;
152
153         if (!tcf_em_tree_match(skb, &head->ematches, NULL))
154                 return -1;
155
156         res->classid = classid;
157         res->class = 0;
158         return tcf_exts_exec(skb, &head->exts, res);
159 }
160
161 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
162 {
163         return 0UL;
164 }
165
166 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
167 {
168 }
169
170 static int cls_cgroup_init(struct tcf_proto *tp)
171 {
172         return 0;
173 }
174
175 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
176         [TCA_CGROUP_EMATCHES]   = { .type = NLA_NESTED },
177 };
178
179 static int cls_cgroup_change(struct net *net, struct sk_buff *in_skb,
180                              struct tcf_proto *tp, unsigned long base,
181                              u32 handle, struct nlattr **tca,
182                              unsigned long *arg)
183 {
184         struct nlattr *tb[TCA_CGROUP_MAX + 1];
185         struct cls_cgroup_head *head = tp->root;
186         struct tcf_ematch_tree t;
187         struct tcf_exts e;
188         int err;
189
190         if (!tca[TCA_OPTIONS])
191                 return -EINVAL;
192
193         if (head == NULL) {
194                 if (!handle)
195                         return -EINVAL;
196
197                 head = kzalloc(sizeof(*head), GFP_KERNEL);
198                 if (head == NULL)
199                         return -ENOBUFS;
200
201                 tcf_exts_init(&head->exts, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
202                 head->handle = handle;
203
204                 tcf_tree_lock(tp);
205                 tp->root = head;
206                 tcf_tree_unlock(tp);
207         }
208
209         if (handle != head->handle)
210                 return -ENOENT;
211
212         err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
213                                cgroup_policy);
214         if (err < 0)
215                 return err;
216
217         tcf_exts_init(&e, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
218         err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e);
219         if (err < 0)
220                 return err;
221
222         err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
223         if (err < 0)
224                 return err;
225
226         tcf_exts_change(tp, &head->exts, &e);
227         tcf_em_tree_change(tp, &head->ematches, &t);
228
229         return 0;
230 }
231
232 static void cls_cgroup_destroy(struct tcf_proto *tp)
233 {
234         struct cls_cgroup_head *head = tp->root;
235
236         if (head) {
237                 tcf_exts_destroy(tp, &head->exts);
238                 tcf_em_tree_destroy(tp, &head->ematches);
239                 kfree(head);
240         }
241 }
242
243 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
244 {
245         return -EOPNOTSUPP;
246 }
247
248 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
249 {
250         struct cls_cgroup_head *head = tp->root;
251
252         if (arg->count < arg->skip)
253                 goto skip;
254
255         if (arg->fn(tp, (unsigned long) head, arg) < 0) {
256                 arg->stop = 1;
257                 return;
258         }
259 skip:
260         arg->count++;
261 }
262
263 static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
264                            struct sk_buff *skb, struct tcmsg *t)
265 {
266         struct cls_cgroup_head *head = tp->root;
267         unsigned char *b = skb_tail_pointer(skb);
268         struct nlattr *nest;
269
270         t->tcm_handle = head->handle;
271
272         nest = nla_nest_start(skb, TCA_OPTIONS);
273         if (nest == NULL)
274                 goto nla_put_failure;
275
276         if (tcf_exts_dump(skb, &head->exts) < 0 ||
277             tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
278                 goto nla_put_failure;
279
280         nla_nest_end(skb, nest);
281
282         if (tcf_exts_dump_stats(skb, &head->exts) < 0)
283                 goto nla_put_failure;
284
285         return skb->len;
286
287 nla_put_failure:
288         nlmsg_trim(skb, b);
289         return -1;
290 }
291
292 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
293         .kind           =       "cgroup",
294         .init           =       cls_cgroup_init,
295         .change         =       cls_cgroup_change,
296         .classify       =       cls_cgroup_classify,
297         .destroy        =       cls_cgroup_destroy,
298         .get            =       cls_cgroup_get,
299         .put            =       cls_cgroup_put,
300         .delete         =       cls_cgroup_delete,
301         .walk           =       cls_cgroup_walk,
302         .dump           =       cls_cgroup_dump,
303         .owner          =       THIS_MODULE,
304 };
305
306 static int __init init_cgroup_cls(void)
307 {
308         int ret;
309
310         ret = cgroup_load_subsys(&net_cls_subsys);
311         if (ret)
312                 goto out;
313
314         ret = register_tcf_proto_ops(&cls_cgroup_ops);
315         if (ret)
316                 cgroup_unload_subsys(&net_cls_subsys);
317
318 out:
319         return ret;
320 }
321
322 static void __exit exit_cgroup_cls(void)
323 {
324         unregister_tcf_proto_ops(&cls_cgroup_ops);
325
326         cgroup_unload_subsys(&net_cls_subsys);
327 }
328
329 module_init(init_cgroup_cls);
330 module_exit(exit_cgroup_cls);
331 MODULE_LICENSE("GPL");