blob: 6db7855b9029ee26dcd97d73c43862a93c35b6c4 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090013#include <linux/slab.h>
Thomas Graff4009232008-11-07 22:56:00 -080014#include <linux/types.h>
15#include <linux/string.h>
16#include <linux/errno.h>
17#include <linux/skbuff.h>
18#include <linux/cgroup.h>
Herbert Xuf8451722010-05-24 00:12:34 -070019#include <linux/rcupdate.h>
Daniel Wagner6a328d82012-10-25 04:16:59 +000020#include <linux/fdtable.h>
Thomas Graff4009232008-11-07 22:56:00 -080021#include <net/rtnetlink.h>
22#include <net/pkt_cls.h>
Herbert Xuf8451722010-05-24 00:12:34 -070023#include <net/sock.h>
24#include <net/cls_cgroup.h>
Thomas Graff4009232008-11-07 22:56:00 -080025
Li Zefan8e8ba852008-12-29 19:39:03 -080026static inline struct cgroup_cls_state *cgrp_cls_state(struct cgroup *cgrp)
Thomas Graff4009232008-11-07 22:56:00 -080027{
Li Zefan8e8ba852008-12-29 19:39:03 -080028 return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
29 struct cgroup_cls_state, css);
30}
31
32static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
33{
34 return container_of(task_subsys_state(p, net_cls_subsys_id),
35 struct cgroup_cls_state, css);
Thomas Graff4009232008-11-07 22:56:00 -080036}
37
Tejun Heo92fb9742012-11-19 08:13:38 -080038static struct cgroup_subsys_state *cgrp_css_alloc(struct cgroup *cgrp)
Thomas Graff4009232008-11-07 22:56:00 -080039{
40 struct cgroup_cls_state *cs;
41
Eric Dumazetcc7ec452011-01-19 19:26:56 +000042 cs = kzalloc(sizeof(*cs), GFP_KERNEL);
43 if (!cs)
Thomas Graff4009232008-11-07 22:56:00 -080044 return ERR_PTR(-ENOMEM);
Thomas Graff4009232008-11-07 22:56:00 -080045 return &cs->css;
46}
47
Tejun Heo0ba18f72012-11-22 07:32:46 -080048static int cgrp_css_online(struct cgroup *cgrp)
49{
50 if (cgrp->parent)
51 cgrp_cls_state(cgrp)->classid =
52 cgrp_cls_state(cgrp->parent)->classid;
53 return 0;
54}
55
Tejun Heo92fb9742012-11-19 08:13:38 -080056static void cgrp_css_free(struct cgroup *cgrp)
Thomas Graff4009232008-11-07 22:56:00 -080057{
Li Zefan8e8ba852008-12-29 19:39:03 -080058 kfree(cgrp_cls_state(cgrp));
Thomas Graff4009232008-11-07 22:56:00 -080059}
60
Daniel Wagner6a328d82012-10-25 04:16:59 +000061static int update_classid(const void *v, struct file *file, unsigned n)
62{
63 int err;
64 struct socket *sock = sock_from_file(file, &err);
65 if (sock)
66 sock->sk->sk_classid = (u32)(unsigned long)v;
67 return 0;
68}
69
70static void cgrp_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
71{
72 struct task_struct *p;
73 void *v;
74
75 cgroup_taskset_for_each(p, cgrp, tset) {
76 task_lock(p);
77 v = (void *)(unsigned long)task_cls_classid(p);
78 iterate_fd(p->files, 0, update_classid, v);
79 task_unlock(p);
80 }
81}
82
Thomas Graff4009232008-11-07 22:56:00 -080083static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
84{
Li Zefan8e8ba852008-12-29 19:39:03 -080085 return cgrp_cls_state(cgrp)->classid;
Thomas Graff4009232008-11-07 22:56:00 -080086}
87
88static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
89{
Li Zefan8e8ba852008-12-29 19:39:03 -080090 cgrp_cls_state(cgrp)->classid = (u32) value;
Thomas Graff4009232008-11-07 22:56:00 -080091 return 0;
92}
93
94static struct cftype ss_files[] = {
95 {
96 .name = "classid",
97 .read_u64 = read_classid,
98 .write_u64 = write_classid,
99 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700100 { } /* terminate */
Thomas Graff4009232008-11-07 22:56:00 -0800101};
102
Tejun Heo676f7c82012-04-01 12:09:55 -0700103struct cgroup_subsys net_cls_subsys = {
104 .name = "net_cls",
Tejun Heo92fb9742012-11-19 08:13:38 -0800105 .css_alloc = cgrp_css_alloc,
Tejun Heo0ba18f72012-11-22 07:32:46 -0800106 .css_online = cgrp_css_online,
Tejun Heo92fb9742012-11-19 08:13:38 -0800107 .css_free = cgrp_css_free,
Daniel Wagner6a328d82012-10-25 04:16:59 +0000108 .attach = cgrp_attach,
Tejun Heo676f7c82012-04-01 12:09:55 -0700109 .subsys_id = net_cls_subsys_id,
Tejun Heo4baf6e32012-04-01 12:09:55 -0700110 .base_cftypes = ss_files,
Tejun Heo676f7c82012-04-01 12:09:55 -0700111 .module = THIS_MODULE,
112};
113
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000114struct cls_cgroup_head {
Thomas Graff4009232008-11-07 22:56:00 -0800115 u32 handle;
116 struct tcf_exts exts;
117 struct tcf_ematch_tree ematches;
118};
119
Eric Dumazetdc7f9f62011-07-05 23:25:42 +0000120static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
Thomas Graff4009232008-11-07 22:56:00 -0800121 struct tcf_result *res)
122{
123 struct cls_cgroup_head *head = tp->root;
Paul Menagee65fcfd2009-05-26 20:47:02 -0700124 u32 classid;
Thomas Graff4009232008-11-07 22:56:00 -0800125
Herbert Xuf8451722010-05-24 00:12:34 -0700126 rcu_read_lock();
127 classid = task_cls_state(current)->classid;
128 rcu_read_unlock();
129
Thomas Graff4009232008-11-07 22:56:00 -0800130 /*
131 * Due to the nature of the classifier it is required to ignore all
132 * packets originating from softirq context as accessing `current'
133 * would lead to false results.
134 *
135 * This test assumes that all callers of dev_queue_xmit() explicitely
136 * disable bh. Knowing this, it is possible to detect softirq based
137 * calls by looking at the number of nested bh disable calls because
138 * softirqs always disables bh.
139 */
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700140 if (in_serving_softirq()) {
Herbert Xuf8451722010-05-24 00:12:34 -0700141 /* If there is an sk_classid we'll use that. */
142 if (!skb->sk)
143 return -1;
144 classid = skb->sk->sk_classid;
145 }
Thomas Graff4009232008-11-07 22:56:00 -0800146
Paul Menagee65fcfd2009-05-26 20:47:02 -0700147 if (!classid)
148 return -1;
149
150 if (!tcf_em_tree_match(skb, &head->ematches, NULL))
151 return -1;
152
153 res->classid = classid;
154 res->class = 0;
155 return tcf_exts_exec(skb, &head->exts, res);
Thomas Graff4009232008-11-07 22:56:00 -0800156}
157
158static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
159{
160 return 0UL;
161}
162
163static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
164{
165}
166
167static int cls_cgroup_init(struct tcf_proto *tp)
168{
169 return 0;
170}
171
172static const struct tcf_ext_map cgroup_ext_map = {
173 .action = TCA_CGROUP_ACT,
174 .police = TCA_CGROUP_POLICE,
175};
176
177static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
178 [TCA_CGROUP_EMATCHES] = { .type = NLA_NESTED },
179};
180
Eric W. Biedermanaf4c6642012-05-25 13:42:45 -0600181static int cls_cgroup_change(struct sk_buff *in_skb,
182 struct tcf_proto *tp, unsigned long base,
Thomas Graff4009232008-11-07 22:56:00 -0800183 u32 handle, struct nlattr **tca,
184 unsigned long *arg)
185{
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000186 struct nlattr *tb[TCA_CGROUP_MAX + 1];
Thomas Graff4009232008-11-07 22:56:00 -0800187 struct cls_cgroup_head *head = tp->root;
188 struct tcf_ematch_tree t;
189 struct tcf_exts e;
190 int err;
191
Minoru Usui52ea3a52009-06-09 04:03:09 -0700192 if (!tca[TCA_OPTIONS])
193 return -EINVAL;
194
Thomas Graff4009232008-11-07 22:56:00 -0800195 if (head == NULL) {
196 if (!handle)
197 return -EINVAL;
198
199 head = kzalloc(sizeof(*head), GFP_KERNEL);
200 if (head == NULL)
201 return -ENOBUFS;
202
203 head->handle = handle;
204
205 tcf_tree_lock(tp);
206 tp->root = head;
207 tcf_tree_unlock(tp);
208 }
209
210 if (handle != head->handle)
211 return -ENOENT;
212
213 err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
214 cgroup_policy);
215 if (err < 0)
216 return err;
217
218 err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
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
232static void cls_cgroup_destroy(struct tcf_proto *tp)
233{
Patrick McHardy47a1a1d2008-11-19 08:03:09 +0000234 struct cls_cgroup_head *head = tp->root;
Thomas Graff4009232008-11-07 22:56:00 -0800235
236 if (head) {
237 tcf_exts_destroy(tp, &head->exts);
238 tcf_em_tree_destroy(tp, &head->ematches);
239 kfree(head);
240 }
241}
242
243static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
244{
245 return -EOPNOTSUPP;
246}
247
248static 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 }
259skip:
260 arg->count++;
261}
262
263static 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, &cgroup_ext_map) < 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, &cgroup_ext_map) < 0)
283 goto nla_put_failure;
284
285 return skb->len;
286
287nla_put_failure:
288 nlmsg_trim(skb, b);
289 return -1;
290}
291
292static 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
306static int __init init_cgroup_cls(void)
307{
Herbert Xuf8451722010-05-24 00:12:34 -0700308 int ret;
309
Ben Blum8e039d82010-03-23 05:24:03 +0000310 ret = cgroup_load_subsys(&net_cls_subsys);
311 if (ret)
Herbert Xuf8451722010-05-24 00:12:34 -0700312 goto out;
313
Herbert Xuf8451722010-05-24 00:12:34 -0700314 ret = register_tcf_proto_ops(&cls_cgroup_ops);
315 if (ret)
316 cgroup_unload_subsys(&net_cls_subsys);
317
318out:
Ben Blum8e039d82010-03-23 05:24:03 +0000319 return ret;
Thomas Graff4009232008-11-07 22:56:00 -0800320}
321
322static void __exit exit_cgroup_cls(void)
323{
324 unregister_tcf_proto_ops(&cls_cgroup_ops);
Herbert Xuf8451722010-05-24 00:12:34 -0700325
Ben Blum8e039d82010-03-23 05:24:03 +0000326 cgroup_unload_subsys(&net_cls_subsys);
Thomas Graff4009232008-11-07 22:56:00 -0800327}
328
329module_init(init_cgroup_cls);
330module_exit(exit_cgroup_cls);
331MODULE_LICENSE("GPL");