2 * net/sched/cls_cgroup.c Control Group Classifier
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.
9 * Authors: Thomas Graf <tgraf@suug.ch>
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>
23 #include <net/cls_cgroup.h>
25 static inline struct cgroup_cls_state *cgrp_cls_state(struct cgroup *cgrp)
27 return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
28 struct cgroup_cls_state, css);
31 static inline struct cgroup_cls_state *task_cls_state(struct task_struct *p)
33 return container_of(task_subsys_state(p, net_cls_subsys_id),
34 struct cgroup_cls_state, css);
37 static struct cgroup_subsys_state *cgrp_create(struct cgroup *cgrp)
39 struct cgroup_cls_state *cs;
41 cs = kzalloc(sizeof(*cs), GFP_KERNEL);
43 return ERR_PTR(-ENOMEM);
46 cs->classid = cgrp_cls_state(cgrp->parent)->classid;
51 static void cgrp_destroy(struct cgroup *cgrp)
53 kfree(cgrp_cls_state(cgrp));
56 static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
58 return cgrp_cls_state(cgrp)->classid;
61 static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
63 cgrp_cls_state(cgrp)->classid = (u32) value;
67 static struct cftype ss_files[] = {
70 .read_u64 = read_classid,
71 .write_u64 = write_classid,
76 struct cgroup_subsys net_cls_subsys = {
78 .create = cgrp_create,
79 .destroy = cgrp_destroy,
80 .subsys_id = net_cls_subsys_id,
81 .base_cftypes = ss_files,
82 .module = THIS_MODULE,
85 * While net_cls cgroup has the rudimentary hierarchy support of
86 * inheriting the parent's classid on cgroup creation, it doesn't
87 * properly propagates config changes in ancestors to their
88 * descendents. A child should follow the parent's configuration
89 * but be allowed to override it. Fix it and remove the following.
91 .broken_hierarchy = true,
94 struct cls_cgroup_head {
97 struct tcf_ematch_tree ematches;
100 static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
101 struct tcf_result *res)
103 struct cls_cgroup_head *head = tp->root;
107 classid = task_cls_state(current)->classid;
111 * Due to the nature of the classifier it is required to ignore all
112 * packets originating from softirq context as accessing `current'
113 * would lead to false results.
115 * This test assumes that all callers of dev_queue_xmit() explicitely
116 * disable bh. Knowing this, it is possible to detect softirq based
117 * calls by looking at the number of nested bh disable calls because
118 * softirqs always disables bh.
120 if (in_serving_softirq()) {
121 /* If there is an sk_classid we'll use that. */
124 classid = skb->sk->sk_classid;
130 if (!tcf_em_tree_match(skb, &head->ematches, NULL))
133 res->classid = classid;
135 return tcf_exts_exec(skb, &head->exts, res);
138 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
143 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
147 static int cls_cgroup_init(struct tcf_proto *tp)
152 static const struct tcf_ext_map cgroup_ext_map = {
153 .action = TCA_CGROUP_ACT,
154 .police = TCA_CGROUP_POLICE,
157 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
158 [TCA_CGROUP_EMATCHES] = { .type = NLA_NESTED },
161 static int cls_cgroup_change(struct sk_buff *in_skb,
162 struct tcf_proto *tp, unsigned long base,
163 u32 handle, struct nlattr **tca,
166 struct nlattr *tb[TCA_CGROUP_MAX + 1];
167 struct cls_cgroup_head *head = tp->root;
168 struct tcf_ematch_tree t;
172 if (!tca[TCA_OPTIONS])
179 head = kzalloc(sizeof(*head), GFP_KERNEL);
183 head->handle = handle;
190 if (handle != head->handle)
193 err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
198 err = tcf_exts_validate(tp, tb, tca[TCA_RATE], &e, &cgroup_ext_map);
202 err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
206 tcf_exts_change(tp, &head->exts, &e);
207 tcf_em_tree_change(tp, &head->ematches, &t);
212 static void cls_cgroup_destroy(struct tcf_proto *tp)
214 struct cls_cgroup_head *head = tp->root;
217 tcf_exts_destroy(tp, &head->exts);
218 tcf_em_tree_destroy(tp, &head->ematches);
223 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
228 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
230 struct cls_cgroup_head *head = tp->root;
232 if (arg->count < arg->skip)
235 if (arg->fn(tp, (unsigned long) head, arg) < 0) {
243 static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
244 struct sk_buff *skb, struct tcmsg *t)
246 struct cls_cgroup_head *head = tp->root;
247 unsigned char *b = skb_tail_pointer(skb);
250 t->tcm_handle = head->handle;
252 nest = nla_nest_start(skb, TCA_OPTIONS);
254 goto nla_put_failure;
256 if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
257 tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
258 goto nla_put_failure;
260 nla_nest_end(skb, nest);
262 if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
263 goto nla_put_failure;
272 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
274 .init = cls_cgroup_init,
275 .change = cls_cgroup_change,
276 .classify = cls_cgroup_classify,
277 .destroy = cls_cgroup_destroy,
278 .get = cls_cgroup_get,
279 .put = cls_cgroup_put,
280 .delete = cls_cgroup_delete,
281 .walk = cls_cgroup_walk,
282 .dump = cls_cgroup_dump,
283 .owner = THIS_MODULE,
286 static int __init init_cgroup_cls(void)
290 ret = cgroup_load_subsys(&net_cls_subsys);
294 ret = register_tcf_proto_ops(&cls_cgroup_ops);
296 cgroup_unload_subsys(&net_cls_subsys);
302 static void __exit exit_cgroup_cls(void)
304 unregister_tcf_proto_ops(&cls_cgroup_ops);
306 cgroup_unload_subsys(&net_cls_subsys);
309 module_init(init_cgroup_cls);
310 module_exit(exit_cgroup_cls);
311 MODULE_LICENSE("GPL");