xref: /linux/net/sched/cls_cgroup.c (revision f884ab15afdc5514e88105c92a4e2e1e6539869a)
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 *cgrp_cls_state(struct cgroup *cgrp)
27 {
28 	return container_of(cgroup_subsys_state(cgrp, net_cls_subsys_id),
29 			    struct cgroup_cls_state, css);
30 }
31 
32 static 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);
36 }
37 
38 static struct cgroup_subsys_state *cgrp_css_alloc(struct cgroup *cgrp)
39 {
40 	struct cgroup_cls_state *cs;
41 
42 	cs = kzalloc(sizeof(*cs), GFP_KERNEL);
43 	if (!cs)
44 		return ERR_PTR(-ENOMEM);
45 	return &cs->css;
46 }
47 
48 static 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 
56 static void cgrp_css_free(struct cgroup *cgrp)
57 {
58 	kfree(cgrp_cls_state(cgrp));
59 }
60 
61 static 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 
70 static 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 
83 static u64 read_classid(struct cgroup *cgrp, struct cftype *cft)
84 {
85 	return cgrp_cls_state(cgrp)->classid;
86 }
87 
88 static int write_classid(struct cgroup *cgrp, struct cftype *cft, u64 value)
89 {
90 	cgrp_cls_state(cgrp)->classid = (u32) value;
91 	return 0;
92 }
93 
94 static struct cftype ss_files[] = {
95 	{
96 		.name = "classid",
97 		.read_u64 = read_classid,
98 		.write_u64 = write_classid,
99 	},
100 	{ }	/* terminate */
101 };
102 
103 struct cgroup_subsys net_cls_subsys = {
104 	.name		= "net_cls",
105 	.css_alloc	= cgrp_css_alloc,
106 	.css_online	= cgrp_css_online,
107 	.css_free	= cgrp_css_free,
108 	.attach		= cgrp_attach,
109 	.subsys_id	= net_cls_subsys_id,
110 	.base_cftypes	= ss_files,
111 	.module		= THIS_MODULE,
112 };
113 
114 struct cls_cgroup_head {
115 	u32			handle;
116 	struct tcf_exts		exts;
117 	struct tcf_ematch_tree	ematches;
118 };
119 
120 static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
121 			       struct tcf_result *res)
122 {
123 	struct cls_cgroup_head *head = tp->root;
124 	u32 classid;
125 
126 	rcu_read_lock();
127 	classid = task_cls_state(current)->classid;
128 	rcu_read_unlock();
129 
130 	/*
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 	 */
140 	if (in_serving_softirq()) {
141 		/* If there is an sk_classid we'll use that. */
142 		if (!skb->sk)
143 			return -1;
144 		classid = skb->sk->sk_classid;
145 	}
146 
147 	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);
156 }
157 
158 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
159 {
160 	return 0UL;
161 }
162 
163 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
164 {
165 }
166 
167 static int cls_cgroup_init(struct tcf_proto *tp)
168 {
169 	return 0;
170 }
171 
172 static const struct tcf_ext_map cgroup_ext_map = {
173 	.action = TCA_CGROUP_ACT,
174 	.police = TCA_CGROUP_POLICE,
175 };
176 
177 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
178 	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
179 };
180 
181 static int cls_cgroup_change(struct net *net, struct sk_buff *in_skb,
182 			     struct tcf_proto *tp, unsigned long base,
183 			     u32 handle, struct nlattr **tca,
184 			     unsigned long *arg)
185 {
186 	struct nlattr *tb[TCA_CGROUP_MAX + 1];
187 	struct cls_cgroup_head *head = tp->root;
188 	struct tcf_ematch_tree t;
189 	struct tcf_exts e;
190 	int err;
191 
192 	if (!tca[TCA_OPTIONS])
193 		return -EINVAL;
194 
195 	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(net, tp, tb, tca[TCA_RATE], &e,
219 				&cgroup_ext_map);
220 	if (err < 0)
221 		return err;
222 
223 	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
224 	if (err < 0)
225 		return err;
226 
227 	tcf_exts_change(tp, &head->exts, &e);
228 	tcf_em_tree_change(tp, &head->ematches, &t);
229 
230 	return 0;
231 }
232 
233 static void cls_cgroup_destroy(struct tcf_proto *tp)
234 {
235 	struct cls_cgroup_head *head = tp->root;
236 
237 	if (head) {
238 		tcf_exts_destroy(tp, &head->exts);
239 		tcf_em_tree_destroy(tp, &head->ematches);
240 		kfree(head);
241 	}
242 }
243 
244 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
245 {
246 	return -EOPNOTSUPP;
247 }
248 
249 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
250 {
251 	struct cls_cgroup_head *head = tp->root;
252 
253 	if (arg->count < arg->skip)
254 		goto skip;
255 
256 	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
257 		arg->stop = 1;
258 		return;
259 	}
260 skip:
261 	arg->count++;
262 }
263 
264 static int cls_cgroup_dump(struct tcf_proto *tp, unsigned long fh,
265 			   struct sk_buff *skb, struct tcmsg *t)
266 {
267 	struct cls_cgroup_head *head = tp->root;
268 	unsigned char *b = skb_tail_pointer(skb);
269 	struct nlattr *nest;
270 
271 	t->tcm_handle = head->handle;
272 
273 	nest = nla_nest_start(skb, TCA_OPTIONS);
274 	if (nest == NULL)
275 		goto nla_put_failure;
276 
277 	if (tcf_exts_dump(skb, &head->exts, &cgroup_ext_map) < 0 ||
278 	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
279 		goto nla_put_failure;
280 
281 	nla_nest_end(skb, nest);
282 
283 	if (tcf_exts_dump_stats(skb, &head->exts, &cgroup_ext_map) < 0)
284 		goto nla_put_failure;
285 
286 	return skb->len;
287 
288 nla_put_failure:
289 	nlmsg_trim(skb, b);
290 	return -1;
291 }
292 
293 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
294 	.kind		=	"cgroup",
295 	.init		=	cls_cgroup_init,
296 	.change		=	cls_cgroup_change,
297 	.classify	=	cls_cgroup_classify,
298 	.destroy	=	cls_cgroup_destroy,
299 	.get		=	cls_cgroup_get,
300 	.put		=	cls_cgroup_put,
301 	.delete		=	cls_cgroup_delete,
302 	.walk		=	cls_cgroup_walk,
303 	.dump		=	cls_cgroup_dump,
304 	.owner		=	THIS_MODULE,
305 };
306 
307 static int __init init_cgroup_cls(void)
308 {
309 	int ret;
310 
311 	ret = cgroup_load_subsys(&net_cls_subsys);
312 	if (ret)
313 		goto out;
314 
315 	ret = register_tcf_proto_ops(&cls_cgroup_ops);
316 	if (ret)
317 		cgroup_unload_subsys(&net_cls_subsys);
318 
319 out:
320 	return ret;
321 }
322 
323 static void __exit exit_cgroup_cls(void)
324 {
325 	unregister_tcf_proto_ops(&cls_cgroup_ops);
326 
327 	cgroup_unload_subsys(&net_cls_subsys);
328 }
329 
330 module_init(init_cgroup_cls);
331 module_exit(exit_cgroup_cls);
332 MODULE_LICENSE("GPL");
333