xref: /openbmc/linux/net/sched/cls_cgroup.c (revision b34e08d5)
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/skbuff.h>
15 #include <linux/rcupdate.h>
16 #include <net/rtnetlink.h>
17 #include <net/pkt_cls.h>
18 #include <net/sock.h>
19 #include <net/cls_cgroup.h>
20 
21 struct cls_cgroup_head {
22 	u32			handle;
23 	struct tcf_exts		exts;
24 	struct tcf_ematch_tree	ematches;
25 };
26 
27 static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
28 			       struct tcf_result *res)
29 {
30 	struct cls_cgroup_head *head = tp->root;
31 	u32 classid;
32 
33 	rcu_read_lock();
34 	classid = task_cls_state(current)->classid;
35 	rcu_read_unlock();
36 
37 	/*
38 	 * Due to the nature of the classifier it is required to ignore all
39 	 * packets originating from softirq context as accessing `current'
40 	 * would lead to false results.
41 	 *
42 	 * This test assumes that all callers of dev_queue_xmit() explicitely
43 	 * disable bh. Knowing this, it is possible to detect softirq based
44 	 * calls by looking at the number of nested bh disable calls because
45 	 * softirqs always disables bh.
46 	 */
47 	if (in_serving_softirq()) {
48 		/* If there is an sk_classid we'll use that. */
49 		if (!skb->sk)
50 			return -1;
51 		classid = skb->sk->sk_classid;
52 	}
53 
54 	if (!classid)
55 		return -1;
56 
57 	if (!tcf_em_tree_match(skb, &head->ematches, NULL))
58 		return -1;
59 
60 	res->classid = classid;
61 	res->class = 0;
62 	return tcf_exts_exec(skb, &head->exts, res);
63 }
64 
65 static unsigned long cls_cgroup_get(struct tcf_proto *tp, u32 handle)
66 {
67 	return 0UL;
68 }
69 
70 static void cls_cgroup_put(struct tcf_proto *tp, unsigned long f)
71 {
72 }
73 
74 static int cls_cgroup_init(struct tcf_proto *tp)
75 {
76 	return 0;
77 }
78 
79 static const struct nla_policy cgroup_policy[TCA_CGROUP_MAX + 1] = {
80 	[TCA_CGROUP_EMATCHES]	= { .type = NLA_NESTED },
81 };
82 
83 static int cls_cgroup_change(struct net *net, struct sk_buff *in_skb,
84 			     struct tcf_proto *tp, unsigned long base,
85 			     u32 handle, struct nlattr **tca,
86 			     unsigned long *arg)
87 {
88 	struct nlattr *tb[TCA_CGROUP_MAX + 1];
89 	struct cls_cgroup_head *head = tp->root;
90 	struct tcf_ematch_tree t;
91 	struct tcf_exts e;
92 	int err;
93 
94 	if (!tca[TCA_OPTIONS])
95 		return -EINVAL;
96 
97 	if (head == NULL) {
98 		if (!handle)
99 			return -EINVAL;
100 
101 		head = kzalloc(sizeof(*head), GFP_KERNEL);
102 		if (head == NULL)
103 			return -ENOBUFS;
104 
105 		tcf_exts_init(&head->exts, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
106 		head->handle = handle;
107 
108 		tcf_tree_lock(tp);
109 		tp->root = head;
110 		tcf_tree_unlock(tp);
111 	}
112 
113 	if (handle != head->handle)
114 		return -ENOENT;
115 
116 	err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
117 			       cgroup_policy);
118 	if (err < 0)
119 		return err;
120 
121 	tcf_exts_init(&e, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
122 	err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e);
123 	if (err < 0)
124 		return err;
125 
126 	err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
127 	if (err < 0)
128 		return err;
129 
130 	tcf_exts_change(tp, &head->exts, &e);
131 	tcf_em_tree_change(tp, &head->ematches, &t);
132 
133 	return 0;
134 }
135 
136 static void cls_cgroup_destroy(struct tcf_proto *tp)
137 {
138 	struct cls_cgroup_head *head = tp->root;
139 
140 	if (head) {
141 		tcf_exts_destroy(tp, &head->exts);
142 		tcf_em_tree_destroy(tp, &head->ematches);
143 		kfree(head);
144 	}
145 }
146 
147 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
148 {
149 	return -EOPNOTSUPP;
150 }
151 
152 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
153 {
154 	struct cls_cgroup_head *head = tp->root;
155 
156 	if (arg->count < arg->skip)
157 		goto skip;
158 
159 	if (arg->fn(tp, (unsigned long) head, arg) < 0) {
160 		arg->stop = 1;
161 		return;
162 	}
163 skip:
164 	arg->count++;
165 }
166 
167 static int cls_cgroup_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
168 			   struct sk_buff *skb, struct tcmsg *t)
169 {
170 	struct cls_cgroup_head *head = tp->root;
171 	unsigned char *b = skb_tail_pointer(skb);
172 	struct nlattr *nest;
173 
174 	t->tcm_handle = head->handle;
175 
176 	nest = nla_nest_start(skb, TCA_OPTIONS);
177 	if (nest == NULL)
178 		goto nla_put_failure;
179 
180 	if (tcf_exts_dump(skb, &head->exts) < 0 ||
181 	    tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
182 		goto nla_put_failure;
183 
184 	nla_nest_end(skb, nest);
185 
186 	if (tcf_exts_dump_stats(skb, &head->exts) < 0)
187 		goto nla_put_failure;
188 
189 	return skb->len;
190 
191 nla_put_failure:
192 	nlmsg_trim(skb, b);
193 	return -1;
194 }
195 
196 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
197 	.kind		=	"cgroup",
198 	.init		=	cls_cgroup_init,
199 	.change		=	cls_cgroup_change,
200 	.classify	=	cls_cgroup_classify,
201 	.destroy	=	cls_cgroup_destroy,
202 	.get		=	cls_cgroup_get,
203 	.put		=	cls_cgroup_put,
204 	.delete		=	cls_cgroup_delete,
205 	.walk		=	cls_cgroup_walk,
206 	.dump		=	cls_cgroup_dump,
207 	.owner		=	THIS_MODULE,
208 };
209 
210 static int __init init_cgroup_cls(void)
211 {
212 	return register_tcf_proto_ops(&cls_cgroup_ops);
213 }
214 
215 static void __exit exit_cgroup_cls(void)
216 {
217 	unregister_tcf_proto_ops(&cls_cgroup_ops);
218 }
219 
220 module_init(init_cgroup_cls);
221 module_exit(exit_cgroup_cls);
222 MODULE_LICENSE("GPL");
223