net: sched: cls_cgroup use RCU
[firefly-linux-kernel-4.4.55.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/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         struct tcf_proto        *tp;
26         struct rcu_head         rcu;
27 };
28
29 static int cls_cgroup_classify(struct sk_buff *skb, const struct tcf_proto *tp,
30                                struct tcf_result *res)
31 {
32         struct cls_cgroup_head *head = rcu_dereference_bh(tp->root);
33         u32 classid;
34
35         classid = task_cls_state(current)->classid;
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 void cls_cgroup_destroy_rcu(struct rcu_head *root)
84 {
85         struct cls_cgroup_head *head = container_of(root,
86                                                     struct cls_cgroup_head,
87                                                     rcu);
88
89         tcf_exts_destroy(head->tp, &head->exts);
90         tcf_em_tree_destroy(head->tp, &head->ematches);
91         kfree(head);
92 }
93
94 static int cls_cgroup_change(struct net *net, struct sk_buff *in_skb,
95                              struct tcf_proto *tp, unsigned long base,
96                              u32 handle, struct nlattr **tca,
97                              unsigned long *arg, bool ovr)
98 {
99         struct nlattr *tb[TCA_CGROUP_MAX + 1];
100         struct cls_cgroup_head *head = rtnl_dereference(tp->root);
101         struct cls_cgroup_head *new;
102         struct tcf_ematch_tree t;
103         struct tcf_exts e;
104         int err;
105
106         if (!tca[TCA_OPTIONS])
107                 return -EINVAL;
108
109         if (!head && !handle)
110                 return -EINVAL;
111
112         if (head && handle != head->handle)
113                 return -ENOENT;
114
115         new = kzalloc(sizeof(*head), GFP_KERNEL);
116         if (!new)
117                 return -ENOBUFS;
118
119         if (head) {
120                 new->handle = head->handle;
121         } else {
122                 tcf_exts_init(&new->exts, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
123                 new->handle = handle;
124         }
125
126         new->tp = tp;
127         err = nla_parse_nested(tb, TCA_CGROUP_MAX, tca[TCA_OPTIONS],
128                                cgroup_policy);
129         if (err < 0)
130                 return err;
131
132         tcf_exts_init(&e, TCA_CGROUP_ACT, TCA_CGROUP_POLICE);
133         err = tcf_exts_validate(net, tp, tb, tca[TCA_RATE], &e, ovr);
134         if (err < 0)
135                 return err;
136
137         err = tcf_em_tree_validate(tp, tb[TCA_CGROUP_EMATCHES], &t);
138         if (err < 0)
139                 return err;
140
141         tcf_exts_change(tp, &new->exts, &e);
142         tcf_em_tree_change(tp, &new->ematches, &t);
143
144         rcu_assign_pointer(tp->root, new);
145         if (head)
146                 call_rcu(&head->rcu, cls_cgroup_destroy_rcu);
147         return 0;
148 }
149
150 static void cls_cgroup_destroy(struct tcf_proto *tp)
151 {
152         struct cls_cgroup_head *head = rtnl_dereference(tp->root);
153
154         if (head) {
155                 tcf_exts_destroy(tp, &head->exts);
156                 tcf_em_tree_destroy(tp, &head->ematches);
157                 RCU_INIT_POINTER(tp->root, NULL);
158                 kfree_rcu(head, rcu);
159         }
160 }
161
162 static int cls_cgroup_delete(struct tcf_proto *tp, unsigned long arg)
163 {
164         return -EOPNOTSUPP;
165 }
166
167 static void cls_cgroup_walk(struct tcf_proto *tp, struct tcf_walker *arg)
168 {
169         struct cls_cgroup_head *head = rtnl_dereference(tp->root);
170
171         if (arg->count < arg->skip)
172                 goto skip;
173
174         if (arg->fn(tp, (unsigned long) head, arg) < 0) {
175                 arg->stop = 1;
176                 return;
177         }
178 skip:
179         arg->count++;
180 }
181
182 static int cls_cgroup_dump(struct net *net, struct tcf_proto *tp, unsigned long fh,
183                            struct sk_buff *skb, struct tcmsg *t)
184 {
185         struct cls_cgroup_head *head = rtnl_dereference(tp->root);
186         unsigned char *b = skb_tail_pointer(skb);
187         struct nlattr *nest;
188
189         t->tcm_handle = head->handle;
190
191         nest = nla_nest_start(skb, TCA_OPTIONS);
192         if (nest == NULL)
193                 goto nla_put_failure;
194
195         if (tcf_exts_dump(skb, &head->exts) < 0 ||
196             tcf_em_tree_dump(skb, &head->ematches, TCA_CGROUP_EMATCHES) < 0)
197                 goto nla_put_failure;
198
199         nla_nest_end(skb, nest);
200
201         if (tcf_exts_dump_stats(skb, &head->exts) < 0)
202                 goto nla_put_failure;
203
204         return skb->len;
205
206 nla_put_failure:
207         nlmsg_trim(skb, b);
208         return -1;
209 }
210
211 static struct tcf_proto_ops cls_cgroup_ops __read_mostly = {
212         .kind           =       "cgroup",
213         .init           =       cls_cgroup_init,
214         .change         =       cls_cgroup_change,
215         .classify       =       cls_cgroup_classify,
216         .destroy        =       cls_cgroup_destroy,
217         .get            =       cls_cgroup_get,
218         .put            =       cls_cgroup_put,
219         .delete         =       cls_cgroup_delete,
220         .walk           =       cls_cgroup_walk,
221         .dump           =       cls_cgroup_dump,
222         .owner          =       THIS_MODULE,
223 };
224
225 static int __init init_cgroup_cls(void)
226 {
227         return register_tcf_proto_ops(&cls_cgroup_ops);
228 }
229
230 static void __exit exit_cgroup_cls(void)
231 {
232         unregister_tcf_proto_ops(&cls_cgroup_ops);
233 }
234
235 module_init(init_cgroup_cls);
236 module_exit(exit_cgroup_cls);
237 MODULE_LICENSE("GPL");