pkt_sched: Schedule qdiscs instead of netdev_queue.
[linux-2.6] / net / sched / act_ipt.c
1 /*
2  * net/sched/ipt.c      iptables target interface
3  *
4  *TODO: Add other tables. For now we only support the ipv4 table targets
5  *
6  *              This program is free software; you can redistribute it and/or
7  *              modify it under the terms of the GNU General Public License
8  *              as published by the Free Software Foundation; either version
9  *              2 of the License, or (at your option) any later version.
10  *
11  * Copyright:   Jamal Hadi Salim (2002-4)
12  */
13
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/skbuff.h>
19 #include <linux/rtnetlink.h>
20 #include <linux/module.h>
21 #include <linux/init.h>
22 #include <net/netlink.h>
23 #include <net/pkt_sched.h>
24 #include <linux/tc_act/tc_ipt.h>
25 #include <net/tc_act/tc_ipt.h>
26
27 #include <linux/netfilter_ipv4/ip_tables.h>
28
29
30 #define IPT_TAB_MASK     15
31 static struct tcf_common *tcf_ipt_ht[IPT_TAB_MASK + 1];
32 static u32 ipt_idx_gen;
33 static DEFINE_RWLOCK(ipt_lock);
34
35 static struct tcf_hashinfo ipt_hash_info = {
36         .htab   =       tcf_ipt_ht,
37         .hmask  =       IPT_TAB_MASK,
38         .lock   =       &ipt_lock,
39 };
40
41 static int ipt_init_target(struct ipt_entry_target *t, char *table, unsigned int hook)
42 {
43         struct xt_target *target;
44         int ret = 0;
45
46         target = xt_request_find_target(AF_INET, t->u.user.name,
47                                         t->u.user.revision);
48         if (!target)
49                 return -ENOENT;
50
51         t->u.kernel.target = target;
52
53         ret = xt_check_target(target, AF_INET, t->u.target_size - sizeof(*t),
54                               table, hook, 0, 0);
55         if (ret) {
56                 module_put(t->u.kernel.target->me);
57                 return ret;
58         }
59         if (t->u.kernel.target->checkentry
60             && !t->u.kernel.target->checkentry(table, NULL,
61                                                t->u.kernel.target, t->data,
62                                                hook)) {
63                 module_put(t->u.kernel.target->me);
64                 ret = -EINVAL;
65         }
66
67         return ret;
68 }
69
70 static void ipt_destroy_target(struct ipt_entry_target *t)
71 {
72         if (t->u.kernel.target->destroy)
73                 t->u.kernel.target->destroy(t->u.kernel.target, t->data);
74         module_put(t->u.kernel.target->me);
75 }
76
77 static int tcf_ipt_release(struct tcf_ipt *ipt, int bind)
78 {
79         int ret = 0;
80         if (ipt) {
81                 if (bind)
82                         ipt->tcf_bindcnt--;
83                 ipt->tcf_refcnt--;
84                 if (ipt->tcf_bindcnt <= 0 && ipt->tcf_refcnt <= 0) {
85                         ipt_destroy_target(ipt->tcfi_t);
86                         kfree(ipt->tcfi_tname);
87                         kfree(ipt->tcfi_t);
88                         tcf_hash_destroy(&ipt->common, &ipt_hash_info);
89                         ret = ACT_P_DELETED;
90                 }
91         }
92         return ret;
93 }
94
95 static const struct nla_policy ipt_policy[TCA_IPT_MAX + 1] = {
96         [TCA_IPT_TABLE] = { .type = NLA_STRING, .len = IFNAMSIZ },
97         [TCA_IPT_HOOK]  = { .type = NLA_U32 },
98         [TCA_IPT_INDEX] = { .type = NLA_U32 },
99         [TCA_IPT_TARG]  = { .len = sizeof(struct ipt_entry_target) },
100 };
101
102 static int tcf_ipt_init(struct nlattr *nla, struct nlattr *est,
103                         struct tc_action *a, int ovr, int bind)
104 {
105         struct nlattr *tb[TCA_IPT_MAX + 1];
106         struct tcf_ipt *ipt;
107         struct tcf_common *pc;
108         struct ipt_entry_target *td, *t;
109         char *tname;
110         int ret = 0, err;
111         u32 hook = 0;
112         u32 index = 0;
113
114         if (nla == NULL)
115                 return -EINVAL;
116
117         err = nla_parse_nested(tb, TCA_IPT_MAX, nla, ipt_policy);
118         if (err < 0)
119                 return err;
120
121         if (tb[TCA_IPT_HOOK] == NULL)
122                 return -EINVAL;
123         if (tb[TCA_IPT_TARG] == NULL)
124                 return -EINVAL;
125
126         td = (struct ipt_entry_target *)nla_data(tb[TCA_IPT_TARG]);
127         if (nla_len(tb[TCA_IPT_TARG]) < td->u.target_size)
128                 return -EINVAL;
129
130         if (tb[TCA_IPT_INDEX] != NULL)
131                 index = nla_get_u32(tb[TCA_IPT_INDEX]);
132
133         pc = tcf_hash_check(index, a, bind, &ipt_hash_info);
134         if (!pc) {
135                 pc = tcf_hash_create(index, est, a, sizeof(*ipt), bind,
136                                      &ipt_idx_gen, &ipt_hash_info);
137                 if (unlikely(!pc))
138                         return -ENOMEM;
139                 ret = ACT_P_CREATED;
140         } else {
141                 if (!ovr) {
142                         tcf_ipt_release(to_ipt(pc), bind);
143                         return -EEXIST;
144                 }
145         }
146         ipt = to_ipt(pc);
147
148         hook = nla_get_u32(tb[TCA_IPT_HOOK]);
149
150         err = -ENOMEM;
151         tname = kmalloc(IFNAMSIZ, GFP_KERNEL);
152         if (unlikely(!tname))
153                 goto err1;
154         if (tb[TCA_IPT_TABLE] == NULL ||
155             nla_strlcpy(tname, tb[TCA_IPT_TABLE], IFNAMSIZ) >= IFNAMSIZ)
156                 strcpy(tname, "mangle");
157
158         t = kmemdup(td, td->u.target_size, GFP_KERNEL);
159         if (unlikely(!t))
160                 goto err2;
161
162         if ((err = ipt_init_target(t, tname, hook)) < 0)
163                 goto err3;
164
165         spin_lock_bh(&ipt->tcf_lock);
166         if (ret != ACT_P_CREATED) {
167                 ipt_destroy_target(ipt->tcfi_t);
168                 kfree(ipt->tcfi_tname);
169                 kfree(ipt->tcfi_t);
170         }
171         ipt->tcfi_tname = tname;
172         ipt->tcfi_t     = t;
173         ipt->tcfi_hook  = hook;
174         spin_unlock_bh(&ipt->tcf_lock);
175         if (ret == ACT_P_CREATED)
176                 tcf_hash_insert(pc, &ipt_hash_info);
177         return ret;
178
179 err3:
180         kfree(t);
181 err2:
182         kfree(tname);
183 err1:
184         kfree(pc);
185         return err;
186 }
187
188 static int tcf_ipt_cleanup(struct tc_action *a, int bind)
189 {
190         struct tcf_ipt *ipt = a->priv;
191         return tcf_ipt_release(ipt, bind);
192 }
193
194 static int tcf_ipt(struct sk_buff *skb, struct tc_action *a,
195                    struct tcf_result *res)
196 {
197         int ret = 0, result = 0;
198         struct tcf_ipt *ipt = a->priv;
199
200         if (skb_cloned(skb)) {
201                 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
202                         return TC_ACT_UNSPEC;
203         }
204
205         spin_lock(&ipt->tcf_lock);
206
207         ipt->tcf_tm.lastuse = jiffies;
208         ipt->tcf_bstats.bytes += skb->len;
209         ipt->tcf_bstats.packets++;
210
211         /* yes, we have to worry about both in and out dev
212          worry later - danger - this API seems to have changed
213          from earlier kernels */
214         ret = ipt->tcfi_t->u.kernel.target->target(skb, skb->dev, NULL,
215                                                    ipt->tcfi_hook,
216                                                    ipt->tcfi_t->u.kernel.target,
217                                                    ipt->tcfi_t->data);
218         switch (ret) {
219         case NF_ACCEPT:
220                 result = TC_ACT_OK;
221                 break;
222         case NF_DROP:
223                 result = TC_ACT_SHOT;
224                 ipt->tcf_qstats.drops++;
225                 break;
226         case IPT_CONTINUE:
227                 result = TC_ACT_PIPE;
228                 break;
229         default:
230                 if (net_ratelimit())
231                         printk("Bogus netfilter code %d assume ACCEPT\n", ret);
232                 result = TC_POLICE_OK;
233                 break;
234         }
235         spin_unlock(&ipt->tcf_lock);
236         return result;
237
238 }
239
240 static int tcf_ipt_dump(struct sk_buff *skb, struct tc_action *a, int bind, int ref)
241 {
242         unsigned char *b = skb_tail_pointer(skb);
243         struct tcf_ipt *ipt = a->priv;
244         struct ipt_entry_target *t;
245         struct tcf_t tm;
246         struct tc_cnt c;
247
248         /* for simple targets kernel size == user size
249         ** user name = target name
250         ** for foolproof you need to not assume this
251         */
252
253         t = kmemdup(ipt->tcfi_t, ipt->tcfi_t->u.user.target_size, GFP_ATOMIC);
254         if (unlikely(!t))
255                 goto nla_put_failure;
256
257         c.bindcnt = ipt->tcf_bindcnt - bind;
258         c.refcnt = ipt->tcf_refcnt - ref;
259         strcpy(t->u.user.name, ipt->tcfi_t->u.kernel.target->name);
260
261         NLA_PUT(skb, TCA_IPT_TARG, ipt->tcfi_t->u.user.target_size, t);
262         NLA_PUT_U32(skb, TCA_IPT_INDEX, ipt->tcf_index);
263         NLA_PUT_U32(skb, TCA_IPT_HOOK, ipt->tcfi_hook);
264         NLA_PUT(skb, TCA_IPT_CNT, sizeof(struct tc_cnt), &c);
265         NLA_PUT_STRING(skb, TCA_IPT_TABLE, ipt->tcfi_tname);
266         tm.install = jiffies_to_clock_t(jiffies - ipt->tcf_tm.install);
267         tm.lastuse = jiffies_to_clock_t(jiffies - ipt->tcf_tm.lastuse);
268         tm.expires = jiffies_to_clock_t(ipt->tcf_tm.expires);
269         NLA_PUT(skb, TCA_IPT_TM, sizeof (tm), &tm);
270         kfree(t);
271         return skb->len;
272
273 nla_put_failure:
274         nlmsg_trim(skb, b);
275         kfree(t);
276         return -1;
277 }
278
279 static struct tc_action_ops act_ipt_ops = {
280         .kind           =       "ipt",
281         .hinfo          =       &ipt_hash_info,
282         .type           =       TCA_ACT_IPT,
283         .capab          =       TCA_CAP_NONE,
284         .owner          =       THIS_MODULE,
285         .act            =       tcf_ipt,
286         .dump           =       tcf_ipt_dump,
287         .cleanup        =       tcf_ipt_cleanup,
288         .lookup         =       tcf_hash_search,
289         .init           =       tcf_ipt_init,
290         .walk           =       tcf_generic_walker
291 };
292
293 MODULE_AUTHOR("Jamal Hadi Salim(2002-4)");
294 MODULE_DESCRIPTION("Iptables target actions");
295 MODULE_LICENSE("GPL");
296
297 static int __init ipt_init_module(void)
298 {
299         return tcf_register_action(&act_ipt_ops);
300 }
301
302 static void __exit ipt_cleanup_module(void)
303 {
304         tcf_unregister_action(&act_ipt_ops);
305 }
306
307 module_init(ipt_init_module);
308 module_exit(ipt_cleanup_module);