xref: /linux/net/sched/act_connmark.c (revision 1b0975ee3bdd3eb19a47371c26fd7ef8f7f6b599)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/act_connmark.c  netfilter connmark retriever action
4  * skb mark is over-written
5  *
6  * Copyright (c) 2011 Felix Fietkau <nbd@openwrt.org>
7 */
8 
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/kernel.h>
12 #include <linux/skbuff.h>
13 #include <linux/rtnetlink.h>
14 #include <linux/pkt_cls.h>
15 #include <linux/ip.h>
16 #include <linux/ipv6.h>
17 #include <net/netlink.h>
18 #include <net/pkt_sched.h>
19 #include <net/act_api.h>
20 #include <net/pkt_cls.h>
21 #include <uapi/linux/tc_act/tc_connmark.h>
22 #include <net/tc_act/tc_connmark.h>
23 #include <net/tc_wrapper.h>
24 
25 #include <net/netfilter/nf_conntrack.h>
26 #include <net/netfilter/nf_conntrack_core.h>
27 #include <net/netfilter/nf_conntrack_zones.h>
28 
29 static struct tc_action_ops act_connmark_ops;
30 
31 TC_INDIRECT_SCOPE int tcf_connmark_act(struct sk_buff *skb,
32 				       const struct tc_action *a,
33 				       struct tcf_result *res)
34 {
35 	const struct nf_conntrack_tuple_hash *thash;
36 	struct nf_conntrack_tuple tuple;
37 	enum ip_conntrack_info ctinfo;
38 	struct tcf_connmark_info *ca = to_connmark(a);
39 	struct tcf_connmark_parms *parms;
40 	struct nf_conntrack_zone zone;
41 	struct nf_conn *c;
42 	int proto;
43 
44 	tcf_lastuse_update(&ca->tcf_tm);
45 	tcf_action_update_bstats(&ca->common, skb);
46 
47 	parms = rcu_dereference_bh(ca->parms);
48 
49 	switch (skb_protocol(skb, true)) {
50 	case htons(ETH_P_IP):
51 		if (skb->len < sizeof(struct iphdr))
52 			goto out;
53 
54 		proto = NFPROTO_IPV4;
55 		break;
56 	case htons(ETH_P_IPV6):
57 		if (skb->len < sizeof(struct ipv6hdr))
58 			goto out;
59 
60 		proto = NFPROTO_IPV6;
61 		break;
62 	default:
63 		goto out;
64 	}
65 
66 	c = nf_ct_get(skb, &ctinfo);
67 	if (c) {
68 		skb->mark = READ_ONCE(c->mark);
69 		goto count;
70 	}
71 
72 	if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, parms->net,
73 			       &tuple))
74 		goto out;
75 
76 	zone.id = parms->zone;
77 	zone.dir = NF_CT_DEFAULT_ZONE_DIR;
78 
79 	thash = nf_conntrack_find_get(parms->net, &zone, &tuple);
80 	if (!thash)
81 		goto out;
82 
83 	c = nf_ct_tuplehash_to_ctrack(thash);
84 	skb->mark = READ_ONCE(c->mark);
85 	nf_ct_put(c);
86 
87 count:
88 	/* using overlimits stats to count how many packets marked */
89 	tcf_action_inc_overlimit_qstats(&ca->common);
90 out:
91 	return READ_ONCE(ca->tcf_action);
92 }
93 
94 static const struct nla_policy connmark_policy[TCA_CONNMARK_MAX + 1] = {
95 	[TCA_CONNMARK_PARMS] = { .len = sizeof(struct tc_connmark) },
96 };
97 
98 static int tcf_connmark_init(struct net *net, struct nlattr *nla,
99 			     struct nlattr *est, struct tc_action **a,
100 			     struct tcf_proto *tp, u32 flags,
101 			     struct netlink_ext_ack *extack)
102 {
103 	struct tc_action_net *tn = net_generic(net, act_connmark_ops.net_id);
104 	struct tcf_connmark_parms *nparms, *oparms;
105 	struct nlattr *tb[TCA_CONNMARK_MAX + 1];
106 	bool bind = flags & TCA_ACT_FLAGS_BIND;
107 	struct tcf_chain *goto_ch = NULL;
108 	struct tcf_connmark_info *ci;
109 	struct tc_connmark *parm;
110 	int ret = 0, err;
111 	u32 index;
112 
113 	if (!nla)
114 		return -EINVAL;
115 
116 	ret = nla_parse_nested_deprecated(tb, TCA_CONNMARK_MAX, nla,
117 					  connmark_policy, NULL);
118 	if (ret < 0)
119 		return ret;
120 
121 	if (!tb[TCA_CONNMARK_PARMS])
122 		return -EINVAL;
123 
124 	nparms = kzalloc(sizeof(*nparms), GFP_KERNEL);
125 	if (!nparms)
126 		return -ENOMEM;
127 
128 	parm = nla_data(tb[TCA_CONNMARK_PARMS]);
129 	index = parm->index;
130 	ret = tcf_idr_check_alloc(tn, &index, a, bind);
131 	if (!ret) {
132 		ret = tcf_idr_create_from_flags(tn, index, est, a,
133 						&act_connmark_ops, bind, flags);
134 		if (ret) {
135 			tcf_idr_cleanup(tn, index);
136 			err = ret;
137 			goto out_free;
138 		}
139 
140 		ci = to_connmark(*a);
141 
142 		nparms->net = net;
143 		nparms->zone = parm->zone;
144 
145 		ret = ACT_P_CREATED;
146 	} else if (ret > 0) {
147 		ci = to_connmark(*a);
148 		if (bind) {
149 			err = 0;
150 			goto out_free;
151 		}
152 		if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
153 			err = -EEXIST;
154 			goto release_idr;
155 		}
156 
157 		nparms->net = rtnl_dereference(ci->parms)->net;
158 		nparms->zone = parm->zone;
159 
160 		ret = 0;
161 	} else {
162 		err = ret;
163 		goto out_free;
164 	}
165 
166 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
167 	if (err < 0)
168 		goto release_idr;
169 
170 	spin_lock_bh(&ci->tcf_lock);
171 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
172 	oparms = rcu_replace_pointer(ci->parms, nparms, lockdep_is_held(&ci->tcf_lock));
173 	spin_unlock_bh(&ci->tcf_lock);
174 
175 	if (goto_ch)
176 		tcf_chain_put_by_act(goto_ch);
177 
178 	if (oparms)
179 		kfree_rcu(oparms, rcu);
180 
181 	return ret;
182 
183 release_idr:
184 	tcf_idr_release(*a, bind);
185 out_free:
186 	kfree(nparms);
187 	return err;
188 }
189 
190 static inline int tcf_connmark_dump(struct sk_buff *skb, struct tc_action *a,
191 				    int bind, int ref)
192 {
193 	unsigned char *b = skb_tail_pointer(skb);
194 	struct tcf_connmark_info *ci = to_connmark(a);
195 	struct tc_connmark opt = {
196 		.index   = ci->tcf_index,
197 		.refcnt  = refcount_read(&ci->tcf_refcnt) - ref,
198 		.bindcnt = atomic_read(&ci->tcf_bindcnt) - bind,
199 	};
200 	struct tcf_connmark_parms *parms;
201 	struct tcf_t t;
202 
203 	spin_lock_bh(&ci->tcf_lock);
204 	parms = rcu_dereference_protected(ci->parms, lockdep_is_held(&ci->tcf_lock));
205 
206 	opt.action = ci->tcf_action;
207 	opt.zone = parms->zone;
208 	if (nla_put(skb, TCA_CONNMARK_PARMS, sizeof(opt), &opt))
209 		goto nla_put_failure;
210 
211 	tcf_tm_dump(&t, &ci->tcf_tm);
212 	if (nla_put_64bit(skb, TCA_CONNMARK_TM, sizeof(t), &t,
213 			  TCA_CONNMARK_PAD))
214 		goto nla_put_failure;
215 	spin_unlock_bh(&ci->tcf_lock);
216 
217 	return skb->len;
218 
219 nla_put_failure:
220 	spin_unlock_bh(&ci->tcf_lock);
221 	nlmsg_trim(skb, b);
222 	return -1;
223 }
224 
225 static void tcf_connmark_cleanup(struct tc_action *a)
226 {
227 	struct tcf_connmark_info *ci = to_connmark(a);
228 	struct tcf_connmark_parms *parms;
229 
230 	parms = rcu_dereference_protected(ci->parms, 1);
231 	if (parms)
232 		kfree_rcu(parms, rcu);
233 }
234 
235 static struct tc_action_ops act_connmark_ops = {
236 	.kind		=	"connmark",
237 	.id		=	TCA_ID_CONNMARK,
238 	.owner		=	THIS_MODULE,
239 	.act		=	tcf_connmark_act,
240 	.dump		=	tcf_connmark_dump,
241 	.init		=	tcf_connmark_init,
242 	.cleanup	=	tcf_connmark_cleanup,
243 	.size		=	sizeof(struct tcf_connmark_info),
244 };
245 
246 static __net_init int connmark_init_net(struct net *net)
247 {
248 	struct tc_action_net *tn = net_generic(net, act_connmark_ops.net_id);
249 
250 	return tc_action_net_init(net, tn, &act_connmark_ops);
251 }
252 
253 static void __net_exit connmark_exit_net(struct list_head *net_list)
254 {
255 	tc_action_net_exit(net_list, act_connmark_ops.net_id);
256 }
257 
258 static struct pernet_operations connmark_net_ops = {
259 	.init = connmark_init_net,
260 	.exit_batch = connmark_exit_net,
261 	.id   = &act_connmark_ops.net_id,
262 	.size = sizeof(struct tc_action_net),
263 };
264 
265 static int __init connmark_init_module(void)
266 {
267 	return tcf_register_action(&act_connmark_ops, &connmark_net_ops);
268 }
269 
270 static void __exit connmark_cleanup_module(void)
271 {
272 	tcf_unregister_action(&act_connmark_ops, &connmark_net_ops);
273 }
274 
275 module_init(connmark_init_module);
276 module_exit(connmark_cleanup_module);
277 MODULE_AUTHOR("Felix Fietkau <nbd@openwrt.org>");
278 MODULE_DESCRIPTION("Connection tracking mark restoring");
279 MODULE_LICENSE("GPL");
280