xref: /openbmc/linux/net/sched/act_ctinfo.c (revision 8e7a49e0) 
1 // SPDX-License-Identifier: GPL-2.0+
2 /* net/sched/act_ctinfo.c  netfilter ctinfo connmark actions
3  *
4  * Copyright (c) 2019 Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
5  */
6 
7 #include <linux/module.h>
8 #include <linux/init.h>
9 #include <linux/kernel.h>
10 #include <linux/skbuff.h>
11 #include <linux/rtnetlink.h>
12 #include <linux/pkt_cls.h>
13 #include <linux/ip.h>
14 #include <linux/ipv6.h>
15 #include <net/netlink.h>
16 #include <net/pkt_sched.h>
17 #include <net/act_api.h>
18 #include <net/pkt_cls.h>
19 #include <uapi/linux/tc_act/tc_ctinfo.h>
20 #include <net/tc_act/tc_ctinfo.h>
21 #include <net/tc_wrapper.h>
22 
23 #include <net/netfilter/nf_conntrack.h>
24 #include <net/netfilter/nf_conntrack_core.h>
25 #include <net/netfilter/nf_conntrack_ecache.h>
26 #include <net/netfilter/nf_conntrack_zones.h>
27 
28 static struct tc_action_ops act_ctinfo_ops;
29 
30 static void tcf_ctinfo_dscp_set(struct nf_conn *ct, struct tcf_ctinfo *ca,
31 				struct tcf_ctinfo_params *cp,
32 				struct sk_buff *skb, int wlen, int proto)
33 {
34 	u8 dscp, newdscp;
35 
36 	newdscp = (((READ_ONCE(ct->mark) & cp->dscpmask) >> cp->dscpmaskshift) << 2) &
37 		     ~INET_ECN_MASK;
38 
39 	switch (proto) {
40 	case NFPROTO_IPV4:
41 		dscp = ipv4_get_dsfield(ip_hdr(skb)) & ~INET_ECN_MASK;
42 		if (dscp != newdscp) {
43 			if (likely(!skb_try_make_writable(skb, wlen))) {
44 				ipv4_change_dsfield(ip_hdr(skb),
45 						    INET_ECN_MASK,
46 						    newdscp);
47 				ca->stats_dscp_set++;
48 			} else {
49 				ca->stats_dscp_error++;
50 			}
51 		}
52 		break;
53 	case NFPROTO_IPV6:
54 		dscp = ipv6_get_dsfield(ipv6_hdr(skb)) & ~INET_ECN_MASK;
55 		if (dscp != newdscp) {
56 			if (likely(!skb_try_make_writable(skb, wlen))) {
57 				ipv6_change_dsfield(ipv6_hdr(skb),
58 						    INET_ECN_MASK,
59 						    newdscp);
60 				ca->stats_dscp_set++;
61 			} else {
62 				ca->stats_dscp_error++;
63 			}
64 		}
65 		break;
66 	default:
67 		break;
68 	}
69 }
70 
71 static void tcf_ctinfo_cpmark_set(struct nf_conn *ct, struct tcf_ctinfo *ca,
72 				  struct tcf_ctinfo_params *cp,
73 				  struct sk_buff *skb)
74 {
75 	ca->stats_cpmark_set++;
76 	skb->mark = READ_ONCE(ct->mark) & cp->cpmarkmask;
77 }
78 
79 TC_INDIRECT_SCOPE int tcf_ctinfo_act(struct sk_buff *skb,
80 				     const struct tc_action *a,
81 				     struct tcf_result *res)
82 {
83 	const struct nf_conntrack_tuple_hash *thash = NULL;
84 	struct tcf_ctinfo *ca = to_ctinfo(a);
85 	struct nf_conntrack_tuple tuple;
86 	struct nf_conntrack_zone zone;
87 	enum ip_conntrack_info ctinfo;
88 	struct tcf_ctinfo_params *cp;
89 	struct nf_conn *ct;
90 	int proto, wlen;
91 	int action;
92 
93 	cp = rcu_dereference_bh(ca->params);
94 
95 	tcf_lastuse_update(&ca->tcf_tm);
96 	tcf_action_update_bstats(&ca->common, skb);
97 	action = READ_ONCE(ca->tcf_action);
98 
99 	wlen = skb_network_offset(skb);
100 	switch (skb_protocol(skb, true)) {
101 	case htons(ETH_P_IP):
102 		wlen += sizeof(struct iphdr);
103 		if (!pskb_may_pull(skb, wlen))
104 			goto out;
105 
106 		proto = NFPROTO_IPV4;
107 		break;
108 	case htons(ETH_P_IPV6):
109 		wlen += sizeof(struct ipv6hdr);
110 		if (!pskb_may_pull(skb, wlen))
111 			goto out;
112 
113 		proto = NFPROTO_IPV6;
114 		break;
115 	default:
116 		goto out;
117 	}
118 
119 	ct = nf_ct_get(skb, &ctinfo);
120 	if (!ct) { /* look harder, usually ingress */
121 		if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
122 				       proto, cp->net, &tuple))
123 			goto out;
124 		zone.id = cp->zone;
125 		zone.dir = NF_CT_DEFAULT_ZONE_DIR;
126 
127 		thash = nf_conntrack_find_get(cp->net, &zone, &tuple);
128 		if (!thash)
129 			goto out;
130 
131 		ct = nf_ct_tuplehash_to_ctrack(thash);
132 	}
133 
134 	if (cp->mode & CTINFO_MODE_DSCP)
135 		if (!cp->dscpstatemask || (READ_ONCE(ct->mark) & cp->dscpstatemask))
136 			tcf_ctinfo_dscp_set(ct, ca, cp, skb, wlen, proto);
137 
138 	if (cp->mode & CTINFO_MODE_CPMARK)
139 		tcf_ctinfo_cpmark_set(ct, ca, cp, skb);
140 
141 	if (thash)
142 		nf_ct_put(ct);
143 out:
144 	return action;
145 }
146 
147 static const struct nla_policy ctinfo_policy[TCA_CTINFO_MAX + 1] = {
148 	[TCA_CTINFO_ACT]		  =
149 		NLA_POLICY_EXACT_LEN(sizeof(struct tc_ctinfo)),
150 	[TCA_CTINFO_ZONE]		  = { .type = NLA_U16 },
151 	[TCA_CTINFO_PARMS_DSCP_MASK]	  = { .type = NLA_U32 },
152 	[TCA_CTINFO_PARMS_DSCP_STATEMASK] = { .type = NLA_U32 },
153 	[TCA_CTINFO_PARMS_CPMARK_MASK]	  = { .type = NLA_U32 },
154 };
155 
156 static int tcf_ctinfo_init(struct net *net, struct nlattr *nla,
157 			   struct nlattr *est, struct tc_action **a,
158 			   struct tcf_proto *tp, u32 flags,
159 			   struct netlink_ext_ack *extack)
160 {
161 	struct tc_action_net *tn = net_generic(net, act_ctinfo_ops.net_id);
162 	bool bind = flags & TCA_ACT_FLAGS_BIND;
163 	u32 dscpmask = 0, dscpstatemask, index;
164 	struct nlattr *tb[TCA_CTINFO_MAX + 1];
165 	struct tcf_ctinfo_params *cp_new;
166 	struct tcf_chain *goto_ch = NULL;
167 	struct tc_ctinfo *actparm;
168 	struct tcf_ctinfo *ci;
169 	u8 dscpmaskshift;
170 	int ret = 0, err;
171 
172 	if (!nla) {
173 		NL_SET_ERR_MSG_MOD(extack, "ctinfo requires attributes to be passed");
174 		return -EINVAL;
175 	}
176 
177 	err = nla_parse_nested(tb, TCA_CTINFO_MAX, nla, ctinfo_policy, extack);
178 	if (err < 0)
179 		return err;
180 
181 	if (!tb[TCA_CTINFO_ACT]) {
182 		NL_SET_ERR_MSG_MOD(extack,
183 				   "Missing required TCA_CTINFO_ACT attribute");
184 		return -EINVAL;
185 	}
186 	actparm = nla_data(tb[TCA_CTINFO_ACT]);
187 
188 	/* do some basic validation here before dynamically allocating things */
189 	/* that we would otherwise have to clean up.			      */
190 	if (tb[TCA_CTINFO_PARMS_DSCP_MASK]) {
191 		dscpmask = nla_get_u32(tb[TCA_CTINFO_PARMS_DSCP_MASK]);
192 		/* need contiguous 6 bit mask */
193 		dscpmaskshift = dscpmask ? __ffs(dscpmask) : 0;
194 		if ((~0 & (dscpmask >> dscpmaskshift)) != 0x3f) {
195 			NL_SET_ERR_MSG_ATTR(extack,
196 					    tb[TCA_CTINFO_PARMS_DSCP_MASK],
197 					    "dscp mask must be 6 contiguous bits");
198 			return -EINVAL;
199 		}
200 		dscpstatemask = tb[TCA_CTINFO_PARMS_DSCP_STATEMASK] ?
201 			nla_get_u32(tb[TCA_CTINFO_PARMS_DSCP_STATEMASK]) : 0;
202 		/* mask & statemask must not overlap */
203 		if (dscpmask & dscpstatemask) {
204 			NL_SET_ERR_MSG_ATTR(extack,
205 					    tb[TCA_CTINFO_PARMS_DSCP_STATEMASK],
206 					    "dscp statemask must not overlap dscp mask");
207 			return -EINVAL;
208 		}
209 	}
210 
211 	/* done the validation:now to the actual action allocation */
212 	index = actparm->index;
213 	err = tcf_idr_check_alloc(tn, &index, a, bind);
214 	if (!err) {
215 		ret = tcf_idr_create_from_flags(tn, index, est, a,
216 						&act_ctinfo_ops, bind, flags);
217 		if (ret) {
218 			tcf_idr_cleanup(tn, index);
219 			return ret;
220 		}
221 		ret = ACT_P_CREATED;
222 	} else if (err > 0) {
223 		if (bind) /* don't override defaults */
224 			return 0;
225 		if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
226 			tcf_idr_release(*a, bind);
227 			return -EEXIST;
228 		}
229 	} else {
230 		return err;
231 	}
232 
233 	err = tcf_action_check_ctrlact(actparm->action, tp, &goto_ch, extack);
234 	if (err < 0)
235 		goto release_idr;
236 
237 	ci = to_ctinfo(*a);
238 
239 	cp_new = kzalloc(sizeof(*cp_new), GFP_KERNEL);
240 	if (unlikely(!cp_new)) {
241 		err = -ENOMEM;
242 		goto put_chain;
243 	}
244 
245 	cp_new->net = net;
246 	cp_new->zone = tb[TCA_CTINFO_ZONE] ?
247 			nla_get_u16(tb[TCA_CTINFO_ZONE]) : 0;
248 	if (dscpmask) {
249 		cp_new->dscpmask = dscpmask;
250 		cp_new->dscpmaskshift = dscpmaskshift;
251 		cp_new->dscpstatemask = dscpstatemask;
252 		cp_new->mode |= CTINFO_MODE_DSCP;
253 	}
254 
255 	if (tb[TCA_CTINFO_PARMS_CPMARK_MASK]) {
256 		cp_new->cpmarkmask =
257 				nla_get_u32(tb[TCA_CTINFO_PARMS_CPMARK_MASK]);
258 		cp_new->mode |= CTINFO_MODE_CPMARK;
259 	}
260 
261 	spin_lock_bh(&ci->tcf_lock);
262 	goto_ch = tcf_action_set_ctrlact(*a, actparm->action, goto_ch);
263 	cp_new = rcu_replace_pointer(ci->params, cp_new,
264 				     lockdep_is_held(&ci->tcf_lock));
265 	spin_unlock_bh(&ci->tcf_lock);
266 
267 	if (goto_ch)
268 		tcf_chain_put_by_act(goto_ch);
269 	if (cp_new)
270 		kfree_rcu(cp_new, rcu);
271 
272 	return ret;
273 
274 put_chain:
275 	if (goto_ch)
276 		tcf_chain_put_by_act(goto_ch);
277 release_idr:
278 	tcf_idr_release(*a, bind);
279 	return err;
280 }
281 
282 static int tcf_ctinfo_dump(struct sk_buff *skb, struct tc_action *a,
283 			   int bind, int ref)
284 {
285 	struct tcf_ctinfo *ci = to_ctinfo(a);
286 	struct tc_ctinfo opt = {
287 		.index   = ci->tcf_index,
288 		.refcnt  = refcount_read(&ci->tcf_refcnt) - ref,
289 		.bindcnt = atomic_read(&ci->tcf_bindcnt) - bind,
290 	};
291 	unsigned char *b = skb_tail_pointer(skb);
292 	struct tcf_ctinfo_params *cp;
293 	struct tcf_t t;
294 
295 	spin_lock_bh(&ci->tcf_lock);
296 	cp = rcu_dereference_protected(ci->params,
297 				       lockdep_is_held(&ci->tcf_lock));
298 
299 	tcf_tm_dump(&t, &ci->tcf_tm);
300 	if (nla_put_64bit(skb, TCA_CTINFO_TM, sizeof(t), &t, TCA_CTINFO_PAD))
301 		goto nla_put_failure;
302 
303 	opt.action = ci->tcf_action;
304 	if (nla_put(skb, TCA_CTINFO_ACT, sizeof(opt), &opt))
305 		goto nla_put_failure;
306 
307 	if (nla_put_u16(skb, TCA_CTINFO_ZONE, cp->zone))
308 		goto nla_put_failure;
309 
310 	if (cp->mode & CTINFO_MODE_DSCP) {
311 		if (nla_put_u32(skb, TCA_CTINFO_PARMS_DSCP_MASK,
312 				cp->dscpmask))
313 			goto nla_put_failure;
314 		if (nla_put_u32(skb, TCA_CTINFO_PARMS_DSCP_STATEMASK,
315 				cp->dscpstatemask))
316 			goto nla_put_failure;
317 	}
318 
319 	if (cp->mode & CTINFO_MODE_CPMARK) {
320 		if (nla_put_u32(skb, TCA_CTINFO_PARMS_CPMARK_MASK,
321 				cp->cpmarkmask))
322 			goto nla_put_failure;
323 	}
324 
325 	if (nla_put_u64_64bit(skb, TCA_CTINFO_STATS_DSCP_SET,
326 			      ci->stats_dscp_set, TCA_CTINFO_PAD))
327 		goto nla_put_failure;
328 
329 	if (nla_put_u64_64bit(skb, TCA_CTINFO_STATS_DSCP_ERROR,
330 			      ci->stats_dscp_error, TCA_CTINFO_PAD))
331 		goto nla_put_failure;
332 
333 	if (nla_put_u64_64bit(skb, TCA_CTINFO_STATS_CPMARK_SET,
334 			      ci->stats_cpmark_set, TCA_CTINFO_PAD))
335 		goto nla_put_failure;
336 
337 	spin_unlock_bh(&ci->tcf_lock);
338 	return skb->len;
339 
340 nla_put_failure:
341 	spin_unlock_bh(&ci->tcf_lock);
342 	nlmsg_trim(skb, b);
343 	return -1;
344 }
345 
346 static void tcf_ctinfo_cleanup(struct tc_action *a)
347 {
348 	struct tcf_ctinfo *ci = to_ctinfo(a);
349 	struct tcf_ctinfo_params *cp;
350 
351 	cp = rcu_dereference_protected(ci->params, 1);
352 	if (cp)
353 		kfree_rcu(cp, rcu);
354 }
355 
356 static struct tc_action_ops act_ctinfo_ops = {
357 	.kind	= "ctinfo",
358 	.id	= TCA_ID_CTINFO,
359 	.owner	= THIS_MODULE,
360 	.act	= tcf_ctinfo_act,
361 	.dump	= tcf_ctinfo_dump,
362 	.init	= tcf_ctinfo_init,
363 	.cleanup= tcf_ctinfo_cleanup,
364 	.size	= sizeof(struct tcf_ctinfo),
365 };
366 
367 static __net_init int ctinfo_init_net(struct net *net)
368 {
369 	struct tc_action_net *tn = net_generic(net, act_ctinfo_ops.net_id);
370 
371 	return tc_action_net_init(net, tn, &act_ctinfo_ops);
372 }
373 
374 static void __net_exit ctinfo_exit_net(struct list_head *net_list)
375 {
376 	tc_action_net_exit(net_list, act_ctinfo_ops.net_id);
377 }
378 
379 static struct pernet_operations ctinfo_net_ops = {
380 	.init		= ctinfo_init_net,
381 	.exit_batch	= ctinfo_exit_net,
382 	.id		= &act_ctinfo_ops.net_id,
383 	.size		= sizeof(struct tc_action_net),
384 };
385 
386 static int __init ctinfo_init_module(void)
387 {
388 	return tcf_register_action(&act_ctinfo_ops, &ctinfo_net_ops);
389 }
390 
391 static void __exit ctinfo_cleanup_module(void)
392 {
393 	tcf_unregister_action(&act_ctinfo_ops, &ctinfo_net_ops);
394 }
395 
396 module_init(ctinfo_init_module);
397 module_exit(ctinfo_cleanup_module);
398 MODULE_AUTHOR("Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>");
399 MODULE_DESCRIPTION("Connection tracking mark actions");
400 MODULE_LICENSE("GPL");
401