xref: /linux/net/ipv4/fib_rules.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		IPv4 Forwarding Information Base: policy rules.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  *		Thomas Graf <tgraf@suug.ch>
11  *
12  * Fixes:
13  *		Rani Assaf	:	local_rule cannot be deleted
14  *		Marc Boucher	:	routing by fwmark
15  */
16 
17 #include <linux/types.h>
18 #include <linux/kernel.h>
19 #include <linux/netdevice.h>
20 #include <linux/netlink.h>
21 #include <linux/inetdevice.h>
22 #include <linux/init.h>
23 #include <linux/list.h>
24 #include <linux/rcupdate.h>
25 #include <linux/export.h>
26 #include <net/inet_dscp.h>
27 #include <net/ip.h>
28 #include <net/route.h>
29 #include <net/tcp.h>
30 #include <net/ip_fib.h>
31 #include <net/nexthop.h>
32 #include <net/fib_rules.h>
33 #include <linux/indirect_call_wrapper.h>
34 
35 struct fib4_rule {
36 	struct fib_rule		common;
37 	u8			dst_len;
38 	u8			src_len;
39 	dscp_t			dscp;
40 	u8			dscp_full:1;	/* DSCP or TOS selector */
41 	__be32			src;
42 	__be32			srcmask;
43 	__be32			dst;
44 	__be32			dstmask;
45 #ifdef CONFIG_IP_ROUTE_CLASSID
46 	u32			tclassid;
47 #endif
48 };
49 
50 static bool fib4_rule_matchall(const struct fib_rule *rule)
51 {
52 	struct fib4_rule *r = container_of(rule, struct fib4_rule, common);
53 
54 	if (r->dst_len || r->src_len || r->dscp)
55 		return false;
56 	return fib_rule_matchall(rule);
57 }
58 
59 bool fib4_rule_default(const struct fib_rule *rule)
60 {
61 	if (!fib4_rule_matchall(rule) || rule->action != FR_ACT_TO_TBL ||
62 	    rule->l3mdev)
63 		return false;
64 	if (rule->table != RT_TABLE_LOCAL && rule->table != RT_TABLE_MAIN &&
65 	    rule->table != RT_TABLE_DEFAULT)
66 		return false;
67 	return true;
68 }
69 EXPORT_SYMBOL_GPL(fib4_rule_default);
70 
71 int fib4_rules_dump(struct net *net, struct notifier_block *nb,
72 		    struct netlink_ext_ack *extack)
73 {
74 	return fib_rules_dump(net, nb, AF_INET, extack);
75 }
76 
77 unsigned int fib4_rules_seq_read(const struct net *net)
78 {
79 	return fib_rules_seq_read(net, AF_INET);
80 }
81 
82 int __fib_lookup(struct net *net, struct flowi4 *flp,
83 		 struct fib_result *res, unsigned int flags)
84 {
85 	struct fib_lookup_arg arg = {
86 		.result = res,
87 		.flags = flags,
88 	};
89 	int err;
90 
91 	/* update flow if oif or iif point to device enslaved to l3mdev */
92 	l3mdev_update_flow(net, flowi4_to_flowi(flp));
93 
94 	err = fib_rules_lookup(net->ipv4.rules_ops, flowi4_to_flowi(flp), 0, &arg);
95 #ifdef CONFIG_IP_ROUTE_CLASSID
96 	if (arg.rule)
97 		res->tclassid = ((struct fib4_rule *)arg.rule)->tclassid;
98 	else
99 		res->tclassid = 0;
100 #endif
101 
102 	if (err == -ESRCH)
103 		err = -ENETUNREACH;
104 
105 	return err;
106 }
107 EXPORT_SYMBOL_GPL(__fib_lookup);
108 
109 INDIRECT_CALLABLE_SCOPE int fib4_rule_action(struct fib_rule *rule,
110 					     struct flowi *flp, int flags,
111 					     struct fib_lookup_arg *arg)
112 {
113 	int err = -EAGAIN;
114 	struct fib_table *tbl;
115 	u32 tb_id;
116 
117 	switch (rule->action) {
118 	case FR_ACT_TO_TBL:
119 		break;
120 
121 	case FR_ACT_UNREACHABLE:
122 		return -ENETUNREACH;
123 
124 	case FR_ACT_PROHIBIT:
125 		return -EACCES;
126 
127 	case FR_ACT_BLACKHOLE:
128 	default:
129 		return -EINVAL;
130 	}
131 
132 	rcu_read_lock();
133 
134 	tb_id = fib_rule_get_table(rule, arg);
135 	tbl = fib_get_table(rule->fr_net, tb_id);
136 	if (tbl)
137 		err = fib_table_lookup(tbl, &flp->u.ip4,
138 				       (struct fib_result *)arg->result,
139 				       arg->flags);
140 
141 	rcu_read_unlock();
142 	return err;
143 }
144 
145 INDIRECT_CALLABLE_SCOPE bool fib4_rule_suppress(struct fib_rule *rule,
146 						int flags,
147 						struct fib_lookup_arg *arg)
148 {
149 	struct fib_result *result = arg->result;
150 	struct net_device *dev = NULL;
151 
152 	if (result->fi) {
153 		struct fib_nh_common *nhc = fib_info_nhc(result->fi, 0);
154 
155 		dev = nhc->nhc_dev;
156 	}
157 
158 	/* do not accept result if the route does
159 	 * not meet the required prefix length
160 	 */
161 	if (result->prefixlen <= rule->suppress_prefixlen)
162 		goto suppress_route;
163 
164 	/* do not accept result if the route uses a device
165 	 * belonging to a forbidden interface group
166 	 */
167 	if (rule->suppress_ifgroup != -1 && dev && dev->group == rule->suppress_ifgroup)
168 		goto suppress_route;
169 
170 	return false;
171 
172 suppress_route:
173 	if (!(arg->flags & FIB_LOOKUP_NOREF))
174 		fib_info_put(result->fi);
175 	return true;
176 }
177 
178 INDIRECT_CALLABLE_SCOPE int fib4_rule_match(struct fib_rule *rule,
179 					    struct flowi *fl, int flags)
180 {
181 	struct fib4_rule *r = (struct fib4_rule *) rule;
182 	struct flowi4 *fl4 = &fl->u.ip4;
183 	__be32 daddr = fl4->daddr;
184 	__be32 saddr = fl4->saddr;
185 
186 	if (((saddr ^ r->src) & r->srcmask) ||
187 	    ((daddr ^ r->dst) & r->dstmask))
188 		return 0;
189 
190 	/* When DSCP selector is used we need to match on the entire DSCP field
191 	 * in the flow information structure. When TOS selector is used we need
192 	 * to mask the upper three DSCP bits prior to matching to maintain
193 	 * legacy behavior.
194 	 */
195 	if (r->dscp_full && r->dscp != inet_dsfield_to_dscp(fl4->flowi4_tos))
196 		return 0;
197 	else if (!r->dscp_full && r->dscp &&
198 		 !fib_dscp_masked_match(r->dscp, fl4))
199 		return 0;
200 
201 	if (rule->ip_proto && (rule->ip_proto != fl4->flowi4_proto))
202 		return 0;
203 
204 	if (fib_rule_port_range_set(&rule->sport_range) &&
205 	    !fib_rule_port_inrange(&rule->sport_range, fl4->fl4_sport))
206 		return 0;
207 
208 	if (fib_rule_port_range_set(&rule->dport_range) &&
209 	    !fib_rule_port_inrange(&rule->dport_range, fl4->fl4_dport))
210 		return 0;
211 
212 	return 1;
213 }
214 
215 static struct fib_table *fib_empty_table(struct net *net)
216 {
217 	u32 id = 1;
218 
219 	while (1) {
220 		if (!fib_get_table(net, id))
221 			return fib_new_table(net, id);
222 
223 		if (id++ == RT_TABLE_MAX)
224 			break;
225 	}
226 	return NULL;
227 }
228 
229 static int fib4_nl2rule_dscp(const struct nlattr *nla, struct fib4_rule *rule4,
230 			     struct netlink_ext_ack *extack)
231 {
232 	if (rule4->dscp) {
233 		NL_SET_ERR_MSG(extack, "Cannot specify both TOS and DSCP");
234 		return -EINVAL;
235 	}
236 
237 	rule4->dscp = inet_dsfield_to_dscp(nla_get_u8(nla) << 2);
238 	rule4->dscp_full = true;
239 
240 	return 0;
241 }
242 
243 static int fib4_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
244 			       struct fib_rule_hdr *frh,
245 			       struct nlattr **tb,
246 			       struct netlink_ext_ack *extack)
247 {
248 	struct net *net = sock_net(skb->sk);
249 	int err = -EINVAL;
250 	struct fib4_rule *rule4 = (struct fib4_rule *) rule;
251 
252 	if (!inet_validate_dscp(frh->tos)) {
253 		NL_SET_ERR_MSG(extack,
254 			       "Invalid dsfield (tos): ECN bits must be 0");
255 		goto errout;
256 	}
257 	/* IPv4 currently doesn't handle high order DSCP bits correctly */
258 	if (frh->tos & ~IPTOS_TOS_MASK) {
259 		NL_SET_ERR_MSG(extack, "Invalid tos");
260 		goto errout;
261 	}
262 	rule4->dscp = inet_dsfield_to_dscp(frh->tos);
263 
264 	if (tb[FRA_DSCP] &&
265 	    fib4_nl2rule_dscp(tb[FRA_DSCP], rule4, extack) < 0)
266 		goto errout;
267 
268 	/* split local/main if they are not already split */
269 	err = fib_unmerge(net);
270 	if (err)
271 		goto errout;
272 
273 	if (rule->table == RT_TABLE_UNSPEC && !rule->l3mdev) {
274 		if (rule->action == FR_ACT_TO_TBL) {
275 			struct fib_table *table;
276 
277 			table = fib_empty_table(net);
278 			if (!table) {
279 				err = -ENOBUFS;
280 				goto errout;
281 			}
282 
283 			rule->table = table->tb_id;
284 		}
285 	}
286 
287 	if (frh->src_len)
288 		rule4->src = nla_get_in_addr(tb[FRA_SRC]);
289 
290 	if (frh->dst_len)
291 		rule4->dst = nla_get_in_addr(tb[FRA_DST]);
292 
293 #ifdef CONFIG_IP_ROUTE_CLASSID
294 	if (tb[FRA_FLOW]) {
295 		rule4->tclassid = nla_get_u32(tb[FRA_FLOW]);
296 		if (rule4->tclassid)
297 			atomic_inc(&net->ipv4.fib_num_tclassid_users);
298 	}
299 #endif
300 
301 	if (fib_rule_requires_fldissect(rule))
302 		net->ipv4.fib_rules_require_fldissect++;
303 
304 	rule4->src_len = frh->src_len;
305 	rule4->srcmask = inet_make_mask(rule4->src_len);
306 	rule4->dst_len = frh->dst_len;
307 	rule4->dstmask = inet_make_mask(rule4->dst_len);
308 
309 	net->ipv4.fib_has_custom_rules = true;
310 
311 	err = 0;
312 errout:
313 	return err;
314 }
315 
316 static int fib4_rule_delete(struct fib_rule *rule)
317 {
318 	struct net *net = rule->fr_net;
319 	int err;
320 
321 	/* split local/main if they are not already split */
322 	err = fib_unmerge(net);
323 	if (err)
324 		goto errout;
325 
326 #ifdef CONFIG_IP_ROUTE_CLASSID
327 	if (((struct fib4_rule *)rule)->tclassid)
328 		atomic_dec(&net->ipv4.fib_num_tclassid_users);
329 #endif
330 	net->ipv4.fib_has_custom_rules = true;
331 
332 	if (net->ipv4.fib_rules_require_fldissect &&
333 	    fib_rule_requires_fldissect(rule))
334 		net->ipv4.fib_rules_require_fldissect--;
335 errout:
336 	return err;
337 }
338 
339 static int fib4_rule_compare(struct fib_rule *rule, struct fib_rule_hdr *frh,
340 			     struct nlattr **tb)
341 {
342 	struct fib4_rule *rule4 = (struct fib4_rule *) rule;
343 
344 	if (frh->src_len && (rule4->src_len != frh->src_len))
345 		return 0;
346 
347 	if (frh->dst_len && (rule4->dst_len != frh->dst_len))
348 		return 0;
349 
350 	if (frh->tos &&
351 	    (rule4->dscp_full ||
352 	     inet_dscp_to_dsfield(rule4->dscp) != frh->tos))
353 		return 0;
354 
355 	if (tb[FRA_DSCP]) {
356 		dscp_t dscp;
357 
358 		dscp = inet_dsfield_to_dscp(nla_get_u8(tb[FRA_DSCP]) << 2);
359 		if (!rule4->dscp_full || rule4->dscp != dscp)
360 			return 0;
361 	}
362 
363 #ifdef CONFIG_IP_ROUTE_CLASSID
364 	if (tb[FRA_FLOW] && (rule4->tclassid != nla_get_u32(tb[FRA_FLOW])))
365 		return 0;
366 #endif
367 
368 	if (frh->src_len && (rule4->src != nla_get_in_addr(tb[FRA_SRC])))
369 		return 0;
370 
371 	if (frh->dst_len && (rule4->dst != nla_get_in_addr(tb[FRA_DST])))
372 		return 0;
373 
374 	return 1;
375 }
376 
377 static int fib4_rule_fill(struct fib_rule *rule, struct sk_buff *skb,
378 			  struct fib_rule_hdr *frh)
379 {
380 	struct fib4_rule *rule4 = (struct fib4_rule *) rule;
381 
382 	frh->dst_len = rule4->dst_len;
383 	frh->src_len = rule4->src_len;
384 
385 	if (rule4->dscp_full) {
386 		frh->tos = 0;
387 		if (nla_put_u8(skb, FRA_DSCP,
388 			       inet_dscp_to_dsfield(rule4->dscp) >> 2))
389 			goto nla_put_failure;
390 	} else {
391 		frh->tos = inet_dscp_to_dsfield(rule4->dscp);
392 	}
393 
394 	if ((rule4->dst_len &&
395 	     nla_put_in_addr(skb, FRA_DST, rule4->dst)) ||
396 	    (rule4->src_len &&
397 	     nla_put_in_addr(skb, FRA_SRC, rule4->src)))
398 		goto nla_put_failure;
399 #ifdef CONFIG_IP_ROUTE_CLASSID
400 	if (rule4->tclassid &&
401 	    nla_put_u32(skb, FRA_FLOW, rule4->tclassid))
402 		goto nla_put_failure;
403 #endif
404 	return 0;
405 
406 nla_put_failure:
407 	return -ENOBUFS;
408 }
409 
410 static size_t fib4_rule_nlmsg_payload(struct fib_rule *rule)
411 {
412 	return nla_total_size(4) /* dst */
413 	       + nla_total_size(4) /* src */
414 	       + nla_total_size(4) /* flow */
415 	       + nla_total_size(1); /* dscp */
416 }
417 
418 static void fib4_rule_flush_cache(struct fib_rules_ops *ops)
419 {
420 	rt_cache_flush(ops->fro_net);
421 }
422 
423 static const struct fib_rules_ops __net_initconst fib4_rules_ops_template = {
424 	.family		= AF_INET,
425 	.rule_size	= sizeof(struct fib4_rule),
426 	.addr_size	= sizeof(u32),
427 	.action		= fib4_rule_action,
428 	.suppress	= fib4_rule_suppress,
429 	.match		= fib4_rule_match,
430 	.configure	= fib4_rule_configure,
431 	.delete		= fib4_rule_delete,
432 	.compare	= fib4_rule_compare,
433 	.fill		= fib4_rule_fill,
434 	.nlmsg_payload	= fib4_rule_nlmsg_payload,
435 	.flush_cache	= fib4_rule_flush_cache,
436 	.nlgroup	= RTNLGRP_IPV4_RULE,
437 	.owner		= THIS_MODULE,
438 };
439 
440 static int fib_default_rules_init(struct fib_rules_ops *ops)
441 {
442 	int err;
443 
444 	err = fib_default_rule_add(ops, 0, RT_TABLE_LOCAL);
445 	if (err < 0)
446 		return err;
447 	err = fib_default_rule_add(ops, 0x7FFE, RT_TABLE_MAIN);
448 	if (err < 0)
449 		return err;
450 	err = fib_default_rule_add(ops, 0x7FFF, RT_TABLE_DEFAULT);
451 	if (err < 0)
452 		return err;
453 	return 0;
454 }
455 
456 int __net_init fib4_rules_init(struct net *net)
457 {
458 	int err;
459 	struct fib_rules_ops *ops;
460 
461 	ops = fib_rules_register(&fib4_rules_ops_template, net);
462 	if (IS_ERR(ops))
463 		return PTR_ERR(ops);
464 
465 	err = fib_default_rules_init(ops);
466 	if (err < 0)
467 		goto fail;
468 	net->ipv4.rules_ops = ops;
469 	net->ipv4.fib_has_custom_rules = false;
470 	net->ipv4.fib_rules_require_fldissect = 0;
471 	return 0;
472 
473 fail:
474 	/* also cleans all rules already added */
475 	fib_rules_unregister(ops);
476 	return err;
477 }
478 
479 void __net_exit fib4_rules_exit(struct net *net)
480 {
481 	fib_rules_unregister(net->ipv4.rules_ops);
482 }
483