xref: /linux/include/net/ip_fib.h (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET  is implemented using the  BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		Definitions for the Forwarding Information Base.
7  *
8  * Authors:	A.N.Kuznetsov, <kuznet@ms2.inr.ac.ru>
9  *
10  *		This program is free software; you can redistribute it and/or
11  *		modify it under the terms of the GNU General Public License
12  *		as published by the Free Software Foundation; either version
13  *		2 of the License, or (at your option) any later version.
14  */
15 
16 #ifndef _NET_IP_FIB_H
17 #define _NET_IP_FIB_H
18 
19 #include <net/flow.h>
20 #include <linux/seq_file.h>
21 #include <linux/rcupdate.h>
22 #include <net/fib_notifier.h>
23 #include <net/fib_rules.h>
24 #include <net/inetpeer.h>
25 #include <linux/percpu.h>
26 #include <linux/notifier.h>
27 #include <linux/refcount.h>
28 
29 struct fib_config {
30 	u8			fc_dst_len;
31 	u8			fc_tos;
32 	u8			fc_protocol;
33 	u8			fc_scope;
34 	u8			fc_type;
35 	/* 3 bytes unused */
36 	u32			fc_table;
37 	__be32			fc_dst;
38 	__be32			fc_gw;
39 	int			fc_oif;
40 	u32			fc_flags;
41 	u32			fc_priority;
42 	__be32			fc_prefsrc;
43 	struct nlattr		*fc_mx;
44 	struct rtnexthop	*fc_mp;
45 	int			fc_mx_len;
46 	int			fc_mp_len;
47 	u32			fc_flow;
48 	u32			fc_nlflags;
49 	struct nl_info		fc_nlinfo;
50 	struct nlattr		*fc_encap;
51 	u16			fc_encap_type;
52 };
53 
54 struct fib_info;
55 struct rtable;
56 
57 struct fib_nh_exception {
58 	struct fib_nh_exception __rcu	*fnhe_next;
59 	int				fnhe_genid;
60 	__be32				fnhe_daddr;
61 	u32				fnhe_pmtu;
62 	__be32				fnhe_gw;
63 	unsigned long			fnhe_expires;
64 	struct rtable __rcu		*fnhe_rth_input;
65 	struct rtable __rcu		*fnhe_rth_output;
66 	unsigned long			fnhe_stamp;
67 	struct rcu_head			rcu;
68 };
69 
70 struct fnhe_hash_bucket {
71 	struct fib_nh_exception __rcu	*chain;
72 };
73 
74 #define FNHE_HASH_SHIFT		11
75 #define FNHE_HASH_SIZE		(1 << FNHE_HASH_SHIFT)
76 #define FNHE_RECLAIM_DEPTH	5
77 
78 struct fib_nh {
79 	struct net_device	*nh_dev;
80 	struct hlist_node	nh_hash;
81 	struct fib_info		*nh_parent;
82 	unsigned int		nh_flags;
83 	unsigned char		nh_scope;
84 #ifdef CONFIG_IP_ROUTE_MULTIPATH
85 	int			nh_weight;
86 	atomic_t		nh_upper_bound;
87 #endif
88 #ifdef CONFIG_IP_ROUTE_CLASSID
89 	__u32			nh_tclassid;
90 #endif
91 	int			nh_oif;
92 	__be32			nh_gw;
93 	__be32			nh_saddr;
94 	int			nh_saddr_genid;
95 	struct rtable __rcu * __percpu *nh_pcpu_rth_output;
96 	struct rtable __rcu	*nh_rth_input;
97 	struct fnhe_hash_bucket	__rcu *nh_exceptions;
98 	struct lwtunnel_state	*nh_lwtstate;
99 };
100 
101 /*
102  * This structure contains data shared by many of routes.
103  */
104 
105 struct fib_info {
106 	struct hlist_node	fib_hash;
107 	struct hlist_node	fib_lhash;
108 	struct net		*fib_net;
109 	int			fib_treeref;
110 	refcount_t		fib_clntref;
111 	unsigned int		fib_flags;
112 	unsigned char		fib_dead;
113 	unsigned char		fib_protocol;
114 	unsigned char		fib_scope;
115 	unsigned char		fib_type;
116 	__be32			fib_prefsrc;
117 	u32			fib_tb_id;
118 	u32			fib_priority;
119 	struct dst_metrics	*fib_metrics;
120 #define fib_mtu fib_metrics->metrics[RTAX_MTU-1]
121 #define fib_window fib_metrics->metrics[RTAX_WINDOW-1]
122 #define fib_rtt fib_metrics->metrics[RTAX_RTT-1]
123 #define fib_advmss fib_metrics->metrics[RTAX_ADVMSS-1]
124 	int			fib_nhs;
125 #ifdef CONFIG_IP_ROUTE_MULTIPATH
126 	int			fib_weight;
127 #endif
128 	struct rcu_head		rcu;
129 	struct fib_nh		fib_nh[0];
130 #define fib_dev		fib_nh[0].nh_dev
131 };
132 
133 
134 #ifdef CONFIG_IP_MULTIPLE_TABLES
135 struct fib_rule;
136 #endif
137 
138 struct fib_table;
139 struct fib_result {
140 	__be32		prefix;
141 	unsigned char	prefixlen;
142 	unsigned char	nh_sel;
143 	unsigned char	type;
144 	unsigned char	scope;
145 	u32		tclassid;
146 	struct fib_info *fi;
147 	struct fib_table *table;
148 	struct hlist_head *fa_head;
149 };
150 
151 struct fib_result_nl {
152 	__be32		fl_addr;   /* To be looked up*/
153 	u32		fl_mark;
154 	unsigned char	fl_tos;
155 	unsigned char   fl_scope;
156 	unsigned char   tb_id_in;
157 
158 	unsigned char   tb_id;      /* Results */
159 	unsigned char	prefixlen;
160 	unsigned char	nh_sel;
161 	unsigned char	type;
162 	unsigned char	scope;
163 	int             err;
164 };
165 
166 #ifdef CONFIG_IP_ROUTE_MULTIPATH
167 #define FIB_RES_NH(res)		((res).fi->fib_nh[(res).nh_sel])
168 #else /* CONFIG_IP_ROUTE_MULTIPATH */
169 #define FIB_RES_NH(res)		((res).fi->fib_nh[0])
170 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
171 
172 #ifdef CONFIG_IP_MULTIPLE_TABLES
173 #define FIB_TABLE_HASHSZ 256
174 #else
175 #define FIB_TABLE_HASHSZ 2
176 #endif
177 
178 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh);
179 
180 #define FIB_RES_SADDR(net, res)				\
181 	((FIB_RES_NH(res).nh_saddr_genid ==		\
182 	  atomic_read(&(net)->ipv4.dev_addr_genid)) ?	\
183 	 FIB_RES_NH(res).nh_saddr :			\
184 	 fib_info_update_nh_saddr((net), &FIB_RES_NH(res)))
185 #define FIB_RES_GW(res)			(FIB_RES_NH(res).nh_gw)
186 #define FIB_RES_DEV(res)		(FIB_RES_NH(res).nh_dev)
187 #define FIB_RES_OIF(res)		(FIB_RES_NH(res).nh_oif)
188 
189 #define FIB_RES_PREFSRC(net, res)	((res).fi->fib_prefsrc ? : \
190 					 FIB_RES_SADDR(net, res))
191 
192 struct fib_entry_notifier_info {
193 	struct fib_notifier_info info; /* must be first */
194 	u32 dst;
195 	int dst_len;
196 	struct fib_info *fi;
197 	u8 tos;
198 	u8 type;
199 	u32 tb_id;
200 };
201 
202 struct fib_nh_notifier_info {
203 	struct fib_notifier_info info; /* must be first */
204 	struct fib_nh *fib_nh;
205 };
206 
207 int call_fib4_notifier(struct notifier_block *nb, struct net *net,
208 		       enum fib_event_type event_type,
209 		       struct fib_notifier_info *info);
210 int call_fib4_notifiers(struct net *net, enum fib_event_type event_type,
211 			struct fib_notifier_info *info);
212 
213 int __net_init fib4_notifier_init(struct net *net);
214 void __net_exit fib4_notifier_exit(struct net *net);
215 
216 void fib_notify(struct net *net, struct notifier_block *nb);
217 
218 struct fib_table {
219 	struct hlist_node	tb_hlist;
220 	u32			tb_id;
221 	int			tb_num_default;
222 	struct rcu_head		rcu;
223 	unsigned long 		*tb_data;
224 	unsigned long		__data[0];
225 };
226 
227 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
228 		     struct fib_result *res, int fib_flags);
229 int fib_table_insert(struct net *, struct fib_table *, struct fib_config *,
230 		     struct netlink_ext_ack *extack);
231 int fib_table_delete(struct net *, struct fib_table *, struct fib_config *,
232 		     struct netlink_ext_ack *extack);
233 int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
234 		   struct netlink_callback *cb);
235 int fib_table_flush(struct net *net, struct fib_table *table);
236 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
237 void fib_table_flush_external(struct fib_table *table);
238 void fib_free_table(struct fib_table *tb);
239 
240 #ifndef CONFIG_IP_MULTIPLE_TABLES
241 
242 #define TABLE_LOCAL_INDEX	(RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
243 #define TABLE_MAIN_INDEX	(RT_TABLE_MAIN  & (FIB_TABLE_HASHSZ - 1))
244 
245 static inline struct fib_table *fib_get_table(struct net *net, u32 id)
246 {
247 	struct hlist_node *tb_hlist;
248 	struct hlist_head *ptr;
249 
250 	ptr = id == RT_TABLE_LOCAL ?
251 		&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
252 		&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
253 
254 	tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
255 
256 	return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
257 }
258 
259 static inline struct fib_table *fib_new_table(struct net *net, u32 id)
260 {
261 	return fib_get_table(net, id);
262 }
263 
264 static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
265 			     struct fib_result *res, unsigned int flags)
266 {
267 	struct fib_table *tb;
268 	int err = -ENETUNREACH;
269 
270 	rcu_read_lock();
271 
272 	tb = fib_get_table(net, RT_TABLE_MAIN);
273 	if (tb)
274 		err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
275 
276 	if (err == -EAGAIN)
277 		err = -ENETUNREACH;
278 
279 	rcu_read_unlock();
280 
281 	return err;
282 }
283 
284 static inline bool fib4_rule_default(const struct fib_rule *rule)
285 {
286 	return true;
287 }
288 
289 static inline int fib4_rules_dump(struct net *net, struct notifier_block *nb)
290 {
291 	return 0;
292 }
293 
294 static inline unsigned int fib4_rules_seq_read(struct net *net)
295 {
296 	return 0;
297 }
298 
299 #else /* CONFIG_IP_MULTIPLE_TABLES */
300 int __net_init fib4_rules_init(struct net *net);
301 void __net_exit fib4_rules_exit(struct net *net);
302 
303 struct fib_table *fib_new_table(struct net *net, u32 id);
304 struct fib_table *fib_get_table(struct net *net, u32 id);
305 
306 int __fib_lookup(struct net *net, struct flowi4 *flp,
307 		 struct fib_result *res, unsigned int flags);
308 
309 static inline int fib_lookup(struct net *net, struct flowi4 *flp,
310 			     struct fib_result *res, unsigned int flags)
311 {
312 	struct fib_table *tb;
313 	int err = -ENETUNREACH;
314 
315 	flags |= FIB_LOOKUP_NOREF;
316 	if (net->ipv4.fib_has_custom_rules)
317 		return __fib_lookup(net, flp, res, flags);
318 
319 	rcu_read_lock();
320 
321 	res->tclassid = 0;
322 
323 	tb = rcu_dereference_rtnl(net->ipv4.fib_main);
324 	if (tb)
325 		err = fib_table_lookup(tb, flp, res, flags);
326 
327 	if (!err)
328 		goto out;
329 
330 	tb = rcu_dereference_rtnl(net->ipv4.fib_default);
331 	if (tb)
332 		err = fib_table_lookup(tb, flp, res, flags);
333 
334 out:
335 	if (err == -EAGAIN)
336 		err = -ENETUNREACH;
337 
338 	rcu_read_unlock();
339 
340 	return err;
341 }
342 
343 bool fib4_rule_default(const struct fib_rule *rule);
344 int fib4_rules_dump(struct net *net, struct notifier_block *nb);
345 unsigned int fib4_rules_seq_read(struct net *net);
346 
347 #endif /* CONFIG_IP_MULTIPLE_TABLES */
348 
349 /* Exported by fib_frontend.c */
350 extern const struct nla_policy rtm_ipv4_policy[];
351 void ip_fib_init(void);
352 __be32 fib_compute_spec_dst(struct sk_buff *skb);
353 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
354 			u8 tos, int oif, struct net_device *dev,
355 			struct in_device *idev, u32 *itag);
356 #ifdef CONFIG_IP_ROUTE_CLASSID
357 static inline int fib_num_tclassid_users(struct net *net)
358 {
359 	return net->ipv4.fib_num_tclassid_users;
360 }
361 #else
362 static inline int fib_num_tclassid_users(struct net *net)
363 {
364 	return 0;
365 }
366 #endif
367 int fib_unmerge(struct net *net);
368 
369 /* Exported by fib_semantics.c */
370 int ip_fib_check_default(__be32 gw, struct net_device *dev);
371 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
372 int fib_sync_down_addr(struct net_device *dev, __be32 local);
373 int fib_sync_up(struct net_device *dev, unsigned int nh_flags);
374 
375 #ifdef CONFIG_IP_ROUTE_MULTIPATH
376 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
377 		       const struct sk_buff *skb);
378 #endif
379 void fib_select_multipath(struct fib_result *res, int hash);
380 void fib_select_path(struct net *net, struct fib_result *res,
381 		     struct flowi4 *fl4, const struct sk_buff *skb);
382 
383 /* Exported by fib_trie.c */
384 void fib_trie_init(void);
385 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
386 
387 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
388 {
389 #ifdef CONFIG_IP_ROUTE_CLASSID
390 #ifdef CONFIG_IP_MULTIPLE_TABLES
391 	u32 rtag;
392 #endif
393 	*itag = FIB_RES_NH(*res).nh_tclassid<<16;
394 #ifdef CONFIG_IP_MULTIPLE_TABLES
395 	rtag = res->tclassid;
396 	if (*itag == 0)
397 		*itag = (rtag<<16);
398 	*itag |= (rtag>>16);
399 #endif
400 #endif
401 }
402 
403 void free_fib_info(struct fib_info *fi);
404 
405 static inline void fib_info_hold(struct fib_info *fi)
406 {
407 	refcount_inc(&fi->fib_clntref);
408 }
409 
410 static inline void fib_info_put(struct fib_info *fi)
411 {
412 	if (refcount_dec_and_test(&fi->fib_clntref))
413 		free_fib_info(fi);
414 }
415 
416 #ifdef CONFIG_PROC_FS
417 int __net_init fib_proc_init(struct net *net);
418 void __net_exit fib_proc_exit(struct net *net);
419 #else
420 static inline int fib_proc_init(struct net *net)
421 {
422 	return 0;
423 }
424 static inline void fib_proc_exit(struct net *net)
425 {
426 }
427 #endif
428 
429 #endif  /* _NET_FIB_H */
430