xref: /linux/include/net/ip_fib.h (revision 9a379e77033f02c4a071891afdf0f0a01eff8ccb)
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 	struct rcu_head		rcu;
126 	struct fib_nh		fib_nh[0];
127 #define fib_dev		fib_nh[0].nh_dev
128 };
129 
130 
131 #ifdef CONFIG_IP_MULTIPLE_TABLES
132 struct fib_rule;
133 #endif
134 
135 struct fib_table;
136 struct fib_result {
137 	__be32		prefix;
138 	unsigned char	prefixlen;
139 	unsigned char	nh_sel;
140 	unsigned char	type;
141 	unsigned char	scope;
142 	u32		tclassid;
143 	struct fib_info *fi;
144 	struct fib_table *table;
145 	struct hlist_head *fa_head;
146 };
147 
148 struct fib_result_nl {
149 	__be32		fl_addr;   /* To be looked up*/
150 	u32		fl_mark;
151 	unsigned char	fl_tos;
152 	unsigned char   fl_scope;
153 	unsigned char   tb_id_in;
154 
155 	unsigned char   tb_id;      /* Results */
156 	unsigned char	prefixlen;
157 	unsigned char	nh_sel;
158 	unsigned char	type;
159 	unsigned char	scope;
160 	int             err;
161 };
162 
163 #ifdef CONFIG_IP_ROUTE_MULTIPATH
164 #define FIB_RES_NH(res)		((res).fi->fib_nh[(res).nh_sel])
165 #else /* CONFIG_IP_ROUTE_MULTIPATH */
166 #define FIB_RES_NH(res)		((res).fi->fib_nh[0])
167 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
168 
169 #ifdef CONFIG_IP_MULTIPLE_TABLES
170 #define FIB_TABLE_HASHSZ 256
171 #else
172 #define FIB_TABLE_HASHSZ 2
173 #endif
174 
175 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh);
176 
177 #define FIB_RES_SADDR(net, res)				\
178 	((FIB_RES_NH(res).nh_saddr_genid ==		\
179 	  atomic_read(&(net)->ipv4.dev_addr_genid)) ?	\
180 	 FIB_RES_NH(res).nh_saddr :			\
181 	 fib_info_update_nh_saddr((net), &FIB_RES_NH(res)))
182 #define FIB_RES_GW(res)			(FIB_RES_NH(res).nh_gw)
183 #define FIB_RES_DEV(res)		(FIB_RES_NH(res).nh_dev)
184 #define FIB_RES_OIF(res)		(FIB_RES_NH(res).nh_oif)
185 
186 #define FIB_RES_PREFSRC(net, res)	((res).fi->fib_prefsrc ? : \
187 					 FIB_RES_SADDR(net, res))
188 
189 struct fib_entry_notifier_info {
190 	struct fib_notifier_info info; /* must be first */
191 	u32 dst;
192 	int dst_len;
193 	struct fib_info *fi;
194 	u8 tos;
195 	u8 type;
196 	u32 tb_id;
197 };
198 
199 struct fib_nh_notifier_info {
200 	struct fib_notifier_info info; /* must be first */
201 	struct fib_nh *fib_nh;
202 };
203 
204 int call_fib4_notifier(struct notifier_block *nb, struct net *net,
205 		       enum fib_event_type event_type,
206 		       struct fib_notifier_info *info);
207 int call_fib4_notifiers(struct net *net, enum fib_event_type event_type,
208 			struct fib_notifier_info *info);
209 
210 int __net_init fib4_notifier_init(struct net *net);
211 void __net_exit fib4_notifier_exit(struct net *net);
212 
213 void fib_notify(struct net *net, struct notifier_block *nb);
214 
215 struct fib_table {
216 	struct hlist_node	tb_hlist;
217 	u32			tb_id;
218 	int			tb_num_default;
219 	struct rcu_head		rcu;
220 	unsigned long 		*tb_data;
221 	unsigned long		__data[0];
222 };
223 
224 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
225 		     struct fib_result *res, int fib_flags);
226 int fib_table_insert(struct net *, struct fib_table *, struct fib_config *,
227 		     struct netlink_ext_ack *extack);
228 int fib_table_delete(struct net *, struct fib_table *, struct fib_config *,
229 		     struct netlink_ext_ack *extack);
230 int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
231 		   struct netlink_callback *cb);
232 int fib_table_flush(struct net *net, struct fib_table *table);
233 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
234 void fib_table_flush_external(struct fib_table *table);
235 void fib_free_table(struct fib_table *tb);
236 
237 #ifndef CONFIG_IP_MULTIPLE_TABLES
238 
239 #define TABLE_LOCAL_INDEX	(RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
240 #define TABLE_MAIN_INDEX	(RT_TABLE_MAIN  & (FIB_TABLE_HASHSZ - 1))
241 
242 static inline struct fib_table *fib_get_table(struct net *net, u32 id)
243 {
244 	struct hlist_node *tb_hlist;
245 	struct hlist_head *ptr;
246 
247 	ptr = id == RT_TABLE_LOCAL ?
248 		&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
249 		&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
250 
251 	tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
252 
253 	return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
254 }
255 
256 static inline struct fib_table *fib_new_table(struct net *net, u32 id)
257 {
258 	return fib_get_table(net, id);
259 }
260 
261 static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
262 			     struct fib_result *res, unsigned int flags)
263 {
264 	struct fib_table *tb;
265 	int err = -ENETUNREACH;
266 
267 	rcu_read_lock();
268 
269 	tb = fib_get_table(net, RT_TABLE_MAIN);
270 	if (tb)
271 		err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
272 
273 	if (err == -EAGAIN)
274 		err = -ENETUNREACH;
275 
276 	rcu_read_unlock();
277 
278 	return err;
279 }
280 
281 static inline bool fib4_rule_default(const struct fib_rule *rule)
282 {
283 	return true;
284 }
285 
286 static inline int fib4_rules_dump(struct net *net, struct notifier_block *nb)
287 {
288 	return 0;
289 }
290 
291 static inline unsigned int fib4_rules_seq_read(struct net *net)
292 {
293 	return 0;
294 }
295 
296 #else /* CONFIG_IP_MULTIPLE_TABLES */
297 int __net_init fib4_rules_init(struct net *net);
298 void __net_exit fib4_rules_exit(struct net *net);
299 
300 struct fib_table *fib_new_table(struct net *net, u32 id);
301 struct fib_table *fib_get_table(struct net *net, u32 id);
302 
303 int __fib_lookup(struct net *net, struct flowi4 *flp,
304 		 struct fib_result *res, unsigned int flags);
305 
306 static inline int fib_lookup(struct net *net, struct flowi4 *flp,
307 			     struct fib_result *res, unsigned int flags)
308 {
309 	struct fib_table *tb;
310 	int err = -ENETUNREACH;
311 
312 	flags |= FIB_LOOKUP_NOREF;
313 	if (net->ipv4.fib_has_custom_rules)
314 		return __fib_lookup(net, flp, res, flags);
315 
316 	rcu_read_lock();
317 
318 	res->tclassid = 0;
319 
320 	tb = rcu_dereference_rtnl(net->ipv4.fib_main);
321 	if (tb)
322 		err = fib_table_lookup(tb, flp, res, flags);
323 
324 	if (!err)
325 		goto out;
326 
327 	tb = rcu_dereference_rtnl(net->ipv4.fib_default);
328 	if (tb)
329 		err = fib_table_lookup(tb, flp, res, flags);
330 
331 out:
332 	if (err == -EAGAIN)
333 		err = -ENETUNREACH;
334 
335 	rcu_read_unlock();
336 
337 	return err;
338 }
339 
340 bool fib4_rule_default(const struct fib_rule *rule);
341 int fib4_rules_dump(struct net *net, struct notifier_block *nb);
342 unsigned int fib4_rules_seq_read(struct net *net);
343 
344 #endif /* CONFIG_IP_MULTIPLE_TABLES */
345 
346 /* Exported by fib_frontend.c */
347 extern const struct nla_policy rtm_ipv4_policy[];
348 void ip_fib_init(void);
349 __be32 fib_compute_spec_dst(struct sk_buff *skb);
350 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
351 			u8 tos, int oif, struct net_device *dev,
352 			struct in_device *idev, u32 *itag);
353 #ifdef CONFIG_IP_ROUTE_CLASSID
354 static inline int fib_num_tclassid_users(struct net *net)
355 {
356 	return net->ipv4.fib_num_tclassid_users;
357 }
358 #else
359 static inline int fib_num_tclassid_users(struct net *net)
360 {
361 	return 0;
362 }
363 #endif
364 int fib_unmerge(struct net *net);
365 
366 /* Exported by fib_semantics.c */
367 int ip_fib_check_default(__be32 gw, struct net_device *dev);
368 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
369 int fib_sync_down_addr(struct net_device *dev, __be32 local);
370 int fib_sync_up(struct net_device *dev, unsigned int nh_flags);
371 
372 #ifdef CONFIG_IP_ROUTE_MULTIPATH
373 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
374 		       const struct sk_buff *skb);
375 #endif
376 void fib_select_multipath(struct fib_result *res, int hash);
377 void fib_select_path(struct net *net, struct fib_result *res,
378 		     struct flowi4 *fl4, const struct sk_buff *skb);
379 
380 /* Exported by fib_trie.c */
381 void fib_trie_init(void);
382 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
383 
384 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
385 {
386 #ifdef CONFIG_IP_ROUTE_CLASSID
387 #ifdef CONFIG_IP_MULTIPLE_TABLES
388 	u32 rtag;
389 #endif
390 	*itag = FIB_RES_NH(*res).nh_tclassid<<16;
391 #ifdef CONFIG_IP_MULTIPLE_TABLES
392 	rtag = res->tclassid;
393 	if (*itag == 0)
394 		*itag = (rtag<<16);
395 	*itag |= (rtag>>16);
396 #endif
397 #endif
398 }
399 
400 void free_fib_info(struct fib_info *fi);
401 
402 static inline void fib_info_hold(struct fib_info *fi)
403 {
404 	refcount_inc(&fi->fib_clntref);
405 }
406 
407 static inline void fib_info_put(struct fib_info *fi)
408 {
409 	if (refcount_dec_and_test(&fi->fib_clntref))
410 		free_fib_info(fi);
411 }
412 
413 #ifdef CONFIG_PROC_FS
414 int __net_init fib_proc_init(struct net *net);
415 void __net_exit fib_proc_exit(struct net *net);
416 #else
417 static inline int fib_proc_init(struct net *net)
418 {
419 	return 0;
420 }
421 static inline void fib_proc_exit(struct net *net)
422 {
423 }
424 #endif
425 
426 #endif  /* _NET_FIB_H */
427