xref: /openbmc/linux/include/net/ip_fib.h (revision 9b93eb47)
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 	u8			fc_gw_family;
36 	/* 2 bytes unused */
37 	u32			fc_table;
38 	__be32			fc_dst;
39 	union {
40 		__be32		fc_gw4;
41 		struct in6_addr	fc_gw6;
42 	};
43 	int			fc_oif;
44 	u32			fc_flags;
45 	u32			fc_priority;
46 	__be32			fc_prefsrc;
47 	struct nlattr		*fc_mx;
48 	struct rtnexthop	*fc_mp;
49 	int			fc_mx_len;
50 	int			fc_mp_len;
51 	u32			fc_flow;
52 	u32			fc_nlflags;
53 	struct nl_info		fc_nlinfo;
54 	struct nlattr		*fc_encap;
55 	u16			fc_encap_type;
56 };
57 
58 struct fib_info;
59 struct rtable;
60 
61 struct fib_nh_exception {
62 	struct fib_nh_exception __rcu	*fnhe_next;
63 	int				fnhe_genid;
64 	__be32				fnhe_daddr;
65 	u32				fnhe_pmtu;
66 	bool				fnhe_mtu_locked;
67 	__be32				fnhe_gw;
68 	unsigned long			fnhe_expires;
69 	struct rtable __rcu		*fnhe_rth_input;
70 	struct rtable __rcu		*fnhe_rth_output;
71 	unsigned long			fnhe_stamp;
72 	struct rcu_head			rcu;
73 };
74 
75 struct fnhe_hash_bucket {
76 	struct fib_nh_exception __rcu	*chain;
77 };
78 
79 #define FNHE_HASH_SHIFT		11
80 #define FNHE_HASH_SIZE		(1 << FNHE_HASH_SHIFT)
81 #define FNHE_RECLAIM_DEPTH	5
82 
83 struct fib_nh_common {
84 	struct net_device	*nhc_dev;
85 	int			nhc_oif;
86 	unsigned char		nhc_scope;
87 	u8			nhc_family;
88 	u8			nhc_gw_family;
89 	unsigned char		nhc_flags;
90 	struct lwtunnel_state	*nhc_lwtstate;
91 
92 	union {
93 		__be32          ipv4;
94 		struct in6_addr ipv6;
95 	} nhc_gw;
96 
97 	int			nhc_weight;
98 	atomic_t		nhc_upper_bound;
99 
100 	/* v4 specific, but allows fib6_nh with v4 routes */
101 	struct rtable __rcu * __percpu *nhc_pcpu_rth_output;
102 	struct rtable __rcu     *nhc_rth_input;
103 	struct fnhe_hash_bucket	__rcu *nhc_exceptions;
104 };
105 
106 struct fib_nh {
107 	struct fib_nh_common	nh_common;
108 	struct hlist_node	nh_hash;
109 	struct fib_info		*nh_parent;
110 #ifdef CONFIG_IP_ROUTE_CLASSID
111 	__u32			nh_tclassid;
112 #endif
113 	__be32			nh_saddr;
114 	int			nh_saddr_genid;
115 #define fib_nh_family		nh_common.nhc_family
116 #define fib_nh_dev		nh_common.nhc_dev
117 #define fib_nh_oif		nh_common.nhc_oif
118 #define fib_nh_flags		nh_common.nhc_flags
119 #define fib_nh_lws		nh_common.nhc_lwtstate
120 #define fib_nh_scope		nh_common.nhc_scope
121 #define fib_nh_gw_family	nh_common.nhc_gw_family
122 #define fib_nh_gw4		nh_common.nhc_gw.ipv4
123 #define fib_nh_gw6		nh_common.nhc_gw.ipv6
124 #define fib_nh_weight		nh_common.nhc_weight
125 #define fib_nh_upper_bound	nh_common.nhc_upper_bound
126 };
127 
128 /*
129  * This structure contains data shared by many of routes.
130  */
131 
132 struct fib_info {
133 	struct hlist_node	fib_hash;
134 	struct hlist_node	fib_lhash;
135 	struct net		*fib_net;
136 	int			fib_treeref;
137 	refcount_t		fib_clntref;
138 	unsigned int		fib_flags;
139 	unsigned char		fib_dead;
140 	unsigned char		fib_protocol;
141 	unsigned char		fib_scope;
142 	unsigned char		fib_type;
143 	__be32			fib_prefsrc;
144 	u32			fib_tb_id;
145 	u32			fib_priority;
146 	struct dst_metrics	*fib_metrics;
147 #define fib_mtu fib_metrics->metrics[RTAX_MTU-1]
148 #define fib_window fib_metrics->metrics[RTAX_WINDOW-1]
149 #define fib_rtt fib_metrics->metrics[RTAX_RTT-1]
150 #define fib_advmss fib_metrics->metrics[RTAX_ADVMSS-1]
151 	int			fib_nhs;
152 	bool			fib_nh_is_v6;
153 	struct rcu_head		rcu;
154 	struct fib_nh		fib_nh[0];
155 #define fib_dev		fib_nh[0].fib_nh_dev
156 };
157 
158 
159 #ifdef CONFIG_IP_MULTIPLE_TABLES
160 struct fib_rule;
161 #endif
162 
163 struct fib_table;
164 struct fib_result {
165 	__be32			prefix;
166 	unsigned char		prefixlen;
167 	unsigned char		nh_sel;
168 	unsigned char		type;
169 	unsigned char		scope;
170 	u32			tclassid;
171 	struct fib_nh_common	*nhc;
172 	struct fib_info		*fi;
173 	struct fib_table	*table;
174 	struct hlist_head	*fa_head;
175 };
176 
177 struct fib_result_nl {
178 	__be32		fl_addr;   /* To be looked up*/
179 	u32		fl_mark;
180 	unsigned char	fl_tos;
181 	unsigned char   fl_scope;
182 	unsigned char   tb_id_in;
183 
184 	unsigned char   tb_id;      /* Results */
185 	unsigned char	prefixlen;
186 	unsigned char	nh_sel;
187 	unsigned char	type;
188 	unsigned char	scope;
189 	int             err;
190 };
191 
192 static inline struct fib_nh_common *fib_info_nhc(struct fib_info *fi, int nhsel)
193 {
194 	return &fi->fib_nh[nhsel].nh_common;
195 }
196 
197 #ifdef CONFIG_IP_MULTIPLE_TABLES
198 #define FIB_TABLE_HASHSZ 256
199 #else
200 #define FIB_TABLE_HASHSZ 2
201 #endif
202 
203 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh);
204 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res);
205 
206 #define FIB_RES_NHC(res)		((res).nhc)
207 #define FIB_RES_DEV(res)	(FIB_RES_NHC(res)->nhc_dev)
208 #define FIB_RES_OIF(res)	(FIB_RES_NHC(res)->nhc_oif)
209 
210 struct fib_entry_notifier_info {
211 	struct fib_notifier_info info; /* must be first */
212 	u32 dst;
213 	int dst_len;
214 	struct fib_info *fi;
215 	u8 tos;
216 	u8 type;
217 	u32 tb_id;
218 };
219 
220 struct fib_nh_notifier_info {
221 	struct fib_notifier_info info; /* must be first */
222 	struct fib_nh *fib_nh;
223 };
224 
225 int call_fib4_notifier(struct notifier_block *nb, struct net *net,
226 		       enum fib_event_type event_type,
227 		       struct fib_notifier_info *info);
228 int call_fib4_notifiers(struct net *net, enum fib_event_type event_type,
229 			struct fib_notifier_info *info);
230 
231 int __net_init fib4_notifier_init(struct net *net);
232 void __net_exit fib4_notifier_exit(struct net *net);
233 
234 void fib_notify(struct net *net, struct notifier_block *nb);
235 
236 struct fib_table {
237 	struct hlist_node	tb_hlist;
238 	u32			tb_id;
239 	int			tb_num_default;
240 	struct rcu_head		rcu;
241 	unsigned long 		*tb_data;
242 	unsigned long		__data[0];
243 };
244 
245 struct fib_dump_filter {
246 	u32			table_id;
247 	/* filter_set is an optimization that an entry is set */
248 	bool			filter_set;
249 	bool			dump_all_families;
250 	unsigned char		protocol;
251 	unsigned char		rt_type;
252 	unsigned int		flags;
253 	struct net_device	*dev;
254 };
255 
256 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
257 		     struct fib_result *res, int fib_flags);
258 int fib_table_insert(struct net *, struct fib_table *, struct fib_config *,
259 		     struct netlink_ext_ack *extack);
260 int fib_table_delete(struct net *, struct fib_table *, struct fib_config *,
261 		     struct netlink_ext_ack *extack);
262 int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
263 		   struct netlink_callback *cb, struct fib_dump_filter *filter);
264 int fib_table_flush(struct net *net, struct fib_table *table, bool flush_all);
265 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
266 void fib_table_flush_external(struct fib_table *table);
267 void fib_free_table(struct fib_table *tb);
268 
269 #ifndef CONFIG_IP_MULTIPLE_TABLES
270 
271 #define TABLE_LOCAL_INDEX	(RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
272 #define TABLE_MAIN_INDEX	(RT_TABLE_MAIN  & (FIB_TABLE_HASHSZ - 1))
273 
274 static inline struct fib_table *fib_get_table(struct net *net, u32 id)
275 {
276 	struct hlist_node *tb_hlist;
277 	struct hlist_head *ptr;
278 
279 	ptr = id == RT_TABLE_LOCAL ?
280 		&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
281 		&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
282 
283 	tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
284 
285 	return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
286 }
287 
288 static inline struct fib_table *fib_new_table(struct net *net, u32 id)
289 {
290 	return fib_get_table(net, id);
291 }
292 
293 static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
294 			     struct fib_result *res, unsigned int flags)
295 {
296 	struct fib_table *tb;
297 	int err = -ENETUNREACH;
298 
299 	rcu_read_lock();
300 
301 	tb = fib_get_table(net, RT_TABLE_MAIN);
302 	if (tb)
303 		err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
304 
305 	if (err == -EAGAIN)
306 		err = -ENETUNREACH;
307 
308 	rcu_read_unlock();
309 
310 	return err;
311 }
312 
313 static inline bool fib4_rule_default(const struct fib_rule *rule)
314 {
315 	return true;
316 }
317 
318 static inline int fib4_rules_dump(struct net *net, struct notifier_block *nb)
319 {
320 	return 0;
321 }
322 
323 static inline unsigned int fib4_rules_seq_read(struct net *net)
324 {
325 	return 0;
326 }
327 
328 static inline bool fib4_rules_early_flow_dissect(struct net *net,
329 						 struct sk_buff *skb,
330 						 struct flowi4 *fl4,
331 						 struct flow_keys *flkeys)
332 {
333 	return false;
334 }
335 #else /* CONFIG_IP_MULTIPLE_TABLES */
336 int __net_init fib4_rules_init(struct net *net);
337 void __net_exit fib4_rules_exit(struct net *net);
338 
339 struct fib_table *fib_new_table(struct net *net, u32 id);
340 struct fib_table *fib_get_table(struct net *net, u32 id);
341 
342 int __fib_lookup(struct net *net, struct flowi4 *flp,
343 		 struct fib_result *res, unsigned int flags);
344 
345 static inline int fib_lookup(struct net *net, struct flowi4 *flp,
346 			     struct fib_result *res, unsigned int flags)
347 {
348 	struct fib_table *tb;
349 	int err = -ENETUNREACH;
350 
351 	flags |= FIB_LOOKUP_NOREF;
352 	if (net->ipv4.fib_has_custom_rules)
353 		return __fib_lookup(net, flp, res, flags);
354 
355 	rcu_read_lock();
356 
357 	res->tclassid = 0;
358 
359 	tb = rcu_dereference_rtnl(net->ipv4.fib_main);
360 	if (tb)
361 		err = fib_table_lookup(tb, flp, res, flags);
362 
363 	if (!err)
364 		goto out;
365 
366 	tb = rcu_dereference_rtnl(net->ipv4.fib_default);
367 	if (tb)
368 		err = fib_table_lookup(tb, flp, res, flags);
369 
370 out:
371 	if (err == -EAGAIN)
372 		err = -ENETUNREACH;
373 
374 	rcu_read_unlock();
375 
376 	return err;
377 }
378 
379 bool fib4_rule_default(const struct fib_rule *rule);
380 int fib4_rules_dump(struct net *net, struct notifier_block *nb);
381 unsigned int fib4_rules_seq_read(struct net *net);
382 
383 static inline bool fib4_rules_early_flow_dissect(struct net *net,
384 						 struct sk_buff *skb,
385 						 struct flowi4 *fl4,
386 						 struct flow_keys *flkeys)
387 {
388 	unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
389 
390 	if (!net->ipv4.fib_rules_require_fldissect)
391 		return false;
392 
393 	skb_flow_dissect_flow_keys(skb, flkeys, flag);
394 	fl4->fl4_sport = flkeys->ports.src;
395 	fl4->fl4_dport = flkeys->ports.dst;
396 	fl4->flowi4_proto = flkeys->basic.ip_proto;
397 
398 	return true;
399 }
400 
401 #endif /* CONFIG_IP_MULTIPLE_TABLES */
402 
403 /* Exported by fib_frontend.c */
404 extern const struct nla_policy rtm_ipv4_policy[];
405 void ip_fib_init(void);
406 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
407 		    struct netlink_ext_ack *extack);
408 __be32 fib_compute_spec_dst(struct sk_buff *skb);
409 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev);
410 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
411 			u8 tos, int oif, struct net_device *dev,
412 			struct in_device *idev, u32 *itag);
413 #ifdef CONFIG_IP_ROUTE_CLASSID
414 static inline int fib_num_tclassid_users(struct net *net)
415 {
416 	return net->ipv4.fib_num_tclassid_users;
417 }
418 #else
419 static inline int fib_num_tclassid_users(struct net *net)
420 {
421 	return 0;
422 }
423 #endif
424 int fib_unmerge(struct net *net);
425 
426 /* Exported by fib_semantics.c */
427 int ip_fib_check_default(__be32 gw, struct net_device *dev);
428 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
429 int fib_sync_down_addr(struct net_device *dev, __be32 local);
430 int fib_sync_up(struct net_device *dev, unsigned char nh_flags);
431 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu);
432 
433 #ifdef CONFIG_IP_ROUTE_MULTIPATH
434 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
435 		       const struct sk_buff *skb, struct flow_keys *flkeys);
436 #endif
437 void fib_select_multipath(struct fib_result *res, int hash);
438 void fib_select_path(struct net *net, struct fib_result *res,
439 		     struct flowi4 *fl4, const struct sk_buff *skb);
440 
441 int fib_nh_init(struct net *net, struct fib_nh *fib_nh,
442 		struct fib_config *cfg, int nh_weight,
443 		struct netlink_ext_ack *extack);
444 void fib_nh_release(struct net *net, struct fib_nh *fib_nh);
445 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *fc_encap,
446 		       u16 fc_encap_type, void *cfg, gfp_t gfp_flags,
447 		       struct netlink_ext_ack *extack);
448 void fib_nh_common_release(struct fib_nh_common *nhc);
449 
450 /* Exported by fib_trie.c */
451 void fib_trie_init(void);
452 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
453 
454 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
455 {
456 #ifdef CONFIG_IP_ROUTE_CLASSID
457 	struct fib_nh_common *nhc = res->nhc;
458 	struct fib_nh *nh = container_of(nhc, struct fib_nh, nh_common);
459 #ifdef CONFIG_IP_MULTIPLE_TABLES
460 	u32 rtag;
461 #endif
462 	*itag = nh->nh_tclassid << 16;
463 #ifdef CONFIG_IP_MULTIPLE_TABLES
464 	rtag = res->tclassid;
465 	if (*itag == 0)
466 		*itag = (rtag<<16);
467 	*itag |= (rtag>>16);
468 #endif
469 #endif
470 }
471 
472 void free_fib_info(struct fib_info *fi);
473 
474 static inline void fib_info_hold(struct fib_info *fi)
475 {
476 	refcount_inc(&fi->fib_clntref);
477 }
478 
479 static inline void fib_info_put(struct fib_info *fi)
480 {
481 	if (refcount_dec_and_test(&fi->fib_clntref))
482 		free_fib_info(fi);
483 }
484 
485 #ifdef CONFIG_PROC_FS
486 int __net_init fib_proc_init(struct net *net);
487 void __net_exit fib_proc_exit(struct net *net);
488 #else
489 static inline int fib_proc_init(struct net *net)
490 {
491 	return 0;
492 }
493 static inline void fib_proc_exit(struct net *net)
494 {
495 }
496 #endif
497 
498 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr);
499 
500 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
501 			  struct fib_dump_filter *filter,
502 			  struct netlink_callback *cb);
503 
504 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nh,
505 		     unsigned char *flags, bool skip_oif);
506 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nh,
507 		    int nh_weight);
508 #endif  /* _NET_FIB_H */
509