xref: /openbmc/linux/include/net/ip_fib.h (revision b7a59557)
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  *		Definitions for the Forwarding Information Base.
8  *
9  * Authors:	A.N.Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  */
11 
12 #ifndef _NET_IP_FIB_H
13 #define _NET_IP_FIB_H
14 
15 #include <net/flow.h>
16 #include <linux/seq_file.h>
17 #include <linux/rcupdate.h>
18 #include <net/fib_notifier.h>
19 #include <net/fib_rules.h>
20 #include <net/inetpeer.h>
21 #include <linux/percpu.h>
22 #include <linux/notifier.h>
23 #include <linux/refcount.h>
24 
25 struct fib_config {
26 	u8			fc_dst_len;
27 	u8			fc_tos;
28 	u8			fc_protocol;
29 	u8			fc_scope;
30 	u8			fc_type;
31 	u8			fc_gw_family;
32 	/* 2 bytes unused */
33 	u32			fc_table;
34 	__be32			fc_dst;
35 	union {
36 		__be32		fc_gw4;
37 		struct in6_addr	fc_gw6;
38 	};
39 	int			fc_oif;
40 	u32			fc_flags;
41 	u32			fc_priority;
42 	__be32			fc_prefsrc;
43 	u32			fc_nh_id;
44 	struct nlattr		*fc_mx;
45 	struct rtnexthop	*fc_mp;
46 	int			fc_mx_len;
47 	int			fc_mp_len;
48 	u32			fc_flow;
49 	u32			fc_nlflags;
50 	struct nl_info		fc_nlinfo;
51 	struct nlattr		*fc_encap;
52 	u16			fc_encap_type;
53 };
54 
55 struct fib_info;
56 struct rtable;
57 
58 struct fib_nh_exception {
59 	struct fib_nh_exception __rcu	*fnhe_next;
60 	int				fnhe_genid;
61 	__be32				fnhe_daddr;
62 	u32				fnhe_pmtu;
63 	bool				fnhe_mtu_locked;
64 	__be32				fnhe_gw;
65 	unsigned long			fnhe_expires;
66 	struct rtable __rcu		*fnhe_rth_input;
67 	struct rtable __rcu		*fnhe_rth_output;
68 	unsigned long			fnhe_stamp;
69 	struct rcu_head			rcu;
70 };
71 
72 struct fnhe_hash_bucket {
73 	struct fib_nh_exception __rcu	*chain;
74 };
75 
76 #define FNHE_HASH_SHIFT		11
77 #define FNHE_HASH_SIZE		(1 << FNHE_HASH_SHIFT)
78 #define FNHE_RECLAIM_DEPTH	5
79 
80 struct fib_nh_common {
81 	struct net_device	*nhc_dev;
82 	int			nhc_oif;
83 	unsigned char		nhc_scope;
84 	u8			nhc_family;
85 	u8			nhc_gw_family;
86 	unsigned char		nhc_flags;
87 	struct lwtunnel_state	*nhc_lwtstate;
88 
89 	union {
90 		__be32          ipv4;
91 		struct in6_addr ipv6;
92 	} nhc_gw;
93 
94 	int			nhc_weight;
95 	atomic_t		nhc_upper_bound;
96 
97 	/* v4 specific, but allows fib6_nh with v4 routes */
98 	struct rtable __rcu * __percpu *nhc_pcpu_rth_output;
99 	struct rtable __rcu     *nhc_rth_input;
100 	struct fnhe_hash_bucket	__rcu *nhc_exceptions;
101 };
102 
103 struct fib_nh {
104 	struct fib_nh_common	nh_common;
105 	struct hlist_node	nh_hash;
106 	struct fib_info		*nh_parent;
107 #ifdef CONFIG_IP_ROUTE_CLASSID
108 	__u32			nh_tclassid;
109 #endif
110 	__be32			nh_saddr;
111 	int			nh_saddr_genid;
112 #define fib_nh_family		nh_common.nhc_family
113 #define fib_nh_dev		nh_common.nhc_dev
114 #define fib_nh_oif		nh_common.nhc_oif
115 #define fib_nh_flags		nh_common.nhc_flags
116 #define fib_nh_lws		nh_common.nhc_lwtstate
117 #define fib_nh_scope		nh_common.nhc_scope
118 #define fib_nh_gw_family	nh_common.nhc_gw_family
119 #define fib_nh_gw4		nh_common.nhc_gw.ipv4
120 #define fib_nh_gw6		nh_common.nhc_gw.ipv6
121 #define fib_nh_weight		nh_common.nhc_weight
122 #define fib_nh_upper_bound	nh_common.nhc_upper_bound
123 };
124 
125 /*
126  * This structure contains data shared by many of routes.
127  */
128 
129 struct nexthop;
130 
131 struct fib_info {
132 	struct hlist_node	fib_hash;
133 	struct hlist_node	fib_lhash;
134 	struct list_head	nh_list;
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 	bool			nh_updated;
154 	struct nexthop		*nh;
155 	struct rcu_head		rcu;
156 	struct fib_nh		fib_nh[0];
157 };
158 
159 
160 #ifdef CONFIG_IP_MULTIPLE_TABLES
161 struct fib_rule;
162 #endif
163 
164 struct fib_table;
165 struct fib_result {
166 	__be32			prefix;
167 	unsigned char		prefixlen;
168 	unsigned char		nh_sel;
169 	unsigned char		type;
170 	unsigned char		scope;
171 	u32			tclassid;
172 	struct fib_nh_common	*nhc;
173 	struct fib_info		*fi;
174 	struct fib_table	*table;
175 	struct hlist_head	*fa_head;
176 };
177 
178 struct fib_result_nl {
179 	__be32		fl_addr;   /* To be looked up*/
180 	u32		fl_mark;
181 	unsigned char	fl_tos;
182 	unsigned char   fl_scope;
183 	unsigned char   tb_id_in;
184 
185 	unsigned char   tb_id;      /* Results */
186 	unsigned char	prefixlen;
187 	unsigned char	nh_sel;
188 	unsigned char	type;
189 	unsigned char	scope;
190 	int             err;
191 };
192 
193 #ifdef CONFIG_IP_MULTIPLE_TABLES
194 #define FIB_TABLE_HASHSZ 256
195 #else
196 #define FIB_TABLE_HASHSZ 2
197 #endif
198 
199 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
200 				 unsigned char scope);
201 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res);
202 
203 #define FIB_RES_NHC(res)		((res).nhc)
204 #define FIB_RES_DEV(res)	(FIB_RES_NHC(res)->nhc_dev)
205 #define FIB_RES_OIF(res)	(FIB_RES_NHC(res)->nhc_oif)
206 
207 struct fib_entry_notifier_info {
208 	struct fib_notifier_info info; /* must be first */
209 	u32 dst;
210 	int dst_len;
211 	struct fib_info *fi;
212 	u8 tos;
213 	u8 type;
214 	u32 tb_id;
215 };
216 
217 struct fib_nh_notifier_info {
218 	struct fib_notifier_info info; /* must be first */
219 	struct fib_nh *fib_nh;
220 };
221 
222 int call_fib4_notifier(struct notifier_block *nb,
223 		       enum fib_event_type event_type,
224 		       struct fib_notifier_info *info);
225 int call_fib4_notifiers(struct net *net, enum fib_event_type event_type,
226 			struct fib_notifier_info *info);
227 
228 int __net_init fib4_notifier_init(struct net *net);
229 void __net_exit fib4_notifier_exit(struct net *net);
230 
231 void fib_info_notify_update(struct net *net, struct nl_info *info);
232 int fib_notify(struct net *net, struct notifier_block *nb,
233 	       struct netlink_ext_ack *extack);
234 
235 struct fib_table {
236 	struct hlist_node	tb_hlist;
237 	u32			tb_id;
238 	int			tb_num_default;
239 	struct rcu_head		rcu;
240 	unsigned long 		*tb_data;
241 	unsigned long		__data[0];
242 };
243 
244 struct fib_dump_filter {
245 	u32			table_id;
246 	/* filter_set is an optimization that an entry is set */
247 	bool			filter_set;
248 	bool			dump_all_families;
249 	bool			dump_routes;
250 	bool			dump_exceptions;
251 	unsigned char		protocol;
252 	unsigned char		rt_type;
253 	unsigned int		flags;
254 	struct net_device	*dev;
255 };
256 
257 int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
258 		     struct fib_result *res, int fib_flags);
259 int fib_table_insert(struct net *, struct fib_table *, struct fib_config *,
260 		     struct netlink_ext_ack *extack);
261 int fib_table_delete(struct net *, struct fib_table *, struct fib_config *,
262 		     struct netlink_ext_ack *extack);
263 int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
264 		   struct netlink_callback *cb, struct fib_dump_filter *filter);
265 int fib_table_flush(struct net *net, struct fib_table *table, bool flush_all);
266 struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
267 void fib_table_flush_external(struct fib_table *table);
268 void fib_free_table(struct fib_table *tb);
269 
270 #ifndef CONFIG_IP_MULTIPLE_TABLES
271 
272 #define TABLE_LOCAL_INDEX	(RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
273 #define TABLE_MAIN_INDEX	(RT_TABLE_MAIN  & (FIB_TABLE_HASHSZ - 1))
274 
275 static inline struct fib_table *fib_get_table(struct net *net, u32 id)
276 {
277 	struct hlist_node *tb_hlist;
278 	struct hlist_head *ptr;
279 
280 	ptr = id == RT_TABLE_LOCAL ?
281 		&net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
282 		&net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
283 
284 	tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
285 
286 	return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
287 }
288 
289 static inline struct fib_table *fib_new_table(struct net *net, u32 id)
290 {
291 	return fib_get_table(net, id);
292 }
293 
294 static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
295 			     struct fib_result *res, unsigned int flags)
296 {
297 	struct fib_table *tb;
298 	int err = -ENETUNREACH;
299 
300 	rcu_read_lock();
301 
302 	tb = fib_get_table(net, RT_TABLE_MAIN);
303 	if (tb)
304 		err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
305 
306 	if (err == -EAGAIN)
307 		err = -ENETUNREACH;
308 
309 	rcu_read_unlock();
310 
311 	return err;
312 }
313 
314 static inline bool fib4_rule_default(const struct fib_rule *rule)
315 {
316 	return true;
317 }
318 
319 static inline int fib4_rules_dump(struct net *net, struct notifier_block *nb,
320 				  struct netlink_ext_ack *extack)
321 {
322 	return 0;
323 }
324 
325 static inline unsigned int fib4_rules_seq_read(struct net *net)
326 {
327 	return 0;
328 }
329 
330 static inline bool fib4_rules_early_flow_dissect(struct net *net,
331 						 struct sk_buff *skb,
332 						 struct flowi4 *fl4,
333 						 struct flow_keys *flkeys)
334 {
335 	return false;
336 }
337 #else /* CONFIG_IP_MULTIPLE_TABLES */
338 int __net_init fib4_rules_init(struct net *net);
339 void __net_exit fib4_rules_exit(struct net *net);
340 
341 struct fib_table *fib_new_table(struct net *net, u32 id);
342 struct fib_table *fib_get_table(struct net *net, u32 id);
343 
344 int __fib_lookup(struct net *net, struct flowi4 *flp,
345 		 struct fib_result *res, unsigned int flags);
346 
347 static inline int fib_lookup(struct net *net, struct flowi4 *flp,
348 			     struct fib_result *res, unsigned int flags)
349 {
350 	struct fib_table *tb;
351 	int err = -ENETUNREACH;
352 
353 	flags |= FIB_LOOKUP_NOREF;
354 	if (net->ipv4.fib_has_custom_rules)
355 		return __fib_lookup(net, flp, res, flags);
356 
357 	rcu_read_lock();
358 
359 	res->tclassid = 0;
360 
361 	tb = rcu_dereference_rtnl(net->ipv4.fib_main);
362 	if (tb)
363 		err = fib_table_lookup(tb, flp, res, flags);
364 
365 	if (!err)
366 		goto out;
367 
368 	tb = rcu_dereference_rtnl(net->ipv4.fib_default);
369 	if (tb)
370 		err = fib_table_lookup(tb, flp, res, flags);
371 
372 out:
373 	if (err == -EAGAIN)
374 		err = -ENETUNREACH;
375 
376 	rcu_read_unlock();
377 
378 	return err;
379 }
380 
381 bool fib4_rule_default(const struct fib_rule *rule);
382 int fib4_rules_dump(struct net *net, struct notifier_block *nb,
383 		    struct netlink_ext_ack *extack);
384 unsigned int fib4_rules_seq_read(struct net *net);
385 
386 static inline bool fib4_rules_early_flow_dissect(struct net *net,
387 						 struct sk_buff *skb,
388 						 struct flowi4 *fl4,
389 						 struct flow_keys *flkeys)
390 {
391 	unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
392 
393 	if (!net->ipv4.fib_rules_require_fldissect)
394 		return false;
395 
396 	skb_flow_dissect_flow_keys(skb, flkeys, flag);
397 	fl4->fl4_sport = flkeys->ports.src;
398 	fl4->fl4_dport = flkeys->ports.dst;
399 	fl4->flowi4_proto = flkeys->basic.ip_proto;
400 
401 	return true;
402 }
403 
404 #endif /* CONFIG_IP_MULTIPLE_TABLES */
405 
406 /* Exported by fib_frontend.c */
407 extern const struct nla_policy rtm_ipv4_policy[];
408 void ip_fib_init(void);
409 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
410 		    struct netlink_ext_ack *extack);
411 __be32 fib_compute_spec_dst(struct sk_buff *skb);
412 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev);
413 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
414 			u8 tos, int oif, struct net_device *dev,
415 			struct in_device *idev, u32 *itag);
416 #ifdef CONFIG_IP_ROUTE_CLASSID
417 static inline int fib_num_tclassid_users(struct net *net)
418 {
419 	return net->ipv4.fib_num_tclassid_users;
420 }
421 #else
422 static inline int fib_num_tclassid_users(struct net *net)
423 {
424 	return 0;
425 }
426 #endif
427 int fib_unmerge(struct net *net);
428 
429 /* Exported by fib_semantics.c */
430 int ip_fib_check_default(__be32 gw, struct net_device *dev);
431 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
432 int fib_sync_down_addr(struct net_device *dev, __be32 local);
433 int fib_sync_up(struct net_device *dev, unsigned char nh_flags);
434 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu);
435 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig);
436 
437 #ifdef CONFIG_IP_ROUTE_MULTIPATH
438 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
439 		       const struct sk_buff *skb, struct flow_keys *flkeys);
440 #endif
441 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
442 		 struct netlink_ext_ack *extack);
443 void fib_select_multipath(struct fib_result *res, int hash);
444 void fib_select_path(struct net *net, struct fib_result *res,
445 		     struct flowi4 *fl4, const struct sk_buff *skb);
446 
447 int fib_nh_init(struct net *net, struct fib_nh *fib_nh,
448 		struct fib_config *cfg, int nh_weight,
449 		struct netlink_ext_ack *extack);
450 void fib_nh_release(struct net *net, struct fib_nh *fib_nh);
451 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *fc_encap,
452 		       u16 fc_encap_type, void *cfg, gfp_t gfp_flags,
453 		       struct netlink_ext_ack *extack);
454 void fib_nh_common_release(struct fib_nh_common *nhc);
455 
456 /* Exported by fib_trie.c */
457 void fib_trie_init(void);
458 struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
459 
460 static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
461 {
462 #ifdef CONFIG_IP_ROUTE_CLASSID
463 	struct fib_nh_common *nhc = res->nhc;
464 #ifdef CONFIG_IP_MULTIPLE_TABLES
465 	u32 rtag;
466 #endif
467 	if (nhc->nhc_family == AF_INET) {
468 		struct fib_nh *nh;
469 
470 		nh = container_of(nhc, struct fib_nh, nh_common);
471 		*itag = nh->nh_tclassid << 16;
472 	} else {
473 		*itag = 0;
474 	}
475 
476 #ifdef CONFIG_IP_MULTIPLE_TABLES
477 	rtag = res->tclassid;
478 	if (*itag == 0)
479 		*itag = (rtag<<16);
480 	*itag |= (rtag>>16);
481 #endif
482 #endif
483 }
484 
485 void fib_flush(struct net *net);
486 void free_fib_info(struct fib_info *fi);
487 
488 static inline void fib_info_hold(struct fib_info *fi)
489 {
490 	refcount_inc(&fi->fib_clntref);
491 }
492 
493 static inline void fib_info_put(struct fib_info *fi)
494 {
495 	if (refcount_dec_and_test(&fi->fib_clntref))
496 		free_fib_info(fi);
497 }
498 
499 #ifdef CONFIG_PROC_FS
500 int __net_init fib_proc_init(struct net *net);
501 void __net_exit fib_proc_exit(struct net *net);
502 #else
503 static inline int fib_proc_init(struct net *net)
504 {
505 	return 0;
506 }
507 static inline void fib_proc_exit(struct net *net)
508 {
509 }
510 #endif
511 
512 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr);
513 
514 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
515 			  struct fib_dump_filter *filter,
516 			  struct netlink_callback *cb);
517 
518 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nh,
519 		     u8 rt_family, unsigned char *flags, bool skip_oif);
520 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nh,
521 		    int nh_weight, u8 rt_family);
522 #endif  /* _NET_FIB_H */
523