ip6_fib.h

  1/* SPDX-License-Identifier: GPL-2.0-or-later */
  2/*
  3 *	Linux INET6 implementation 
  4 *
  5 *	Authors:
  6 *	Pedro Roque		<roque@di.fc.ul.pt>	
  7 */
  8
  9#ifndef _IP6_FIB_H
 10#define _IP6_FIB_H
 11
 12#include <linux/ipv6_route.h>
 13#include <linux/rtnetlink.h>
 14#include <linux/spinlock.h>
 15#include <linux/notifier.h>
 16#include <net/dst.h>
 17#include <net/flow.h>
 18#include <net/ip_fib.h>
 19#include <net/netlink.h>
 20#include <net/inetpeer.h>
 21#include <net/fib_notifier.h>
 22#include <linux/indirect_call_wrapper.h>
 23#include <uapi/linux/bpf.h>
 24
 25#ifdef CONFIG_IPV6_MULTIPLE_TABLES
 26#define FIB6_TABLE_HASHSZ 256
 27#else
 28#define FIB6_TABLE_HASHSZ 1
 29#endif
 30
 31#define RT6_DEBUG 2
 32
 33struct rt6_info;
 34struct fib6_info;
 35
 36struct fib6_config {
 37	u32		fc_table;
 38	u32		fc_metric;
 39	int		fc_dst_len;
 40	int		fc_src_len;
 41	int		fc_ifindex;
 42	u32		fc_flags;
 43	u32		fc_protocol;
 44	u16		fc_type;        /* only 8 bits are used */
 45	u16		fc_delete_all_nh : 1,
 46			fc_ignore_dev_down:1,
 47			__unused : 14;
 48	u32		fc_nh_id;
 49
 50	struct in6_addr	fc_dst;
 51	struct in6_addr	fc_src;
 52	struct in6_addr	fc_prefsrc;
 53	struct in6_addr	fc_gateway;
 54
 55	unsigned long	fc_expires;
 56	struct nlattr	*fc_mx;
 57	int		fc_mx_len;
 58	int		fc_mp_len;
 59	struct nlattr	*fc_mp;
 60
 61	struct nl_info	fc_nlinfo;
 62	struct nlattr	*fc_encap;
 63	u16		fc_encap_type;
 64	bool		fc_is_fdb;
 65};
 66
 67struct fib6_node {
 68	struct fib6_node __rcu	*parent;
 69	struct fib6_node __rcu	*left;
 70	struct fib6_node __rcu	*right;
 71#ifdef CONFIG_IPV6_SUBTREES
 72	struct fib6_node __rcu	*subtree;
 73#endif
 74	struct fib6_info __rcu	*leaf;
 75
 76	__u16			fn_bit;		/* bit key */
 77	__u16			fn_flags;
 78	int			fn_sernum;
 79	struct fib6_info __rcu	*rr_ptr;
 80	struct rcu_head		rcu;
 81};
 82
 83struct fib6_gc_args {
 84	int			timeout;
 85	int			more;
 86};
 87
 88#ifndef CONFIG_IPV6_SUBTREES
 89#define FIB6_SUBTREE(fn)	NULL
 90
 91static inline bool fib6_routes_require_src(const struct net *net)
 92{
 93	return false;
 94}
 95
 96static inline void fib6_routes_require_src_inc(struct net *net) {}
 97static inline void fib6_routes_require_src_dec(struct net *net) {}
 98
 99#else
100
101static inline bool fib6_routes_require_src(const struct net *net)
102{
103	return net->ipv6.fib6_routes_require_src > 0;
104}
105
106static inline void fib6_routes_require_src_inc(struct net *net)
107{
108	net->ipv6.fib6_routes_require_src++;
109}
110
111static inline void fib6_routes_require_src_dec(struct net *net)
112{
113	net->ipv6.fib6_routes_require_src--;
114}
115
116#define FIB6_SUBTREE(fn)	(rcu_dereference_protected((fn)->subtree, 1))
117#endif
118
119/*
120 *	routing information
121 *
122 */
123
124struct rt6key {
125	struct in6_addr	addr;
126	int		plen;
127};
128
129struct fib6_table;
130
131struct rt6_exception_bucket {
132	struct hlist_head	chain;
133	int			depth;
134};
135
136struct rt6_exception {
137	struct hlist_node	hlist;
138	struct rt6_info		*rt6i;
139	unsigned long		stamp;
140	struct rcu_head		rcu;
141};
142
143#define FIB6_EXCEPTION_BUCKET_SIZE_SHIFT 10
144#define FIB6_EXCEPTION_BUCKET_SIZE (1 << FIB6_EXCEPTION_BUCKET_SIZE_SHIFT)
145#define FIB6_MAX_DEPTH 5
146
147struct fib6_nh {
148	struct fib_nh_common	nh_common;
149
150#ifdef CONFIG_IPV6_ROUTER_PREF
151	unsigned long		last_probe;
152#endif
153
154	struct rt6_info * __percpu *rt6i_pcpu;
155	struct rt6_exception_bucket __rcu *rt6i_exception_bucket;
156};
157
158struct fib6_info {
159	struct fib6_table		*fib6_table;
160	struct fib6_info __rcu		*fib6_next;
161	struct fib6_node __rcu		*fib6_node;
162
163	/* Multipath routes:
164	 * siblings is a list of fib6_info that have the same metric/weight,
165	 * destination, but not the same gateway. nsiblings is just a cache
166	 * to speed up lookup.
167	 */
168	union {
169		struct list_head	fib6_siblings;
170		struct list_head	nh_list;
171	};
172	unsigned int			fib6_nsiblings;
173
174	refcount_t			fib6_ref;
175	unsigned long			expires;
176
177	struct hlist_node		gc_link;
178
179	struct dst_metrics		*fib6_metrics;
180#define fib6_pmtu		fib6_metrics->metrics[RTAX_MTU-1]
181
182	struct rt6key			fib6_dst;
183	u32				fib6_flags;
184	struct rt6key			fib6_src;
185	struct rt6key			fib6_prefsrc;
186
187	u32				fib6_metric;
188	u8				fib6_protocol;
189	u8				fib6_type;
190
191	u8				offload;
192	u8				trap;
193	u8				offload_failed;
194
195	u8				should_flush:1,
196					dst_nocount:1,
197					dst_nopolicy:1,
198					fib6_destroying:1,
199					unused:4;
200
201	struct list_head		purge_link;
202	struct rcu_head			rcu;
203	struct nexthop			*nh;
204	struct fib6_nh			fib6_nh[];
205};
206
207struct rt6_info {
208	struct dst_entry		dst;
209	struct fib6_info __rcu		*from;
210	int				sernum;
211
212	struct rt6key			rt6i_dst;
213	struct rt6key			rt6i_src;
214	struct in6_addr			rt6i_gateway;
215	struct inet6_dev		*rt6i_idev;
216	u32				rt6i_flags;
217
218	/* more non-fragment space at head required */
219	unsigned short			rt6i_nfheader_len;
220};
221
222struct fib6_result {
223	struct fib6_nh		*nh;
224	struct fib6_info	*f6i;
225	u32			fib6_flags;
226	u8			fib6_type;
227	struct rt6_info		*rt6;
228};
229
230#define for_each_fib6_node_rt_rcu(fn)					\
231	for (rt = rcu_dereference((fn)->leaf); rt;			\
232	     rt = rcu_dereference(rt->fib6_next))
233
234#define for_each_fib6_walker_rt(w)					\
235	for (rt = (w)->leaf; rt;					\
236	     rt = rcu_dereference_protected(rt->fib6_next, 1))
237
238#define dst_rt6_info(_ptr) container_of_const(_ptr, struct rt6_info, dst)
239
240static inline struct inet6_dev *ip6_dst_idev(const struct dst_entry *dst)
241{
242	return dst_rt6_info(dst)->rt6i_idev;
243}
244
245static inline bool fib6_requires_src(const struct fib6_info *rt)
246{
247	return rt->fib6_src.plen > 0;
248}
249
250/* The callers should hold f6i->fib6_table->tb6_lock if a route has ever
251 * been added to a table before.
252 */
253static inline void fib6_clean_expires(struct fib6_info *f6i)
254{
255	f6i->fib6_flags &= ~RTF_EXPIRES;
256	f6i->expires = 0;
257}
258
259/* The callers should hold f6i->fib6_table->tb6_lock if a route has ever
260 * been added to a table before.
261 */
262static inline void fib6_set_expires(struct fib6_info *f6i,
263				    unsigned long expires)
264{
265	f6i->expires = expires;
266	f6i->fib6_flags |= RTF_EXPIRES;
267}
268
269static inline bool fib6_check_expired(const struct fib6_info *f6i)
270{
271	if (f6i->fib6_flags & RTF_EXPIRES)
272		return time_after(jiffies, f6i->expires);
273	return false;
274}
275
276/* Function to safely get fn->fn_sernum for passed in rt
277 * and store result in passed in cookie.
278 * Return true if we can get cookie safely
279 * Return false if not
280 */
281static inline bool fib6_get_cookie_safe(const struct fib6_info *f6i,
282					u32 *cookie)
283{
284	struct fib6_node *fn;
285	bool status = false;
286
287	fn = rcu_dereference(f6i->fib6_node);
288
289	if (fn) {
290		*cookie = READ_ONCE(fn->fn_sernum);
291		/* pairs with smp_wmb() in __fib6_update_sernum_upto_root() */
292		smp_rmb();
293		status = true;
294	}
295
296	return status;
297}
298
299static inline u32 rt6_get_cookie(const struct rt6_info *rt)
300{
301	struct fib6_info *from;
302	u32 cookie = 0;
303
304	if (rt->sernum)
305		return rt->sernum;
306
307	rcu_read_lock();
308
309	from = rcu_dereference(rt->from);
310	if (from)
311		fib6_get_cookie_safe(from, &cookie);
312
313	rcu_read_unlock();
314
315	return cookie;
316}
317
318static inline void ip6_rt_put(struct rt6_info *rt)
319{
320	/* dst_release() accepts a NULL parameter.
321	 * We rely on dst being first structure in struct rt6_info
322	 */
323	BUILD_BUG_ON(offsetof(struct rt6_info, dst) != 0);
324	dst_release(&rt->dst);
325}
326
327struct fib6_info *fib6_info_alloc(gfp_t gfp_flags, bool with_fib6_nh);
328void fib6_info_destroy_rcu(struct rcu_head *head);
329
330static inline void fib6_info_hold(struct fib6_info *f6i)
331{
332	refcount_inc(&f6i->fib6_ref);
333}
334
335static inline bool fib6_info_hold_safe(struct fib6_info *f6i)
336{
337	return refcount_inc_not_zero(&f6i->fib6_ref);
338}
339
340static inline void fib6_info_release(struct fib6_info *f6i)
341{
342	if (f6i && refcount_dec_and_test(&f6i->fib6_ref)) {
343		DEBUG_NET_WARN_ON_ONCE(!hlist_unhashed(&f6i->gc_link));
344		call_rcu_hurry(&f6i->rcu, fib6_info_destroy_rcu);
345	}
346}
347
348enum fib6_walk_state {
349#ifdef CONFIG_IPV6_SUBTREES
350	FWS_S,
351#endif
352	FWS_L,
353	FWS_R,
354	FWS_C,
355	FWS_U
356};
357
358struct fib6_walker {
359	struct list_head lh;
360	struct fib6_node *root, *node;
361	struct fib6_info *leaf;
362	enum fib6_walk_state state;
363	unsigned int skip;
364	unsigned int count;
365	unsigned int skip_in_node;
366	int (*func)(struct fib6_walker *);
367	void *args;
368};
369
370struct rt6_statistics {
371	__u32		fib_nodes;		/* all fib6 nodes */
372	__u32		fib_route_nodes;	/* intermediate nodes */
373	__u32		fib_rt_entries;		/* rt entries in fib table */
374	__u32		fib_rt_cache;		/* cached rt entries in exception table */
375	__u32		fib_discarded_routes;	/* total number of routes delete */
376
377	/* The following stat is not protected by any lock */
378	atomic_t	fib_rt_alloc;		/* total number of routes alloced */
379};
380
381#define RTN_TL_ROOT	0x0001
382#define RTN_ROOT	0x0002		/* tree root node		*/
383#define RTN_RTINFO	0x0004		/* node with valid routing info	*/
384
385/*
386 *	priority levels (or metrics)
387 *
388 */
389
390
391struct fib6_table {
392	struct hlist_node	tb6_hlist;
393	u32			tb6_id;
394	spinlock_t		tb6_lock;
395	struct fib6_node	tb6_root;
396	struct inet_peer_base	tb6_peers;
397	unsigned int		flags;
398	unsigned int		fib_seq; /* writes protected by rtnl_mutex */
399	struct hlist_head       tb6_gc_hlist;	/* GC candidates */
400#define RT6_TABLE_HAS_DFLT_ROUTER	BIT(0)
401};
402
403#define RT6_TABLE_UNSPEC	RT_TABLE_UNSPEC
404#define RT6_TABLE_MAIN		RT_TABLE_MAIN
405#define RT6_TABLE_DFLT		RT6_TABLE_MAIN
406#define RT6_TABLE_INFO		RT6_TABLE_MAIN
407#define RT6_TABLE_PREFIX	RT6_TABLE_MAIN
408
409#ifdef CONFIG_IPV6_MULTIPLE_TABLES
410#define FIB6_TABLE_MIN		1
411#define FIB6_TABLE_MAX		RT_TABLE_MAX
412#define RT6_TABLE_LOCAL		RT_TABLE_LOCAL
413#else
414#define FIB6_TABLE_MIN		RT_TABLE_MAIN
415#define FIB6_TABLE_MAX		FIB6_TABLE_MIN
416#define RT6_TABLE_LOCAL		RT6_TABLE_MAIN
417#endif
418
419typedef struct rt6_info *(*pol_lookup_t)(struct net *,
420					 struct fib6_table *,
421					 struct flowi6 *,
422					 const struct sk_buff *, int);
423
424struct fib6_entry_notifier_info {
425	struct fib_notifier_info info; /* must be first */
426	struct fib6_info *rt;
427	unsigned int nsiblings;
428};
429
430/*
431 *	exported functions
432 */
433
434struct fib6_table *fib6_get_table(struct net *net, u32 id);
435struct fib6_table *fib6_new_table(struct net *net, u32 id);
436struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
437				   const struct sk_buff *skb,
438				   int flags, pol_lookup_t lookup);
439
440/* called with rcu lock held; can return error pointer
441 * caller needs to select path
442 */
443int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
444		struct fib6_result *res, int flags);
445
446/* called with rcu lock held; caller needs to select path */
447int fib6_table_lookup(struct net *net, struct fib6_table *table,
448		      int oif, struct flowi6 *fl6, struct fib6_result *res,
449		      int strict);
450
451void fib6_select_path(const struct net *net, struct fib6_result *res,
452		      struct flowi6 *fl6, int oif, bool have_oif_match,
453		      const struct sk_buff *skb, int strict);
454struct fib6_node *fib6_node_lookup(struct fib6_node *root,
455				   const struct in6_addr *daddr,
456				   const struct in6_addr *saddr);
457
458struct fib6_node *fib6_locate(struct fib6_node *root,
459			      const struct in6_addr *daddr, int dst_len,
460			      const struct in6_addr *saddr, int src_len,
461			      bool exact_match);
462
463void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *arg),
464		    void *arg);
465void fib6_clean_all_skip_notify(struct net *net,
466				int (*func)(struct fib6_info *, void *arg),
467				void *arg);
468
469int fib6_add(struct fib6_node *root, struct fib6_info *rt,
470	     struct nl_info *info, struct netlink_ext_ack *extack);
471int fib6_del(struct fib6_info *rt, struct nl_info *info);
472
473static inline
474void rt6_get_prefsrc(const struct rt6_info *rt, struct in6_addr *addr)
475{
476	const struct fib6_info *from;
477
478	rcu_read_lock();
479
480	from = rcu_dereference(rt->from);
481	if (from)
482		*addr = from->fib6_prefsrc.addr;
483	else
484		*addr = in6addr_any;
485
486	rcu_read_unlock();
487}
488
489int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
490		 struct fib6_config *cfg, gfp_t gfp_flags,
491		 struct netlink_ext_ack *extack);
492void fib6_nh_release(struct fib6_nh *fib6_nh);
493void fib6_nh_release_dsts(struct fib6_nh *fib6_nh);
494
495int call_fib6_entry_notifiers(struct net *net,
496			      enum fib_event_type event_type,
497			      struct fib6_info *rt,
498			      struct netlink_ext_ack *extack);
499int call_fib6_multipath_entry_notifiers(struct net *net,
500					enum fib_event_type event_type,
501					struct fib6_info *rt,
502					unsigned int nsiblings,
503					struct netlink_ext_ack *extack);
504int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt);
505void fib6_rt_update(struct net *net, struct fib6_info *rt,
506		    struct nl_info *info);
507void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
508		     unsigned int flags);
509
510void fib6_age_exceptions(struct fib6_info *rt, struct fib6_gc_args *gc_args,
511			 unsigned long now);
512void fib6_run_gc(unsigned long expires, struct net *net, bool force);
513void fib6_gc_cleanup(void);
514
515int fib6_init(void);
516
517#if IS_ENABLED(CONFIG_IPV6)
518/* Add the route to the gc list if it is not already there
519 *
520 * The callers should hold f6i->fib6_table->tb6_lock.
521 */
522static inline void fib6_add_gc_list(struct fib6_info *f6i)
523{
524	/* If fib6_node is null, the f6i is not in (or removed from) the
525	 * table.
526	 *
527	 * There is a gap between finding the f6i from the table and
528	 * calling this function without the protection of the tb6_lock.
529	 * This check makes sure the f6i is not added to the gc list when
530	 * it is not on the table.
531	 */
532	if (!rcu_dereference_protected(f6i->fib6_node,
533				       lockdep_is_held(&f6i->fib6_table->tb6_lock)))
534		return;
535
536	if (hlist_unhashed(&f6i->gc_link))
537		hlist_add_head(&f6i->gc_link, &f6i->fib6_table->tb6_gc_hlist);
538}
539
540/* Remove the route from the gc list if it is on the list.
541 *
542 * The callers should hold f6i->fib6_table->tb6_lock.
543 */
544static inline void fib6_remove_gc_list(struct fib6_info *f6i)
545{
546	if (!hlist_unhashed(&f6i->gc_link))
547		hlist_del_init(&f6i->gc_link);
548}
549
550static inline void fib6_may_remove_gc_list(struct net *net,
551					   struct fib6_info *f6i)
552{
553	struct fib6_gc_args gc_args;
554
555	if (hlist_unhashed(&f6i->gc_link))
556		return;
557
558	gc_args.timeout = READ_ONCE(net->ipv6.sysctl.ip6_rt_gc_interval);
559	gc_args.more = 0;
560
561	rcu_read_lock();
562	fib6_age_exceptions(f6i, &gc_args, jiffies);
563	rcu_read_unlock();
564}
565#endif
566
567struct ipv6_route_iter {
568	struct seq_net_private p;
569	struct fib6_walker w;
570	loff_t skip;
571	struct fib6_table *tbl;
572	int sernum;
573};
574
575extern const struct seq_operations ipv6_route_seq_ops;
576
577int call_fib6_notifier(struct notifier_block *nb,
578		       enum fib_event_type event_type,
579		       struct fib_notifier_info *info);
580int call_fib6_notifiers(struct net *net, enum fib_event_type event_type,
581			struct fib_notifier_info *info);
582
583int __net_init fib6_notifier_init(struct net *net);
584void __net_exit fib6_notifier_exit(struct net *net);
585
586unsigned int fib6_tables_seq_read(const struct net *net);
587int fib6_tables_dump(struct net *net, struct notifier_block *nb,
588		     struct netlink_ext_ack *extack);
589
590void fib6_update_sernum(struct net *net, struct fib6_info *rt);
591void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt);
592void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i);
593
594void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val);
595static inline bool fib6_metric_locked(struct fib6_info *f6i, int metric)
596{
597	return !!(f6i->fib6_metrics->metrics[RTAX_LOCK - 1] & (1 << metric));
598}
599void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
600			    bool offload, bool trap, bool offload_failed);
601
602#if IS_BUILTIN(CONFIG_IPV6) && defined(CONFIG_BPF_SYSCALL)
603struct bpf_iter__ipv6_route {
604	__bpf_md_ptr(struct bpf_iter_meta *, meta);
605	__bpf_md_ptr(struct fib6_info *, rt);
606};
607#endif
608
609INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_output(struct net *net,
610					     struct fib6_table *table,
611					     struct flowi6 *fl6,
612					     const struct sk_buff *skb,
613					     int flags));
614INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_input(struct net *net,
615					     struct fib6_table *table,
616					     struct flowi6 *fl6,
617					     const struct sk_buff *skb,
618					     int flags));
619INDIRECT_CALLABLE_DECLARE(struct rt6_info *__ip6_route_redirect(struct net *net,
620					     struct fib6_table *table,
621					     struct flowi6 *fl6,
622					     const struct sk_buff *skb,
623					     int flags));
624INDIRECT_CALLABLE_DECLARE(struct rt6_info *ip6_pol_route_lookup(struct net *net,
625					     struct fib6_table *table,
626					     struct flowi6 *fl6,
627					     const struct sk_buff *skb,
628					     int flags));
629static inline struct rt6_info *pol_lookup_func(pol_lookup_t lookup,
630						struct net *net,
631						struct fib6_table *table,
632						struct flowi6 *fl6,
633						const struct sk_buff *skb,
634						int flags)
635{
636	return INDIRECT_CALL_4(lookup,
637			       ip6_pol_route_output,
638			       ip6_pol_route_input,
639			       ip6_pol_route_lookup,
640			       __ip6_route_redirect,
641			       net, table, fl6, skb, flags);
642}
643
644#ifdef CONFIG_IPV6_MULTIPLE_TABLES
645static inline bool fib6_has_custom_rules(const struct net *net)
646{
647	return net->ipv6.fib6_has_custom_rules;
648}
649
650int fib6_rules_init(void);
651void fib6_rules_cleanup(void);
652bool fib6_rule_default(const struct fib_rule *rule);
653int fib6_rules_dump(struct net *net, struct notifier_block *nb,
654		    struct netlink_ext_ack *extack);
655unsigned int fib6_rules_seq_read(const struct net *net);
656
657static inline bool fib6_rules_early_flow_dissect(struct net *net,
658						 struct sk_buff *skb,
659						 struct flowi6 *fl6,
660						 struct flow_keys *flkeys)
661{
662	unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
663
664	if (!net->ipv6.fib6_rules_require_fldissect)
665		return false;
666
667	memset(flkeys, 0, sizeof(*flkeys));
668	__skb_flow_dissect(net, skb, &flow_keys_dissector,
669			   flkeys, NULL, 0, 0, 0, flag);
670
671	fl6->fl6_sport = flkeys->ports.src;
672	fl6->fl6_dport = flkeys->ports.dst;
673	fl6->flowi6_proto = flkeys->basic.ip_proto;
674
675	return true;
676}
677#else
678static inline bool fib6_has_custom_rules(const struct net *net)
679{
680	return false;
681}
682static inline int               fib6_rules_init(void)
683{
684	return 0;
685}
686static inline void              fib6_rules_cleanup(void)
687{
688	return ;
689}
690static inline bool fib6_rule_default(const struct fib_rule *rule)
691{
692	return true;
693}
694static inline int fib6_rules_dump(struct net *net, struct notifier_block *nb,
695				  struct netlink_ext_ack *extack)
696{
697	return 0;
698}
699static inline unsigned int fib6_rules_seq_read(const struct net *net)
700{
701	return 0;
702}
703static inline bool fib6_rules_early_flow_dissect(struct net *net,
704						 struct sk_buff *skb,
705						 struct flowi6 *fl6,
706						 struct flow_keys *flkeys)
707{
708	return false;
709}
710#endif
711#endif