tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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
25struct 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
55struct fib_info;
56struct rtable;
57
58struct 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
72struct 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
80struct 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
103struct 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
129struct nexthop;
130
131struct 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 refcount_t 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[];
157};
158
159
160#ifdef CONFIG_IP_MULTIPLE_TABLES
161struct fib_rule;
162#endif
163
164struct fib_table;
165struct 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
178struct 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
207struct fib_rt_info {
208 struct fib_info *fi;
209 u32 tb_id;
210 __be32 dst;
211 int dst_len;
212 u8 tos;
213 u8 type;
214 u8 offload:1,
215 trap:1,
216 offload_failed:1,
217 unused:5;
218};
219
220struct fib_entry_notifier_info {
221 struct fib_notifier_info info; /* must be first */
222 u32 dst;
223 int dst_len;
224 struct fib_info *fi;
225 u8 tos;
226 u8 type;
227 u32 tb_id;
228};
229
230struct fib_nh_notifier_info {
231 struct fib_notifier_info info; /* must be first */
232 struct fib_nh *fib_nh;
233};
234
235int call_fib4_notifier(struct notifier_block *nb,
236 enum fib_event_type event_type,
237 struct fib_notifier_info *info);
238int call_fib4_notifiers(struct net *net, enum fib_event_type event_type,
239 struct fib_notifier_info *info);
240
241int __net_init fib4_notifier_init(struct net *net);
242void __net_exit fib4_notifier_exit(struct net *net);
243
244void fib_info_notify_update(struct net *net, struct nl_info *info);
245int fib_notify(struct net *net, struct notifier_block *nb,
246 struct netlink_ext_ack *extack);
247
248struct fib_table {
249 struct hlist_node tb_hlist;
250 u32 tb_id;
251 int tb_num_default;
252 struct rcu_head rcu;
253 unsigned long *tb_data;
254 unsigned long __data[];
255};
256
257struct fib_dump_filter {
258 u32 table_id;
259 /* filter_set is an optimization that an entry is set */
260 bool filter_set;
261 bool dump_routes;
262 bool dump_exceptions;
263 unsigned char protocol;
264 unsigned char rt_type;
265 unsigned int flags;
266 struct net_device *dev;
267};
268
269int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
270 struct fib_result *res, int fib_flags);
271int fib_table_insert(struct net *, struct fib_table *, struct fib_config *,
272 struct netlink_ext_ack *extack);
273int fib_table_delete(struct net *, struct fib_table *, struct fib_config *,
274 struct netlink_ext_ack *extack);
275int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
276 struct netlink_callback *cb, struct fib_dump_filter *filter);
277int fib_table_flush(struct net *net, struct fib_table *table, bool flush_all);
278struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
279void fib_table_flush_external(struct fib_table *table);
280void fib_free_table(struct fib_table *tb);
281
282#ifndef CONFIG_IP_MULTIPLE_TABLES
283
284#define TABLE_LOCAL_INDEX (RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
285#define TABLE_MAIN_INDEX (RT_TABLE_MAIN & (FIB_TABLE_HASHSZ - 1))
286
287static inline struct fib_table *fib_get_table(struct net *net, u32 id)
288{
289 struct hlist_node *tb_hlist;
290 struct hlist_head *ptr;
291
292 ptr = id == RT_TABLE_LOCAL ?
293 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
294 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
295
296 tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
297
298 return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
299}
300
301static inline struct fib_table *fib_new_table(struct net *net, u32 id)
302{
303 return fib_get_table(net, id);
304}
305
306static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
307 struct fib_result *res, unsigned int flags)
308{
309 struct fib_table *tb;
310 int err = -ENETUNREACH;
311
312 rcu_read_lock();
313
314 tb = fib_get_table(net, RT_TABLE_MAIN);
315 if (tb)
316 err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
317
318 if (err == -EAGAIN)
319 err = -ENETUNREACH;
320
321 rcu_read_unlock();
322
323 return err;
324}
325
326static inline bool fib4_has_custom_rules(const struct net *net)
327{
328 return false;
329}
330
331static inline bool fib4_rule_default(const struct fib_rule *rule)
332{
333 return true;
334}
335
336static inline int fib4_rules_dump(struct net *net, struct notifier_block *nb,
337 struct netlink_ext_ack *extack)
338{
339 return 0;
340}
341
342static inline unsigned int fib4_rules_seq_read(struct net *net)
343{
344 return 0;
345}
346
347static inline bool fib4_rules_early_flow_dissect(struct net *net,
348 struct sk_buff *skb,
349 struct flowi4 *fl4,
350 struct flow_keys *flkeys)
351{
352 return false;
353}
354#else /* CONFIG_IP_MULTIPLE_TABLES */
355int __net_init fib4_rules_init(struct net *net);
356void __net_exit fib4_rules_exit(struct net *net);
357
358struct fib_table *fib_new_table(struct net *net, u32 id);
359struct fib_table *fib_get_table(struct net *net, u32 id);
360
361int __fib_lookup(struct net *net, struct flowi4 *flp,
362 struct fib_result *res, unsigned int flags);
363
364static inline int fib_lookup(struct net *net, struct flowi4 *flp,
365 struct fib_result *res, unsigned int flags)
366{
367 struct fib_table *tb;
368 int err = -ENETUNREACH;
369
370 flags |= FIB_LOOKUP_NOREF;
371 if (net->ipv4.fib_has_custom_rules)
372 return __fib_lookup(net, flp, res, flags);
373
374 rcu_read_lock();
375
376 res->tclassid = 0;
377
378 tb = rcu_dereference_rtnl(net->ipv4.fib_main);
379 if (tb)
380 err = fib_table_lookup(tb, flp, res, flags);
381
382 if (!err)
383 goto out;
384
385 tb = rcu_dereference_rtnl(net->ipv4.fib_default);
386 if (tb)
387 err = fib_table_lookup(tb, flp, res, flags);
388
389out:
390 if (err == -EAGAIN)
391 err = -ENETUNREACH;
392
393 rcu_read_unlock();
394
395 return err;
396}
397
398static inline bool fib4_has_custom_rules(const struct net *net)
399{
400 return net->ipv4.fib_has_custom_rules;
401}
402
403bool fib4_rule_default(const struct fib_rule *rule);
404int fib4_rules_dump(struct net *net, struct notifier_block *nb,
405 struct netlink_ext_ack *extack);
406unsigned int fib4_rules_seq_read(struct net *net);
407
408static inline bool fib4_rules_early_flow_dissect(struct net *net,
409 struct sk_buff *skb,
410 struct flowi4 *fl4,
411 struct flow_keys *flkeys)
412{
413 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
414
415 if (!net->ipv4.fib_rules_require_fldissect)
416 return false;
417
418 skb_flow_dissect_flow_keys(skb, flkeys, flag);
419 fl4->fl4_sport = flkeys->ports.src;
420 fl4->fl4_dport = flkeys->ports.dst;
421 fl4->flowi4_proto = flkeys->basic.ip_proto;
422
423 return true;
424}
425
426#endif /* CONFIG_IP_MULTIPLE_TABLES */
427
428/* Exported by fib_frontend.c */
429extern const struct nla_policy rtm_ipv4_policy[];
430void ip_fib_init(void);
431int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
432 struct netlink_ext_ack *extack);
433__be32 fib_compute_spec_dst(struct sk_buff *skb);
434bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev);
435int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
436 u8 tos, int oif, struct net_device *dev,
437 struct in_device *idev, u32 *itag);
438#ifdef CONFIG_IP_ROUTE_CLASSID
439static inline int fib_num_tclassid_users(struct net *net)
440{
441 return net->ipv4.fib_num_tclassid_users;
442}
443#else
444static inline int fib_num_tclassid_users(struct net *net)
445{
446 return 0;
447}
448#endif
449int fib_unmerge(struct net *net);
450
451static inline bool nhc_l3mdev_matches_dev(const struct fib_nh_common *nhc,
452const struct net_device *dev)
453{
454 if (nhc->nhc_dev == dev ||
455 l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex)
456 return true;
457
458 return false;
459}
460
461/* Exported by fib_semantics.c */
462int ip_fib_check_default(__be32 gw, struct net_device *dev);
463int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
464int fib_sync_down_addr(struct net_device *dev, __be32 local);
465int fib_sync_up(struct net_device *dev, unsigned char nh_flags);
466void fib_sync_mtu(struct net_device *dev, u32 orig_mtu);
467void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig);
468
469/* Fields used for sysctl_fib_multipath_hash_fields.
470 * Common to IPv4 and IPv6.
471 *
472 * Add new fields at the end. This is user API.
473 */
474#define FIB_MULTIPATH_HASH_FIELD_SRC_IP BIT(0)
475#define FIB_MULTIPATH_HASH_FIELD_DST_IP BIT(1)
476#define FIB_MULTIPATH_HASH_FIELD_IP_PROTO BIT(2)
477#define FIB_MULTIPATH_HASH_FIELD_FLOWLABEL BIT(3)
478#define FIB_MULTIPATH_HASH_FIELD_SRC_PORT BIT(4)
479#define FIB_MULTIPATH_HASH_FIELD_DST_PORT BIT(5)
480#define FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP BIT(6)
481#define FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP BIT(7)
482#define FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO BIT(8)
483#define FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL BIT(9)
484#define FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT BIT(10)
485#define FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT BIT(11)
486
487#define FIB_MULTIPATH_HASH_FIELD_OUTER_MASK \
488 (FIB_MULTIPATH_HASH_FIELD_SRC_IP | \
489 FIB_MULTIPATH_HASH_FIELD_DST_IP | \
490 FIB_MULTIPATH_HASH_FIELD_IP_PROTO | \
491 FIB_MULTIPATH_HASH_FIELD_FLOWLABEL | \
492 FIB_MULTIPATH_HASH_FIELD_SRC_PORT | \
493 FIB_MULTIPATH_HASH_FIELD_DST_PORT)
494
495#define FIB_MULTIPATH_HASH_FIELD_INNER_MASK \
496 (FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP | \
497 FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP | \
498 FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO | \
499 FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL | \
500 FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT | \
501 FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
502
503#define FIB_MULTIPATH_HASH_FIELD_ALL_MASK \
504 (FIB_MULTIPATH_HASH_FIELD_OUTER_MASK | \
505 FIB_MULTIPATH_HASH_FIELD_INNER_MASK)
506
507#define FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK \
508 (FIB_MULTIPATH_HASH_FIELD_SRC_IP | \
509 FIB_MULTIPATH_HASH_FIELD_DST_IP | \
510 FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
511
512#ifdef CONFIG_IP_ROUTE_MULTIPATH
513int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
514 const struct sk_buff *skb, struct flow_keys *flkeys);
515#endif
516int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
517 struct netlink_ext_ack *extack);
518void fib_select_multipath(struct fib_result *res, int hash);
519void fib_select_path(struct net *net, struct fib_result *res,
520 struct flowi4 *fl4, const struct sk_buff *skb);
521
522int fib_nh_init(struct net *net, struct fib_nh *fib_nh,
523 struct fib_config *cfg, int nh_weight,
524 struct netlink_ext_ack *extack);
525void fib_nh_release(struct net *net, struct fib_nh *fib_nh);
526int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
527 struct nlattr *fc_encap, u16 fc_encap_type,
528 void *cfg, gfp_t gfp_flags,
529 struct netlink_ext_ack *extack);
530void fib_nh_common_release(struct fib_nh_common *nhc);
531
532/* Exported by fib_trie.c */
533void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri);
534void fib_trie_init(void);
535struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
536bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
537 const struct flowi4 *flp);
538
539static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
540{
541#ifdef CONFIG_IP_ROUTE_CLASSID
542 struct fib_nh_common *nhc = res->nhc;
543#ifdef CONFIG_IP_MULTIPLE_TABLES
544 u32 rtag;
545#endif
546 if (nhc->nhc_family == AF_INET) {
547 struct fib_nh *nh;
548
549 nh = container_of(nhc, struct fib_nh, nh_common);
550 *itag = nh->nh_tclassid << 16;
551 } else {
552 *itag = 0;
553 }
554
555#ifdef CONFIG_IP_MULTIPLE_TABLES
556 rtag = res->tclassid;
557 if (*itag == 0)
558 *itag = (rtag<<16);
559 *itag |= (rtag>>16);
560#endif
561#endif
562}
563
564void fib_flush(struct net *net);
565void free_fib_info(struct fib_info *fi);
566
567static inline void fib_info_hold(struct fib_info *fi)
568{
569 refcount_inc(&fi->fib_clntref);
570}
571
572static inline void fib_info_put(struct fib_info *fi)
573{
574 if (refcount_dec_and_test(&fi->fib_clntref))
575 free_fib_info(fi);
576}
577
578#ifdef CONFIG_PROC_FS
579int __net_init fib_proc_init(struct net *net);
580void __net_exit fib_proc_exit(struct net *net);
581#else
582static inline int fib_proc_init(struct net *net)
583{
584 return 0;
585}
586static inline void fib_proc_exit(struct net *net)
587{
588}
589#endif
590
591u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr);
592
593int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
594 struct fib_dump_filter *filter,
595 struct netlink_callback *cb);
596
597int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nh,
598 u8 rt_family, unsigned char *flags, bool skip_oif);
599int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nh,
600 int nh_weight, u8 rt_family);
601#endif /* _NET_FIB_H */