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/inet_dscp.h>
21#include <net/inetpeer.h>
22#include <linux/percpu.h>
23#include <linux/notifier.h>
24#include <linux/refcount.h>
25#include <linux/ip.h>
26#include <linux/in_route.h>
27
28struct fib_config {
29 u8 fc_dst_len;
30 dscp_t fc_dscp;
31 u8 fc_protocol;
32 u8 fc_scope;
33 u8 fc_type;
34 u8 fc_gw_family;
35 /* 2 bytes unused */
36 u32 fc_table;
37 __be32 fc_dst;
38 union {
39 __be32 fc_gw4;
40 struct in6_addr fc_gw6;
41 };
42 int fc_oif;
43 u32 fc_flags;
44 u32 fc_priority;
45 __be32 fc_prefsrc;
46 u32 fc_nh_id;
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
58struct fib_info;
59struct rtable;
60
61struct 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
75struct 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
83struct fib_nh_common {
84 struct net_device *nhc_dev;
85 netdevice_tracker nhc_dev_tracker;
86 int nhc_oif;
87 unsigned char nhc_scope;
88 u8 nhc_family;
89 u8 nhc_gw_family;
90 unsigned char nhc_flags;
91 struct lwtunnel_state *nhc_lwtstate;
92
93 union {
94 __be32 ipv4;
95 struct in6_addr ipv6;
96 } nhc_gw;
97
98 int nhc_weight;
99 atomic_t nhc_upper_bound;
100
101 /* v4 specific, but allows fib6_nh with v4 routes */
102 struct rtable __rcu * __percpu *nhc_pcpu_rth_output;
103 struct rtable __rcu *nhc_rth_input;
104 struct fnhe_hash_bucket __rcu *nhc_exceptions;
105};
106
107struct fib_nh {
108 struct fib_nh_common nh_common;
109 struct hlist_node nh_hash;
110 struct fib_info *nh_parent;
111#ifdef CONFIG_IP_ROUTE_CLASSID
112 __u32 nh_tclassid;
113#endif
114 __be32 nh_saddr;
115 int nh_saddr_genid;
116#define fib_nh_family nh_common.nhc_family
117#define fib_nh_dev nh_common.nhc_dev
118#define fib_nh_dev_tracker nh_common.nhc_dev_tracker
119#define fib_nh_oif nh_common.nhc_oif
120#define fib_nh_flags nh_common.nhc_flags
121#define fib_nh_lws nh_common.nhc_lwtstate
122#define fib_nh_scope nh_common.nhc_scope
123#define fib_nh_gw_family nh_common.nhc_gw_family
124#define fib_nh_gw4 nh_common.nhc_gw.ipv4
125#define fib_nh_gw6 nh_common.nhc_gw.ipv6
126#define fib_nh_weight nh_common.nhc_weight
127#define fib_nh_upper_bound nh_common.nhc_upper_bound
128};
129
130/*
131 * This structure contains data shared by many of routes.
132 */
133
134struct nexthop;
135
136struct fib_info {
137 struct hlist_node fib_hash;
138 struct hlist_node fib_lhash;
139 struct list_head nh_list;
140 struct net *fib_net;
141 refcount_t fib_treeref;
142 refcount_t fib_clntref;
143 unsigned int fib_flags;
144 unsigned char fib_dead;
145 unsigned char fib_protocol;
146 unsigned char fib_scope;
147 unsigned char fib_type;
148 __be32 fib_prefsrc;
149 u32 fib_tb_id;
150 u32 fib_priority;
151 struct dst_metrics *fib_metrics;
152#define fib_mtu fib_metrics->metrics[RTAX_MTU-1]
153#define fib_window fib_metrics->metrics[RTAX_WINDOW-1]
154#define fib_rtt fib_metrics->metrics[RTAX_RTT-1]
155#define fib_advmss fib_metrics->metrics[RTAX_ADVMSS-1]
156 int fib_nhs;
157 bool fib_nh_is_v6;
158 bool nh_updated;
159 bool pfsrc_removed;
160 struct nexthop *nh;
161 struct rcu_head rcu;
162 struct fib_nh fib_nh[] __counted_by(fib_nhs);
163};
164
165
166#ifdef CONFIG_IP_MULTIPLE_TABLES
167struct fib_rule;
168#endif
169
170struct fib_table;
171struct fib_result {
172 __be32 prefix;
173 unsigned char prefixlen;
174 unsigned char nh_sel;
175 unsigned char type;
176 unsigned char scope;
177 u32 tclassid;
178 dscp_t dscp;
179 struct fib_nh_common *nhc;
180 struct fib_info *fi;
181 struct fib_table *table;
182 struct hlist_head *fa_head;
183};
184
185struct fib_result_nl {
186 __be32 fl_addr; /* To be looked up*/
187 u32 fl_mark;
188 unsigned char fl_tos;
189 unsigned char fl_scope;
190 unsigned char tb_id_in;
191
192 unsigned char tb_id; /* Results */
193 unsigned char prefixlen;
194 unsigned char nh_sel;
195 unsigned char type;
196 unsigned char scope;
197 int err;
198};
199
200#ifdef CONFIG_IP_MULTIPLE_TABLES
201#define FIB_TABLE_HASHSZ 256
202#else
203#define FIB_TABLE_HASHSZ 2
204#endif
205
206__be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
207 unsigned char scope);
208__be32 fib_result_prefsrc(struct net *net, struct fib_result *res);
209
210#define FIB_RES_NHC(res) ((res).nhc)
211#define FIB_RES_DEV(res) (FIB_RES_NHC(res)->nhc_dev)
212#define FIB_RES_OIF(res) (FIB_RES_NHC(res)->nhc_oif)
213
214struct fib_rt_info {
215 struct fib_info *fi;
216 u32 tb_id;
217 __be32 dst;
218 int dst_len;
219 dscp_t dscp;
220 u8 type;
221 u8 offload:1,
222 trap:1,
223 offload_failed:1,
224 unused:5;
225};
226
227struct fib_entry_notifier_info {
228 struct fib_notifier_info info; /* must be first */
229 u32 dst;
230 int dst_len;
231 struct fib_info *fi;
232 dscp_t dscp;
233 u8 type;
234 u32 tb_id;
235};
236
237struct fib_nh_notifier_info {
238 struct fib_notifier_info info; /* must be first */
239 struct fib_nh *fib_nh;
240};
241
242int call_fib4_notifier(struct notifier_block *nb,
243 enum fib_event_type event_type,
244 struct fib_notifier_info *info);
245int call_fib4_notifiers(struct net *net, enum fib_event_type event_type,
246 struct fib_notifier_info *info);
247
248int __net_init fib4_notifier_init(struct net *net);
249void __net_exit fib4_notifier_exit(struct net *net);
250
251void fib_info_notify_update(struct net *net, struct nl_info *info);
252int fib_notify(struct net *net, struct notifier_block *nb,
253 struct netlink_ext_ack *extack);
254
255struct fib_table {
256 struct hlist_node tb_hlist;
257 u32 tb_id;
258 int tb_num_default;
259 struct rcu_head rcu;
260 unsigned long *tb_data;
261 unsigned long __data[];
262};
263
264struct fib_dump_filter {
265 u32 table_id;
266 /* filter_set is an optimization that an entry is set */
267 bool filter_set;
268 bool dump_routes;
269 bool dump_exceptions;
270 bool rtnl_held;
271 unsigned char protocol;
272 unsigned char rt_type;
273 unsigned int flags;
274 struct net_device *dev;
275};
276
277int fib_table_lookup(struct fib_table *tb, const struct flowi4 *flp,
278 struct fib_result *res, int fib_flags);
279int fib_table_insert(struct net *, struct fib_table *, struct fib_config *,
280 struct netlink_ext_ack *extack);
281int fib_table_delete(struct net *, struct fib_table *, struct fib_config *,
282 struct netlink_ext_ack *extack);
283int fib_table_dump(struct fib_table *table, struct sk_buff *skb,
284 struct netlink_callback *cb, struct fib_dump_filter *filter);
285int fib_table_flush(struct net *net, struct fib_table *table, bool flush_all);
286struct fib_table *fib_trie_unmerge(struct fib_table *main_tb);
287void fib_table_flush_external(struct fib_table *table);
288void fib_free_table(struct fib_table *tb);
289
290#ifndef CONFIG_IP_MULTIPLE_TABLES
291
292#define TABLE_LOCAL_INDEX (RT_TABLE_LOCAL & (FIB_TABLE_HASHSZ - 1))
293#define TABLE_MAIN_INDEX (RT_TABLE_MAIN & (FIB_TABLE_HASHSZ - 1))
294
295static inline struct fib_table *fib_get_table(struct net *net, u32 id)
296{
297 struct hlist_node *tb_hlist;
298 struct hlist_head *ptr;
299
300 ptr = id == RT_TABLE_LOCAL ?
301 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX] :
302 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX];
303
304 tb_hlist = rcu_dereference_rtnl(hlist_first_rcu(ptr));
305
306 return hlist_entry(tb_hlist, struct fib_table, tb_hlist);
307}
308
309static inline struct fib_table *fib_new_table(struct net *net, u32 id)
310{
311 return fib_get_table(net, id);
312}
313
314static inline int fib_lookup(struct net *net, const struct flowi4 *flp,
315 struct fib_result *res, unsigned int flags)
316{
317 struct fib_table *tb;
318 int err = -ENETUNREACH;
319
320 rcu_read_lock();
321
322 tb = fib_get_table(net, RT_TABLE_MAIN);
323 if (tb)
324 err = fib_table_lookup(tb, flp, res, flags | FIB_LOOKUP_NOREF);
325
326 if (err == -EAGAIN)
327 err = -ENETUNREACH;
328
329 rcu_read_unlock();
330
331 return err;
332}
333
334static inline bool fib4_has_custom_rules(const struct net *net)
335{
336 return false;
337}
338
339static inline bool fib4_rule_default(const struct fib_rule *rule)
340{
341 return true;
342}
343
344static inline int fib4_rules_dump(struct net *net, struct notifier_block *nb,
345 struct netlink_ext_ack *extack)
346{
347 return 0;
348}
349
350static inline unsigned int fib4_rules_seq_read(struct net *net)
351{
352 return 0;
353}
354
355static inline bool fib4_rules_early_flow_dissect(struct net *net,
356 struct sk_buff *skb,
357 struct flowi4 *fl4,
358 struct flow_keys *flkeys)
359{
360 return false;
361}
362#else /* CONFIG_IP_MULTIPLE_TABLES */
363int __net_init fib4_rules_init(struct net *net);
364void __net_exit fib4_rules_exit(struct net *net);
365
366struct fib_table *fib_new_table(struct net *net, u32 id);
367struct fib_table *fib_get_table(struct net *net, u32 id);
368
369int __fib_lookup(struct net *net, struct flowi4 *flp,
370 struct fib_result *res, unsigned int flags);
371
372static inline int fib_lookup(struct net *net, struct flowi4 *flp,
373 struct fib_result *res, unsigned int flags)
374{
375 struct fib_table *tb;
376 int err = -ENETUNREACH;
377
378 flags |= FIB_LOOKUP_NOREF;
379 if (net->ipv4.fib_has_custom_rules)
380 return __fib_lookup(net, flp, res, flags);
381
382 rcu_read_lock();
383
384 res->tclassid = 0;
385
386 tb = rcu_dereference_rtnl(net->ipv4.fib_main);
387 if (tb)
388 err = fib_table_lookup(tb, flp, res, flags);
389
390 if (!err)
391 goto out;
392
393 tb = rcu_dereference_rtnl(net->ipv4.fib_default);
394 if (tb)
395 err = fib_table_lookup(tb, flp, res, flags);
396
397out:
398 if (err == -EAGAIN)
399 err = -ENETUNREACH;
400
401 rcu_read_unlock();
402
403 return err;
404}
405
406static inline bool fib4_has_custom_rules(const struct net *net)
407{
408 return net->ipv4.fib_has_custom_rules;
409}
410
411bool fib4_rule_default(const struct fib_rule *rule);
412int fib4_rules_dump(struct net *net, struct notifier_block *nb,
413 struct netlink_ext_ack *extack);
414unsigned int fib4_rules_seq_read(struct net *net);
415
416static inline bool fib4_rules_early_flow_dissect(struct net *net,
417 struct sk_buff *skb,
418 struct flowi4 *fl4,
419 struct flow_keys *flkeys)
420{
421 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
422
423 if (!net->ipv4.fib_rules_require_fldissect)
424 return false;
425
426 memset(flkeys, 0, sizeof(*flkeys));
427 __skb_flow_dissect(net, skb, &flow_keys_dissector,
428 flkeys, NULL, 0, 0, 0, flag);
429
430 fl4->fl4_sport = flkeys->ports.src;
431 fl4->fl4_dport = flkeys->ports.dst;
432 fl4->flowi4_proto = flkeys->basic.ip_proto;
433
434 return true;
435}
436
437#endif /* CONFIG_IP_MULTIPLE_TABLES */
438
439static inline bool fib_dscp_masked_match(dscp_t dscp, const struct flowi4 *fl4)
440{
441 return dscp == inet_dsfield_to_dscp(RT_TOS(fl4->flowi4_tos));
442}
443
444/* Exported by fib_frontend.c */
445extern const struct nla_policy rtm_ipv4_policy[];
446void ip_fib_init(void);
447int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
448 struct netlink_ext_ack *extack);
449__be32 fib_compute_spec_dst(struct sk_buff *skb);
450bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev);
451int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
452 u8 tos, int oif, struct net_device *dev,
453 struct in_device *idev, u32 *itag);
454#ifdef CONFIG_IP_ROUTE_CLASSID
455static inline int fib_num_tclassid_users(struct net *net)
456{
457 return atomic_read(&net->ipv4.fib_num_tclassid_users);
458}
459#else
460static inline int fib_num_tclassid_users(struct net *net)
461{
462 return 0;
463}
464#endif
465int fib_unmerge(struct net *net);
466
467static inline bool nhc_l3mdev_matches_dev(const struct fib_nh_common *nhc,
468const struct net_device *dev)
469{
470 if (nhc->nhc_dev == dev ||
471 l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex)
472 return true;
473
474 return false;
475}
476
477/* Exported by fib_semantics.c */
478int ip_fib_check_default(__be32 gw, struct net_device *dev);
479int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force);
480int fib_sync_down_addr(struct net_device *dev, __be32 local);
481int fib_sync_up(struct net_device *dev, unsigned char nh_flags);
482void fib_sync_mtu(struct net_device *dev, u32 orig_mtu);
483void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig);
484
485/* Fields used for sysctl_fib_multipath_hash_fields.
486 * Common to IPv4 and IPv6.
487 *
488 * Add new fields at the end. This is user API.
489 */
490#define FIB_MULTIPATH_HASH_FIELD_SRC_IP BIT(0)
491#define FIB_MULTIPATH_HASH_FIELD_DST_IP BIT(1)
492#define FIB_MULTIPATH_HASH_FIELD_IP_PROTO BIT(2)
493#define FIB_MULTIPATH_HASH_FIELD_FLOWLABEL BIT(3)
494#define FIB_MULTIPATH_HASH_FIELD_SRC_PORT BIT(4)
495#define FIB_MULTIPATH_HASH_FIELD_DST_PORT BIT(5)
496#define FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP BIT(6)
497#define FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP BIT(7)
498#define FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO BIT(8)
499#define FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL BIT(9)
500#define FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT BIT(10)
501#define FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT BIT(11)
502
503#define FIB_MULTIPATH_HASH_FIELD_OUTER_MASK \
504 (FIB_MULTIPATH_HASH_FIELD_SRC_IP | \
505 FIB_MULTIPATH_HASH_FIELD_DST_IP | \
506 FIB_MULTIPATH_HASH_FIELD_IP_PROTO | \
507 FIB_MULTIPATH_HASH_FIELD_FLOWLABEL | \
508 FIB_MULTIPATH_HASH_FIELD_SRC_PORT | \
509 FIB_MULTIPATH_HASH_FIELD_DST_PORT)
510
511#define FIB_MULTIPATH_HASH_FIELD_INNER_MASK \
512 (FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP | \
513 FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP | \
514 FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO | \
515 FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL | \
516 FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT | \
517 FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
518
519#define FIB_MULTIPATH_HASH_FIELD_ALL_MASK \
520 (FIB_MULTIPATH_HASH_FIELD_OUTER_MASK | \
521 FIB_MULTIPATH_HASH_FIELD_INNER_MASK)
522
523#define FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK \
524 (FIB_MULTIPATH_HASH_FIELD_SRC_IP | \
525 FIB_MULTIPATH_HASH_FIELD_DST_IP | \
526 FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
527
528#ifdef CONFIG_IP_ROUTE_MULTIPATH
529int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4,
530 const struct sk_buff *skb, struct flow_keys *flkeys);
531
532static void
533fib_multipath_hash_construct_key(siphash_key_t *key, u32 mp_seed)
534{
535 u64 mp_seed_64 = mp_seed;
536
537 key->key[0] = (mp_seed_64 << 32) | mp_seed_64;
538 key->key[1] = key->key[0];
539}
540
541static inline u32 fib_multipath_hash_from_keys(const struct net *net,
542 struct flow_keys *keys)
543{
544 siphash_aligned_key_t hash_key;
545 u32 mp_seed;
546
547 mp_seed = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_seed).mp_seed;
548 fib_multipath_hash_construct_key(&hash_key, mp_seed);
549
550 return flow_hash_from_keys_seed(keys, &hash_key);
551}
552#else
553static inline u32 fib_multipath_hash_from_keys(const struct net *net,
554 struct flow_keys *keys)
555{
556 return flow_hash_from_keys(keys);
557}
558#endif
559
560int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
561 struct netlink_ext_ack *extack);
562void fib_select_multipath(struct fib_result *res, int hash);
563void fib_select_path(struct net *net, struct fib_result *res,
564 struct flowi4 *fl4, const struct sk_buff *skb);
565
566int fib_nh_init(struct net *net, struct fib_nh *fib_nh,
567 struct fib_config *cfg, int nh_weight,
568 struct netlink_ext_ack *extack);
569void fib_nh_release(struct net *net, struct fib_nh *fib_nh);
570int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
571 struct nlattr *fc_encap, u16 fc_encap_type,
572 void *cfg, gfp_t gfp_flags,
573 struct netlink_ext_ack *extack);
574void fib_nh_common_release(struct fib_nh_common *nhc);
575
576/* Exported by fib_trie.c */
577void fib_alias_hw_flags_set(struct net *net, const struct fib_rt_info *fri);
578void fib_trie_init(void);
579struct fib_table *fib_trie_table(u32 id, struct fib_table *alias);
580bool fib_lookup_good_nhc(const struct fib_nh_common *nhc, int fib_flags,
581 const struct flowi4 *flp);
582
583static inline void fib_combine_itag(u32 *itag, const struct fib_result *res)
584{
585#ifdef CONFIG_IP_ROUTE_CLASSID
586 struct fib_nh_common *nhc = res->nhc;
587#ifdef CONFIG_IP_MULTIPLE_TABLES
588 u32 rtag;
589#endif
590 if (nhc->nhc_family == AF_INET) {
591 struct fib_nh *nh;
592
593 nh = container_of(nhc, struct fib_nh, nh_common);
594 *itag = nh->nh_tclassid << 16;
595 } else {
596 *itag = 0;
597 }
598
599#ifdef CONFIG_IP_MULTIPLE_TABLES
600 rtag = res->tclassid;
601 if (*itag == 0)
602 *itag = (rtag<<16);
603 *itag |= (rtag>>16);
604#endif
605#endif
606}
607
608void fib_flush(struct net *net);
609void free_fib_info(struct fib_info *fi);
610
611static inline void fib_info_hold(struct fib_info *fi)
612{
613 refcount_inc(&fi->fib_clntref);
614}
615
616static inline void fib_info_put(struct fib_info *fi)
617{
618 if (refcount_dec_and_test(&fi->fib_clntref))
619 free_fib_info(fi);
620}
621
622#ifdef CONFIG_PROC_FS
623int __net_init fib_proc_init(struct net *net);
624void __net_exit fib_proc_exit(struct net *net);
625#else
626static inline int fib_proc_init(struct net *net)
627{
628 return 0;
629}
630static inline void fib_proc_exit(struct net *net)
631{
632}
633#endif
634
635u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr);
636
637int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
638 struct fib_dump_filter *filter,
639 struct netlink_callback *cb);
640
641int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nh,
642 u8 rt_family, unsigned char *flags, bool skip_oif);
643int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nh,
644 int nh_weight, u8 rt_family, u32 nh_tclassid);
645#endif /* _NET_FIB_H */